Quantitative Risk Assessment in Fire Safety

Quantitative Risk Assessment
in Fire Safety
Quantitative Risk Assessment
in Fire Safety
Ganapathy Ramachandran
and David Charters
First published 2011
by Spon Press
2 Park Square, Milton Park, Abingdon, Oxon OX14 4RN
Simultaneously published in the USA and Canada by Spon Press
270 Madison Avenue, New York, NY 10016, USA
Spon Press is an imprint of the Taylor & Francis Group, an informa business
© 2011 Ganapathy Ramachandran and David Charters
The right of Ganapathy Ramachandran and David Charters to be
identified as authors of this work has been asserted by them in accordance
with sections 77 and 78 of the Copyright, Designs and Patents Act 1988.
Typeset in Goudy by
HWA Text and Data Management, London
Printed and bound in Great Britain by
CPI Anthony Rowe, Chippenham, Wiltshire
All rights reserved. No part of this book may be reprinted or reproduced or
utilised in any form or by any electronic, mechanical, or other means, now
known or hereafter invented, including photocopying and recording, or in
any information storage or retrieval system, without permission in writing
from the publishers.
This publication presents material of a broad scope and applicability.
Despite stringent efforts by all concerned in the publishing process, some
typographical or editorial errors may occur, and readers are encouraged
to bring these to our attention where they represent errors of substance.
The publisher and author disclaim any liability, in whole or in part, arising
from information contained in this publication. The reader is urged to
consult with an appropriate licensed professional prior to taking any
action or making any interpretation that is within the realm of a licensed
professional practice.
British Library Cataloguing in Publication Data
A catalogue record for this book is available from the British Library
Library of Congress Cataloging-in-Publication Data
Ramachandran, G. (Ganapathy)
Quantitative risk assessment in fire safety / Ganapathy Ramachandran &
David A. Charters.
p. cm.
Includes bibliographical references and index.
1. Fire risk assessment. 2. Quantitative research. I. Charters,
David A. (David Anderson), 1949– II. Title.
TH9176.R36 2009
363.37’7—dc22 2009022649
ISBN13: 978-0-419-20790-0 (hbk)
ISBN13: 978-0-203-93769-3 (ebk)
Contents
List of figures vi
List of tables viii
About the authors x
Preface xi
Acknowledgements xiii
1 Introduction 1
2 Qualitative and semi-quantitative risk assessment techniques 34
3 Quantitative risk assessment techniques 70
4 Acceptance criteria 175
5 Initiation 196
6 Design fire size 239
7 Fire spread beyond room of origin 255
8 Performance and reliability of detection, alarm and suppression 268
9 Performance and reliability of human response and evacuation 295
10 Performance and effectiveness of fire service intervention 310
11 Whole project analysis 327
12 Interactions 341
13 Combining data from various sources – Bayesian techniques 352
Index 359
Figures
1.1 Breakdown of different fire risk assessment methods 13

1.2 Schematic of full quantitative fire risk assessment process 17
1.3 An F–n curve 20
2.1 An example of combustible materials 36
2.2 An example of a car fire 37
2.3 Summary report for an unstructured fire risk assessment 39
2.4 Record sheet from a structured qualitative fire risk assessment
method 41
2.5 An example of a matrix method flow chart 44
3.1 Venn diagram 75
3.2 Venn diagram of union set 76
3.3 Venn diagram of intersection set 77
3.4 An example of a fault tree 79
3.5 General form of an event tree 89
3.6 Event tree of the early stages of fire event development 90
3.7 Simplified event tree for bus garage fires 92
3.8 General form of a fault tree 95
3.9 Fault tree for failure to detect a fire within five minutes of ignition 96
3.10 Average time (min) and area damaged (m2) 104
3.11 Damage and building size – textile industry, UK 107
3.12 Fire growth within room of origin 108
3.13 Discovery time and fatality rate 108
3.14 Density function curve of fire loss 112
3.15 Textile industry, United Kingdom – probability distribution of area
damaged 113
3.16 Pareto distribution of area damage – retail premises 114
3.17 The survivor probability distribution of fire loss for each class
in the textile industry 118
3.18 Probability tree for textile industry 123
3.19 Room layout and corresponding graph 126
3.20 Probabilistic network of fire spread of Room 1 to C2 127
3.21 Equivalent fire spread network with five-minute unrated doors 128
3.22 Typical consequence probability distribution 138
List of figures vii
3.23 Correlation of the number of non-fatal injuries per fire and the
area damaged by fire, heat and smoke 139
3.24 Ticket hall after the fire at King’s Cross underground station 139
3.25 Fire testing of a shutter 140
3.26 Example of a full-scale fire experiment 141
3.27 Typical schematic of building fire control volume model 142
3.28 Schematic of mass flow between control volumes 143
3.29 Schematics of a plume in natural ventilation 147
3.30 Typical schematic of building fire heat transfer processes 147
3.31 Typical results for a three-layer control volume model compared
with experimental data 148
3.32 Example of computational fluid dynamics 151
3.33 Example of discrete evacuation analysis 152
4.1 Life safety acceptance criteria shown on an F–n Curve. 180
4.2 Experience per 106 person years of fires with N or more fatalities 184
4.3 Targets for acceptability for societal risk from process industries 185
4.4 Multiple fatality fires in buildings 186
4.5 Proposed target for acceptibility from multiple fatality fires in all
buildings other than dwellings for different populations 187
5.1 Probability tree for rooms and offices 218
5.2 Sprinkler alternative probability tree 229
7.1 Effect of beam deflection on a fire-resisting wall 265
9.1 Occupants’ response – room of fire origin 296
9.2 Occupants’ response – room other than room of fire origin 297
10.1 Effectiveness of early intervention by fire brigade – example
based on UK fire statistics, non-sprinklered buildings 323
10.2 Effectiveness of early intervention by fire brigade – example
based on UK fire statistics, sprinklered buildings 324
11.1 Typical quantified fire risk assessment process 330
11.2 Typical output from SAFiRE 337
12.1 Damage and building size 343
12.2 Damage and compartment size 344
Tables
2.1 Typical matrix method record sheet 45

2.2 Hazard /accident frequency 46
2.3 Accident trigger probability 46
2.4 Accident frequency categories 47
2.5 Accident severity 47
2.6 Risk classification 48
2.7 Risk tolerability 48
2.8 Hazard frequency 54
2.9 Accident severity (life safety) 54
2.10 Accident probability 54
2.11 Cost of asset damage or destroyed 54
2.12 Cost of performance penalties or operational loss 55
2.13 Life safety risk rating 55
2.14 Business risk rating 56
2.15 Overall station risk rating 56
2.16 Example of part of a risk ranking spreadsheet (most likely fire event) 57
A2.1 Relative S values for active systems 65
A2.2 Values of components 66
A2.3 Sample of a summary sheet 67
3.1 Relationship between random processes and probability modelling 72
3.2 Truth table for an OR gate 78
3.3 Truth table for an AND gate 78
3.4 Frequency of ignition per year 87
3.5 Frequency of ignition normalised per building 88
3.6 Frequency of ignition normalised by floor area 88
3.7 General data on the reliability of fire protection systems 93
3.8 Data used to quantify the probability of failure to detect a fire
within five minutes of ignition 97
3.9 Annual chance of a fire outbreak for various occupancies 99
3.10 Probable damage in a fire – parameters of Equation (3.41) 100
3.11 Industrial buildings – annual damage 102
3.12 Retail premises – assembly areas – frequency distribution of area
damage 115
List of tables ix
3.13 Average loss per fire at 1966 prices 117
3.14 Pathways for fire spread equivalent network assuming
five-minute unrated corridor doors 129
3.15 Pathways for the fire spread equivalent network assuming
self–closing 20-minute rated corridor doors 129
3.16 Passive fire protection measures in hospitals 130
3.17 Probabilities of success and failure (standard normal distribution) 167
4.1 Typical types of acceptance criteria 177
4.2 Number of deaths per building year and per occupant year 178
4.3 Societal risk criteria for structural collapse due to fire 192
4.4 Discounted cash flow for bus garage sprinkler system 194
5.1 Spinning and doubling industry – places of origin of fires and
sources of ignition 198
5.2 Fire growth parameters 204
5.3 Expected values of the fire growth rate θ for different scenarios
– paper, printing and publishing industries, UK 207
5.4 Maximum values of the fire growth rate θ for different scenarios
– paper, printing and publishing industries, UK 210
5.5 Growth rate and doubling time 212
5.6 Probability of flashover 221
5.7 Fire extent of spread – textile industry, UK 227
5.8 Location of casualties – single and multiple occupancy dwellings 231
5.9 Fatal casualties in dwellings by whereabouts of casualties and
cause of death 232
6.1 Horizontal fire-damaged area in retail premises, 1978 240
6.2 Area of direct burning and number of sprinkler heads opening for
different fractiles, sprinklered retail premises 241
6.3 Pareto parameters 243
6.4 Design fire size 244
6.5 Pareto distribution of area damage – rooms without and with
sprinklers 246
6.6 Design fire size 247
7.1 Probability of fire spreading beyond room of origin 256
7.2 Material ignited first – probability of fire spreading beyond room
of origin (chemical and allied industries) 258
8.1 Experiment 1: Summary of results and overall rankings 283
8.2 Experiment 2: Summary of results and overall rankings 284
8.3 Effectiveness of sprinklers – non-industrial buildings 290
9.1 Design evacuation time and fatality rate for single and multiple
occupancy dwellings 307
10.1 Direct loss in relation to attendance time – non-dwellings, UK 318
10.2 Percentage of small fires and fires extinguished by sprinklers
and by fire brigade 320
10.3 Control time and arrival time: regression coefficients 322
11.1 Risk–cost benefit analysis 339
About the authors
Ganapathy Ramachandran, PhD, DSc, FIFireE, MSFPE, held senior scientific

appointments for 23 years (1965–1988) at the Fire Research Station,
Borehamwood, Herts, UK. Since retiring in November 1988, he has been
practising as a private consultant mainly in research problems in the application
of statistical, probabilistic and economic techniques to fire risk evaluation, fire
protection engineering and fire insurance. He has published several papers on
these topics including chapters in the four editions of the SFPE Handbook of Fire
Protection Engineering. He is the author of Economics of Fire Protection (Spon 1998)
and co-author of Evaluation of Fire Safety (Wiley 2004).
Ramachandran was a Visiting Professor in Glasgow Caledonian University,
University of Hertfordshire and University of Manchester. For the past 12 years
he is a Visiting Professor in the University of Leeds. For three and a half years (up
until July 2010), he was Technical Director, Fire and Risk Engineering at Faber
Maunsell/AECOM.
David Charters, BSc, PhD, CEng, FIFireE, MIMechE, MSFPE is Director of
Fire Engineering at BRE Global (incorporating the Fire Research Station, FRS)
where he has been responsible for projects such as research into the use of lifts
and escalators for emergency evacuation and quantitative fire risk assessments
for nuclear and non-nuclear fire engineering projects. He has over 20 years of
fire safety experience. Prior to joining BRE Global, he was a Director of Arup
Fire responsible for the fire engineering of projects including; Channel Tunnel
Rail Link (CTRL) tunnels, City of Manchester Stadium and Attatürk Airport.
Previously he was Chief Engineer and Manager; Fire and Safety at NHS Estates,
Department of Health where he was responsible for policy advice to ministers
and the FIRECODE suite of guidance. Prior to this he was a Senior Fire Risk
Assessment Engineer with the UK Atomic Energy Authority.
In 2005, David became International President of the Institution of Fire
Engineers. He was appointed Visiting Professor in Fire Risk Analysis at the
University of Ulster in 2003. In 1994, David was awarded a PhD in Fire Growth
and Smoke Movement in Tunnels from the University of Leeds. He is co-author of
the chapter on Building Fire Risk Analysis in the SFPE Handbook on Fire Protection
Engineering. David was also responsible for quantitative fire risk assessments of
London Underground stations following the King’s Cross fire and Tartan Alpha
following the Piper Alpha disaster.
Preface
As viable alternative for prescriptive rules, particularly for large and complex
buildings, performance-based fire safety codes and building design methods
are being developed and applied in many countries. These rules, codes and
methods are mainly based on a qualitative assessment of fire risk supported by
experimental data, case studies, deterministic (scientific) models and professional
engineering practice. Statistical data provided by real fires are rarely analysed and
the results (evidence) produced by such an analysis are seldom included in the
risk assessment. Quantitative methods of risk assessment discussed in this book,
on the other hand, explicitly consider statistical data on real fires, in addition to
experimental data, and take account of uncertainties governing the occurrence of
a fire, spread of fire, damage caused in a fire, reliability of passive and active fire
protection systems and evacuation of building occupants.
Fire safety regulations, codes and standards do provide some unquantified
levels of safety particularly for the occupants of a building, but these levels may or
may not be adequate for some large, tall and complex buildings. Also, these levels
may not be acceptable to property owners who have to consider also property
damage and consequential losses such as business interruption and loss of profits.
Given that there are no quantitative criteria for fire risk in buildings, it is not
clear whether the safety levels provided by prescriptive rules are acceptable to
the society at large. Criteria for determining acceptable safety levels for property
owners and the society are discussed in Chapter 4.
Buildings can be designed according to acceptable levels for life safety and
property protection by applying quantitative methods of fire risk assessment.
These methods are discussed in detail in Chapter 3. Methods applicable to
some particular problems in fire safety engineering are discussed in other
chapters. These problems include initiation of fires (Chapter 5), design fire size
(Chapter 6), flashover and spread of a fire beyond room of fire origin (Chapter 7)
and performance and reliability of detection, alarm and suppression systems
(Chapter 8). Building designs should sufficiently consider response and evacuation
capability of occupants as discussed in Chapter 9.
Prescriptive rules specified in fire regulations, codes and standards should be
further verified, validated and improved, if necessary, in the light of quantitative
risk assessment. Otherwise, fire safety engineers will not be able to recommend
xii Preface
enhanced fire protection necessary for large tall and complex buildings. This is
because owners of such buildings are generally reluctant to spend more money
than the cost required for complying with prescriptive rules and performance-based
codes. The owners may not appreciate at present that enhanced fire protection may
be more cost-effective than protection provided by prescriptive rules and codes.
Enhanced fire protection for large, tall and complex buildings, can be
economically attractive if, in addition to cost of fire protection, monetary values
of the cost per life saved property/business loss are also considered in a cost–
benefit analysis. From among alternative fire protection strategies identified by a
performance-based code, a building owner may select a strategy which is the most
cost-effective in terms of costs and benefits of fire protection measures considered
and interactions and synergies between these measures (Chapter 12). The owner
would also consider the costs and benefits due to insurance/self-insurance options.
A detailed framework for carrying out a cost-benefit analysis as described above
has been discussed in Ramachandran’s book The Economics of Fire Protection
(1998). Quantitative risk assessment is an integral part of this framework.
As discussed above, the four major stakeholders or decision makers in the fire
safety field are property owners, architects and designers, consultant firms engaged
in fire safety engineering and government departments and organisations involved
in the development and enforcement of fire safety regulations, codes and standards.
Public fire and rescue services constitute the fifth major stakeholder. These bodies
have to provide adequate fire cover to properties in their geographical areas by
providing a sufficient number of strategically located fire stations, with enough
firefighters and other resources. This problem should necessarily consider the fire
risk in the properties in each area and the effectiveness of fire-protection measures
and fire brigade performance in reducing the risk – see Chapter 10. Taking into
account the interactions and synergies (Chapter 12) with fire protection measures,
a fire and rescue service can identify an economically optimum fire cover strategy
for any area.
The fire insurance industry is the fifth major stakeholder that has to estimate
appropriate premiums that can be charged for different types of properties. If
an insurance firm underestimates the premiums to be collected, it might face
bankruptcy in a market-driven economic environment involving keen competition
with other insurance firms. In the national interest the firm should promote fire
safety by offering sufficient rebates in premiums for fire protection systems and
self-insurance deductibles. Traditionally, most of the fire insurance underwriters
adopt semi-quantitative points schemes discussed in Chapter 2, Section 2.2.2.
The underwriters should validate and improve their premium calculation
methods by applying statistical models described in Chapter 3, section 3.3 and in
Chapter 11 of Ramachandran’s book The Economics of Fire Protection (1998). The
statistical models would provide more accurate estimates of the ‘risk premium’ and
the ‘safety loading’ to be added to this premium. An insurer firm can add another
loading towards expenses and profits to estimate the total premium to be charged.
The authors hope this book will provide most of the methods and tools for
quantitative risk assessment needed by stakeholders in the fire safety field.
Acknowledgements
Ganapathy Ramachandran (‘Ram’) would like to thank Liz Tattersall for word
processing his Chapters 5, 6, 7, 8, 9, 10 and 12, and sections in Chapters 2, 3,
4 and 13. Liz coped admirably well with all the statistical and mathematical
functions and formulae in these chapters and sections. Ram’s wife, Radha, gave
him considerable assistance in checking the typescript of the manuscript and
printed page proofs for spotting mistakes to be corrected.
David Charters would like to thank Dr Roth Phylaktou for the permission
to incorporate content from the MSc Module on Fire Risk Assessment and
Management at the University of Leeds. He would also like to thank Dominic
Vallely and James Holland (formerly and currently respectively) of Network
Rail, Paul Scott, Fermi Ltd and Matthew Salisbury, MSA for their support in the
development of the book. David would like to thank BRE Global and all the other
people who gave permission for material in this book. He would also like to thank
his sons, Jack and Theo for their encouragement and support.
The authors acknowledge the following permissions to reproduce material in this

book.
Figures 1.2 and 11.1 are reproduced from D. A. Charters, ‘A review of fire
risk assessment methods’, Interflam 04, reproduced with permission of Interscience
Communications.
Figures 1.3, 4.1, 2.5 and Tables 2.1, 2.2, 2.3, 2.4, 2.5, 2.6 and 2.7 are reproduced
with permission of Agetro Ltd
Figures 2.1, 2.2, 3.24, 3.25, 3.32 and 3.33 are reproduced with permission of
BRE Global Ltd.
Table A.2.1. is reproduced from E. C. Wessels (1980) ‘Rating techniques for
risk assessment’, Fire International, 67, 80–89. with permission from Keyways
Publishing Ltd.
Tables A.2.2. and A.2.3. are reproduced from P. Stollard (1984), ‘The
development of a points scheme to assess fire safety in hospitals’, Fire Safety
Journal, 7, 2, 145–153 with permission of Elsevier.
xiv Acknowledgements
Figures 3.1, 3.2 and 3.3 are reproduced from D. A Charters, Fire Risk
Assessment and Management MSc module reproduced with permission of the
Energy and Resources Research Institute, University of Leeds.
Figures 3.5, 3.6, 3.8, 3.9, Tables 3.7 and 4.2 are reproduced from D. A Charters
and G. Ramachandran, committee draft of BSI PD 7974 ‘Application of fire
safety engineering principles to the design of buildings’, Part 7 ‘Probabilistic risk
assessment’. Permission to reproduce extracts from PD 7974 ‘Application of fire
safety engineering principles to the design of buildings’ is granted by BSI. British
Standards can be obtained in PDF or hard copy formats from the BSI online shop:
www.bsigroup.com/Shop or by contacting BSI Customer Services for hardcopies
only: Tel: +44 (0)20 8996 9001, Email: [email protected].
Figure 3.7 and Table 4.4 are reproduced from D. A. Charters ‘Fire safety
assessment in bus transportation’, Fire Safety in Transport, reproduced with
permission of the Institution of Mechanical Engineers.
Figures 3.10 to 3.19, 5.2, 12.1 and 12.2 are repoduced from D.J .Rasbash, G.
Ramachandran, B. Kandola, J. M. Watts and M. Law (2004) Evaluation of Fire
Safety with permission of John Wiley and Sons Ltd (Figures 3.10, 3.11, 12.1 and
12.2 were previously published in G. Ramachandran (1998) The Economics of Fire
Proctection published by Spon)
Figures 3.20 and 3.21 and Tables 3.14 and 3.15 are reproduced from W. T. C
Ling and R. B. Williamson (1985) ‘The modeling of fire spread through probabilistic
networks’, Fire Safety Journal, 9, 287–300 with permission of Elsevier, and from
D.J. Rasbash, G. Ramachandran, B. Kandola, J. M. Watts and M. Law (2004)
Evaluation of Fire Safety with permission of John Wiley and Sons Ltd.
Figure 3.28, 3.29, 3.30 and 3.31 are reproduced from D.A. Charters, ‘Control
volume modelling of tunnel fires’ in A. Beard and R. Carvel (eds) The Handbook
of Tunnel Fire Safety, reproduced with permission of Thomas Telford Publishing.
Figures 4.2, 4.3, 4.4 and 4.5 are reproduced from D.J. Rasbash (1984), ‘Criteria
for acceptability for use with quantitative approaches to fire safety’, Fire Safety
Journal, 8, 141–158 with permission of Elsevier.
Figure 7.1 is reproduced from BS5950, Part 8, 1990, Section 4 with permission
of British Standards Institution.
Table 11.1 is reproduced from D. A. Charters, M. Salisbury and S. Wu, ‘The
development of risk-informed performance based codes’, 5th International
Conference on Performance Based Codes and Design Methods, reproduced with
permission of The Society of Fire Protection Engineers.
Figure 11.2 is reproduced from D. A. Charters, ‘Quantitative fire risk assessment
in the design of major multi-occupancy buildings’, Interflam 01, reproduced with
permission of Interscience Communications.
Every effort has been made to seek permission to reproduce copyright material
before the book went to press. If any proper acknowledgement has not been made,
we would invite copyright holders to inform us of the oversight.
1 Introduction
We all take risks all the time, whether it is crossing the road, driving to work or
watching television. The risk may vary from being knocked down, to being in a
car accident or suffering ill health due to lack of exercise. The same can be said
of fire safety in buildings. As long as we occupy buildings where there is a chance
that ignition sources and combustible materials may be present together, there will
be a risk of death and injury due to fire in addition to property damage. We need
not be fatalistic, however, this simply identifies the need to manage the risk. It also
indicates that, although we should work towards reducing risk, the ultimate goal
of zero risk is not currently a realistic expectation.
As Benjamin Franklin once said, ‘But in this world nothing is certain, but death
and taxes’. It follows that whilst we live there is a risk of death and the only way
of not dying is not to live in the first place. This may be s elf-evident, but it is very
important when we start to consider specific risks that we consider them in the
context of other risks.
There is a practical benefit to looking at risks in context. Society may decide
that it would like to dedicate more resources to addressing one risk than another.
For example, for healthcare, it may typically cost about £20,000 to save a life,
whereas for fire safety in buildings, it may cost more than say £1million to save a
life (Charters 1996). Therefore, society (and/or its representatives) could decide
to put more resources into healthcare than into fire safety in buildings. Equally,
society may be more concerned about the suffering of people killed and injured
by multi-fatality fires than it is about the provision of every possible healthcare
intervention to all patients, irrespective of need or prognosis.
For fire safety in buildings, the annual fire statistics indicate that there is a finite
level of fire risk in buildings (Office of the Deputy Prime Minister 2005). This may
also indicate that if a building complies with the appropriate fire safety standards,
then its level of fire risk is broadly tolerable (or possibly acceptable). It could also
be said that applying the fire standards to a non-standard building could result in
intolerable levels of fire risk. However, no criteria for fire risk in buildings have
been set in the UK (British Standards Institute PD 7974 Part 7 2003).
For the fire safety engineering of a non-standard building, this means that the
level of risk should be designed to be the same or lower than that for an equivalent
standard building (Office of the Deputy Prime Minister 2005; British Standards
2 Quantitative Risk Assessment in Fire Safety
Institute BS 7974 Code of Practice 2001). Since there are no quantitative risk
criteria, this means comparison of a non-standard building with a compliant
building. Similarly, since regulations are not generally framed in terms of risk,
this results in an assessment of physical hazards and the balancing of a qualitative
arguments about risk. So, in conclusion, we could say that, with respect to fire
safety in buildings, we are all taking a risk but rarely, if ever, is it calculated.
The risk of an undesirable event can be defined as the combination of:
• the severity of its outcome (its consequences); and

• how often it might happen (its frequency) (British Standards Institute PD
7974 Part 7 2003).
If we simply consider the potential consequences of fires, then we may not

be adequately addressing fires with lower consequences but whose risk is much
higher due to their frequency. It is easy to imagine very severe fire events in
buildings that have no fire precautions. We can identify ignition sources and
combustible material, estimate how quickly a fire would grow and how quickly
untenable conditions would develop. We can then consider the potential number
of occupants, their likely behaviour and how quickly they might escape from the
building. It is less easy to imagine very severe events in buildings with many fire
precautions, but they can, and do, occur. Severe fire events like these seem to
catch us by surprise and are characterised by a series of failures of the many fire
precautions. The lower frequency of these events leads to the sense of surprise.
The fact that a series of failures had to occur before the severe event occurred
indicates that we had ‘defence in depth’. Defence in depth is where many systems
are present, but only a few need to work to achieve a safe outcome. Finally, this
indicates that all fire precautions have a level of unreliability and occasionally
many of the fire precautions may fail at the same time, leading to the severe event.
Fortunately, these very severe events are relatively rare, but this rarity means that
it is very difficult to assess their frequency directly. For example, if we have a
thousand buildings of a certain type and we want to address fires that may occur
once every million building years, then on average this fire could be expected only
to happen once every thousand years. This is a long period over which to collect
and analyse data and make a decision about the adequacy of the fire precautions.
What makes it even harder is that this ‘one in a million building years’ event
could happen tomorrow and the day after. This makes the direct estimation of
high consequence/low frequency events very uncertain. Therefore, we use events
that happen more often to estimate the frequency of rarer events. The frequency
of ignition is the usual starting point for fire safety in buildings. To estimate the
frequency of the severe fire events we need to understand how ignition may, or may
not, lead to them. This involves the study of the reliability of fire safety systems.
For example, probabilities can be attributed to the following series of events:
• Does the fire grow?

• If yes, is the fire detected early?
Introduction 3
• If yes, is the fire extinguished using first aid fire fighting?
• Is the fire suppressed and/or vented by automatic systems?
• Does the fire spread beyond the compartment of fire origin?
In this way we can assess the impact of fire retardant or non-combustible

materials, fire detection, extinguishers, sprinklers, vents, compartmentation etc
on reducing the frequency of severe fires if, say, all these fire precautions fail.
We can combine the frequencies and consequences of these severe events to
estimate their levels of risk. The levels of risk can be expressed as a frequency
distribution (or ‘Fn’ curve) of risks, with lower risk events near the origin and
higher risk events to the upper right-hand side. See Section 1.4 on acceptance
criteria for risk and the role of Fn curves. Some events may have trivial or no
consequences and so can be discounted from the analysis. Other events are high
risk and should be addressed.
Because there are no quantitative criteria for the fire risk in buildings in the UK,
it is not possible to say in absolute terms when the risk is low enough. However, if
we know where a compliant standard building lies on the distribution, it is possible
to do it by comparison. So we can say that we are safe enough, when the risk is
lower, than a compliant standard building.
1.1 Fire engineering

Fire engineering can be defined as (BS 7974 2001):
the application of scientific and engineering principles to the protection of

people, property and the environment from fire.
Fire engineering can address one or more objectives. These objectives can
include:
• life safety;
• property loss prevention/business continuity; and
• environmental protection.
Fire engineering is generally more demanding, technically and in terms of

resources, than the application of simple prescriptive fire safety guidance used
to support building regulations (ADB 2007). Therefore, fire engineering is
generally used where simple fire safety guidance may not adequately address
the fire scenarios or issues of concern. Simple fire safety guidance may not
adequately address the fire scenarios or issues of concern when the building
is large, complex or unique, or when application of the simple prescription
conflicts with the function of the building (usually rendering the fire precaution
highly unreliable) or when application of the simple rules is not the most cost-
effective approach. Other factors that might indicate where fire engineering is
typically used include:
• when there is an atrium;
• when there are multiple purpose groups in one building;
• where a highly innovative design is used to facilitate the function of the
building; or
• when there are unique or challenging fire hazards.
Examples where fire engineering is typically used include the larger assembly
buildings, hotels, hospitals, industrial and commercial premises, transport
interchanges and tunnels, landmark, heritage and headquarters buildings, ships
and offshore installations.
1.2 Deterministic approaches

Deterministic approaches use quantitative analysis of physical processes, such
as smoke movement and evacuation, to aid decision-making during the design
process (BS 7974 2001). To understand the deterministic approaches, a more
detailed consideration of the design process may be of benefit.
Design can be characterised as an essentially creative, largely intuitive, often
divergent process where many design parameters are manipulated in three and
sometimes four dimensions, to meet multiple design objectives. Design objectives
may include function, efficiency, cost, buildability, safety, durability, reliability,
aesthetics etc. Design parameters may include the location of the site, the size
and interconnection of different spaces, means of access and egress, the number
of levels, the type of construction etc. For fire safety design, the main design
objective is life safety (occupants, fire fighters and others in the vicinity), but other
objectives such as property protection/business continuity and the environment
may also be included. In many respects, the use of simple prescriptive guidance
suppresses the design aspect of fire safety and can encourage the perception that
there is only one solution and ‘this is it’. However, the nature of many modern
buildings means that simple fire safety rules are increasingly difficult and costly to
apply. In these circumstances alternative fire safety design solutions are required
to ensure that an appropriate level of safety is achieved. This is where analysis
is used to assess the level of safety and indicate whether the fire safety design
objective(s) have been met. Analysis can be characterised as an essentially logical,
structured, rigorous process which takes certain input parameters, undertakes a
series of operations/calculations and produces one or more output parameters. For
fire safety analysis, the input parameters could include the material contents and
their burning characteristics, size of a space, the number, width and distribution
of its exits, the number of occupants, a walking speed and a flow rate through the
exit. Deterministic analysis can be defined as the use of point values for the above
variables and a purely physical model. Through the use of deterministic egress
calculation methods a single value of the output parameter of time for occupants
to move through an exit can be calculated (see Section 3.6 on consequence
analysis). This single value is regarded as an exact value, ignoring the uncertainties
governing the input and output parameters.
Introduction 5
Decision-making can be characterised as an essentially convergent process
of identifying an issue that needs to be addressed, gathering information on it
(including that from analysis) and reaching a conclusion based on the evidence.
For fire safety, the issue could be the number and width of exits required for egress
from a large and complex space. Based on knowledge of the number and width of
exits for smaller/simpler spaces, the basis of the input data, analytical model and
results of the analysis, a decision on the appropriate number of exits for a large and
complex space can be made (Charters 2000).
Quantified fire risk assessment may best be used where deterministic fire safety
engineering may not adequately address the fire scenarios of concern. This tends
to occur when the (life safety, loss prevention and/or environmental protection)
consequences of a fire may be very high/intolerable.
1.3 Probabilistic approaches

Fire risk assessment allows the performance and reliability of fire precautions to be
explicitly taken into account in the fire safety engineering of a building (PD 7974
Part 7 2003). It can be used to:
• select fire scenarios for deterministic analysis; and/or

• quantify levels of fire risk (to life, property, business and/or the environment).
In the selection of fire scenarios, an event tree for a broad class of occupancies
and first order data estimates of frequencies and consequences may be used to
identify and define an appropriate fire scenario for deterministic fire engineering
analysis. It is important that the scenario selected provides a reasonably severe
challenge to the fire safety design, yet is reasonably credible in terms of its
frequency. Typically, this may include the failure of fire prevention, reaction to
fire of materials, natural fire breaks, first aid fire fighting and non-fire resisting
construction. Additional scenarios can be analysed to assess the dependence on a
particular fire safety system such as sprinklers, smoke control or compartmentation.
Fire risk assessment can also be useful in quantifying the levels of risk for an
individual building design:
• in demonstrating equivalent life safety to a code-compliant building or

satisfaction of fire risk criteria;
• for cost-benefit analysis of property protection/business continuity; and/or
• in assessing the environmental impact of large fires.
Therefore, circumstances where fire risk assessment could be useful include

where:
1 Life safety is affected by buildings:

• containing a very large number of people who are unfamiliar with the
building;
• containing a large number of sleeping occupants;
• containing a large number of vulnerable people;
• with very high fire growth rates;
• with restricted means of escape; and/or
• with other unusual features which could be significantly detrimental to
fire safety.
Examples where fire risk assessment is useful for life safety in buildings and
structures may include larger multi-occupancy buildings, assembly buildings,
hotels, hospitals, industrial and commercial premises, tunnels, offshore
installations and ships.
2 Property protection/business continuity is affected by buildings:

• with a high value;
• with high value operations; and/or
• with high value public images.
Premises where fire risk assessment is useful for property protection/business
continuity may include certain landmark buildings, headquarters, leisure,
assembly and heritage buildings as well as some industrial, commercial and
transport premises (Charters and McGrail 2002).
3 Environmental protection is affected by buildings with a high potential impact

on the environment.
Premises where fire risk assessment is useful in addressing risks to the
environment from fire include industrial and transport premises, particularly
where hazardous goods may be present in significant quantities (Charters and
McGrail 2002). Transportation of dangerous and hazardous goods by road, rail
and other means can cause significant environmental damage.
Fire risk assessment is essential where deterministic fire safety engineering cannot
adequately address the fire scenarios of concern. This tends to occur when
the consequences in terms of life safety, loss prevention and/or environmental
protection of a fire may be intolerable.
Factors that might indicate where fire risk assessment would be essential
include:
• When the reliability and performance of protection systems is critical. Fire

risk assessment should be used to assess the ‘defence in depth’ of a design that
relies heavily on a single or limited number of fire safety systems.
• When the variability of input parameters has a significant impact on the
results. Fire risk assessment should be used where there are significant
variations in variables like the number of people, their characteristics, fire
growth rates etc and deterministic analysis shows that credible combinations
of the variables are not acceptably safe (Charters and McGrail 2002).
Introduction 7
• When a wide range of fire scenarios is deemed to be necessary. Fire risk
assessment should be used in complex, high value, mission critical buildings,
where high levels of defence in depth (multiple fire safety systems) are
required.
1.4 Background to the development of fire risk assessment

Historically, fire has always been one of the major threats to life. In earliest times,
the fires occurred in the natural environment of forests and grasslands. More
recently, fires involving combustible materials in the built environment have
caused the greatest risk to life.
In early civilisations, the buildings were generally small and single storey and so
escape to a place of safety for their occupants was usually quite easy. The property
lost could then be replaced relatively quickly due to their simple construction. As
civilisations became more organised and sophisticated, settlements became larger.
Buildings began to have upper storeys and fires in kitchens and living rooms and
later candles and oil lamps added to the risks of fire. Throughout history disasters
like the Great Fire of London in 1666 have struck such communities causing
massive loss of property and life.
The response to various fire disasters has been to learn specific lessons. Over
time this has resulted in the development of building regulations, considerations of
compartment size, awareness of the need for a coherent package of fire precautions,
the importance of the reaction to fire of building contents and human behaviour
in fire. However, the more fire precautions that were placed in buildings did not
seem to guarantee safety and fire disasters continued to occur even in apparently
safe buildings. Equally, it could be argued that fire risk in buildings could have
increased in the absence of fire regulations, codes and standards.
The nuclear industry was the first to learn this lesson in the 1950s with the
fire at the Windscale reactor. For other building types in the UK, it was the King’s
Cross underground fire that raised the question of why an apparently safe building
type (there had been few life loss fires on the underground) could suddenly suffer
such a severe fire. The answer of course was that some of the fire precautions
present work in most fires and so they result in a safe outcome (or a near miss).
However, very occasionally, all the measures present will fail in such a way as
to lead to a fire disaster. Therefore, the way of dealing with the stochastic (or
random) nature of fire is through probabilistic fire risk assessment.
In spite of the many fire precautions in modern buildings, fire statistics still
show the extent of the challenge. Fires in a range of countries result in the loss of
(World Fire Statistics Centre 2007):
• 0.5 to 2 people per 100,000 of population per year; and

• 0.5 to 1.5 per cent of gross domestic product per year.
1.4.1 Corporate governance
Corporate governance is the way in which companies are directed and controlled
for the benefit of internal and external stakeholders in a business. It is, therefore,
the mechanism through which management is held accountable for performance
and through which stakeholders are able to monitor and intervene in the
operations of management.
The most widely accepted principles of corporate governance are provided
by the Organisation for Economic Cooperation and Development (OECD) and
include:
• protect stakeholders rights;

• ensure equitable treatment of stakeholders;
• recognise the rights of stakeholders in law and encourage cooperation
between businesses and stakeholders to create wealth, jobs and sustainable
enterprises;
• ensure timely and accurate disclosure of material matters regarding the
business;
• ensure strategic guidance of the company through effective monitoring of
management for the business and the stakeholders.
To further consider the requirement for fire safety, it is worth identifying the
many and varied groups of people who have an expectation of fire safety. These
groups of stakeholders can be categorised as:
• staff
• public
• customers
• board of directors
• regulators, and
• other external stakeholders.
Staff expect a safe working environment. This expectation may be balanced

by the desire for job security. In other words staff have a high interest in fire
safety, but not at a ridiculously high cost. An example of a workforce whose safety
related expectations had not been met is that of the offshore oil industry. There
were many instances of safety related industrial action following the Piper Alpha
disaster in 1988.
The public expect safety for themselves whenever they interact in any way
with a business. They also expect others to be safe. The public has a strong ‘social’
awareness of safety and in particular fires affecting a large number of people
(known as ‘consequence aversion’). An example of the failure to ensure public
safety is the King’s Cross underground station fire in 1987.
Customers expect to be safe whilst on the business’s premises. Here it is not the
actual level of safety that is important; it is the perception of safety that matters.
Introduction 9
Hence, road tunnels may be statistically safer than the open road, but public
knowledge of this fact does not affect public fear of tunnel fires following the
multiple fatality fires at Mont Blanc, Tauern and Kaprun in the late 1990s.
Boards have expectation of safety, both as stakeholders and as individuals who
are ultimately responsible and liable if things do go wrong. The Health and Safety
at Work Act 1974 effectively establishes board members as negligent if risks are
not as low as reasonably practicable (ALARP). Criminal law is increasingly being
directed at board members to establish corporate liability. Chief Executives, such
as that at London Underground Ltd at the time of the King’s Cross fire, are being
forced to resign.
Regulators try to balance all the different requirements of the various groups
of society. The regulators recognise the national interest in a strong economy.
The balance between ever-improving safety and reasonable social/economic cost
is enshrined in the ALARP principle/legal precedent where risks can be said to
be ALARP when the cost of further risk reduction far outweighs the reduction in
the risk.
1.4.2 Risk management

Early interpretation of these principles concentrated on financial aspects, however,
the Combined Code (1998) and the Turnbull report (1999) widened the emphasis
on internal controls to business-wide risk management activity. Not surprisingly,
risk management incorporates all the usual features of a management system:
• establish a strategic policy (individual policies such as fire safety may be

required to support this);
• put in place an organisational structure, resources, competencies and
responsibilities to deliver the policy;
• plan and implement risk assessments leading to the implementation of control
measures to reduce risk;
• monitor the use and effectiveness of the control measures and investigate
incidents;
• audit and review the system to confirm that it continues to address risks and
identify opportunities for improvement.
These risks may cover a range of areas including health and safety, finance,
quality, environment, security, IT, resources, reputation, regulations and business
continuity. Fire may impact on all of these areas and the key is to identify, analyse
and control the risks from fire.
A hazard can be defined as something with the potential to cause harm,
for example a pile of combustible waste is a fire hazard. Risk can be defined as
the combination of the frequency of an unwanted event, such as a fire and its
consequences, i.e. how often it might occur and how bad the outcome might be.
Once the risks have been assessed, a hierarchy of risk reduction measures can
be applied:
1 Elimination: remove the hazard;
2 Prevention: reduce the likelihood of the event occurring;
3 Mitigation: reduce the severity of the event;
4 Control of consequences: emergency and contingency planning and crisis
management.
The remaining business risk can then either be:
1 Transferred: by means such as insurance, or

2 Absorbed: with a budget allocation to cover the expected level of loss.
1.4.3 Legislation
One of the main drivers for fire risk assessment is legislation specifically relating to
fire safety. This varies from country to country but in many countries it typically
includes:
• building regulations;
• fire precautions in existing buildings (normally workplaces); and
• other safety legislation.
1.4.3.1 Building regulations

The objectives of building regulations vary tremendously from one country to the
next. Typical fire safety objectives could include:
1 to ensure satisfactory provision of means of giving an alarm of fire and a

satisfactory standard of means of escape for persons in the event of a fire in
a building;
2 that fire spread of the internal linings of buildings is inhibited;
3 to ensure the stability of buildings in the event of fire; to ensure that there is
a sufficient degree of fire separation within buildings and between adjoining
buildings; and to inhibit the unseen spread of fire and smoke in concealed
spaces in buildings;
4 that external walls and roofs have adequate resistance to the spread of fire
over the external envelope, and that spread of fire from one building to
another is restricted;
5 to ensure satisfactory access for fire appliances to buildings and the provision
of facilities in buildings to assist fire fighters in the saving of life of people in
and around buildings.
Guidance on how to comply with building regulations can be found in a variety

of guides. These guides are intended for the some of the more common building
situations. However, many recognise that there are alternative ways of complying
with the requirements of building regulations. The means for developing
Introduction 11
alternative design solutions is commonly known as fire safety engineering. Some
guides indicate that a fire engineering approach may be the only practical way to
achieving a satisfactory standard of fire safety in some large and complex buildings.
Further they may indicate that the factors that should be taken into account
include:
• the anticipated probability of a fire occurring; and

• the anticipated fire severity.
These factors clearly imply the use of fire risk assessment.

Furthermore, alternative solutions are accepted on the basis of equivalency,
within whose definition is often an explicit reference to the level of fire risk. For
equivalency, there is a need to:
…demonstrate that a building, as designed, presents no greater risk to

occupants than a similar type of building designed in accordance with well-
established codes.
Therefore, fire risk and fire risk assessment can play a major role in the design
of new buildings. Further information on fire engineering can be found in many
guides, for example PD7974 (2001) contains guidance on the application of
probabilistic fire risk assessment.
1.4.3.2 Fire regulations for existing buildings

The fire regulations for existing buildings are many and varied. Some of the more
recent examples include requirements for:
• a fire risk assessment to be undertaken;

• identification and recording of the significant details of those at particular
risk;
• provision and maintenance of fire precautions to safeguard those in the
workplace;
• provision of information, instruction and training to employees about fire
precautions in the workplace.
With many relatively low risk buildings to assess, the fire risk assessment process
used to support this kind of legislation relies.
1.4.3.3 Other safety legislation

Other safety legislation is based on the development and approval of safety cases.
One of the main pieces of evidence in support of a safety case is risk assessment.
Some of the legislation that might include a fire risk assessment or a risk assessment
that includes fire hazards concerns:
• rail transport;
• major chemical sites;
• offshore oil and gas installations; and
• nuclear power installations.
1.4.4 Fire safety management
1.4.4.1 Fire safety policy

The fire safety policy for an organisation is a document that lays out the
commitment of the organisation to fire safety on behalf of its stakeholders. Fire
safety policies may contain references to supporting documentation such as
legislation, procedures and risk reduction plans. Policy statements are usually
endorsed by the Chief Executive of an organisation, to demonstrate commitment
from the highest level.
1.4.4.2 Management structure

To support the delivery of the policy, there is need for a fire safety management
structure. This structure will range from the Chief Executive and an Executive
Director responsible for fire safety to all members of the organisation.
Other roles may include a senior manager who may be responsible for:
• supervising the day-to-day upkeep of the policy;

• ensuring staff participate in fire safety training;
• carrying out fire drills;
• coordinating and directing staff action in accordance with an emergency plan;
• receiving reports of fires;
• preparing reports to the board of directors on fire safety.
Members of organisations also have general duties and responsibilities and

these include:
• understanding the nature of fire hazards;

• awareness of fire hazards in the building;
• practising and promoting fire prevention;
• knowing and following procedures for when a fire breaks out.
There may also be a role for a specialist fire safety advisor. This role may include:
• assisting management in the interpretation and application of legislation and

guidance;
• advising management of their responsibilities;
• undertaking fire safety audits;
• undertaking fire risk assessments;
Introduction 13
• liaising with staff, designers and approval authorities when specifying fire
precautions;
• training staff in fire safety.
Specialist fire safety advisors may be internal or external to an organisation.

If however, they are internal, their position in the management structure should
be adjacent to the normal management lines, so that they are independent of
operations managers who are responsible for fire safety in their area.
1.4.4.3 Audit and risk assessment

Lord Kelvin said ‘you cannot predict what you cannot measure’. Equally, what
cannot be measured or predicted cannot be managed. Therefore, the two crucial
measurement processes in the feedback loop of the risk management process are
audit and risk assessment.
Audit is a structured interview process which includes a range of questions and
seeks evidence of compliance. It was first developed in the financial industry for
verifying accounts, but is equally applicable to the assessment of any management
process such as risk and quality assurance.
Risk assessment on the other hand tends to address the physical situation and
seek to measure/predict and assess the acceptability of risk in a particular place.
Over time a range of fire risk assessment methods have been developed to address
risk in a range of applications. These methods range from simple qualitative fire
risk assessment methods for the Fire Safety Order to the complex quantitative
and probabilistic methods applied in the nuclear industry. Figure 1.1 shows a
breakdown of fire risk assessment methods and terms.
The following subsections outline each of these types of fire risk analysis.
1.4.5 Qualitative fire risk assessment methods

Qualitative methods rely on identification of factors that affect risk. These factors
may be those that affect the level of hazard, such as combustibles or the level
of fire precautions intended to mitigate those hazards such as fire detection. An
assessment of all these factors is undertaken against a benchmark set of values
Fire risk assessment
Qualitative methods Semi-quantitative methods Quantitative methods
Structured Unstructured Points schemes Matrix Probabilistic Full QRA
Figure 1.1 Breakdown of different fire risk assessment methods

and a judgement made as to whether the factor is higher, lower or about the
same as the benchmark. A review is then undertaken of all the factors to judge
how overall the area being assessed compares with the benchmarks. If additional
mitigation measures are judged to be required, this is a user input or a choice
from a prescribed list. Assessments under the fire regulations for existing buildings
usually fall into this category.
1.4.5.1 Unstructured methods

An unstructured method of qualitative fire risk assessment might follow the five-
step risk assessment process outlined below:
1 identify potential fire hazards;

2 identify those in danger from fire;
3 evaluate the risks and whether the current fire precautions are adequate;
4 record findings and actions and communicate to employees;
5 keep the assessment under review and revise it when necessary.
The method is unstructured as it has no prescribed standards for what types

and levels of precautions are appropriate for various levels of hazard. An example
of an unstructured fire risk assessment can be seen in Section 2.1.1.
1.4.5.2 Structured methods

Structured methods of qualitative fire risk assessment use the same process as
unstructured methods except that in addition they tend to prescribe:
• different levels of hazard;

• different levels of precaution;
• the acceptability of different combinations of hazard and precaution;
• appropriate risk reduction alternatives.
A good example of this approach is that of FIRECODE HTM 86 (1994) ‘Fire risk
assessment in hospitals’. This method identifies different types of fire hazard such
as ‘smoking’ and ‘fire hazard rooms’ and whether the level is ‘acceptable’, ‘high’ or
‘very high’. It then goes on to identify different types of fire precaution including
‘alarm and detection’ and ‘compartmentation’ and whether the level of provision is
‘unacceptable’, ‘inadequate’, ‘acceptable’ or ‘high’. Certain combinations of hazard
ratings and precautions ratings are deemed to be acceptable. Other combinations
lead to the need to reduce risk by a range of prescribed alternatives.
This kind of qualitative flexible prescription identifies areas of high hazard and/
or inadequate precautions and prescribes one or more prescriptive solutions. This
is relatively easy and consistent to apply, taking only a matter of hours to survey
and days to report. This method does not suffer from the apparent acceptability
of inappropriate fire strategies. However, the method cannot easily address non-
Introduction 15
standard situations, limits options for upgrade and may not be applicable to new
designs.
1.4.6 Semi-quantitative fire risk assessment methods

Semi-quantitative fire risk assessment methods place some numerical value on
the level of risk, without that numerical value representing a precise value of risk.
1.4.6.1 Points schemes

Points schemes use similar factors to other qualitative methods, but inject some
numeracy to result in a score. Typically each aspect of fire safety (or factor) is
scored from a range and then multiplied by a weighting to provide an overall
score for that factor. The overall score for each factor is then added together to
provide an overall score for the area being assessed. This overall score can then be
compared with a benchmark score to assess the level of risk. Risk reduction can
be effected by improving any number of factors and recalculating the overall score
for the area.
The 1980s version of FIRECODE HTM 86 (1987) for fire risk assessment in
UK hospitals and the ADAC (Allgemeiner Deutscher Automobil-Club) European
fire risk assessment of tunnels fall into this category.
Points schemes have the advantages that they are relatively easy and quick to
use and provide a degree of resolution of the level of risk by producing a number.
Their disadvantages include the relatively arbitrary values given to the weightings,
the implicit fire safety strategy that they represent and the freedom to develop
high scoring fire safety solutions that do not work in practice.
1.4.6.2 Matrix methods

Matrix methods are used where the subjective nature and lack of quantitative
information from qualitative methods are likely to lead to poor decision making
with significant consequences. Matrix methods usually take the form of a
structured brainstorm in a workshop. The Fire HAZard ANalysis (HAZAN) is
often the first task in as asset’s fire risk assessment. The objectives of the HAZAN
workshop are to:
• identify fire hazards and accidents that could occur in the asset;
• estimate the consequences according to an accident severity matrix;
• rate the hazards by using a risk classification matrix;
• screen out minor fire hazards; and
• provide a list of identified major fire hazards.
The major fire hazards will then be subject to a Quantitative Risk Assessment
(QRA). An example of a matrix and the output tables can be seen in Section
2.2.1.
The advantages of matrix methods are that they place risks in broad categories
by breaking down difficult questions into smaller parts, the judgements can largely
be evidence based, they are auditable and mathematically robust and based on
consensus. The disadvantages are that the risk categories are broad, judgements
are often subjective and dependent on who actually attends the workshop and the
dynamic of the brainstorm process. This is why they are often used to prioritise
hazards and events as a precursor for full quantitative fire risk assessment.
1.4.7 Quantitative fire risk assessment

Quantitative risk assessment methods predict a discrete level of risk. They tend
to use a combination of probabilistic and/or physical modelling to predict the
frequency of certain events and their likely consequences. Most methods can be
characterised as probabilistic or full quantitative. This approach assumes that
event consequences from the asset being risk assessed will be like the others from
which the statistics are derived.
1.4.7.1 Probabilistic risk assessment

Probabilistic methods simply use statistical analysis to predict the frequency or
probability of an unwanted event. The consequences of the unwanted event
are taken directly from the statistical information, e.g. probability of fire spread
beyond the item first ignited or probability of area damaged being greater than x.
Probabilistic methods have the advantage that they have a sound theoretical
basis and provide a numerical prediction of risk. The disadvantages are that they
require a high degree of technical input, are reliant of fire statistics being available
and assume that the asset being assessed is the same as a homogeneous set from
which the fire statistics were taken.
1.4.7.2 Full quantitative risk assessment

Full quantitative fire risk assessment predicts the level of consequences from
unwanted events as well as their frequency. Figure 1.2 illustrates the process.
How often is a severe fire likely to occur?
One of the main challenges in assessing fire risk is the fact that the kinds of very
severe events that we are interested in are relatively infrequent. For example, the
King’s Cross fire happened after approximately 10,000 station-years of experience.
If severe fire events occurred all the time, their frequency would be easy to
predict. Thankfully, they are relatively rare. This rarity means that we are unable
to predict these extreme events explicitly from historical data. So to predict the
frequency of fire events that have not happened yet, we have to break the event
down from ignition to outcome into sub-events which occur frequently enough
for there to be meaningful data available. The means for constructing event and/
Introduction 17
Hazard
identification
Frequency Consequence
analysis analysis
Risk Risk
evaluation reduction
Risk NO
acceptable
YES
END
Figure 1.2 Schematic of the full quantitative fire risk assessment process
or fault trees and these were developed for probabilistic safety analysis in the
nuclear industry.
How severe is a fire likely to be?
How severe a fire event is likely to be can be evaluated in a number of ways.

The prediction of event consequences tends to rely on modelling which in many
respects is more mature and more widely used in processes like fire engineering.
For fire risk assessment, this modelling tends to involve predicting:
• how quickly a fire will grow;

• how the heat and smoke will move about;
• how quickly the occupants will be aware of, react to and evacuate;
• how tenable the environment is for occupants before evacuation is complete.
Full quantitative fire risk assessment methods have the advantage that they
have a sound theoretical basis and provide a numerical prediction of risk. The
disadvantages are that they require a high degree of technical input and are reliant
on fire statistics and appropriate models being available.
1.4.8 Other methods

A wide range of fire risk assessment methods are available including:
• checklists
• GOFA
• Delphi
• HAZID
• HAZOP.
These methods are described in the technical literature and share many of the
same characteristics as the methods described above, with the possible exception
of the Delphi method, described below.
Delphi
The Delphi method is a way of gaining a group opinion without the difficulties
inherent in meetings (Linstone and Turoff 1975). The group never meet and all
communication is via a group controller. The group controller selects the members
of the group, presents the basic problem and informs the individuals in the group of
progress. When the main elements of a problem have been described, a consensus
is sought on the value to be ascribed to each issue.
The consensus process is via a series of rounds of voting. After each round
the controller returns each panel member with their score and a measure of
the group score. The members are then asked if they wish to revise their score
until the controller judges that consensus has been reached or that consensus is
unattainable.
The advantages of the method are the lack of group pressure present in a
meeting when operating in an area with no data or a very difficult decision. The
disadvantages include extended time implications, greater dependency on the
group controller to avoid unconscious bias, empirical and subjective assessment
of risk and a range of other doubts about the quality of the outcome. For example,
recent work by Carvel et al. (2001) found that there were significant variations
between the estimates of a panel and actual experimental values.
1.5 Examples of the adoption of fire risk assessment
1.5.1 Introduction
This section describes several disasters that have shaped the provision of safety
from fire in different industries and situations. The nuclear and chemical industries
in particular, for example, consider their physical neighbours whilst dealing with
complexities which are beyond the understanding of the vast majority of the public
and even many of those industries’ employees. Prevalent in all the ‘disaster led’
revisions of safety provision below is the term ‘safety case’, a term whose semantics
express the move away from explaining ‘why something is not dangerous’ to a
documented appraisal of ‘why something is safe’.
These industries are not only dealing with the risk from fire, of course, as the
‘release’ catastrophes of Bhopal (1984) and Chernobyl (1986) illustrate only too
well. Indeed, the first recognisable quantitative risk assessment was undertaken
over 200 years ago. The concepts of insurance and domestic risk management
have a history almost as long as the concept of risk itself, but the real milestone
Introduction 19
came with the development of probability theory by Pascal in 1657. This seems to
have initiated a flurry of activity culminating in the first quantitative assessment
of risks by Laplace in 1792.
Laplace also helpfully described the application of probability theory as (1814):
…common sense with a little mathematics.
1.5.2 Nuclear installations
1.5.2.1 Windscale fire 1957

On 10 October 1957, a fire occurred at the Windscale nuclear power plant. The
fire spread to the plant’s air filters, there was a loss of containment and a release
of radioactive material into the local area which led to a ban on milk produced
locally.
The safety philosophy at this time was one of reaction and prevention. For
every incident, changes were made to the design and operation of plant so that
the recent event could not happen again. This resulted in more and more safety
systems being incorporated but still incidents (usually different in nature from the
previous one) would occur. To address this problem a different safety philosophy
was required – one that recognised that all systems can fail and that combinations
of initiating events and system performance on demand are important and can
lead to very different outcomes for events. This led to the use of methods for
predicting the probability of unwanted events that had been developed in the
communications, aviation and electronics industries after World War Two (Green
and Bourne 1972).
One of the key findings from the inquiry into the incident was that there
was need for ‘better communication between the plant’s management and
local interests’. This, in turn, ultimately led to a way of expressing society’s risk
expectations in the Farmer curve (later known as Fn curves, see Farmer 1967).
Implicitly, the best way of using the Farmer curve was to analyse the level of risk
from a plant or activity and compare it with the levels on the curve.
1.5.2.2 Farmer curve

The Farmer curve was the first quantitative expression of risks to society. The
graph plots the number of reactor years against the size of release of iodine 131 in
curies. The curve indicates that relatively small releases can be tolerated every 100
to 1,000 reactor years but that larger releases cannot be tolerated unless they are
every million to 10 million reactor years. This type of curve later became known as
Fn or frequency–consequence curves (Allen et al. 1992), see Figure 1.3.
Frequency data
Frequency (Events >N per year) 1.00E-01
1.00E-02 Intolerable
1.00E-03
1.00E-04
ALARP
1.00E-05
1.00E-06
1.00E-07 Negligible
1.00E-08
1 10 100 1000 10000
Fatalities (N)
Negligible Intolerable Detailed analysis
Figure 1.3 An F–n Curve
1.5.2.3 Safety cases and probabilistic risk assessments (PRAs)

The Nuclear Installations Act, 1965 (as amended) (HSE 2002) requires any
operator of a defined nuclear installation to be licensed and gives the HSE the
powers to ‘attach to the licence such conditions as may appear … to be necessary
or desirable in the interests of safety’. There are 36 standard Licence Conditions
attached to all Nuclear Site Licences of which Licence Condition 14 requires the
licensee to:
make and implement adequate arrangements for the production and

assessment of safety cases consisting of documentation to justify safety
during design, construction, manufacture, commissioning, operation and
decommissioning phases of the installation.
A safety case can be defined as the totality of documented information and

arguments which substantiate the safety of the plant, activity, operation or
modification in question. It provides a written demonstration that relevant
standards have been met and that risks have been reduced as low as reasonably
practicable (ALARP).
The safety case for a plant as a whole should be a living document which is
subject to review, change and amendment as time elapses. For example, the safety
cases may change due to important changes to the plant, its mode of operation,
or the understanding of safety-related issues. It may also change in the light of
operating experience.
The safety case for a nuclear plant (or modification etc) should be based upon
robust design, defence-in-depth and deterministic analysis of normal operation and
fault behaviour. The latter should consider faults that are reasonably foreseeable
during the lifetime of the plant and for which provisions have been designed into
the plant to prevent or mitigate them. This is known as design base accident
Introduction 21
(DBA) analysis. In addition, a deterministic analysis should be undertaken of more
severe faults and failures which have not been specifically protected against in the
design and which, in the extreme, could lead to large releases of radioactivity. This
constitutes severe accident analysis.
To supplement and support the deterministic analysis, a probabilistic safety
analysis (PSA) may be required. The requirement, depth and level of the PSA
should be commensurate with the significance of the hazard. The PSA will provide
the means to identify failure scenarios, confirm the effectiveness of defence-in-
depth provisions, search for weaknesses in the design, show there is reasonable
balance of risk for all hazards and operations, and derive numerical estimates of
risk. The PSA provides quantitative input to the ALARP case and it can provide
estimates of relative benefits (in terms of risk reduction) of improvements.
This shows how quantitative risk assessment (of all hazards including fire) is
intrinsic to, and underpins, much of the safety case regime for the licensing of
nuclear installations. The most common methods used are HAZOP (similar to
matrix methods), deterministic analysis (similar to consequence analysis) and
PSA (similar to full quantitative risk assessment).
The nuclear industry uses safety assessment principles (SAPs) to help judge the
acceptability of risk assessments. The SAPs contain some 330 principles, the vast
majority of which relate to engineering or operational good practices as well as
dose and risk criteria. The overall requirement is that risks should be reduced to
ALARP: the criteria are for the guidance of inspectors and are not design targets
for licensees. To support SAPs, more detailed internal guidance in the form of
Technical Assessment Guides has been produced, some of which are available on
the HSE website (HSE 2003).
1.5.3 Chemical plant and offshore installations
1.5.3.1 Flixborough, 1974

On Saturday 1 June 1974, the Flixborough works of Nypro (UK) Ltd were
practically demolished by a massive explosion, killing 28 and injuring 36 people.
The plant manufactured caprolactam, which is the basic ingredient of Nylon 6.
Originally, this was achieved by the hydrogenation of phenol but the process
was changed in 1972 and cyclohexane was oxidised to produce cyclohexanone,
bringing with it various new problems regarding safety. Any escape of cyclohexane,
which was circulated through reactors under pressure at a temperature of 155°C
could prove potentially dangerous.
Following the detection of a cyclohexane leakage in Reactor 5, a large crack was
discovered in the vessel. It was decided at a management meeting that the reactor
should be removed and inspected, and that a by-pass pipe should be installed
between Reactors 4 and 6 in order to allow the plant to continue to function. It
was inadequacies in the design and installation of this pipe that led to it rupturing
and releasing large quantities of cyclohexane vapour which subsequently exploded.
1.5.3.2 COMAH regulations
This section describes the Control of Major Accident Hazards Regulations
1999, which came into force on 1 April 1999 (HSE 1999). They implemented
the SEVESO II Directive and replaced the Control of Industrial Major Accident
Hazards (CIMAH) Regulations 1984.
COMAH applies mainly to the chemical industry, but also to some
storage activities, explosives and nuclear sites, and other industries where
threshold quantities of dangerous substances identified in the Regulations are kept
or used.
A major accident is defined as an occurrence (including in particular, a major
emission, fire or explosion) resulting from uncontrolled developments in the
course of the operation of any establishment and leading to serious danger to
human health or the environment, immediate or delayed, inside or outside an
establishment, and involving at least one dangerous substance.
The COMAH regulations have a regime that is similar in many ways to that of
the nuclear industry in that the Competent Authority (CA) sets down its policies,
procedures and guidance for the handling and assessment of safety reports. The
assessment involves the exercise of professional judgement by inspectors and
a framework within which these judgements are made. This is intended as a
practical tool for inspectors, to help achieve consistency in the approach to safety
report assessment.
A safety report has to contain certain information, which relates to the major
accident hazards and how major accidents are prevented or how the consequences
of such an accident are limited. The information provided has to be sufficient
to meet the purposes of the report and to help assessors gather this information
the CA has drawn up assessment criteria. There are about 130 criteria set out in
six groups. The predictive criteria deal with the identification of major accident
hazards and risk analysis. They cover:
• principles of risk assessment and the use of appropriate data;

• identification of major hazards and accident scenarios;
• likelihood of a particular major accident scenario or the conditions under
which they occur including initiating and event sequences; and
• consequence assessment.
The safety report as a whole should enable a view to be taken on the suitability
and sufficiency of the risk assessment for drawing soundly based conclusions. It
should be clear that the operator’s approach to demonstrating compliance with
the ‘all necessary measures’ requirement, is fit for purpose (HSE 2003).
1.5.3.3 Piper Alpha, 1988

At 21:45 on 6 July 1988, one of two condensate injection pumps tripped on the
Piper Alpha platform in the North Sea (Cullen 1990). Due to a failure in the
Introduction 23
Permit to Work system, the night shift attempted to restart the other pump which
had been shut down for maintenance. Unknown to them, a pressure safety valve
had been removed from the line and a blank flange assembly had been fitted that
was not leak-tight.
The initial explosion caused extensive damage. It led immediately to a large
crude oil fire in B Module, the oil separation module, which engulfed the north
end of the platform in dense black smoke. This fire which extended in C Module
and down to the 68ft level was fed by oil from the platform and by a leak in the
main line to the shore, to which pipelines from Claymore and Tartan platforms
were connected. The fire spread rapidly, unchecked by the firewater system which
had failed to operate partly because the fire pumps were on manual due to diving
operations.
At about 22:20 there was a second major explosion which caused a massive
intensification of the fire. This was due to the rupture of the riser on the gas
pipeline from Tartan as a result of the concentration and high temperature of
the crude oil fire. In total 226 persons were aboard the rig, most of these in the
accommodation part of the platform. Only 61 people survived the disaster.
1.5.3.4 The Offshore Installations (Safety Case) Regulations, 1992

Prior to 1988, the Department of Energy was responsible for both production of
oil and gas and for safety in the North Sea. This potential conflict of interest
was resolved by enforcing the above regulations through the Health and Safety
Executive (HSE).
The Offshore Installations (Safety Case) Regulations 1992 (HMSO 1992) were
in addition to and amended the Offshore Installations (Safety Representatives
and Safety Committee) Regulations 1989. The Regulations are applicable to
fixed and mobile installations, combined operations and abandonment of fixed
installations. Similar to the nuclear and COMAH regimes they require a safety
case that is reviewed, in this case by the Offshore Safety Division of the HSE.
The risk assessment methods are also similar to the nuclear installations, but
modelling pool and jet fires as hazards in their own right rather than simply as
mechanisms for the release of radioactivity or as a failure mode of safety critical
systems.
A key part of an offshore safety case is demonstrating that all hazards with the
potential to cause a major accident have been identified, their risk evaluated, and
that measures have been or will be taken to reduce the risks to people affected by
those hazards to the lowest level that is reasonably practicable (ALARP) (HSE
2003). Acceptable safety cases will demonstrate that a structured approach has
been taken which:
a. identifies all major accident hazards. The identification methods should

be appropriate to the magnitude of the hazards involved and a systematic
process should be used to identify where a combination or sequence of events
could lead to a major accident;
b. evaluates the risks from the identified major accident hazards. Any criteria
for eliminating less significant risks should be explained and in deciding
what is reasonably practicable, relevant good practice and sound engineering
principles should be taken into account. In addition, human factors need to
be accounted for and safety critical tasks should be analysed to determine the
demands on personnel;
c. describes how any quantified risk assessment has been used, and how
uncertainties have been taken into account;
d. identifies and describes the implementation of the risk reduction measures.
The reasoning for or against the choice of risk reduction measures to be
implemented should be clear.
1.5.4 Transport
1.5.4.1 King’s Cross fire, 1987
On Wednesday 18 November 1987, 31 people lost their lives when a fire occurred
in escalator 4 at the King’s Cross underground station (Fennell 1988). Although
there had been a smoking ban in place since 1985, ignition, probably due to smokers’
materials, occurred at about 19:25. The fire probably grew from the ignition of
detritus before involving grease and the plywood skirting board. Between 19:43
and 19:45, a modest escalator fire was transformed into a ‘flashover’ that erupted
in the ticket halls. A new mechanism, the ‘trench effect’ (where the flames laid
flat against the wooden escalator steps) led to the flaming and a large amount of
dense black smoke in the ticket hall that caused horrendous injuries and killed 31
people.
A combination of factors, including poor communications between members of
staff and London Fire Brigade, uncertainty about evacuating passengers, confusion
about who was in charge, and a failure to operate the water-fog equipment, meant
that the fire was able to take hold, eventually propagating violently up the escalator
through the ‘trench effect’.
The safety ethos of the time, as summed up by Lord Fennell (1988), was that:
• ‘there have been smoulderings … year in year out’;

• ‘there had been some escalator fires, … but no passenger had ever been
burned …’; and
• ‘a safe environment is not one in which there is an absence or low number of
serious injury incidents …’.
With the benefit of hindsight these statements clearly did not adequately
describe the nature of fire safety at that time. The dilemma for the operator
was that a significant body of historical experience indicated that such an event
was highly unlikely, but on the other hand the potential ignition of combustible
materials mean that it is difficult to call this kind of event incredible.
Introduction 25
1.5.4.2 The Railways (Safety Case) Regulations 2000
Prior to the King’s Cross fire, the underground system in London was regulated by
the Railway Inspectorate. Company standards were in place and the Inspectorate
would inspect and audit these standards and investigate any significant incidents,
e.g. derailment.
From a fire perspective, London Underground Ltd had a very good record with
relatively few large fires, injuries or fatalities over its 100-year or so history. There
were, however, some occasional but large escalator fires and a large number of
small fires, usually ignition of rubbish by the third rail arcing.
One of the initial regulatory responses to the King’s Cross fire was to extend
certification under the Fire Precautions Act 1971 to cover underground stations
(the S ub-surface Railway Station Regulations, also known as ‘Section 12’).
Subsequently, the Railways (Safety Case) Regulations 2000 (HMSO 2003)
replaced the 1994 Regulations and made them the sole responsibility of the HSE.
Here, a safety case is described as a document produced by a railway operator
which describes their operations, analyses the hazards and risks from those
operations and explains the control measures, such as procedures and managerial
systems that the operator has put in place to manage those risks.
Railways operators are required to prepare and submit safety cases to HSE (HSE
2003). The safety case needs to provide sufficient specific information to describe
the nature and extent of the operation and must demonstrate that the operator
has undertaken adequate risk assessment for all operations, identified risk control
measures, and has systems in place to ensure the measures are implemented
and maintained. HSE inspectors form judgements about the completeness of
a safety case and the adequacy of the arguments presented to show that risks
have been properly controlled. The criteria represent what is currently accepted
as good practice. The criteria have been published to make them widely known
throughout the railway industry and help develop a common understanding of
the requirements for producing safety cases and to make the process by which the
HSE assess them transparent. The criteria used to assess risk assessment aspects
of Railway Safety Cases are:
• The Safety Case should give details of the duty holder’s organisation and
arrangements for identification of hazards and assessment of risk.
• The Safety Case should justify the methodologies used for the identification
of hazards and assessment of risk with particular reference to any assumptions
and data used, together with the methods of calculation.
• The Safety Case should describe the significant findings of the risk assessments
and demonstrate that the control measures are adequate to control the risk to
a level as low as reasonably practicable.
• The Safety Case should describe the duty holder’s arrangements to review
risk assessments in the light of new information, new technology, incidents,
or other changes that may affect risks, and to ensure that the risk assessments
remain valid.
1.5.5 Tunnels
1.5.5.1 Mont Blanc fire

On 24 March 1999 a fire occurred in the Mont Blanc tunnel between France
and Italy, killing 38 people (Lacroix 2001). The tunnel is 11,600m long with bi-
directional single lane traffic and had a transverse ventilation system (air inlet and
exhaust) along the length of the tunnel.
At 10:53 an HGV stopped halfway through the tunnel and was seen to be on
fire. The HGV pulled a refrigerated trailer containing 9 tons of margarine and 11
tons of flour and was insulated by polyurethane foam. The driver was unable to
reach his cab extinguisher and did not attempt first aid fire fighting. At 10:57 the
alarm was raised by tunnel emergency telephone. The fire brigades entered the
tunnel at 10:57 and 11:11 from France and Italy (respectively), but were unable
to effect extinguishing operations due to thick black smoke.
There were delays in mobilising emergency response on the Italian side of the
tunnel and in configuring the ventilation system to manage the smoke. The fire
was between 75 and 190MW at its peak and burnt for 53 hours. Many of the dead
were found in their vehicles and four cars had attempted U -turns.
1.5.5.2 Tauern fire

On 29 May 1999, in the Tauern Road Tunnel in Salzberg, Austria, a rear-end
collision of a lorry caused a large fire that killed eight people (Eberl 2001). The
tunnel is 6,400m long with bi-directional single lane traffic and a fully transverse
ventilation system.
At 04:48 ignition occurred after a road traffic accident. The lorry contained
spray cans and paints and there may have been a fuel spill from one of the cars.
The fire grew steadily, the ventilation system was configured in emergency mode
with a clear layer through which many people escaped to the adjacent safety
tunnel. Ten to 15 minutes after ignition, a series of explosions were seen to lead to
a rapid increase in heat and smoke and smoke stratification was lost.
The fire was extinguished by 11:00 the following day and this revealed that four
people had died in the original road traffic accident and four people were found to
have died from the effects of the fire in a vehicle. They had been seen to evacuate
the tunnel and then return to pick up papers from vehicles.
1.5.5.3 The European Tunnel Directive

Standards for road tunnels in the UK are set in a guide BD78/99 published by the
Highways Agency. Most new road tunnel projects are now subject to some form
of fire risk assessment using either matrix methods, probabilistic analysis and/or
full quantification.
In 2004, the EU published a Directive for road tunnels on the Trans-European
road network. It remains to be seen how this will be enacted by Member States,
what implications it will have for other road tunnels (via the ALARP principle)
Introduction 27
and how, and to what extent, fire risk assessment will form part of the safety
regime in future.
1.5.6 Summary
This section shows how quantitative fire risk assessment was first developed
several centuries ago. It again found favour in addressing risk issues in the nuclear
industry where many of the technical foundations were established.
As fire disasters have affected other activities, these approaches and their
regulatory regime have been implemented in different ways to suit different needs.
In the nuclear industry great effort is placed on a robust quantification of fire
hazards and risks against a background of approved codes of practice; whereas
in the railway industry, fire safety is normally addressed by accepted company
standards with qualitative risk assessment to satisfy the fire regulations for large
numbers of simple existing buildings and quantitative risk assessment methods
used by exception to address particular significant issues.
Buildings generally use qualitative techniques and semi-quantitative
techniques, but quantitative approaches are being increasingly used to better
inform fire safety decisions in a wide range of buildings such as utility buildings,
hospitals, airports and railway stations.
1.6 General principles of fire risk assessment
1.6.1 Introduction
There are a wide range of fire risk assessment models and approaches available
(Ramachandran 1979/80, 1988, 2002, Beck and Yung 1994, Fraser-Mitchell
1997, Frantzich et al. 1997, Watts 1996, Fitzgerald 1985, Charters and Marrion
1999). However, until recently fire risk assessment was rarely used in the design
of buildings and so this section proposes that, for fire risk assessment to become
an integral part of fire engineering design, it needs to satisfy at least three
requirements. It needs to:
• be consistent with the regulatory paradigm;

• be based on engineering and scientific principles; and
• add value to the design process.
This section also discusses the proposition that there are two things that
quantified fire risk assessment does not need to do to become an integral part of
the design process. It does not need to:
• address all aspects of building fire safety; or

• recreate exact fire events.
These five issues are discussed in more detail below.

1.6.2 Consistency with the regulatory paradigm
To be consistent with the regulatory paradigm means that quantified fire risk
assessment needs to address the concerns and objectives of fire regulations. In
England and Wales, the legislation that governs the design of new buildings is the
Building Regulations 1991. The objectives for fire safety are contained in Part B of
these regulations (ADB 2007) and are divided into five areas:
• B1 Means of warning and escape

• B2 Internal fire spread (linings)
• B3 Internal fire spread (structure)
• B4 External fire spread
• B5 Access and facilities for the fire service.
The objectives are stated as functional ‘Requirements’. For example, B1

requires that:
The building shall be designed and constructed so that there are appropriate
provisions for the early warning of fire, and appropriate means of escape in
case of fire from the building to a place of safety outside the building capable
of being safely and effectively used at all material times.
The prescriptive guidance on how to meet this requirement is contained in the

ADB (2007). For a specific type of building this guidance provides information
on the:
• number of occupants
• number of exits
• overall exit width
• travel distance to an exit
• location of exits.
For fire risk assessment to be accepted as ‘evidence tending to show’ that a

design fulfils the functional requirement of the legislation, it is essential that it
addresses the issue or issues of concern (the functional requirement) and uses the
same parameters as the prescriptive guidance.
1.6.3 Based on engineering and scientific principles

To satisfy engineering and scientific principles, models for quantified fire risk assess-
ment need to have a good theoretical and evidential basis. General principles such as:
• the conservation of mass and energy;

• that risk is the combination of the frequency and severity of an unwanted
event; and
Introduction 29
• fire safety is concerned with the time and location of certain events such as
evacuation, untenability and/or structural collapse, are essential in judging
the theoretical robustness of a model.
The evidence and data on which the model is based is also important. For
probabilistic data such as the frequency of fires, reliability of systems and any data
on the frequency of the unwanted outcome(s) of concern, we need to assess:
• What is the set of cases that the data is drawn from?

• What case is the data measuring?
• How similar is the building being designed to the cases considered?
• Will variations in statutory controls and design practice skew the data?
Equally, for physical data, we need to ask: is it based on standard tests or ad-
hoc tests, large or small scale, new or as used, single or multiple tests, in a similar
enclosure etc? The answers to these questions and the nature of the data will
affect the way in which the data is used, and more significantly, the way in which
the answer is used.
An example of the kind of difficulty that can arise occurred with the use of a
well-known points scheme for fire risk assessment. The scheme covered a building
occupancy where evacuation of occupants is difficult and often hazardous, so
the fire strategy was usually based on a significant degree of compartmentation.
However, in attempting to include all fire safety systems in the assessment, the
scheme allowed a very low level of compartmentation to be compensated for by
a very high degree of fire signs, notices and emergency lighting. Although this
‘alternative strategy’ may work in some occupancies, it was clearly inappropriate
here and this occurred because the theoretical model did not adequately represent
the system it was modelling.
This example also implies that there may be fire systems whose need is self-
evident, where the impact of their level of performance on the level of risk cannot
be predicted using current knowledge. Fire signs, notices and emergency lighting
all fall into this category and are, therefore, normally best served by prescriptive
system standards.
However, no matter how hard we try to model all aspects, to make the fire risk
assessment feasible, simplifications and assumptions are necessary. So, as with all
engineering analysis, simplifications or assumptions must be clear and supported
by evidence. When undertaking quantified fire risk assessment against absolute
criteria and comparative analysis, where the ranking of solution is important,
it is possible to make simplifications and assumptions err on the side of safety.
However, in comparative analysis, where the quantitative difference between
cases is important, care should be taken that the assumptions do not bias the
comparison in one direction or another.
There also needs to be a balance in approach to the method of quantified risk
assessment. For example, some models incorporate very sophisticated physical sub-
models for fire growth, smoke movement, detection, egress, structural response
etc and use single point data for highly variable variables. They often also link the
physical sub-models together with simple ‘yes/no’ Boolean logic. This may be an
attempt to use the best available knowledge, but there may quickly come a point
where the ‘span of sophistication’ is so great that the higher level of sophistication
becomes unhelpful. Equally, other models use very sophisticated statistical models
to quantify the risk from fire. However, these methods often make highly simplified
assumptions about the physical environment, which mean that they cannot take
the particular nature of a specific building into account when assessing the risk.
Both of these types of model have been used in support of national fire safety
policy decisions but have not been used extensively on specific projects. Therefore,
it is recommended that quantitative fire risk models should treat the frequency
and the consequence sides of the fire risks predicted with an appropriate and
balanced emphasis, depending on the aspect of fire safety being analysed.
1.6.4 Adding value to the design process

To add value to the design process, the analysis needs to be feasible, usually on
several different design options, during the ‘design window’. Design is often an
iterative process and fire safety analysis needs to be undertaken in a finite period of
time for its results to be of use. The design iteration time step can be anything from
a few days to a few months depending on the building, but the trend is generally
downward and so detailed quantified fire risk assessment that takes weeks to
perform is going to be of limited use in the design of most buildings.
Just as crucially, the results of the analysis also need to improve the design
decision-making process in terms of the safety, aesthetics, function and/or cost of
the finished building. Therefore, as with the regulatory paradigm, the quantified
risk analysis needs to have the same currency as the architect. Fortunately, the
regulatory paradigm and the design parameters of interest to the architect have a
high degree of overlap and so it should be relatively easy to use the same design
parameters to address issues of concern to the architect.
The above two requirements combined demand that quantified fire risk
assessment be quick enough to use, but sufficiently sophisticated that it can
provide the necessary detail and level of accuracy for design decision making.
1.6.5 Not addressing all aspects of building fire safety

A common expectation of quantified fire risk assessment methods is that they
should address all aspects of building fire safety. It is certainly true that many fire
safety systems impact on the level of fire safety in more than one way. For example,
a sprinkler system will suppress a fire, reducing property damage and reducing the
risk to occupants. However, independent aspects of fire safety can be addressed
using different techniques, e.g. radiation modelling of external fire spread and
prescriptive standards for egress, such as travel distances, as long as the resultant
level of those particular fire precautions does not impact adversely on a separate
fire safety objective.
Introduction 31
The same is also true of quantified risk assessment. This does, however, imply
that those aspects not addressed by quantified fire risk assessment need to be
addressed by prescriptive standards or by deterministic performance-based fire
design. It also implies that the resulting package of fire precautions needs to form
a coherent whole.
1.6.6 Not recreating exact fire events

Another common expectation of quantified fire risk assessment is that it should
recreate, exactly, past fire events. Examination of structural analysis shows that it
does not exactly model the forces in, or the forms, of a structural member in, say, a
bridge. So fire risk analysis, like any other safety critical engineering analysis, only
needs to represent actual events to the extent necessary to provide a sound basis
for design decision-making.
References
ADB (2007), The Approved Document to Part B of the Building Regulations 2000, As
amended 2002, (ADB), Office of the Deputy Prime Minister, The Stationery Office,
London.
Allen F R, Garlick A R, Hayns M R and Taig A R (1992), The Management of Risk to Society
from Potential Accidents, Elsevier, London.
Beck V B and Yung D (1994), The development of a risk-cost assessment model for the
evaluation of fire safety in buildings, Proceedings of 4th International Symposium of Fire
Safety Science, International Association of Fire Safety Science, Ottawa.
BS 7974 (2001), Code of Practice on the Application of fire safety engineering principles to
the design of buildings, British Standards Institute, London.
Carvel R O, Beard A N and Jowitt P W (2001), The influence of longitudinal ventilation
systems on fires in tunnels, Tunnelling and Underground Space Technology, 16,(1) 3–21.
Charters D (2000), What does quantified fire risk assessment need to do to become an
integral part of design decision making?, Proceedings of Conference on Engineered Fire
Protection Design, Society of Fire Protection Engineers, San Francisco.
Charters D and McGrail D (2002), Assessment of the environmental sustainability of
different performance based fire safety designs, Proceedings of the 4th International
Conference on Performance-Based Codes and Fire Safety Design Methods, Society of Fire
Protection Engineers, Melbourne.
Charters D and Wu S (2002), The application of ‘simplified’ quantitative fire risk assessment
to major transport infrastructure, Society of Fire Protection Engineers Symposium on
Risk, New Orleans.
Charters D, Paveley J and Steffensen F-B (2001), Quantitative fire risk assessment in the
design of a major multi-occupancy building, Proceedings of Interflam 2001, Interscience
Communications, Cambridge.
Charters D A (1992), Fire risk assessment in rail tunnels, Proceedings of the International
Conference on Safety in Road and Rail Tunnels, Independent Technical Conference, Basle.
Charters, D A (1996), Quantified assessment of hospital fire risks, Proceedings of Interflam
96, Interscience Communications, Cambridge.
Charters D A and Marrion C (1999), A simple method for quantifying fire risk in the
design of buildings, Proceedings of the Society of Fire Protection Engineers Symposium on
Risk, Uncertainty, and Reliability in Fire Protection Engineering, Society of Fire Protection
Engineers, Baltimore, MD.
Combined Code (1998), Corporate Governance, Organisation for Economic Cooperation
and Development (OECD), Paris.
Cullen W (1990), The Public Inquiry into the Piper Alpha Disaster, Department of Energy,
HMSO, London.
Eberl G (2001), The Tauern Tunnel incident: what happened and what has been learned,
Proceedings of the Fourth International Conference on Safety in Road and Rail Tunnels,
Independent Technical Conferences, Tenbury Wells.
Farmer F R (1967), Siting criteria – a new approach, International Atomic Energy Agency
(IAEA) Symposium, April.
Fennell D (1988), Investigation into the King’s Cross Underground Fire, Department of
Transport, HMSO, London.
FIRECODE HTM 86 (1987) Fire Risk Assessment in Hospitals, Department of Health,
HMSO, London.
FIRECODE HTM 86 (1994) Fire Risk Assessment in Hospitals, Department of Health,
HMSO, London.
Fitzgerald, R W (1985), An engineering method for building fire safety analysis, Fire Safety
Journal, 9, 233.
Frantzich H, Magnusson S -E, Holmquist B and Ryden J (1997), Derivation of partial safety
factors for fire safety evaluation using reliability index β method, Proceedings of 5th
International Symposium of Fire Safety Science, International Association of Fire Safety
Science, Melbourne.
Fraser-Mitchell, J (1997), Risk assessment of factors relating to fire protection in dwellings,
Proceedings of 5th International Symposium of Fire Safety Science, International Association
of Fire Safety Science, Melbourne.
Green A E and A J Bourne (1972), Reliability Technology, Wiley-Interscience, New York.
HMSO (1965), Nuclear Installations Act, 1965, HMSO, London.
HMSO (1988), The Tolerability of Risk from Nuclear Power Stations, Health and Safety
Executive, HMSO, London.
HMSO (1992), The Offshore (Safety Case) Regulations 1992, HMSO, London.
HMSO (2003), The Railways (Safety Case) Regulations 2000 (as amended 2003), HMSO,
London.
HSE (1999),The Control of Major Accident Hazards (COMAH) Regulations 1999, Health
and Safety Executive, London.
HSE (2003), Good Practice and Pitfalls in Risk Assessment, Research Report 151, Health and
Safety Executive, London.
Lacroix D (2001), The Mont Blanc Tunnel fire: what happened and what has been learned,
Proceedings of the Fourth International Conference on Safety in Road and Rail Tunnels,
Independent Technical Conferences, Tenbury Wells.
Laplace P-S (1814), Essai philosophique sur les probabilities.
Linstone H A and Turoff M (1975), The Delphi Method: Techniques and Applications,
Addison-Wesley, Glenview, IL.
Office of the Deputy Prime Minister (2005), UK Fire Statistics, Office of the Deputy Prime
Minister, London.
PD 7974 Part 7 (2003), Application of fire safety engineering principles to the design of
buildings, Part 7 Probabilistic Risk Assessment, British Standards Institute, London.
Introduction 33
Ramachandran G (1979/80), Statistical methods in fire risk evaluation, Fire Safety Journal,
2, 125–154.
Ramachandran G (1988), Probabilistic approach to fire risk evaluation, Fire Technology, 24,
3, 204–226.
Ramachandran G (2002), Stochastic models of fire growth, SFPE Handbook of Fire
Protection Engineering, 3rd Edition, Section 3, Chapter 15, Society of Fire Protection
Engineers, Baltimore, MD.
Turnbull N (1999), Internal Control: Guidance for Directors on the Combined Code, (The
Turnbull Report), The Stationery Office, London.
Watts, J M (1996), Fire risk assessment using multiattribute evaluation, Proceedings of 5th
International Symposium of Fire Safety Science, International Association of Fire Safety
Science, Melbourne.
World Fire Statistics Centre (2007), Information Bulletin No 23, TheGeneva Association,
Geneva.
2 Qualitative and semi-quantitative
risk assessment techniques
This chapter describes some of the qualitative and semi-quantitative fire

risk assessment techniques used in fire safety. The qualitative techniques are
categorised as unstructured and structured. The semi-quantitative methods
described include matrix methods and points schemes.
2.1 Qualitative techniques

Most fire risk assessments of buildings are undertaken to satisfy legislation and
in the vast majority of cases that means the fire regulations in existing buildings
which addresses life safety in occupied buildings. There can also be many other
reasons for undertaking a fire risk assessment of a building. These include:
• life safety
• asset protection
• mission/business continuity
• environmental protection
• heritage
• public image
• post-fire analysis.
The initial qualitative assessment process in fire safety engineering (also

known as the Qualitative Design Review (QDR) or Fire Engineering Design Brief
(FEDB)) can be likened to a qualitative risk assessment process based on a table-
top exercise.
Qualitative methods rely on identification of factors that affect risk. These
factors may be those that affect the level of hazard, such as combustibles or the
level of fire precautions intended to mitigate those hazards, such as fire detection.
An assessment of all these factors is undertaken against a benchmark set of values
and a judgement made as to whether the factor is higher, lower or about the
same as the benchmark. A review is then undertaken of all the factors to judge
how, overall, the area being assessed compares to the benchmarks. If additional
mitigation measures are judged to be required, this is a user input or a choice from
Qualitative and semi-quantitative risk assessment techniques 35
a prescribed list. Assessments under the Fire Safety Order in the UK fall into this
category.
Qualitative fire risk assessment methods can be placed into two categories:
unstructured and structured.
2.1.1 Unstructured methods

An unstructured method of qualitative fire risk assessment would follow the five-
step risk assessment process outlined in the guidance to the Fire Safety Order
2005 (TSO 2005). These steps are:
1 Identify potential fire hazards.

2 Identify those in danger from fire.
3 Evaluate the risks and whether the current fire precautions are adequate.
4 Record findings and actions and communicate to employees.
5 Keep the assessment under review and revise it when necessary.
The method is unstructured as it has no prescribed standards for what types

and levels of precautions are appropriate for various levels of hazard. This can
make judging adequate compliance rather subjective.
2.1.2 Checklists
Quite often checklists are used to help identify fire hazards in unstructured
qualitative risk assessments and in fully quantitative fire risk assessments. The
advantage of a checklist is that it indicates to the non-professional what may
constitute a fire hazard and it reminds the professional risk assessor of the range
of fire hazards. The disadvantages can be that they are used too literally to record
every piece of combustible material and they can lead to unusual fire hazards being
missed, because any checklist can never be totally comprehensive. The fire hazard
lists below are examples of checklists. Fire hazards may include combustible
materials and ignition sources.
Combustible material (Figure 2.1) can include:
• products based on flammable liquids, e.g. paints, varnish, thinners;

• flammable liquids and solvents such as petrol, white spirit and paraffin;
• flammable chemicals;
• wood;
• paper and card;
• plastics, rubber, foam such as polystyrene and polyurethane used in packaging
and furniture respectively;
• flammable gases, such as LPG;
• textiles;
• packaging materials;
• waste materials such as wood shavings, offcuts, dust, paper and textiles.
Figure 2.1 An example of combustible materials
Ignition sources (Figure 2.2) can include:
• matches and smokers’ materials;

• naked flames;
• electrical, gas or oil fired heaters;
• hot processes such as welding and grinding;
• cooking;
• engines or boilers;
• machinery;
• faulty or misused electrical equipment;
• lighting equipment such as halogen lamps;
• hot surfaces and the obstruction of equipment ventilation;
• friction from loose bearings or drive belts;
• static electricity;
• metal impacts such as metal tools striking each other;
• arson.
Lists of hazards such as these can form the basis of checklists and be used as
part of a hazard identification process.
Figure 2.2 An example of a car fire, most car fires are ignited by electrical faults or arson
A number of factors are important in determining the level of hazard from a

combustible material. For combustible materials, the extent of the hazard depends
on their:
• ignitability – e.g. certain flammable liquids are easier to ignite than others;
thin items tend to be easier to ignite than thick items (see Drysdale 2002
p208);
• reaction to fire – e.g. rate of flame spread, heat release, smoke production,
toxicity etc (BS EN 13501-1 2000);
• amount – the higher the fire load the greater the potential rate of heat release
and fire severity;
• orientation – vertical, corner and ceiling surfaces and high-racked storage
tend to increase the rate of flame spread (see Drysdale 2002 p232);
• location – materials that are grouped together or near exits or large numbers
of people may present an increased risk.
A number of factors are important in determining the level of risk represented

by an ignition source. For ignition sources the extent of the hazard that they
present may depend on their:
• frequency of occurrence – more frequent ignition sources such as cooking and

hot works tend to increase the level of risk;
• ignition energy – higher ignition energy may mean increased initial fire
growth rate;
• temperature – higher temperate ignition sources may result in increased
initial fire growth rate (see Drysdale 2002 pp191 and 212);
• number – multiple ignition sources from, say, arson may be more dangerous
than a single source;
• timing – ignition when many people are present or no-one is present may
result in increased life and property fire risks (respectively);
• location – ignition at the bottom of an item or near an exit may increase the
risk.
Unstructured approaches are simple to apply and may be very general in

nature. This may mean that they can be highly subjective and may rely heavily on
the expertise and judgement of the risk assessor.
Some guidance for fire risk assessment in existing buildings requires that fire
risk assessments are ‘suitable and sufficient’ without prescribing an approach. This
guidance describes a unstructured qualitative form of fire risk assessment as the
regulations apply to a wide range of workplaces, many of which are small and
simple in nature.
A summary report for an unstructured fire risk assessment can be seen in Figure
2.3. The risk assessment consisted of a survey of the building, locating all hazards
and precautions, designed or installed, with relation to fire safety. In addition to
this, staff members in the different occupancies were asked some questions in
relation to safety procedures and checks were made concerning the maintenance
of hand-held fire extinguishers. All these factors were then evaluated to assess
whether the risks from fire were adequately addressed.
Where available, the maintenance records for fire safety related equipment
were inspected. However, during the survey of the building, no equipment was
tested.
There are a number of alternative frameworks for carrying out fire safety risk
assessments. The assessment in this building was carried out using the five-step
process above. The checklist, which defines the manner in which the assessment
was carried out is outlined below:
1 Assess the fire risks in the workplace.

2 Check that a fire can be detected in a reasonable time and that people can
be warned.
3 Check that people who may be in the building can get out safely.
4 Provide reasonable fire-fighting equipment.
5 Check that those in the building know what to do if there is a fire.
6 Check and maintain your fire safety equipment.
The building was surveyed on a floor-by-floor and occupancy-by-occupancy

basis. Within each area, the hazard to persons within the occupancy and persons
in other occupancies were considered.
CFire Solutions Fire Safety Risk Assessment Schedule

Client: Bristol Holdings Date of 21 July 2002 / 13
assessment: October 2002
Site: Febrile House, Assessment officer: Cherry Stone
Umbridge
Location of Ground Title off occupancy: Fenda
occupancy: Floor
Description of Open plan office
occupancy:
Fire risks Both escape doors from this space are not remote from
identified each other (i.e. they are in the same space).
Fire detection and Manual call points provided.

alarm provisions
Means of escape Travel distance to alternative escape doors is short

in the event of fire (< 18m). No emergency lighting in tenant’s area. Fire
doors (main escape doors) appear to be FD30 with self
closers but door fit was inadequate with a 10mm gap
between doors.
Fire suppression Manual fire extinguishers (water and CO2).

equipment
Fire safety No written procedures noted.

procedures and
management
Maintenance Water extinguishers checked March 2002 and CO2

of fire safety extinguishers checked Jan 2001. Fire alarm tested
provisions weekly.
Additional notes In general, due to the area being small, means of

escape is considered to be adequate if emergency
lighting were installed. The concern is that a fire could
grow undetected to a point where manual fire fighting
is ineffective and smoke could spread into both escape
stairways.
Figure 2.3 Summary report for an unstructured fire risk assessment
During the survey of the premises, representatives from all available tenants
were asked questions, including:
• Do you have an evacuation procedure?

• What do you do when the alarm sounds?
• Have you had fire drills?
• When was the last drill conducted?
2.1.3 Structured methods

Structured methods of qualitative fire risk assessment use the same process as
unstructured methods except that, in addition, they tend to prescribe:
• different levels of hazard;

• different levels of precaution;
• the acceptability of different combinations of hazard and precaution; and,
• appropriate risk reduction alternatives.
A good example of this approach is that of FIRECODE HTM 86 (1994) ‘Fire risk
assessment in hospitals’. This method identifies different types of fire hazard such
as ‘smoking’ and ‘fire hazard rooms’ and whether the level is ‘acceptable’, ‘high’ or
‘very high’. It then goes on to identify different types of fire precaution including
‘alarm and detection’ and ‘compartmentation’ and whether the level of provision is
‘unacceptable’, ‘inadequate’, ‘acceptable’ or ‘high’. Certain combinations of hazard
ratings and precautions ratings are deemed to be acceptable. Other combinations
lead to the need to reduce risk by a range of prescribed alternatives.
Figure 2.4 shows a summary assessment sheet for a fire risk assessment to HTM
86 (1994). Like the unstructured fire risk assessment shown in Figure 2.3, this
risk assessment consisted of a survey of the building, locating all hazards and
precautions, with relation to fire safety. Similarly, staff members in the different
occupancies were asked some questions in relation to safety procedures and checks
were made concerning the maintenance of hand-held fire extinguishers. All these
factors were then evaluated to assess whether the risks from fire were adequately
addressed.
Again, where available, the maintenance records for fire safety related
equipment were inspected. However, during the survey of the building, no
equipment was tested.
As before, the building was surveyed on a floor-by-floor and occupancy-by-
occupancy basis. Within each area, the hazard to persons within the occupancy
and persons in other occupancies were considered.
During the survey of the premises, representatives from all available tenants
were asked the following questions:
• Do you have an evacuation procedure?

• What do you do when the alarm sounds?
Assessment record Meddleston Primary Health Care NHS Trust

10/04/1999
Victoria High Standard Acceptable High risk/ Very high Inadequate Unacceptable
Ground Floor Risk/hazard Hazard risk/hazard
Pickersgill A Block 10 or HTM 85
Standard
1 Patients X
HAZARDS
Ignition sources
2 Smoking X
3 Fire started by patients X
4 Arson X
5 Work processes X
6 Fire hazard rooms X
7 Equipment X
8 Non patient access areas X
9 Lightning X
Combustible materials
10 Surface finishes X
11 Textile and furniture X
12 Other materials X
PRECAUTIONS
Prevention
13 Management X
14 Training X
15 Fire notice and signs X
Communications
16 Observation X
17 Alarm and detection X
Means of escape
18 Single direction of escape X

19 Travel distance X
20 Refuge X
21 Stairways
22 Height above ground level X
23 Escape lighting X
24 Staff X
25 Escape bed lifts
Containment
26 Elements of structure X
27 Compartmentation X
28 Sub-division of roof and ceiling voids
29 External envelope protection X
30 Smoke control
Figure 2.4 Record sheet from a structured qualitative fire risk assessment method
(HTM 86 1994)
• Have you had fire drills?
• When was the last drill conducted?
The main difference of the structured compared with the unstructured fire risk
assessment is the evaluation and presentation of the findings. In this case, each
assessment sheet was completed from information gained both during the survey
and through desktop analysis of drawings and policy statements. This section of
the report has further broken down the assessments into three distinct areas:
• adequate assessment areas;

• compensated assessment areas; and
• inadequate assessment areas.
The results of the building survey, review of documentation and records and
of ad hoc audit questions are then expressed against the prescriptive standards on
the summary assessment sheet in Figure 2.4.
This kind of qualitative flexible prescription identifies areas of high hazard and/
or inadequate precautions and prescribes one or more prescriptive solutions. This
is relatively easy and consistent to apply, taking only a matter of hours to survey
and days to report. This method does not suffer from the apparent acceptability
of inappropriate fire strategies. However, the method cannot easily address non-
standard situations, limits options for upgrade and is not applicable to new designs.
The structured fire risk assessment approach was adopted for hospitals because
they comprise a more homogeneous estate and in this case, structured methods
may be more efficient and provide greater consistency of assessment.
2.1.4 Fire hazard definitions

Another example of qualitative prescription in fire risk assessment can be found
in this extract from BS 5588 : Part 11 (1997) ‘Code of practice for shops, offices,
industrial, storage and other similar buildings’, Section 8.1.1.
Examples of different types of hazard are as follows:
• Normal hazard. Where any outbreak of fire is likely to remain localised or

is likely to spread only slowly, and where there is little hazard of any part of
the building structure igniting readily.
• Low hazard. Where there are very few flammable and no explosive
materials present and where the hazard of a fire breaking out and smoke
and fumes spreading rapidly is minimal.
• High hazard. Where there are:
• Materials stored or handled in such quantities or dispositions that
they would be likely, if ignited, to cause a rapid spread of fire, smoke
or fumes. For example, processes handling large quantities of highly
flammable liquids, gases or solids, such as polyurethane foam;
• Unusual circumstances relating to the occupants;
• Certain areas which, due to their function, may present a greater risk
of fires occurring and developing than elsewhere.
A similar type of approach is adopted in BS9999: 2008
2.2 Semi-quantitative techniques

Semi-quantitative fire risk assessment methods place some numerical value on
the level of risk, without that numerical value representing a precise value of risk.
2.2.1 Matrix methods

The matrix method, so called for its use of matrix-like tables, was developed from
the HAZard and OPerablity (HAZOP) method used in the chemical industry.
HAZOP is a technique for systematically considering deviations from the design
intent of a system and was developed in the chemical industry to identify failure
modes of complex process plant (Mannan 1980, Chemical Industry Safety and
Health Council 1977). HAZOP studies identify deviations by the application of
a series of guide words such as ‘too much’, ‘too little’ to the process parameters,
such as flow and pressure. The process is best suited to group sessions with multi-
disciplinary participants. The technique is completely general and can be applied
to processes or procedures of any type and complexity.
The matrix method below incorporates many of the features of HAZOP such
as systematic hazard identification with a multi-disciplinary group, but in a much
more general way. It then takes the process further by classifying the frequency
and consequences of identified events in categories representing an order of
magnitude, e.g., one per year to once every ten years etc.
Matrix methods are usually based on a ‘structured brainstorm’ in a workshop
environment (Mannan1980). Rather than a formal ‘walking of the halls’ with
clipboard in hand, they make use of the collective experience and knowledge
of people involved in an asset (i.e., the building or other construction under
consideration) or people who have related generic knowledge e.g. the organisation’s
safety officer. Often this includes representatives of the design team and the
management team.
It is a method which tries to identify high risk areas (i.e., specific locations or
specific activities) so that more precise techniques, which are too costly to apply to
the whole facility, can be efficiently targeted. Often these high risk areas are then
subject to a quantitative risk assessment.
2.2.1.1 Matrix methodology

The matrix method of fire risk analysis is best illustrated by a flow chart which is
specifically designed for a rail facility (see Figure 2.5).
ASSET fire hazard

analysis
Identify flow types involved Identify critical tunnel

1. Operation 1. Activities
2. Maintenance 2. System
3. Defect management 3. Areas
Identify possible fire hazards to:

• Public and ASSET personnel
• ASSET operations
Identify triggers that may cause

accidents related to the hazards
Estimate the potential frequency Estimate the potential consequences

of the hazard and accident of accidents, i.e.:
trigger occurring • Major injuries and fatalities
• Disruption to ASSET operations
Rate the fire hazards associated

with the identified accidents and
compare with tolerability criteria
Screen out minor fire hazards
Provide a list of identified major

fire hazards in a table format
Figure 2.5 An example of a matrix method flow chart

Each of the stages are done in turn. It is a ‘one off’ and not an iterative approach,
i.e., this is very much a snapshot risk assessment method rather than a method
of ongoing management. A table is used to record findings as shown in Table 2.1.
This table is completed by the workshop attendees.
Tables 2.2–2.7 present typical ‘look-up’ matrices, which are used by the
workshop teams to objectively assess fire hazards. These tables will be created
before the matrix workshop. Note that Tables 2.2, 2.3 and 2.5 are industry-specific
to different degrees.
Hazards considered can be defined to include a narrow or wide range, such as:
• flammable liquids;
• flammable gases;
• recognised hazardous loads;
• HGV vehicles;
• multi-passenger vehicles;
• cars and light vans.
Other hazards include:
• trespass and arson;

• asset equipment.
Further hazards identified by the workshop team are also included in the
analysis. The workshop will normally be limited to fire hazard analysis, however,
asset operations and effectiveness of equipment are often the subject of a further
HAZOP workshop and evaluation.
Table 2.1 Typical matrix method record sheet

Tolerability/
remarks
Accident frequency
Risk classification
Hazard frequency
Trigger probablity
Accident severity
Hazard
no Fire hazard Accident
Potential Causes Potential Trigger
Table 2.2 Hazard /accident frequency
Category Description Definition Guide frequency Example
A Frequent Likely to occur repeatedly (more than once a week) > 1 per 100 hrs Stoppage due to flow congestion
Normal operations
B Occasional Likely to occur from time to time (once a year or 1 per 100 to 1 per Temporary loss of control function
more) 10,000 hours
C Probable Likely to occur once on the system during its 1 per 10,000 to 1 per Failure of active system, e.g. ventilation fan
operational life (e.g. once in 30 years) million hours
D Improbable Unlikely to occur during the system design life (i.e., 1 per million to 1 per Major chemical spillage in asset
less than one in 100 years) 108 hours
E Remote Very unlikely to occur during system design 1 per 108 hours to 1 Major earthquake
life (i.e., less than 1% likelihood of happening per 1010 hours
throughout the life of the system)
F Incredible Extremely unlikely to occur during the life of the < 1 per 1010 hours Meteor impact
system (i.e. once in a million years).
Table 2.3 Accident trigger probability

Category Description Definition Guide frequency Examples
A Frequent Likely to exist repeatedly on the system > 10% Traffic driving too fast
B Occasional Often exist somewhere on the system 1% – 10% Maintainer working in the plant room
C Probable Likely to exist from time to time on the system 0.01% – 1% Staff working in asset during running hours
D Improbable Likely to exist only on rare occasion on the system 0.001% – 0.01% Trespasser in asset (possible arsonist)
E Remote Unlikely to exist even on rare occasion on the system 0.0001% – 0.001% Control room unavailable, e.g. due to fire
F Incredible Extremely unlikely ever to exist on system < 0.0001% Two separate serious fires in asset
Table 2.4 Accident frequency categories
Accident frequency Hazard cause frequency
A B C D E F
Accident A A B C D E F
trigger B B B C D E F
C C C D E F F
probability D D D E F F F
E D D F F F F
F E E F F F F
Table 2.5 Accident severity

Category Description General description Examples
I Catastrophic Possible multiple deaths, or Vehicle strikes maintenance
probable death of passenger crew
or member of the public, or Passenger(s) fall from moving
total system loss vehicle
Uncontrolled incident Collapse or loss of asset
leading to: Assets needing repair or
Asset unavailable (asset recommissioning before flow
cannot operate) can resume
Incidents resulting in asset
unavailability for > 24 hours
II Critical Probable major injury to Trespasser struck by vehicle
one or more passengers Collision of vehicles in asset
or members of the public, Assets needing inspection
or probable major injury before flow can resume
and possible death of staff, Controlled evacuation from
contractor or trespasser or vehicles into asset
major system damage Incidents resulting in asset
Loss of control function unavailability for < 24 hours
Asset cannot operate
III Marginal Probable minor injuries to Maintainer strikes head on
one or more passengers or sharp object.
public or staff, contractors Strain injury to asset staff
or trespassers, or possibility Controlled evacuation of
of major injury to a single vehicles from asset
individual Signalling fault reported by TO
Major service disruption Fire alarm and FP system
Third party action operation in Control Centre
requiring halt to operations Control Centre fault requiring
whilst remedial action investigation
(rescue, repair, inspection)
takes place
IV Negligible Trivial injuries to Unavailability of FP system
passengers or staff, with due to maintenance / defect
no more than possibility of management
minor injury
Unscheduled maintenance
required without
interruption to service.
Table 2.6 Risk classification
Risk classification Accident severity category
I II III IV
Accident A A A A B
frequency B A A B C
category C A B C C
D B C C D
E C C D D
F C D D D
Table 2.7 Risk tolerability

Category Description Definition
A Intolerable Risk shall be reduced by whatever means possible.
B Undesirable Risk shall only be accepted if risk reduction is not
practicable.
C Tolerable Risk shall be accepted subject to confirmation that risk
is as low as reasonably practicable.
D Acceptable Risk shall be accepted subject to endorsement of the
supporting hazard analyses.
2.2.1.2 Quantification for matrix methods

Matrix methods rely on each category of frequency or consequence representing
an order of magnitude. The workshop attendees, assisted by historical information
and their own experience, can then identify a range of fire hazards using a structured
brainstorming technique. Having identified a reasonably comprehensive a list of
fire hazards, these are rationalised to eliminate duplicates and trivial events. Each
fire hazard is then evaluated in turn by the workshop attendees against each of the
parameters e.g. frequency, consequence etc.
In the tables, ratings for each parameter are represented by letters and these are
then combined using a risk matrix to identify the level of risk using capital roman
numerals. These associations are not arbitrary and rely on the basic principles of
risk assessment. To illustrate this, we will consider the numerical rating system used
in the second example in section 2.2.1.3. Here each category of each parameter
represents an order of magnitude in the scale of interest and is represented by an
integer number rising as the scale rises.
The total risk rating for life safety is given by:
Total life risk = F + S + P (2.1)

where: F is the rating for the frequency of initiating event (usually ignition)
S is the likely severity of the outcome, and
P is the probability that that severity will result from that initiating event.
Clearly there is a range of events that could occur from any particular initiating
event. The workshop may feel that a particularly severe outcome is likely from an
event, but this may be offset, to a degree, by a relatively low probability and vice
versa. This helps guard against unintentional bias in the process. For example, small
fires with large consequences may happen frequently and be categorised as 1 and 3
respectively. Equally, large fires may happen infrequently and be categorised as 3 and
1 respectively. In this case both would have a risk category of 4 and so it is important
that the frequency and consequence ratings are consistent with each other.
It often helps to define the nature of the events being quantified as most likely
or reasonable worst case or both (depending on the objective of the workshop) to
help participants converge on a shared view of the ratings.
The process is addition rather than multiplication, because each number
represents an order of magnitude:
Risk = 10F × 10S × 10P = 10(F + S + P) (2.2)
For the purposes of prioritisation we can simply use Equation (2.1).

For the estimation of total business risk it is slightly more complicated because
business risk is a combination of asset damage and performance loss ‘damage’ that
may have different values for the same event.
Total business risk is given by:
Total business risk = F + P + (log10[10Cd × 10Cp]) (2.3)
where Cd and Cp are the asset damage and performance loss ratings respectively.
Thus for example:
3 + 2 + log10(103 × 103) = 8.3
In the example spreadsheet discussed at the end of this section the results can
appear to be c ounter-intuitive at first sight.
2.2.1.3 Matrix workshops

The staged objectives of such workshops are, in sequence, to:
• identify fire hazards and accidents that could occur in an asset;

• estimate the consequences (by making use of ‘accident severity matrices’;
• rate hazards (by using a risk classification matrix);
• screen out minor fire hazards; and
• provide a list of identified major fire hazards.
Once all the participants have been briefed on the asset under assessment and
the process of assessment, a structured brainstorming process is used to fill in the
first five columns of the record sheet (Table 2.1). The keys to good brainstorming
are recognised as:
• say the first thing that enters your head;

• no discussion; and
• no critical comment.
Once the fire hazards and accidents have been identified, they can be sorted
and duplicates amalgamated and irrelevances discarded, i.e. a list of hazards is
created. These would be entered into the first column of Table 2.1.
The next stage is to estimate which of the matrix categories each fire hazard best
fits. This can best be done with a combination of historical data and experience
from the key stakeholders around the table. One of the major beneficial side effects
of the matrix workshop is the final column of tolerability where comments on the
acceptability of hazards or their mitigation can be recorded for future reference.
This often leads to greater mutual understanding of a project and its hazards
amongst design and management teams alike.
Once all the fire hazards have been rated, they can be sorted in terms of priority
and the minor fire hazards noted and the major hazards addressed further by
quantitative fire risk assessment.
Often matrix methods are used to collect information on hazards and key event
progressions that are then used in a full quantitative fire risk assessment. In a
quantitative fire risk assessment following a matrix method risk assessment:
• Matrix methods are used where the subjective and lack of quantitative
information from qualitative methods may lead to poor decision making with
significant consequences.
• Matrix methods can be applied to any building or situation where there
are hazards with high potential consequences and/or where the risks of an
activity may be high. That is why matrix methods are usually a precursor to
full quantitative fire risk assessment.
Fires are important in terms of safety risk and commercial cost, both of which
need managing effectively. The importance of managing fire risk for both safety
and commercial reasons is emphasised by current fire losses and trends:
• There are hundreds of thousands fire events per year in most countries.
• Significant fires have progressively increased over recent years with
commercially significant fires accounting for most of the increase.
• The annual cost to national economies is measured in billions of pounds per
year.
Case study
This study summarises the proceedings of a fire safety risk assessment workshop
in support of a rail Safety Risk Framework. The workshop was held at a railway
station.
The declared objectives of the workshop were to:
1 Provide a company-wide preliminary indication of the nature, scale and

source of fire risk, i.e. what proportion of the assets had high, medium and
low levels of fire risk.
2 Identify any high consequence and/or high likelihood events requiring
immediate attention.
3 Produce a risk ranked list of assets to enable detailed risk assessments to be
undertaken.
The workshop used a structured, formal assessment process similar to that of

HAZID (HAZard IDentification). It was structured in that the brainstorming and
risk rating of hazards followed a process defined by the headings of the matrix
table. It was formal in that the matrix table, including comments, is a written
record that forms part of the risk register and should be a living document for the
system, i.e. it is updated regularly and when aspects of the asset or its operation
are changed.
After confirming the assets under consideration, the workshop systematically
assessed the assets and the results were recorded on a spreadsheet.
The main parameters that the workshop addressed are described below:
• asset type;
• location;
• failure mode/hazard;
• cause;
• consequence;
• for three scenarios:
• unwanted activation of fire alarm;
• most likely fire;
• worst case fire;
• to assess for each scenario their:
• frequency;
• severity;
• probability;
• asset loss;
• performance loss;
• current controls;
• comments/high priorities;
• actions;
• asset type.
The workshop focused on the particular railway station. The fire risks contained
within the station are complex and varied and are categorised by distinct locations
which in general, may include:
• platforms;
• concourses;
• retail;
• non-public areas;
• neighbouring buildings, as appropriate, which may include:
• offices;
• hotels;
• hospitals;
• construction sites;
• car parks;
• connecting underground stations.
The failure modes/hazards under consideration were ‘fire and/or unwanted fire
alarm activation’.
The workshop identified the main potential causes of the failure mode/hazard
for each location. This comprised typical combustibles/ignition sources, e.g.
‘overhead line equipment, rubbish, electrical faults, vehicles, arson, smoking’.
The workshop identified the general scale of event in terms of life safety, asset
loss and performance loss, e.g. ‘death/injury, loss of building, closure of line’.
The workshop then assessed the risks under three main categories:
• unwanted activation of fire alarm;

• most likely fire; and
• worst case fire.
For unwanted activation of fire alarm, the workshop assessed the frequency and
consequences of unwanted activation of fire alarm, e.g. an unwanted alarm within
a retail area leading to evacuation.
For most likely fire, the workshop considered the frequency and consequences
of fires that tend to be more common but not necessarily very severe, e.g. a small
fire in a kitchen or office that is confined to the room or item of origin.
For worst case fire, the workshop considered the frequency and consequences
of fires that are likely to be much less frequent and probably much more severe,
e.g. a fire that involves the whole of a carriage on a platform.
For each of the three event categories, five parameters were assessed:
• F hazard frequency
• P incident probability
• S accident severity
• Cd cost of asset damaged or replacement cost.
• Cp cost of performance penalty or operational loss.
Each parameter was given a rating. Since precise values for these parameters are
not available, each rating represents a range of values. For unwanted activation of
fire alarm, life safety and business (asset and performance), the total fire risk rating
was calculated.
The current controls were identified by the workshop in terms of compliance
with main fire legislation and standards, e.g. the fire safety order and/or company
standards.
The workshop also identified any high priorities, significant fire scenarios,
queries or further comments for the record.
The assessments of the assets under consideration were recorded on a
spreadsheet. The basis of the assessment was determined using Tables 2.8 – 2.12.
The following fictional characters may have participated in the workshop:
• Station Manager
• Duty Station Manager
• Fire Systems Manager
• Safety Risk Manager
• Project Sponsor / Company Fire Adviser
• Workshop Facilitator / Fire Risk Engineer
• Project Manager
• Workshop Secretary
The fire hazards for the railway station have been assessed and the results are
summarised in Tables 2.8 and 2.9, listing the locations in descending order of risk
rating scores in terms of life safety and possible business cost. These overall scores
have been determined from the results obtained for the three hazard scenarios
considered.
A risk rating of 10 represents a potential maximum level of risk occurrence
of 1 × 10–1 fatality equivalents per year which equates to one fatality in every 10
years. Further examples are risk ratings of 9 and 4 which represent potential levels
of risk of 1 × 10–2 and 1 × 10–7 fatality equivalents per year respectively. These
equate to one fatality in every 100 years for a risk rating of 9, and one fatality in
every 10,000,000 years for a risk rating of 4. A ‘fatality equivalent’ equals one
fatality or 10 major injuries. Table 2.13 highlights that the link works, with a risk
rating of 10, represents the highest safety risk for this station.
A risk rating of 13 represents a potential maximum loss of £1,000,000 per
year (or £10,000,000 once in ten years), 12 represents £100,000 per year and 11
represents £10,000 per year etc.
The figures in Table 2.14 are asset damage and performance penalty costs
combined. The performance loss system is a financial penalty system for late trains
and so may represent some of these issues. Not explicitly, these may be in the
minds of the workshop members when they rate the risks. Table 2.14 highlights
that link works, with a risk rating of 13.3, represents the highest business risk. The
retail outlets, with a risk rating of 13, are indicated as representing the second
highest risk.
Table 2.8 Hazard frequency, F
Occurrence frequency, F Range Rating
Never < 1 in 10,000 years 0
Remote 1 in 1,000 to 1 in 9,999 years 1
Rare 1 in 100 to 1 in 999 years 2
Infrequent 1 in 10 to 1 in 99 years 3
Occasional 1 in 1 to 1 in 9 years 4
Frequent 1 to 10 times per year 5
Common > 10 times per year 6
Table 2.9 Accident severity (life safety), S

Rating
None 0
Minor injuries 1
Major injuries 2
One fatality 3
Multiple fatalities 4
Table 2.10 Accident probability, P

Likelihood No. of fires before accident Rating
Never > 10,000 0
Very unlikely 1,000 to 9,999 1
Improbable 100 to 999 2
Occasional 10 to 99 3
Near certain 1 to 9 4
Table 2.11 Cost of asset damage or destroyed, Cd (£)

Rating
< 100 1
100 to 999 2
1,000 to 9,999 3
10,000 to 99,999 4
100,000 to 999,999 5
1,000,000 to 9,999,999 6
> 10,000,000 7
Table 2.12 Cost of performance penalties or operational loss, CP (£)
Rating
< 100 1
100 to 999 2
1,000 to 9,999 3
10,000 to 99,999 4
100,000 to 999,999 5
1,000,000 to 9,999,999 6
> 10,000,000 7
Table 2.13 Life safety risk rating

Location Risk rating
Link works 10.0
Food preparation basement 9.1
Retail outlets 9.0
East side offices (including station control room) 8.3
Non–public areas – west side offices and south west offices 8.3
Platforms 9–11 8.3
Platform areas 2–8 8.3
Concourse and forecourt 8.0
Platform 1 and access road 8.0
Clothes store (above platforms 9–11) 7.0
Car parks 6.3
Parcel post 6.3
Underground station 5.0
Public highway 5.0
Hotel Way 4.0
Overall risk ratings for the entire station for both business and life safety risk
are presented in Table 2.15. The values indicate that the overall score is largely
dependent on the greatest single risk due to the logarithmic nature of the process
(see above).
As could be expected, the life safety rating is dominated by the risk presented
by fires and not unwanted alarm activation. Fires dominate the business risk rating
whilst the risk to asset and performance are similar. There may be more scope to
reduce risks, cost effectively, for some aspects more than others.
Table 2.14 Business risk rating
Location Risk rating
Link works 13.3
Retail outlets 13.0
Platforms 9–11 12.5
Platform areas 2 to 8 12.5
Food preparation basement 12.3
Platform 1 and access road 12.1
Concourse and forecourt 11.5
East side offices (including station control room) 11.5
Non–public areas – west side offices and south west offices 11.5
Parcel post 11.0
Underground station 11.0
Car parks 9.3
Clothes store (above platforms 9–11) 9.3
Public highway 9.3
Hotel Way 8.0
Table 2.15 Overall station risk rating

Risk rating
Overall life safety risk rating 10.1
a. Unwanted activation 8.1
b. Fires 10.1
Overall business risk rating 13.6
2.1 a. Asset 7.9
b. Performance 8.0
2.2 a. Unwanted activation 12.6
b. Fires 13.6
An example of the full results of the workshop can be found in the spreadsheet
presented in Table 2.16 (courtesy of Mr J. Holland, Network Rail).
2.2.1.4 Summary
The matrix method can be applied to many assets and the results presented
graphically. For example, a histogram of number of areas against ratings can
be generated to assess the distribution of risks facing an organisation or project
and which risks should take priority for further assessment and/or mitigation.
Table 2.16 Example of part of a risk ranking spreadsheet (most likely fire event)
Ref Location Causes Consequences F S P Tot Cd Cp Tot Current controls Comments/
priority
1 Platforms/ Rubbish, smoking, Serious injury 3 2 2 7 3 3 8.3 Company standards,
concourse trains, electrical or death procedures, staff,
vehicles and contents, PA system, means
retail, arson of escape
2 Storage Rubbish, smoking, Serious injury 3 2 2 7 3 3 8.3 Controlled access, Time limit
arson, food, packaging, or death staff, automatic to current
electrical, rats detection occupancy
and warning, system,
compartmentation,
extinguishers
3 Parking Vehicles, rubbish, Serious injury 2 1 3 6 1 1 6.3 Highway code, CCTV, Check
arson staff fuel spill
facilities
4 Passages Rubbish, smoking, Serious injury 1 1 3 6 1 1 5.3 Company standards,
arson, gas main, CCTV, means of
contractors escape, staff, manual
call points
5 etc etc etc
Legend: F = Frequency
S = Severity (life safety)
P = Probability
Cd = Asset damage or destroyed
Cp = Performance penalties or loss
Distributions of life safety and business fire risk for a major business can also be
generated.
It is important to note that the range of each category is significant and that
the risk categories calculated by combinations of several categories represent
even wider ranges. These are intended for prioritisation purposes only, based on
evidence and views of stakeholders and other experts. Subsequent quantified fire
risk assessment will probably predict a level of risk consistent with the matrix
method, but any backward comparison is probably only of academic interest.
Matrix methods tend to be used when the risks and/or consequences involved
with an activity are potentially more critical. They are often a precursor to full
quantitative risk analysis of a subset of the hazards identified. Matrix methods
have a more robust technical basis but their results are based on stakeholder
judgements and so should be viewed in a comparative rather than absolute way.
2.2.2 Points schemes

Points schemes use similar factors to other qualitative methods, but inject some
numeracy to result in a score. Typically, each aspect of fire safety (or factor) is
scored from a range and then multiplied by a weighting to provide an overall
score for that factor. The overall scores for each factor are then added together
to provide an overall score for the area being assessed. This overall score can
then be compared with a benchmark score, provided by the regulator and/or their
stakeholder group, to assess the level of risk.
2.2.2.1 Gretener method

During the 1960s, the Association of Cantonal Fire Insurance Houses, the
umbrella organisation of public legal insurance in Switzerland, felt the need for
assessing objectively, on a standard basis, the fire risk and necessary protection
measures particularly for industrial and other special types of buildings whose
construction and purpose necessitate more than the normal protective measures.
The initiative for developing a standard method for this purpose was taken up by
M. Gretener, Head of the Fire Prevention Service for Industry and Commerce in
Zurich. He presented the first comprehensive paper dealing with this problem
at the third International Fire Protection Seminar of the Association for the
Promotion of German Fire Protection held in Eindhoven – see Gretener (1968).
The basic features of this method of fire risk evaluation are as follows.
Gretener proposed the formula
P⋅A
R= (2.4)
N ⋅S⋅F
where R is a numerical measure of fire risk. The parameter P quantifies the
inherent or potential hazard or risk of the building and includes factors such as fire
load, combustibility, size of fire area with or without fire venting system, building
height and tendency to produce smoke, toxic or corrosive gases. The parameter
A is the activation factor representing the tendency for a fire to start and may be
quantified by the probability of a fire starting.
The parameter N refers to ‘normal’ fire precautions such as fire brigade, fire
extinguishers, trained personnel, hydrants and water supplies. The parameter S is
concerned with the existence of ‘special’ protective measures such as automatic
fire detection systems and sprinklers and F with the fire resistance of the building.
The parameters in the numerator of Equation (2.4) are factors enhancing fire risk
and those in the denominator are factors reducing the risk.
In the estimation of R, the component excluding A denoted by
P
B= (2.5)
N ⋅S⋅F
is the ‘effective’ or ‘endangering’ fire risk, representing the extent of spread or
damage if a fire occurs. For each type of building, a certain ‘normal’ or ‘maximum’
value, Bmax, is specified for the risk factor B, taking into account insurance
requirements and factors such as the distance of the nearest fire station and the
average times taken by the fire brigade to attend and extinguish fires. Depending
on the potential fire risk represented by the risk component P under given
conditions, fire protection requirements according to the components N, S and F
are evaluated such that the value of B does not exceed Bmax. An advantage of this
method is that the planner of a building project is not strictly tied to specifically
prescribed protective measures but a range of possible measures which can ensure
that the risk does not exceed Bmax A disadvantage of this method is that the basic
equations are dimensionally meaningless.
Risk to life depends primarily on the number of people in a building. It also
depends on the physical and psychological conditions of the occupants of the
building. For example, life risk is greater in an old people’s home than in a
residential building. Hence, a slightly smaller value is specified for Bmax where
there is a greater risk to human life. The value of Bmax is also adjusted depending
on the presence or absence of factors affecting the parameter A quantifying the
probability of fire starting.
Each of the risk parameters P, N, S and F are evaluated by enumerating the
factors affecting the parameter and assigning points to these factors. These points
are then multiplied to provide the overall risk value for the parameter. For example,
as mentioned earlier, the potential fire risk quantified by P depends on factors such
as fire load, combustibility, smoke factor, corrosion factor, shape, size and height
of a fire compartment, presence or absence of heat and smoke ventilation systems
and height of the building or number of storeys. The parameter S depends on the
existence of protection measures such as guard service, automatic fire detection
systems and sprinklers.
The points to be assigned to the factors are specified in tables developed for
the parameters P, N, S and F. Appendix A2.1 is an example showing the values
of S suggested by the European Insurance Committee, CEA. The value of N
is set equal to unity for a building with ‘normal’ or ‘standard’ fire precautions
without any ‘extra’ or ‘special’ protection measures. Tables and methods for
calculating P, N, S and F are discussed in details by Bürgi (1973), Kaiser (1980)
and Wessels (1980). Modified points schemes have been suggested by Purt
(1971) and Cluzel and Sarrat (1979); in the latter method, different values, one
for people safety and another for property protection, are allocated to evaluate
the performances of protection measures. As mentioned above, with some
modification to suit local conditions, the Gretener method has been applied to
fire protection and insurance problems in Switzerland, Austria, Italy, France,
Belgium and the Netherlands. A number of commercially available computer
models of the Gretener method have been developed in the last decade and
these include FREM (Watts 1995a), RiskPro (2000), FRAME (2000) and Risk
Design (FSD 2000).
2.2.2.2 Fire safety in hospitals – United Kingdom

A more complex points scheme is the fire risk assessment method for hospitals
developed by the University of Edinburgh during the 1980s under the sponsorship
of the Department of Health and Social Services – see Stollard (1984). The Fire
Precautions Act 1971 – A Guide to Fire Precautions in Hospitals (1976/79)
provided the basis for identifying 17 factors contributing to the ‘norm’ against
which assessment can be made. Based on a ‘Delphi’ exercise involving fire safety
experts, three more factors were added – staff, patients and visitors, fire brigade –
20 factors in all.
The Delphi group considered a hierarchy of fire safety in order to determine the
relative importance of the 20 components of fire safety identified. This exercise
estimated contributory values of objectives to policy, of tactics to objectives and
of the 20 components to tactics. By the multiplication of these matrices, a vector
for the contribution of components to overall policy was produced. Each number
in this vector was then expressed as a decimal fraction of the whole to give a set of
relative values for the contribution of components. This vector was then modified
to give a new set of values of component contribution which took into account the
interactions between components evaluated in a matrix form. The 20 components
and their relative values are given in Column 1 of the table in Appendix A2.2.
Revised values of the components after considering the interactions are given in
Column 2 of this table.
The patient areas of a hospital were divided into fire zones or survey volumes.
Each zone was then assessed by a surveyor to assess the deficiency in each of the
20 components, in comparison with the ‘ideal’ or ‘norm’, on a six-point scale 0–5
with 0 representing 100 per cent deficiency and 5 indicating no deficiency. Work
sheets were used for this assessment. The grade for each component was then
multiplied by the relative (decimal fraction) contribution of that component to
the overall fire safety policy to calculate a single score between 0 and 500. The
scores for the components were then added. A total score between 450 to 500
was considered as good, 350 to 450 as acceptable, 280 to 350 as unacceptable
and less than 280 as definitely unacceptable. An example of the summary sheet
is given in Appendix A2.3. The method was superceded in 1994, because the
arbitrary interaction between factors could lead to an acceptable risk score and an
inadequate fire strategy (see section 1.4.6.1).
The hierarchical approach described above was further developed at the
University of Ulster for application to dwellings – see Shields and Silcock (1986)
and Donegan, Shields and Silcock (1989). This approach has been refined and
implemented for the assessment of fire risk in telecommunication facilities – see
Budnick, Kushler and Watts (1993).
2.2.2.3 Risk assessment schedules

Schedule rating is a risk assessment method similar to a points scheme developed
in the USA by the insurance industry for calculating fire insurance premium rates,
particularly for industrial and commercial properties. For these properties, unlike
occupancies such as dwellings for which class rates are determined, a premium
rate is specifically calculated for each property by the application of a schedule
or formula. The specific rate is designed to measure the relative quantity of fire
hazard present in the property. In the past, many insurance companies in the
United Kingdom adopted a tariff system developed by the Fire Offices’ Committee
which specified premium rates for different classes within an occupancy type,
based on fire resistance and fire protection measures such as sprinklers.
Based on various physical hazards, the schedules establish an arbitrary point
from which insurance premium rates are built up. A schedule of additions and
reductions is computed for a particular property and the difference applied to the
arbitrary point. Schedule rating provides a plan for quantifying the fire hazards in
any particular property; it is an empirical standard for the measurement of relative
quantity of fire risk. Schedule rating takes into consideration various factors such
as construction and occupancy with a view to determining which features either
enhance or reduce the probability of loss.
Examples of the schedules developed in the USA are the Mercantile
Schedule, the Dean Schedule and the Specific Commercial Property Evaluation
Schedule (SCOPES) of the Insurance Services Office (ISO). SCOPES is the
most commonly used insurance rating schedule according to which a percentage
occupancy charge is determined for a building from tabulated charges for classes
of occupancy, modified by factors reflecting the specific fire risk in the building.
The basic building grade is a function of the resistance to fire of structural walls
and floor and roof assemblies. The building fire insurance rate is the product of
occupancy charges and building grade modified by factors such as exposure to fire
in nearby buildings and protection provided by portable fire extinguishers, fire
alarm systems, sprinklers etc. Tabulated values and conversion factors are based
on actuarial analyses of fire losses (claims) paid by insurers and reported to the
insurance industry.
The Fire Safety Evaluation System (FSES) is a schedule approach, developed
in the USA in the late 1970s to determine equivalencies to the National Fire
Protection Association (NFPA) 101 Life Safety Code for certain institutional
occupancies. The technique was developed at the Centre for Fire Research at
the National Bureau of Standards (now National Institute of Standards and
Technology) in co-operation with the US Department of Health and Human
Services. It was adapted to new editions of the Life Safety Code and published
in NFPA 101A: Alternative Approaches to Life Safety (1995). The FSES was
developed to provide a uniform method of evaluating health care facilities to
determine what fire safety measures would provide a level of fire safety equivalent
to that provided by the NFPA Life Safety Code.
The FSES subdivides a building into ‘fire zones’ for evaluation. A fire zone is
defined as a space separated from other parts of the building by floors, fire barriers
or smoke barriers. When a floor is not partitioned by fire or smoke barriers, the
entire floor is the fire zone. In application, every zone in the facility should be
evaluated.
The FSES begins with a determination of occupancy risk factor, calculated
as the product of the assigned values for the risk parameters of the following
five characteristics: patient mobility, patient density, fire zone location, ratio of
patients to attendants and average patient age. Variations of these parameters
have been assigned relative weights, as indicated in Table 3–1 of the worksheets
for health care occupancies, in Chapter 3 of NFPA 101A. A hardship adjustment
for existing buildings is applied to the occupancy risk factor; this modifies the risk
in existing buildings to 60 per cent of that for an equivalent new building.
Thirteen fire safety parameters were selected for offsetting calculated occupancy
risk by safety features. These parameters and their respective ranges of values are
reproduced from Table 3–4 in Chapter 3 of NFPA 101A. This table is designed to
be used for selecting appropriate values for each safety parameter by carrying out
a survey and inspecting the fire zone.
An important concept of the FSES is redundancy through simultaneous use
of alternative safety strategies. The purpose is to ensure that failure of a single
protection device or system will not result in a major fire loss. Three fire safety
strategies are identified: containment, extinguishment and people movement.
Table 3–5 in Chapter 3 of NFPA 101A indicates which fire safety parameters
apply to each fire safety strategy. The limited value of automatic sprinklers for
people movement safety is adjusted for by using one-half of the parameter value
in this column.
Following the procedure described above, the FSES determines if the fire zone
considered possesses a level of fire safety equivalent to that of the NFPA Life
Safety Code. This is done by comparing the calculated level for each fire safety
strategy with stated mandatory minimum requirements. The FSES and other fire
risk ranking methods have been discussed in detail by Watts (1995b).
2.2.2.4 Points schemes for road tunnels

An example of this is the application of the ADAC (2002) tunnel fire points
scheme to a Victorian road tunnel in the UK. The ADAC points scheme is based
on a particular national standard for new road tunnels and awards the majority of
its points for ventilation systems and means of escape.
Physical constraints meant that the Victorian road tunnel could not score
highly in these categories and so received a low score (i.e. implying a high, if
not unacceptable, risk). However, the tunnel operators had implemented a range
of alternative arrangements which meant that the tunnel was much safer than
implied by the points scheme. These measures were not recognised in the points
scheme and so made no difference to the score achieved by the tunnel even
though it could be seen from historical data that they would make a significant
contribution to the reduction of risk.
In some situations, such as existing hospitals, it has been found in hindsight
that the application of the simple rules is not the most cost-effective approach.
One notable example of this was a small urban Victorian hospital where the ward
corridor was turned into a type of hospital street (with significant implications in
terms of compartmentation, fire dampers and stopping) which required a massive
investment with (given other fire precautions in the hospital) relatively little
improvement in safety.
Points schemes remain popular with those looking for a quick method to apply
to a large number of buildings. This is particularly true when their originators are
not aware of points schemes’ limitations or the availability of equally easy to use
alternative methods that are more technically sound.
2.2.2.5 Merits and demerits of points schemes

As discussed by Ramachandran (1982), points schemes follow simple numerical
approaches and are found easy to apply in practical problems concerned with the
evaluation of fire risk and effectiveness of fire safety measures. They are useful
for comparing fire risk in a building against a ‘norm’ or prescribed ‘standard’ and
identifying components of fire safety which are deficient in the building. Points
schemes allow a certain amount of flexibility in building design and in providing
comparisons between premises which are not basically dissimilar in general
occupancy and use. The latter is of assistance in making decisions on the allocation
of scarce resources for fire safety improvement.
Though based on practice, experience and expert opinion, points schemes are
empirical methods with somewhat arbitrary or subjective allocation of points,
scores or ranks. It might be argued that statistical information provided by real
fires and case studies is consciously present in the minds of those who contribute to
the allocation. But there is room for argument and serious disagreement between
people in the determination of points or ratings for different factors enhancing
or reducing fire risk. In the points schemes, equal weights are generally attached
to all the factors when adding or multiplying the points allocated to the factors
for evaluating the overall numerical measure for fire risk in the whole or part of
a building. This is not a satisfactory procedure. Points schemes do not consider
sufficiently the interactions between fire safety measures.
The numbers generated by a points scheme are subjective qualitative indicators
of relative fire risk. They are not objective quantitative estimates of absolute fire
risk and effectiveness of fire protection measures. Hence, points schemes cannot
provide quantitative inputs required for any cost benefit analysis of fire protection
measures. With a statistical/probabilistic analysis based on factual data provided
by real fires, case studies and experiments, it is possible to trace back from the final
risk figure the independent quantitative contributions from different factors and
their interactions. Scores or ranks can be easily derived from such an analysis in
order to facilitate the presentation and application of results.
Points schemes are, therefore, best suited to specific building types because they
can be judged on the same criteria and then compared with each other (rather like
testing in schools of pupils at specific ages; a benchmarking exercise).
Four limitations of points schemes should be borne in mind. They:
• are based on scoring and weighting systems that may be largely arbitrary;
• contain an implicit fire strategy, so valid alternative fire strategies may not
achieve the appropriate score;
• may be applied subjectively which may have an impact if comparison is to
be made over several regions with different assessors, i.e. one assessor may
generally be more generous or strict in assessing certain components;
• have no structure between factors which means that inappropriate fire strategies
may achieve a good score whilst the actual level of risk is not acceptable. The
healthcare example of this was that a ward, with no compartmentation and
excellent emergency lighting, exit signs and fire notices, would achieve an
acceptable score. When the fire safety of patients depends on protection by
compartmentation before progressive horizontal evacuation, an acceptable
score is dangerously inappropriate.
For fire safety, the main objective is usually life safety, but other objectives
such as property protection/business continuity, heritage, image/reputation and
the environment may also be important to organisations. In many respects the use
of points scheme risk assessment methods suppresses the decision-making aspect
of fire safety and can encourage the perception that there is only one solution
and ‘this is it’. However, the nature of many modern buildings means that points
scheme fire risk assessment methods are increasingly difficult to apply. The literal
application of the implicit solution in specific buildings may not be the most cost/
risk effective and so can have significant cost implications.
Due to the limitations of points schemes, the original version of HTM 86
(1987) for fire risk assessment of hospitals was replaced by a structured qualitative
approach (see Section 2.1.2).
For fire safety engineering, conditions where points scheme fire risk assessment
methods may not adequately address the issues of concern tend to occur when
the building is large, complex or unique, or when application of the simple risk
assessment solutions conflict with the function of the building (usually rendering
the fire precaution highly unreliable).
Appendix A2.1 Relative S values for active systems
Active System S
FIRE DETECTION
Guard service
Control rounds minimum twice per night 1.03
Control rounds every two hours 1.05
Alarm boxes (manual) 1.04
Automatic fire detection
Detectors in relation with public fire service E1 1.20
E2 1.15
E3 1.11
E4 1.07
E5 1.04
E6 1.02
Automatic extinction installation
Detection value 1.10
ALARM TRANSMISSION
Central station in building 1.03
Automatic transmission 1.04
Simultaneous alarm 1.05
Security line 1.03
INHOUSE FIRE FIGHTING
Fire-fighting teams 1.05
Voluntary fire service 1.17
Professional fire service 1.28
AUTOMATIC EXTINCTION
Sprinkler installation, class I, in relation with public fire service E1 2.40
E2 2.40
E3 2.39
E4 2.38
E5 2.36
E6 2.33
Sprinkler installation, class II, in relation with public fire service E1 1.62
E2 1.61
E3 1.60
E4 1.59
E5 1.57
E6 1.55
CO2 installation not yet
Halon Installation determined
FIRE VENTILATION
PUBLIC FIRE SERVICE
The efficiency of automatic detection and extinction is E Time Distance
considered to be a function of the operation of the public (min) (km)
fire service. S-values have therefore been determined in
relation to the “E-factor” for the public fire service 1 ≤10 ≤1
2 >10–15 >1–3
3 >15–20 >3–6
4 >20–30 >6–10
5 >30–40 >10–15
6 >40 >15
Appendix A2.2 Values of components
1 2
01 Staff 0.0866 0.0889
02 Patients and 0.0646 0.0643
Visitors
03 Factors affecting 0.0586 0.0656
smoke movement
04 Protected areas 0.0565 0.0555
05 Ducts, shafts and 0.0443 0.0400
cavities
06 Hazard protection 0.0676 0.0649
07 Interior finish 0.0500 0.0497
08 Furnishings 0.0592 0.0625
09 Access to protected 0.0448 0.0407
areas
10 Direct external 0.0436 0.0412
egress
11 Travel distance 0.0478 0.0488
12 Staircases 0.0509 0.0488
13 Corridors 0.0511 0.0509
14 Lifts 0.0356 0.0342
15 Communications 0.0487 0.0506
systems
16 Signs and Fire 0.0401 0.0406
Notices
17 Manual firefighting 0.0328 0.0302
equipment
18 Escape lighting 0.0411 0.0462
19 Automatic 0.0316 0.0329
suppression
20 Fire Brigade 0.0445 0.0435
1.0000 1.0000
Appendix A2.3 Sample of a summary sheet
Health authority:
Building: Victoria Hospital
Survey volume: Ward 1
Date of survey:
Surveyor:
Number of bedspaces: 30
Component Grade Percentage Score

Contribution
01 Staff 012345 X 9 27
02 Patients and Visitors 012345 X 6 12
03 Factors affecting smoke movement 012345 X 7 21
04 Protected areas 012345 X 6 30
05 Ducts, shafts and cavities 012345 X 4 16
06 Hazard protection 012345 X 7 35
07 Interior finish 012345 X 5 15
08 Furnishings 012345 X 6 18
09 Access to protected areas 012345 X 4 16
10 Direct external egress 012345 X 4 16
11 Travel distance 012345 X 5 15
12 Staircases 012345 X 5 25
13 Corridors 012345 X 5 25
14 Lifts 012345 X 3 15
15 Communications systems 012345 X 5 10
16 Signs and Fire Notices 012345 X 4 12
17 Manual firefighting equipment 012345 X 3 12
18 Escape lighting 012345 X 5 10
19 Automatic suppression 012345 X 3 0
20 Fire Brigade 012345 X 4 20
TOTAL SCORE (out of 500) 350
Additional comments
References
ADAC (2002), Personal correspondence.
BS 5588: Part 11 (1997), Code of Practice for Shops, Offices, Industrial, Storage and Other
Similar Buildings, British Standards Institute, London.
BS 9999 (2008), Code of Practice for Fire Safety in the Design, Management and Use of
Buildings, British Standards Institute, London.
BS EN 13501-1 (2000), Reaction to Fire Testing, British Standards Institute, London.
Budnick, E K, Kushler, B D and Watts, J M, Jr (1993), Fire risk assessment: a systematic
approach for telecommunication facilities, Annual Conference on Fire Research: Book of
Abstracts, NISTIR 5280, National Institute of Standards and Technology, Gaithersburg,
MD.
Bürgi, K (1973), A method of evaluating fire risks and protective measures, Fourth
International Fire Protection Seminar on Methods of Evaluating Fire Hazards of
Industrial and Other Objects, Zurich.
Chemical Industry Safety and Health Council, (1977), A Guide to Hazard and Operability
Studies, Chemical Industry Safety and Health Council, London.
Cluzel, D and Sarrat, P (1979), Méthode ERIC: Evaluation du risque incendie par le calcul.
Proceedings of Conseil International du Bâtiment Symposium on Systems Approach to Fire
Safety in Buildings, Vol I, II, CIB, Rotterdam.
Donegan, H A, Shields, T J and Silcock, G W (1989), A mathematical strategy to relate
fire safety evaluation and fire safety policy formulation for buildings, Fire Safety Science
– Proceedings of the Second International Symposium, Hemisphere Publishing Corporation,
New York.
Drysdale D (2002), An Introduction to Fire Dynamics, 2nd edn, John Wiley and Sons,
Chichester.
FRAME (2000), Fire Risk Assessment Method for Engineering, http://user.online.
be.notr.034926/webengels.doc.htm, 7 September.
FSD (2000) Risk Design, http://www.fsd.se/eng/index.html, 7 September.
Gretener, M (1968), Attempt to calculate the fire risk of industrial and other objects, Third
International Fire Protection Symposium, Eindhoven.
HTM 86 (1987), Fire Risk Assessment in Hospitals, Department of Health, HMSO, London.
HTM 86 (1994), Fire Risk Assessment in Hospitals, FIRECODE, NHS Estates, HMSO,
London.
Kaiser, J (1980), Experiences of the Gretener method. Fire Safety Journal, 2, 213–222.
Mannan, S (ed.) (1980), Lees’ Loss Prevention in the Process Industry, Butterworth, Oxford.
NFPA 101A (1995), Alternative Approaches to Life Safety, National Fire Protection
Association, Quincy, MA.
Purt, G (1971), The evaluation of fire risk as basis for the planning of automatic fire
protection systems, Sixth International Seminar for Automatic Fire Detection, IENT,
Aachen.
Ramachandran, G (1982), A review of mathematical models for assessing fire risk, Fire
Prevention, 149, May, 28–32.
RiskPro (2000), Simco Consulting, Wantirna, Australia.
Shields, T J and Silcock, G W (1986), An application of the hierarchical approach to fire
safety, Fire Safety Journal, 2, 3, 235–242.
Stollard, P (1984), The development of a points scheme to assess fire safety in hospitals,
Fire Safety Journal, 7, 2, 145–153.
TSO (2005)The Regulatory Reform (Fire Safety) Order, The Stationery Office, London
Watts, J M, Jr, (1995a), ‘Software review: fire risk evaluation model’, Fire Technology, 31, 4,
369–371.
Watts, J M, Jr (1995b), Fire risk ranking, SFPE Handbook of Fire Protection Engineering,
2nd Edition, Section 5, Chapter 2, National Fire Protection Association, Quincy, MA.
Wessels, E C (1980), Rating techniques for risk assessment, Fire International, 67, 80–89.
3 Quantitative risk assessment
techniques
Due to uncertainties caused by several factors affecting fire safety in buildings, it

may be realistic to treat these factors as non-deterministic random phenomena.
This generally means adopting a probabilistic approach to the evaluation of fire risk
and assessment of the fire protection requirements of a building (Ramachandran
1988a). In this approach, there are essentially four types of models that are used
in the quantitative analysis of fire risk in which probabilities enter the calculations
explicitly. These are:
• statistical methods;
• logic tree analysis;
• stochastic models; and/or
• sensitivity analysis.
These models are discussed below. Further types of analysis, or variations on

the types of analysis shown, can be used as appropriate and some data can be
found in the tables in this chapter.
The analysis of statistics provided by real fires is the basis of most probabilistic
fire risk assessment, from the frequency of ignition to the conditional probability
of failure of a fire protection system. Statistical analysis takes data that has been
collected on building fires and transforms it into information that can be used
to predict the likelihood of occurrence of future events and their consequences.
This can take the form of the simple assessment of the average probability of
an event (occurrence of fire) over a set of buildings and the damage caused
and their probability distribution over a period of time or a complex regression
analysis.
Statistical analysis has the advantages that it is based on actual events and that
the results are usually simple to apply. It is, however, based on historical data which
is then averaged. This assumes that future performance can be predicted from past
experience and that an average measure can be applied to a particular building
with some adjustment. In most cases these assumptions are reasonable and there
is less uncertainty in undertaking a risk assessment based on historical data than
to take no account of the probability of failure of the various fire precautions in a
building design.
Quantitative risk assessment techniques 71
Another limitation of statistical analysis is that it is often not possible to collect
sufficient data to directly predict, with confidence, the kind of high consequence,
low frequency events, such as multiple fatality fires, that are of concern. Statistical
data is much better for more frequent events such as ignition, and the conditional
probability of success or failure of fire precautions. These individual pieces of
information can be used to predict the frequency of low frequency events, by
using logic trees and other techniques. Alternatively, other statistical methods
have been developed to deal with such insufficient outcome data, such as extreme
value theory (Rabash et al. 2004).
For most practical problems in fire protection, it may be sufficient to carry out
a probabilistic fire risk assessment based on one or more logic trees. These provide
a simple method for estimating the frequency of occurrence of an undesirable
event (or events) known as an outcome. Such events include the fire reaching
flashover stage or spreading beyond the room of origin and smoke causing visual
obscuration on an escape route. In this approach, the sub-events leading to
the outcome are identified and placed in their sequential order. This process is
continued until a basic event (usually ignition) or set of basic events is identified,
for which the probabilities can be estimated from statistical data (DOE (1996).
Probabilities associated with sub-events are then combined in a suitable logical
manner to derive the frequency of occurrence of the outcomes of concern. The
calculation procedure is facilitated by the use of logic diagrams or trees, which
provide a graphical representation of a sequence of sub-events.
There are normally two types of logic trees used in a probabilistic risk assessment,
event trees and fault trees, and these are described in Section 3.2. To understand
these methods it is first important to understand probability theory.
3.1 Theoretical basis
3.1.1 Probability theory

Sections 3.3 to 3.5 introduce statistical analysis, probability distributions and
other associated techniques such as regression analysis, Bayes’ theorem or Monte
Carlo analysis. This section looks at set theory and Boolean algebra – the basis of
logic trees which are used to quantify the probability of events.
An experiment is any process where an observation is made. Experiments
comprise three parts: the conditions, performance and outcome. If, under identical
conditions, the performance of an experiment produces the same outcome then the
process is said to be deterministic. If, under identical conditions, the performance
of an experiment can produce any one of a finite or infinite number of different
outcomes then the process is said to be random.
Probability theory is used to analyse random processes. For example, if an
experiment can result in N different outcomes and there is no reason to suppose
that any one of them should be favoured more than any other, then if nE of them
possess the property E the probability of E is given by Johnston and Matthews
(1982):
n
P (E ) = E (3.1)
N
This concept of probability was developed through assessing chances of winning
at gambling and can be assigned before the experiment so there is often no need to
perform the experiment at all.
If an experiment is repeated a large number of times under exactly the same
conditions, the relative frequency with which an event E occurs settles down
to a constant limiting value P(E) which is the probability of E. The stability of
relative frequency in the long run brings regularity into the uncertainty involved
in random experiments. This regularity may be described by a probability model
– see Table 3.1.
In real studies, interest is rarely restricted to a single possible outcome, but
more usually to a group of them. Such a composite result is represented in the
model by a compound event comprising the set of corresponding simple events.
For an event set ε containing simple events (e1, e2, e3, …, en) the probability of
a compound event E contained in ε is the sum of the probabilities of the simple
events in E. Therefore:
0 ≤ P (E ) ≤ 1 (3.2)
If E and F are events in ε then:
1 E and F are mutually exclusive if they contain no simple events in common.

2 E implies F if E is wholly contained in F.
3 If the outcome of an experiment is a simple event in either E or F or both (i.e.
at least one occurs), this can be expressed as E ∪ F.
4 If the outcome of an experiment is a simple event in both E and F (they
happen simultaneously), this can be expressed as E ∩ F.
5 The set containing all simple events not in E is the complement of E and is
written E .
Probabilities can be added together:

If E and F are mutually exclusive events in ε then:
P (E ∪ F ) = P (E ) + P (F ) (3.3)
Table 3.1 Relationship between random processes and probability modelling

Real life Model
Random experiment Probability model
Any possible outcome Simple event
List of possible outcomes Event set
Long–run relative frequency Probability
If E and F are any two events in ε then:
P (E ∪ F ) = P (E ) + P (F ) − P (E ∩ F ) (3.4)
Since E and E are mutually exclusive, then:
P (E ) + P (E ) =1 (3.5)
Conditional probability is an important concept in the analysis of fire risk

and is described as follows. The probability of an event E given that an event F is
known to have happened is the conditional probability of E given F, defined as:
P (E ∩ F )
P (E F ) = (3.6)
P (F )
E and F are independent if P(E/F) = P(E), that is, the occurrence of E is in no
way affected by the occurrence of F.
If E and F are mutually exclusive then they are totally dependent, since
knowledge that F has happened tells us for certain that E has not happened, so
P(E/F) = 0 which is not the same as the unconditional P(E).
Only events that can happen simultaneously are in contention for being
independent.
Probabilities can also be multiplied:
If E and F are independent events in ε then:
P(E ∩ F ) = P(E ).P(F ) (3.7)
If E and F are any events in ε then:
P(E ∩ F ) = P(E ).P(F E ) = P (F ).P (E F ) (3.8)
In view of the two equations above, the addition rule becomes:
P(E ∪ F ) = P (E ) + P (F ) − P (E ).P (F E ) (3.9)
or if E and F are independent:
P (E ∪ F ) = P (E ) + P (F ) − P(E ).P(F ) (3.10)
Both the addition and product rules readily extend to more than two events.
The application of probability theory becomes more apparent in the sections on
fault and event trees.
When all the simple events of an event set ε have the same probability, for
example when considering a random sample of items from a batch, the calculation
of probabilities of compound events in ε reduces to counting how many simple
events there are in the compound event and in ε. Therefore, knowledge of
formulae for permutations and combinations is desirable.
The factorial expression n! is used to mean
1 × 2 × 3 × … (n – 3) × (n – 2) × (n – 1) × n.
Therefore, 5! = 1 × 2 × 3 × 4 × 5 = 120, 1! = 1 and for convenience we

define 0! = 1 also.
Permutations are the number of ways in which r items can be selected from n
distinct items, taking notice of the order of selection and is given by:
n!
n
Pr = (3.11)
( n − r )!
For example, the number of ways in which 2 items can be selected from 4
distinct items, taking notice of the order of selection, is:
4!
4
P2 = = 4 × 3 = 12
(3.12)
2!
Combinations are when the order of selection is not important but only the
final content of the r items selected. Therefore, the different permutations which
contain the same r items need only be counted as 1 combination of the n items
taken r at a time. The number of different combinations of n items taken r at a
time is:
n!
n
Cr = (3.13)
( n − r )! r !
For example, using Equation (3.13) the number of ways 6 numbers can be
drawn from 49 in a lottery is:
49 49 !
C6 = = 1.4 ×107
( 49 − 6 )! 6 !
or a one in 14 million chance of success with one ticket.
3.1.2 Set theory and Boolean algebra
3.1.2.1 Set theory

Set theory is an elementary and intuitive concept that is best expressed by way of
lists, that is a list of all the elements of a set. For, example:
A = (1, 2, 3, 4, 5, 6, 7, 8, 9)
B = (2, 4, 6, 8, 10, 12)
C = (2, 4, 6, 8)
D = (even numbers)
Sets are normally denoted by upper case letters. The other way of expressing a
set is by a Venn diagram (see Figure 3.1).
Two sets are equal if they contain the same elements. For example:
(Football, Hockey, Chess) = (Hockey, Chess, Football)
The membership of a set can be expressed for the example in Figure 3.1:
1 is a member of set A, but 1 is not a member of set B.
Also for the example in Figure 3.1:
C is a subset of B because all of the elements of C are also elements of B.
There are two special sets that need to be considered:
1 The set that contains no elements at all is call the null set and is denoted by
0.
2 The set that contains all potential objects under consideration is called the
universal set and is denoted by 1.
The universal set is often only implicitly used in set theoretical calculations and
manipulations and it enables the identification of:
All the elements that do not belong to set A is the complement of A.
D
9 A
1 12
B
3
7 6
2
5 8 10
4 C
Figure 3.1 Venn diagram

Sets can also be defined in terms of other sets. For example, a set C could be
defined as the set consisting of all of the elements in a set A together with all of
the elements in a set B and contain no other elements:
If A = (1, 2, 3)
and B = (1, 4, 7)
then C = (1, 2, 3, 4, 7)
Note that elements will only appear once in the list of elements in a set. The set
C is defined as the union of sets A and B (Figure 3.2) and is expressed as:
C = A + B (3.14)
This is the notation commonly used in switching theory and is the one adopted
by most fault tree analysts (there are two other notations used in set theory).
Similarly, we can define C as the set that consists of all the elements that
appear in both A and B and contains no other elements:
If A = (1, 2, 3)
and B = (1, 4, 7)
then C = (1)
2
7
1
3 B
Figure 3.2 Venn diagram of union set

C
2
7
1
3 B
Figure 3.3 Venn diagram of intersection set
The set C is defined as the intersection of sets A and B (Figure 3.3) and is
expressed as:
C = A . B (3.15)
This notation is used in fault tree analysis because union and intersection in
sets have many of the same properties as addition and multiplication of numbers.
3.1.2.2 Algebra of events

An algebra of events is used to link probability theory (associated with the
outcomes of experiments/trials) with events. For example, if the trial is the throw
of a die, then the event A1 can be defined:
A1 = (lands on 1)
Similarly:
A2 = (lands on 2) etc.
The universal set in this algebra is the set/event that encompasses all of the
possible outcomes of the experiment in question. It is often called the event space.
The null set in this algebra corresponds to an outcome that cannot occur. So, for
the above die example, the sample space is the set:
U = (lands on 1, lands on 2, lands on 3, lands on 4, lands on 5, lands on 6)
and an example of the null set would be:

0 = (lands on 7)
If the experiment is performed and the event A occurs, then we can express this
as:
A=1
Conversely, if A does not occur this can be expressed as:
A = 0 (or not A equals one; A = 1)
A probability can be associated with each potential outcome. This is expressed as:
P (A = 1)
that is, the probability that outcome A will be the result of the experiment.
Combinations of events can be described in the same way as the combination
of sets. Therefore, A + B is the compound event ‘the occurrence of event A or
event B or both’. The truth table shown in Table 3.2 can be used to represent the
outcome. This is equivalent to an OR gate in an event tree.
Similarly, A . B can be defined as the compound event ‘the occurrence of both
event A and event B’. The truth table for this is shown in Table 3.3.
Fault trees contain two classes of events. Firstly, those events which form
termination points in the fault tree and generally will consist of compound failures;
these events are called basic or base events. Secondly, each gate of the fault tree
can be associated with a combined event. The system failure for which the fault
tree was developed is called the top event.
Table 3.2 Truth table for an OR gate

A B A+B
0 0 0
1 0 1
0 1 1
1 1 1
Table 3.3 Truth table for an AND gate

A B A.B
0 0 0
1 0 0
0 1 0
1 1 1
For example in Figure 3.4:
G, X, Y are base events

G1, G2 TOP are combined events where:
G1 = G + X
G2 = G + Y
TOP = G1 . G2
Therefore:
• an AND gate corresponds to a Boolean product;

• an OR gate corresponds to a Boolean sum; and
• an EXCLUSIVE OR gate corresponds to a combination of Boolean sums,
products and complements but can usually be approximated by Boolean sum.
Example of its use in fault tree analysis
The main interest in fault tree analysis is to gain qualitative and quantitative
information about the top event. To do this the Boolean expression for the top
event must be manipulated into a form where:
1 all of the variables are base events;

2 the expression is in a standard format.
Top event
AND
G1 G2
OR OR
G X G Y
Figure 3.4 An example of a fault tree

One way of expressing this for a top event is:
Top = C1 + C2 + C 3 ...C n (3.16)
where C1 is a combination event formed by the product of base events:
C1 = A1 . A2 . A 3 ... A n (3.17)
where A1 is a base event and n ≥ 1.

These equations express the condition that if all the base events of a combination
event exist simultaneously, then the combination event and so the top event will
be produced. Note only one combination event is required to exist for the top
event to be produced. These equations are known as the ‘sum of product’ form.
Boolean manipulation
In many ways, union and intersection operations behave in a similar way to

addition and multiplication. These similarities can be seen below:
• The order the operations are performed is:
before + unless overridden by a bracket
For example, in A + B . C the operation order is to first evaluate B .C and call

the result say X. Then evaluate A + X. However, if the expression was written
(A + B) . C then the operation order would be to evaluate (A + B) first and
call the result say Y and then evaluate Y . C.
• Both operations are independent of the order of the variables:
A + B = B + A A . B = B . A (3.18)
• Both operations are independent of order within themselves:
(A + B) + C = A + (B + C) (A . B) . C = A . (B . C) (3.19)
• Both operations obey the laws of multiplication:
(A + B) . (C + D) = A . C + A . D + B . C + B . D (3.20)
• Both operations interact with binary values:
A + 0 = A A . 1 = A (3.21)
There are also several areas where Boolean algebra varies from normal
arithmetic:
• Complementary events:
For any event A there exists a complementary event, A such that:
A + A =1 (either A or A must occur) (3.22)
A ⋅ A = 0 (A and A cannot co-exist) (3.23)
Complementation should occur before . and + operations.
• Idempotent laws
A + A = A A . A = A (3.24)
• Absorption law
A + A . B = A (3.25)
This law can be proved using truth tables or Venn diagrams.
If there are no complement events in an expression, the top event expression

can be manipulated into the sum of product form:
1 Examine all the parts of the expression that are not bracketed and try to
reduce their length using the idempotent and absorption laws.
2 Take parts of the expression of the form:
(A + … + N) . (M + … + Z)
and expand using multiplication. Note that A, B and so on will not generally
be base events and will generally be unbracketed expressions.
3 Repeat stages 1 and 2 until the expression cannot be reduced further.
This process is the way of analysing fault trees.

For example, a top event may be written:
Top = (G . X) + (G . Y)
= G . G + G . Y + X . G + X . Y (expand)
= G + G . Y + G . X + X . Y (G = G . G)
= G + X . Y (absorb)
Other processes of Boolean manipulation are also available. If complementary

expressions are present, these have to be modified to produce the standard form of
expression before Boolean reduction can commence:
1 Apply De Morgan’s laws of complementary combination events until all the
complementary operations act on the base events. De Morgan’s laws are:
A + E = A ⋅E (3.26)
A ⋅E = A + E (3.27)
2 Examine all the parts of the expression that are not bracketed and try to
reduce their length using the following expressions:
A+A=A=A.A
A+A.B=A
A. A =0
A+ A =1
A.B.C+ A .B= A .B+B.C
A.B+ A .B=B
3 Expand the inner most level of brackets using multiplication.
4 Repeat stages 2 and 3 until the expression cannot be reduced further.
Again further methods are available in the literature.
Minimal cut sets
One of the most important results of the analysis is a qualitative expression of all
the ways in which the top event can occur. This information is easily obtained
from the sum of products form of the expression for the top event. For example:
Top = G + X . Y
So if G occurs, the top event results. Equally, if X and Y occur, the top event
results. Conversely, if X occurs on its own or Y occurs on its own, the top event does
not result. This can be seen as an expression of redundancy or defence in depth.
Therefore:
• A cut set can be defined as a combination of base events, the simultaneous

existence of which will produce the top event.
• A minimal cut set can be defined as a cut set that has no other cut set as a
proper subset.
So for the above example:
G
GX
GY
GXY
XY
are all cut sets but only:
G
XY
are minimal.
For fault tree analysis of a top event, Boolean expressions can be manipulated
to the sum of products form. Where NOT logic is involved, the sum of product
form can be obtained where:
1 each product is a minimal cut set;

2 all of the minimal cut sets are found as products.
Johnston and Matthews (1982) showed that in other expressions where NOT
logic is present, other forms of analysis are necessary. Normally, fault tree analysis
software is used to produce the minimal cut sets. However, knowledge of set
theory and Boolean algebra is essential to the understanding and construction of
fault trees.
3.1.3 Reliability and availability of systems

Most fire protection and detection systems are installed because they are needed
to satisfy the Building Regulations or at the request of an insurance company
covering the risk. The deterministic approach to fire safety engineering assumes
that the installed system will work on the day. Deterministic fire safety engineering
does not quantitatively address the reliability of systems. This section considers
system reliability; it shows how reliabilities can be calculated and suggests values of
reliability for different systems and hazards. For completeness, a brief introduction
to reliability theory is given by Finucane and Pinkney (1988).
3.1.3.1 Reliability
Reliability is a measure of the ability of an item to perform its required function in
the desired manner under all relevant conditions, and on the occasions or during
the time intervals when it is required so to perform (Green and Bourne 1972).
Reliability is normally expressed as a probability. For example a system that fails
randomly in time but once a year on average will have a probability of failing (PF)
in any one particular month of a 1/12, i.e. PF = 0.0833. Conversely the probability
of success (PS), i.e. not failing during that particular month, is 11/12 = 0.9167
which is the same as 1 – PF, i.e. PS = 1 – PF and by transposition PF = 1 – PS.
Mathematically these expressions can be expressed as:
PS = e –t/T and PF = 1 – e–t/T (3.28)
where t = the time interval during which success is required and

T = mean time between failures.
For values where t/T = 0.1 or less, then PF is approximately equal to t/T.
PS = 1 – PF (3.29)
PS = 1 – t/T (3.30)
T −t
Ps = (3.31)
T
The exponential probability distribution defined in Equation (3.28) is applicable
to the major middle useful period of the life of a system. Other probability
distributions such as Weibull are applicable to the early period and old age related
wear out failures at the end of a system’s life.
For example, if the mean time between failures is one year and the time interval
during which success is required is one year, then the probability of failure PF is not
actually 12/12, i.e. 1, but
PF = 1 – e –t/T where t/T = 1 (3.32)

= 1 – 0.37
= 0.63
So there is a 63 per cent chance of failure in a year and the probability of

success is given by:
PS = 1 – PF (3.33)
= 1 – 0.63
= 0.37
So there is a 37 per cent chance of not failing in any one particular year.
In practice, when considering the reliability of fire protection systems, it is easier
to talk in terms of unreliability or probability of failure (PF). Taking, for example,
the previously discussed case where the mean time between failures was one year,
PF = 0.0833 and PS = 0.9167. If the mean time between failures were improved by
a factor of 10, i.e. to 10 years, then PF changes from 0.0833 to 0.00833 but PS only
changes from 0.9167 to 0.99167. For a system where failure creates a potential
hazard, e.g. failure of a compartment wall or suppression system, the probability of
failure PF is a more direct measure of the risk involved.
3.1.3.2 Availability
Availability is the proportion of the total time that a system is performing in the
desired manner. For protection or warning systems such as a fire alarm system,
failure of the system does not, in itself, create an immediate hazard. Only if the
failure exists when a fire occurs does an unprotected hazard result.
If we take the original example of a system with a mean time between failures of
one year and assume that the fault is immediately alarmed, but it takes one week
to repair it, on average, the system is out of action for one week per year, i.e. its
unavailability is 1/52 = 0.019 and its availability is 51/52 = 0.981. Alternatively,
assume that the fault is not alarmed, but is only revealed when a comprehensive
weekly test is performed. In this case, the outage time can vary from near zero
(i.e. fault occurs immediately prior to test) to nearly one week (i.e. fault occurs
immediately after the test). The average outage will therefore be half a week.
The unavailability from this cause will therefore be 0.5/52 = 1/104 = 0.0096. It
should be noted that this is half the probability of failure PF for a similar one-week
period. The total outage time will be the sum of the two types of outage, i.e. from
immediately revealed faults and from faults only revealed at regular test intervals.
As with reliability, the unavailability is a more sensitive indicator of how well a
system performs.
Assume the original system with a one year mean time between failures is a
fire alarm system and that the total outage is, on average, one week per fault.
The unavailability will therefore be 1/52. Assume that fires occur randomly in the
protected area, again with an average mean time between fires of one year. The
probability of a fire occurring within the particular week when the equipment is
dead is therefore 1/52 per fire. Since there is only one fire per year on average,
only once in 52 years (mean time between hazard) is there likely to be a fire at the
same time as the fire alarm system is not working. In other words, the mean time
between undetected fires is the mean time between fires, divided by the fractional
dead time of the fire alarm system.
Mean time between fires

Mean time between hazards =
Unavailability of fire alarm system
1 year
= (3.34)
1 / 52
= 52 years
3.1.3.3 Factors influencing system reliability

When considering the reliability of any system, various factors have to be taken
into account. For example, the quality of the components used in the system
and their suitability for the particular application; the stress imposed on these
components by the designer; additional stresses imposed by the environment
in which the system is installed; the tolerance of the design to variations in
component performance; the test and certification procedures adopted for the
system, and the time intervals between these tests. All these factors could cause
a consequent, and possibly unacceptable, reduction in the system reliability (BS
5760). Evaluation of the reliability of a whole system based on the reliabilities of
its components is a complex statistical problem and like many others in risk and
engineering, it is not yet fully researched.
When considering reliability issues, care needs to be taken when analysing
data and interpreting the results. According to UK fire statistics, in a significant
proportion of fires, the sprinklers may not operate due to the fact that the fire is
‘small’ such that the heat generated may be insufficient for activating the sprinkler
heads. Mechanical defect and the system having been turned off are main reasons
for the non-operation of sprinklers. Although sprinklers operate in only 9 per cent
of all fires they do so in 87 per cent (= 39/45) of the cases in which their action
is required. This denotes a probability of 0.87 for sprinkler operation in ‘big’ or
‘growing’ fires. Some of the fires in which sprinklers operate are extinguished by
the system itself, and some by the fire brigade.
3.1.4 Frequency of ignition

The frequency of ignition is one of the key parameters of most probabilistic risk
assessments. It is usually the initiating event in most event trees and can be a
base event in fault trees. A more detailed summary of the statistical derivation of
frequency of ignition can be found in Section 3.3 Statistical models. The annual
probability/frequency of ignition for a building increases as a ‘fractional power’ of
the size of the building expressed in terms of total floor area.
Statistical studies (Ramachandran 1970, 1979/80, 1988a) have shown that the
frequency (or probability during a period) of ignition is approximately given by:
Fi = aA bb (3.35)
where a and b are constants for a particular type of building related to occupancy.
The parameter a includes the ratio of the number of fires in a period over the
number of buildings at risk (n/N) while b measures the increase in the value of
Fi for an increase in Ab, denoting the total floor area of the building. For a full
derivation of this power function see Section 3.3.1 and Equation (3.40).
A value of unity for b would indicate that the probability of fire starting is
directly proportional to the size of the building; this would also imply that all parts
of a building have the same risk of fire breaking out. This is not true, since different
parts have different types and numbers of ignition sources. Hence, the probability
of fire starting is not likely to increase in direct proportion to building size so that
b would be less than unity. If two buildings are considered, one twice the size of
the other, the probability for the larger building will be less than two times the
probability for the smaller building. These theoretical arguments are confirmed by
actuarial studies on frequency of insurance claims as a function of the financial
value (size) of the risk insured (Ramachandran 1979/80, Benktander 1973).
Based on fire statistics and a special survey as mentioned earlier, the Home
Office estimated the values of a and b for major groups of buildings; Rutstein
(1979) (Table 3.4). For all manufacturing industries in the UK with Ab (m2), the
values of a and b were estimated as 0.0017 and about 0.53 (respectively).
Actuarial studies (Benktander 1973) in some European countries confirm
that the value of b is about 0.5 for industrial buildings. For a particular building
Table 3.4 Frequency of ignition per year
Probability of fire per year
Occupancy a b
Industrial buildings
Food, drink and tobacco 0.0011 0.60
Chemical and allied 0.0069 0.46
Mechanical engineering and other metal goods 0.00086 0.56
Electrical engineering 0.0061 0.59
Vehicles 0.00012 0.86
Textiles 0.0075 0.35
Timber, furniture 0.00037 0.77
Paper, printing and publishing 0.000069 0.91
Other manufacturing 0.0084 0.41
All manufacturing industry 0.0017 0.53
Other occupancies
Storage 0.00067 0.5
Shops 0.000066 1.0
Offices 0.000059 0.9
Hotels etc. 0.00008 1.0
Hospitals 0.0007 0.75
Schools 0.0002 0.75
the ‘global’ value of Fi given by Equation (3.35) can be adjusted. The ratio of
number of fires over the number of buildings at risk provides an overall measure,
unadjusted for building size, of the probability of fire starting (see Table 3.5).
Using data for the years 1968 to 1970 (North 1973) , a figure of 0.092 was
estimated for all manufacturing industries in the UK for the risk of having a fire
per annum, per establishment; an establishment can have more than one building.
An estimate for the probability of fire starting according to building size is also
given by number of fires starting per unit of floor area (see Table 3.6). It should be
noted that the figures in Tables 3.4, 3.5 and 3.6 are now quite dated, but they are
the best available.
3.2 Logic trees

For most practical problems in fire protection, it may be sufficient to carry out a
probabilistic fire risk assessment based on one or more logic trees. These provide a
simple method for estimating the probability of occurrence of an undesirable event
(or events) known as an outcome. Such events include the fire reaching flashover
Table 3.5 Frequency of ignition normalised per building
Occupancy Probability of fire starts per occupancy(fires.y–1)
Industrial 4.4 × 10–2
Storage 1.3 × 10–2
Offices 6.2 × 10–3
Assembly entertainment 1.2 × 10–1
Assembly non–residential 2.0 × 10–2
Hospitals 3.0 × 10–1
Schools 4.0 × 10–2
Dwellings 3.0 × 10–3
Table 3.6 Frequency of ignition normalised by floor area

Occupancy Probability of fire starting
(starts per y–1 m–2 floor area)
Offices 1.2 × 10–5
Storage 3.3 × 10–5
Public assembly 9.7 × 10–5
stage or spreading beyond the room of origin and smoke causing visual obscuration
on an escape route. In this approach, the sub-events leading to the outcome are
identified and placed in their sequential order. This process is continued until a
basic event (usually ignition) or a set of basic events is identified for which the
probabilities can be estimated from statistical data. Probabilities associated with
sub-events are then continued in a suitable logical manner to derive the probability
of occurrence of the outcome of concern. The calculation procedure is facilitated
by the use of logic diagrams or trees which provide a graphical representation of a
sequence of s ub-events.
Normally two types of logic trees are used in a probabilistic risk assessment:
event trees and fault trees.
3.2.1 Event tree analysis

Event trees are most useful when there is little data on the frequency of outcomes
of concern (usually because they are infrequent) e.g. multiple fire deaths. Event
trees can be used to predict the frequency of infrequent events by the logical
connection of a series of much more frequent sub-events, for which data is
available.
Event trees work forward from an initiating event (such as ignition) to generate
branches defining events and paths resulting from secondary (or nodal) events to
Nodal Nodal Outcome

event 1 event 2
Frequency
PS2
F1
PS1
Initiating event F2
PF2
F
PS2
F3
Yes
PF1
F4
PF2
No
Figure 3.5 General form of an event tree
give a whole range of outcomes. Some of the outcomes may represent a very low
risk event, others may represent very high risk events.
The construction of an event tree starts by defining an initiating event leading
to the final outcome, following a series of branches, each denoting a possible
outcome of a chain of events. Figure 3.5 is an example of an event tree representing
a range of outcomes resulting from an initiating event via two nodal events. Care
should be taken that the event tree reflects the actual order of events in real fires
and that all the nodal events of importance have been included.
It is crucial that all the sub-events included in an event tree are independent.
That is, they can all occur in the same event and so are not mutually exclusive.
The frequency associated with each branch (outcome) is given by multiplying
out the initiating frequency F and the relevant conditional probabilities of success
and/or failure (PS and PF respectively). For example:
F2 = F .PS1 .PF 2 (3.36)
Figure 3.6 shows how an event tree could be applied to the early stages of a fire.
The initiating event is ignition. The two nodal events are ‘Is the fire restricted
to the item first ignited?’ and ‘Is the fire detected less than five minutes from
ignition?’ The outcomes in descending order are:
1 A fire where ignition occurs, but the fire does not grow beyond the item first
ignited.
2 A fire where it grows beyond the item first ignited, but is detected in less than
five minutes from ignition.
3 A fire where it grows beyond the item first ignited and is not detected in less
than five minutes from ignition.
Is the fire
Is the fire restricted detected in less Outcome
to the item first Frequency
than 5 minutes
ignited? from ignition
0.6 0.14
Ignition:
0.24 fires per 0.5 0.05
year
Yes
0.4
0.5 0.05
No
Figure 3.6 Event tree of the early stages of fire event development
The frequencies of the outcomes can be calculated as above and indicate that
although ignition can be expected just under once in four years, the frequency of
events where a fire would be expected to grow and not be detected is about once
in twenty years. This could be used to measure the benefit of materials that are
fire retardant and ignition sources that are low in number and energy. This event
tree could also be used to demonstrate that an alternative mode of fire detection
is equivalent to that of a code-compliant solution. Care should be taken to ensure
that the conditional probability of the first nodal event does not include events
that can only follow the second nodal event, e.g. first aid fire fighting.
3.2.1.1 Example of probabilistic fire risk assessment using an event tree

This subsection shows how event tree analysis can be used as part of a probabilistic
fire risk assessment (Charters 1992). Fire risk assessment can be used to assess the
risk/cost benefit of fire precautions for property protection. The following example
is a risk assessment carried out for a major bus operator (Charters and Smith 1992,
Charters 1998).
A major bus operator expressed concern about the risk to business from fires
in bus garages. In particular, the operator was interested in whether or not they
should install sprinkler systems in their existing bus garages or take some other
action. Obviously, the cost of this would be considerable and so the bus operator
commissioned a study to quantify the benefits in terms of property protection.
This risk assessment involves:
1 identifying events that could give rise to the outcome of concern;

2 estimating how often the events happen;
3 estimating what the severity of the outcome of those events would be; and
4 assessing the implications of the level of risk.
3.2.1.2 Identifying events
The events of concern are fires causing significant damage to vehicles and property
in bus garages. From operating experience, fire safety judgement and full-scale fire
tests, these events were narrowed down to one ‘reasonable worst case’ event: a
seat fire at three points on a double deck bus parked amongst others.
The risk parameter chosen for the study was the cost of fires per calendar year.
This could allow the bus operator to put these risks in context with historical data
on other risks.
3.2.1.3 Estimating the frequency of events

To estimate how often the fire event happens, historical data was collected on
how often fires occur on buses in garages. Because fires on buses are relatively
infrequent, there was insufficient information to estimate how often the event
occurs. Therefore, an event tree was constructed to help estimate the missing
information.
An event tree is a logic diagram, which predicts the possible outcomes from an
initial event (see Figure 3.7). For example, an initial event of ‘seat fire in the lower
saloon of a double deck bus’ may have outcomes of ‘damage less than £200,000’
and ‘damage greater than £500,000’. The likelihood of each outcome depends on
other factors such as ‘Is the fire noticed at an early stage?’, ‘Does the fire spread to
neighbouring buses?’ or ‘Is the fire put out with fire extinguishers?’
The conditional probability of each of these other factors is estimated using
historical data. Therefore, using the likelihood of an initial event and the
probabilities of the other factors, an estimate can be made of how often an event
occurs (Rutstein and Cooke 1978, Rutstein and Gilbert 1978, Charters 1997).
3.2.1.4 Estimating the severity of the outcome

There are several ways to estimate the severity of the outcome: from historical
information, using simple analytical methods, using computer models and/or using
full-scale tests. Each approach has its advantages and disadvantages. Historical
data describes what the outcomes have been in the past but may not be complete
or relevant. Simple analytical methods can predict the severity of outcomes cost
effectively but the answer is only as good as the assumptions made. Computer
models can predict the severity of outcomes more closely but can be expensive and
time consuming. Full-scale testing probably gives the most accurate assessment of
the severity of outcomes but it is usually even more costly and time consuming.
In this case the severity of the outcome (i.e. losses due to damaged buses/
garage) depended heavily on the spread of fire from bus to bus and the effective
spray density of different sprinkler systems. Therefore, a combination of full-scale
testing and computer modelling was used to predict fire growth, fire spread and
the effectiveness of sprinklers (Fardell and Kumar 1991). Typical data can be
found in Table 3.7.
Cost category
Probability
Is the fire Is the fire Does the fire Do Do the Is damage Is damage
noticed at extinguished spread to sprinklers fire brigade less than more than
the early stage using neighbouring control control the £200,000? £500,000?
of development? extinguishers? buses? the fire? fire?
A 0.56
0.93
0.2 A 0.006
0.95
0.8 B 0.024
A 0.0001
0.6 0.1
0.8
0.3 0.0004
B
0.9
0.05
C 0.0002
0.066 0.2
0.7 B 0.0009
YES 0.8
A 0.0075
0.95
A 0.0002
0.2 0.5
0.05
B 0.0002
0.5
A 0.0608
0.2
NO 0.95
B 0.24
0.8
A 0.0003
0.8 0.1
0.3 0.0029
B
0.9
0.4
0.05 0.0026
0.2 C
0.7 0.01
B
0.8
A 0.076
0.95
A 0.002
0.2 0.5
Total probabilities of outcome 0.05
costs 0.5 B 0.002
Less than £200,000 A 0.71
£200,000–£500,000 B 0.28
More than £500,000 C 0.0028
Figure 3.7 Simplified event tree for bus garage fires
3.2.1.5 Results
The risk assessment indicated that, for the event identified, a higher than ordinary
hazard level of sprinkler spray density was necessary to prevent fire spread from
bus to bus. The frequencies of fires in bus garages was about 0.1 per year. The
fire risk was then calculated for the bus garages with and without sprinklers. The
difference between the two figures is the benefit rate from reduced property losses
by fitting sprinkler systems. This was approximately £2,000 per year (but this
varied with the size of garage).
Historical accident data indicate that the predicted risk of damage is pessimistic;
very few records of such fire damage could be found. Having quantified the benefits
of sprinklers in reducing risks in bus garages, how can we tell whether they would
represent a good investment in fire safety? The answer of course is cost-benefit
analysis (Health and Safety Executive 1989, Ramachandran 1998b).
Table 3.7 General data on the reliability of fire protection systems
General fire data Reference
Probability of fire occurring Industrial 0.5 DOE (1996)
and not being reported Commercial 0.8 Appleton (1992)
to the local authority fire Dwellings 0.8 Crossman and
service (UK). Zachary (1974)
& Reynolds and
Bosley (1995)
Average 0.8 All the above
Frequency of reported fires Industrial 4.4 × 10–2
per occupancy per year. Storage 1.2 × 10–2
Shops 8.4 × 10–3
Offices 5.7 × 10–3
Hotels etc. 3.7 × 10–2
Dwellings 2.7 × 10–3
Probability of a reported Industrial 0.004 Scoones (1995)
fire causing property loss in Other commercial 0.001
excess of £1M (1992 prices). Educational 0.003
Typical probability of fire Beyond room of 0.1 Government
spread for reported fires. origin Statistics Service
To other buildings 0.2 (1992)
Fire alarm and detection systems
Improvement in probability General value 0.5–0.6
of early detection in
buildings with AFDA.
Reliability of alarm box, General value 0.95–1.0
wiring and sounders.
Reliability of detectors. [a3] Commercial smoke 0.90
Domestic smoke 0.75
Aspirating smoke 0.90
Heat 0.90
Flame 0.50
Automatic fire suppression systems
Overall reduction in loss due General value 50%
to provision of sprinklers.
Probability of successful Maximum 0.95
sprinkler operation.[a1] General: 0.90
Property protection 0.80
Life safety 0.75
Minimum
Probability of successful General value 0.90
operation of other AFS
systems.
Continued …
Table.3.7 continued
General fire data Reference
Smoke control systems(mechanical and natural)
Probability of system General value 0.90
operating as designed, on
demand.
Passive fire systems
Probability that fire-resisting Masonry walls 0.75
structures will achieve at Partition walls 0.65
least 75% of the designated Glazing 0.40
fire resistance standard. Suspended ceiling 0.25
Probability of fire doors being General value 0.30
blocked open.
Probability of self-closing General value 0.20
doors failing to close
correctly on demand
(excluding those blocked
open).
3.2.1.6 Summary
A study to assess the benefits of installing sprinkler systems in bus garages indicated
that there were business continuity and property protection benefits to the operator.
However, the cost-benefit analysis and the operator’s contingency plans meant
that there was no cost-benefit for the consequence case for installing sprinklers
in bus garages. As a result of the risk assessment, the operator implemented other
forms of safeguard and fire precaution.
3.2.2 Fault tree analysis

Fault trees trace the root causes of a given final or ‘top event’ of concern by working
backwards logically to base events. A fault tree is a graphical representation of
logical relations between an undesirable top event and primary cause events. The
construction of a fault tree starts with the definition of the top event identified at
the hazard identification stage. The tree is constructed by placing various cause
events in correct sequential order. This is generally done by working backwards
from the top event and specifying the events’ causes, faults or conditions that could
lead to the occurrence of the top event, working backwards from each of these
which, in effect, become secondary top events and so on. This process is continued
and terminated when a final set of base (or root) events, faults or conditions are
identified. A diagrammatic representation of the process would then generate the
branches of a tree. Probabilities or frequencies are assigned to the root events.
The events in a fault tree are connected by logic gates, which show what
combination of the constituent events could cause the particular top event. These
are mainly AND gates in which all the constituent events have to occur and OR
gates in which only one of the constituent events needs to occur to cause the
specific top event. The probability of occurrence of the top event is calculated
using Boolean algebra. Simple fault trees can be calculated directly using Boolean
algebra, but more complex fault trees require that ‘minimum cut sets’ or ‘path sets’
be established using ‘Boolean reduction’ techniques.
Figure 3.8 shows a general fault tree and the use of the logic underlying the
AND and OR gates. Computer software is available that can speed up the use of
complex fault trees.
An example of a simple fault tree applied to fire detection is given in Figure
3.9. Here the top event is ‘failure to detect fire within 5 minutes of ignition’. The
causes of this top event can be followed through the four root causes for which
data can be generated.
OR gates are usually calculated by adding the root probabilities together and
subtracting their multiplied value:
POR = (PA + PB ) − PA .PB (3.37)
AND gates are calculated by multiplying the root probabilities together.

The top events of fault trees can very often supply the conditional probabilities
for event trees.
PAND = PA .PB (3.38)
Top event
of concern
AND
Intermediate event failure of

route events 2 and/ or 3
Root event 1
OR
Root event 2 Root event 3
Figure 3.8 General form of a fault tree

Failure to detect fire within
5 minutes of ignition
AND
Failure to detect fire within

5 minutes from ignition Failure by staff to observe a
by automatic fire detection fire within 5 minutes of
system ignition
OR OR
No automatic detector Failure of detector to Area not observed by staff Staff fail to observe fire
present detect fire
AND AND
A C
Detector present Failure to detect fire Staff present Fail to detect fire
A B C D
Figure 3.9 Fault tree for failure to detect a fire within five minutes of ignition
3.2.2.1 Example of probabilistic fire risk assessment applied to the detection

of fires in wards by people
Traditional hospital wards were open in layout with the staff able to see most areas
of the ward. This allowed staff to detect a fire by observation and so no automatic
fire detection was required. A move towards patient privacy means that the vast
majority of wards now comprise single-bed rooms. This means that staff no longer
have visual access to the majority of the ward. Therefore, what level of automatic
fire detection is required to maintain current safety levels?
The acceptance criterion is ‘equivalency’, i.e. the level of safety provided by
the alternative fire safety solution is as good, if not better, than the standard
prescriptive code solution.
A fault tree was constructed to assess the probabilistic nature of fire detection
in hospital wards and is shown in Figure 3.9. Here the top event is the ‘failure to
detect a fire within five minutes of ignition’. The causes of this top event can be
followed through the four root causes for which data can be generated.
These root causes are:
• no automatic detector present;

• automatic detector fails to detect the fire;
• no staff member present; and
• staff member fails to detect the fire.
This assumes that fires detected by patients are neglected so that the analysis
errs on the side of safety.
The two cases of interest are:
1 good visual access and no automatic fire detection (AFD);

2 poor visual access but with AFD throughout.
The fault tree minimum cut set was developed using Boolean algebra and then
quantified for both cases using the data in Table 3.8.
The results of the analysis for the two cases are:
1 probability of failure to detect fire within five minutes (good visual access and
no AFD) = 0.3;
2 probability of failure to detect fire within five minutes (poorer visual access
but with AFD throughout) = 0.1.
The results from Case 1 are consistent with those recorded generally for the
probability of failure to detect a fire within 5 minutes of ignition in hospitals wards
(NHS Estates 1996).
Table 3.8 Data used to quantify the probability of failure to detect a fire within five
minutes of ignition
Root event Case 1 Case 2 Comments
A 1.0 0.0 Automatic detection present/not present
Č 0.0 1.0
B 0.1 0.1 British Standards Institute PD 7974 Part 7 (2003)
C 0.3 0.8 Charters, Barnett et al. (1997)
Č 0.7 0.2
D 0.001 0.001 Probability that trained operators will not detect
a warning signal (a low value errs on the side of
safety); Cullen (1990)
Sensitivity analysis indicates that the probability of failure of AFD could be as
high as 0.26 before the results reverse (a factor of 2.6 greater than the probability
given by fire statistics).
Equally, the proportion of area covered by AFD could be as low as 63 per cent
before the results reverse. Because there is relatively little additional cost between
63 per cent and 100 per cent AFD coverage, the principle of practicability indicates
that full coverage of AFD is the appropriate solution in this case.
Therefore, the alternative fire safety solution (Case 2) provides equivalency
with the standard prescriptive guidance solution (Case 1).
3.3 Statistical models
3.3.1 Power functions

For any type of building, the probability of fire starting or frequency of fire
occurrence will increase with the number of ignition sources and hence with the
size of the building expressed in terms of total floor area, A, in square metres.
This probability or frequency denoted by F(A), as proposed by Ramachandran
(1979/80), can be estimated for any period by applying the following formula:
n S ( A) (3.39)
F ( A) = n
N S
N
( A)
where:
n = number of fires during the period, say a year
N = number of all buildings of the type considered at risk i.e. involved and not
involved in fires
Sn (A) = proportion of buildings of size A involved in fires
SN (A) = proportion of buildings of size A at risk.
The parameters n and Sn (A) can be estimated by analysing data on real fires in
the buildings of the type considered for a period of years. Such data are provided
by fire brigades in the United Kingdom. But a special survey of all buildings of the
type considered at risk needs to be carried out to estimate N and SN (A). To obtain
approximate estimates of N and SN (A), it may be possible to analyse some other
statistics such as the distribution of manufacturing units according to employment
size which may be combined with an estimate of average area occupied by each
person employed in the manufacturing industry considered.
Statistical studies reviewed by Ramachandran (1970, 1979/80, 1988a) have
shown that F (A) is approximately given by the ‘power’ functions
F ( A) = K Aα (3.40)
where K and α are constants for a particular type of building. The parameter K
includes the ratio (n/N) in Table 3.9 while α measures the increase in the value of
Table 3.9 Annual chance of a fire outbreak for various occupancies

Hazard Number of Number of fires P = n/N
buildings* annually+
(N) (n)
Industry 183 377 8 075 4.4 × 10–2
Houses 14 202 359 38 142 5.7 × 10–4
Commercial – shops 664 817 5 574 8.4 × 10–3
Commercial – offices 152 430 866 5.7 × 10–3
Assembly – entertainment 12 540 1 446 1.2 × 10–1
Assembly – non-residential 143 019 2 810 2.0 × 10–2
Residential – clubs, hotels etc 36 609 1 352 3.7 × 10–2
Residential – institutions – 803 –
Storage 199 612 2 420 1.2 × 10–2
Notes
* Source: 108th Report of the Commissioners of HM Inland Revenue
+ Fires during 1967
F(A) for an increase in A. Equation (3.40) is the same as equation (3.35) but with
a different notation; K=a, A=Ab and α=b. To estimate K and α, as mentioned
in section 3.1.4, it is necessary to analyse fire statistics and data collected through
a special survey of buildings at risk. This exercise was carried out by the Home
Office in the United Kingdom to estimate K and α for major groups of buildings –
see Table 3.4 for the values of K=a and α=b with A in m2]
It may be seen from the figures in Table 3.4 that the value of α (or b) is less
than unity for most of the occupancy groups, e.g. 0.53 for all manufacturing
industries. Actuarial studies in some European countries, based on frequency of
fire insurance claims as a function of the financial value (size) of the risk insured,
confirm that the value of α is about 0.5 for industrial buildings – see Benktander
(1973).
A value less than unity for α indicates that the probability of fire starting
does not increase in direct proportion to building size; it increases according
to a fractional power. If two buildings are considered, one twice the size of the
other, the probability of fire starting in the larger building will be less than two
times the probability for the smaller building. If these two are industrial
buildings, for example, the probability for the larger building will be 1.44 (=20.53)
times the probability for the smaller building. The value of α will be less than
unity, since different parts of a building will have different types and number
of ignition sources. A value equal to unity for α would imply that all parts of
a building have the same risk of fire breaking out. Actuarial studies reviewed
by Benktander (1973) support the theoretical arguments mentioned above.
These theoretical studies are confirmed by Ramachandran (1970, 1979/80 and
1988a).
The actuarial and statistical studies mentioned above have also revealed that,
in the event of a fire occurring, the probable or expected or average area damage
is approximately given by the ‘power’ function
D(A) = C Aβ (3.41)
where, as in Equation (3.40), A is the total floor area (size) of a building and C and
β are constants for a particular risk category or type of building. These constants
can be estimated by fitting the power function in Equation (3.41) if, as for fires in
the United Kingdom, data are available for area damage, D(A), and building size,
A. Based on the survey by the Home Office, UK mentioned earlier, Rutstein (1979)
estimated the values of C and β for major groups of buildings – see Table 3.10.
A fire in a large building is more likely than one in a small building to be
discovered and extinguished before involving the whole building. The proportion,
D(A)/A destroyed in a large building can, therefore, be expected to be smaller
than the proportion destroyed in a small building. Hence, this proportion or
damage rate would decrease with increasing values of A; in other words, the value
Table 3.10 Probable damage in a fire – parameters of Equation (3.41)

Occupancy Probable damage in a fire
C (m2) β
Industrial buildings
Food, drink and tobacco 2.7 0.45
Chemical and allied 11.8 0.12
Mechanical engineering and other metal goods 1.5 0.43
Electrical engineering 18.5 0.17
Vehicles 0.80 0.58
Textiles 2.6 0.39
Timber, furniture 24.2 0.21
Paper, printing and publishing 6.7 0.36
Other manufacturing 8.7 0.38
All manufacturing industry 2.25 0.45
Other occupancies
Storage 3.5 0.52
Shops 0.95 0.50
Offices 15.0 0.00
Hotels etc. 5.4 0.22
Hospitals 5.0 0.00
Schools 2.8 0.37
of β would be less than unity as revealed by the figures in Table 3.10. This result is
supported by statistical and actuarial studies.
Provision of reliable fire suppression equipment, if it operates satisfactorily
when a fire occurs, would reduce the damage rate, D(A)/A, and the value of β.
Consider, for example, the value of 0.45 for β estimated by Rutstein (1979) for
industrial buildings without sprinklers. For industrial buildings with sprinklers,
Rutstein’s estimate of 16m2 for average area damage in a building of total floor
area 1500m2 indicates a value of 0.27 for β. This result is based on the assumption
that the value of C denoting initial conditions is 2.25m2 whether a building is
sprinklered or not.
Based on data for fire spread provided by UK fire statistics, Ramachandran
(1988a) estimated an initial damage of 4.43m2 for the parameter C and average
damage of 187m2 in an ‘average’ textile industry building of total floor area 8000m2
without sprinklers. This provided a value of 0.42 for β. With an average damage of
31m2 and C = 4.43, β was estimated to be about 0.22 for a textile industry building
equipped with sprinklers for fires in which the heat produced was sufficient to
activate the system. For an unsprinklered textile industry building, Rutstein
(1979) estimated β = 0.39; his estimate of 2.6m2 for C apparently included very
small fires. The estimate for expected damage, D(A), is very sensitive to the value
used for the parameter C.
In a later study with C = 4.43, Ramachandran (1990) estimated values of 0.68
and 0.60 for β for unsprinklered and sprinklered textile industry buildings. These
estimates were based on maximum area damage (worst-case) in a fire of 2000m2 in
an unsprinklered building and 1000m2 in a sprinklered building, both buildings of
an average size of 8000m2 total floor area.
For any type of building or a risk category within a type, the product of F(A)
given by Equation (3.40) and D(A) given by Equation (3.41) provides an estimate
of fire risk expressed on an annual (yearly) basis. Table 3.11 is an example showing
the estimates of annual damage (risk) for industrial buildings of different sizes with
and without sprinklers.
Equations similar to (3.40) and (3.41) with the total financial value V at
risk instead of total floor area A, and their product are used for determining
approximate ‘risk premiums’ for fire insurance – see Benktander (1973). These
equations can be converted to provide estimates of F(A) and D(A) in terms of
total floor area A by using an estimate for v (= V/A) denoting value density per
square metre of floor area. Area damage can also be converted to financial loss by
using an approximate estimate for loss per square metre of fire damage. At 1978
prices, this rate of loss was £140 per m2 for all manufacturing industries – see
Rutstein (1979).
3.3.2 Exponential model

It is necessary to discover or detect the existence of a fire in a building soon after
the start of ignition in order to commence early and safe evacuation and reduce
the number of fatal or non-fatal casualties. Early detection of fire would also
Table 3.11 Industrial buildings – annual damage
Probable damage D(A) Annual damage
Building size Probability of Without With Without With
(A) fire starting sprinklers sprinklers sprinklers sprinklers
(m2) F(A) (m2) (m2) (m2) (m2)
1000 0.0661 50.37 14.53 3.33 0.96
2000 0.0955 68.81 17.52 6.57 1.67
3000 0.1184 82.58 19.54 9.78 2.31
4000 0.1379 94.00 21.12 12.96 2.91
5000 0.1552 103.92 22.43 16.13 3.48
8000 0.1991 128.40 25.47 25.56 5.07
10,000 0.2241 141.97 27.05 31.81 6.06
20,000 0.3236 193.93 32.62 62.76 10.56
50,000 0.5259 292.89 41.77 154.03 21.97
100,000 0.7594 400.11 50.37 303.84 38.25
500,000 1.7819 825.48 77.79 1470.92 138.61
1,000,000 2.5730 1127.66 93.80 2901.48 241.34
F(A) = 0.0017A 0.53
(annual probability); D(A) = 2.25 A
0.45
(without sprinklers); D(A) = 2.25A0.27(with
sprinklers)
reduce property damage by enabling the commencement of fire fighting when the
fire is small in size such that the fire can be controlled and extinguished quickly.
To assess the value of early detection in reducing damage to life and property, it is
necessary to estimate area damage, A(T), as a function of duration of burning, T.
For this purpose, Ramachandran (1980) proposed the following exponential model
based on statistics of real fires attended by fire brigades in the United Kingdom:
A(T) = A(0) exp (θT) (3.42)
where A(0) is the area initially ignited and θ the fire growth parameter.
It should be emphasised that A(T) in Equation (3.42) is the final (cumulative)
size of a fire in terms of area damage at the time (T) of its extinguishment. A(T)
is not the fire size at any intermediate time T. Fire statistics do not and cannot
provide information on the size of a fire at any specific time, say, when the fire
brigade arrives at the scene of a fire.
The model in Equation (3.42) is based on some scientific theories and
experimental results according to which heat output in a fire grows exponentially
with time. It is applicable to the period after the onset of ‘established burning’
when a fire has a steady and sustained growth. It is not applicable to the initial
and very early stage of a fire which, although small in size, can be very variable
in length of time; this stage can last for hours (smouldering) or it can be over in
minutes.
During the very early stage of fire development, commencing with the ignition
of the first object or material, area damage at time T may be directly proportional
to T2. This model is generally recommended by fire safety engineers according
to whom the heat release rate of a fire during the growth phase is proportional
to T2. This result is based on a series of fire tests and analysis of some real fires.
This conclusion may be true for the initial stage of a growing fire, when only one
object or material in a room is involved. It is unlikely to be realistic for the later
stage of fire growth, when several objects are involved. For the later stage, the
exponential model in Equation (3.42) appears to be more realistic, particularly
due to the stochastic nature of fire spread from object to object – see next section.
The uncertainties caused by the random nature of fire spread are not taken into
account by a T -square curve.
The exponential model in Equation (3.42) is also supported by statistical analysis
of large samples of real fires. Fire statistics available in the United Kingdom provide,
for each fire, information on area damage, A(T), and the duration of burning, T. An
estimate for T can be obtained as the sum of the following five periods:
• T1 – ignition to detection or discovery of fire;

• T2 – detection to calling of fire brigade;
• T3 – call to arrival of the brigade at the scene of the fire;
• T4 – arrival to the time when the fire was brought under control by the brigade;
• T5 – control to extinction of a fire.
The fifth period, T5, need not be included in the sum since the growth of a fire
will be practically negligible during this period.
An estimate of T1 is given by the brigade according to the following classification:
1 Discovered at ignition (T1 = 0).

2 Discovered under 5 minutes after ignition.
3 Discovered between 5 and 30 minutes after ignition.
4 Discovered more than 30 minutes after ignition.
Average values of 2, 17 and 45 minutes can be adopted for the second, third
and fourth classes of T1.
In a pilot study concerned with the economic value of automatic fire
detectors, Ramachandran (1980) applied the exponential model in Equation
(3.42) to data on fires in the textile industry during 1978. He estimated the
values of 4.69m2 and 0.0632 for the parameters A(0) and θ respectively. Based
on these parameter values, the exponential growth of fire has been depicted
in Figure 3.10 to show the economic value of automatic detectors connected
and not connected to the fire brigade. In a later study of textile industry fires,
Ramachandran (1988a) estimated the values of θ as 0.083 if not sprinklered
and 0.031 if sprinklered. These overall values were applicable to fire spread
in a building. The value of 4.43m2 for A(0) related to item first ignited and
corresponded to the initial stage taken as zero time and the commencement of
140
R Detector connected to the fire brigade
130 B Detector not connected to the fire brigade
G Fire not discovered immediately after ignition
120
Duration of burning up to
110 the arrival of the fire brigade
Duration of burning up to
100 the time fire is brought under control
Area damaged (m2)
90
80
70
60
50
40
30
20
10 R B G
R B G
10 20 30 40 50 60
Time (min)
Figure 3.10 Average time (min) and area damaged (m2)
established burning. For fire spread within a room, θ was estimated to be 0.196
if not sprinklered and 0.117 if sprinklered.
Ramachandran and Chandler (1984) applied the exponential model for
assessing the economic value of early detection of fires in industrial and commercial
premises. In this study, the model was expanded for estimating the values of θ
separately for the four periods T1, T2, T3 and T4. Fires were classified according
to whether they occurred in production or storage areas. A further classification
related to whether or not fires were tackled by first-aid fire-fighting before the
arrival of the brigade. Public/assembly areas replaced production areas in the
distributive trades group.
The following were the main conclusions in the above study:
1 Early detection would reduce the damage, especially in premises without first-
aid fi
re-fighting or training in its use.
2 There is a clear need to achieve early fire detection in storage areas.
3 Early detection should be followed up by quick action to extinguish the fire.
4 Automatic detection systems connected to the fire brigade only marginally
increase the savings from early detection as compared with systems not
connected directly to the fire brigade.
The exponential model in Equation (3.42) provides an estimate of the ‘doubling
time’ given by:
d = (1/θ)loge2 = (0.6931) / θ (3.43)
which is the parameter generally used for characterising and comparing rates of
fire growth of different objects or materials. This is the time taken by a fire to
double in size and is a constant for the exponential model. For example, if it takes
5 minutes for the area damage to increase from 10m2 to 20m2, it will also take only
5 minutes for the damage to increase from 30m2 to 60m2, 50m2 to 100m2, 100m2
to 200m2 and so on. For an example considered earlier, with θ equal to 0.117 and
0.196 for sprinklered and unsprinklered rooms, the corresponding doubling times
are 5.9 minutes and 3.5 minutes respectively.
Assuming an exponential model, Bengtson and Ramachandran (1994)
estimated the fire growth rate, θ, for four types of occupancies – railway properties,
public car parks, road tunnels and power stations. They showed clearly that the
maximum value (upper confidence limit) of growth rates in individual fires is
distinct from the maximum value of the average growth rate in all fires. The former
maximum is considerably higher than the latter and represents more realistically
the worst case scenario.
With appropriate assumptions about the ratio of vertical rate of fire spread to
horizontal rate, fire growth rates and doubling times, as discussed above in terms
of area damage (horizontal spread), can be converted to fire growth rates and
doubling times in terms of volume destroyed – see Ramachandran (1986). As one
might expect, doubling time in terms of volume involved is shorter than doubling
time in terms of area.
Bengtson and Laufke (1979) used the exponential model and a combination
of quadratic (T2) and exponential models to estimate the fire area and time when
sprinklers operate in different hazard categories. The authors also discussed the
estimation of the time to flashover at different room volumes with and without
installed fire ventilation system. Other topics discussed by them include operation
time of smoke detectors, fire brigade efforts on extinguishing a fire and effects on
evacuation of people. Bengtson and Hagglund (1986) described the application of
an exponential fire growth curve in fire engineering problems.
3.3.3 Regression analysis

Simple regression is concerned with the estimation of the expected or average
value of a dependent variable y for a given value of an independent variable x. In
the context of fire protection, y can be probability or frequency of fire occurrence,
area damage, financial loss or fatality rate per fire. The independent variable x can
be a factor such as building size or time taken to detect or discover a fire which
affects the dependent variable y. If y has a linear relationship with x which can be
represented by the straight line
y = m x + c (3.44)
the problem is to estimate the values of the parameters m and c.

The problem of fitting the straight line in (3.44) can be solved graphically in
the first instance by plotting the pairs of observations (yi, xi) available for a sample
of fires which occurred in the type of building or risk category analysed. The
subscript i refers to the ith fire in the sample. The value of the dependent variable
yi for the ith fire corresponds to the value of the independent variable xi for that
fire. It can then be tested whether a straight line can be drawn approximately
to pass through the scatter of points representing the pairs of observations (yi,
xi). If this is possible then the parameter m is called the slope of the line and
measures the ratio between the change in the value of y, say, from y1 to y2 and the
corresponding change from x1 to x2 in the value of x:
m = (y2 – y1) / (x2 – x1) (3.45)
The constant c which is the value of y when x = 0 is the intercept on the y axis
and is called the y intercept.
If the graphical analysis reveals a linear relationship between y and x, the values
of the parameters m and c providing the ‘best’ linear fit can be estimated by applying
the method known as ‘least squares’. Computer packages are available for this
method. With the values of m and c thus estimated, Equation (3.44) can be used
to estimate the expected, mean or average value of y for a given particular value
of x. Computer packages also provide an estimate of the standard deviation of the
‘residual error’ which can be used to obtain the ‘upper confidence limit’ denoting
the maximum value and the ‘lower confidence limit’ denoting the minimum value
of the expected (mean) value of y.
In some cases, the straight line relationship may be applicable to the logarithm
of y and logarithm of x or logarithm of y and x. For example, according to Equation
(3.41) the logarithm of the expected value of area damage, D(A), has a linear
relationship with the logarithm of total floor area (size), A, of a building. Figure 3.11
reproduced from a previous study (Ramachandran, 1990) is an example according
to which if sprinklers are installed in a building of total floor area 10,000m2, area
damage can be expected to reduce to 1100m2 from 2300m2. This figure also reveals
the fact that, for an acceptable damage of 2300m2 a sprinklered building of total
floor area 33,000m2 would be equivalent in damage to a non-sprinklered building
of total floor area 10,000m2. This acceptable size for a sprinklered building may be
reduced to 28,000m2 to take account of the probability of, say, 0.1 of the system
not operating in a fire – see Ramachandran (1998b). Thus, an acceptable size
of 10,000m2 for an unsprinklered building can be increased to 28,000m2 if the
building is provided with sprinklers.
According to the exponential model in Equation (3.42), the logarithm of
damage, A (T), has linear relationship with the duration of burning, T. This
relationship for fire spread within a room depicted in Figure 3.12 has been
reproduced from a previous study by Ramachandran (1990).
1000
Nonsprinklered building
Damage (m2)
Sprinklered building
100
10
1
10 25 50 100 250 500 1000 2500 5000 10,000 33,000
Total floor area of building (m2)
Figure 3.11 Damage and building size – textile industry, UK
According to Ramachandran (1993), the probability of one or more deaths in a

fire or fatality rate per fire, P, is approximately given by the straight line:
P = K + λD (3.46)
where D is the time taken to discover or detect a fire and K and λ are constants
for a given type of building. The relationship between P and D, shown in Figure
3.13, for single and multiple occupancy dwellings, was based on data for the
period 1978–1991. Respectively for these two types of buildings, λ was estimated
to be 0.0008 and 0.0006 and K to be 0.0016 and 0.0015. The fatality rate per fire
for that period was about 0.012 for both types. This rate corresponded to average
fire discovery times of 13 and 18 minutes for single and multiple occupancy
dwellings. If it is assumed that the discovery time of an automatic detector is one
minute, on average, the fatality rate would reduce to (K + λ) if all the dwellings
were fully protected by automatic detection systems. Under such protection
the fatality rate per fire in single and multiple occupancy dwellings would have
18
17
16
15
Sprinklered room
14
13
Duration of burning (min)
12 Nonsprinklered room
11
10
9
8
7
6
5
4
3
2
1
0
1 10 100 1000
Area damaged (m 2)
Figure 3.12 Fire growth within room of origin
35
30
Number of deaths per 1000 fires
25 ing
w ell
yd
nc
20 pa
ccu
gl eo
15 Sin ing
well
cyd
an
10 cup
oc
le
ltip
5 Mu
0
10 20 30 40
Discovery time (min)
Figure 3.13 Discovery time and fatality rate (current risk level)
reduced to about 0.002, i.e. 2 deaths per 1000 fires, from 0.012, i.e. 12 deaths
per 1000 fires which was the average life risk level prevailing during the period
1978–1991.
In the simple single linear regression described above, it is assumed that the
value of the dependent variable, y, would be significantly affected by the magnitude
of a single factor (independent variable), x. This may not be strictly true since
a number of factors may jointly affect y, each factor contributing some amount
towards y. For example, the area likely to be damaged in a fire may be affected
by building size, building height, compartment size, ventilation, fire load, number
of compartments, number of floors, fire resistance and the presence or absence
of fire protection measures such as automatic detectors, sprinklers and smoke
control systems. There are also other factors such as fire brigade attendance time
and control time, rate of fire spread and so on. Some factors will affect property
damage, some life damage (e.g. number of escape routes, widths of escape routes)
and some both property and life damage. Once these factors are identified, their
contribution to the damage can be estimated by performing a multiple regression
analysis, with data on damage and factors for each fire for a sample of fires. Such
data should be available, and if not, should be collected or estimated and their
numerical values used in the analysis.
If p factors (independent variables) are considered in a multiple regression,
their contributions to damage (dependent variable) quantified by the regression
parameters βj ( j = 1, 2, …, p) are estimated by the model
Z = β0 + β1 w1 + β2 w2 + … + βp wp (3.47)
where z is the logarithm of damage, y, to base e, wj is the numerical value (or its
logarithm) of the jth factor. A preliminary simple regression analysis of data for Z
and wj may be carried out to determine whether the value of wj to be used in a
multiple regression is its numerical value or its logarithm. For a qualitative factor
such as sprinklers, the value +1 may be assigned if the building is equipped with
sprinklers or –1 if not equipped. For quantitative factors the parameter βj measures
the increase in the value of z for unit increase in the value of wj. The constant β0
measures the fixed effect, not depending on the factors included in the model; it is
an average value for the effects of factors not included in the model. The model in
Equation (3.47) assumes that damage in a fire has a log normal distribution – see
Section 3.3.4.
In the application of the model in Equation (3.47), for the ith fire, zi is the
logarithm of damage and wij is the corresponding value of the jth factor. If data are
available for n fires, and p factors, the n sets of (p + 1) values provided by zi (i = 1,
2, …, n) and wij (i = 1, 2, …, n ; j = 1, 2, …, p) are used in a least squares multiple
regression analysis to estimate the parameters βj (j = 0, 1, 2, …, p). Computer
packages are available for performing this analysis. Once the parameters βj are
estimated, the expected value of the logarithm of damage can be estimated for
any given set of values for the factors wj ( j = 1, 2, …, p) with the aid of the
Equation (3.47).
Most computer software packages on multiple regression provide an estimate
of the correlation between the dependent variable z and each of the independent
variables wj ( j = 1, 2, …, p). An independent variable (factor) whose correlation
with z is very low (close to zero) can be excluded from the analysis and the
parameters βj of the other factors re-evaluated. The contribution to damage, z,
from a factor with low correlation, will be negligible.
Software packages also provide estimates of the correlation between independent
variables. If two independent variables wj and wk are highly correlated, such a high
degree of interaction will confuse the interpretation of the predicted value of z
due to ‘collinearity’. In such a case, only one of the two variables, wj or wk, may be
included in the final analysis.
In both the single and multiple regression analysis, if the variable z = loge y
where y is the dependent variable such as area damage, computer packages will
provide estimates of the regression parameters m and c in Equation (3.44) or
regression parameters βj ( j = 0, 1, 2, …, p) in Equation (3.47). These parameters
and equations would provide an estimate of the expected value, µ, of z for given
values of independent variable x or independent variables wj (j = 1, …, p). The
median value of damage, y, on the original scale is given by exp (µ). The probability
of exceeding this median is 50 per cent assuming that the ‘residual error’ in the
regression analysis has a normal distribution. Computer packages provide an
estimate of the standard deviation σ, of the residual error. The maximum value of
z is given by the upper confidence limit (µ + t σ) and the corresponding maximum
value of damage y by exp (µ + t σ). For any probability level, the value of t can be
obtained from a table of the standard normal distribution. For example, if t = 1.96,
the probability of z or y exceeding the maximum is 0.025. The expected value of
y is given by
exp [µ + (σ2/2)]
Instead of area damage, the probability ps of fire spread beyond the room of
origin may be used as the dependent variable in a single or multiple regression
model. In this case, the ‘logit’ given by
Ps = ½ log [ ps / (1 – ps ) ] (3.48)
should be used in the estimation process, instead of ps, for rendering the effects of
factors approximately additive. In the logit model, the probability of area damage
exceeding, say, 100m2 or financial loss exceeding, say, £100,000 can be used for ps.
Baldwin and Fardell (1970) applied the logit model to estimate the influence
of various factors on the probability of a fire spreading beyond the room of origin.
According to this study, there were significant differences in this probability
between buildings used for different purposes and between some single storey
and multi-storey buildings. The biggest factor affecting fire spread was the time
of discovery of the fire, the chance of spread at night being twice that of the
day; this was probably because of delays in the discovery of fires. The chance of
spread was considerably smaller for modern buildings than for older buildings,
particularly for multi-storey buildings. This was, perhaps, the result of increased
building (fire) control and safety consciousness. The fire brigade attendance time
had no influence on fire spread.
Shpilberg (1975) applied the logit model to quantify the relative effects of
types of building construction, number of storeys, sprinkler protection, type of fire
department and the subjective Factory Mutual Overall Rating on the probability
of loss size. His object was to predict the probability of loss being above or
below $10,000 given the particular characteristics of a group of risks. The logit
transformation, Equation (3.48), was applied to the probability of loss exceeding
$10,000. For purposes of illustration, Shpilberg used all fire loss claims in industrial
property classified as ‘machine shops’ paid by Factory Mutual during 1970–1973.
In particular, the overall rating adopted by Factory Mutual was found to be of
great value for predicting size and degree of loss, i.e. the fraction of the value
of the property that was lost. Sprinklers were also found to be a major factor in
determining both expected size and degree of loss.
3.3.4 Probability distributions

The ‘power function’ in Equation (3.41) provides an approximate estimate for
the expected value of area damage when a fire occurs in a building of given size.
A more accurate but complex method is provided by probability distributions.
The probability distribution for financial loss x or area damage d expresses
mathematically the probabilities with which the loss or damage in a fire would
reach various amounts. This distribution is usually referred to as the ‘parent
distribution’ in extreme value theory (Section 3.3.5). This distribution provides
the expected value, standard deviation and other statistical parameters of x or d.
The nature of the probability distribution of loss x has been investigated by
Ramachandran (1974, 1975) and Shpilberg (1974) and other authors mentioned
in these papers. According to these studies, fire loss distribution is skewed (non-
normal) due to the fact that many fires are small and only a few fires are large. In
general the transformed variable z (= log x) i.e. logarithm of loss, has a probability
distribution belonging to the ‘exponential type’. This type, defined by Gumbel
(1958) with reference to the limiting (asymptotic) behaviour of a random variable
at the tail of its probability distribution, includes exponential, normal, log normal,
chi square, gamma and logistic distributions. Among these distributions, normal
and exponential for z have been widely recommended by actuaries, based on the
analyses of data on fire insurance claim amounts. These correspond to log normal
and Pareto for loss x on the original scale. Figure 3.14 is a general sketch of the
probability density function, f(z), of a normal distribution for z (= log x). The
function f(z) is the derivative of the cumulative distribution function F(z).
If figures for financial loss are available for all the fires which occurred in a
risk category, standard statistical methods or a graphical method can be applied
to identify the probability distribution which would provide the best fit for the
data analysed. But in most countries these data are generally available only for
0.4
0.3
f (z)
0.2
0.1
0
−3.0 −2.0 −1.0 zp 0.0 zq 1.0 z 2.0 3.0
z (= log10 x)
Loss x in units of thousand dollars
Figure 3.14 Density function (f (z)) curve of fire loss
large fires which, in the United Kingdom for example, are currently defined as
fires costing £50,000 or more in property damage. The threshold level which was
£10,000 until 1973 has been gradually increased over the years due to inflation and
the need to keep the number of large fires to be reported by insurance companies
at a manageable level. This led to the development of extreme value statistical
models discussed in Section 3.3.5.
However, a probability distribution can be constructed for area damage for
which, particularly in the UK, data are available for all fires. The probability of
area damage being less than or equal to d is given by the cumulative distribution
function G(d) and the probability of damage exceeding d by [1 – G(d)]. Figure
3.15 is an example (textile industry) based on fire brigade data and shows the
relationship between d and [1 – G(d)] for a building with sprinklers and a building
without sprinklers (Ramachandran, 1988a). The area damage is on a log scale
since, as revealed by several statistical studies, this random variable, like financial
loss, has a skewed probability distribution such as log normal. The values of the
parameters of this distribution vary from one type of building to another and
with the effectiveness of fire protection measures. Using Equation (3.42), the
x (horizontal) axis can be converted to describe the probability distribution for
duration of burning (in minutes).
It appears from Figure 3.15 that an initial damage of 3m2 is likely to occur
before the heat generated in a fire is sufficient to activate a sprinkler system.
For both types of building, the probability of damage exceeding 3m2 is 0.58. It is
0.2
0.5
1
Sprinklered
5
10
Nonsprinklered
20
40
50
100 [1 − G (d )]
60
70
80
95
99
99.99
1 2.7 10 20 55 100 148 403 1000 2981
Area damaged (m2)
Figure 3.15 Textile industry, United Kingdom – probability distribution of area damaged
G(d) = Probability of damage being less than or equal to d
1 − G(d) = Probability of damage exceeding d
apparent that, in the range greater than 3m2, the successful operation of sprinklers
would reduce the probability of damage exceeding any given value. For example,
the probability of damage in a fire exceeding 100m2 is about 0.18 if the building
has no sprinklers and 0.08 if the building is equipped with sprinklers. Also, for a
given probability level, say, 0.08 for [1 – G(d)] or 0.92 for G(d), the damage would
be 500m2 if not sprinklered, compared with 100m2 if sprinklered.
A log normal distribution was fitted to the raw data pertaining to Figure 3.15,
disregarding fires with damage less than 1m2 and following a method appropriate
for ‘censored’ samples – see Ramachandran (1988a). For the range exceeding 1m2,
values of 0.02 and 2.46 were obtained for the mean and standard deviation of
logarithm (z) of area damage in a sprinklered building. The expected (average)
damage was calculated as 41.64m2. For a non-sprinklered building the mean and
standard deviation of z were 0.75 and 2.87 leading to an expected damage of
216.67m2.
Figure 3.16 is an example based on Pareto distribution for area damage
which is the same as exponential distribution for logarithm of damage. If this
Q (d ) Log Q (d )
Q (d ) - Probability of damage exceeding d 1.0 0
0.1 −1
0.01 −2 No
ns
pr
ink
ler
ed
ro
Data points for nonsprinklered room om
Data points for sprinklered room Sp
0.001 −3 rin
kle
re
d
ro
om
0.0001 −4
−1 0 1 2 3 4
Logarithm of area damage
0.1 1 10 100 1000 10,000

Area damage (m2)
Figure 3.16 Pareto distribution of area damage – retail premises (public areas)
distribution is appropriate, as in Figure 3.16, logarithm of damage and logarithm

of the survivor function [1 – G(d)], denoted as Q(d) in the figure, should have
approximately a straight line relationship. Values for plotting the points in Figure
3.16 were obtained from the figures for the frequency distribution of damage in
Table 3.12.
If occupants in a building, attempting to escape from a fire, are exposed to
untenable (lethal) conditions caused by a combustion product, some of them
may sustain fatal injuries. The number of deaths occurring in a fire has a discrete
(discontinuous) probability distribution applicable for a random variable which
takes integer values. Poisson is one such distribution, which has been widely used
in the statistical literature for modelling the occurrence of rare events. According
to an extended form of this distribution, the probability p(x, t) of exactly x deaths
occurring in a fire due to an exposure period of t minutes as discussed by Rasbash,
Ramachandran et al. (2004), is given by
p (x, t) = exp (–δt) (δt)x / x! (3.49)
where
x! = x (x – 1) (x – 2) … 2.1
Table 3.12 Retail premises – assembly areas – frequency distribution of area damage
Number of fires in assembly areas
Area damage (m ) 2
Without sprinklers With sprinklers
less than 1 4197 154
(48.9) (31.3)
2–4 1987 37
(24.7) (14.7)
5–9 619 9
(17.1) (10.7)
10–19 463 13
(11.5) (4.9)
20–49 430 6
(6.2) (2.2)
50–99 221 4
(3.5) (0.5)
100–199 127 –
(2.0)
200–499 100 1
(0.8)
500–999 29 –
(0.4)
1000 and above 34 –
Total number of fires 8207 224

Note: Figures within brackets are percentages of fires exceeding upper limits of damage ranges in the
first column.
If t is the average value of t in a population of fires, δt = P which, as defined

with reference to Equation (3.46), is the fatality rate per fire or probability of one
or more deaths.
As discussed with reference to Equation (3.46), data are available for a number
of years for estimating the fatality rate per fire, P, for any type of occupancy.
Deterministic (scientific) models may provide, for any occupancy type, an estimate
of the parameter δ which is the increase in the probability of death for every extra
minute of exposure to untenable conditions caused by any combustion product. It
may be noted that the probability of no death is given by x = 0 in Equation (3.49)
or by exp (–δt) or exp (–P) which may be approximated to (1 – P) if P is small.
For any type of occupancy such as hotels, hospitals and office buildings with
a large number of people at risk, figures for the frequency distribution of number
of deaths for a number of years may provide sufficient data for fitting the Poisson
distribution in Equation (3.49). Other distributions, e.g. negative binomial, may
provide a better fit to these data. The ‘global’ distribution thus identified can be
adjusted to take account of the number of people at risk in a particular building.
3.3.5 Extreme value distributions
Large losses fall at the ‘tail’ of the parent distribution of loss, discussed in the
previous section. These losses constitute a very small percentage of the total
number of fires in a risk category as shown in Figure 3.14 by the area with dots with
values of z greater than zl at the upper tail. Hence, these losses are not amenable
to analysis by standard statistical methods. Extreme order theory, as developed
by Ramachandran in a series of papers, provides a mathematical framework for
making the best use of the information provided by large losses – for a review of
these studies see Ramachandran (1982, 1988b). The asymptotic theory of extreme
values, discussed in these studies, provides approximate results for an ‘exponential
type’ distribution. According to this theory, the number of fires (n) occurring
during a period should be large, say, more than 100. Also, preferably, at least 20
large losses should be available for analysis. Due to these requirements, in some
cases, it may be necessary to consider fires occurring in a group of buildings with
similar fire risks over, say, four or five years. In such a case, the loss figures should
be corrected for inflation using, say, indices for retail prices.
For a more detailed summary about extreme value theory see Rasbash,
Ramachandran et al. (2004). Basic features of this theory are as follows. The
logarithms of losses in n fires, occurring in a risk category over a period of years,
constitute a sample of observations generated by the parent distribution F(z). If
these loss figures are arranged in decreasing order of magnitude, the logarithm of
the mth loss in this arrangement may be denoted by z(m)n which is referred to as
an extreme order statistic. For the largest value, the subscript m takes the value
one (first rank). Over repeated samples (periods), z(m)n is a random variable with
an extreme value probability distribution. Extreme order theory is concerned with
the individual probability distributions generated by extreme order statistics of
varying rank m and their joint distribution.
In the absence of any knowledge about the exact nature of the parent
distribution, the parameters of the extreme value distribution of z(m)n can be
estimated from observations on z(m)n in repeated samples. Three methods available
for this purpose, as discussed by Ramachandran (1982), involve corrections due
to the varying value of n, number of fires, from period to period. The estimated
values of the parameters for different ranks (m) would describe the behaviour of
the tail of the parent distribution as a function of n. The parameters would also
provide an indication of the nature of the parent distribution. Parent distributions
satisfying this behaviour can be fitted to the large losses and the errors estimated,
in order to select a distribution which would provide the best fit.
Another application of the extreme value theory is concerned with the
estimation of the mean (µ) and standard deviation (σ) of z, logarithm of loss, in all
n fires, large and small. But this estimation has to be based on, say, r consecutive
large losses, m = 1 to r, above a threshold level. Information on financial loss may
be available only for r large fires out of n fires. To obtain the best estimates of µ
and σ in all n fires from r large fires, Ramachandran has developed two methods
– generalised least squares and maximum likelihood. The first method provides
Table 3.13 Average loss per fire at 1966 prices (£000s)
Sprinklered Sprinklered Non- Non-
single storey multi-storey sprinklered sprinklered
single storey multi-storey
Textiles 2.9 3.5 6.6 25.2
Timber and furniture 1.2 3.2 2.4 16.5
Paper, printing and publishing 5.2 5.0 7.1 16.2
Chemical and allied 3.6 4.3 4.3 18.2
Wholesale distributive trades – 4.7 3.8 19.4
Retail distributive trades – 1.4 0.4 12.4
‘unbiased’ estimates but involves complex calculations for which a computer

program has been developed. The second method is quite easy to apply and would
only require a pocket calculator. This method would provide ‘biased’ estimates
but formulae have been developed to adjust the results for biases. Both the
methods require an assumption, such as log normal, to be made about the parent
distribution.
Assuming a log normal distribution and applying the generalised least squares
method, Rogers (1977) estimated the average losses due to fires in industrial and
commercial buildings with and without sprinklers. His results are reproduced in
Table 3.13. Figure 3.17 is an example based on his investigation.
3.4 Stochastic models
3.4.1 Stochastic nature of fire spread

The statistical models discussed in the previous section are useful for assessing
fire risk in a group or category of buildings with similar risk of fire starting and
causing damage. The quantitative estimates of risk provided by these models are
generally applicable to a building of ‘average’ characteristics belonging to the risk
category considered. These estimates can be adjusted to provide an approximate
assessment of fire risk in a particular building within that category or group – see
Chapter 5.
However, it may be desirable to assess the risk in a particular building based
mostly on the characteristics of that building. These characteristics include:
• building layout
• design
• fire load, i.e.
• amount of combustible materials and objects; and
• arrangement of these objects, i.e. distance between the objects
• ventilation
0.01
0.05
0.1
0.2
0.5
1
2
100 x [1 − F (z )]
2
1
5
3
10
20 4
30
40
50
60
70
103 104 105 106
Total loss ( ) x − 1966 prices
Line Subpopulation Parameters z = log x

m = mean (average) value of z
m = −0.616
1 Sprink/singlestorey s = 1.024 s = standard deviation of z
2 Sprink/multistorey m = −1.419
s = 1.340
m = −0.334
3 Nonsprink/singlestorey s = 1.062
m = 0.401
4 Nonsprink/multistorey s = 0.992
F (z ) (cumulative distribution for z ) = Probability of loss less than or equal to z

V (x ) (cumulative distribution for x ) = Probability of loss less than or equal to x
Survivor probability = 1 − F(z ) = 1 − V (x ) = Probability of loss exceeding x or z
Figure 3.17 The survivor probability distribution of fire loss for each class in the textile
industry
• fire resistance of the structural boundaries; and

• installed fire protection measures such as sprinklers.
Fire risk in a particular building can be assessed by applying a stochastic

model of fire spread. Detailed reviews of stochastic models of fire growth were
carried out by Ramachandran (1995a, 2002) and Rasbash, Ramachandran et al.
(2004). Basic features of three of these models are discussed in the next three
subsections.
For the following reasons the spread of a real fire in a building is a stochastic
(probabilistic) phenomenon. It is not a deterministic process, obeying in exact
terms the course predicted by scientific theories and experimental results.
A fire in a room usually starts with the ignition of one of the objects. Then
it spreads to other objects depending on the distances between the objects and
other factors such as fire load and ventilation. This process produces a chain of
ignitions that can lead to fully developed fire conditions, defined as ‘flashover’. If
such conditions occur, the fire can spread beyond the room, depending on the fire
resistance of the barriers (walls, floor, ceiling).
There will, however, be a chance that the fire chain can break at some stage
for various reasons with the fire being extinguished before spreading further –
statistics of real fires support this hypothesis.
As described above, a fire passes through several stages during its development
with a chance of being extinguished during any stage. It stays for a random length
of time in each stage before moving to the next stage. Its movement (spread) from
stage to stage is governed by ‘transition probabilities’ which are functions of time
since the start of ignition. This hypothesis forms the basis of a stochastic model of
fire spread.
Due to uncertainties caused by several factors, the spread of fire is really
a stochastic phenomenon, although a fire experiences certain deterministic
(physio-chemical and thermodynamic) processes during its development. A
stochastic model is a non-deterministic model which provides a probabilistic
prediction of spatial spread of fire as a function of time – see Ramachandran
(1991).
3.4.2 Random walk

In a simple stochastic representation, the fire process involving any single object
or number of objects can be regarded as a random walk. The fire takes a random
step every short period either to spread with a probability λ or to be extinguished
(or burn out) with a probability µ (= 1 – λ). The parameter λ denotes the success
probability of the fire and µ the success probability of an extinguishing agent.
The process stops when the fire is extinguished; extinguishment is an ‘absorbing
boundary’ to the random walk.
If the fight between the fire and the extinguishing agent goes on continuously
every minute, a random walk as described above will lead to an exponential model
(Ramachandran, 1985) for the random variable t:
Q (t) = exp (–µt); c = µ – λ (3.50)
where Q(t) is the probability of duration of burning exceeding t minutes. The fire-
fighting effort is adequate if c is positive with µ > λ and hence greater than ½; it
is inadequate if c is negative with µ less than λ and hence less than ½. If c = 0,
such that µ = λ = ½ there is an equal balance between fire-fighting efforts and
the propensity of fire to spread.
The area destroyed (d) is also a random variable whose logarithm is directly
proportional to t as a first approximation – see Equation (3.42). This assumption
would transform Equation (3.50) to the Pareto distribution:
ϕ(d) = d -w ; d ≥ 1 (3.51)

denoting the probability of damage exceeding d (see Section 3.3.4). The use of
the Pareto distribution for fire damage was proposed by Mandelbrot (1964) who
derived this distribution following a random walk process.
Ramachandran (1969) found that a constant value for µ in Equation (3.50)
or for w in Equation (3.51) was unrealistic for fires which were fought and
extinguished at some stage. He suggested a distribution with the ‘failure rate’
µ in Equation (3.50) increasing for large values of t. The failure rate can be
decreasing in the early stages of fire development, denoting a success for fire
in spreading but it would eventually increase since fire-fighting efforts would
succeed ultimately.
Consider the spread of fire as a random walk along spaces arranged in a linear
sequence. Every minute, the fire takes a random step to move forwards to an
adjacent space with a probability of λ or stay in its present space with a probability
of µ (= 1 – λ). It does not move backwards to the space from which it has moved.
In a simple model, λ and µ are regarded as constants, although, realistically, they
are functions of time since the commencement of burning.
Under the above assumptions, if the random walk process starts in the first
space with probability one, the probabilities of the fire being in the first two spaces
after one minute are µ and λ respectively. After two minutes, the fire is in one
of the first three spaces. At that time, the fire may be in the first space with the
probability of µ2 or it may be in the third space with the probability of λ2, since it
should have moved to the second phase after one minute with the probability of
λ. Hence, after two minutes, the probability of the fire remaining in the second
space is 2λµ (= 1 – λ2 – µ2).
Generalising, after t minutes, the probability of the fire being in the rth space
is the rth term in the binomial expansion of (µ + λ)t. After t minutes, the
probabilities of fire burning in the first and (t + 1)th spaces are µt and λt. If µ is a
small quantity, (1 – µ) or λ is an approximation for exp (–µ) in Equation (3.50)
such that λt is an approximation for exp (–µt). Hence the probability of duration
of burning exceeding t minutes is λt approximately. The probability of the fire
being extinguished during the (t + 1)th minute is λt.µ.
3.4.3 Markov model

Stages of fire growth can generally be defined as states or spatial modules
(Watts, 1986), phases (Morishita, 1977), or realms (Berlin, 1980). Fire spreads,
moves or makes a transition from state to state. Mathematically, if a fire is in
state ai at the nth minute, it can be in state aj at the (n + 1)th minute according
to the transition probability λij(n). The probability of remaining in state ai at
the nth minute without making a move to another state is denoted by λii(n).
In a Markov process with stationary transition probabilities, the value of λij(n)
is a constant λij not varying with the time denoted by n. With m states the
transition probabilities per minute are most conveniently handled in the matrix
form:
λ11 λ12 λ13 … λ1m
λ21 λ22 λ23 … λ2m
P= …
λm1 λm2 λm3 … λmm
m
where ∑
j=1
λ ij =1; i =1, 2,..........m
The probabilities of the fire being in different states at time n can be expressed
as the vector
Pn = (q1n q2n … qmn)
where qin is the probability of the fire burning in state ai at time n. Since a fire can
m
be in one of the m states, for any time n, ∑q
i=1
in = 1. The vector Pn +1, given by the
matrix product Pn.P, expresses the probabilities of fire burning in different states
one transition (minute) later. If the fire starts in state a1, the first element in the
vector P0 for the initial time denoted by q10 is unity and the rest of the other
elements in this vector are zero. With this initial condition, the elements in the
vectors Pn for different times n can be obtained by performing the matrix
multiplication Pn.P repeatedly starting with P0.P.
As an example, consider a Markov model of fire growth in a room in which
state ai represents i objects burning. Suppose, with 3 objects, i.e. m = 3, and no
extinguishment, the process terminates with the occurrence of flashover at state a3
when all the three objects are burning. There is no recession in growth and hence
there is no transition to a lower state from a higher state. The problem now is to
estimate the transition probabilities per minute in the matrix P. This is possible by
applying scientific and statistical models to experimental data on heat output or
release rate for the three objects and other information such as distances between
the objects.
With the assumptions mentioned above, let the transition matrix be
0.5 0.3 0.2

P= 0 0.8 0.2
0 0 1.0
The process starts with the ignition of one of the objects with the other two
objects not yet ignited. We may express the probability distribution of the system
at the initial time 0 as the vector
P0 = (1 0 0)
By performing the matrix multiplication P0.P, the probability distribution of the
system after one minute is:
P1 = (0.5 0.3 0.2)
By performing the matrix multiplication successively, it may be seen that
P2 = P1P = (0.25 0.39 0.36)

P3 = P2P = (0.125 0.387 0.488)
P10 = P9.P = (0.0010 0.1064 0.8926)
P15 = P14.P = (0.0001 0.0352 0.9647)
and so on. Hence after 15 minutes, the probability of flashover (state a3), with all
the three objects burning, increases to such a high figure as 0.9647.
Following the model described above, Berlin (1980) estimated λij for six realms
for residential occupancies:
• o-fire state
n
• sustained burning
• vigorous burning
• interactive burning
• remote burning; and
• full room involvement.
The realms were defined by critical events characterised by heat release rate,
flame height and upper room gas temperature. Estimation of λij for different i and j
was based on data from over a hundred full-scale fire tests. The first realm, no fire,
was an ‘absorbing state’ since all fires eventually terminate in this state.
Berlin applied his model to a smouldering fire in a couch with cotton cushions.
He estimated the probabilities of maximum extent of flame development as
0.33, 0.07, 0.02 and 0.58 for a fire reaching the second, third, fourth and fifth
realms respectively but not growing beyond these realms. Berlin also discussed
other fire effects such as probability of self-termination and distribution of fire
intensity. The fire growth model of Beck (1987) was based on the six realms
defined by Berlin.
The state transition model (STM) is a particular (simple) version of a Markov
model with stationary (constant) transition probabilities. An event tree such as
in Figure 3.18 constitutes a simple STM in which fire in a room is described as
developing through four successive stages or states E1 to E4. A fire can ‘jump’
to E4 from E1 or E2 without passing through E2 and E3 but such ‘jumps’ have
not been considered in this simple STM. The probability Ei is the probability of
confinement of fire (extinguishment) in the ith state if the fire has spread beyond
the previous states. The probabilities E1, E2, E3 and E4 add up to unity. These
probabilities are limiting probabilities of a fire being extinguished ultimately in the
four states. Estimates of these probabilities for most of the building types and some
risk categories can be obtained by collecting and analysing data such as those
compiled by the fire brigades in the UK.
Based on estimates for Ei (i = 1 to 4), the probabilities λi and µi (i = 1 to 3)
can be estimated by applying the formulae given at the foot of Figure 3.18; λi
+ µi = 1. The parameters λi and µi are respectively ‘conditional’ probabilities
of fire spreading beyond and being extinguished in the ith state given that the
Sprinklered States Non-sprinklered

µ1 µ1
E1 0 .7 2 1 0 .4 9 E1
0.72 0.49
λ1 λ1
0.28 0.51
µ2 µ2
E2 0.19 2 0.23 E2
0.68 0.45
λ2 λ2
0.32 0.55
µ3 µ3
E3 0.07 3 0 .2 1 E3
0.78 0.75
λ3 λ3
0.22 0.25
µ4 µ4
E4 0 .0 2 4 0 .0 7 E4
1.0 1.0
Figure 3.18 Probability tree for textile industry

E1 = Probability of confinement to item first ignited.
E2 = Probability of spreading beyond item first ignited but confinement to contents of room of fire
origin.
E3 = Probability of spreading beyond item first ignited and other contents, but confinement to room
of fire origin and involvement of structure.
E4 = Probability of spreading beyond room of fire origin, but confinement to building.
E1=µ1; E2 =λ1⋅µ2; E3=λ1⋅λ2⋅µ3; E4=λ1⋅λ2⋅λ3⋅µ4; µ4=1

fire has spread to this state. These parameters are values to which the transition
probabilities of spread and extinguishment would tend ultimately over a period
of time; they are not probabilities per minute. The parameters λi and µi can be
expressed on a per minute basis by estimating the duration for which their values
in the event tree are applicable. It may be noted that µ4 = 1 and λ4 = 0 since fire
spread beyond the building of origin is not considered in this model.
Ramachandran (1985) applied an STM similar to Figure 3.18 in the overall
framework but expanded it to include sub-realms (periods), each of a fixed duration
of 5 minutes, for evaluating the spreading and extinguishment probabilities
λi and µi as functions of time. He also considered the probability wi of burning
(remaining) in the ith state without extinguishment or spreading; in this case,
λi + µi + wi = 1. Statistics on real fires provided by the fire brigades were used
for estimating the transition probabilities mentioned above and the probability
distributions of duration of burning in each stage. Four materials ignited first in
the bedroom of a dwelling were used as examples. The states defined by Aoki
(1978) were based on the spatial extent of fire spread and his analysis was similar
to that of Ramachandran (1985). Morishita (1977) considered eight phases of
spatial spread which included spread to the ceiling.
The STM approach can also be adopted for evaluating the probability of fire
spreading from room to room in a building – see Ramachandran (1995). Each
room or corridor in a building has an independent probability of fire spreading
beyond its structural boundaries. This probability for a room or compartment is
the product of probability of flashover and the conditional probability of structural
(thermal) failure, given flashover. Using these probabilities, estimated for different
rooms and corridors, fire spread in a building can be considered as a discrete
propagation process of burning among points which abstractly represent rooms,
spaces or elements of a building.
For example, three adjoining rooms R1, R2 and R3 provide the following four
states with the fire commencing with the ignition of objects in R1:
• only R1 is burning
• R1 and R2 (not R3) are burning
• R1 and R3 (not R2) are burning; and,
• all the three rooms are burning.
There is no transition from the first to the fourth state or from second to the
third state or from third to the second state. There is also no transition from the
second or third or fourth state to the first state, i.e. no recession of fire growth.
A transition from the second to the fourth state involves the spread of fire
to R3 from R1 or R2. The probability for this transition is, therefore, the sum of
probabilities for spread from R1 to R3 and R2 to R3. Likewise, the probability of
transition from the third to the fourth state is the sum of probabilities for spread
from R1 to R2 and R3 to R2. A fire can burn in the same state without transition
to another state. The process terminates when the fourth state is reached. With
the assumptions mentioned above, a transition matrix can be formed specifying
the probability of fire spread from one room to another. For estimating the average
time for transition to the fourth state, denoting the burning of all the three rooms,
Morishita (1985) proposed a method based on partitioning of the transition
matrix.
An event tree such as the one in Figure 3.18 can provide a ‘global’ estimate
of the probability of flashover for a room or compartment belonging to any
occupancy type and area of fire origin separately for the two cases – unsprinklered
and sprinklered. This method and other methods of estimating the probability
of flashover are discussed in Section 5.3. Given flashover, the probability of
compartment failure depends on the severity likely to be attained when a fire
occurs in the compartment and the fire resistance of the structural boundaries
of the compartment. Estimation of this probability is discussed in Sections 3.7.3
Safety factors and 3.7.4 Beta method.
The major weakness of the Markov model discussed above is the assumption
that the transition probabilities are constants, remaining unchanged regardless of
the number of transitions representing the passage of time. The length of time a
fire burns in a given state affects future fire spread. For example, the probability
of a wall burn-through increases with fire severity which is a function of time.
The time spent by fire in a particular state may also depend on how that state was
reached, i.e. whether the fire was growing or receding. Some fires may grow quickly
and some may grow slowly depending on high or low heat release. In a Markov
model with constant transition probabilities, no distinction is made between a
growing fire and a dying fire. An application of a Markov model with time-varying
transition probabilities would require large amounts of data and involve complex
computation procedures.
3.4.4 Network models

As explained in the previous section, the STM can provide for each room in a
building a cumulative probability, ps, at time ts, when the structural boundaries of
the room are breached. The duration ts is the sum of tf representing the time to
the occurrence of flashover and tb representing the time for which the structural
barriers of the room can withstand fire severity attained during the post-flashover
stage. The probability ps is the product of probability pf of flashover and probability
pb of structural failure given flashover. The pairs of values (ps and ts) for different
rooms can then be used in a stochastic model in which a building is represented
by a network with rooms or compartments as nodes and the links between these
nodes as possible paths for fire spread.
Consider, as an example, the simple layout of Figure 3.19(a) relating to four
rooms and the corresponding graph shown in Figure 3.19(b) which also shows the
probability (pij) of fire spread between each pair of rooms (i, j). The probability pij
refers to ps as defined above, whereas Dusing et al. (1979) and Elms and Buchanan
(1981) considered only the barrier failure probabilities pb ignoring the probability
of flashover pf. The specific problem considered by these authors was to compute
1 4
(a)
2
p12 p24
1 p32 p23 4
p13 p34
3
(b)
Figure 3.19a Room layout; b Corresponding graph
the probability of fire spread from room 1 to room 4 which might follow any of the
four paths
(1) → (2) → (4)

(1) → (3) → (4)
(1) → (2) → (3) → (4) and
(1) → (3) → (2) → (4)
Using the event space method, Elms and Buchanan considered first all possible
‘events’ or combinations of fire spreading or not spreading along various links. If
aij represents spread of fire along link ij, and a ij represents fire not spreading along
the link, then one event might be
[a12 , a13 , a23 , a32 , a24 , a34 ]
There will be 26 = 64 events which will all be exclusive as any pair of events
will contain at least one link for which fire spreads in one event and does not
spread in the other. The probability of each event occurring is the product of the
probabilities of its elements assuming that the elements are independent. Thus,
for the example given above, the event probability will be
p12(1 – p13)p23 (1 – p32)(1 – p24)p34
and the overall probability is the sum of all 64 event probabilities.

Representing the completed event space as a tree with 64 branches, Elms
and Buchanan adopted a procedure known as ‘depth-first search of a graph’ for
identifying or searching possible paths of links leading to node (room) 4 from
node 1. The calculation was carried out for each pair of rooms and the results
assembled in a ‘fire spread matrix’ with unit values for the diagonal elements. The
core of this model is a probabilistic network analysis to compute the probability
of fire spreading to any compartment within a building. The dimension of time
was not explicitly considered in this model, although it was implicit in many of
the functions used. In a similar network model proposed by Platt (1989), the
probability of spread is dependent on time.
Ling and Williamson (1986) proposed a model in which a floor plan is first
transformed into a network. Each link in their network represents a possible route
of fire spread and those links between nodes corresponding to spaces separated by
walls with doors are possible exit paths. The space network is then transformed
into a probabilistic fire spread network as in the example in Figure 3.20 with four
rooms, Rm 1 to Rm 4 and two corridor segments c1 and c2. With Rm 1 and Rm 1´
with a ‘prime’ denoting the pre-flashover and post-flashover stages the first link is
represented by
Rm 1 → Rm 1´
(pf, tf)
where pf represents the probability of flashover and tf represents the time to

flashover.
In Figure 3.20, three different types of links are identified. The first corresponds
to the fire growth in a compartment, the second to the fire breaching a barrier
element, and the third to fire spread along the corridors. To each link i, a pair of
numbers (pi, ti) is assigned with pi representing the distributed probability that a
fire will go through link i and ti representing the time distribution that it will take
for such a fire to go through link i. The section of the corridor, c1, opposite Rm
1 is treated as a separate fire compartment and is assigned a (pf, tf) for the link
from c1 to c1´. The number pair (ps, ts) represents the probability and time for the
Rm 1 Rm 1′ C1 C1′ Rm 3 Rm 3′ Rm 4 Rm 4′
(p f, t f) (p b, t b) (p f, t f) (p b, t b) ( p f, t f) (p b, t b) (p f, t f)
(p b, t b)
(p b, t b) (ps, ts) (ps, t s)
(p b, t b)
(p f, t f) (p b, t b) (p f, t f)
Rm 2 Rm 2′ C2 C2′
Figure 3.20 Probabilistic network of fire spread of Room 1 to C2

Fire growth within compartment;
Fire breaches barrier elements \\\\
Fire spread along corridor
p re-flashover spread of fire along the corridor from c1 to c2. Once full involvement
occurs in the section c1 of the corridor outside Rm 1 (i.e. node c1´ is reached) the
fire spread in the corridor is influenced more by the ventilation in the corridor
and by the contribution of Rm 1 than by the materials properties of the corridor
itself. Thus there is a separate link, c1´ to c2 which has its own (ps, ts). The number
pair (pb, tb) represents the probability of failure of the barrier element with tb
representing the endurance of the barrier element.
Once one has constructed the probabilistic network, the next step is to ‘solve’ it
by obtaining a listing of possible paths of fire spread with quantitative probabilities
and times associated with each path. For this purpose, Ling and Williamson adopted
a method based on ‘emergency equivalent network’ developed by Mirchandani
(1976) to compute the expected shortest distance through a network. (The word
‘shortest’ has been used instead of ‘fastest’ to be consistent with the literature.)
This new ‘equivalent’ network would yield the same probability of connectivity
and the same expected shortest time as the original probabilistic network. In this
method, each link has a Bernoulli probability of success and the link delay time is
deterministic.
It must be noted that there are multiple links between nodes in the equivalent
fire spread network. For example, the door between Rm 1 and the corridor could
be either open or closed at the time that fire flashed over in Rm 1. Ling and
Williamson assumed, as an example, that there is a 50 per cent chance of the
door being open and that an open door has zero fire resistance. Furthermore, they
assumed that the door, if closed, would have a 5-minute fire rating. With further
assumptions, they constructed the equivalent fire spread network (Figure 3.21)
with 12 possible paths for the example in Figure 3.20 to find the expected shortest
time for the fire in Rm 1 to spread to the portion of corridor c2. A similar network
was constructed for the case with self-closing 20-minute fire rated doors which
had ten possible paths. For the two equivalent networks, all the possible paths are
listed in Tables 3.14 and 3.15 with increasing time and with all the component
links identified. Each of the paths describe a fire scenario. For instance, the
scenario for path 1 in Table 3.14 would be: the fire flashes over, escapes from Rm
2 (0.5, 0)
Rm 1 Rm 1′ C1
(1, 5)
1 (0.5, 10) 3 (0.25, 12.5)
5
4
(0.5, 7.5)
7
(0.3, 15)
9
(1, 35)
8
(0.86, 25)
11
(0.5, 0)
6 (0.09, 5) (1, 10)
10
Rm 2 Rm 2′ C2
12
(1, 5)
Figure 3.21 Equivalent fire spread network with five-minute unrated doors
1 through an open door into the corridor c1 and spreads along the corridor to c2.
The probability for that scenario is 0.13. The time of 17.5 mins is the sum of 10
mins for flashover and 7.5 mins for fire spread from c1 to c2.
Ling and Williamson derived a formula for calculating, from the figures in
Tables 3.14 and 3.15, the probability of connectivity which is 0.5 for both the
networks. This probability is a direct result of the assumed probability of 0.50 for
flashover in the room of fire origin and the occurrence of unity probabilities in the
remaining links which make up certain paths through the network. According to
another formula, the expected shortest time is 29.6 minutes for Figure 3.21 which
increases to 47.1 minutes due to the presence of the 20-minute fire rated door. The
equivalent fire spread network thus facilitates an evaluation of design changes and
affords ready comparison of different strategies to effect such changes.
Following Ling and Williamson (1986), Connolly and Charters (1997) applied
a network model to evaluate the effectiveness of passive fire protection measures
in contributing to fire safety in hospitals. The particular objective considered by
Table 3.14 Pathways for fire spread equivalent network assuming five-minute unrated
corridor doors
Paths Component links Probability pi Time ti (minutes)
1 1–2–4 1/8 = 0.13 17.5
2 1–2–5 1/16 = 0.06 22.5
3 1–3–4 1/4 = 0.25 22.5
4 1–6–10–11 1/44 = 0.02 25.0
5 1–3–5 1/8 = 0.13 27.5
7 1–6–10–12 1/22 = 0.05 30.0
8 1–7–10–12 3/40 = 0.08 35.0
9 1–8–10–11 3/14 = 0.21 40.0
10 1–8–10–12 3/7 = 0.43 50.0
11 1–9–10–11 1/4 = 0.25 55.0
12 1–9–10–12 1/2 = 0.50 60.0
Table 3.15 Pathways for the fire spread equivalent network assuming self–closing 20-minute
rated corridor doors
Paths Component links Probability pi Time ti (minutes)
1 1–2–3 1/4 = 0.25 37.5
2 1–2–4 1/8 = 0.13 42.5
3 1–5–9–10 1/22 = 0.05 45.0
4 1–6–9–10 3/20 = 0.15 55.0
5 1–7–9–10 3/7 = 0.43 65.0
6 1–8–9–10 1/2 = 0.50 75.0
Table 3.16 Passive fire protection measures in hospitals
Fire safety measure tshortest (mins)
tf = 5 mins Safety
factor
Datum
30-minute wall, 5-minute doors, 50/50 chance of being open 27.7 1.00
(i) Lock store room door 30.0 1.08
(ii) Lock store room door, s/c on ward door 30.4 1.10
(iii) Store room door 30 min FR, s/c on store room door 33.7 1.22
(iv) Firecode 30 min FR doors both with s/c 34.8 1.25
Note: s/c = self-closer assumed to ensure the door is closed 80% of time; FR = fire-resisting
them was to prevent a fire starting in a store room from spreading to the ward.
They assumed a minimal level of fire protection for the datum case analysed. The
fire resistance of the wall separating the store from the ward was 30 minutes. The
doors to the store room and to the ward had a fire resistance of only 5 minutes. It
was assumed further that there was a 50 per cent chance of the doors being open
in which event their fire resistance would be zero. The probability of flashover, pf,
was set equal to the probability of fire spreading beyond the item first ignited. The
time, tf, for the occurrence of flashover was not specified.
For the datum level of fire protection described above, Connolly and Charters
(1997) calculated the expected shortest time, tm, for fire to spread from the store
to the ward as
tm = 1.22tf + 21.6 minutes
This result suggested that fire spread was likely to take at least 21.6 minutes for
the nature of fire protection defined in the datum case. Fire protection measures
considered by the authors included:
• locking of store room door; and

• fitting self-closing device on ward door and upgrading the fire resistance of
doors to 30 minutes.
A summary of their findings, reproduced in Table 3.16, is applicable to situations

where time to flashover is small, say, 5 minutes.
3.5 Monte Carlo simulation
3.5.1 Introduction
Fire safety engineers are required to deal with complex fire scenarios which
include human reactions and behaviour in addition to physical and chemical fire
processes evolved by a variety of burning materials. Physical models representing
such scenarios involve intractable mathematical relationships which cannot be
solved analytically. Also, sufficient and realistic experimental or statistical data
are unlikely to be available for estimating all the parameters of a physical model.
For such complex models, only numerical solutions can be obtained by applying a
simulation procedure step by step.
Simulation involves the construction of a working mathematical model,
representing a dynamic system in which the processes or interactions bear a
close resemblance or relationship to those of the specific or actual system being
simulated or studied. The model should include realistic input parameters
capable of generating outputs that are similar or analogous to those of the system
represented. Then, by varying the numerical values of the input parameters,
it would be possible to predict the time-varying behaviour of the system
and determine how the system will respond to changes in structure or in its
environment. Such simulation experiments can be performed on a computer by
developing an appropriate software package.
Simulation models can be either discrete or continuous. As time progresses, the
state of a building changes continuously as a small fire develops into a big fire. The
physical and chemical processes involved in such a fire growth lend themselves
to a continuous simulation model. On the other hand, discrete simulations are
more appropriate for determining ‘design times’ concerned with fire detection
and fighting and building evacuation. These times define critical events occurring
discretely during a sequence of clear-cut stages. In a continuous model, changes
in the variables are directly based on changes in time. Phillips (1995) discussed in
detail the various aspects of computer simulation for fire protection engineering
together with some examples.
3.5.2 Monte Carlo simulation

Monte Carlo analysis is a simulation technique applicable to problems involving
stochastic or probabilistic parameters. For example, some input parameters such as
compartment size and ventilation factor may be of deterministic nature such that,
for each of these parameters, a range of possible values can be used in simulation
experiments. On the other hand, some input parameters may be random variables,
taking values according to probability distributions during the course of fire
development. Examples of such variables are:
• rate of flame spread and fire growth

• temperature of the fire
• smoke concentration
• ambient air temperature
• wind speed and wind direction
• number of doors open
• number of windows open; and
• the response of occupants to fire alarms.
Consider, as an example, a stochastic parameter xi whose value at time T
during the course of fire development is xi(T). The exact value of xi(T) may
not be known but it might be possible to estimate its mean µi(T) and standard
deviation σi(T) and the form of its probability distribution. Suppose this
distribution is normal, such that the standardised counterpart t of xi(T) has
a standard normal distribution. Then, with ti = 1.96, the probability that the
value of the stochastic parameter xi at time T is less than or equal to the value
given by the following equation is 0.975:
xi(T) = µi(T) + σi(T) . ti (3.52)
The probability of the value of the stochastic parameter exceeding the value
given by the above equation is 0.025. This particular value of xi(T) can be
regarded as the probable maximum, while the value corresponding to ti = –1.96
in the above equation would be the probable minimum. The probability of the
value of the stochastic parameter being less than this minimum is 0.025. Instead
of the maximum or minimum value, a series of random values of xi(T) can be
generated by ‘spinning the Monte Carlo wheel’ in the computer and selecting,
at random, values of the standard normal variable ti. Virtually every computer is
equipped with a subroutine that can generate random numbers. This process will
provide a random sample for estimating the time-varying relationship between
the input parameter xi and an output variable yj. The output variable may be a
quantity such as area damage representing property damage or number of fatal
or non-fatal casualties representing life loss. Methods have been developed for
generating random values for most of the well-known probability distributions
such as Normal as well as any empirical distribution.
The probability distribution of an output variable yj can now be estimated with
the aid of random sample values of several input variables xi generated by Monte
Carlo simulation. Some input variables may be of deterministic nature and some of
stochastic or probabilistic type. It would be possible to regress the output yj on the
input variables using a multiple linear regression analysis technique. In this analysis,
as discussed in Section 3.3.3, it may be necessary to use the logarithm of yj and the
logarithms of some of the input variables or other appropriate transformations of
the variables to reduce to a linear form the relationship between the output and
input variables. There are, however, computer packages available to identify the
non-linear relationship and perform a non-linear multiple regression analysis. The
multiple regression equation then would provide an estimate of the expected value
of the output yj for a given set of random or extreme (maximum or minimum)
values of the input variables xi at any time T during the period of fire development.
Monte Carlo simulation can be used to generate sample values for constructing
the probability distribution of an input variable which might not be known due
to lack of data or whose mathematical structure is too difficult to be derived
theoretically. This method would provide the mean, standard deviation and other
parameters of the variable which can be used to confirm or reject theoretical
results.
The object of Monte Carlo simulation is to take account of uncertainties
governing the input and output variables involved in the fire safety system and
estimate the effects of input variables on the output variables. Suppose that, at a
given time, the output variables yj (j = 1, 2, …, N) are dependent on the input
variables xi (i = 1, 2, …, n) according to a set of functions:
yj = fj (x1, x2, …, .xn) (3.53)
Then, in the neighbourhood of x1, x2, …, xn, yj can be evaluated approximately

by expanding the function in Equation (3.53) in a Taylor series and then omitting
all terms after the second. This method would provide the variance-covariance
matrices for the input and output variables – see Phillips (1995).
Suppose the following linear hypothesis is valid:
yj = ao + a1 x1 + a2 x2 + … +an xn (3.54)
If xi (i = 1, 2, …, n) are independent random variables with mean x i and

2
variance σ i , the mean and variance of yj as given by Taylor series expansion are
yj = a0 + a1 x1 + a2 x + an xn (3.55)
σ2j = a12σ12 + a22σ22 +a2n σ2n (3.56)
For the input variable xi, consider as an example, the rate of heat output Q
that may increase with time T according to a T2 or exponential function. This
function will provide an estimate of Q at time T which may be regarded as the
expected or mean value µq(T) of Q. But Q is a random variable, since it is affected
by ventilation and other factors. Hence, as discussed earlier:
Q (T ) = µ q (T ) + σq (T ) ⋅ t (3.57)
where σq(T) is the standard deviation of Q(T) and the random variable t may be
assumed to have a standard normal distribution. Experimental data would provide
an estimate of σq(T) for any material or object. Random variables Q(T) can then
be generated by simulating random values of t.
The mass loss rate of fuel, m, is another input variable, whose mean value
and standard deviation can be estimated directly from experimental data or by
considering the relationship
Q = m ∆H (3.58)
where ∆H is the effective heat of combustion of the fuel, usually assumed to

have the value 18,800 kilojoules per kilogram. Q is measured in kilowatts and m
in kilograms per second.
The parameters m and Q are directly correlated with the rate at which the floor
area of a compartment is destroyed per unit of time – see Ramachandran (1995b)
and Section 5.2. Area damage is an output variable which is also affected by other
input variables such as fire load, compartment dimensions, ventilation factor and
delays in detecting and commencing fire fighting.
3.5.3 Simulation models – examples

Based on physical equations of motion, Evers and Waterhouse (1978) developed a
computer model for simulating and predicting the movement of smoke from a fire
in a building. The building was considered as a series of spaces or nodes, each at a
specific pressure with air movement between them, from areas of high to areas of
low pressure. The inflow and outflow of air from each node, through paths such as
windows, doors and ventilation openings, were determined in order to analyse the
smoke flow and examine the way in which the smoke concentration increases with
time at each node. The principal stochastic inputs to the calculations were the
number of doors and windows left open, burnt down or broken, the ambient wind
and temperature conditions and the location and severity of the fire. The values
of these stochastic variables were sampled at random from specified statistical
distributions. The model was applied to two types of buildings – the main building
at the Fire Research Station, Borehamwood and law courts in London.
Coward (1975) used a simulation method to estimate the statistical distribution
of fire severity from data obtained from surveys of office rooms. Fire severity was
defined as the time taken for an equivalent exposure in a standard furnace test. A
physical model was used to investigate how fire severity varies with the fire load,
the area of ventilation and the dimensions of a compartment. Values of these
factors were sampled at random from their frequency distributions ascertained
from the survey data. Each distribution used in the simulation was built up by
a pseudo-random number series, independent of all other series. The program
generated a model population of 20,000 office rooms. From the data provided by
such a large sample, fire severity, S (in units of time (minutes)), was estimated to
have the exponential distribution
p = exp (–0.04S) (3.59)
where p is the probability that any room will have a severity greater than S.
Accordingly, the average severity for an office room was estimated as 25 (=
1/0.04) minutes.
The mathematical structure of the Comparison of Risk Indices by Simulation
Procedures (CRISP), developed by the Fire Research Station (see Phillips, 1992)
is based on systems of simultaneous differential equations:
dx i
= f ( x1 ,…, x i ,…, x n ) (3.60)
dT
In these equations, the rate at which a state variable xi on the right-hand side
changes with time T is expressed by the differential coefficient on the left-hand
side. When the differential equations are solved, they yield a solution which
represents the evolution of the system over time.
In CRISP, the history of a ‘reference fire’ is analysed into a sequence of five
stages:
• initiation;
• accelerating growth;
• decelerating growth;
• full fire equilibrium; and
• extinction.
These stages are modelled with the aid of physical parameters such as
combustion, heat and temperature. Alarm and suppression come into action
during the last stage. Physiological effects and human behaviour are also taken
into consideration. Monte Carlo tests are conducted by carrying out a sequence of
runs of the model using sample values of parameters drawn from their probability
distributions. From the model output, it may be possible to estimate probability
densities representing fire conditions or number of casualties.
The Building Fire Simulation Model (BFSM) developed by the National Fire
Protection Association, USA (see Fahy, 1985) allows the user to examine the
interrelationships among fire development, spread of combustion products and
people movement in residential occupancies. Data from full-scale fire tests are
used in the model. Fire growth is defined in terms of six discrete stages called
‘realms’. The realms are:
• the n on-fire state;

• sustained burning;
• vigorous burning;
• interactive burning;
• remote burning; and
• full room involvement.
These realms are based on measurable criteria such as heat release rate and air
temperature. The levels of combustion products estimated by the model are based
on the realm the fire is in and the time it is in that realm. The ability of people to
escape depends on these estimated levels of combustion products throughout the
building.
Fire Risk Assessment by Simulation (Fire sim), developed in Norway, combines
simulation techniques with statistics to calculate the expected annual fire risk of
industrial plants (see Hansen-Tangen and Baunan, 1983). A large number of fires
are simulated to estimate the percentage damage of the total value. The expected
frequency of fires for the plant is estimated from statistics.
Sasaki and Jin (1979) carried out simulations of urban fires by applying
probabilistic percolation theory. From fire statistics, probability of fire spread was
obtained as a function of distance between buildings. The model provided an
estimate of number of burnt buildings per fire incident in Tokyo.
3.5.4 Merits and demerits of simulation

Simulation is the best tool for obtaining solutions to problems lacking sufficient
data for applying statistical, probabilistic, deterministic and other types of
mathematical models. It may be expensive and time-consuming to develop a
computer program/package for a simulation model but, once developed, the
package can provide rapid and inexpensive useful information about the time-
varying behaviour of the system it represents. It is a cheaper alternative method
than costly and time-consuming exercises involving statistical data collection
surveys, small or large scale experiments on fire characteristics of different
materials or evacuation drills. A computer program based on a simulation
model permits all the critical factors, processes and events that characterise
a real system to be internally represented so that alternative fire and human
behaviour patterns may be identified by varying the numerical values of the input
parameters. Simulation is the only practical tool to deal with uncertainties and
probabilities governing the behavioural patterns of a complex system consisting
of several interactive factors varying with time.
Simulation does not produce a general solution. It does not identify all the
possible behaviour patterns. Instead, simulation gives one time history of system
operation corresponding to initial conditions and coefficients of the model
parameters, whose numerical values are selected or specified. Several sets of
numerical values for initial conditions and coefficients will have to be used as
inputs for predicting the time-varying history of the system. Hence, simulation
exercises of a complex system can be time-consuming and expensive. The results
of simulation exercises in fire safety engineering should be checked and validated
against data provided by real fires.
Simulation is not the model to be used for preliminary evaluations and
comparisons. Analytical models, e.g. stochastic, are cheaper to use than simulation,
particularly if good-quality data are available in sufficient quantities for evaluating
the major component parts of a system. It is, however, necessary to base analytical
models on assumptions which are realistic and relevant according to scientific
theories supported by experimental results.
3.6 Consequence analysis
3.6.1 Introduction
Consequence analysis is intended to assess how severe a fire event or set of fire
events are likely to be.
The ideal consequence analysis should:
• be physically similar to the system or building being assessed; and
• cover the full range of possible fire events.
The implication of these objectives is that the ideal consequence analysis is

an infinite number of full-size experiments with real people and assets which
experience all possible fire events over an infinite period of time. Clearly this is
not practical or ethical, so like many other forms of quantitative risk assessment
we seek to find the information that we need to assess event consequences in a
number of other ways. These other methods include:
• historical data;
• disasters and near misses;
• experiments and fire tests;
• modelling.
3.6.2 Historical data

Historical data is where information about previous fire events has been
collected. Typically, historical data can be national for a certain building type or
for a specific organisation or business. The fire report, filled in by fire brigades, is
an example of the latter and this collects data on the consequences of each fire
event in terms of number of injuries and fatalities and area damaged by burning,
heat and smoke.
Figure 3.22 shows an example of a consequence probability distribution.
The advantage of historical data is that it is a measure of the fire consequences
from real fire events in real buildings. If these buildings are similar to the one being
assessed, this can be very useful information.
The disadvantages of historical data include:
• nder-reporting or bias in reporting or analysis;

u
• lack of physical similarity or homogeneity in the set being studied;
• short sample duration compared with the events of concern.
To a degree, under-reporting and bias can be dealt with by ensuring that the
risk analysis is undertaken with a consistent data set. For example, if a study shows
that small fire events are under-reported, the data may still be useful in assessing
the risks for large fire events. Bias is a little harder to deal with, but an assessment
of the nature and extent of the bias may lead to use of data, whose pessimism (i.e.
it errs on the side of safety or a robust decision) there is a strong case for.
Lack of physical similarity or homogeneity in the set from which the data is
collected can lead to consequences from one set of circumstances being used to
assess a physically quite different set of circumstances. It is worth noting that, in
insurance risk assessments, if the set is a portfolio of the insured, homogeneity is less
important in assessing companies’ risk exposure and hence in setting premiums and
excesses.
0.45
0.40
0.35
0.30
0.25
Probability
0.20
0.15
0.10
0.05
0
1 2 5 10 20 50 100 200
Cost (£ x 10-3)
Figure 3.22 Typical consequence probability distribution
Short sample duration can mean that there has been insufficient time to collect
enough data to cover the consequences of the events of concern. For example,
1,000 buildings studied for 10 years represents 104 building years of experience.
This may provide a reasonable basis for consequences whose events have a
frequency of 10–3 per building year or greater, but is clearly not adequate to address
consequences for events whose frequency is of the order of 10–6 per building year.
For these reasons most quantitative fire risk assessments include other forms of
consequence analysis.
An example of the application of historical data is the assessment of
consequences from small fires. In many fires consequences can occur when the fire
is too small to cause untenability in the assessment area. In these cases, statistical
analysis is used to predict the consequences of small fires, e.g. people who die as a
result of intimate contact with the fire through accident or attempts to extinguish
the fire. Figure 3.23 shows a typical correlation of the area damaged by fire/heat/
smoke and the probability that occupants will be injured.
3.6.3 Disasters and near misses

Disasters like the King’s Cross (see Figure 3.24) and Piper Alpha fires can provide
significant insight into the various failure modes and potential severity of fire
events. They can provide insight into the way that the fire behaved, what the main
mechanisms of event escalation were, what the main hazards/damage mechanisms
were and how people responded to the event.
However, it is essential to bear in mind that they are only a single event for
a specific facility and therefore are of relatively limited use when assessing the
infinite number of possible fire events that could occur in other different facilities.
If we address any one disaster too literally, there is a danger that we are constantly
addressing the last disaster rather than providing a broad understanding of the
potential consequences of fires in a building.
0.4
0.3
casualties per fire
No. of non-fatal
0.2
0.1
0
0 5 10 15 20 25
Estimate of area damaged by fire, heat, smoke, etc
No. of non-fatal casualties per fire vs estimate of area damaged by fire, heat, smoke, etc
(shopping malls, indoor markets, etc)
Figure 3.23 Correlation of the number of non-fatal injuries per fire and the area damaged
by fire, heat and smoke
Figure 3.24 Ticket hall after the fire at King’s Cross underground station
3.6.4 Experiments and fire tests

Experiments and fire tests can provide insight into the physical behaviour of
materials, items or spaces in a building.
Fire tests are applied to a wide range of building materials and contents and
so can provide insight into the likely behaviour of materials when exposed to
a specific physical hazard (for example a radiant heat flux or time/temperature
curve). Fire tests are primarily intended for the classification of building products
and so may not fully represent the physical processes (many are small or reduced
scale to minimise economic burdens on manufacturers) or situation of interest,
although assessments can be made where the situation of interest varies only
slightly from the standard test. Figure 3.25 shows the fire testing of a shutter.
Figure 3.25 Fire testing of a shutter
Experiments, if designed to address a specific item, can provide much

more relevant information to a specific building and/or type of fire event. Fire
experiments are more likely to be full or large scale and so more representative
of the physical processes. For example, Figure 3.26 shows a fire experiment to
assess the likely fire growth and smoke spread from an engine compartment fire
in a bus. Experiments, if ethically sound, may also involve people in evacuation
exercises and desktop event simulations. The disadvantage of experiments is
that they are usually large and expensive and so, like disasters, they only usually
represent a small sample of events or a limited part of the event progression.
Therefore, the use of test and experimental data usually require some form of
modelling.
3.6.5 Modelling
3.6.5.1 Introduction
Modelling is an approach for predicting various consequences of fires in buildings.
There are three main approaches:
1 simple calculations such as those in CIBSE TM19 (1995);

2 zone (or control volume) models;
3 field (or computational fluid dynamic) models.
Figure 3.26 Example of a full-scale fire experiment
Simple calculation procedures are available in many fire engineering textbooks

and handbooks and an example in this section is the addressing of fire tenability.
This section will mostly discribe control volume modelling before discussing more
complex computational models.
Control volume models are often known as ‘zone’ modelling. The approach
works by dividing the building fire ‘system’ in to a series of control volumes (or
zones). Each control volume represents a part of the system that is homogeneous
in nature, that is, it is assumed to have the same properties throughout (e.g.
temperature, velocity, density, species concentration etc). Conservation equations
are applied to each control volume to predict how ‘source terms’, such as the fire or
plume, and processes between control volumes, such as radiative and convective
heat transfer, affect the control volume’s properties. Figure 3.27 shows how the
building fire domain can be divided up into different control volumes (Drysdale
1999).
Control volume modelling has the capacity to predict various aspects of fires in
buildings and is an approach used in many fields of engineering. Indeed, control
volume modelling has been developed and applied to fire in compartments since
the 1970s (Quintiere 2003). However, any radical departure by the fire system
from the conceptual basis of the control volume model can seriously affect the
accuracy and validity of the approach.
Heat to structure
mvent mplume
Make up air
Radiated heat
Figure 3.27 Typical schematic of building fire control volume model
3.6.5.2 Application of control volumes to fires

Control volume modelling of fires is built on three types of model: conservation
equations, source terms and mass and heat transfer sub-models. These models
are described in brief below; an extensive review of control volume equations as
applied to compartment fire can be found in the literature e.g. Quintiere (1989).
The kinds of assumptions that are typical of control volume modelling include:
• all properties in the control volume are homogeneous;

• the gas is treated as an ideal gas (usually as air);
• combustion is treated as a source term of heat and mass;
• mass transport times within a control volume are instant;
• heat transfer to building contents such as vehicle is neglected;
• the section of the building is constant;
• the pressure in the building is assumed to be constant;
• frictional effects at boundaries are not explicitly treated.
When applying control volume equations to building fires, consideration should

be given to the unique nature of some fire phenomena in buildings, for example:
• An assumption that hot layer properties are homogeneous along the length of
the spaces will only be tenable for relatively short spaces.
• Ambient and forced ventilation flows in buildings may affect air entrainment
in plumes.
• The relative velocities of hot and cold layers may mean that shear mixing
effects at the interface may not be negligible.
Conservation equations
The main basis of control volume modelling is the conservation of fundamental

properties. The concept of conservation is applied to mass, energy and momentum.
However, momentum is not normally explicitly applied since the information
needed to calculate velocities and pressures is based on assumptions at the
boundaries.
The conservation of mass for a control volume states that the rate of change
of mass in the volume plus the sum of the net mass flow rates out to the volume
is zero:
d ( ρz )
A + ∑ mo = 0 (3.61)
dt
where: A is the area of the control volume
ρ is the density of the gas in the control volume
z is the height of the control volume, and
m is a net mass flow rate out of the control volume.
Figure 3.28 shows how this differential equation for the conservation of mass
can be applied to a control volume (Rylands et al. 1998).
The mass contained within the control volume at time t is given by the following
equation (Rylands et al. 1998):
M(t ) = M(t − ∆t ) + ∆t( m in − m out + m in_hot + m in_cool ) (3.62)
where: M is the mass in the control volume

∆t is a small interval of time
min, mout, min_hot and min_cool are the mass flow rates into and out of
the control volume due to convection and shear mixing effects.
Hot layer min_hot
Velocity u
min M mout
Cool layer min_cool
Figure 3.28 Schematic of mass flow between control volumes

Similar equations can be used for the conservation of species (that is mass
concentrations such as carbon monoxide) and energy (Quintiere 2003).
The mass concentration of a species Y can be conserved by modifying Equation
(3.61):
dY
ρzA + ∑ m o (Y − Ycv ) = ω (3.63)
dt
where: Y is the mass concentration of a species in the flow
Ycv is the mass concentration in the control volume, and
ω is the mass production of that species due to the fire.
The mass concentration of a species depends strongly on plume entrainment

(see ‘Mass and heat transfer sub-models’ below) and the mass production rate of
that species by the fire. The mass production rate of a species will vary significantly
depending on the type of fuel and the air/fuel ratio near the fire. For example,
the mass production rate of carbon dioxide depends on the rate of heat release,
the amount of carbon in the fuel and the air/fuel ratio. Limited knowledge of the
detailed chemical reactions in real fires’ combustion means that a simplified one-
step chemical model and studies of stoichiometry can be used to estimate species
production. However, uncertainties around the application of stoichiometry mean
that most control volume models use a mass conversion factor of the mass flow
rate of burnt fuel. For carbon monoxide these mass conversion factors can vary
tremendously depending on the type of fuel and whether the fire is fuel bed or
ventilation controlled. Therefore, results of large-scale fire tests are used to derive
mass conversion factors of CO.
The conservation of energy for a control volume combines Equation (3.61) and
the equation of state, p = ρRT, such that
dT dp
ρC p zA − zA + C p ∑ m (T − Tcv ) = m f χ∆H − Qnet_loss (3.64)
dt dt
where: Cp is the specific heat capacity of the gas in the control volume
p is the pressure in the control volume
T is the temperature of gas in the flow
Tcv is the temperature of gas in the control volume
mf is the rate at which fuel is volatilised
χ is the combustion efficiency
∆H is the heat of combustion (taken as positive), and
Qnet_loss is the net rate at which heat is lost to the boundary.
This model assumes that there is sufficient oxygen to react with the fuel with
a combustion efficiency factor to adjust for incomplete combustion. There is also
difficulty in dealing with combustion in vitiated layers. Often in control volume
models, the rate of heat release is a user input from which the mass flow rate of fuel
is derived.
If the rate of temperature changes in the control volume is low, the first term
can be neglected. This leads to a simpler quasi-steady analysis for growing fires. In
well-ventilated conditions (some transport buildings are well ventilated in terms
of fire dynamics), the second term can also be neglected. This leaves an enthalpy
term for flows into and out of the control volume and two source terms. For flows
out to the control volume the lumped parameter assumption means that T = Tcv.
Source terms
The main source terms in fire modelling are the rate heat release and the mass
flow rate of fuel. Where the fire source is known and well controlled, such as a gas
burner, precise values can be used. In most fire safety situations the rate of heat
release and mass flow rate of fuel is the result of a spreading fire over a variety of
material and surfaces. In these circumstances, an empirical model to give the rate
of heat release at time t can be used (CIBSE TM19 1995):
q f = α(t − t o )2 (3.65)
where: qf is the rate of heat release of the fire

α is the fire growth coefficient
to is the time between ignition and fire growth (incubation
period).
Similar source term models exist for the mass flow rate of fuel, for example:
qf
mf = (3.66)
∆Hχ
Given the conversion rate of fuel to carbon monoxide (a function of the materials
and how well ventilated the fire is) the mass concentration of CO can be estimated.
These empirical models provide approximations and so consideration should
be given so that values for the coefficient of fire growth, combustion efficiency and
mass conversion of rates are appropriate.
Mass and heat transfer sub-models
Mass and heat transfer models are an essential feature of control volume models.
These models may include:
• entrainment in plumes;
• flows through openings;
• mixing between layers;
• convective heat transfer to surfaces;
• radiative heat transfer;
• conductive heat transfer;
• other effects.
Entrainment in plumes has been shown to have a critical effect on the
tenability and development of fire hazards. There is a range of models available
to predict flame heights and mass entrainment in plumes (CIBSE TM19 1995,
Heskestad 2003). It is essential that the model used is appropriate to the physical
situation of the real fire. For example, entrainment in an axisymmetric plume may
be appropriate for a fire in the floor in the middle of a building section, but may
not be appropriate for a spill plume from a window or a plume attached to one of
the building walls. The effect on plume properties of any ambient ventilation and
for a situation where flames reach the hot layer should also be taken into account.
Figure 3.29 shows a schematic of a building fire plume.
Flows through openings are crucial to the modelling of fires in compartments,
such as doors and windows of vehicles, flows into and out of building ventilation
system openings and entrance/exits. Detailed models for flows through openings
and building ventilation systems can be found in the literature, e.g. Emmons
(2003).
Mixing between layers can occur in one of three ways:
1 a cold flow injected into a hot layer;

2 shear mixing associated with lateral layer flows;
3 mixing due to building wall flows.
Cold flows injected into a hot layer can be resolved by computational fluid
dynamics or physical modelling research. Shear mixing of layers has been studied
to a certain extent and some correlations are available for counter flow between
layers and back-layering against forced ventilation. Mixing due to wall flows has
been studied in compartment fires. None of these three phenomena is as critical as
the primary buoyant mixing in the plume (Emmons 1991, Vantelon 1991).
Convective heat transfer to surfaces is one of the main processes of heat
loss between the hot layer control volume and the building lining. Convective
heat transfer to ceilings has been studied extensively. Convective flows will vary
along the building walls and ceiling depending on their position with respect to
the fire. Where convective heat transfer is dependent on local boundary layer
temperatures (rather than hot layer control volume temperature) an adiabatic
wall temperature approach can be used. Convective heat transfer for ceilings and
walls in building fires has not been developed and so most control volume models
use natural convection correlations (Atreya 2003, Evans 2003).
Radiative heat transfer theory is generally sufficient for control volume models
of building fires. Grey body radiation from uniform temperature hot gas layers
can be predicted, although emissivity values require careful consideration. For
radiation from flames, empirical data is used because complex temperature
distributions for radiation from flames and the role of soot are not well understood
(Tien et al. 2002, McCaffrey 2003).
Conductive heat transfer through the building linings should be balanced with
the radiative and convective heat transfer from the hot gas layer control volumes.
This entails a numerical (or graphical) solution to a set of partial differential
Roof
Floor Fire
Figure 3.29 Schematic of a plume in natural ventilation
equations. Most control volume models consider conductive heat transfer in one
dimension only and that the building linings are thermally thick or conductive
heat transfer is to an infinite ambient heat sink (Carslaw and Jaegar 1959, Rockett
and Milke 2003).
Figure 3.30 shows a schematic of the kinds of heat transfer processes that may
be relevant to fires in buildings.
Flames Plume
Cool Hot
layer layer
Cool Mixing Hot

wall layer wall
Mixing
wall Radiation
Convection
Conduction
Ambient
wall
Figure 3.30 Typical schematic of building fire heat transfer processes

From the above it can be seen that other effects can be incorporated into control
volume building fire models easily, as long as there is an adequate description of
the phenomenon and there is appropriate data.
3.6.5.3 Application of control volume models in fire safety

Recent fires in buildings have illustrated how, for fires in buildings, the products
of combustion are confined and this may result in fast smoke movement and
rapid threat to life. Figure 3.26 shows the typical control volumes associated with
a 2-layer building fire model. A review of building fire control volume models
indicated a number that are being used to predict building fire smoke movement
(Quintiere 2003).
There are also likely to be significant differences between the source term and
heat and mass transfer sub-models used in each of the models and these should
be reviewed in selecting a model. For example, some models incorporate modified
plume sub-models to address flame impingement of the hot layer or building roof.
Figure 3.31 shows typical results from a transient, 3-layer, building fire control
volume model. It shows a typical vertical temperature profile against experimental
data.
Control volume models are being increasingly used to inform design decisions for
building fire safety and the next two subsections indicate two typical applications.
1.0
0.6
Height (h/H)
0.2
0 200 400 600 800 1000 1200

Temperature (K)
Test Model
Figure 3.31 Typical results for a three-layer control volume model compared with
experimental data
3.6.5.4 Evacuation
Evacuation analysis predicts how quickly occupants are likely to exit an assessment
area. This uses simple models for time to detection, pre-movement time and
movement time. The total time to escape is given by
te = ta + t p + tm (3.67)
where: te is the total time to escape (s)

ta is the time from ignition to alarm (s)
tp is the p re-movement time for occupants (s)
tm is the time for occupants to move to a place of relative safety (s).
The time for occupants to move to a place of safety depends on the number of
occupants, the travel distance, the exit width and the speed of movement and flow
rate of occupants. For each evacuation scenario, the time to move will be limited
by travel distance or exit width.
The time for the last occupant to move this distance is given by
dt
tm( d ) = (3.68)
um
where um is the velocity of movement (taken to be 1 m/s).
The time for the last occupant to flow through the exits is given by
Np
tm( w ) = (3.69)
∑ w .n e p
where: we is the exit width (m)

Np is the number of people (people)
np is the flow rate of people through an exit (people/m/s).
The movement time, tm used in the analysis is then taken to be the larger of the
travel distance and exit width limited times (Equations (3.68) and (3.69)).
The number of n on-fatal injuries (for te > tu) is given by:
Np
N i = (t e − t u ). (3.70)
tm
where: tu is the time for conditions to become untenable
From the value of Ni, the number of fatalities can be calculated using factors
based on fire statistics.
3.6.5.5 Fire tenability

By analysing the response of people escaping a building fire to the hazards that the
fire creates, it is possible to gain a much better understanding of the way that a
building and its fire precautions will perform (Charters 1992, Charters et al 1994).
Control volumes models are particularly well suited to providing information on
fire hazards so that tenability can be assessed.
The main hazards from building fires are (Purser 2003):
• heat (hyperthermia);
• toxic gases (hypoxia); and
• thermal radiation.
The level of visibility in a building fire, although not directly hazardous, may
reduce occupants’ way-finding capability and so increase their exposure to other
fire hazards.
Tenability to heat and toxic gases is strongly dose-related and so fractional
effective dose techniques are used. For example, the heat fractional effective dose
per second can be calculated by
1
Fheat = (3.71)
(60e (5.2−0.027T )
)
where: T is the temperature of the smoke (oC).
This relationship indicates that a temperature of 100oC could be tolerated for
12 minutes and 60oC for 35 minutes.
For toxicity (CO) the fractional effective dose per second is given by
C
Ftoxicity = (3.72)
90
where: C is the percentage concentration of CO.
This relationship indicates that 0.5 per cent CO could be tolerated for 3
minutes. A maximum concentration criterion of 1 per cent CO is also applied
and consideration should be given to selecting a CO conversion factor that will
be conservative.
For thermal radiation, a criterion of 2.5 kW/m2 is taken. Below this value
thermal radiation can be tolerated over extended periods of time and above the
level tolerability is measured in a few tens of seconds.
The visibility criterion may be either 5m or 10m depending on the nature of the
building and its ease of way-finding. For example, for a small space in a building, a
5m visibility criterion may be appropriate. Similarly, for a large space in a building
with exit doors clearly marked, a 10m visibility criterion may be appropriate.
Therefore, the hazard/time output for each control volume from the smoke
movement model can be integrated with a semi-infinite stream of people moving
away from the fire. As each individual moves down the building, they move from
one control volume to the next. Their dose can then be calculated based on the
level of hazard in each control volume as they passed through it.
3.6.6 Complex computational analysis
There is a range of computational methods available for analysing fluid or people
flow in similar ways. The advantages of these methods are that they model the
flow processes in much greater detail than control volume/hydraulic evacuation
models. The disadvantage is that they are relatively expensive to undertake and
so tend to be used for a limited number of fire risk assessments and for a small
number of cases. Figures 3.32 and 3.33 show examples of computational fluid
dynamics analysis and discrete evacuation analysis.
3.7 Dealing with uncertainty

This section describes the importance of pessimistic assumptions, sensitivity
analysis and/or uncertainty analysis in testing the robustness of a decision based
on an estimate of risk.
3.7.1 Sensitivity analysis
Sensitivity analysis can be used to draw useful conclusions in the first instance or
to assess the robustness of a decision based on probabilistic risk assessment.
Probabilistic risk assessment, like all fire engineering analysis, uses analysis
techniques and data to answer questions regarding fire safety. The analysis
techniques and data may have simplifying assumptions and limitations that mean
that they may not replicate the details of actual events. However, if meaningful
conclusions are to be drawn from an analysis, they should be sufficiently
representative that the correct fire safety decision is taken.
Figure 3.32 Example of computational fluid dynamics

Figure 3.33 Example of discrete evacuation analysis
If the results of the probabilistic risk analysis are well within the acceptance
criteria, then sensitivity analysis may not be needed. If, however, the results of
the probabilistic risk analysis are close to the acceptance criteria, then variations
in the variables may have a significant effect on the conclusions from the analysis
and sensitivity analysis should be used to assess this.
The first step of sensitivity analysis is to identify the variable(s) that are likely
to have the greatest impact on the results of the analysis. The variables can be
identified as those where:
• a small change may be magnified due to its role in an equation or analysis;

and/or
• its value may be subject to the greatest variability or uncertainty.
For example, a variable that has a value to the 1/3 power in an equation may
not have a large impact on the final results of the analysis. Variations in another
variable, which is to the 4th power in an equation, may have a significant impact
on the results of the analysis. If a variable is the only one in an equation or it is
used several times in the analysis, then it too may have a significant impact on the
results of the analysis.
The variables identified as potentially having a significant impact on the results
of the analysis can then be investigated in one of three ways:
• a single variable with an alternative value;

• a single variable over a range of values;
• a multiple point assessment of multiple variables.
A sensitivity analysis of a single variable with an alternative value is the

simplest approach. The analysis is repeated with an alternative (usually more
onerous) value to assess whether the conclusions of the analysis are robust. If
the assessment criteria are still satisfied, then the conclusions of the analysis are
further reinforced.
Sensitivity analysis based on a single alternative value of a variable is often not
very conclusive. Therefore, sensitivity analysis of a variable over a range of values
is used. The analysis is repeated and a graph produced showing the variation of the
results of the analysis against values of the variable. This provides much greater
insight into the relationship between the variable and the output of the analysis. If
the results of the analyses lie across the acceptance criteria then a critical value of
the variable can be identified and an assessment made of its implications.
Advanced methods of sensitivity analysis are available that allow more than one
variable to be varied at a time (see the Section 3.5 on Monte Carlo simulation).
The results of this analysis can then be presented in a table or, after applying
regression analysis, as a mathematical expression.
Estimates of risk should assess the implications of assumptions and bias in the
risk analysis. For life safety against absolute risk criteria, it is often possible to deal
with assumptions and bias by making them pessimistic so that they err on the side
of safety, i.e. such assumptions that tend to increase the predicted frequency and/
or consequence of an event. For comparative life risk analysis, the same approach
can be used as long as the assumptions or bias do not clearly favour one option
over another. For financial objectives, there is no natural ‘side of safety’ to err
on, since all options have costs and financial benefits this type of analysis should
normally be of best estimates rather than values of realistic or credible pessimism.
Sensitivity analysis is often used to test how robust a decision is to variations in
values that may have been subject to assumptions and/or bias.
There is guidance in British Standards Institute (2003) PD7974 Part 7 on
data sources and quality. Again the robustness of a decision to variations in data
values can be assessed using sensitivity analysis. The following suggest some key
questions that should be considered with respect to sources and quality of data.
Applicability:
• What is the set of cases that the data is drawn from?

• What case is the data measuring?
• How similar is my system to the cases considered?
• If the data is from another country, will variations in statutory controls or
design practices skew the data?
Quality:
• How old is the data (10 years may be considered a cut-off age for high-quality
data, but this will depend on the system and the extent and rate of its change)?
• Is collaborative data available?
• Is the data from statistical studies or based on engineering judgement?
Results:
• Do the answers look realistic?

• How sensitive are the results to questionable data?
3.7.2 Uncertainty analysis

The kinds of uncertainty that are built into the estimated levels of fire risk arise
from:
• the choice of tangible and credible failure cases from the wide range of events
that could be envisaged;
• the frequency of ignition for each case;
• the probability of failure or effectiveness of mitigation measures;
• the fire’s location, growth rate and peak heat release rate;
• the level of hazard generated by the fire;
• the validity of the smoke movement and human behaviour;
• uncertainties in the values used in the models for entrainment, vent flow,
number and distribution of people and their pre-movement times etc.;
• the accuracy of toxicology models in translating hazard levels and distributions
into lethal effects.
There are also uncertainties arising from the hazard identification process.
These are:
• failure to identify all the significant failure events;

• failure to include significant events that were identified.
These are problems of incompleteness rather than uncertainty and as such

cannot be addressed easily using uncertainty analysis.
Uncertainty analysis uses Monte Carlo analysis (see Section 3.5) or Latin
hypercube sampling to process probability distributions that represent the
uncertainty in a parameter or parameters. For application to fire risk assessment,
the output parameter would be the level of risk.
The method for estimating uncertainties in risk assessment is:
1 Identify the parameters that contribute most to the variations in risk levels,
with a few key parameters usually being chosen (Charters 1999), e.g.
a. the fire growth rate
b. the time to detect a fire
c. the number of people present
d. their p re-movement time.
2 A probability distribution is derived for each parameter.
3 Monte Carlo analysis randomly selects a value of each parameter and runs the
model and this is repeated a large number of times.
4 The output parameter results are then used to create a probability distribution
for which statistical properties like the mean, standard deviation and 95th
percentile can be determined.
Monte Carlo analysis depends on a large sample for its accuracy, whereas Latin
hypercube sampling selects from discrete segments of the distribution and so its
results are less sensitive to the number of calculations undertaken.
Where there is good data on the variability of a parameter, uncertainty analysis
can provide useful insight into the robustness of a decision based on an estimate.
Where there is uncertainty surrounding a point value due to lack of data, this may be
amplified in the ‘generation’ of a probability distribution, it is not clear whether un-
certainty analysis increases or decreases the robustness of the decision. In these cases,
sensitivity analysis provides much clarity and a clear audit trail in terms of robustness.
3.7.3 Safety factors
For many fire safety engineering components or subsystems, the performance may
be formulated in terms of two random variables, X and Y. The variable X represents
stress and Y the strength. Taking the compartment in a building as an example,
X is the severity of fire to which the structural boundaries of the compartment
are exposed and Y the fire resistance of the boundaries. Both fire severity and fire
resistance are usually expressed in units of time. Another example is concerned
with building evacuation, in which X is the time taken by a combustion product to
produce an untenable condition on an escape route and Y the time since the start
of ignition taken by an occupant to get through the escape route.
In the first example mentioned above, the compartment would ‘fail’ with
consequential damage to life and property if X exceeds Y, particularly during the
post-flashover stage. In the second example, ‘egress failure’ would occur with fatal
or non-fatal casualties if Y exceeds X. The objective of fire safety design is to
reduce the probability of failure to an acceptably small level. For estimating this
probability, two methods are generally adopted. The first method discussed in
this section involves partial safety factors and is semi-probabilistic. The second
method, discussed in the next section, is probabilistic and involves probability
distributions of X and Y; it is also known as the Beta method.
3.7.3.2 Characteristic values
The first step in this analysis is to select appropriate values for X and Y which
are typical or characteristic values representing the two random variables. These
values can be, for example, the mean or average values µx and µy of X and Y
or other statistical parameters such as median (50th percentile) or mode (the
most probable value with the highest relative frequency). A value corresponding
to some other percentile, e.g. 80th or 90th or 95th can also be selected as a
characteristic value for X or Y.
Consider a design problem in which failure would occur if X > Y or success if
Y ≥ X. For example, thermal failure of the compartment would occur if severity S
exceeds resistance R and success if R ≥ S. It is usual to provide a structural element
with minimum fire resistance, Rp, which is greater than the maximum severity, Sq,
likely to be encountered during the post-flashover stage. Rp and Sq can be regarded
as the characteristic values Rk and Sk of R and S.
Suppose µr and σr are the mean and standard deviation of fire resistance R and
µs and σs the mean and standard deviation of fire severity S. If the values of these
parameters are known, we can write
Rp = µr – tr σr (3.73)
Sq = µs + ts σs (3.74)
If VR and VS are the coefficients of variations given by
VR = σr / µr; VS = σs / µs (3.75)
we have
Rp = µr (1 – VR tr) (3.76)
Sq = µs (1 + VS ts) (3.77)
According to the Chebyshev inequality (La Valle, 1970), whatever may be the
probability distribution of S, the probability of fire severity exceeding Sq given by
Equation (3.74) is less than or equal to (1 / t 2s ). For instance, ts = 2 guarantees
a safety margin of at least 75 per cent (= 1 – 1 ). The probability of severity
22
exceeding Sq in this case is at most 0.25. The values of ts and Sq may be selected
according to any specified safety margin. It may be seen, for example, that ts =
3.16 would provide a safety margin of at least 90 per cent. The probability of
severity exceeding Sq in this case is at most 0.10. In the case of minimum fire
resistance, if tr = 3.16, the probability of resistance being less than Rp given by
Equation (3.73) would be at most 0.10 and the probability of resistance exceeding
Rp would be at least 0.90.
Suppose the probability distributions of R and S are also known, in addition to
their means and standard deviation. If these are normal, for example, the values
of tr and ts for any specified probability levels can be obtained from tables of the
standard Normal distribution. For example, ts = 1.96 corresponds to the fractile
value of 0.975 of the probability distribution of fire severity. In this case, the
probability of severity exceeding the value of Sq given by Equation (3.74) would
be 0.025. If ts = 2.33 which corresponds to the fractile value 0.99, the probability
of severity exceeding Sq is 0.01. The probability of fire resistance being less than
the value of Rp, given by Equation (3.73), would be 0.025 if tr = 1.96 and 0.01 if
tr = 2.33.
The mean, maximum or any other value representing the characteristic value
Sk of fire severity likely to be attained in a compartment, can be estimated with the
aid of an analytical model such as
Te = c . w . q (3.78)
where c is a constant depending on the thermal properties of the compartment

boundaries, w the ventilation factor and q the fire load density. The ventilation
factor is given by
Af
w= (3.79)
1
( AT AV h )2
where Af is the floor area of the compartment, AT the area of the bounding
surfaces of the compartment including the area of ventilation openings (AV) and
h the weighted mean ventilation height. With area in m2, h in m and q in MJ/m2,
fire severity Te is expressed in minutes. Formulae (3.78) and (3.79) relating to
equivalent time of fire exposure have been recommended by the CIB (1986).
In Equation (3.78) the parameters c and w may be regarded as constants
for any compartment with known or given structural (thermal) characteristics,
dimensions and area and height of ventilation openings. Fire load density q may
be considered as a random variable such that the mean severity μs is estimated by
inserting the average value μq of fire load density:
µs = c . w. µq (3.80)
The standard deviation σs of fire severity is given by:
σs = c . w . σq (3.81)
where σq is the standard deviation of fire load density. Then, from Equations
(3.75), (3.80) and (3.81), it may be seen that the coefficient of variation VS of
severity is equal to that of fire load density given by σq/µq.
The fire resistance required for a structural element of a compartment
may be based on the criterion that the minimum fire resistance Rp given by
Equation (3.73), exceeds the maximum severity Sq given by Equation (3.74). A
standard fire resistance test would indicate whether the structural element meets
this criterion or not. However, the fire resistance would be a random variable in a
real fire – see Ramachandran (1990). The variability depends on materials used.
For example, fire resistance of a gypsum board wall would have a greater variability
than the resistance of a concrete block wall. The resistance of a steel wall would
depend on the thickness of insulation, total mass of insulation and steel, average
perimeter of protective material and a factor representing the insulation heat
transmittance value for the material.
Fire resistance of a compartment composed of different structural elements
would not be the same as the fire resistance of any of these elements. Fire resistance
of a compartment is affected by weakness caused by penetrations, doors or other
openings in barriers. Sufficient data are not available for estimating realistically
the mean µr and standard deviation σr of the fire resistance of a compartment
in an actual fire. The values of these parameters can only be assumed according
to data provided by standard fire resistance tests and other experiments. These
tests and experiments can provide some indication of the standard deviation σr or
coefficient of variation VR as defined in Equation (3.75). For the sake of simplicity,
fire resistance may be assumed to have the same probability distribution as that of
fire severity, e.g. Normal.
The mean fire resistance µr required for a compartment is an output to be
estimated according to the input values µs and σs of fire severity. The output µr
should satisfy the design criterion that the minimum fire resistance Rp, as given by
Equation (3.73), exceeds the maximum severity Sq as given by Equation (3.74). Rp
and Sq include safety margins provided by the standard deviations σr and σs and
the parameters tr and ts.
As defined in Equations (3.78) and (3.79), fire severity is the product of several
factors. Based on data from fire tests, fire resistance in some cases is also expressed
as the product of some factors, e.g. thin wall steel members (Homer, 1979). In
all such cases, it may be considered necessary to take account of uncertainties
governing all the factors. Generally, if a variable y is a product of several variables
x1, x2, x3, … which are mutually independent, the mean of y is approximately given
by the product
y = x1 ⋅ x2 ⋅ x3 … (3.82)
where y , x1 , x2 , x3 , … are the means of the variables. The coefficient of variation

of y is approximately given by:
V 2y = V 1 + V 22 + V 23 + ....... (3.83)
2
where V1, V2, V3, … are the coefficients of variation of x1, x2, x3, ….
As derived by Hahn and Shapiro (1967), the results in Equations (3.82) and
(3.83) are based on an application of truncated Taylor series expansion of the
function
y = f (x1, x2, …) (3.84)
The second and higher derivatives of the function are neglected in the above-
mentioned expansion. Ramachandran (1998a) discussed the above results in a
detailed report on the various aspects of probabilistic evaluation of structural fire
protection.
For the second example relating to building evacuation, the design criterion
is that the total evacuation time H, Y as defined previously, should not exceed
the time F, X as defined earlier, taken by a combustion product, e.g. smoke, to
travel from the place of fire origin and produce an untenable condition, e.g. visual
obscuration on an escape route. The total time H is the sum of three periods.
In sequential order, the first period D is the time taken to detect or discover the
existence of a fire after it started. The second period B is known as ‘recognition
time’ or ‘gathering phase’ in human behaviour studies. This period is the elapsed
time from discovery of fire to the commencement of evacuation. The third period
E, known as ‘design evacuation time’, is the time taken by an occupant to reach
the entrance to an escape route, e.g. protected staircase, after leaving his/her place
of occupation.
The time period D depends on the presence or absence of automatic fire
detection systems or suppression systems such as sprinklers. A characteristic
value for D can be estimated from fire statistics or detector tests, together with
its standard deviation. Human behaviour studies suggest a characteristic value
of 2 minutes for B. But this parameter for any occupancy type can be estimated
by carrying out evacuation exercises. For any type of building, the characteristic
value of E and its standard deviation can be estimated from fire drills or computer
models of evacuation. A value of 2.5 minutes for E has been recommended in
British Standard BS 5588. The actual value of E would depend on building type
and the physical (disabled etc.) and mental conditions of the occupants, apart
from other factors such as widths of staircases and exits. Deterministic models and
associated computer packages can be used to estimate the characteristic value and
standard deviation of F for any type of building. By reducing the rate of growth of
fire and smoke, sprinklers would increase the value of F if they fail to extinguish a
fire. Sprinklers also have a high probability of extinguishing a fire in which case F
will have an infinite or high value.
The mean value µh of total evacuation time H is the sum of the mean values of
D, B and E. The standard deviation σh of H is given by:
σ2h = σ2d + σ2b + σ2e (3.85)
where σd, σb and σe are the standard deviations of D, B and E. For any escape
route and place of fire origin, the mean value µf is the sum of the means of the
F values for different combustion products. By considering different places of
fire origin, escape routes and combustion products, the overall mean value of F
can be estimated for any building or any floor of the building. An estimate of
this mean is given by the sum of mean values of F for all the factors mentioned
above. Following Equation (3.85), the square of the standard deviation of the
overall value of F is the sum of squares of the standard deviations of the factors.
The formula in Equation (3.83) is applicable for estimating approximately the
coefficient of variation of the overall value of H or F.
Ramachandran (1993) discussed in detail the model described above for
building evacuation and has derived equations similar to (3.73) to (3.77).
3.7.3.3 Design values

In practical fire safety engineering, it is necessary to determine design values
which include partial safety factors αx and αy to account for uncertainties in the
estimation of characteristic values for the random variables X and Y. The sources
of uncertainties are mainly parameters included in, or excluded from, analytical
models, data used, hypotheses and assumptions. The corrections for uncertainties
should be in the direction of greater safety after assigning values greater than unity
for the partial safety factors αx and αy.
Consider first the fire protection given by the fire resistance of the structural
boundaries of a compartment. With the partial safety factor αr greater than unity,
the design value Rd for fire resistance can be estimated by
Rd = Rk / αr (3.86)
where Rk is the characteristic value. Rd will be less than Rk according to Equation

(3.86). This design condition will also be satisfied if the minimum value Rp in
Equation (3.73) is considered as the design value and the mean value µr as the
characteristic value. In this case, from Equations (3.76) and (3.86) αr is the
reciprocal of (1 – VRtr).
The formula for the design value Sd for fire severity is
Sd = αs.Sk (3.87)
where Sk is the characteristic value and αs greater than unity is the partial safety
factor. Accordingly, Sd will be greater than Sk. This design condition will also be
satisfied if the maximum value Sq in Equation (3.74) is considered as the design
value and the mean value µs as the characteristic value. In this case, from Equations
(3.77) and (3.87), αs is equal to (1 + Vsts).
For example, if the estimate of Rk is correct to 15 per cent, then
αr = 1.176, Rd = 0.85Rk
It may also be seen that, if VR = 0.2 and a value of 1.96 is adopted for tr, αr =
1.64.
Likewise, if the estimate of Sk is known within 25 per cent, then
αs = 1.25, Sd = 1.25Sk
Also, if VS = 0.2 and ts = 1.96, then αs = 1.39.
Since the design requirement is Rd ≥ Sd, from Equations (3.86) and (3.87)
Rk ≥ αr . αs . Sk (3.88)
Equation (3.88) provides a method for adjusting the characteristic value Sk of

fire severity to take account of uncertainties with the aid of partial safety factors αr
and αs. For adopting to reliability requirements differing from the average or normal
requirements, additional adjustment factors can be included on the right-hand
side of Equation (3.88) as additional (multiplicative) partial safety factors. The
adjustments for a particular building or type of building should reflect the increase
or decrease in fire risk from the average risk, compartment size, effectiveness of
sprinklers (if installed), efficiency of fire brigade and other such factors affecting
fire severity.
For the evacuation model, the design value Hd for the total evacuation time H
is given by
Hd = Hk . αh (3.89)
where Hk is the characteristic value and αh the partial safety factor greater than
unity. The maximum total evacuation time
Hq = µh (1 + Vh th)
can be considered as the design value and the mean μh as the characteristic value.
In this case,
αh = (1 + Vh th)
where Vh is the coefficient of variation of H and µh a parameter similar to tr in

Equation (3.73) or ts in Equation (3.74).
The design value for the combustion product time F is given by
Fd = Fk / αf (3.90)
where Fk is the characteristic value and αf the partial safety factor greater than
unity. The minimum value of F given by
Fp = µf (1 – Vf tf)
can be considered as the design value and the mean value μf as the characteristic
value. In this case, αf is the reciprocal of (1 – Vf tf). The parameter Vf is the
coefficient of variation of F and tf is a constant similar to th.
Since the design criterion for successful evacuation is Hd ≤ Fd,
Hk ≤ Fk / αf . αh or Fk ≥ Hk . αf . αh (3.91)
The purpose of including the partial safety factors αf and αh in the design
process is to ensure that the maximum or any other design value for the total
evacuation time H does not exceed the minimum or any other design value for the
combustion product time F.
In the semi-probabilistic approach discussed in this section, the choices for
the values of the partial safety factors are usually based on the expert judgement
of the fire safety engineer and the quality of information available to him/her for
estimating the values of the parameters. Instead of adopting such empirical and
intuitive methods, the partial safety factors can be derived from the probability
distributions of the variables involved. This method, based on the ‘design point’,
was described in detail by Ramachandran (1998a).
3.7.4 Beta method
3.7.4.1 Probabilistic design criterion

In a probabilistic procedure, the deterministic design criterion, Y ≥ X, discussed in
Section 3.7.3.1 is modified to:
P (Y ≥ X) ≥ 1 – Pg (3.92)
where P (Y ≥ X) denotes the probability of strength Y being greater than or equal

to stress X; this is equivalent to the probability of success. Pg is a (small) target
probability (risk) acceptable to a property owner or society at large. The value
of Pg depends on consequences in terms of damage to life and property if failure
occurs. The probability of failure should be less than Pg:
P (Y < X) < Pg (3.93)
If Y is fire resistance R and X is fire severity S the probabilistic design criterion

for compartment success is
P (R ≥ S) ≥ 1 – Pg (3.94)
Probability of compartment failure should be less than Pg:
P (R < S) < Pg (3.95)
For building evacuation, X is the time F taken by a combustion product to

produce an untenable condition on an escape route and Y is the total evacuation
time H. In this case, Equation (3.92) is modified to
P (H ≤ F) ≥ 1 – Pg (3.96)
for egress success. Probability of egress failure should be less than Pg:
P (H > F) < Pg (3.97)
Probabilistic methods are concerned with the evaluation of Pg and (1 – Pg) for
different combinations of X and Y. The evaluation procedure takes account of
uncertainties through the probability distributions of X and Y.
3.7.4.2 Univariate approach

In this approach, discussed by Ramachandran (1998a), the stress variable X only is
considered as a random variable with the strength variable Y treated as a constant.
This is the approach traditionally adopted by fire safety engineers for determining
the fire resistance required for a structural element. The cumulative probability
distribution function of severity S is denoted by Fs(x), the probability of severity
being less than or equal to x. If the fire resistance R of a structural element is
set equal to x, the probability of severity exceeding R is [1 – Fs(R) ] which is the
probability of failure of the element.
Consider first the exponential probability distribution for fire severity S:
Fs(x) = 1 – exp (–λs x) (3.98)
According to a property of this distribution, λs is the reciprocal of the mean

value μs of fire severity. Baldwin (1975) estimated μs = 25 minutes for office
buildings such that λs = 0.04. It may be seen from Equation (3.98) that if R = x =
25 minutes, the probability of failure, would be
1 – Fs(R) = exp (–λsR)

= exp (–1) = 0.37
which is not a small quantity. But the probability of failure would reduce to 0.09 if
R = 60 minutes and to 0.03 if R = 90 minutes and so on.
If fire severity S has a normal distribution with mean μs and standard deviation
σs, the standardised random variable t given by
t = (S – μs) / σs (3.99)
has a standard normal distribution with mean zero and standard deviation unity.
From Equation (3.99)
S = μs + t σs (3.100)
If the fire resistance R of a structural element is set equal to S with t = 0 such

that R = μs, the probability of success or failure of the element in a fire is 0.5. But
if R = S with t = 1.96 in Equation (3.100), the probability of success given by the
cumulative distribution function of t is 0.975; the probability of failure is 0.025.
For t > 1.96, the probability of failure would be less than 0.025. For t = 2.33,
the probability of success would be 0.99 with 0.01 for the probability of failure.
Probabilities of success and failure for different values of t can be obtained from a
table of standard Normal distribution. Using this table, the fire resistance required
to meet any target level for the probability of failure can be determined by using,
in Equation (3.100), the value of t corresponding to this level.
3.7.4.3 Bivariate approach

In this approach, more commonly known as the Beta method, both the stress and
strength variables are considered as random variables affected by uncertainties –
see Ramachandran (1998a) and Magnusson (1974). The difference (Y – X) is the
‘safety margin’ which is also referred to as the ‘state function’. The expected value
of the random variable
Z = Y – X (3.101)
is given by
μz = μy – μx (3.102)
where μy and μx are the mean values of Y and X. The standard deviation of z is
given by
1
σz = (σ2y + σ2x ) 2 (3.103)
where σy and σx are the standard deviations of Y and X. The ‘safety index’ β is
given by
β = μz / σz (3.104)
Consider first the determination of fire resistance required for a structural

element to satisfy a specified level for the probability of failure. If the mean and
standard deviation of fire resistance R are μr and σr and the mean and standard
deviation of fire severity S are μs and σs, the mean and standard deviation of the
state function z = R – S are
μz = μr – μs (3.105)
1
σz = (σ2r + σ2s ) 2 (3.106)
The ‘safety index’ β is given by

1
β = µ z / σz = (µ r − µ s ) / (σ2r + σ2s ) 2 (3.107)
The fire resistance required may be set according to μr given by
1
µ r = µ s + β(σ2r + σ2s ) 2 (3.108)
If R and S have normal distributions, the parameter β has a standard normal

distribution. In this case, the value of β corresponding to any target level for
probability of failure can be obtained from a table of standard Normal distribution.
This value can then be inserted in Equation (3.108) to provide the fire resistance
μr required for the structural element. As discussed in Section 3.7.4.2 in terms of
the variable t, the probability of structural failure would be 0.5 if β = 0 and μr =
μs, less than 0.5 if β is positive and μr > μs and greater than 0.5 if β is negative and
μr < μs. The probability of failure would be 0.025 if β = 1.96, 0.01 if β = 2.33 and
0.001 if β = 3.09. For a selection of values of β, probabilities of structural success
and failure are given in Table 3.17.
Consider now the determination of the total evacuation time H that will satisfy
a specified level for the probability of egress failure – see Ramachandran (1993).
The state function in this case may be written as z = F – H such that the mean
and standard deviation of z are
μz = μf – μh (3.109)
1
σz = (σ2f + σ2h ) 2 (3.110)
The parameters μf and σf are the mean and standard deviation of F and μh and
σh are the mean and standard deviation of H. The safety index β is given by
β = µ z / σz
(3.111)
1
= (µ f − µ h ) / (σ2f + σ ) 2 2
h
The total evacuation time required may be set according to μh in the following
equation:
1
µ h = µ f − β(σ2f + σ2h ) 2 (3.112)
If F and H have normal distributions, as discussed earlier, β has a standard

normal distribution. The probability of egress failure would be 0.5 if β = 0 and μh
= μf, less than 0.5 if β is positive and μh < μf and greater than 0.5 if β is negative
and μh > μf .The probability of failure would be 0.025 if β = 1.96, 0.01 if β = 2.33
and 0.001 if β = 3.09. The figures in Table 3.17 can be used in conjunction with
Equation (3.112) to determine the total evacuation time according to a specified
level for the probability of egress failure. It should be noted that β has a positive
sign attached to it in Equation (3.108) but a negative sign in Equation (3.112).
To satisfy the condition specified in Equation (3.112), it may be necessary to
install automatic detectors and/or sprinklers if the building is not already equipped
with these devices. These devices would reduce the detection time D and hence
reduce μh. Sprinklers would also increase the combustion product time μf. The
total evacuation time μh can also be reduced by providing additional or wider
staircases which will reduce the design evacuation time E.
If egress failure occurs, there is a probability, K, that one or more deaths may
occur. This probability can be estimated by analysing fire statistics. According to an
analysis of these statistics for the period 1978 to 1988 by Ramachandran (1993),
the average detection time for single and multiple occupancy dwellings was 10
minutes. With B = 2 minutes and E = 3 minutes, the average total evacuation
time, μh, was 15 minutes. The mean value of combustion product time, μf, for
causing death, was assumed to be 15 minutes such that the probability of egress
failure was estimated to be 0.5. With a fatality rate per fire of 0.013, the value of K
was estimated as 0.026 ( = 0.013/0.5).
3.7.4.4 Safety factor

Corresponding to the safety index β, a safety factor θ may be defined as the
ratio between the mean values of the stress and strength variables. In the case of
structural failure
θ = μr / μs (3.113)
such that, from Equation (3.107)

1
β = ( θ −1) / (VR2 θ2 + Vs2 ) 2 (3.114)
where VR and VS are coefficients of variation given by
VR = σr / μr ; VS = σs / μs
For facilitating calculations, Equation (3.114) may be inverted to give
1 + β (VR2 + Vs2 − β2 VR2 Vs2 )½ (3.115)

θ=
1 − β2 VR2
Equation (3.115) has a solution only if VR < 1/β.

If it is is assumed that VR = Vs = r, then Equation (3.115) reduces to
1 + β r ( 2 − β2 r 2 )½
θ=
(3.116)
1 − β2 r 2
For r = 0.15, the values of θ corresponding to those of β given in Table 3.17 for
different failure probabilities.
In the safety factor approach, the mean value of fire resistance, μr, should be
set equal to or greater than the value given by θμs. Suppose, for example, the
probability of structural failure should be less than 0.005. In this case, from Table
3.17 for Pg = 0.005, β = 2.5758 and θ = 1.7934 if r = 0.15. Hence, to achieve
Table 3.17 Probabilities of success and failure (standard normal distribution)

Probability of success Probability of failure β θ
(1–Pg ) (Pg ) (r = 0.15)
0.0001 0.9999 –3.7190 0.3993
0.0005 0.9995 –3.2905 0.4573
0.0010 0.9990 –3.0902 0.4848
0.0020 0.9980 –2.8782 0.5145
0.0025 0.9975 –2.8070 0.5245
0.0050 0.9950 –2.5758 0.5576
0.0100 0.9900 –2.3263 0.5941
0.0250 0.9750 –1.9600 0.6494
0.0500 0.9500 –1.6449 0.6990
0.1000 0.9000 –1.2816 0.7587
0.2000 0.8000 –0.8416 0.8355
0.3000 0.7000 –0.5244 0.8945
0.4000 0.6000 –0.2533 0.9477
0.5000 0.5000 0.0000 1.0000
0.6000 0.4000 0.2533 1.0552
0.7000 0.3000 0.5244 1.1180
0.8000 0.2000 0.8416 1.1969
0.9000 0.1000 1.2816 1.3181
0.9500 0.0500 1.6449 1.4307
0.9750 0.0250 1.9600 1.5398
0.9900 0.0100 2.3263 1.6832
0.9950 0.0050 2.5758 1.7934
0.9975 0.0025 2.8070 1.9064
0.9980 0.0020 2.8782 1.9437
0.9990 0.0010 3.0902 2.0626
0.9995 0.0005 3.2905 2.1869
0.9999 0.0001 3.7190 2.5043
the desired target, the mean fire resistance, μr, should be set equal to or greater
than 1.79μs.
For the evacuation model, the safety factor is given by
θ = μf / μh (3.117)
such that, following Equation (3.114)
1
β = ( θ −1) / (Vf2 θ2 + Vh2 ) 2
where Vf and Vh are the coefficients of variation of F and H given by
Vf = σf / μf ; Vh = σh / μh
Also,
1 + β (Vf2 + Vh2 − β2 Vf2 Vh2 )½ (3.118)

θ=
1 − β2 Vf2
Equation (3.116) is applicable if Vf = Vh = r.
If r = 0.15, from Table 3.17, β = 2.3263 and θ = 1.6832, for a target maximum
value of 0.01 for the probability of egress failure. To achieve this target, the mean
value, μh, of total evacuation time should not exceed 0.59μf. This result follows
from Equation (3.117) according to which μf should be greater than θμh or μh
should be less than (μf /θ). Under such a protection for life safety, the fatality rate
per fire would be less than 0.00026 (= 0.01 × 0.026) if the probability, K, of one
or more deaths occurring given egress failure is 0.026 as estimated earlier. The
value of K would vary from one type of occupancy to another.
Thus, the fatality rate per fire in single and multiple occupancy dwellings can
be reduced to 0.00026 from the level of 0.013 estimated from the data for 1978–
1988 if the total evacuation time, μh, is reduced to 9 minutes (= 0.59 × 15) from
the level of 15 minutes. With B = 2 minutes and E = 3 minutes, the detection or
discovery time, D, should be reduced to 4 minutes from the level of 10 minutes.
3.7.4.5 Log normal safety index

For structural fire resistance, Esteva and Rosenblueth (1971) put forward a design
format based on the state variable
y = loge (R/S) (3.119)
which is applicable if resistance R and severity S have log normal probability

distributions. Approximate values of the mean y and standard deviation σy of y
are given by
y = log e (µ r / µ s )
1
σy = (VR2 + VS2 ) 2
where, as defined earlier, μr and VR are the mean and coefficient of variation of R
and μs and Vs are the mean and coefficient of variation of S.
The safety index corresponding to the state variable y in Equation (3.119) is
βER = y / σy (3.120)
1
= log e (µ r / µ s ) / (VR2 + V )
S
R 2
The fire resistance required may be determined according to μr given by

1
log e µ r = log e µ s + βER (VR2 + Vs2 ) 2 (3.121)
The safety factor θER is given by

 1
θ ER = (µ r / µ s ) = exp  βER (VR2 + Vs2 ) 2  (3.122)
 
 
The mean fire resistance μr should be set equal to or greater than βER . μs.
Values of βER for different probabilities of structural failure are the same as
those in Table 3.17. θER = 1 if βER = 0, less than 1 if βER is negative and greater
than 1 if βER is positive. If VR = Vs = r:
θ ER = exp  βER r 2  (3.123)

 
Calculations based on Equation (3.123) show that, for any target probability of
failure less than 0.3, the value of θ given by Equation (3.115) is marginally greater
than the corresponding value of θER. Hence, in this range of failure probability
which is of interest in structural fire safety design, an assumption of normal
distributions for R and S would provide a slightly greater safety margin than an
assumption of log normal distributions.
It is a somewhat complex statistical problem to construct an appropriate safety
index if both R and S have exponential probability distributions or they have
different distributions. The safety index proposed in Equation (3.107) or (3.120)
would be sufficient for all practical purposes. Ramachandran (1998a) discussed in
detail other problems such as ‘design point’, full probabilistic approach, extreme
value technique and determination of tolerable failure probability.
References
Aoki, Y (1978), Studies on Probabilistic Spread of Fire, Research Paper No. 80, Building
Research Institute, Tokyo.
Atreya A (2003), Convection heat transfer, Section 1 Chapter 3, SFPE Handbook of Fire
Protection Engineering, 3rd edn, SFPE/NFPA, Gaithersburg, MD.
Baldwin, R (1975), Economics of Structural Fire Protection, Current Paper CP 45/75.,
Building Research Establishment, Watford.
Baldwin, R and Fardell, L G (1970), Statistical Analysis of Fire Spread in Buildings. Fire
Research Note 848, Fire Research Station, Borehamwood.
Beck, V R (1987), A cost-effective decision-making model for fire safety and protection,
Fire Safety Journal, 12, 121–138.
Bengtson, S and Hagglund, B (1986), The use of a zone model in fire engineering
application, Fire Safety Science: Proceedings of the First International Symposium,
Hemisphere Publishing Corporation, New York.
Bengtson, S and Laufke, H (1979/80), Methods of estimation of fire frequencies, personal
safety and fire damage, Fire Safety Journal, 2, 167–180.
Bengtson, S and Ramachandran, G (1994), Fire growth rates in underground facilities,
Fire Safety Science: Proceedings of the Fourth International Symposium, National Institute
of Standards and Technology, Gaithersburg, MD.
Benktander, G (1973), Claims frequency and risk premium rate as a function of the size
of the risk, ASTIN Bulletin, 7, 119–136.
Berlin, G N (1980), Managing the variability of fire behaviour, Fire Technology, 16, 287–
302.
British Standards Institute (2003), PD 7974 Code on the application of fire safety
engineering principles to the design of buildings, Part 7: Probabilistic risk assessment,
British Standards Institute, London.
BS 5760 Reliability of systems, equipment and components, British Standards Institute,
London.
Carslaw H S and Jaeger J C (1959), Conduction of Heat in Solids, 2nd edn, Oxford
University Press, Oxford.
Charters D (1999), What does quantified fire risk assessment have to do to become an
integral part to design decision-making, International Symposium on Uncertainty,
Risk and Reliability, SFPE, Bethesda, MD.
Charters D A (1992a), Fire risk assessment of rail buildings, Proceedings of the 1st
International Conference on Safety in Road and Rail Buildings, ITC, Basel.
Charters, D A (1992b), Fire Safety Assessment of Bus Transportation, C437/037, Institution
of Mechanical Engineers, London.
Charters D A (1997), Risk assessment – the reliability and performance of systems,
FIREX, NEC, Birmingham.
Charters, D A (1998), Fire safety at any price?, Fire Prevention, 313, October, 12–15.
Charters D A and Smith, F M (1992), The Effects of Materials on Fire Hazards and Fire
Risk Assessment, C438/017, Institution of Mechanical Engineers, London.
Charters D A, Gray W A and McIntosh A C (1994), A computer model to assess fire
hazards in buildings, Fire Technology, 30, 1, 135–155.
Charters D A, Barnett J, et al. (1997), Assessment of the probabilities that staff and/or
patients will detect fires in hospitals, Proceedings of the Fifth International Symposium of
Fire Safety Science, Society of Fire Protection Engineers, Melbourne.
CIB W14 (1986), Design guide: structural fire safety, Fire Safety Journal, 10, 2, 77–154.
CIBSE TM19 (1995), Relationships for Smoke Control Calculations, Chartered Institution
of Building Service Engineers, London.
Connolly, R J and Charters, D A (1997), The use of probabilistic networks to evaluate
passive fire protection measures in hospitals, Proceedings of the Fifth International
Symposium on Fire Safety Science, Society of Fire Protection Engineers, Melbourne.
Coward, S K D (1975). A Simulation Method for Estimating the Distribution of Fire Severities
in Office Rooms. Current Paper CP 31/75. Building Research Establishment, Watford.
Crossman R and Zachary W (1974), Occupant response to domestic fire incidents, US
National Fire Protection Association Annual Conference.
Cullen W(1990), The Public Inquiry into the Piper Alpha Disaster, HMSO, London.
DOE (1996), Data for the Application of Probabilistic Risk Assessments to the Evaluation of
Building Fire Safety – Appendix II, Department of Environment, London.
Drysdale, D, (1999) An Introduction to Fire Dynamics, 2nd edn, Wiley, Chichester.
Dusing J W A, Buchanan A H and Elms D G (1979), Fire Spread Analysis of Multi-
compartment Buildings, Research Report 79/12, Department of Civil Engineering,
University of Canterbury, New Zealand.
Elms D G and Buchanan A H (1981), Fire Spread Analysis of Buildings, Research Report
R35, Building Research Association of New Zealand, Auckland.
Emmons H (1991), The ceiling jet in fires, Proceedings of the 3rd International Symposium
of Fire Safety Science, Elsevier, London.
Emmons H (2003), Vent flows, Section 2 Chapter 5, SFPE Handbook of Fire Protection
Engineering, 3rd edn, SFPE/NFPA, Gaithersburg, MD
Esteva L and Rosenblueth E (1971), Use of reliability theory in building codes, Conference
on Application of Statistics and Probability to Soil and Structural Engineering, Hong
Kong, September.
Evans D (2003),Ceiling jet flows, Chapter 2 Section 4, SFPE Handbook of Fire Protection
Engineering, 3rd edn, SFPE/NFPA, Gaithersburg, MD.
Evers E and Waterhouse A (1978), A Computer Model for Analysing Smoke Movement in
Buildings, Current Paper CP 69/78, Building Research Establishment, Watford.
Fahy R F (1985), Building fire simulation model – an overview, Fire Safety Journal, 9,
189–203.
Fardell P J and Kumar S (1991), Fires in Buses: A Study of Life Threat and the Efficacy of
Bus Garage Sprinkler Protection for London Buses Ltd. Part A Life Threat Experimental
Studies and Predictive Computer Modelling, LPC Draft Report.
Finucane M and Pinkney D (1988), Reliability of Fire Protection and Detection Systems,
SRD R431, UKAEA, Harwell.
Government Statistics Service (1992), Fire Statistics United Kingdom 1990, HMSO,
London.
Green A E and Bourne A J (1972), Reliability Technology, Wiley, London.
Gumbel, E J (1958), Statistics of Extremes, Columbia University Press, New York.
Hahn, G J and Shapiro, S S (1967), Statistical Models in Engineering, Wiley, New York.
Hansen-Tangen E and Baunan T (1983), Fire risk assessment by simulation – Fire sim,
Fire Safety Journal, 5, 205–212.
Health and Safety Executive (1989), Quantified Risk Assessment: Its Input to Decision
Making, HMSO, London.
Heskestad G (2003), ‘Fire plumes’, Section 2, Chapter 2, SFPE Handbook of Fire Protection
Engineering, 3rd edn, SFPE/NFPA, Gaithersburg, MD.
Homer R D (1979), The protection of cold-form structural elements against fire,
Proceedings of International Conference on Thin-Wall Structures, Wiley, New York.
Johnston B D and Matthews R H (1982), Noncoherent Structure Theory: A Review and its
Use in Fault Tree Analysis, SRD R 245, UKAEA, Harwell.
La Valle I H (1970), An Introduction to Probability, Decision and Inference, Holt, Rinehart
and Winston, New York.
Ling W T C and Williamson R B (1986), The modelling of fire spread through probabilistic
networks, Fire Safety Journal, 9, 287–300.
Magnusson S E (1974), Probabilistic Analysis of Fire Exposed Steel Structures, Bulletin 27,
Lund Institute of Technology, Lund, Sweden.
Mandelbrot, B (1964), Random walks, fire damage amount and other Paretian risk
phenomena, Operations Research, 12, 582–585.
McCaffrey B (2003), Flame height, Section 2 Chapter 1, SFPE Handbook of Fire Protection
Engineering, 3rd edn, SFPE/NFPA, Gaithersburg, MD
Mirchandani, P B (1976). Computations and Operations Research, 3, Pergamon Press,
Oxford.
Morishita, Y (1977), Establishment of Evaluating Method for Fire Safety Performance Report,
Research Project on Total Evaluating System on Housing Performances, Building
Research Institute, Tokyo.
Morishita, Y (1985), A stochastic model of fire spread, Fire Science and Technology, 5, 1,
1–10.
NHS Estates (1996), Fire Practice Note 9 – NHS Healthcare Fire Statistics 1994/5, HMSO,
London.
North, M A (1973), The Estimation of Fire Risk of Various Occupancies, Fire Research Note
989, Fire Research Station, Borehamwood.
Phillips, W G B (1992), The Development of a Fire Risk Assessment Model, Information
Paper IP 8/92, Building Research Establishment, Watford.
Phillips, W G B (1995), Computer simulation for fire protection engineering. SFPE
Handbook of Fire Protection Engineering, 2nd edn, Section 5, Chapter 1, National Fire
Protection Association, Quincy, MA.
Platt, D G (1989), Modelling Fire Spread: A Time-based Probability Approach, Research
Report 89/7, Department of Civil Engineering, University of Canterbury, New
Zealand.
Purser, D A (2003), Toxicity assessment of combustion products, SFPE Handbook of Fire
Protection Engineering, SFPE/NFPA 3rd Edition.
Quintiere J G (1989), ‘Fundamentals of enclosure fire zone models’, Journal of Fire
Protection Engineering, Volume 1 (3).
Quintiere J G (2003), ‘Compartment fire modelling’, Section 3 Chapter 5, SFPE Handbook
of Fire Protection Engineering, 3rd edn, SFPE/NFPA, Gaithersburg, MD.
Ramachandran, G (1969), The poisson process and fire loss distribution, Bulletin of the
International Statistical Institute, 43, 2, 234–236.
Ramachandran, G (1970), Fire Loss Indexes, Fire Research Note 839, Fire Research
Station, Borehamwood.
Ramachandran, G (1974). Extreme value theory and large fire losses. ASTIN Bulletin, 7,
3, 293–310.
Ramachandran, G (1975), Extreme order statistics in large samples from exponential
type distributions and their application to fire loss, Statistical Distributions in Scientific
Work, vol 2, D Reidel Publishing, Dordrecht.
Ramachandran, G (1979/80), Statistical methods in risk evaluation, Fire Safety Journal,
2, 125–145.
Ramachandran, G (1980), Economic Value of Automatic Fire Detectors, Information Paper
IP27/80, Building Research Establishment, Fire Research Station, Borehamwood.
Ramachandran, G (1982), Properties of extreme order statistics and their application to
fire protection and insurance problems, Fire Safety Journal, 5, 59–76.
Ramachandran, G (1985), Stochastic modelling of fire growth, Fire Safety: Science
and Engineering, ASTM STP 882, American Society for Testing and Materials,
Philadelphia, PA.
Ramachandran, G (1986), Exponential model of fire growth, Fire Safety Science.
Proceedings of the First International Symposium, Hemisphere Publishing Corporation,
New York.
Ramachandran, G (1988a), Probabilistic approach to fire risk evaluation, Fire Technology,
24, 3, 204–226.
Ramachandran, G (1988b), Extreme value theory, Section 4, Chapter 4, SFPE Handbook
of Fire Protection Engineering, 1st edn, National Fire Protection Association, Quincy,
MA (reproduced in 2nd edn, 1995, Section 5, Chapter 3).
Ramachandran, G (1990), Probability-based fire safety code, Journal of Fire Protection
Engineering, 2, 3, 75–91.
Ramachandran, G (1991), Non-deterministic modelling of fire spread, Journal of Fire
Protection Engineering, 3, 2, 37–48.
Ramachandran, G (1993), Probabilistic evaluation of design evacuation time. Proceedings
of the CIB W14 International Symposium on Fire Safety Engineering, University of Ulster,
Belfast.
Ramachandran, G (1995a), Stochastic models of fire growth, Section 3, Chapter 15,
SFPE Handbook of Fire Protection Engineering, 2nd edn, National Fire Protection
Ramachandran, G (1995b), Heat output and fire area, Proceedings of the International
Conference on Fire Research and Engineering, CRC Press, Boca Raton, FL.
Ramachandran, G (1998a). Probabilistic Evaluation of Structural Fire Protection – A
Simplified Guide. Fire Note 8, Fire Research Station, Building Research Establishment,
Borehamwood.
Ramachandran, G (1998b). The Economics of Fire Protection. E & F N Spon, London.
Ramachandran, G (2002), Stochastic models of fire growth, Section 3, Chapter 15, SFPE
Handbook of Fire Protection Engineering, 3rd edn, National Fire Protection Association,
Quincy, MA.
Ramachandran, G and Chandler, S E (1984), Economic Value of Early Detection of Fires
in Industrial and Commercial Premises, Information Paper IP13/84. Building Research
Establishment, Fire Research Station, Borehamwood.
Rasbash, D, Ramachandran, G, Kandola, B, Watts, J M and Law, M, (2004), Evaluation
of Fire Safety, Wiley, Chichester.
Reynolds C and Bosley K (1995), Domestic First Aid Firefighting, Research Report 65, Fire
Research and Development Group, London.
Rockett J A and Milke J A (2003), ‘Conduction of heat in solids’, SFPE Handbook of Fire
Protection Engineering, 3rd edn, National Fire Protection Association, Quincy, MA.
Rogers, F E (1977), Fire Losses and the Effect of Sprinkler Protection of Buildings in a Variety
of Industries and Trades, Current Paper CP 9/77, Building Research Establishment, Fire
Research Station, Borehamwood.
Rutstein, R (1979), The estimation of the fire hazard in different occupancies, Fire
Surveyor, 8, 2, 21–25.
Rylands S, David P, McIntosh A C and Charters D A (1998), Predicting fire and smoke
movement in buildings using zone modelling, Proceedings of the 3rd International
Conference on Safety in Road and Rail Buildings, International Technical Conferences,
Nice.
Sasaki, H and Jin, T (1979), Probability of Fire Spread in Urban Fires and their Simulations,
Report No 47, Fire Research Institute of Japan, Tokyo.
Scoones K, (1995), FPA large fire analysis’, Fire Prevention Journal, January/February, Fire
Protection Association (Loss Prevention Council), Moreton in Marsh.
Shpilberg, D C (1974), Risk Insurance and Fire Protection: A System Approach. Part 1:
Modelling the Probability Distribution of Fire Loss Amount, Technical Report 22431,
Factory Mutual Research Corporation, Norwood, MA.
Shpilberg, D C (1975), Statistical decomposition analysis and claim distribution for
industrial fire losses, Twelfth ASTIN Colloquium, Portugal, September–October 1975.
Tien C L, Lee K Y and Stretton A J (2003), ‘Radiation heat transfer’, Section 1 Chapter
4, SFPE Handbook of Fire Protection Engineering, 3rd edn, National Fire Protection
Vantelon et al. (1991), Investigation of fire induced smoke movement in tunnels and
stations: An application to the Paris Metro, Proceedings of the 3rd International
Symposium of Fire Safety Science, Elsevier, London.
Watts, J M, Jr (1986), Dealing with uncertainty: some applications in fire protection
engineering, Fire Safety Journal, 11, 127–134.
4 Acceptance criteria
We all take risks all the time. Whether it is crossing the road, driving to work or
watching television. The risk may vary from being knocked down, to being in a car
accident or suffering ill health due to lack of exercise. The same can be said of fire
safety in buildings. As long as we occupy buildings where there is a chance that
ignition sources and combustible materials may be present together, there will be
a risk of death and injury due to fire. We need not be too fatalistic, however, this
simply identifies the need to manage the risk. It also indicates that although we
should work towards reducing risk, the ultimate goal of zero risk is not a realistic
expectation.
They say that there are two certainties in life: death and taxation. It follows
that whilst we live there is a risk of death and the only way of not dying is not
to live in the first place. This may be self-evident, but is very important when
we start to consider specific risks that we consider them in the context of other
risks.
There is a practical benefit to looking at risks in context. Society may decide
that it would like to dedicate more resources to addressing one risk than another.
For example, for healthcare, it may typically cost about £20k to save a life, whereas
for fire safety in buildings, it may cost more than say £1million to save a life.
Therefore, society (or its representatives) could decide to put more resources into
healthcare than into building fire safety
Equally, society may decide that the risk of death by fire is much worse than
the risk of not being treated for a potentially fatal condition, in which case the
opposite conclusion is valid. This is called ‘risk acceptability’ and there are several
ways in which a risk may be accepted:
1 Ignorance. If the existence of a risk is not known, then it may be accepted for
many years. Smoking, asbestos and the fire safety of certain types of sandwich
panel fall into this category. However, once these risks were known, there was
a duty to address them.
2 Negligible. If the risks are so low that they can be considered negligible, they
can be accepted. What is negligible is difficult to define, but for health and
safety, risks that are well below 1 death in a million years for a member of the
public are generally considered negligible.
Risks that are above those that are negligible can be considered tolerable or
intolerable:
1 Tolerable. Risks may be considered tolerable where the benefits of the activity
causing the risk are perceived to outweigh the risks. The convenience of
travelling by air, rail or car could be seen in this way. Even though the
benefits are seen to outweigh the risks, there is still a need to reduce the
risks until the costs of further risk reduction far outweigh reduction in
risk. This principle is known as reducing the risks until they are ‘as low as
reasonably practicable’ (ALARP) and has been a legal precedent in health
and safety for many years.
2 Intolerable. There are, however, levels of risk that could not be tolerated
under any circumstances and these can be considered intolerable. This could
be the risk of death of a member of the public above say one in a thousand
per person year.
For fire safety in buildings, the annual fire statistics indicate that there is
a finite level of fire risk in buildings. This may also indicate that if a building
complies with the appropriate fire safety standards, then its level of fire risk is
broadly tolerable (or possibly acceptable). It could also be said that applying the
fire standards to a non-standard building could result in intolerable levels of fire
risk. However, no explicit numerical criteria for fire risk in buildings have been
set in the UK.
For the fire safety engineering of a non-standard building, this means that
the level of risk should be designed to be the same or lower than that for an
equivalent standard building. Because there are no quantitative risk criteria,
this means comparison of a non-standard building with a compliant building.
Because the regulations are not generally framed in terms of risk, this results in
an assessment of physical hazards and the balancing of qualitative arguments
about risk.
Types of acceptance criteria

The acceptability of fire risks is often seen to be contentious, with engineers,
approval bodies, fire authorities, psychologists, sociologists, politicians and many
others contributing to the debate. Neither is the acceptability of risks a fixed
phenomenon. Through history it can be seen that risks that were once accepted
or tolerated become unacceptable from time to time. However, without some
form of acceptance criteria it is not possible to judge whether a building is
acceptably safe.
There are three general approaches to risk acceptability:
1 Risk comparison
2 Absolute risk criteria
3 Economic basis.
Acceptance criteria 177
Approach 3 is also typically a form of risk comparison. This can be seen in Table
4.1 which contains a summary of the approaches as applied to fire safety.
Qualitative and semi-quantitative risk assessment methods

Since these methods do not explicitly predict a level of fire risk, risk acceptance is
generally by comparison of the results of the analysis with an accepted standard. For
example, the acceptability of a risk identified in a qualitative fire risk assessment in
an office is generally made with reference to what would normally be expected in
such an office in the relevant guidance document (see Section 2.1).
For the points schemes of Section 2.2, the risk acceptability is represented by
a threshold score or range of scores which lead to risk reduction or no action. For
matrix methods, risk may be acceptable by comparison to the rating of a compliant
design or by the low level of risk implicit in the score and these can be expressed
in matrices of risk acceptance.
Risk acceptance criteria for quantitative methods is altogether more
complicated, but can provide a more informed and evidence-driven test of how
risks should be treated.
4.1 Comparative criteria

The simplest and most common basis for assessing whether a building is acceptably
safe is by comparison with other risks via historical data. Extensive lists of risks
have been drawn up for many types of activity. Table 4.2 contains levels of fire risk
in different building types in England and Wales. The concept is that risks that are
significantly lower than those attached to common everyday activities accepted
by society are safe.
For fire risks in buildings, it can often be difficult to establish the level of risk
in absolute terms. However, it may be relatively straightforward to demonstrate
that the design provides a level of risk equivalent to that in a building which
conforms to more prescriptive codes (life safety or financial). Since the study
is purely comparative, many errors contained in assumptions or data regarding
ignition frequencies or reliability of systems will be cancelling in nature and have
Table 4.1 Typical types of acceptance criteria

Fire safety objectives
Life safety Financial
Comparative Level of risk equivalent to Comparison of design
code-compliant solution, alternatives, (cost/benefit
Analysis e.g. Approved Document B analysis).
method (AD B).
Absolute Number of casualties per Acceptable average loss per
occupant year. year.
Table 4.2 Number of deaths per building year and number of deaths per occupant year
Average per year 1995–1999
Occupancy No buildings No occupants No deaths No injuries No fires Death/ Death/
building/year occupant/yearc
a
Further education 1051 845617 0.0 17 535 <2.4E–04 < 3.0E–07
a
Schools 34731 10503100 0.0 51 1669 <7.2E–06 < 2.4E–08
Licensed premises 101081 –a 2.8 262 3317 2.7E–05 –
a
Public recreation buildings 45049 – 1.3 48 2581 2.8E–05 –
a
Shops 354475 – 3.3 284 5671 9.2E–06 –
a
Hotels 28371 389174 2.5 116 1021 8.8E–05 6.4E–06
Hostels 9829 –a 0.5 60 1338 5.1E–05 –
a
Hospitals 3486 – 3.3 113 3063 9.3E–04 –
a
Care homes 29080 – 4.5 130 1616 1.5E–04 –
b
Offices 209627 4107000 0.3 219 1988 1.2E–06 7.3E–08
Factories 170972 –a 4.3 286 5299 2.5E–05 –
a
All above occupancies 987752 15844891 22.5 1584 28096 2.3E–05 6.5E–06
Notes
a Number of occupants equals the sum of the number of employees and other occupants.
b Number of occupants equals the number of employees only.
c It may be more appropriate to use the number of deaths per occupant for large or complex buildings.
no significant influence on the decision. This can be confirmed by sensitivity
analysis.
Before it can be demonstrated that a solution offers an equivalent level of risk
as a prescriptive code, the intent of that code needs to be clearly understood. It
is important to understand the intentions of each recommendation, as particular
provisions may have more than one objective. Alternative risk management
solutions can be developed to address the specific underlying objectives. The fire
risk assessor should demonstrate that the solution proposed will be at least as
effective as the conventional approach.
The limitations of this approach can be summarised as follows:
• It does not distinguish between risks that are accepted and those that are
tolerated (for example the risk from potato poisoning is accepted since there
has only been one recorded death, but the risk of choking is tolerated because
of the benefits of eating and drinking).
• It takes no account of changes that may make historical risk levels out of date
(the technology and regulation of air travel have changed significantly over
the last 50 years, therefore basing acceptance criteria on the data from the
last 50 years may be misleading).
• It does not address risk reduction following a major incident.
• It neglects other factors that affect the acceptability of the risk, e.g. involuntary
and societal risks.
Risks are more likely to be accepted if the activity is voluntary, i.e. the person
exposed to the risk perceives that they can control their level of exposure. For
example, driving a car can be perceived as a voluntary activity, whereas being a
passenger on a train can be seen as an involuntary activity.
There are several ways of expressing risk criteria and one of the most common
contains three concepts:
1 There is an upper limit which cannot be tolerated on any grounds and

activities whose risk levels cannot be lowered should be stopped.
2 There is a lower limit where the level of risk appears to be trivial and the use
of resources to reduce it would be wasteful.
3 All activities whose risk lies between the upper and lower limits should only
be carried out if the benefits justify the risks.
The third concept leads to the principle that the risks between the upper
and lower limits should be as low as reasonably practicable (ALARP) which is
enshrined in the UK Health and Safety at Work Act 1974 Part 1 under the general
duties of persons concerned with premises:
It shall be the general duty of each person who has, to any extent control of
premises … to take such measures … as so far as is reasonably practicable …
that they are safe and without risks to health.
As low as practicable can be defined as the point where the cost of further risk
reduction far outweighs the resultant reduction in risk.
This can mean that the levels of risk and cost of several potential safeguards
or sets of safeguards are evaluated against a baseline solution (i.e. a benchmark).
The level of risk reduction relative to the baseline (usually an existing or code-
compliant case) can then be evaluated. Care should be taken that the event
scenarios quantified cover an appropriately broad range of events. For example,
if a very severe fire growth and peak heat release rate is the only event quantified
in the evaluation of say, a new rail tunnel design, then the alternatives are very
unlikely to show any risk/cost benefit unless they change the event being modelled
in some way.
The cost of the alternative solution can then be evaluated, in terms of ‘cost
per life saved’, i.e. the cost of saving a life using the alternative solution. This
can provide a very effective measure of the risk/cost effectiveness of a range of
alternative solutions. The cost per life saved can be then be compared with other
alternatives or against benchmarks such as safeguards costing more than £10
million pounds for a life saved are not risk/cost-effective. The cost of saving a
life through a clinical intervention is generally of the order of £20,000. This is
conceptually and ethically quite different to the ‘value of a life’ described in the
absolute criteria section below (see Figure 4.1).
Comparative criteria in fire engineering use the level(s) of risk predicted for
a code-compliant building as a benchmark for a non-standard building to assess
equivalency (i.e. the level of risk is the same or lower). The main advantage of
comparative criteria is that assumptions made during the analysis cancel each
other out and so decisions made on the output should be much more reliable.
The main disadvantage is that the code-compliant building may not provide an
acceptable level of risk.
Frequency data
Frequency (Events >N per year)
1.00E-01
1.00E-02 Intolerable
1.00E-03
1.00E-04
ALARP
1.00E-05
1.00E-06
1.00E-07 Negligible
1.00E-08
1 10 100 1000 10000
Fatalities (N)
Negligible Intolerable Detailed analysis
Figure 4.1 Life safety acceptance criteria shown on an F–n Curve.

4.2 Absolute criteria
Absolute acceptance criteria for life safety are derived from society’s perception of
the risk. The perceived level of risk is often very different from the estimated level
of the risk. Extensive research by psychologists and social scientists indicates that
there is a range of factors, values and beliefs that influence the public’s perception
of risk. Theses factors include whether the risk:
• is known or unknown in nature;

• is dreaded or not;
• is voluntary or involuntary;
• affects an individual or many people at the same time.
Risks that are unknown or known and dreaded or not can be plotted on a matrix.
An example of a risk that is known and neutral in terms of dread is that of motor
vehicles. An example of a risk that is known and dreaded is that of crime and an
example of a risk that is unknown and dreaded is that of nuclear power. Generally
speaking, risks that are unknown and/or dreaded are less acceptable to society than
those that are known and not dreaded. Our perception of these risks can change
with information, events and the way that they are portrayed in the media.
The degree to which a risk is voluntary also has a significant impact on the
level of risk that may be tolerated. A voluntary risk may be one that an individual
chooses to be exposed to and for which they feel they have a large degree of
control. An example of this may be horse riding for leisure.
An involuntary risk is one where there is little alternative choice and the
degree of risk is controlled by others. An example of this is air travel where the
destination is on the other side of the globe. Generally, society is more willing to
accept higher levels of risk in activities that are seen as voluntary rather than those
that are seen as involuntary. Often the degree to which an activity is perceived to
be voluntary is different to the actual situation.
For example, initially car driving may be seen as voluntary ‘risk taking’ but in
reality the level of risk depends on many other factors including other road users,
the skill of road designers, equipment maintainers, those who are near roads, the
closeness and efficiency of emergency and medical services etc.
The difference between an individual following a certain pattern of activity or
many people at the same time has a significant impact on tolerance. The former
are known as individual risks and the latter as societal risks. Society is generally
much more concerned about societal risks than it is about individual risks. Again,
the different way that society (and the media) treats 3,000 largely individual road
deaths each year compared with individual rail events with between 10 and 20
deaths is very clear.
Absolute life risk criteria fall into two categories: individual and societal, where:
• Individual risk is the frequency at which an individual may be expected to

sustain a given level of harm from the realisation of specified hazards. This
is usually related to a specific pattern of life. For fire safety this may be the
individual risk of someone who works in an industrial or office building or of
a shopper who visits a retail development once a week.
• Societal risk is the relationship between frequency of occurrence and
the number of people in a given population suffering from a specified
level of harm from the realisation of specified hazards. This is important
because multi-fatality disasters are particularly repugnant to society. This
may be expressed as the frequency with which ten or more people may die
from fires. This is normally significantly lower than an individual level of
risk.
4.2.1 Individual risk levels

The levels of risk to individual members of the public from the activities on major
industrial sites (Health and Safety Executive 1988, 1989) are:
1 Maximum tolerable risk to individual member of the public (death per year) is
10–4, i.e. one death can be expected from the operation of this site once every
10,000 years (or from 10,000 such sites, once per year). That is one of the
implications of this type of risk criterion which is fine if the level of activity is
constant and well known. This issue is more relevant for societal risks where
acceptance criteria may be used to gain acceptance for one plant, but are the
criteria still valid 100 plant designs later?
2 General acceptable risk to individual member of the public (death per year)
is 10–6.
The nature of the above risks is that they are largely involuntary, although
there is an overall benefit to society from the activity.
Therefore, the generally acceptable levels of individual risk for a member of the
public at home from a fire is 10–5 death per year. The generally acceptable levels of
individual risk of the public elsewhere from a fire is 10–6 death per year.
4.2.2 Societal risk levels

Many studies have been carried out on the quantification of risk in the nuclear
power industry. Accidents, particularly in the reactor, could cause not only early
deaths but delayed deaths, sickness and genetic damage to people outside the
plant. One of the earliest studies was the one carried out by Rasmussen (1975)
for the US Nuclear Regulatory Commission, known as WASH 1400. Rasmussen
quantified the main hazards of the pressurised water reactors in use. He also
collected information on fatalities caused by a range of man-made and natural
disasters in order to compare potential nuclear hazard and hazards for which there
was everyday experience. He expressed the annual expectation of fatalities as the
probability of a number equalling or exceeding a certain value, N. Later, Fryer and
Griffiths (1979) referred to this relationship as f(N) line.
Rasbash (1984) reviewed various studies on the f(N) relationship to suggest
criteria for acceptability for use with quantitative approaches to fire safety.
According to him, two points were not taken into account in the conclusion of
Rasmussen (1975), that societal risk due to the installation of 100 nuclear reactors
would be small compared with the societal risk to which the nation (USA) had
become accustomed from well-known man-made hazards including those of fires
and explosions. Firstly, the people who would be killed after a nuclear accident
would, for most of the part, be living quite near the nuclear reactor. The societal
risk is, therefore, suffered disproportionately within the population. Secondly, the
large majority of the people, for other hazards considered by Rasmussen, were
obtaining a substantial benefit from the risk activity, they for the most part being
major beneficiaries of the risk activity. For nuclear power stations, those at risk
were most likely to be obtaining only marginal benefit. For a proper comparison
to be made, Rasbash suggested that the number of people killed by fires and
explosions should be those who were not using the building or being employed at
the risk activity concerned.
Reviewing studies on the quantification of the hazard of major industrial
processes, Rasbash estimated that the societal risk was about 1 × 10–4 downwards
for an incident that would cause about 100 deaths outside the plant. Much of
the hazard to the general public caused by process industries is due to fire and
explosion associated particularly with the release of flammable liquefied gas or a
large amount of flammable liquid in a vaporised or dispersed state. Rasbash put
forward the view that risks associated with such hazards, both to local individuals
and to local communities, should not be out of line with the fire and explosion
hazard normally experienced by these people. In addition, the amount of risk that
these individuals or communities should be expected to tolerate should be related
to the benefit they obtain from the risk activity.
Rashbash’s suggestions for societal risk were based upon experience of large
kill-size fires and explosions in the UK, as shown in Figure 4.2. The full lines
in this figure were based on actual fires in the UK and the extrapolation on
experience in rich countries worldwide. Analysing this graph, Rasbash suggested
that the total societal risk for all non-dwelling fires with N or more fatalities should
be proportional to the size of the community that is threatened. On this basis
and including other factors, he put forward target f(N) lines as in Figure 4.3 for
threatened communities of different sizes expressing the boundary of acceptability.
This figure shows that a target that would not be acceptable to a community of,
say, 1000, would be acceptable to a community of 10,000 and so on.
The approach discussed above has given rise to a criticism that a hazard
could be made acceptable to a community if more people were crammed into
the area at risk. For example, if a fire hazard associated with a plant were such
that it could produce a multiple-fatality fire disaster which could kill 100 or more
people 3 times in 10 million years (X in Figure 4.3), it would be unacceptable to a
community of 10,000 but acceptable for 100,000. This criticism would be justified
if cramming ten times the number of people into the threatened area did not affect
the expected number of fatalities that might occur in an incident.
10 A
Experience per 106 person years of fires with fatalities ≥ N 1 B
10–1
10–2
A B
C
10–3
10–4
C
10–5
10–6
1 2 5 10 20 50 100 200 500 1000
N
A United Kingdom experience D type fire 1963–73 (dwellings)
B United Kingdom experience D type fire 1963–73 (not dwellings)
C Points based on biggest fire in British Isles 1949–78
C Points based on four biggest fires in rich countries 1949–78 (excluding Japan)
C Dubious point base on biggest peacetime type fire in rich countries 1879–1978
Figure 4.2 Experience per 106 person years of fires with N or more fatalities
In fact, the number of fatalities in a societal incident engendered by a nearby

process plant is directly proportional to the density of population. Cramming
ten times as many people into the area that would normally carry 10,000 would
increase the number of expected fatalities from greater than 100 to greater than
1000 and would push the relevant risk point from X to Y. It will be seen that Y
indicates a risk even less acceptable to a community of 100,000 than X is for a
community of 10,000. In fact, to make the hazard more acceptable, it is necessary
to reduce the density of population by a factor of 10 and shift the calculated risk
point from X to Z. Thus risk which is not acceptable in an urban area could be
acceptable in a rural area.
10–3
10–4
Accident frequeny (events/year with consequences ≥ N)
10–5 Unacceptable
W
Population base
10–6
106
Z X Y´
10–7
105
10–8
104
Acceptable
10–9
103
W
10–10
1 10 100 1000 10000
N fatalities
X, Y´, Z Sext text, p. XXX
Estimated risk from LNG Terminal at Canvey Island7
W W Risk of pressurised water nuclear reaction1, 4
Figure 4.3 Targets for acceptability for societal risk from process industries. (Marginal
benifit to those at risk)
4.2.3 Multiple-death fires

Fires can cause not only direct damage to human life in terms of fatal and non-
fatal casualties but also indirect losses, e.g. distress and financial loss to the families
of the victims and to society at large. The aggregate disutility or consequences due
to fire deaths would be generally low for single-death fires and high for multiple-
death incidents (Ramachandran, 1998). The disutility associated with a single fire
with, say, 10 deaths is greater than the total disutility caused by 10 fires each with
a single death. Catastrophes with several deaths have serious social, economic and
political consequences.
Hence, it is necessary to analyse the characteristics of multiple-death fires and
estimate the probability of occurrence of such fires, particularly in large buildings
with several people at risk. Such an analysis would provide target values for
accepting the occurrence of multiple-death fires. Consider, for example, Figure
4.4, produced by Rasbash (1984/1985). Information given in this figure would
provide target values, firstly, for the whole country and, thereafter, for different
kinds of premises. For this purpose, an f(N) distribution curve such as Figure 4.5,
produced by Rasbash, should be drawn to represent a risk no greater than the risk
prevalent at the present time. The lines AA, BB and CC in this figure represent
the total target risk respectively for populations of 106, 107 and 5.6 × 107, the
last line being equal to the population of the UK at that time. There is a need to
update f(N) curves such as those in Figures 4.4 and 4.5 using more recent data on
the frequency distribution of fire deaths which are available for many countries.
Multiple-fatality fires generally occur in buildings other than dwellings to
which a figure such as Figure 4.5 is applicable. The figure provides target values
for the acceptability of the total risk for different population sizes. It is necessary
to determine how this total risk should be distributed amongst different types of
non-dwellings at risk. For this evaluation, it is necessary to have information on
B
A Dwellings
UK 1963–73
UK 1960–72
US 1970–82
Experience per 106 person years of fires with fatalities ≥ N
10–1 Buildings not dwellings

UK 1960–82
US 1970–82
C Rich world 1946–82
D Rich world 1883–1982
10–2 All fires
C
10–3
A B F
D
E
10–4
F
E
10 –5
1 10 100 1000
Figure 4.4 Multiple fatality fires in buildings

10
1
C
Target expectation of fires with fatalities ≥ N
10–1 B
10–2 A
Population
10–3
5.6x107
10–4
107
B
10–5
A 106
10–6
1 2 5 10 20 50 100 200 500 1000
N
Figure 4.5 Proposed target for acceptibility from multiple fatality fires in all buildings other
than dwellings for different populations
the distribution, numbers and size of various types of such buildings. Information
of this kind is scarce.
Rasbash (1984) attempted to tackle the problem mentioned above using
information on the number of people that will be in the buildings and obtained
a preliminary assessment of a number of premises of different kinds and different
sizes. He corrected these figures by assuming that the number of fires occurring
in a given occupancy would be approximately proportional to the square root
of the number of people that would be at risk. He then obtained a relationship
between mean values of target probability and size of building, expressed in terms
of number of people at risk.
The relationship mentioned above suggested that, for small buildings with less
than 15 people, the target probability of having a fire which could kill more than
5 people was about 5 × 10–7 per annum. This probability (per annum) was about
1 × 10–6 for a building with 15–100 people, 2 × 10–6 for a building with 100–500
people and 4 × 10–6 for a building with more than 500 people. Similar calculations
produced target probabilities for a fire which could kill more than 15, 100 or 500
people for buildings of the four population sizes. Zero probabilities were assigned
for cases such as more than 15 fatalities in a building with less than 15 people. In
general, the probabilities (per annum) ranged from about 10–6 for premises where
5 or more people may be killed in a fire to 10–7 to 10–8 for premises where 100 or
more may be killed.
The levels of societal risk are generally expressed on an F-n curve similar to
that in Figure 4.1.
When judging probabilistic risk assessments against the above criteria, the
predicted level of risk should generally be significantly below the criterion. If the
predicted level of risk only just satisfies the criterion then care should be taken
to ensure that the assumptions made in the study clearly err on the side of safety.
Some regulators have attempted to use the ‘value of a life’ to assess whether to
invest in safety. Different methods have been developed for assessing the monetary
value of a life (Ramachandran 1998). The value used can be taken from damages
awarded in courts (based on future earnings) and a notional sum for pain, grief
and suffering. This results in values in the order of £500,000 to £1million per life.
This approach does not take account of the preferences of the people at risk nor
the value of non-income earners.
Alternatively, the behaviour of individuals in the market place and/or social
survey questionnaires can be used as a value of life through an empirical economic
evaluation. This approach usually leads to a value that is approximately an order
of magnitude higher than that driven by court compensation payments.
Absolute criteria do not necessarily take account of the actual estimate or
recorded level of the risk and so they are an expression of perception. However,
for that reason they are important as they allow decision makers to take account of
people’s perceptions of risk in a highly effective but largely qualitative way.
Absolute criteria in fire engineering set specific levels of risk so that risks can
be assessed as negligible, tolerable or intolerable. The main advantage of absolute
criteria is that they provide a benchmark for judging acceptability. The main
disadvantage is that there is little or no link between the current standards in the
codes and the absolute criteria.
4.3 Financial objectives
4.3.1 Introduction
Chapter 1 discussed corporate governance and in conjunction with this, an
organisation or facility can decide, given its investments, competitive position,
insurance cover, contingency plans etc., that it can tolerate certain levels of loss or
interruption with certain return periods. These are usually expressed in terms of a
financial loss per year or level of financial loss and a frequency (or return period).
Fires cause fatal and non-fatal injuries to occupants of buildings and inflict direct
material damage to the buildings and their contents. Some fires may also cause
indirect/consequential losses, such as loss of production, profits, employment and
exports although, at the national level, these losses do not contribute significantly
to total fire loss. This is due to the fact that the loss of a specific unit of productive
capacity may be spread among the remaining capacity in the nation. On average
per year, fires occurring in the United Kingdom cause direct material damage
amounting to about £1200 million and indirect loss amounting to about £120
million. The direct and indirect losses in the UK represent about 0.21 per cent of
the Gross Domestic Product.
Target levels have to be set for acceptable limits to life and property damage
caused by fires. Target levels for life risk were discussed in the previous section. This
section is concerned with target levels for property damage and financial losses to
owners of particular industrial and commercial properties. Most of these losses can
be claimed from fire insurance companies if the properties are adequately insured
for direct and consequential losses. However, there is a need to set an acceptable
limit to property damage, since the damage caused by a large fire can seriously
disrupt or even bankrupt the industrial or commercial activity of a property owner.
4.3.2 Target level for property damage

A property owner can set an acceptable limit to property damage in terms of
physical extent of fire spread. For example, they may set an acceptable upper limit
for the probability of a fire spreading beyond the room of origin. This limit may be,
say, 0.05 such that, at most, only 5 per cent of the fires occurring in the property
may spread beyond the room of origin. This limit would depend on the type of
room, e.g. office, storage room, production area in an industrial building, assembly
or customer area in a department store. Life safety can also be considered in the
determination of this limit.
To achieve the target probability of fire spreading beyond the room of origin, a
property owner may adopt adequate passive and active fire safety measures. Some
of these measures may be in addition to those required under fire regulations. For
example, the fire resistance periods of the structural boundaries of compartments
may be 90 minutes, although only 60 minutes resistance is required under the fire
regulations. Also, the sizes of compartments may be limited to the maximum size
permissible for an unsprinklered compartment but sprinklers installed in these
compartments act as additional protection. Fire regulations permit larger sizes for
sprinklered compartments.
The property owner can also set an acceptable upper limit to property damage
in terms of area damage. The chance (probability) of area damage in an actual
fire exceeding this upper limit should be a small value. Based on the probability
distribution of area damage, the acceptable upper limit for area damage can
be determined with due regard to consequences in terms of damage to life and
property. This criterion has been adopted in determining a design fire size for
smoke ventilation systems. For covered sprinklered shopping complexes, for
example, 10m2 has been suggested as an acceptable design fire – see Morgan and
Chandler (1981). This corresponds to a heat output of 5MW or 0.5MW per m2.
The chance of damage in sprinklered public areas of retail premises exceeding
10m2 was estimated to be 0–4 per cent. In these areas, in the absence of sprinklers,
14–17 per cent of fires were estimated to be larger than 10m2. In non-public
areas of retail premises, the design fire size of 10m2 was exceeded in about 12 per
cent of all fires in sprinklered premises and in about 19 per cent in unsprinklered
premises. Carrying out a similar analysis based on the probability distribution of
area damage, Bengtson and Ramachandran (1995) determined design fire sizes for
underground facilities. Determination of design fire size is discussed in more detail
in Chapter 6.
Determination of an acceptable limit for area damage would also depend on the
density per m2 of financial value at risk in the building structure and its contents.
The financial loss in a fire can be assumed to be the product of the value density
per m2 and area damage in m2.
Depending on assets and other economic factors, a property owner may be
able to allocate in their annual budget, a certain amount of money to meet
financial losses in small fires which will not disrupt their business activity. The
property owner may be able to accept self-insurance for such small losses less
than a ‘deductible’ level. But they should purchase insurance cover for losses
exceeding this level. Losses exceeding a higher threshold level might disrupt
their business activity permanently or temporarily for a long time. To determine
the lower and higher threshold levels mentioned above for direct material
damage to the building and contents, the property owner should take account
of consequential losses and obtain insurance cover accordingly. Losses between
the two levels might cause only a minor disruption of the business activity, with
small consequential losses.
To determine the amounts which should be allocated annually for self-
insurance and insurance, the property owner should carry out a risk assessment
and estimate the annual frequency (number) of fires likely to occur in their
property for each of the three classes demarcated by the lower and higher
threshold levels mentioned above. The risk assessment would also enable the
property owner to determine fire prevention and protection measures which will
reduce the risk. This assessment would provide inputs to a cost-benefit analysis
of fire safety measures and self-insurance and insurance requirements. Based on
such an analysis, the property owner can determine an economically optimum
package of fire protection and insurance strategies, taking into account tax
allowances, savings in insurance premiums and other benefits for fire protection
measures included in the package. This economic problem was discussed in
detail by Ramachandran (1998).
4.3.3 Target levels for consequential loss

A property owner should take into account the consequential losses such as
loss of production or profits etc. when determining the level of additional safety
measures above those required under fire regulations. Apart from some general
factors such as repairing fire-hit premises, using more power and materials and
hiring additional people, consequential losses may be affected by special factors
concerned with a particular industrial or commercial activity. Specialised
equipment such as that controlled electronically or by computer and tailor-made
driers or centrifuges used in the manufacture of pharmaceuticals, if damaged
by fire, cannot be replaced easily and quickly. Loss of laboratory facilities may
seriously interrupt testing and quality control programmes. In some industries,
even minor fire damage at a critical point can cause a long period of interruption.
Factors affecting consequential losses in different industries were discussed in
detail by Ramachandran (1998).
For the reasons mentioned above, it is necessary to identify parts or areas
of an industrial or commercial property which have potentially high risks of
consequential losses. For each of such parts or areas of a property, target values
should be specified for damage and associated probability, depending on the
magnitude of consequential loss. To achieve these targets, adequate fire prevention
and protection measures should be determined, taking into account the costs and
benefits.
Target levels for consequential losses can be determined by applying utility/
disutility theory techniques (Ramachandran 1998) which take into account
factors such as the assets of a property owner and their attitude to fire risk, i.e.
extent of risk-aversion in addition to fire damage. The disutility function of
the property owner can be constructed by considering different levels of direct
property damage. As discussed in the previous section, a range for these levels
can be identified by determining the minimum damage, a loss less than which will
cause no disruption to the business activity and the maximum, a loss exceeding
which might cause serious disruption.
4.3.4 Target levels for structural failure

During the 1970s, civil engineers started adopting a limit state approach based
on reliability methods, instead of a safety factor approach, and target safety
probabilities for designing structures. For example, a range of 10–6 to 10–14 per
annum was suggested for annual failure probabilities for reinforced concrete and
steel buildings. Such results for buildings with 10 to 1000 people at risk were based
on consequences in terms of multiple fatalities. These structural safety concepts
were extended to fire safety design during the 1980s to propose hypothetical target
probabilities for failure of primary structural members during a fire. Based on a
lifetime of 50 years for a building, a target failure probability of the order of 10–8 per
annum or less was estimated for a situation in a building where a collapse might
bring about the deaths of hundreds of people. See Table 4.3.
Collapse of a structure during a fire occurs very rarely. A structural element
(wall, floor or ceiling) of a compartment (or room) is more likely to experience
thermal failure and allow a fire to spread to an adjacent compartment or area. If
such a failure occurs with probability Pc, one or more deaths might occur with a
probability Pd. Consistent with the target probability set for Pd, a target value may
be determined for Pc the probability of compartment failure.
As discussed earlier, with a change in the notation, Pc is the product of PA and
PB where PA is the probability of flashover or serious fire and PB the ‘conditional
Table 4.3 Societal risk criteria for structural collapse due to fire
General acceptable risk for General acceptable risk for

10 or more deaths 100 or more deaths
(deaths per building year) (deaths per building year)
Deaths per building year 5 × 10–7 5 × 10–8
Deaths per occupant year 1 × 10–8 1 × 10–9
probability’ of compartment failure given flashover, i.e. PC = PA × PB. Depending

on PA and the target set for PC, a target value may be assigned for PB and the fire
resistance required to meet this target determined accordingly.
In the CIB report (1986) ‘Design Guide: Structural Fire Safety’, illustrative
examples are given for PB with PC = 10–6 per annum for different values of PA
and for different compartment sizes. PA is the product of annual probability of fire
occurrence and the probability of fire developing into a large size.
Using the financial damage techniques and data in this book, it is possible to
estimate the risk of damage that may result from a fire. This information may then
be used to estimate potential monetary losses and enable cost-benefit analysis to
be undertaken to establish the relative value of installing additional or alternative
fire protection measures.
These financial criteria in terms of levels of loss or interruption should be set
in conjunction with the organisation concerned and/or their financiers or insurer.
4.3.5 Cost-benefit analysis

For financial risks and acceptance criteria such as ALARP, cost-benefit analysis
can be crucial in assessing the acceptability of risks. Various methods can be used
in cost-benefit analysis of fire protection measures (Ramachandran 1998). What
follows is a case study of a bus garage to illustrate the use of cost-benefit analysis
in fire risk analysis (Charters 1998).
The first step was to determine the total costs of the sprinkler installation.
This not only included the initial installation costs, but also covered the annual
running costs. The following list, whilst not meant to be exhaustive, covers the
main costs included in this case:
• design fees;
• installation/construction;
• commissioning/training of staff;
• maintenance/running etc.
The capital cost for the sprinkler system was £25,000 with an annual
maintenance cost of £100. The benefits of the new installation were then listed,
including:
• reduced property loss;
• reduced consequential losses;
• reduced insurance premiums;
• improved life safety etc.
The benefit rates from the quantified fire risk assessment were added to the
difference in insurance premium to give the total benefit rate of £2,500 per garage
year.
This is the figure used in the investment appraisal. Table 4.4 shows the
discounted cash flow over a 30-year period. The discount factor used is 10 per
cent. This is the norm for commercial premises and is spread over a 30-year life
span (the normal life of the sprinkler system). The financial figures in Table 4.4
do not represent those of any particular garage or operator, but may be typical of
some circumstances.
The cost benefit analysis showed a small positive net present value at the end
of 30 years. The positive figure indicated that, strictly speaking, the installation of
bus garage sprinkler systems did not represent a good investment. However, the
smallness of the value indicated that this was a marginal case. In the light of the
risk assessment, the bus operator decided that they had sufficient redundancy and
diversity of bus supply through ownership (in several garages), leasing and buying
and insurance not to require bus garage sprinklers. However, the risk assessment
had highlighted several other areas, such as fire safety management and the
separation of the IT centre that were much more cost-effective and these were
implemented.
This study to assess the benefits of installing sprinkler systems in bus garages
indicated that there were business continuity and property protection benefits to
the operator. However, the cost-benefit analysis and the operator’s contingency
plans meant that there was no cost-benefit or consequence case for installing
sprinklers the bus garage. As a result of the risk assessment, the operator did
implement other forms of safeguard and fire precaution.
4.4 Other objectives

Other fire safety objectives can include the protection of heritage and the
protection of the environment. Fire risk acceptance criteria may be provided
by relevant guidance either explicitly or implicitly. Often there are no absolute
criteria and so these can only be agreed by consensus between all the relevant
regulators and stakeholders.
Table 4.4 Discounted cash flow for bus garage sprinkler system
Year Capital cost Annual cost Total cost Savings Net costs/ Discount factor NPV of costs/ Cumulative
(£) (£/yr) (£/yr) (£/yr) savings (10%) savings NPV
(£/yr) (£)
0 25000 25000 0 25000 1 25000 25000
1 100 100 –2500 –2400 0.9091 –2182 22818
2 100 100 –2500 –2400 0.8265 –1983 20835
3 100 100 –2500 –2400 0.7513 –1803 19032
4 100 100 –2500 –2400 0.6830 –1639 17392
5 100 100 –2500 –2400 0.6209 –1490 15902
…
26 100 100 –2500 –2400 0.0839 –201 3014
27 100 100 –2500 –2400 0.0763 –183 2831
28 100 100 –2500 –2400 0.0693 –166 2664
29 100 100 –2500 –2400 0.0630 –151 2513
30 100 100 –2500 –2400 0.0573 –138 2375
Total 28000 –75000 –47000 2375
References
Bengtson, S and Ramachandran, G (1995), Design fires in underground facilities, Book of
Abstracts, First European Symposium on Fire Safety Science, ETH, Zurich.
Charters, D A (1998), Fire safety at any price?, Fire Prevention 313, October, 12–15.
Fryer, L S and Griffiths, R F (1979), United Kingdom Atomic Energy Authority, Safety and
Reliability Directorate, Report SRD R149, UKAEA, Harwell.
Health and Safety Executive (1988), The Tolerability of Risk from Nuclear Power Stations,
HMSO, London
Health and Safety Executive (1989), Quantified Risk Assessment: Its Input into Decision
Making, HMSO, London
Morgan, H P and Chandler, S E (1981), Fire sizes and sprinkler effectiveness in shopping
complexes and retail premises, Fire Surveyor, 10, 5, 23–28.
Ramachandran, G (1998), The Economics of Fire Protection, E & F N Spon, London.
Rasbash, D J (1984), Criteria for acceptability for use with quantitative approaches to fire
safety, Fire Safety Journal, 8, 141–158.
Rasmussen, N C (1975), An Assessment of Accident Risks in US Commercial Power Plants,
WASH 1400, US Nuclear Regulatory Commission, Washington, DC.
5 Initiation
5.1 Frequency of ignition/probability of fire starting
5.1.1 Global estimation

During a short period, say a year, fires do not occur frequently in a particular
building. Hence, it is necessary to consider a group or type of buildings with similar
fire risks to evaluate the annual frequency of ignition or probability of fire starting.
This frequency or probability, for a given type of building, will increase with the
number of ignition sources and, hence, with the size of the building.
As discussed in Section 3.2.1, the annual frequency of ignition or probability
of fire starting would increase approximately with the size of a building according
to a ‘power’ relationship. The building size may be expressed in terms of total
floor area, A, in square metres or total financial value, V, at risk in the building
and contents. This relationship is described in Equation (3.40) where F(A) is the
annual frequency of ignition or probability of fire starting in a building of total
floor area, A. The parameter α in this equation denoting the power is generally
less than unity (one) for the reasons discussed in Section 3.2.1 – see also Table 3.4
with a=b according to equation 3.35.
The values of the parameters K and α in Equation (3.40) are ‘global’ estimates
applicable to a given type of building. The values of these two parameters for
buildings in the United Kingdom given in Table 3.4 were estimated more than
20 years ago. Hence, these parameters need to be re-evaluated by carrying out
again surveys of buildings at risk and analysing relevant statistical data relating to
fires which occurred during recent years. This exercise should be carried out by
a national organisation for different types of buildings and, if possible, for major
groups or fire risk categories within each type.
The probability of an accidental (not arson) fire starting in a building depends
on the presence or absence of causes or sources which can be classified into
two broad groups – human and non-human. The first group consists mainly of
children playing with fire, e.g. matches, careless disposal of matches and smokers’
materials and misuse of electrical and other appliances. The second group
includes defects in, or faulty connections to, appliances using electricity, gas
and other fuels. The appliances may be further classified according to cooking,
Initiation 197
space heating, central heating and other uses. This group also includes causes
such as mechanical heat or sparks in industrial buildings, natural occurrences
and spontaneous combustion. Some materials in a building, e.g. latex foam and
finely powdered rubber, could be ignited even by a low energy smouldering
sources.
The nature and number of ignition sources and materials will vary from
one part of a building to another. In an industrial building, for example, three
major parts can be identified – production, storage and other areas. Given that a
building is involved in a fire, the conditional probabilities reflecting the relative
or comparative risks due to various causes in different parts of the building can be
estimated from group statistics such as those in Table 5.1. In this example, if a fire
occurs, the conditional probability of fire starting due to, say, smoking materials in
the stores/stock room is 0.0129 (= 15/1162).
5.1.2 Particular buildings

The conditional probabilities of fire starting, based on figures such as those in
Table 5.1, would pertain to an average or reference building in the type or risk
category considered. For a particular building in any type or risk category, an
estimate of the conditional probability (given fire) for the ith cause in the jth part
of the building is given by
Iij Pij (5.1)
where Pij is the probability for this cause and part revealed by figures such as those
in Table 5.1. The parameter Iij will be assigned the value zero if the ith cause is
totally absent in the jth part of the building considered for risk evaluation. If the
cause is present, Iij should be given a positive value depending on the extent to
which this cause can be responsible for starting a fire in the jth part; this value
can be greater than unity. A value equal to unity can be assigned if the building is
similar to the ‘average building’ in this respect.
Taking smokers’ materials as an example, it should be possible to determine,
for the building type considered, a global quantitative measure, Sij, denoting
the exposure of, say, a storage area at risk from fire due to such a cause. This
measure may be the total number of cigarettes etc consumed by all smokers per
day, per m2 of floor area. A similar quantitative measure, sij, should be evaluated
with respect to the consumption of smoking materials in the storage area of the
particular building subjected to risk evaluation. Then the ratio sij/Sij is an estimated
value of the parameter Iij for the particular building considered. This ratio should
be adjusted to take account of factors such as smoking lobbies and publicity
measures, e.g. notices, circular letters making people aware of the risk of fire due
to smoking materials. The assignment of a value to the parameter Iij has to be
somewhat subjective, with its accuracy depending on the extent and accuracy of
relevant information used in the calculations.
Table 5.1 Spinning and doubling industry – places of origin of fires and sources of ignition
Production and
maintenance Storage areas
Dust Loading Miscell-
extractor Store/ bay, aneous
(not Other stock packing Other
Sources of ignition cyclone) areas Assembly room dept areas areas Total
A Industrial appliances
Dust extractor (electrical) 14 3 – – – – – 17
Dust extractor (other fuels) 12 – – – – – – 12
Other appliances (electrical) 6 111 – – – – – 117
Other appliance (other fuels) – 22 – 1 – – 2 25
B Welding and cutting equipment – 10 – 6 – – 7 23
C Motor (not part of other appliances) – 7 – – – – – 7
D Wire and cable 1 12 – – – – 2 15
E Mechanical heat or sparks (electrical) 27 194 – – – – – 221
Mechanical heat or sparks (other) 52 387 – 2 – – – 441
F Malicious or intentional ignition – 9 – 3 – – 3 15
Doubtful – 13 – 7 – – – 20
G Smoking materials 2 29 1 15 1 – 7 55
H Children with fire e.g. matches 3 4 – 12 2 4 5 30
J Others 4 29 2 3 2 – 12 52
K Unknown 11 78 – 14 – 9 112
Total 132 908 3 63 5 4 47 1162
Initiation 199
Each possible cause or source of ignition in each part of the building considered
should be identified and its Iij value estimated. The aggregate probability of fire
starting for the building is then
F(A) ∑ ∑ Iij Pij (5.2)

i j
where F(A) is the global probability of fire starting in a building of total floor area
A (in m2) estimated by Equation (3.40). The value given by the double summation
part, excluding F(A) in Equation (5.2), can be greater or less than unity depending
on the extent to which the various causes are present or absent in the building.
It will be equal to unity only if the building considered is more or less identical to
the average characteristics of the underlying population of buildings in regard to
causes or ignition sources.
The method described above was proposed by Ramachandran (1979/80, 1988).
It is similar to the allocation approach used in fire risk assessments of nuclear
power plants – see Apostolakis (1982).
5.1.3 Special factors affecting frequency of fire occurrence
5.1.3.1 Types and total amounts of fuel used

By comparing the number of fires in buildings attributed to various types of fuel,
with the total amount of fuel used, it is possible to estimate correlations and
predict trends. The number of fires caused by a type of fuel over a period of years
may be plotted against the number of units of this fuel used during that period.
This procedure followed by Chandler (1968) gave an approximately straight line
correlation for electrical fires occurring in the UK between the years 1956–1966.
When this trend was extrapolated, it was estimated that 25,500 fires would occur
in 1970 associated with an output of (210 × 109) kWh of electricity. The number
of fires which actually did occur in 1970 was in good agreement with the estimated
number.
For gas, the number of fires during 1957–1966 did not vary linearly with the
amount of town gas sold. In fact, the trend showed there was a reduction in the
fire frequency per 109 therms of gas sold. However, extrapolating on the trend did
predict 7,000 fires in 1970 and the number which occurred was 7,100. Fires due to
solid fuel showed a reduction in number because of a reduction in the amounts of
solid fuel sold. With oil, the fire frequency per million tons of oil delivered dropped
through the period 1955–1966. This was apparently due to the advent of central
heating which is much safer than portable oil heating.
5.1.3.2 Socio-economic factors

Various studies carried out in the United States demonstrated that fire incidence
(with its consequences in terms of deaths and injuries) is related, though this does
not necessarily imply causality, to a combination of factors – see, for example,
Bertrand and McKenzie (1976), Munson and Oates (1977) and Gunther (1975).
These factors reflected poor and substandard housing, overcrowding, social
class, race, lack of family stability and proportions of the young or elderly in the
population.
In the United Kingdom, a detailed analysis by Chandler (1979) in relation to
fires in London, revealed strong correlations between fire incidence and housing
and social factors. The housing factors included tenancy (owner occupied, private
rented etc) and the lack of amenities. People who own their homes might be
expected to be more careful than those who live in rented properties. The social
indicator most strongly correlated with fire incidence was proportion of children
in care, which was thought to reflect family instability. Fire frequency appeared to
be independent of rateable value of a property per person and the age distribution
of the population. However, the incidence of casualties was generally highest
among the young and elderly. Updated results of London analysis were included
in a later study by Chandler, Chapman and Hollington (1984) with reference to
Birmingham and Newcastle-upon-Tyne.
A summary of the results obtained in the UK studies mentioned above,
along with other human aspects of fires in buildings, is contained in a paper by
Ramachandran (1985). As pointed out in this paper, stress could be a primary
factor responsible for carelessness leading to increasing fire incidence. Stress
could result from socio-economic factors such as poor housing conditions and
low income, both of which are interrelated. Personal circumstances force people
into sharing facilities or overcrowded households. High fire incidence may be a
function of an unquantifiable lack of social identity and stability.
5.1.3.3 Weather conditions

Severe weather conditions during winter can cause an increase in number of
fires and fire casualties. Chandler (1982), for example, analysed data for fires in
dwellings in the UK during the severe winter of 1978–1979. He found that in all
regions, temperature and vapour pressure were significantly correlated with total
fires and fires due to space heating, electric blankets and wire and cable, including
leads. The same was true of life risk fires due to space heating. Fires due to cooking
appliances were not generally influenced by severe weather conditions.
Chandler’s analysis suggested that in the temperature range 0°C down to –5°C
there were an extra 30 fires per week in England and Wales for every degree drop
in temperature. This result was in general agreement with the assessment based
on fire frequencies for the 1962–1963 winter – see Gaunt and Aitken (1964). Fires
involving casualties, rescues or escapes were dependent on temperature and even
more so those fires due to space heating.
Initiation 201
5.2 Probable rate of fire growth
5.2.1 Introduction
Successful prediction of the course of a fire provides an indication of the size
of the fire at a given time, the rate of fire growth, the time available for escape
or suppression, the type of suppressive action that will be effective and other
attributes that define fire risk in a particular type of building with known materials
and ignition sources. The ability to predict the course of the fire would enable one
to predict the effect of changes in the initial conditions, design of the building
and passive and active fire protection systems, materials and ignition sources.
Thus, a fire safety engineer would be able to select the best combination of design
features, materials and fire protection devices, providing desirable or acceptable
levels of life safety and property protection compatible with economic, amenity
and aesthetic requirements.
Hence, a central problem in the design of a building for fire safety and provision
of fire protection measures is to predict the development of a fire in the building
as a function of time and estimate the rate at which the fire grows in the room
of origin and subsequently spreads to other parts of a building. This rate depends
initially on the heat output from the material or object first ignited, apart from
other factors such as fire load, room dimensions and ventilation. The subsequent
rate of fire growth depends on the heat transfer (spread) from the object first
ignited and heat output and heat transfer properties of other objects within and
outside the room of fire origin.
To estimate the rate of fire growth as a function of time and space, several
deterministic mathematical models, based on scientific theories relating to heat
output and other physical quantities, have been developed and validated in the
light of experimental data. These models are mainly of three types – zone, field
and simulation models. Computer software packages have also been developed for
these models.
Several factors cause uncertainties in the patterns of development of actual
(not experimental) fires in a building and hence, in the rate of fire growth. These
uncertainties can be evaluated to some extent by using the computer software
package of a deterministic model and performing simulations for different fire
scenarios. Based on such simulations, it might be possible to calculate the average
and other statistical parameters providing inputs for estimating, in probabilistic
terms, the rate of growth of an actual fire occurring in a building with several
objects. Such an exercise would, however, be time-consuming and expensive.
A considerably cheaper and more realistic method for estimating fire growth
rate in probabilistic terms is provided by a non-deterministic statistical model
based on data relating to real fires compiled by the fire brigades. One such model
is the exponential model discussed in Section 3.3.2. This model, as argued in
Section 3.3.2, is more realistic than the T2 curve based mostly on data provided by
fire tests. The exponential model is applicable to the period of fire growth after the
onset of ‘established burning’, when more and more objects in a room are involved,
whereas the T2 model is really applicable to the initial stage of fire growth when
only one object is involved. The exponential model directly takes into account the
uncertainties governing the development of a real fire in a building. The rate of
fire growth estimated by this model is applicable to a wider (general) range of fire
conditions than a restricted range to which a T2 curve or a deterministic model is
applicable.
This section is concerned with the application of the exponential model for
estimating the average and other statistical parameters of the rate of fire growth
within and beyond the room of origin. The scenarios discussed for any occupancy
type are those generated by different areas of fire origin and intervention by
fire brigade and sprinklers. A statistical method is described for estimating the
probabilistic upper confidence limit for the fire growth rate for predicting the
worst case scenario. Some examples are given for illustrating the application and
use of the model in fire safety engineering problems.
5.2.2 Stages of fire growth

A fire in a room or compartment usually starts (first stage) from a very small
ignition source and, as the floor area affected gradually increases, it becomes
sizeable but still small. After some time, established burning occurs with sustained
growth and steady burning (second stage) during which the rate of increase of fire
size suddenly changes and the phase of rapid growth occurs. The fire will continue
to grow in the second stage until it reaches a maximum intensity (flashover) when
most of the objects (fuel) in the room are completely involved. During the fully
developed post-flashover period (third stage), the fire can penetrate the structural
barriers of the room of origin and ignite objects in the adjacent rooms. The fourth
stage is the ‘decaying period’ when the rate of fire growth decreases and the fire
burns out.
The pattern of fire growth described above generally applies to ‘free burning’
fires not involving any fire fighting. Some of these fires may undergo a decaying
process during any of the stages mentioned above due to the consumption of
all available fuel or other reasons. This process, leading to an eventual burnout
of a fire, is referred to as self-termination or self-extinguishment in fire science
literature. Fire incidence data contain information on some of these small fires
which were reported to the fire brigade but were out on or before the arrival of
the brigade. Some small fires not reported to the fire brigade are extinguished
by sprinklers, if installed or by first-aid means such as portable fire extinguishers,
buckets of water or sand and smothering. The growth of a developing fire can
be retarded, controlled or terminated at some stage by the successful operation
of sprinklers, if installed. Fire brigades extinguish all the fires which develop
beyond the ‘infant mortality’ stage and are not put out by sprinklers, portable fire
extinguishers or other first-aid means.
Initiation 203
5.2.3 The exponential model
The exponential model is applicable to the second and third stages of fire
development described above. This is the period when a fire grows steadily after
the commencement of established burning. During this period, heat output from a
fire increases exponentially with time – see Thomas (1974) and Friedman (1978).
This hypothesis is supported by experimental results – see Labes (1966).
The area damaged in a fire is approximately proportional to heat output. Hence,
Ramachandran (1980) proposed the exponential model in Equation (3.42) for
making the best use of statistical information on real fires attended and compiled
by the fire brigades in the United Kingdom. This equation is reproduced below in
order to facilitate the discussion in this section:
A(T) = A(0) exp(θ T) (5.3)
where A(T) is the total floor area (m2) damaged in T (minutes) counted from the
commencement of established burning.
The parameter A(0) is the floor area initially ignited at time T = 0 when
established burning commences. The parameter θ measures the increase in the
value of logarithm of area damage, to base e, for a unit increase in the value of T.
Thus, θ quantifies the rate of fire growth per minute.
Consider any type of building and area of fire origin such as the production or
storage area of an industrial building and the customer area of a department store.
For any such fire risk classification, an average value of A(0) can be estimated by
carrying out a fire load survey. An average value of A(0) can also be estimated
for a particular building by considering the objects in the area considered and the
floor area occupied by these objects. Fire tests may be able to provide an estimate
for the average value of A(0).
As discussed in Section 3.3.2, an average value of A(0) can also be estimated
by analysing area damage, A(T), and corresponding duration of burning, T,
for a sample of fires which occurred in the risk category considered. By fitting
a straight line to logarithm of A(T), base e, and T, the values of logarithm
of A(0), base e, and θ can be estimated graphically or by applying the least
squares method. This analysis would provide the average or expected value for
logarithm of A(0) and hence for A(0). It would also provide an estimate for the
average value of θ.
The value of θ estimated by a linear regression analysis as described above is
an average value for the entire duration of burning. An average value of θ can
be estimated for each of the component periods constituting the total duration
of burning. This was explained in Section 3.3.2 with reference to four periods
involving detection or discovery of a fire, calling of fire brigade, fire brigade arrival
at the scene of the fire and the time taken by the brigade to bring the fire under
control. To estimate the value of θ separately for the four periods mentioned above,
the model in Equation (3.42) or Equation (5.3) was expanded, by expanding the
term θT in the exponential to
Table 5.2 Fire growth parameters
Production Storage Other
2 2
Industry A(0)(m ) θA θB A(0)(m ) θA θB A(0)(m2) θA θB
Food, drink, tobacco 0.504 0.020 0.013 0.694 0.017 0.049 0.327 0.042 0.026
Chemicals and allied 0.225 0.038 0.033 0.628 0.048 0.035 0.218 0.027 0.044
Metal Manufacture 0.341 0.033 0.026 1.160 0.017 0.045 0.425 0.032 0.041
Mechanical, instrument 0.248 0.038 0.038 0.619 0.018 0.072 0.225 0.042 0.045
and electrical engineering
Textiles 0.304 0.047 0.029 1.793 0.037 0.037 0.215 0.032 0.053
Clothing, footwear, leather 0.723 0.038 0.064 1.346 0.025 0.039 0.315 0.028 0.075
and fur
Timber, furniture etc 0.485 0.046 0.046 0.949 0.037 0.052 0.566 0.030 0.037
Paper, printing and 0.213 0.044 0.052 0.985 0.027 0.044 0.235 0.023 0.060
publishing
Initiation 205
θ1 T1 + θ2 T2 + θ3 T3 + θ4 T4 (5.4)
With the logarithm of A(T) as the dependant variable, a multiple regression

analysis was performed to estimate the values of θ1, θ2, θ3 and θ4.
In another study, Ramachandran (1986) added the first three time periods
mentioned above to denote the period tA until the arrival of the fire brigade at the
scene of the fire: tA = T1 + T2 + T3. The fourth period, T4, relating to the arrival
time to the time when the fire was fought and brought under control by the brigade
was redefined as tB. The fire growth parameters for the above two periods may be
denoted by θA and θB. Estimates for these two parameters are reproduced in Table
5.2 for some industrial buildings and three areas of fire origin – production, storage
and other areas. The table also contains estimates for A(0).
Fire fighting by the brigade can be expected to reduce the rate of fire growth.
This hypothesis is supported by figures in Table 5.2 for the production areas of
some industries; in these cases θB is less than θA. But for the storage and other
areas of most of the industries, θB is greater than θA. These results do not indicate
that the fire brigades are ineffective in controlling the rate of fire growth. They
indicate, perhaps, the fact that a fire would be growing fast when the fire brigade
arrives at the scene of the fire. The fire size at that time can be expected to be
bigger than the average size before the arrival of the brigade. The results in Table
5.2 do not take into account the size of the fire at the time of brigade arrival, data
for which were not available. Fires would normally continue to grow faster and
faster if they are not extinguished by the fire brigade.
The results in Table 5.2 also do not take into account the interaction between
the time periods tA and tB. If tA is reduced due to early discovery or detection of
a fire and/or due to quick attendance by the fire brigade, the fire will be in its
early stage of growth when fire fighting by brigade commences and hence can be
controlled quickly. A reduction in tA would, therefore, shorten the control time,
tB, as well thus reducing the total duration of burning, T, and area damage. If
installed, sprinklers also would reduce tA and tB. Sprinklers have a high probability
of extinguishing a fire without the need for fire brigade intervention. If a sprinkler
system operates but does not extinguish a fire, it can control the spread of the fire
and reduce its rate of growth.
A fire requiring fire brigade intervention would be a fully developed fire with
growth rate increasing as a function of time before the arrival of the brigade and
for some time after its arrival. Likewise, a fire in which the sprinkler system is
activated would have survived the ‘infant mortality’ stage and commenced
established burning. The growth rate would be increasing in such fires as well.
Fires which are not extinguished by sprinklers would develop into bigger fires
requiring fire brigade action.
For the reasons discussed in the above three paragraphs, the conflicting effects
of several factors confound the application of the exponential model expanded on
the basis of an equation such as Equation (5.4). Such confounding effects produce
interactions between independent variables in a multiple regression analysis and
lead to the statistical problem known as ‘multicollinearity.’ Taking into account
this problem, the additive model in Equation (5.4) needs to be modified but this
is not possible, since fire brigade statistics do not contain the data required for
this purpose. A modified model can be developed and applied by combining
fire brigade data with data provided by experiments and simulations based on
deterministic models.
In another application of the exponential model of fire growth, Ramachandran
(1992) attempted to construct a general picture of fire development by using fire
statistics and evaluating fire growth rates for the following four main scenarios:
1 fires which are extinguished (or self-terminated) by first-aid methods without

any intervention by sprinklers or fire brigades;
2 fires extinguished by sprinklers;
3 fires in buildings with sprinklers extinguished by fire brigades;
4 fires in buildings without sprinklers extinguished by fire brigades.
In addition, for each of the four categories mentioned above, the fire growth
parameter θ was estimated separately for each of the following three extent of fire
spread classifications:
1 confined to item first ignited;

2 spread beyond item but confined to room of origin;
3 spread beyond room of origin.
For the above investigation, only summary (not raw) data for fires during 1984–
1986 were available for analysis. These tables contained, for each scenario, only
means and standard deviations for logarithm of area damage and time variables.
Hence, it was not possible to estimate the parameter A(0) in Equation (5.3) which
was assumed as one square metre. Under this assumption, the mean value of θ was
estimated by the ratio:
θ= y / T (5.5)
where y is the mean value of logarithm (base e) of area damage, A(T) and T
is the mean value of total duration of burning, T. Equation (5.5) follows from
Equation (5.3) with A(0) = 1.
The investigation described above was carried out for four industrial groups of
buildings, retail distributive trade, wholesale distributive trade and office buildings.
As an example, results for the industrial group Paper, Printing and Publishing are
given in Table 5.3 to show the variation in the expected (average) value of the fire
growth parameter θ for different scenarios. The corresponding ‘doubling times’
(minutes) are shown within brackets. As defined in Equation (3.43) the doubling
time is given by (0.6931) / θ.
For each fire-fighting scenario and for each area of fire origin, the figures in
Table 5.3 provide some indication of the increase in the rate of fire growth as fire
spreads within the room of origin. The growth rate for a fire spreading beyond the
Initiation 207
room of origin and the overall growth rate for the building are affected by the fire
resistance of the structural barriers of the room. The frequencies of such fires were
small and even nil in small fires which did not require the intervention of the fire
brigade or sprinklers and in growing fires extinguished by sprinklers.
In storage and other areas equipped with sprinklers, the fire brigades had
been effective in reducing the probability of fire spreading beyond the room of
Table 5.3 Expected values of the fire growth rate θ for different scenarios – paper, printing
and publishing industries, UK
Fire-fighting scenario
Extent
Area of fire of fire No sprinklers No sprinklers Sprinklers Sprinklers
origin spread No fire brigade Fire brigade No fire brigade Fire brigade
Production (a) 0.0385 0.0138 0.1523 0.0235
(18.0) (50.2) (4.6) (29.5)
(b) 0.0668 0.0392 0.1080 0.0576
(10.4) (17.7) (6.4) (12.0)
(c) – 0.0459 – 0.0376
(–) (15.1) (–) (18.4)
(d) 0.0440 0.0307 0.1330 0.0431
(15.8) (22.6) (5.2) (16.1)
Storage (a) 0.0347 0.0235 – 0.0369
(20.0) (29.5) (–) (18.8)
(b) 0.0397 0.0471 – 0.0276
(17.4) (14.7) (–) (25.1)
(c) – 0.0378 – –
(–) (18.3) (–) (–)
(d) 0.0366 0.0401 – 0.0297
(18.9) (17.3) (–) (23.4)
Other (a) 0.0128 0.0091 0.0349 –
areas (54.2) (76.2) (19.9) (–)
(b) 0.0536 0.0316 0.0509 0.0397
(12.9) (21.9) (13.6) (17.5)
(c) – 0.0381 – –
(–) (18.2) (–) (–)
(d) 0.0240 0.0301 0.0437 0.0392
(28.9) (23.0) (15.9) (17.7)
Notes:
– no or few fires (less than 5)
(a) Confined to item first ignited
(b) Spread beyond item but confined to room of origin
(c) Spread beyond room of origin
(d) Building (overall)
The figures within brackets are ‘doubling times’ in minutes.
origin. Structural fire resistance and Fire Service fire fighting, acting together,
appear to have reduced the fire growth rate in the production areas equipped
with sprinklers. In most of the scenarios in which the fires were confined to
the room of fire origin, whether the buildings were equipped with sprinklers
or not, fire brigade action had reduced the growth rate in fires requiring the
intervention of the brigade.
The figures in Table 5.3 provide some support to the fact that, in buildings with
sprinklers, the rate of fire growth will be increasing before and for some time after
sprinklers are operated. Sprinklers have to deal with fires growing in size.
5.2.4 Maximum rate of fire growth

Every fire individually provides an estimate of the growth rate as given by the ratio
θ = Y / T (5.6)
where Y is given by
Y = y – loge A(0) (5.7)
and y is the logarithm (base e) of area damage, A(T), A(0) the area initially
ignited and T the total duration of burning of the fire. In the model discussed
in the previous section, A(0) is one square metre such that Y = y. Under this
assumption, the ratio in Equation (5.6) reduces to
θ = y / T (5.8)
In a population of fires, the growth rate θ has an expected or average value

given by the ‘ratio estimate’
θ= Y / T = ( y − log e A(0) ) / T (5.9)
where Y and T are the mean values of Y and T and y the mean value of loge A(T).
With A(0) = 1, Equation (5.9) reduces to Equation (5.5). According to statistical
theory (Frishman, 1975), the standard deviation of θ, as defined in Equations
(5.6) and (5.8) is the square root of

2 2 2 2
σ2θ = (σ2Y . T −σ2T . Y ) / T (σ2T + T ) (5.10)
where σ2Y is the variance of Y and σ2T the variance of T. The variance of Y is the
same as the variance, σ2Y , of y since A(0) is a constant.
Equation (5.10) reduces to
Initiation 209
2 2
σ (1−ρ ) 2
σ
σ2θ = Y
2
= 2
(5.11)
2 2
σ +T
T σ +T
T
where ρ is the coefficient of correlation between Y and T and σ2 is the ‘residual’

variance. The standard deviation of the residual error, σ, can be estimated by
performing the regression analysis based on the equation
loge A(T) = loge A(0) + θ T (5.12)
which follows from Equation (5.3). The residual error is the difference between
the logarithm of the observed value of area damage in a fire and the corresponding
value for that fire, with duration of burning, T, predicted or estimated by Equation
(5.12) according to the estimated value of θ.
As discussed in the previous section with reference to the results in Table 5.3,
due to the non-availability of raw data, it was not possible to carry out a regression
analysis to estimate the standard deviation, σ, of the residual error. Hence, the
formula for σ2θ given in Equation (5.10) was used to estimate the maximum rate of
fire growth providing the worst case scenario. The maximum rate is estimated by
the upper confidence limit, uθ given by
uθ = θ + t σθ (5.13)
where θ , as defined in Equations (5.5) and (5.9), is the average value of θ for
individual fires defined in Equation (5.6). If θ is assumed to have a normal
distribution in a population of fires, the random variable t has a standard normal
distribution. Under this assumption, the probability of growth rate in a real fire
exceeding the maximum rate uθ given by Equation (5.13) is 0.025 if t = 1.96, 0.01
if t = 2.33 and 0.005 if t = 2.58. For any desired probability level for the maximum
growth rate, the corresponding value of t can be obtained from a table of the
standard Normal distribution.
The figures given in the example in Table 5.3 are the expected values, θ ,
of the growth rates, θ, for individual fires. With t = 1.96 in Equation (5.13),
the corresponding estimates for maximum values of individual growth rates
are given in Table 5.4 together with corresponding estimates for doubling
times. As mentioned above, for any scenario described in this table, the
probability of growth rate in a real fire exceeding the maximum rate given in
the table is 0.025.
5.2.5 Individual growth rate and average growth rate

As mentioned at the beginning of the previous section, every fire provides an
estimate of the growth rate according to Equation (5.6). This individual rate, θ,
varies randomly in a population of fires and fluctuates around the expected or
average value, θ , estimated by Equation (5.9). The fluctuation of θ is according
to the standard deviation σθ estimated by Equation (5.10).
Table 5.4 Maximum values of the fire growth rate θ for different scenarios – paper, printing
and publishing industries, UK
Fire–fighting scenario
Extent
Area of fire of fire No sprinklers No sprinklers Sprinklers Sprinklers
origin spread No fire brigade Fire brigade No fire brigade Fire brigade
Production (a) 0.2045 0.0523 0.4213 0.0637
(3.4) (13.3) (1.6) (10.9)
(b) 0.1777 0.0646 0.1845 0.0868
(3.9) (10.7) (3.8) (8.0)
(c) – 0.0758 – 0.0613
(–) (9.2) (–) (11.3)
(d) 0.1995 0.0792 0.3250 0.0848
(3.5) (8.8) (2.1) (8.2)
Storage (a) 0.2216 0.0597 – 0.2116
(3.1) (11.6) (–) (3.3)
(b) 0.0888 0.0748 – 0.1060
(7.8) (9.3) (–) (6.5)
(c) – 0.0698 – –
(–) (9.9) (–) (–)
(d) 0.1659 0.0721 – 0.0897
(4.2) (9.6) (–) (7.7)
Other (a) 0.05 0.0299 0.1049 –
areas (13.9) (23.2) (6.6) (–)
(b) 0.1330 0.0681 0.0898 0.0418
(5.2) (10.2) (7.7) (16.6)
(c) – 0.0761 – –
(–) (9.1) (–) (–)
(d) 0.0853 0.0542 0.0979 0.0671
(8.1) (12.8) (7.1) (10.3)
Notes:
– no or few fires (less than 5)
(a) Confined to item first ignited
(b) Spread beyond item but confined to room of origin
(c) Spread beyond room of origin
(d) Building (overall)
The figures within brackets are minimum values of doubling times in minutes corresponding to
maximum values of the fire growth rates.
Initiation 211
Each sample of fires provides an estimate of the average rate θ . In samples of
fires, θ can be expected to vary randomly according to its standard deviation σθ
given by
σθ = σ / n. σT (5.14)
where σ is the standard deviation of the ‘residual’ error mentioned with reference
to Equation (5.11) and n the number of observations (fires) in the sample analysed.
The parameter σT is the standard deviation of the total duration of burning, T.
If θ is assumed to have a normal distribution, the maximum value of θ in
repeated samples is given by
u θ = θ + t σθ (5.15)
where the variable t has a standard Normal distribution. In repeated samples, the
probability of average growth rate, θ , exceeding u θ given by Equation (5.15) is
0.025 if t = 1.96, 0.01 if t = 2.33 and 0.005 if t = 2.58. It may be observed that, if
only one sample is available for analysis, the expected value of the average rate θ
is the same as the expected value, θ , of individual rate, θ.
The distinction between the maximum values of individual and average growth
rates can be explained with the aid of the following simple example. Consider the
following sample of n (= 5) observations (x): 4, 9, 12, 15, 20. Calculations show
that the mean ( x ) and standard deviation σx of x are 12 and 5.4. The standard
deviation σ x of the mean x is equal to
σ x / n = 5.4 / 5 = 2.41
Hence, assuming a normal distribution, the maximum value of the individual

observation x is
ux = x + t σx
= 12 + 1.96 × 5.4 = 22.58
if t = 1.96.
The maximum value of the average value x is
u x = x + t σx
= 12 + 1.96 × 2.41 = 16.72
if t = 1.96.
The distinction between the maximum values of individual and average growth
rates was discussed by Bengtson and Ramachandran (1994) in an investigation
concerned with fire growth rates in underground facilities such as railway
properties, public car parks, road tunnels and subways and power stations. The
Table 5.5 Growth rate and doubling time
Occupancy type Number Area initially Average growth rate in all fires (β) Growth rate in an individual fire (θ)
of fires ignited A(0)
Expected Standard Maximum Expected Standard Maximum
(sq metres)
value deviation rate value deviation rate
Railway properties
Fires in all places 776 1.0002 0.0376 0.0021 0.0417 0.0376 0.0352 0.1066
(18.4) (16.6) (18.4) (6.5)
Fires in public places 214 0.8 0.0454 0.0029 0.0511 0.0454 0.029 0.1022
(15.3) (13.6) (15.3) (6.8)
Fires in basement 66 0.82 0.0273 0.0039 0.0349 0.0273 0.0226 0.0716
(25.4) (19.9) (25.4) (9.7)
Public car parks
Fires in all places 692 1.02 0.0362 0.0025 0.0411 0.0362 0.0318 0.0985
(19.1) (16.9) (19.1) (7.0)
Fires in basement 165 1.26 0.0366 0.0058 0.0480 0.0366 0.0327 0.1007
(18.9) (14.4) (18.9) 6.9)
Road tunnels and subways 107 1.00 0.0220 0.0024 0.0267 0.0220 0.0176 0.0565
(31.5) (26.0) (31.5) (12.3)
Power stations 115 0.93 0.0208 0.0029 0.0265 0.0208 0.021 0.0620
(33.3) (26.2) (33.3) (11.2)
The figures within brackets are corresponding doubling times in minutes.
Initiation 213
results obtained in this investigation are reproduced in Table 5.5, where the
average growth rate is denoted by β.
As one would expect from Equations (5.10) and (5.14), the standard deviation
σθ of individual growth rate is higher than the standard deviation σθ of average
growth rate. Individual growth rates have wider fluctuations around the expected
value than average growth rates. The maximum value of individual growth rate
represents more realistically the worst case scenario.
5.2.6 Applications
As discussed in Section 3.3.2 and previous sections of this chapter, the exponential
model provides a tool for estimating the area likely to be damaged in a fire as
a function of total duration of burning until the fire is extinguished by first-aid
means, sprinklers or fire brigades. The damage can be reduced by early discovery
of the fire by occupants of the building or by early detection by automatic
detectors. Early discovery or detection should be followed by quick action to call
the fire brigade and, if possible, fight the fire with the aid of first-aid means such as
portable fire extinguishers, buckets of water or sand and smothering.
The model can also provide some statistical basis for estimating the reduction
in area damage and the consequential reduction in life risk due to quick response
and early attendance by the fire brigade to arrive at the fire scene and commence
fire fighting. The model may be useful in evaluating the economic value of bringing
the fire under control quickly by improving the fire-fighting tactics adopted by the
fire brigades. For the reasons mentioned above, the exponential model can provide
inputs to fire brigade problems concerned with the determination of appropriate
fire cover for a geographical area in terms of number, location and size (manpower
and equipment) of fire stations.
In Section 3.3.2 and previous sections of this chapter, the variable A(T) was
used to represent the area damaged by direct burning. This information can be
used to estimate the rate of growth of a real fire in terms of heat output – see
Ramachandran (1995). The rate, (dL/dT), at which fire load (L) in a compartment
is destroyed in a fire can be expected to be equivalent to m (kg/sec), the rate at
which fuel mass is consumed. The fire load contained in A square metres of floor
area is approximately A L where L (kg/m2) is the fire load density.
It follows, therefore, that
m = (dL /dT) = L . (dA /dT) (5.16)
where (dA/dT) can be estimated by Equation (5.3) with T in seconds:

dA
= A(0).θ.exp (θT) (5.17)
dT
Hence,
m = mo exp (θT) (5.18)

where
mo = L A (0) θ (5.19)
is the loss rate of fuel mass at the initial time corresponding to the commencement
of established burning. Also, as shown in Equation (3.58), the rate of heat output
Q, is directly proportional to m.
In deriving the above equations it has been assumed that m and Q increase
exponentially with duration of burning since the commencement of established
burning. In the next section, an application of these equations is discussed with
reference to the time of occurrence of flashover.
Based on the rate of growth of fire in terms of heat output as discussed above,
the rate of growth of smoke can be estimated by ascertaining the correlations
between the two rates. Quantity of smoke produced is correlated with heat output.
Smoke can be expected to grow exponentially with time faster than heat with a
growth parameter two or more times the parameter θ for heat development. Rate
of growth of smoke can also be estimated directly, to some extent, using data on
total area damage instead of area damage by direct burning. Fire incident reports
compiled by the fire brigades in the UK also contain information on total area
damage but this includes water damage in addition to smoke damage.
Butcher (1987) attempted to establish the relationship between fire area and
heat output. He used the results of a series of large-scale fire tests staged at the
Fire Research Station, UK in 1966, in which a selection of fire loadings and two
levels of window opening were considered. The size of the fire compartment was
85.5m2. Time and temperature information for these tests was available from
which Butcher derived the time–temperature curve for a compartment with the
largest fire load density of 60kg/m2. The value of heat output estimated from this
curve was combined with the progressive area increases obtained by using the
results on exponential fire growth produced by Ramachandran (1986). The heat
output thus obtained for each fire area, at the appropriate time, was integrated to
provide a value for the total heat output for the growing and spreading fire for any
time value in the fire’s history.
Based on the above analysis, Butcher showed that, for the example considered
by him, it would take 22 minutes for a spreading and growing fire to reach a heat
output of 5MW. He questioned the validity of using, for heat output, a constant
value of 5MW over an area of 10m2 or 0.5MW per m2 for designing smoke
ventilation systems. Since there can be an appreciable delay in a fire reaching
the full 5MW heat output, calculating smoke temperature from the 5MW value
can give a false picture of smoke movement for the early period of fire when the
occupants of a building are attempting to escape.
As pointed out by Butcher, the concept of a constant design fire size such as
5MW over 10m2 area or 0.5MW per m2 for designing smoke ventilation systems
is somewhat misleading. The final fire size attained is less relevant if escape of
occupants takes place during the early stages of a fire. The design fire size need
not have a single value. It should be a function of time in order to consider the
Initiation 215
interaction between smoke movement and the movement of escaping occupants.
Such a functional relationship between fire size and time would also provide a tool
for determining the optimum time for the operation of detectors and sprinklers to
facilitate safe evacuation. There are also uncertainties involved in the operating
times of detectors, sprinklers and smoke ventilation systems. Hence, there is a
need to determine a probability–time-based design fire size for these fire protection
systems. This problem is investigated in the next chapter.
The fire growth rate derived from statistics of real fires, as discussed in this
section, involves a number of materials or objects and structural elements of a
building. The expected and maximum values of this growth rate can, therefore,
provide tools for testing whether the growth rates for different scenarios generated
by a deterministic model satisfactorily reflect real fire situations. These tools can
be improved by combining the statistically determined fire growth rates with those
estimated by experiments involving physical quantities such as heat output. For
this purpose, it would be necessary to estimate first a composite growth rate for a
room or a building based on experimental results which are generally for individual
materials. This might be a complex exercise. Statistical and experimental growth
rates can be merged by applying Bayesian statistical technique – see Ramachandran
(1998).
Fire growth rates can be estimated for different materials ignited first in order
to describe the early stage in the development of a fire before, say, the arrival of
the fire brigade at the fire scene – see Ramachandran (1986). Fire spread from
material to material in a room would depend on room dimensions, ventilation
and environmental conditions apart from the arrangement (overcrowding etc.) of
the materials. Taking into account these and other factors, the statistical growth
rate for any scenario, such as those discussed in this section, should be modified
or adjusted.
5.3 Probability of flashover
5.3.1 Introduction
Generally, the structural elements of a compartment would only be affected
if a fire grows into a fully developed stage. This stage, defined as ‘flashover’, is
reached when the atmosphere temperature at the ceiling exceeds 520°C and the
heat output attains a certain high level. This level depends on factors such as
total area of the compartment walls, floor and ceiling, area of window (opening),
height of window and heat transfer (loss) coefficient of the wall. Formulae based
on scientific theories are available for estimating the heat output required for the
occurrence of flashover in a compartment – see, for example, Walton and Thomas
(1988).
The time taken by a fire to reach the flashover stage depends on the rate at
which the fire grows. This rate depends on whether the compartment is protected
by sprinklers or not, apart from the factors mentioned above and the combustible
nature of materials or objects in the compartment. The rate also depends on
whether the compartment has or does not have an installed ventilation system.
Bengtson and Laufke (1979) have suggested a model for estimating the time to
flashover for different room volumes and for the two cases, with and without an
installed ventilation system.
The heat output produced in a compartment at the time of flashover would
have destroyed a certain floor area of the compartment. By evaluating this area,
the time to flashover since the onset of established burning can be estimated by
applying the exponential model in Equation (5.3) together with Equation (3.58)
and Equations (5.16) to (5.19).
The models discussed above (including the exponential model) are essentially
deterministic in nature. In a real fire, area damage, duration of burning and heat
output or fire severity are all random variables due to uncertainties caused by several
factors. Hence, an element of probability governs the occurrence of flashover
and the time of its occurrence. To estimate this probability, Ramachandran has
developed three models which are discussed briefly below.
5.3.2 Exponential model of fire growth

The first model (Ramachandran, 1995) provides a mechanism for coupling
deterministic rate for increase in heat output in a compartment with exponential
fire growth rate based on area damage in real fires. Applying this model to the
deterministic rate of heat output required for causing flashover, the time of
occurrence of flashover and the total floor area damaged at that time are predicted.
The probability distribution of area damage is then used to provide an estimate or
probability of flashover. Essential features of this model are as follows.
As discussed earlier and expressed in Equation (3.58), the rate of heat output, Qf
(in kw), at the time of flashover, is directly proportional to the rate of consumption
of fuel mass, mf (in kg/sec) at the time of flashover:
Qf = mf ∆H (5.20)
where ∆H is the effective heat of combustion of the fuel, usually assumed to have
the value 18,800 kilojoules per kilogram (kJ/kg). Hence, applying the exponential
model it is reasonable to assume that
Qf = Qo exp (θh Tf) (5.21)
where:
Qf = rate of heat output at the time of flashover
Qo = rate of heat output at initial time, To
θh = increase per second in the value of logarithm (base e) of rate of
heat output Q
Tf = time (in seconds) taken for the occurrence of flashover since the
commencement of established burning at the initial time, To.
As in Equation (5.20),
Initiation 217
Qo = mo ∆H (5.22)
where, following Equation (5.19),
mo = L A(0) . θa (5.23)
The parameter mo is the loss rate of fuel mass at the initial time, L (kg/m2)
is the fire load density and A(0) (in m2) is the area initially ignited at the
commencement of established burning. The fire growth parameter θa is the
increase per second in the value of logarithm (base e) of area damage. The value
of θa can be estimated by (θ/60) where θ is the value of the fire growth parameter
estimated by Equation (5.3) in units of minutes.
The value of θh in Equation (5.21) may be equated to θa in Equation (5.23) i.e.
θh = θa, assuming that heat output grows exponentially in time at the same rate
as for area damage. The worst case scenario may also be considered by using the
maximum value of θa, estimation of which has been discussed in Section 5.2.4.
For any compartment, the rate of heat output Qf, required for the occurrence of
flashover with a temperature of 520°C can be estimated by the following formula
suggested by Walton and Thomas (1988):
Qf = 610 (hK. AT. Ao . h o )½ (5.24)
where
hK = wall heat loss coefficient [(kW/m)/K]
AT = total area (m2) of the compartment excluding area of window
opening
Ao = area of window opening (m2)
ho = height of window opening (m).
If the fire load density of the compartment is L , the rate of heat output at
initial time, Qo, can be estimated with the aid of Equations (5.22) and (5.23) and
an assumed or estimated mean value for A(0), the area initially ignited at the
commencement of established burning.
With values for Qf, Qo and θh estimated as discussed above, the time, Tf,
taken for the occurrence of flashover can be estimated with the aid of Equation
(5.21). The probability of occurrence of flashover can then be estimated by
considering the probability attached to the time, Tf, for the occurrence of
flashover and the corresponding (cumulative) area, Af, destroyed. With T = Tf
in seconds, Af is given by Equation (5.3) with θ = θa calculated for time units
in seconds. The probability of damage exceeding Af can then be estimated by
evaluating the parameters of the probability distribution of area damage for the
compartment and occupancy type considered. This probability is equivalent to
the probability, Pf, of flashover. According to analyses of data for actual fires
attended, particularly by fire brigades in the UK, area damage has a log normal
or Pareto probability distribution.
Consider, as an example, an office room of width 3.6m, depth 6.1m and height
3m. It has a self-closing fire-resistant door which need not be considered as an
opening. The room has a window (opening) of height 1.5m and width 1.2m. The
wall-lining material is 0.016m gypsum plaster on metal lath. The wall heat loss
coefficient, hk, may be assumed to have the value 0.03 [(kW/m)/K] for the period
after the occurrence of established burning. With hk = 0.03, AT = 100.32m2, Ao
= 1.8m2 and ho = 1.5m, the heat output rate, Qf , required for the occurrence of
flashover is 1570kW according to Equation (5.24). Using Equation (5.20), the loss
rate of fuel mass at the time of flashover, mf , is 0.0835 kg/sec.
For an office room, the fire load density, L , is 23kg/m2. For an unsprinklered
office room, according to an analysis carried out by Ramachandran (1992), the
fire growth parameter θ, defined in Equation (5.3) in terms of minutes, has an
average (expected) value of 0.04 and maximum value of 0.06. Hence, in terms
of seconds, the fire growth parameter θa, defined in Equation (5.23), has the
average value of 0.0007 and maximum value of 0.001. The values mentioned
above are applicable to the period after the occurrence of established burning. It is
assumed that established burning would occur when fire load on one square metre
Sprinklered Non-sprinklered
(Insufficient data available)
Area Probability of Probability of Area
damage confinement States confinement damage
(m2) (Ei ) (Ei ) (m2)
µ1 µ1
1 0.22 5.83
0.22
λ1 λ1
0.78
µ2 µ2
2 0.30 12.28
0.29
λ2 λ2
0.71
µ3 µ3 12.21
3 0.35
0.63
λ3 λ3
0.37
µ4 µ4
4 0.20 194.77
1.0 1.0
Average area damage (m2)
47.46
States
1 Confinement to items first ignited
2 Spread beyond items first ignited, but confinement to contents of room of fire origin
3 Spread beyond items first ignited, but confinement to room of fire origin with involvement of structure
4 Spread beyond room of fire origin
Figure 5.1 Probability tree for rooms and offices

Initiation 219
is consumed or destroyed by fire such that A(0) = 1. A fire is unlikely to reach the
flashover stage if established burning does not occur. About 22 per cent of fires in
office rooms do not spread beyond the item first ignited – see Figure 5.1.
Consider the worst case scenario in an actual fire in an office room with the
maximum value of 0.001 (per second) for θa. With L = 23kg/m2, A(0) = 1 and
θa = 0.001, mo = 0.023kg/s according to Equation (5.23). Then, from Equation
(5.22), Qf = 432kW. Hence, with θh = θa, from Equation (5.21),
Qf = 432 exp (0.001 Tf) (5.25)
Therefore, from Equation (5.26), with Qf = 1570 kW, Tf may be estimated to

have the value 1290 seconds or 21.5 minutes. Hence, if a fire in an office room
reaches the established burning stage with a minimum damage of one square
metre, it will take a further 21.5 minutes to reach the flashover stage.
When flashover occurs, from Equation (5.3), with θ = 0.06, A(0) = 1 and
T = 21.5, an area of 3.63m2 would have been affected by direct burning (heat).
According to Ramachandran (1993b), area damage (d) in a fire in an office room
has a Pareto probability distribution:
Q(d) = c d-λ (5.26)
where Q(d) is the probability of area damage exceeding d. For an unsprinklered

office room, C = 0.84m2 and λ = 0.67 such that the probability of damage
exceeding 3.63m2 or probability of flashover is 0.35.
Area damage in a sprinklered office room has a Pareto probability distribution
of the form in Equation (5.26) with C = 0.28m2 and λ = 0.70. Hence, for a
sprinklered office room, the probability of damage exceeding 3.63m2 or probability
of flashover is 0.11.
5.3.3 Event probability tree model

The event tree model was discussed in detail in Section 3.2.1. It was also discussed
in Section 3.4.3 with reference to the stochastic state transition model (STM) – see
Figure 3.18. Figure 5.1 is an event tree relating to an office room. The probabilities
Ei (i = 1, 2, 3, 4), λi (i = 1, 2, 3) and µi (i = 1, 2, 3, 4) in this figure were defined
in Section 3.4.3. The values of these probabilities are based on an analysis of fires
attended by fire brigades – see Ramachandran (1992).
The parameters λi and µi are limiting values to which these probabilities for
the ith stage would tend over a period of time. They are constants and are not
functions of time since the start of the fire. Figures for area damage are averages
for the four states (the figure 12.21m2 is the combined average for the second and
third states).
The product (λ1. λ2) estimates the probability of a fire spreading beyond the
item first ignited to other items in the room but remaining confined to the room.
This probability (product) may be considered as the probability of flashover in the
statistical sense. Under this assumption, it may be seen from Figure 5.1 that the
probability of flashover for an office room without sprinklers is 0.55 (= 0.78 × 0.71).
The probability λ3 (= 0.37) is the conditional probability of thermal failure of the
structural boundaries of the room given that flashover has occurred. The product
λ1 . λ2 . λ3 (= 0.20), as indicated in the figure for the parameter E4 for the fourth
state, is the probability of fire spreading beyond the room of origin but confined
to the building of origin. Fires spreading beyond the building of origin were not
considered in this model.
The probability of flashover provided by an event tree model, such as Figure
5.1, is an average or expected value for all office rooms. It applies to an ‘average’ or
‘reference’ office room with average dimensions, fire load, ventilation conditions
and physical characteristics, discussed in Section 5.3.2. Some research, data
collection and analysis are needed to adjust the global probability of flashover
given by the product (λ1 . λ2) for a particular room in an occupancy type, say, an
office room with known dimensions and other characteristics mentioned above.
This problem will not arise if the method proposed in Section 5.3.2 for a particular
room is applied together with the probability distribution of area damage.
Based on event trees such as Figure 5.1, the probabilities of flashover
have been estimated for a few industrial buildings and areas of fire origin
(Ramachandran, 1992) and areas of non-industrial buildings (Ramachandran,
1999). The results obtained in these two reports are reproduced in Table 5.6.
The parameter K in the last column is the ratio between the probabilities of
flashover in unsprinklered and sprinklered rooms. This ratio can be used to
determine the reduction in fire resistance that can be allowed for a sprinklered
room – see Section 12.2.1.5.
5.3.4 Probability distribution of area damage

For any type of building and room of fire origin, the global probability of flashover
can also be estimated directly by evaluating the probability distribution of area
damage. As mentioned in the previous section, the global probability would only
be applicable to a room with average characteristics relating to heat loss coefficient,
room dimensions and ventilation factor. It is not applicable to any particular room.
In this global approach, the necessary first step is to determine, for any building
type and room of fire origin, a design area damage, Af, affected by direct burning
(heat) which is likely to cause flashover. Flashover has been defined in fire
science literature as the beginning of the fully developed stage of a fire when the
atmosphere temperature at the ceiling exceeds 500°C. As discussed in Section
5.3.2, a certain rate of heat output, Qf , is required to cause flashover.
Suppose, for example, for a range of room sizes and other characteristics of
an office room, the design value for Qf has been estimated to be 2200kW. The
corresponding design value for the rate of consumption of fuel mass, mf , is 0.117
kg/s since, from Equation (5.20), mf = Qf /∆H where ∆H = 18,800 kJ/kg is the
effective heat of combustion of the fuel. The rate mf is applicable to a short interval
of time around the time Tf of flashover. For an exponential model
Initiation 221
Table 5.6 Probability of flashover
Sprinklered building Unsprinklered building Para-
Occupancy type, meter
Prob. of Prob. of
area of fire origin λ1 λ2 flashover λ1 λ2 flashover K*
Textile industry
Production 0.28 0.36 0.10 0.57 0.44 0.25 2.50
Storage – – 0.28 0.81 0.78 0.63 2.25
Other areas 0.34 0.35 0.12 0.58 0.57 0.33 2.75
Chemical etc
industry
Production 0.20 0.20 0.04 0.61 0.41 0.25 6.25
Storage 0.37 0.41 0.15 0.81 0.67 0.54 3.60
Other areas 0.29 0.31 0.09 0.61 0.61 0.37 4.11
Paper etc
industry
Production 0.23 0.26 0.06 0.49 0.39 0.19 3.17
Storage 0.29 0.34 0.10 0.78 0.72 0.56 5.60
Other areas 0.28 0.36 0.10 0.73 0.71 0.52 5.20
Timber etc
industry
Production 0.36 0.39 0.14 0.80 0.69 0.55 3.93
Storage 0.38 0.42 0.16 0.84 0.83 0.70 4.38
Other areas 0.30 0.40 0.12 0.76 0.76 0.58 4.83
Retail trade
Assembly 0.28 0.32 0.09 0.67 0.60 0.40 4.44
Storage 0.24 0.25 0.06 0.82 0.76 0.62 10.33
Other areas 0.27 0.41 0.11 0.71 0.68 0.48 4.36
Wholesale trade
All areas 0.26 0.42 0.11 0.85 0.75 0.64 5.82
Office premises
Rooms used as
offices – – – 0.78 0.71 0.55 –
Other areas 0.21 0.43 0.09 0.65 0.65 0.42 4.67
All areas 0.42 0.60 0.25 0.63 0.84 0.53 2.12
Hotels
All areas 0.15 0.47 0.07 0.70 0.71 0.50 7.14
Pubs, clubs,
restaurants
All areas 0.41 0.63 0.26 0.74 0.84 0.62 2.38
Hospitals
Other areas 0.13 0.92 0.12 0.46 0.65 0.30 2.50
Flats
Other areas 0.10 0.40 0.04 0.47 0.79 0.37 9.25
Probability of flashover = λ1 . λ2
K* = Ratio between the probabilities of flashover in unsprinklered and sprinklered rooms
mf = mf . θa (5.27)
where mf is the total mass equal to the total fire load (Lf) affected by heat at
the time of flashover. Hence, Lf = 117kg (= 0.117/0.001) since, as discussed in
Section 5.3.2, the maximum value of θa for an office room is 0.001. Since Lf = Af.
L where L = 23kg/m2 is the fire load density for an office room, the total floor
area, Af, affected at the time of flashover is 5.1m2.
According to Equation (5.3), with θ = 0.06 in terms of minutes, area damage
of 5.1m2 and flashover would occur in 27.1 minutes since the commencement of
established burning. This result also follows from Equation (5.25) with Qf = 2200
according to which Tf = 1628 seconds.
Equation (5.27) follows from the fact that the mass m contained in A m2 is
equal to A L and hence, at time T, the mass destroyed is given by
m(T) = L A(T) = L A(0) exp (θT) (5.28)
Therefore
dm
m(T) = = L A(0) . θ . exp (θT)
dt
= m (T) θ (5.29)
Having determined the design area damage, Af, for flashover, the next problem
is to estimate the probability of damage exceeding Af. Using Equation (5.26) for
Pareto probability distribution, the probability of damage in an unsprinklerd office
room exceeding 5.1m2 may be calculated to be 0.28. This probability (0.28) may
be considered to be the global probability of flashover for an office room without
sprinklers. The global probability of 0.55 for flashover to occur in an office room
without sprinklers, given in Table 5.6, has been estimated purely from fire statistics
and has not been adjusted for room dimensions, ventilation, fire load and other
physical characteristics.
In the example considered above, the rate of heat output of 2200kw at the time
of flashover has been estimated to be over a floor area of 5.1m2. This is equivalent
to about 432 kW per m2. This result has also been obtained in Section 5.3.2 where,
for the office room considered, the rate of heat output of 1570 kw at the time of
flashover was estimated to be over 3.63m2. This statistical property is a feature of
the exponential model according to which the heat output rate Qf per unit area
is a constant and is equal to Qf /A(0). For the example (office room) considered,
as expressed in Equation (5.25), Qo = 432 kw and A(0) = 1. It follows that the
loss rate of fuel mass m per unit area is also a constant and equal to mo/A(0) =
0.023kg/s for the example considered.
5.3.5 Hotel bedroom

Lawson and Quintiere (1986) performed an analysis of fire growth in the bedroom
of a hotel. The authors considered a particular scenario in which a fire started from
Initiation 223
a cigarette lighter in the centre of a bed with polyurethane foam mattress. The room
door was left one-quarter open as the occupant fled the room. For mass loss rate of
fuel, the authors used the following exponential function for times up to 250 sec:
m = α exp (αt), α = 0.03 (5.30)
where m was expressed in g/sec and t in seconds. In kg/sec, the Equation (5.30)
may be rewritten as
m = 0.00003 exp (0.03t) (5.31)
Mass loss rates for times from 250 secs to 300 secs were calculated using a
steady state burning formula for wood cribs.
It may be inferred from the above analysis that a steady state, or established
burning, would occur in 250 secs when the fuel mass loss rate, m, reaches a value
of 0.05424 kg/sec according to Equation (5.31). This value is m in the exponential
model discussed in this chapter. Using the value of 15.7 kJ/g or 15,700 kJ/kg used by
the authors for the parameter ∆H for polyurethane foam mattress, from Equation
(5.22), the value of Qo at the time of commencement of established burning may
be calculated to be 851.6kW. From Equation (5.29), the mass mo involved at the
time of established burning is 1.808 kg (= mo /θ). This mass would occupy an area
of 0.11m2 since the average fire load density, L , for a hotel bedroom is 310MJ/
m2 or 17 kg/m2 approximately, according to European data – see CIB W14 Design
Guide for Structural Fire Safety (1986). Data are not available for the fire load
density of hotel bedrooms in the USA.
According to the above analysis, the area affected, A(0), at the time of
commencement of established burning is 0.11m2. This value may be applicable
for the particular scenario considered. A higher value may be applicable for
the average value of A(0) over a range of scenarios for fires starting with the
ignition of different objects in a hotel bedroom. A value of one square metre has
been assumed for A(0) in the exponential model discussed in this chapter. The
corresponding value for mo for an office room has been estimated as 0.023 kg/s.
By carrying out a fire load survey it would be possible to estimate a more accurate
value for A(0) for any occupancy type and room or area of fire origin. This would
give more accurate values for mo (= A(0) . L ) and mo.
For hotels in the UK, data on area damage and duration of burning are available
for estimating the fire growth parameter θ in Equation (5.3) and hence for θa in
Equation (5.23) or for θh in Equation (5.21). But this investigation has not yet
been carried out. Area damage in UK hotels has a Pareto probability distribution.
For unsprinklered areas of hotels, the parameter λ in Equation (5.26) has the value
of 0.66 for assembly areas, 0.77 for bedrooms and 0.64 for storage and other areas
– see Ramachandran (1993b). Respectively for these three areas, the parameter c
in Equation (5.26) has the values of 0.71m2, 0.54m2 and 0.38m2. For storage and
other areas of hotels with sprinklers, the values of the parameters λ and c are 0.63
and 0.11 respectively.
5.3.6 Compartment size
If the size (floor area) of a compartment is increased, the total fire load in the
compartment would also increase, which will increase the potential for a fire to
reach a high level of severity. Hence, one can expect the probability of flashover
to increase with an increase in compartment size. In Section 5.3.3, using the event
tree in Figure 5.1 the probability of flashover in an unsprinklered office room was
estimated to be 0.55. This global probability is applicable to an average room size
of 50m2 estimated from a sample of fires in office buildings. In Section 5.3.2, the
probability of flashover in an unsprinklered office room of floor area 22m2 was
estimated to be 0.35. In estimating this probability, room dimensions including
height, ventilation and other factors, were considered but these factors were not
taken into account in the event tree method.
It can, however, be argued that the probability of flashover would decrease
with increasing size of a room or compartment. Ramachandran (1990) provided
statistical support for this hypothesis. The larger the floor area of a compartment,
the longer it takes, generally, for a fire to involve a number of objects and produce
sufficient heat to cause flashover. The extra time thus available would increase
the chance of extinguishment of the fire by first-aid means or by the fire brigade.
A larger room, generally, has a greater non-uniformity in the arrangement
of objects and hence in that of fire load and lesser degree of overcrowding of
objects. Probability of fire spread would decrease with increasing distances
between objects.
In the particular unsprinklered office room of floor area 22m2 considered
in Section 5.3.2, the value of 0.35 estimated for the probability of flashover is
associated with heat output produced by an area damage of 3.63m2. In Section
5.3.4, the probability of flashover for an unsprinklered office room was estimated
to be 0.28. This probability is for a design value of 2200kW for the heat output
rate required for causing flashover. This design value is associated with an area
damage of 5.1m2. The flashover heat output rate of 2200kW and area damage
of 5.1m2 can be expected to apply to a compartment larger than 22m2 with a
flashover heat output of 1570kW and area damage of 3.63m2.
As discussed above, the total floor area damaged, df, when flashover occurs
can be expected to increase with increasing compartment size. The increase can,
perhaps, be quantified approximately by the following ‘power’ function:
df = CAβ (5.32)
where A is the compartment size. With C = 1 and df = 3.63m2 for A = 22m2, the
value of β in Equation (5.32) can be estimated to be 0.42. With this value of β in
Equation (5.32), a damage of 5.1m2 at flashover would occur in a compartment
of size 48m2.
According to the statistical property of a probability distribution, the
probability of damage in a fire exceeding any value specified for the damage,
d, would decrease with increasing values for d. This would be apparent from
Initiation 225
Equation (5.26) for Pareto distribution. It may, therefore, be inferred that
the probability of flashover would decrease with increasing area damage for
increasing compartment size. In a larger compartment, area damage at flashover
would be higher but the probability of this damage level being exceeded or
probability of flashover would be lower. This paradox can be explained by
the fact that, due to uncertainties caused by several factors, area damage in a
real fire is a random variable with a probability distribution. It is unrealistic to
assume that area damage in a real fire can be predicted in exact terms satisfying
deterministic formulae. The concept of probability distribution of area damage
has been applied in the determination of design fire size for designing smoke
ventilation systems – see Section 6.3.
5.3.7 Time before established burning

It may be desirable to consider the duration of burning since the start of the fire
before the occurrence of established burning. This initial stage of fire growth,
although small in size, can be very variable in length of time; it can last for hours
or it can be over in minutes – see Butcher (1987) and Ramachandran (1988).
Ignoring such uncertainties, one may consider the deterministic model such as in
Equation (5.30) for the very early stage of fire growth. According to this equation,
for the scenario considered, it would take 250 seconds or 4.2 minutes before
established burning occurs.
Alternatively, one may consider the following T2 curve:
Q = α T 2
(5.33)
For an office room with medium fire growth, α = 0.012. Hence, on average it
would take about 190 seconds or 3.2 minutes for a fire in the room considered in
Section 5.3.2 with an initial heat output rate, Qo, of 432 kW.
5.4 Probable damage in a fire
5.4.1 Introduction
As discussed in Section 5.1, the probability of fire starting or frequency of fires
occurring during a period is the first component of fire risk. The second component,
which is the subject matter of this section, is the probable damage if and when a
fire occurs. The damage can be measured in terms of any of the following four
attributes:
1 Extent of spread
2 Floor area destroyed
3 Financial loss
4 Life loss – number of fatal and non-fatal casualties.
The probable damage in a fire in a building would depend on the levels of
passive fire protection measures such as compartmentation, fire resistance of
structural boundaries (walls, floor, ceiling) and means of escape facilities (number
and widths of staircases, travel distance etc.) and on the presence or absence
of active measures such as automatic fire detection and sprinkler systems. The
probable damage is also affected by the successful operation and reliability of the
fire protection measures when a fire occurs and their effectiveness in reducing the
damage.
5.4.2 Extent of fire spread and floor area damaged

The probable damage in a fire can be estimated by considering different categories
of fire spread and the probabilities and average or expected floor area damage
associated with these categories. The probabilities and area damage can be
estimated with the aid of statistical data provided by real fires. Such statistics
produced by the fire brigades in the UK enable the extent of fire spread to be
classified as follows:
1 Confined to item first ignited;

2 Spread beyond item but confined to room of origin:
• confined to contents only;
• structure involved.
3 Spread beyond room but confined to floor of fire origin;
4 Spread beyond floor but confined to the building of fire origin;
5 Spread beyond the building of fire origin.
A fire starting in a room can spread upward to the next floor without involving
the entire floor of origin. It is not possible to estimate the number of such cases.
Hence, in the example shown in Table 5.7, the third and fourth categories have
been combined to denote the event of fire spreading beyond the room of origin
but confined to the building of origin. Fires spreading beyond the building of origin
have not been included in this table.
A fire can spread beyond the room of origin without involving the structural
boundaries of the room but such cases are rare occurrences. A fire generally
reaches the post-flashover stage, involving all the contents in a room, and then
spreads beyond the room by attacking the structural boundaries and causing their
thermal failure. Fire spread due to the collapse or destruction of the structural
boundaries only occurs very rarely.
For each category of spread, the area damage shown in Table 5.7 is the
average value for the category. The percentage figure for each category denotes
the probability attached to the category and to the corresponding average area
damage.
In the case of a sprinklered building, the percentage figures include one-third
of fires in these buildings which were estimated to be extinguished by the system,
but not reported to the fire brigades (Rogers, 1977). In other words, fire brigades
Table 5.7 Fire extent of spread –textile industry, UK
Sprinklered Un-sprinklered
Extent of spread Percentage of Average area Financial loss Percentage of Average area Financial loss
fires damage (m2) (£) fires damage (m2) (£)
Confined to item first ignited 72 4.43 3278 49 4.43 3278
Spread beyond item but confined to
room of fire origin:
i) contents only 19 11.82 8747 23 15.04 11130
ii) structure involved 7 75.07 55,552 21 197.41 146,083
Spread beyond room 2 1000.00 740,000 7 2000.00 1,480,000
Average damage 30.69 22710 187.08 138,440
Notes:
Fires considered in the case of sprinklers are those in which the system operated.
In the case of sprinklers, the percentage for the first category, confined to item first ignited, includes one-third of fires extinguished by the system but not reported to the
fire brigade.
The financial loss per m2 is £740 at 1999 prices updating for inflation, Home Office estimate of £225 per m2 at 1978 prices for textile industry (Maclean, 1979)
only attend two-thirds of fires in sprinklered premises. Most of the unreported
small fires in sprinklered buildings were confined to the items first ignited. No
insurance claims were made for compensating the financial losses in these small
fires. It is apparent that, if they operate, sprinklers would increase the probability
of a fire being confined to the item first ignited and thus reduce the probability of
the fire spreading beyond the room of origin. Consequently, as shown in Table 5.7,
sprinklers reduce considerably the overall damage expected in a fire. In many fires,
the heat generated may not be sufficient to activate the sprinkler heads. There are
also other causes for non-operation of sprinklers in fires – see Section 8.3.2.
Fire statistics collected by the US Fire Administration provide figures for
probabilities and dollar losses for different categories of fire spread. In a study
concerned with residential fire loss, Gomberg, Buchbinder and Offensend (1982)
estimated dollar losses for different spread categories which were the same for both
sprinklered and non-sprinklered buildings. They differentiated the probabilities of
extinction to reflect the effectiveness of sprinklers. Their study also included the
effectiveness of smoke detectors and life loss (fatalities and injuries).
Gomberg, Buchbinder and Offensend (1982) used probability trees to assess
the final extent of flame spread and the consequences in terms of dollar loss and
life loss. Three possible levels of spread were considered – confined to the object
of origin (O), spread beyond this object but confined to part of the room of origin
(< R) and spread beyond room (≥ R). Figure 5.2 is an example reproduced from
this study. The ‘suppression size’ in this figure denotes the fire size at the start
of a suppression activity. As with UK fire statistics, the US database does not
provide probabilities for suppression size since only the final size after a fire was
extinguished is recorded in fire reports. Hence, expert judgement was used to
assess the suppression size.
The overall floor area expected to be damaged in a fire can also be estimated
directly by fitting the probability distribution of area damage instead of evaluating
the probabilities and average damage for each of the extent of spread categories.
This distribution is log normal or Pareto as discussed in Section 3.3.4.
5.4.3 Financial loss

As shown in Table 5.7, the expected value for financial loss in a fire can be
determined approximately by estimating the expected value for area damage
(m2) and using an estimate for financial loss per m2. Data on fire insurance claims
(losses) compiled by insurance companies, particularly for large losses, if collated
with fire brigade reports on such fires, can provide estimates for financial losses per
m2 for different occupancy groups and risk categories.
Insurance sources can also provide data for estimating the probability
distribution of financial loss which, as discussed in Section 3.3.4, is log normal or
Pareto. By identifying and fitting an appropriate distribution to observed values
of losses, the expected value, standard deviation and other parameters of fire loss
can be evaluated for any occupancy group or risk category. For this estimation,
standard statistical techniques can be applied if loss data are available for all fires
0 .000163 .020013 $ 385
.947
0 <R .000917 .101283 $2028
.487 .050
≥R .007251 .101283 $8697
.003
Functional <R <R .000917 .062472 $2028
.127 .513 .991
≥R .007251 .101283 $8697
.009
≥R ≥R .007251 .101283 $8697
.000 .000
0 .000343 .019530 $ 385
.936
0 <R .001935 .060962 $2028
G .459 .052
Interior ≥R .015298 .098836 $8697
living space .011
Nonfunctional <R <R .001935 .060962 $2028
.306 .017 .541 .929
≥R .015298 .098836 $8697
.071
≥R ≥R .015298 .098836 $8697
.000 .000
0 .000343 .019530 $ 385
.936
0 <R .001935 .060962 $2028
.459 .052
≥R .015298 .098836 $8697
.011
None <R <R .001935 .060962 $2028
.856 .541 .929
≥R .015298 .098836 $8697
.071
≥R ≥R .015298 .098836 $8697
.000 .000
Scenario Detector Suppression Extent of Fatalities Injuries Dollar loss

group presence size damage per fire per fire per fire
Figure 5.2 Sprinkler alternative probability tree

or extreme value techniques (Section 3.3.5) if data are only available for large
losses.
The financial loss per m2 may be assumed to be equal to financial value v per
m2 at risk in a building and its contents. Under this assumption, if the floor area
destroyed in a fire is D m2, the financial loss is Dv. In the estimation of the value
density, v per m2, the implicit assumption is that the total financial value V at
risk in the building and contents is spread uniformly over the floor area, A, of the
building such that v = V/A.
Using the value density, v, the power function in Equation (3.41) in terms of
floor area A can be transformed to
D(V) = c´ Vβ (5.34)
where
c´ = c v-β (5.35)
Equation (5.34) expresses the fact that the financial loss, D(V), in a fire
increases approximately according to a power of the total financial value, V, at
risk in the building and its contents. According to statistical studies on actuarial
problems in fire insurance, the value of the power, β, is less than unity for most of
the building types – see Ramachandran (1970, 1979/80, 1988) and Benktander
(1973).
The power function in Equation (3.40) can also be transformed to
F (V) = k´ Vα (5.36)
where
k´ = k v-α (5.37)
V is the total value at risk and v the value density per m2. Equation (5.36)
expresses the fact that the annual probability of frequency of fire occurrence
increases approximately according to a power of the total financial value, V, at
risk in the building and its contents. According to the statistical and actuarial
studies mentioned above and in Section 3.3.1 the value of α is less than unity for
most of the building types.
5.4.4 Life loss
5.4.4.1 Location of fire casualties

According to Table 5.8, based on UK Fire Statistics for the period 1978 to 1991,
most of the casualties in single occupancy dwellings were found in the room of fire
origin. A high percentage of the casualties were also found elsewhere on the floor
Initiation 231
Table 5.8 Location of casualties – single and multiple occupancy dwellings
Number of persons
Whereabouts of casualty and occupancy type Fatal Non-fatal
Single occupancy dwellings
Room of origin of fire 3539 26259
(58.2) (44.6)
Elsewhere on floor of origin 1216 15500
(20.0) (26.3)
Floors above floor of origin of fire 1267 14835
(20.8) (25.2)
Floors below floor of origin of fire 54 2330
(1.0) (3.9)
Total 6076 58924
(100.00) (100.0)
Multiple occupancy dwellings
Room of origin of fire 2347 15353
(66.8) (35.0)
Elsewhere on floor of origin 823 18245
(23.4) (41.6)
Floors above floor of origin of fire 330 9066
(9.4) (20.6)
Floors below floor of origin of fire 16 1233
(0.4) (2.8)
Total 3516 43897
(100.0) (100.0)
The numbers within brackets denote the percentage number of casualties
Source: Fire Statistics United Kingdom, 1978–1991
of origin or floors above the floor of origin. A comparatively smaller number of

casualties were found in floors below the floor of fire origin. Location of casualties
in multiple-occupancy dwellings had a similar pattern except that an almost equal
number of non-fatal casualties were found in the room and floor of fire origin.
It is understandable that occupants in the floors above the floor of fire origin
have greater risk than those in floors below. Fire, smoke and toxic gases generally
spread upwards and are more likely to be encountered by people in upper floors if
they remain in their places of occupation or attempt to escape to safe places in or
outside the building involved in fire. People in lower floors have a greater chance
of avoiding combustible products and escaping safely.
It is apparent that, while fire is a major threat to occupants in its immediate
vicinity, it is generally smoke and toxic gases which pose a greater threat than
flame (heat) to occupants who are remote from the fire. Smoke and fumes travel
faster than fire to occupied areas and escape routes. Even a small fire can generate
considerable amounts of smoke and other combustible products and threaten
a greater number of occupants outside the room of origin. Building fires spread
mostly by convection (advance of flame, smoke and hot gases) rather than by the
destruction of the structural boundaries of a room.
5.4.4.2 Nature of injuries

Consistent with the location of casualties, a high percentage of fatalities in the
room of fire origin were caused by burns, apart from gas or smoke, which was the
major cause, accounting for more than 50 per cent of the fatalities in dwellings –
see Table 5.9. Statistical studies and surveys carried out in the UK in the 1970s
revealed that not only were a large proportion of fatal and non-fatal fire casualties
being reported in the category ‘overcome by smoke and toxic gases’ rather than
heat and burns but also that there was a four-fold increase in the former category
between 1955 and 1971.
It is an accepted fact that toxic products of combustion are the major causes of
incapacitation and death in fires (Berl and Halpin, 1976; Harland and Woolley,
1979). In many fires, death or injury is not due to the immediate toxic effects
of exposure to these products but results from the victim being prevented from
escaping due to irritation and visual obscuration caused by dense smoke or to
incapacitation caused by narcotic gases. Consequently, the victim remains in the
fire and sustains fatal or non-fatal injury due to a high dose of toxic products,
inhaled during the prolonged exposure, or due to burns. High percentages of
fatalities and non-fatal casualties are trapped by smoke or fire, because they are
unaware of the fire (asleep etc.) or for other reasons. Survivors from fires may also
experience pulmonary complications and burn injuries which can lead to delayed
death.
Table 5.9 Fatal casualties in dwellings by whereabouts of casualties and cause of death
Cause of death
Whereabouts of casualty and Overcome by gas Burns or scalds Other or
occupancy type or smoke* unknown causes
Single occupancy
Room of origin of fire 1653 953 328
Floor of origin of fire 731 133 116
Elsewhere 868 130 118
Total 3252 1216 562
Multiple occupancy
Room of origin of fire 1217 510 200
Floor of origin of fire 504 66 67
Elsewhere 217 37 40
Total 1938 613 307
Source: Fire Statistics United Kingdom 1978–1988
*Including cases where burns and overcome by gas or smoke were joint causes of death.
A breakdown of figures for causes of death as in the table has not been published for the years 1989
to 1991.
Initiation 233
Increasing fire risk due to smoke and other combustion products led to the
commencement of intensive research on combustion toxicology during the
1970s. These studies have ranged from fundamental laboratory-based thermal
decomposition experiments to large-scale fires with comprehensive gas analysis,
bioassay and detailed pathology of fire victims. Models developed include the
‘mass loss’ model and the ‘fractional effective dose’ model. These models require
as inputs the rates of generation of life-threatening combustion products and
estimate the times when tenability limits are exceeded, resulting in incapacitation
or death. Purser (2002) carried out a very detailed review of various studies and
models on toxicity assessment of combustion products.
5.4.4.3 Causes of fatal and non-fatal casualties in fires

The majority of fatalities in accidental fires in dwellings are due to causes such as
careless disposal of smokers’ materials, ignition of matches – mostly by children
playing with them, incidents with space heating – mainly misuse or placing articles
too close to them and misuse of cooking appliances. Electricity is the major fuel
in regard to deaths caused by the misuse of space heaters or cooking appliances.
A contributory factor for fire deaths during severe winter is the use of portable
heating appliances to supplement central heating.
The leading causes of non-fatal casualties in accidental dwelling fires are
careless disposal of smokers’ materials and misuse of cooking appliances, mostly
those using electricity and some using gas. The next most common specific causes
are electric and gas space heating appliances, electrical wiring, electric blankets
and bedwarmers. Television sets, washing machines and dishwashers are other
minor causes of non-fatal casualties.
5.4.4.4 Materials first ignited

During the past few years, fires in which textiles, upholstery and furnishings were
the materials first ignited, accounted for a large number of deaths. Within this
group, the major items were bedding, upholstery or covers and clothing. According
to a study carried out in 1976, the chance of a fatality in fire involving furniture
and furnishings was twice that in other fires in houses. The majority of these
fires involved upholstery and bedding and were caused by smokers’ materials,
electric blankets and space heating. The main hazard appeared to be to people in
armchairs and beds, using potential sources of ignition (smoking, space heating
etc.) failing to respond to a fire in their vicinity through being asleep or otherwise
incapacitated.
During the past three decades, smoke-related casualties in homes have
increased partly due to an increase in the use of synthetic materials in furnishings
and upholstered furniture. This is also partly due to changes in living styles which
have led to more furnishings and upholstery material being used in homes.
Change from natural to synthetic materials has brought certain benefits in
fire performance. Natural materials tend to be prone to smouldering from small
ignition sources, particularly when in contact with a lighted cigarette, whereas
synthetic materials tend to be more resistant to this type of ignition. However,
the synthetic fabrics are mainly thermoplastics and when subjected to a flame
can burn rapidly with the fabric ‘melting’ to expose the flammable infill fibres
and foams. Natural fabrics (wool, cotton etc.) tend to form carbonaceous chars
during flame exposure which can act as an effective barrier to the penetration
of fire.
5.4.4.5 Casualty rate per fire

Casualty rate per fire is a simple but very useful yardstick for measuring life risk
due to fires in any type of building. This rate is the ratio between the number of
fatal or non-fatal casualties and the number of fires. Data for estimating these
rates for any type of building are available from Fire Statistics United Kingdom,
published annually by the Home Office.
According to UK fire statistics, both the fatal and non-fatal casualty rates per
fire for dwellings do not vary significantly from year to year. In fact, there is some
indication that the fatality rates are gradually declining over the years. This is
perhaps due to effective performance of fire fighting and protection strategies
including fire safety codes, regulations and standards. This is also due to the
fact that the number of fires, particularly in multiple-occupancy dwellings, has
gradually increased over the years but the number of deaths has not increased.
This result indicates the need for increasing fire prevention activities aimed at
reducing the frequency of fire occurrence.
The non-fatal casualty rate per fire in dwellings has gradually increased over
the years.
The casualty rate per fire indicates the probability of one or more people
becoming a casualty in the event of a fire occurring. It should be multiplied by
the annual frequency (number) of fires occurring to express it on an annual basis.
As discussed in Section 3.3.4, the number of deaths likely to occur in a fire is
a random variable, taking integer values according to a discrete (discontinuous)
probability distribution. Well-known examples of such a distribution are Poisson
and negative binomial. The extended form of Poisson described in Equation (3.49)
provides an estimate of the probability of x deaths occurring in a fire if occupants
of a building are exposed for t minutes to lethal conditions caused by a combustion
product.
The parameter δ in Equation (3.49) quantifies the increase in the probability
of death for every extra minute of exposure to lethal conditions caused by a
combustion product. If, on average, occupants of a building are exposed to such
conditions for t minutes, the probability of x deaths occurring in a fire is given by
Equation (3.49) with t = t. For a particular building, the values of δ and t should be
estimated for each combustion product such as heat or smoke. It may be possible
to estimate these values for a particular building by performing simulations with
models discussed by Purser (2002) or other deterministic models. The results for
Initiation 235
different combustion products may then be combined and used in Equation (3.49)
to provide estimates of the probabilities of death for different values of x.
The parameter λ in Equation (3.46) quantifies the increase in the fatality rate
per fire, P, for every minute of delay in discovering or detecting the existence of
a fire in a building. The value of λ was estimated to be about 0.0007 for single
and multiple-occupancy dwellings. The value of the parameter K denotes the fact
that the fatality rate or probability of one or more deaths occurring would be
about 0.0015 even if a fire in these buildings is detected immediately after ignition,
i.e. if D = 0. Hence, with an average discovery time of 15 minutes, the overall
fatality rate in these buildings was about 0.012 [= 0.0015 + (15 × 0.0007)]. The
fatality rate would have reduced to 0.0022 (= K + λ) if all the single and multiple
occupancy dwellings were fully protected by automatic detection systems.
According to Ramachandran (1993a), the value of λ for non-fatal casualties
was 0.0042 for single-occupancy dwellings and 0.0047 for multiple-occupancy
dwellings. Respectively, for these two types of dwellings, the value of K was 0.0938
and 0.1092. Hence, with average fire discovery times of 9 and 10.6 minutes, the
overall non-fatal casualty rate was 0.1314 for single-occupancy dwellings and
0.1592 for multiple-occupancy dwellings. If all these dwellings were protected
by automatic detection systems, the non-fatal casualty rate per fire would have
reduced to 0.098 for single-occupancy dwellings and 0.1139 for multiple-
occupancy dwellings.
5.4.4.6 Other measurements of life risk

The f(N) relationship for investigating life risks due to various types of hazards
was discussed in Section 4.2.2. This function expresses the annual frequency of
N or more fatalities. There is a need to update the f(N) curves using recent data
on the frequency distribution of number of fire deaths. Such data are available for
several countries.
Fatal Accident Frequency Rate (FAFR) is another measurement of life risk
which is the number of fatalities that occur during a hundred million man-hours
of exposure to an occupation or activity. FAFR has been calculated for various
industrial occupations such as nuclear and chemical industries and non-industrial
activities such as travelling by bus, train, car or air, canoeing and rock climbing. In
regard to fire safety in the chemical industry, for example, FAFR for the industry as
a whole has been estimated to be 4. On this basis, it has been arbitrarily proposed
that no single activity which any person is carrying out should contribute more
than 10 per cent to the FAFR, i.e. 0.4 (Kletz, 1976).
North (1973) calculated rough values of FAFR for many occupancies in the
UK. These only covered the years 1967 to 1969 and, in cases where deaths were
infrequent, the estimates had wide confidence limits. For fire deaths in a dwelling,
the FAFR was 0.19 and in hotels 3.6. The latter figure was probably distorted by
some serious multiple-fatality fires that happened during 1967 to 1969.
An American report (Balanoff, 1976) allowed estimates to be obtained for
FAFR for firemen on or following activities at the fireground. The report indicated
86 deaths per 100,000 firemen per annum which provided an estimate of 42 for
FAFR. About half of the deaths were due to heart failure and 10 per cent due to
each of building collapse, burns and smoke inhalation.
5.4.5 Total performance

In some cases it would be useful to express the probable damage (P) in the
event of a fire occurring as the product of two components Q and C, where Q
is the probability of occurrence of an undesirable event and C is the probable
consequences or damage if the undesirable event occurs:
P=Q×C
The total (joint) performance (effectiveness) and not the individual performance
of a building design and installed fire protection measures should ensure that the
value of P will not exceed a level acceptable to society. The acceptable level would
depend on consequences in terms of life loss and injury, and property damage.
Undesirable events would include:
1 fire spreading beyond room of origin;

2 ‘failure’ of a fire-resistant compartment;
3 visual obscuration due to smoke;
4 incapacitation due to burns and toxic gases.
References
Apostolakis, G (1982), Data analysis in risk assessments, Nuclear Engineering and Design,
71, 375–381.
Balanoff, T (1976), Fire Fighter Mortality Report, International Association of Fire Fighters,
Washington, DC.
Bengtson, S and Ramachandran, G (1994), Fire growth rates in underground facilities.
Proceedings of the Fourth International Symposium on Fire Safety Science, International
Association of Fire Safety Science, Ottawa.
Benktander, G (1973), Claims frequency and risk premium rate as a function of the size of
the riskl ASTIN Bulletin, 7, 119–136.
Berl, W G and Halpin, B M (1976), Fire related fatalities: an analysis of their demography,
physical origins and medical causes, Fire Standards and Safety, ASTM STP 614, American
Society for Testing and Materials, Philadelphia, PA.
Bertrand, A L and McKenzie, L S (1976). The Human Factor in High Risk Urban Areas: A
Pilot Study in New Orleans, Louisiana, US. Department of Commerce, Washington DC.
Butcher, G (1987), The nature of fire size, fire spread and fire growth, Fire Engineers’ Journal,
47, 144, 11–14.
Chandler, S E (1968), Estimated Fire Frequencies In Buildings Based On Expected Fuel Usage,
Fire Research Note 716, Fire Research Station, Borehamwood.
Initiation 237
Chandler, S E (1979), The Incidence of Residential Fires in London: The Effect of Housing and
Other Social Factors, Information Paper IP 20/79. Building Research Establishment, Fire
Chandler, S E (1982), The effects of severe weather conditions on the incidence of fires in
dwellings, Fire Safety Journal, 5, 1, 21–27.
Chandler, S E, Chapman, A and Hollington, S J (1984), Fire incidence, housing and social
conditions – the urban situation in Britain, Fire Prevention, 172, 15–20.
Friedman, R (1978), Quantification of threat from a rapidly growing fire in terms of relative
material properties, Fire and Materials, 2, 1, 27–33.
Frishman, F (1975), On the arithmetic means and variances of products and ratios of
random variables, Statistical Distributions in Scientific Work, Volume 1, Models and
Structures, D Reidel, Dordrecht.
Gaunt, J E and Aitken, I S (1964), Causes of Fires in Dwellings in London, Birmingham and
Manchester and their Relationship with the Climatic Conditions during the First Quarter of
1963, Fire Research Note 538, Fire Research Station, Borehamwood.
Gomberg, A, Buchbinder, B and Offensend, F J (1982), Evaluating Alternative Strategies for
Reducing Residential Fire Loss – The Fire Loss Model, Report NBSIR 82-2551, National
Bureau of Standards Centre for Fire Research, Gaithersburg, MD.
Gunther, P (1975), Fire-cause patterns for different socioeconomic neighborhoods in
Toledo, Ohio, Fire Journal, 3, 52–58.
Harland, W A and Woolley, W D (1979), Fire Fatality Study, University of Glasgow,
Information Paper IP 18/79, Building Research Establishment, Fire Research Station,
Borehamwood.
Kletz, T A (1976), The application of hazard analysis to risks to the public at large, World
Congress of Chemical Engineering, Session A5, Amsterdam, July 1976.
Labes, W G (1966), The Ellis Parkway and Gary Dwelling burns, Fire Technology, 2, 4,
287–297.
Lawson, J R and Quintiere, J (1986), Example illustrating slide rule estimates of fire growth,
Fire Technology, 22, 1, 45–53.
Munson, M J and Oates, W E (1977), Community Characteristics and Incidence of Fire: An
Empirical Analysis, Princeton University Press, Princeton, NJ.
North, M A (1973), The Estimated Fire Risk of Various Occupancies, Fire Research Note 989,
Fire Research Station, Borehamwood.
Purser, D A (2002), Toxicity assessment of combustion products, Chapter 2–6, SFPE
Handbook of Fire Protection Engineering, National Fire Protection Association, Quincy,
MA.
Ramachandran, G (1970), Fire Loss Indexes, Fire Research Note 839, Fire Research Station,
Borehamwood.
Ramachandran, G (1979/80), Statistical methods in risk evaluation, Fire Safety Journal, 2,
125–145.
Ramachandran, G (1980), Economic Value of Automatic Fire Detectors, Information Paper
IP27/80, Building Research Establishment, Fire Research Station, Borehamwood.
Ramachandran, G (1985), The human aspects of fires in buildings – a review of research
in the United Kingdom, Fire Safety Science and Engineering, ASTM STP 882 American
Society for Testing and Materials, Philadelphia, PA.
Ramachandran, G (1986), Exponential model of fire growth, Fire Safety Science: Proceedings
of the First International Symposium. Hemisphere Publishing Corporation, New York.
Ramachandran, G (1988), Probabilistic approach to fire risk evaluation, Fire Technology,
24, 3, 204–226.
Ramachandran, G (1990), Probability based fire safety code, Journal of Fire Protection
Ramachandran, G (1992), Statistically Determined Fire Growth Rates for a Range of Scenarios;
Part 1: An Analysis of Summary Data. Part 2: Effectiveness of Fire Protection Measures:
Probabilistic Evaluation, unpublished report to the Fire Research Station, Borehamwood.
Ramachandran, G (1993a), Early detection of fire and life risk, Fire Engineers Journal, 53,
171, 33–37.
Ramachandran, G (1993b), Fire resistance periods for structural elements – the sprinkler
factor, Part 3, Proceedings of the CIB W14 International Symposium on Fire Safety
Engineering, University of Ulster, Jordanstown.
Ramachandran, G (1995), Heat output and fire area, Proceedings of the International
Conference on Fire Research and Engineering, SFPE, Orlando.
Ramachandran, G (1999), Reliability and Effectiveness of Sprinkler Systems for Life Safety,
unpublished report to the Fire Research Station, Borehamwood.
Thomas, P H (1974), Fires in Model Rooms: CIB Research Programmes, Current Paper
CP32/74, Building Research Establishment, Fire Research Station, Borehamwood.
Walton, W D and Thomas P H (1988), Estimating temperatures in compartment fires,
Section 2, Chapter 2, SFPE Handbook of Fire Protection Engineering, 1st edn. National
Fire Protection Association, Quincy, MA.
6 Design fire size
6.1 Introduction
Fire protection measures cannot be expected to cope with all possible sizes of
fires. The best that can be achieved is to design these measures around a large
or maximum size that is likely to be encountered. Determination of such a size,
defined as the ‘design fire size’, is fundamental to the design of fire detection,
sprinkler and smoke ventilation systems, particularly for buildings where the
fire safety engineering approach is more common, e.g. retail premises, shopping
complexes, atrium buildings. The design fire size is expressed in terms of the
amount of heat output likely to be produced in a large fire. This amount depends
on fire load and other factors affecting the growth of fire in the space considered.
Deterministic formulae and models supported by experimental results are
available for calculating the heat output in a large fire in a compartment of given
dimensions, fire load and ventilation factor. However, due to uncertainties caused
by several factors, and the possibility of several scenarios, probabilities are attached
to quantities of heat output produced in an actual (not experimental) fire over the
period of fire growth. The Technical Report ISO/TR 13387-2: 1999(E) describes a
systematic approach to the identification of significant fire scenarios that need to
be considered in fire safety design. Clark and Smith (2001) established a database
of specific fire characteristics for a series of realistic fire scenarios based on results
of experimental studies.
The chance of heat output in the actual fire exceeding a large value is small.
An acceptable value for this small chance or probability can be determined for
any type of property by considering consequences in terms of damage to life and
property.
Heat output is correlated with area damage (Ramachandran, 1995a). Hence,
the probability of heat output exceeding a specified large value can be ascertained
by estimating the probability for the corresponding large value for area damage.
Statistical data, compiled by the fire brigades in the UK, provide information on
area damage in real fires and its probability distribution. This information has been
used for estimating design fire sizes for shopping complexes, railway properties,
public car parks, road tunnels and power stations. These design sizes and their
estimation are discussed in Section 6.2.
The design sizes mentioned above are based on the assumption of a steady-state
fire and are constants independent of time. They do not take into account the
interactions between smoke and escaping occupants. Hence, such design sizes for
designing fire protection systems have been considered to be unrealistic by some
experts, particularly for protecting life. There is a need to develop probability–
time based design sizes which are more realistic. A framework for this purpose is
outlined in Section 6.3.
6.2 Current concept of design fire size

As mentioned in the previous section, the design fire size currently used for
designing a fire protection system for any type of occupancy is a constant quantity
and not a function of time (duration) since the commencement of ignition. It is
based on a large value for area damage in an actual fire, the chance of exceeding
which is small. Such an extreme (large) value can be considered as average
area damage during the post-flashover period of a fire in a compartment. The
heat output corresponding to this average area damage is estimated and used in
designing a fire protection system. The following is a summary of studies based on
the concept of constant design fire size.
Morgan and Chandler (1981) analysed statistical data for fire damaged areas
in sprinklered retail premises. The results obtained by these authors are shown
in Table 6.1. Since the sample size (number of fires) analysed was small, it was
considered that a zero probability that a fire in a public area would exceed 10m²
was not credible. Accordingly an ‘informed guess’ was made that less than 4 per
cent would exceed 10m² in public areas. The heat output corresponding to 10m²
of area damage was assumed to be 5 MW which was equivalent to 0.5 MW per m².
Based on this study, heat output of 5 MW over a fire area of 10m² was suggested
as an acceptable design fire for designing smoke ventilation systems for covered
shopping complexes. Morgan and Chandler also concluded that, in the absence of
sprinklers, 14 to 17 per cent of fires in public areas of retail premises was larger than
10m² of area damage. In non-public areas, the design size of 10m² was exceeded in
about 12 per cent of fires in sprinklered premises and 19 per cent in others.
Table 6.1 Horizontal fire-damaged area in retail premises, 1978

Number of incidents
Area (m²) Public areas Non–public areas Total
0 or minimal 27 28 55
1–10 25 41 66
11–50 – 8 8
51–100 – – –
Over 100 – 1 1
Total 52 78 130
Design fire size 241
The results of Morgan and Chandler (1981), discussed above, were confirmed to
some extent by Law (1986) who analysed data on ‘area of direct burning’ for public
and storage areas of sprinklered retail premises. The data related to fires which
occurred during the period 1984–87. She also analysed 1981–87 data on number
of sprinkler heads operating in public and storage areas (area damage records were
not available for 1981–83). The results obtained by Law are reproduced in Table
6.2 which indicate that area damage in sprinklered public areas exceed 10m² in
about 7 per cent of fires. More than four sprinkler heads operate in less than 5
per cent of incidents. Accordingly, it appears that the probability of heat output
exceeding 5 MW is less than 5 per cent. More than ten heads operate in less than
2 per cent of fires.
The design principles for smoke ventilation in enclosed shopping centres
outlined in the report by Morgan and Gardner (1990) are based upon a 12m
perimeter (3m × 3m) 5 MW fire. Factors affecting the design fire size are
discussed in this report. Should a different design fire be considered, for whatever
reasons, the equations, figures etc. given in this report may no longer apply and
advice should be sought from experts. Other fire sizes have occasionally been
specified by designers for both sprinklered and unsprinklered shops. The problem
of unsprinklered shops has been discussed in detail by Gardner (1988), who has
shown the importance of considering ‘flashover’ in such units and the consequent
need to consider potentially very large fires.
The procedure of selecting a fixed size of fire for designing smoke ventilation
systems has been adopted for occupancies other than retail premises which are also
commonly associated with atrium buildings such as offices and hotel bedrooms
(Hansell and Morgan, 1994). An atrium can be defined as any space penetrating
more than one storey of a building, where the space is fully or partially covered.
Most atria within shopping centres may be considered as part of the shopping
mall and treated accordingly. For offices, Morgan and Hansell (1985) suggested
a design size of 16m², 14m perimeter if sprinklered and 47m², 24m perimeter if
unsprinklered. For hotel bedrooms, Hansell and Morgan (1985) recommended the
floor area of the largest bedroom as the design size. The corresponding design heat
Table 6.2 Area of direct burning and number of sprinkler heads opening for different
fractiles, sprinklered retail premises
Public areas Storage areas

Area of direct Area of direct
burning Number of burning Number of
Fractile sprinkler heads sprinkler heads
(%) (m²) opening (m²) opening
20 3 2 6 3
10 7 3 12 4
5 16 4 50 4
Note: The fractile is the percentage number of fires in which the area of direct burning or number of
sprinkler heads opening was exceeded.
outputs are approximately 1MW for sprinklered office, 6MW for unsprinklered
office and 1MW for hotel bedroom. The hotel bedroom fire represents a fully
developed unsprinklered fire. Where sprinklers are present, a value of 500KW (for
a 6m perimeter fire) may be more appropriate.
In the above studies, the design size for an occupancy type has been identified
from the frequency distribution (table or graph) of area damage. A better method
would be to fit an appropriate probability distribution to such data. Statistical
studies have shown that area damage in fires in most of the occupancies has either
a log normal or Pareto probability distribution – see Section 3.3.4.
Ramachandran (1992a) found the Pareto distribution to be more appropriate
than log normal in an investigation concerned with design fires in underground
facilities. Design sizes were estimated for four types of occupancies: railway
properties, public car parks, road tunnels and power stations. The results were
based on data for the three years 1985–87 for the first two types of buildings and for
the years 1979 and 1984–87 for the other two types. These statistics were provided
by the fire statistics section of the Home Office, United Kingdom. The analysis is
described below (a summary of the results was presented by Ramachandran and
Bengtson (1995c) at the First European Symposium on Fire Safety Science held
in Zurich in August 1995).
The Pareto distribution has the following form:
φ (d) = k d –λ ; k = mλ (6.1)
where φ (d) is the probability of area damage exceeding d and m the minimum
damage. The parameters k and λ are constants for any occupancy type. They also
depend on factors such as whether a building is sprinklered or not. The parameters
can be estimated by applying the least square method to fit the following straight
line to data:
log φ(d) = log k – λ log d (6.2)
If data on area damage, d, are available in the form of a frequency distribution

table, φ(d) is the proportion of fires (out of the total number) exceeding d.
Approximate values of k and λ can be obtained graphically by plotting the pairs of
values (log φ(d), log d) for different values of d.
Table 6.3 contains the values of λ, k and m estimated for the four occupancies
considered. The results relate only to fires in buildings without sprinklers. They
also relate to area damaged by direct burning (fire area). Fires starting in lifts and
stairs were excluded.
The design size can be estimated by assigning an appropriate value for φ(d). For
example, if a level of 0.1 is acceptable for φ(d), the design size, D, from Equation
(6.1) is given by
0.1= k D–λ (6.3)

Table 6.3 Pareto parameters
Parameters Average
Number damage
Categories of fires λ k m (m2) (m2)
Railway properties
Fires in all places 826 0.8084 1.4038 1.5213 21.71
Fires in public places 220 0.6818 0.6182 0.4939 22.58
Fires in basement 73 0.9371 0.2692 0.2465 5.04
Public car parks
Fires in all places 709 1.2738 1.1575 1.1216 4.67
Fires in basement 165 1.2762 1.4950 1.3704 5.87
Road tunnels and subways 122 1.0094 0.5219 0.5251 3.3
Power stations 133 0.82 0.4118 0.3389 6.04
or by
D = (10k)1/λ
If 0.05 is an acceptable value for φ(d) then
D = (20k) 1/λ (6.4)
Area damage would exceed the value of D given by Equation (6.3) in 10 per
cent of fires and the value given by Equation (6.4) in 5 per cent of fires. The design
sizes based on these two equations are given in Table 6.4. They are applicable to
premises without sprinklers.
Design fire sizes, expressed in terms of area damage, can be converted to heat
output rate by applying the method discussed by Ramachandran (1995a). From
Equations (5.3), (5.18) and (5.19), the loss rate of fuel mass, m (kg/sec) is given
by
m = L ⋅ θ ⋅ A(T ) (6.5)
where L is the average fire load density in kg/m², θ the fire growth parameter per
second in the exponential model in Equation (5.3) and A(T) the fire area (m²)
destroyed in T seconds. The heat output rate, Q (kW), corresponding to m , as
shown in Equation (3.58) is given by
Q = m ∆H (6.6)
where ∆H is the effective heat of combustion of the fuel usually assumed to have
the value 18,800 kilojoules per kilogram.
Table 6.4 Design fire size (D)m²
Occupancy category ϕ(D) = 0.1 ϕ(D) = 0.05 (m²)
(m²) (m²)
Railway properties
Fires in all places 26 62
Fires in public places 14 40
Fires in basement 43 46
Public car parks
Fires in all places 47 12
Fires in basement 48 14
Road tunnels and subways 45 10
Power stations 46 13
Consider now power stations for which the average fire load density, L , is
600MJ/m² (or 33 kg/m²) according to Table A1.3.13 (page 116) of ‘Design Guide:
Structural Fire Safety’ (CIB W14 1986). For this occupancy type, the expected
and maximum values of θ (in minutes) in an individual fire are 0.0208 and 0.0620
(Bengtson and Ramachandran, 1994) – see Table 5.5. Expressed in seconds, the
expected and maximum values of θ are 0.000347 and 0.001033.
According to Table 6.4, the design fire size for power stations is 6m² for φ(d)
= 0.1. In this case, with A(T) = 6m², calculations would show the expected and
maximum values of m are 0.0687 kg/sec and 0.2045 kg/sec. The corresponding
expected and maximum values of Q are 1292 kW and 3845 kW. With the design
size of 13m² for φ(d) = 0.05, the expected and maximum values of m are 0.1489
kg/sec and 0.4432 kg/sec. The corresponding expected and maximum values of Q
are 2799 kW and 8332 kW. Estimates for fire load densities are not available for
the other three occupancy types mentioned in Tables 6.3 and 6.4.
Average fire load densities in shopping centres and department stores vary from
380 MJ/m² for textile items to 585 MJ/m² for items such as foods, furniture, carpets
etc. – see Table A.1.3.5 (page 108) of ‘Design Guide: Structural Fire Safety’ (CIB
W14 1986). Hence, one could assume an average fire load density of 480 MJ/
m² or 26kg/m² for shopping centres. According to Ramachandran (1992b), for
fires in assembly (customer) areas of retail premises extinguished by sprinklers, the
average and maximum values of the fire growth parameter θ in minutes (Equation
(5.3)) are 0.0608 and 0.1043. The corresponding values of θ in seconds (Equation
(6.5)) are 0.001013 and 0.001738. The maximum value would represent the rate
of growth in a fast-growing fire.
For a design fire size of 10m², with A(T) = 10 and L = 26, inserting the
values of θ (in seconds) mentioned in Equation (6.5), the average and maximum
values of m are 0.2634 kg/sec and 0.4519 kg/sec. Hence, from Equation (6.6), the
average and maximum values of heat output rate Q are 4952 kW and 8496 kW.
These results provide support to the design heat output of 5MW currently used
for ventilation systems in shopping centres. This value should be regarded as an
average, since the maximum heat output can be 8.5MW.
According to Equation (6.5), m per m² estimated by m / A(T ) is a constant,
given by L .θ which, in the above example for shopping centres, is 0.02634 kg/sec
for a fire with average fire growth and 0.04519 kg/sec for a fast-growing fire. The
corresponding constant values of Q are 495 kW/m² and 850 kW/m². Realistically,
m and Q per m² are not constants since the fire growth parameter θ is not a
constant during the period of fire development. But θ has been assumed to be a
constant in the simple exponential model in Equations (5.3) and (6.5).
In an investigation concerned with the ‘sprinkler factor’ for reducing fire
resistance requirement for a sprinklered compartment, Ramachandran (1993)
fitted the Pareto distribution (Equation (6.1)) for area damage in fires in three
types of occupancies – office buildings, retail premises and hotels. The results
obtained by him are reproduced in Tables 6.5(a) and 6.5(b). The parameters λ0
and k0 apply to rooms without sprinklers and λs and ks to rooms with sprinklers.
Mo and Ms are the minimum area damage, m, in Equation (6.1). Using the results
in Tables 6.5(a) and 6.5(b), the design sizes, D, have been estimated and given in
Table 6.6 for two values, 0.1 and 0.05, for φ(d) in Equation (6.1). The estimates
are based on Equations (6.3) and (6.4).
Sprinklers, if they operate satisfactorily, would cool the heat and smoke
produced, retard the rate of growth of heat and smoke and, hence, reduce the
probability of area damage exceeding a specified level. Considering sprinklered
office rooms, as an example, using Equation (6.1) with values k = 0.2778 and λ
= 0.6987, probability of damage exceeding 24m² is 0.03 while the corresponding
probability, according to Table 6.6, is 0.1 if the room is without sprinklers. Similarly,
the probability of damage exceeding 68m² is 0.015 if the office room is sprinklered
and 0.05 if the room is unsprinklered. For the reasons mentioned above, as one
would expect, the design sizes for sprinklered rooms are considerably smaller than
the design sizes for unsprinklered rooms.
Judging from the figures in Table 6.6, the design size of 47m² suggested by
Hansell and Morgan (1985) for unsprinklered offices is reasonable, but the design
size of 16m² suggested by these authors for sprinklered offices appears to be an
overestimate. Ferguson (1985) expressed doubt about the choice of a 16m² design
fire for office atrium buildings. It may be argued that it would be safer to provide
a larger safety margin and use a higher estimate for the design fire size for any
occupancy type and room type. A more realistic estimate for the design size for any
type of occupancy and room should be evaluated by considering the consequences,
particularly to life risk, and determining an appropriate level for the probability
quantified by Equation (6.1).
In a letter to the editor of Fire Safety Journal, Law (1986) drew attention to the
data used by Morgan and Hansell (1985). According to Law, the data exhibited
the following ‘exponential’ relationship for large fires in offices of area 10m² or
more:
P = 60A–0.63 for sprinklered offices

Table 6.5(a) Pareto distribution of area damage – rooms without sprinklers
Office buildings Retail premises Hotels
Assembly Assembly Storage and
Parameters Office rooms Other rooms areas Storage areas Other areas areas Bedrooms other areas
Number of fires 1860 4369 8207 5144 7194 518 1205 3821
λ0 0.6686 0.7146 0.6947 0.7304 0.8936 0.6603 0.7734 0.6392
M0 (m²) 0.7749 0.4647 0.5968 1.1583 0.7942 0.5907 0.4543 0.2176
k0 0.8432 0.5783 0.6987 1.1133 0.8139 0.7063 0.5432 0.3772
Table 6.5(b) Pareto distribution of area damage – rooms with sprinklers

Office buildings Retail premises Hotels
Parameters Office rooms Other rooms Assembly Storage areas Other areas Storage and
areas other areas
Number of fires 18 127 224 354 183 35
λs 0.6987 0.8711 0.8644 0.8858 0.6991 0.6310
Ms (m²) 0.1599 0.2646 0.4156 0.2852 0.2142 0.0322
ks 0.2778 0.3141 0.4681 0.3291 0.3405 0.1144
Table 6.6 Design fire size D (m²)
Without sprinklers With sprinklers
ϕ(D) = 0.1 ϕ(D) = 0.05 ϕ(D) = 0.1 ϕ(D) = 0.05
Occupancy
type Room type (m²) (m²) (m²) (m²)
Office 24 68 4 12
Office rooms
buildings
Other 12 31 4 8
rooms
Assembly areas 16 45 6 13
Storage areas 27 70 4 8
Retail
premises Other 10 23 6 16
areas
Assembly areas 19 55 – –
Bedrooms 9 22 – –
Hotels
Storage and 8 24 1.24 3.71
other areas
P = 180A–0.78 for non sprinklered offices (6.7)
P is the proportion (%) of fires exceeding a given area A (m²). With A denoted
by d and P by φ(d), Equation (6.7) is essentially the same as the Pareto distribution
shown in Equation (6.1). Equation (6.1) is applicable to all fires, small and large,
while Equation (6.7) is only applicable to large fires.
According to the figures in Tables 6.5(a) and 6.5(b), the values of the
parameter λ for office rooms are 0.67 if unsprinklered and 0.70 if sprinklered.
The corresponding figures of Law are 0.78 and 0.63, respectively, which appear
to be unrealistic. From theoretical considerations one would expect the value of
λ for a sprinklered room to be higher than that for an unsprinklered room. This
hypothesis is supported by the figures in Tables 6.5(a) and 6.5(b) except those for
‘other rooms’ of retail premises. For storage and other areas of hotels, the estimate
of λ for a sprinklered case is almost the same as the estimate for the unsprinklered
case.
6.3 Probability–time based design fire size
6.3.1 Introduction
The loss rate of fuel mass m , and rate of heat output, Q , are not constants but
are functions of the time elapsing since the start of ignition. These two rates would
be generally small quantities during the initial stage of a fire involving the object
or item first ignited. As the fire spreads to other combustible items in the room,
there will be a progressive increase in the fire area and heat output. That is to say,
the heat output of unit area of the fire will start small and it will increase with time
until a final value such as 0.5MW/m² is reached. After this, the heat output will
decay as the fuel on that particular unit area is consumed.
For the reasons mentioned above, Butcher (1987) pointed out that a fire does
not reach its fully developed size but grows and only reaches, say, the 0.5MW/m²
value after an interval of time has elapsed. The 5MW value of heat output over
10m² of fire area is reached only after an appreciable delay. Hence, the concept of
using a constant design size of 5MW heat output for a fire situation is somewhat
unrealistic. Assigning the full 5MW per 10m² or 0.5MW/m² to the early stages of
a fire, where occupants are attempting to escape, can cause a serious error. During
this period, the heat of the fire will be low; hence the smoke temperature will be
low and the buoyancy movement sluggish. Calculating the smoke temperature
from the 5MW value for heat output could give a false picture of smoke movement
for the early period of a fire. But, as argued by Holt (1987), safety margins may be
badly eroded by adopting smaller design sizes.
Several factors can cause uncertainties (randomness) in the determination
of a design size for any type of occupancy or room. The presence or absence of
sprinklers is also a source of uncertainty. Factors influencing the design fire were
discussed by Morgan and Gardner (1990). The uncertainties can be quantified
by probabilities and statistical parameters such as mean and standard deviation.
For the reasons mentioned above, there is a need to develop a probability–
time based model to determine more realistic and accurate design fire sizes. A
framework for this purpose is outlined in the next subsection. The framework
takes into consideration the interaction between movement of heat/smoke
and movement of escaping occupants. In the following three subsections, the
framework is applied to determine design fire sizes for detectors, sprinklers and
ventilation systems.
6.3.2 Framework
The first step is to develop a fire growth curve which will realistically express the
relationship between heat output and time. For this purpose, it is not unreasonable
to assume that heat output does not increase significantly before the occurrence of
established burning. Heat output would grow ‘steadily’ after the commencement
of established burning. Such a steady growth is depicted by an exponential model,
according to which heat output increases exponentially with time – see Section
5.2.3.
Heat output is positively correlated with fire area (Ramachandran, 1995a).
Heat output is directly proportional to fire area or area damage by direct burning.
Hence, as discussed in Section 5.2.6 (Equations (5.18) and (5.19)), the exponential
model in Equation (5.3) can be used to estimate the loss rate of fuel mass as a
function of time and then Equation (3.58) is used to obtain a similar relationship
for the rate of heat output, Q :
Q = L ⋅ θ∆H ⋅ A( 0 )exp(θT ) (6.8)

The parameters in Equation (6.8) have already been defined with reference to
Equations (5.3), (6.5) and (6.6).
The next step is to estimate the design value for time, T, required for the
operation of a fire protection system such as detectors, sprinklers or smoke
ventilation system. This design time should be sufficient for the occupants of a
building to escape to a safe place, without being killed or injured, before the escape
routes are blocked by heat, smoke or toxic gases. For such a successful evacuation
of any type of building, the time of operation, according to the current design heat
output, may or may not be sufficiently safe.
Consider, for example, the design value of 5MW (5000kw) for Q currently
used for smoke ventilation systems in shopping centres. As discussed in Section
6.2, for sprinklered customer areas of retail premises, the maximum value of the
fire growth rate θ estimated by Ramachandran (1992b) is 0.001738 per second.
In this estimation, one square metre was used for A(0), the area ignited at the
commencement of established burning. Q = 5000 kW, L = 26 kg/m², ∆H =
18,800 kJ/kg and θ = 0.001738, calculation based on Equation (6.8) would show
that the minimum value of T applicable to a fast growing fire is 1020 seconds or 17
minutes. It may also be calculated that, with θ = 0.001013 per second, T = 2283
seconds or 38 minutes for a fire growing at an average rate. For two scenarios,
Butcher (1987) estimated the time taken by a fire to reach a heat output of 5MW
as 16 and 22 minutes.
The values of T estimated above were based on statistically determined fire
growth rates in real fires involving several objects, data for which were provided
by fire brigades. They are not based on a deterministic model of fire growth applied
to individual objects with parameter values estimated from experimental data. A
question now arises as to whether time periods ranging from 16 to 38 minutes for
the operation of a smoke ventilation system in a real fire (not experimental fire)
provide an adequate safety margin for the successful evacuation of occupants.
Should the system operate earlier at a lower heat output?
To answer the above question, it is necessary to consider the interaction
between the movement of a combustion product and the movement of occupants
– see Ramachandran (1993, 1995b) and Section 12.2. For safe evacuation, it is a
fundamental condition that the total time, H, taken by an occupant or group of
occupants to reach a safe place in the building, or outside the building, should be
less than the time, F, taken by a combustion product, e.g. smoke, to travel from
the place of fire origin and produce an untenable condition on the escape route
selected. Some occupants may sustain fatal or non-fatal injuries if egress failure
occurs with H > F.
As discussed in Section 3.7.3.2, the total evacuation time, H, is the sum of three
component periods. The first period, D, is the time taken to discover the existence
of a fire by an occupant (Section 8.1.1) or by an automatic detection system
(Section 8.1.2). The second period, B, is the ‘recognition time’ or ‘gathering phase’
discussed in Section 9.1. The third period, E, is the time taken by an occupant to
travel from their place of work and reach a safe place, such as the entrance to a
protected staircase or exit to the outside of the building – see Section 9.4. Thus
H = D + B + E. The time, F, can relate to a particular combustion product such
as heat, smoke or a toxic gas.
As discussed in Section 3.7.3.3, it would be necessary to estimate and use in
a probabilistic evacuation model, the means (averages) and standard deviations
of H and F over several fire and evacuation scenarios likely to occur in an actual
fire. Then ‘partial safety factors’ can be estimated to derive the design value for H
according to Equation (3.89). A similar calculation procedure would provide the
design value for F (Equation (3.90)). The values of the partial safety factors for H
and F should be such that the design criterion H < F for successful evacuation is
satisfied by the design value of H (Equation (3.91)).
The above model, based on partial safety factors, is a semi-probabilistic
approach. A better probabilistic model for estimating the design value of H is
provided by the ‘Beta method’, discussed in detail in Section 3.7.4. This method
involves the probability distributions of H and F and their joint distribution in
addition to their means and standard deviations. In this method, the design value
of H is evaluated according to Equation (3.93) to meet a (small) target probability
specified for egress failure. This target probability should take into consideration
consequences in terms of fatal and non-fatal injuries likely to be sustained by the
occupants attempting to escape from a fire.
Subject to the limit specified for the design value of the total evacuation
time, H, appropriate design values can be estimated for the operation times of
automatic fire detection systems (Section 6.3.3), sprinkler systems (Section 6.3.4)
and ventilation systems (Section 6.3.5). The design value for sprinkler operation
time should also take into consideration the fact that sprinklers have the potential
to increase the value of F even if they do not extinguish a fire. Sprinklers would
reduce the rate of growth of heat and smoke and thus delay the occurrence of an
untenable condition on an escape route. The value of F can be infinity or very
high if sprinklers extinguish a fire.
The total fire area, A(T), likely to be damaged when a fire protection system
operates according to the design time, Td, can be estimated with the aid of
Equation (5.3). The corresponding value of m will be given by Equation (6.5)
and Q by Equation (6.6). All the parameters A(T), m and Q depend on the
fire growth rate θ which, in turn, depends on the occupancy type and whether
the growth is slow, medium or fast. The probability attached to A(T) in a real fire
can be estimated if data are available for evaluating the probability distribution of
area damage (log normal, Pareto etc.) in the presence of the fire protection system
considered.
6.3.3 Detectors
Automatic detection systems are designed to detect heat and/or smoke from a
fire in its early stages of growth, give an audible signal and call the fire brigade
if directly connected to the brigade. Such a signal would enable first-aid fire
fighting to commence early so that the fire could be controlled quickly and
prevented from causing extensive damage. Unlike sprinklers, which both detect
fires and actively participate in fire fighting, detectors are passive and play no
role in fire control.
Although it is possible to calculate from test results the response time of a
heat/smoke detector under known conditions of ceiling height, detector spacing
and fire/smoke intensity (total heat/smoke release rate), the time of operation of
a detector head in an actual fire depends on many factors. The time when a fire
product, heat, smoke or radiation, reaches a detector head depends on the rate of
spread of the product which is controlled by the room/building configuration and
environmental conditions. The factors mentioned above cause uncertainties in
the performance of a detector, which may or may not operate in an actual fire; if it
operates, it may do so at a random time. Detectors would fail to operate if the heat
or smoke generated is insufficient to activate the system.
According to Bengtson and Laufke (1979/80), operating times for heat detectors
range from 2 minutes in ‘extra high hazard’ (XHH) occupancies such as plastic
goods factory, to about 20 minutes for ‘light hazard’ (XLH) which includes flats
and other residential premises. The operating times of smoke detectors range from
0.5 min (XHH) to 2.25 min (XLH) for wood materials and to 0.75 min (XLH)
for polystyrene. For wood materials, glowing fires give out most smoke but for
polystyrene, apparently flaming conditions produce sufficient smoke for a quicker
response time. According to some tests relating to dwelling fires quoted by Custer
and Bright (1974), detection times for smouldering upholstery fires are long, for
both rate-of-rise and fixed temperature detectors. In another test, involving a
rapidly developing fire in a trash barrel, the rate-of-rise detector operated at 2
minutes while the fixed temperature unit responded at 5 minutes and the photo-
electric detector in 8 minutes.
Nash et al. (1971) carried out some tests involving high stacked storage, using
various types of detectors. In a series of similar tests, heat detectors operated
between 1 min 16 sec and 3 min 58 sec. On ignition, ionisation chamber detectors
operated between 1 min 5 sec and 4 min 30 sec while optical detectors took over
3 minutes to operate. Infra-red detectors operated in about 3 minutes and laser
beam detectors took about 5 minutes to operate, if well above the fire.
The studies mentioned above provide some indication of the operating times
of detection systems currently used, according to designs recommended in fire
safety codes and standards. For example, the normal operating temperature of
a heat detector head would be 65°C. The first step is to investigate whether the
operating time of an existing detector system, currently designed for an occupancy
type, would be consistent with the design value for the total evacuation time, H,
discussed in Section 6.3.2. This is to ensure the safe evacuation of the occupants.
If the operating time of an existing detector system added to the other two time
periods, B and E, would increase the value of H beyond its design value, this
time should be reduced to a new design time for the operation of the detector
system. This design time would provide, for example, an estimate of the operating
temperature of a heat detector head, which should be used in the design of the
detector head.
Practically, with the existing technology, a detector system for any part of a
building, e.g. kitchen, office room, etc. can be designed according to any level of
physical parameters such as temperature and heat output from a fire. But care
should be taken to ensure that the detector does not increase the frequency or
probability of occurrence of false or unwanted alarms. For example, if the sensitivity
of a heat detector head is increased by lowering the operating temperature, the
head may pick up signals given by spurious fires from sources such as cigarette
smoking and cooking which are normal activities. On the other hand, if the
sensitivity is decreased by increasing the operating temperature, this will increase
the risk of genuine fires being undetected. Sufficient research has not yet been
carried out to determine an operating temperature which will provide an optimum
balance between the detection of genuine fires and non-detection or blocking of
false alarms.
6.3.4 Sprinklers
Sprinklers are generally required to operate at an average temperature of 68°C
but there are special requirements for certain occupancies and important aspects
such as the flow of hot gases in fires, which can determine the siting of sprinkler
heads to achieve acceptable operating times. As with detection systems, several
factors cause uncertainties in the activation and operating times of sprinklers in
actual fires, although scientific (deterministic) methods have been developed for
estimating the response time. Based on factors such as rate of temperature rise,
height of upper fire surface above the floor and height of the premises, Bengtson
and Laufke (1979/80) estimated sprinkler operation times varying from 2.5
minutes for ‘extra high hazard’ occupancies to 16.8 minutes for ‘light hazard’. The
operation time of sprinklers in experimental fires have been estimated in several
studies carried out particularly by the Fire Research Station, UK, Factory Mutual
Research Corporation, USA and National Institute of Standards and Technology,
USA. In some fires, the heat produced may not be sufficient to activate a sprinkler
system.
As discussed in Section 6.3.3 with reference to detectors, it should be
investigated first whether the operating time of sprinklers currently used is such
that it will lead to a total evacuation time less than the design value of H. If
this condition is not satisfied, the operating time of a sprinkler system should be
reduced from its current level. However, the design value of H can be increased
for a building protected by sprinklers since, as mentioned before, sprinklers have
the potential to increase the time, F, taken by a combustion product to produce
an untenable condition on an escape route. At the same time, it is necessary
to ensure that with H < F, the safety margin provided by the difference (H –
F) is such that it would meet the target probability specified for egress failure.
For some buildings, e.g. residential buildings, it may be necessary to install fast-
response sprinkler systems which operate quicker than sprinkler systems currently
used in industrial buildings. The design value for sprinklers should also take into
consideration whether or not a detection system has been installed in the building.
6.3.5 Ventilation systems
Deterministic models and experimental data may be able to provide an estimate
of the time of operation of a ventilation system designed for smoke from a fire with
a heat output of, say, 5MW. But in a real fire, as pointed out in Section 6.3.2, the
operation time of a 5MW ventilation system installed in a shopping centre can
vary from 16 to 38 minutes, depending on several factors. It may be necessary to
design the system according to a lower heat output operating much earlier.
It is arguable whether a vent should operate before the operation of a sprinkler,
if installed, or after the operation of the sprinkler. There are indications from
current research that the effects of venting on the opening of the first sprinklers
and their capacity to control the fire are likely to be small. There are also
indications that the earlier vents are opened, the more likely they would be
effective in preventing smoke-logging in a sprinklered building. In the initial
stage of fire growth, a vent should, perhaps, operate before a sprinkler if life
safety is the dominant objective, e.g. in hotels, shopping centres, office buildings.
In industrial buildings, the first sprinkler may operate before the opening of any
vent. The operation times of vents and sprinklers can be appropriately adjusted
subject to the condition that the total evacuation time, H, does not exceed its
design value.
References
British Standard BS 7974 (2002), The Application of Fire Safety Engineering Principles to the
Design of Buildings, British Standards Institute, London.
Bengtson, S and Ramachandran, G (1994), Fire growth rates in underground facilities,
Proceedings of the Fourth International Symposium on Fire Safety Science, International
Association of Fire Safety Science, Ottawa.
Butcher, G (1987), The nature of fire size, fire spread and fire growth, Fire Engineers Journal,
47, 144, 11–14.
Clark P and Smith D A (2001), Characterisation of fires for design purposes: a database
for fire safety engineers, Proceedings of the Interflam 2001 Conference, Interscience
Communications, London.
Custer, R L P and Bright, R G (1974), Fire Detection: The State-of-the-Art. NBS Technical
Note 839. National Bureau of Standards, Washington DC.
Ferguson, A (1985), Fire and the atrium, Architect’s Journal, 181, 7, 63–70.
Gardner, J P (1988), Unsprinklered shopping centres: design fire sizes for smoke ventilation.
Fire Surveyor, 17, 6, 41–47.
Hansell, G O and Morgan, H P (1985), Fire sizes in hotel bedrooms – implications for
smoke control design, Fire Safety Journal, 8, 3, 177–186.
Hansell, G O and Morgan, H P (1994), Design Approaches for Smoke Control in Atrium
Buildings, Report BR258, Building Research Establishment, Fire Research Station,
Borehamwood.
Holt, J E (1987), Fire growth and design size (Letter to the Editor), Fire Engineers Journal,
September 1987, 26.
International Standards Organisation (1999), Fire Safety Engineering – Part 2: Design Fire
Scenarios and Design Fires, Technical Report ISO/TR 13387-2 Fire Edition, 1999-10-15,
ISO, Geneva .
Law, M (1986), Letter to the Editor, Fire Safety Journal, 10, 67–68.
Law, M (1995), The origins of the 5MW design fire, Fire Safety Engineering, April, 17–20.
Morgan, H P and Chandler, S E (1981), Fire sizes and sprinkler effectiveness in shopping
complexes and retails premises, Fire Surveyor, 10, 5, 23–28.
Morgan, H P and Hansell, G O (1985), Fire sizes and sprinkler effectiveness in offices –
implications for smoke control design, Fire Safety Journal, 8, 3, 187–198.
Morgan, H P and Gardner, J P (1990), Design Principles for Smoke Ventilation in Enclosed
Shopping Centres, Report BR186, Building Research Establishment, Fire Research
Station, Borehamwood.
Nash, P, Bridge, N W and Young, R A (1971), Some Experimental Studies of the Control of
Developed Fires in High-racked Storages by a Sprinkler System, Fire Research Note 866,
Fire Research, Borehamwood.
Ramachandran, G (1992a), Fires in Certain Types of Buildings – Growth Rates and Design
Sizes Report submitted to Brandskyddslaget, Enskede, Sweden.
Ramachandran, G (1992b). Statistically Determined Fire Growth Rates for a Range of
Scenarios: Part 1: An Analysis of Summary Data. Part 2: Effectiveness of Fire Protection
Measures. Probabilistic Evaluation Report to the Fire Research Station, Borehamwood.
Ramachandran, G (1993), Fire resistance periods for structural elements – the sprinkler
factor, Proceedings of the CIB W14 International Symposium on Fire Safety Engineering,
University of Ulster, Jordanstown
Ramachandran, G (1995a), Heat output and fire area, Proceedings of the International
Conference on Fire Research and Engineering, SPFE, Orlando, FL.
Ramachandran, G (1995b). Probability-based building design for fire safety. Part 1: Fire
Technology, 31, 3, 265–275; Part 2: Fire Technology, 31, 4, 355–368.
Ramachandran, G and Bengtson, S (1995c), Design fires in underground facilities, Book of
Abstracts, First European Symposium on Fire Safety Science, ETH, Zurich, August.
7 Fire spread beyond room of origin
7.1 Probability of fire spread beyond room of origin

A fire may spread beyond the room of origin if and when a structural member of
the room – wall, floor or ceiling – reaches a limit state (condition) and experiences
‘thermal failure’ or ‘collapse’ in the extreme case. These two modes of structural
failure can normally occur if and when ‘flashover’ occurs and the fire produces
intense heat during the post-flashover stage. Due to uncertainties caused by
several factors, a probability is attached to the occurrence of flashover in a real
fire starting in a room. Evaluation of this probability has been discussed in detail
in Section 5.3. A probability is also attached to the thermal failure of a structural
member of a room if flashover occurs. Evaluation of this probability is discussed
in Section 7.2.
In the event of a fire, collapse of the structural barriers of the room of origin
rarely occurs according to fire statistics. Fires spread beyond the room of origin to
an adjacent room or space mostly by convection (advance of flames and hot gases)
by penetrating through a structural element, which may ‘fail’ thermally. A fire
may also spread beyond the room of origin through a door or window left open or
through some other opening. Such cases can also be considered as thermal failure
of a structural element since the fire resistance of the element will reduce to zero
if a door or window, forming part of the element, is left open.
As discussed above, the probability, Q, of a fire spreading beyond the room
of origin due to the thermal failure of a structural element is the product of the
following two components (see Ramachandran 1995):
A – probability of flashover
B – conditional probability of structural failure given flashover
Q = A × B (7.1)
As discussed in Section 5.3, statistics on real fires attended by fire brigades

provide estimates for A, probability of flashover – see Table 5.6. This probability,
as described in Section 5.3.3, is the product of the parameters λ1 and λ2 of a
stochastic state transition model (STM) discussed in Section 3.4.3 – see also
Figures 3.19 and 5.1. These statistics also provide estimates for Q – see Table 7.1.
Q is the product of the three parameters λ1, λ2 and λ3 of an STM.
According to Equation (7.1), an estimate of B, the conditional probability of
structural failure given flashover, is given by
B = Q/A (7.2)
Table 7.1 Probability of fire spreading beyond room of origin

Occupancy type
Area of fire origin Sprinklered building Unsprinklered building

Textile industry
Production 0.04 0.12
Storage 0.04 0.25
Other areas 0.04 0.15
Chemical etc industry
Storage 0.08 0.20
Paper etc industry
Storage 0.05 0.21
Timber etc industry
Storage 0.04 0.28
Retail trade
Assembly 0.02 0.13
Storage 0.02 0.22
Wholesale trade
All areas 0.03 0.23
Office premises
Rooms used as offices – 0.20
All areas 0.03 0.15
Hotels
All areas Less than 1% 0.14
Pubs, clubs, restaurants
All areas 0.07 0.17
Hospitals
All areas 0.06 0.04
Flats
All areas 0.01 0.07
Fire spread beyond room of origin 257
The value of the probability B given by Equation (7.2) is based on fire statistics.
This value, for the following two reasons, does not provide a satisfactory indication
of the probability of thermal failure and of the fire resistance of the structural
barriers of the room. A ‘room’ as recorded in the fire brigade reports on fires is not
necessarily a fire compartment. The figures for number of fires that spread beyond
the room of origin, estimated from fire brigade reports, include a small number
that spread by destruction of barrier elements, although most of the fires would
have spread by convection (advance of flames and hot gases) by penetration
through the barrier elements.
Equation (7.1) provides a method for determining an acceptable level for B,
the probability of failure of a structural element or compartment by specifying
an acceptable level for Q, the probability of fire spreading beyond room
(Ramachandran, 1995). An acceptable value for Q can be determined, with due
regard to likely consequences in terms of damage to life and property, if a fire
spreads beyond the room of origin. The above method would then provide an
estimate of the fire resistance required for the structural elements to meet the
acceptable level determined for the probability B. Subject to the value specified for
Q, the fire resistance for a sprinklered compartment can be reduced, depending on
the extent to which sprinklers are likely to reduce A, the probability of flashover.
This would increase the value of B but the value of the product Q would be
maintained at the same specific level.
Materials in a room involved in a fire can be expected to have an influence on
the rate of fire growth and the probability of fire spreading beyond the room. The
occurrence of established burning and, to some extent, the occurrence of flashover
would depend very much on the material first ignited. This material is specified in
the reports on fires attended by the fire brigades in the UK. Table 7.2, based on a
past study (unpublished), is an example based on these data for the chemical and
allied industries. Most of these fires occurred in unsprinklered buildings but some
fires could have occurred in sprinklered buildings. The study did not consider
separately sprinklered and unsprinklered buildings. For the industry considered,
the overall probability of fire spreading beyond the room of origin is 0.07. Among
known materials first ignited, the leading materials with probabilities exceeding the
overall probability are, in order, paper (0.16), unspecified waste (0.15), packaging
(0.13) and structure (0.10).
Other factors affecting the probability of fire spread beyond the room of origin
are time of occurrence of the fire, i.e. day or night; age of the building, i.e. modern
or old; and whether a building is single storey or multi-storey. To estimate the
influence of these factors, Baldwin and Fardell (1970) applied the logit model
(Equation (3.48)) discussed in Section 3.3.3. According to this study, fire brigade
attendance time had no influence on fire spread but this factor appears to have
some effect, as discussed in Section 10.1.
Table 7.2 Material ignited first – probability of fire spreading beyond room of origin
(chemical and allied industries)
Number of fires Probability of fire
Total number of spreading beyond spread beyond room
Material ignited first fires room of origin of origin
Gases 86 4 0.05
Liquids 436 29 0.07
Textiles 76 3 0.04
Furniture 5 – –
Structure 30 3 0.10
Fittings 30 – –
Food 13 – –
Paper 106 17 0.16
Packaging 61 8 0.13
Lagging 107 1 0.01
Dust, powder, flour etc 96 4 0.04
Electrical insulation 279 15 0.05
Unspecified waste 13 2 0.15
Other 664 33 0.05
Unknown 82 29 0.35
Total 2084 148 0.07
7.2 Performance and reliability of compartmentation
7.2.1 Probabilistic approach

Fire resistant compartmentation has long been the core of fire safety measures
specified in building regulations, codes and standards. A building is regarded as
being composed of compartments, perfectly isolated from one another, and the
spread of fire as taking place by successive destruction (or possibly thermal failure)
of the compartment boundaries – walls, floor and ceiling.
If the boundaries of a compartment are of sufficient fire resistance, it is argued,
the compartment will not ‘fail’ for a specified length of time, defined as fire
resistant period, R. This performance will reduce the probability of fire spreading
beyond the compartment within the period R and thus provide sufficient time for
the response and control of the fire by the fire brigade.
Compartment failure can occur if and when a fire grows to a ‘flashover’ stage.
During the post-flashover stage, heat or severity, S, measured in time units
(minutes) can attain a high level and cause a progressive deterioration of the
structural boundaries (walls, floor and ceiling) which might reach ‘limiting states’
and violate performance criteria relating to any of the following three aspects. The
first aspect, stability, is one of the characteristics of the load-bearing capacity of a
structural element. Strength and ductility are the other two characteristics. The
second aspect, integrity, is concerned with the strength of a structural element to
prevent the penetration (spread) of fire from the fire compartment to an adjacent
compartment or space. The third aspect is the thermal insulation provided for a
structural element to prevent the spread of a fire/heat from the fire compartment
to an adjacent compartment or space.
To delay the time taken by a structural element to reach limiting states, the
element is provided with a period of fire resistance, R (minutes), greater than a
high level of severity, S. Calculation of such a design value for R can be based
on partial safety factors (Section 3.7.2) which is essentially a non-probabilistic
method.
A semi-probabilistic method, currently followed by fire safety engineers for
estimating the maximum severity, is based on an analytical (deterministic) model
such as the ‘equivalent time of fire exposure’, Te, discussed in Section 3.7.3.2 –
see Equation (3.78). Te is estimated by a function of density, per m2, q, of fire
load (weight of combustible materials), thermal inertia, c, of the compartment
boundaries and ventilation factor, w, estimated through Equation (3.79).
Maximum severity is estimated by inserting for q in Equation (3.78), the value of
fire load density corresponding to a high fractile value such as 80 per cent or 90
per cent of the fire load density distribution.
Consider, as an example, an office room of brick construction without sprinklers.
The room has a floor area (Af) of 21.96m2 (= 6.1 × 3.6) and height 3m. The total
area of all the bounding surfaces including the ventilation area is
AT = 2(18.3 + 10.8 + 21.96) = 102.12m2
With a height (h) of 1.5m and width 1.2m, the ventilation area (Av) is 1.8m2.
The ventilation factor w in Equation (3.79) is
21.96
w= = 1.4636
(102.12×1.8×1.225)1/2
According to the CIB Design Guide (CIB W14 1986), the value of c for the
office room considered can be taken as 0.07. Also, according to this publication,
for office rooms, the fire load density corresponding to 80 per cent fractile of fire
load density distribution is 570 MJ/m2. Using the above values of the parameters in
Equation (3.78), maximum severity may be estimated to be 58.4 minutes. Hence,
the structural members of the office room considered should be provided with a
fire resistance exceeding 59 minutes. The probability of compartment failure in
this case is assumed to be 0.2, i.e. 20 per cent. It may be safer to consider a higher
fractile of 90 per cent with a fire load density of 740MJ/m2, such that the fire
resistance exceeds the maximum severity of 75.8 or 76 minutes. The probability of
compartment failure in this case may be assumed to be 0.1.
The method discussed above, used by some fire safety engineers, is simple but
somewhat subjective since it involves an arbitrary selection of the fractile value
and assumption about the probability of compartment failure. The probabilistic
safety margin provided by this method is really an unknown quantity.
A better approach, used by some fire safety engineers, is a probabilistic method
based on the probability distribution of fire severity, S, considered as a random
variable in a real fire occurring in a compartment. Fire resistance, R, is regarded
as a constant (not variable) in this method. This ‘univariate’ approach has been
described in Section 3.7.4.2, using exponential and normal distributions.
Sufficient data may not be available to estimate the probability distribution
of S. In such cases, the method based on Equation (3.78) can be used for the
estimation of the mean, µs, and standard deviation, σs – see Ramachandran
(1998a). For the example (office room) considered earlier, the mean value and
standard deviation of fire load density, q, are respectively 420MJ/m2 and 370MJ/
m2. With these values, calculations based on Equation (3.78) show that the values
of µs and σs are respectively 43 minutes and 38 minutes. Normal or exponential
distribution may be assumed.
A much better method is the ‘bivariate’ method regarding both fire severity, S,
and fire resistance, R, as random variables in a real fire occurring in a compartment.
This method, more commonly known as the Beta method, has been discussed
in detail in Section 3.7.3.4. Under this method, the fire resistance required for
a structural element or compartment is set equal to or greater than the value
given by Equation (3.108) to meet a target failure probability specified through
the safety index β.
The probabilistic methods discussed above recognise the fact that, in a real fire,
S and R are random variables affected by uncertainties caused by several factors,
some of which cannot be controlled. The performance of a structural element
or compartment in a real fire would be different from its performance in a fire
resistance test or a large-scale compartment test carried out under known and
controlled conditions. Due to end and rotational restraints, often present in a
building, the fire resistance attained in a real fire can be significantly different from
the resistance achieved in a fire resistance test.
Fire resistance is also affected by weakness caused by penetrations, doors,
windows or other openings in the structural barriers of a compartment. Pipes,
cables, etc for central heating, television, telephone and other services generally
pass through holes in walls, ceilings or floors. The openings around such holes
should be well sealed, otherwise fire and smoke can spread through the openings,
thus reducing the fire resistance of the structural element.
The fire resistance of a door, even if it is rated according to a test, is generally
less than that of the wall on which the door is located. Consequently, the fire
resistance of the wall will be less than the fire resistance for which it has been
designed and tested. The fire resistance of the wall will be practically zero if the
door is left open at the time of a fire; heat, smoke and toxic gases spread quickly
through open doors. Doors, particularly in office and industrial buildings and
department stores, are likely to be kept open for facilitating passage of people
and goods and during warm weather conditions. Several mechanical devices are
available for closing a door automatically and positively when a fire occurs. Such
a device can be coupled with an automatic smoke detector system with a control
unit ( Langdon-Thomas and Ramachandran, 1970).
Apart from uncertainties governing the development of a real fire, there are also
uncertainties associated with the values used for the parameters of a deterministic
model. For example, uncertainties quantified by mean, standard deviation and
probability distribution are associated with the parameters c and w in Equation
(3.78) in addition to the uncertainties associated with the fire load density, q. The
values of these parameters depend on the compartment size. It is doubtful whether
Equation (3.78) would be applicable to a large compartment.
Uncertainties are also associated with the parameters of a model used for
estimating the fire resistance, R, of a structural member. For example, according to
an analysis of fire tests of thin wall steel members (Homer, 1979),
0.8
 f ⋅ h ⋅ m
R= 
 g 
where h = thickness of insulation, m = total mass of insulation and steel structural

member, g = average perimeter of protective material and f = a factor representing
the insulation heat transmittance value for the material. Uncertainties are
associated with the values used for f, h, m and g and the power (exponent) 0.8.
Parameters not included in the model can cause some uncertainties.
Ramachandran (1998a) reviewed in detail the basic features of the
probabilitistic methods discussed in this section and Sections 3.7.3 and 3.7.4 for
determining the fire resistance required for a structural element or compartment.
This simplified guide also contains a brief discussion about the full probabilistic
approach, determination of the ‘design point’ based on first order (second moment)
reliability theory and the application of extreme value theory. Further applications
of extreme value theory and the reliability approach (next section) were discussed
in a recent paper (Ramachandran 2003).
7.2.2 Reliability approach

For any target level specified for the probability of failure, the probabilistic methods
discussed in the previous section can be applied to determine the fire resistance
required for any of the six structural elements (four walls, floor and ceiling) or for
the compartment as a whole. Two questions arise. Firstly, with a change in the
notation, if the structural elements have fire resistance periods (minutes) of T1, T2,
…, T6, what would be the fire resistance period of the compartment as a whole? In
order to achieve a fire resistance period of T minutes for the compartment, what
should be the fire resistance period of each of the six structural elements?
The above two questions can be answered by developing and applying reliability
techniques to compartment failure. In reliability technology, a compartment is a
‘system’ composed of structural elements as ‘components’. Reliability is defined
as the probability that a component or system will perform its designed-for
function without failure in a specified environment for a designed period. In
structural fire protection, a structural element or compartment constructed in
an environment such as an office building, department store, hotel or hospital
should be reliable in not failing within a specified fire resistance period when a
fire occurs.
Reliability of structural fire protection and of other fire protection devices, such
as automatic detection and sprinklers, is an under-researched topic on which only
very few investigations have been carried out so far – see Ramachandran (1998b).
This problem involves complex statistical (probabilistic) analyses requiring large
amounts of data which are not available at present. Attempts are, however, made
in this section to provide a framework for carrying out a reliability analysis of
structural (compartment) fire protection.
Consider first the thermal failure of a structural element or compartment.
If the ith element (i = 1, 2, …, 6) has a reliability (probability) Ri(Ti) of not
failing thermally within Ti minutes, [I – Ri (Ti)] is the probability of failing after Ti
minutes. In other words, there is a probability [I – Ri (Ti) ] that, after Ti minutes,
heat will penetrate through the structural element to an adjacent compartment
or space. Consider the simpler case with Ti equal to a constant T such that all the
six structural elements have the same fire resistance, T according to fire resistance
tests. But the elements may have different reliabilities, Ri (T), due to differences
in weakness caused by openings (doors, etc) and other factors.
Thermal failure of the compartment would occur if any of the six structural
elements fail thermally and allow the fire to spread to an adjacent compartment
or space. Under this assumption, the elements (components) are in a ‘series’
arrangement in reliability technology in regard to the compartment considered
as a ‘system’. In this case, according to a fundamental theorem in reliability
theory, the reliability, Rc(T), of the compartment in not failing within T minutes
is the product of the reliabilities of the structural elements in not failing within T
minutes.
Rc(T) = R1(T) . R2(T) . … . R6(T) (7.3)
It will be apparent from Equation (7.3) that the reliability of the compartment
is likely to be less than the reliability of any of the structural elements, since the
values of Ri(T) (i = 1, 2, …, 6) are all less than one. The value of Rc(T) will be
significantly reduced if the values of Ri(T) are considerably less than one. If, for
example, the reliabilities of the elements, for a given value of T, are 0.95, 0.96,
0.96, 0.97, 0.98 and 0.99, the reliability of the compartment will be 0.82. Such
a low probability of 0.82 for compartment success and a high probability of 0.18
for compartment failure may not be acceptable due to consequences in terms of
damage to life and property.
Suppose, for example, an acceptable minimum value for compartment reliability
is 0.999 with a failure probability less than 0.001. Such a high level of reliability
may be necessary due to the fact that the building considered is big with a large
number of occupants. A compartment reliability of 0.999 can be achieved by
providing fire resistance to the structural elements sufficiently high that the value
of the product of their reliabilities exceeds 0.999. This condition can be met, for
example, if the reliabilities of each of the four walls is 0.9998 and the reliabilities
of the floor and ceiling are 0.9999:
(0.9998)4. (0.9999)2 = 0.999
To determine the fire resistance required for a structural element to meet a

specified target level for reliability, consider a wall in the above example which
should have a reliability of 0.9998. For this purpose, it may be considered desirable
to apply the Beta method discussed in Section 3.7.4. Following this method,
Equation (3.108) may be applied, assuming that both the fire resistance and fire
severity have normal probability distribution.
For a success probability of 0.9998, straight interpolation of the figures for
0.9995 and 0.9999 in Table 3.18, would give an approximate value of 3.61 for
β but a more accurate value is 3.54 according to a table of standard Normal
distribution. If the compartment considered is an office room, as discussed in the
example in Section 7.2.1, the values of the mean, µs, and standard deviation, σs,
of fire severity are 43 and 38 minutes respectively. The coefficient of variation
Vs, is 0.88 (= 38/43). Such a high value for Vs indicates that the survey of fire
loads in offices quoted in the CIB Design Guide (CIB W14 1986) had produced
inaccurate estimates. Results with a coefficient of variation exceeding 0.2 are
generally considered by statisticians to be inaccurate.
For purposes of illustration, assume that µs is 45 minutes and σs is 9 minutes
with a coefficient of variation of 0.2. The mean fire resistance, µr, is the parameter
to be estimated. For the coefficient of variation VR, of fire resistance, 0.15 may
be considered to be an acceptable value if data are not available for estimating
this parameter. Accordingly, the standard deviation, σr, of fire resistance is 0.15µr.
Then, according to Equation (3.108), an estimate for µr is given by
1
(7.4)
µ r = 45 + 3.54 ( 9)2 + ( 0.15µ r )2  2
1
= 45 + 3.54 81 + 0.0225µ 2r  2
Converting Equation (7.4) into a quadratic, the following equation would

provide an estimate of µr:
0.718µ 2r − 90µ r + 1009.9404 = 0 (7.5)
Since the fire resistance should exceed 45 minutes, an estimate of its value is
113 minutes according to a solution for Equation (7.5). This value (113 minutes)
for minimum fire resistance for a wall can be expected to meet a success probability,
i.e. reliability, exceeding 0.9998 or failure probability less than 0.0002.
The value of the safety factor, θ, in Equation (3.113) is, hence, 2.51 (= 113/45)
as also given by Equation (3.115) with VR = 0.15, VS = 0.2 and β = 3.54. If data
are not available for estimating σr and σs, Equations (3.113) and (3.115) may
be applied for estimating µr by estimating µs according to an equation such as
Equation (3.78) and assuming realistic values for the coefficient of variation VR
and VS for fire resistance and severity.
The value of β is 3.719 (Table 3.18) for a reliability or success probability of
0.9999 for the floor and ceiling. With values of 0.15 and 0.2 for the coefficients
of variation VR and VS, of fire resistance and severity, Equation (3.115) would
provide an estimate of 2.66 for θ in Equation (3.113). Hence, the floor and ceiling
should be provided a fire resistance exceeding 120 minutes (= 2.66 × 45) to meet
a reliability exceeding 0.9999 or failure probability less than 0.0001.
According to the probabilistic analysis discussed above, fire resistance
exceeding 120 minutes for all the six structural elements can be expected
to provide a reliability exceeding 0.999 for the compartment considered as an
example. Similar calculations can be carried out to determine the fire resistance
required for all the compartments in the office building considered or in any other
type of building. A much higher level of fire resistance, with a very high reliability,
would be required for the compartments of a building which is very big or tall with,
say, over one or two thousand occupants. Such huge buildings would also require
adequate means of escape, fire detection and suppression systems and fire warning
and communication systems apart from structural fire resistance (Ramachandran
2008).
The formula in Equation (7.3) for components in a ‘series’ arrangement is only
valid if the failure of one component does not lead to the failure of any other
component in the same system, i.e. the components are independent. This may
not be true in the case of thermal failure of a compartment, involved in a fire,
with walls, floor and ceiling as components. Progressive deterioration of a wall
under severe heat might, in some buildings, affect the performance of the floor or
ceiling. A fire-resisting wall may be affected by the deflection of a beam in a fire
– see Figure 7.1. Joints and other constructional features are likely to cause such
a dependency.
There is a need to carry out further research to modify the formula in Equation
(7.3), particularly for structural fire protection, to take account of the interactions
between structural elements such as columns and beams with regard to their
performance in a real fire. It is necessary to identify these interactions and
quantify them in order to identify interactions which may exercise critical effects
on the probability of compartment success or failure in a fire – see Ramachandran
(1998c).
Only the ‘thermal failure’ of a compartment in a fire has been considered in the
probabilistic and reliability analysis discussed so far in this chapter. There is also a
need to investigate the ‘collapse’ (total destruction) of the structural barriers of a
compartment in a fire, although this may be a rare event with a very low probability
of occurrence. The risk or probability of occurrence of collapse should be reduced
to a very low level, particularly for the compartments of a big or tall building with a
Before fire After fire
L/100
(at midspan)
Internal fire
wall
Figure 7.1 Effect of beam deflection on a fire-resisting wall

Source: British Standard BS 5950: Part 8, 1990, section four, page 15
large number of occupants. The application of extreme value theory for estimating
the probability of collapse of a tall building against a catastrophic fire and the fire
resistance required to reduce this probability to acceptable level was discussed by
Ramachandran (2008).
Generally, the fire load contained in the normal contents (furniture etc) of a
compartment is unlikely to produce a highly intense heat necessary to cause the
collapse of the structural boundaries of the compartment during a short period
after the occurrence of flashover. But the boundaries can collapse over a longer
period of time after flashover due to progressive deterioration caused by intense
heat. Collapse of the structural boundaries can occur immediately after the start
of a fire, if the fire causes explosion by igniting a large volume of gas if present in
the compartment.
A compartment would collapse if any of the structural elements collapse,
particularly due to the interactions between the elements. This assumption
appears to be reasonable and would lead to a ‘series’ arrangement of components
(structural elements) in reliability technology with regard to the collapse of a
compartment considered as a system. Probabilistic and reliability techniques need
to be further developed to predict the occurrence of collapse of a compartment or
a building due to fire or explosion.
7.3 Performance and reliability of building structure

As shown by the figures in Tables 7.1 and 7.2, depending on the occupancy type,
area of fire origin and material ignited first, about 15 per cent of fires spread beyond
the room of origin. This percentage would be reduced by a factor of more than
two if the building was sprinklered. The spread, of course, would also depend on
other materials involved in the room of fire origin and other materials in the floor
of fire origin and in other floors in the building. A fire spreading beyond the room
of origin can thus spread throughout the building. Some fires can spread beyond
the building of origin and ignite other buildings in the immediate vicinity but the
probability of occurrence of such a fire is small according to fire statistics.
A stochastic model such as network model, discussed in Section 3.4.4, can
be applied to estimate the probability of fire spread throughout a building. Such
an extreme event with a low probability would progressively occur over a long
period in minutes, depending on the materials involved and the building size. But
it can occur during a shorter period if, in addition to the ignition sources normally
present in the building, a highly flammable fuel from a source outside the building
is thrust into the building accidentally or deliberately. A recent example of this
outside source is the large quantities of jet (aviation) fuel from the two aircraft
which were hijacked by terrorists and crashed on the twin towers of the World
Trade Center (WTC) in New York on 11 September 2001.
In the WTC disaster, as the burning jet fuel spread across several floors of the
two buildings, it ignited much of the buildings’ contents, causing simultaneous
fires on several floors. Over a period of many minutes, this heat induced additional
stresses into the structural frames, damaged by the impact of the aircraft, while
simultaneously softening and weakening these frames. This additional loading and
the resulting damage were sufficient to induce the collapse of both structures.
The progressive collapse of the WTC towers was a result of the combined effects
of the impacts of the aircraft and ensuing fires. Collapse of a building can also
progressively occur over a short period of time (minutes) if an explosion occurs in
the compartment of fire origin due to the ignition of a large volume of gas present
in the compartment.
Deterministic and non-deterministic (probabilistic or stochastic) models
developed so far do not appear to be capable of assessing the performance in
a fire of the structure of an entire building, although the models can assess, as
discussed earlier, the performance of the structure of individual compartments in
the building. The performance of a building can, perhaps, be evaluated by further
developing and applying the reliability technique discussed in Section 7.2.2.
For this purpose, a building may be considered as a system composed of several
compartments and floors as components. In this context, reliability may be defined
as the probability of the entire building not failing thermally for a specified period
of minutes. The structural members of the building should be designed to meet
this specified reliability standard. This standard should take into consideration the
likely damage to life and property if the structure fails. Estimating the performance
or reliability of a building structure in not collapsing in a fire is a more complex
problem.
The performance of a building structure, with regard to thermal failure or
collapse of structural members, cannot be considered in isolation from other passive
fire protection measures, e.g. means of escape facilities, and active measures, e.g.
detection and sprinkler systems installed in the building. The total performance of
the building should be evaluated, taking into account the interactions between all
the structural and passive measures and active measures. The total performance
should be such that the unreliability, i.e. failure probability, is less than a specified
target probability. The target levels for structural failure and for other aspects
such as life risk, property damage, consequential losses and societal risk have been
discussed in detail in Chapter 4.
As discussed in Section 4.2.3, in determining the target levels and fire
protection for a large building, consideration should be given to the probability of
occurrence of a multiple-death fire which would depend on the number of people
at risk in the building. Consideration should also be given to the probability of a
fire spreading beyond the building of origin and involving other buildings in the
neighbourhood, leading to a conflagration causing catastrophic damage to life and
property and consequential losses. To reduce the probability of occurrence of a
conflagration, the distances between buildings in an area should exceed a certain
‘critical distance’ and the street widths should be designed accordingly.
References
Baldwin, R and Fardell, L G (1970), Statistical Analysis of Fire Spread in Buildings, Fire
CIB W14 (1986). Design guide: structural fire safety. Fire Safety Journal, 10, 2, 77–154.
Homer, R D (1979), The protection of cold-form structural elements against fire, Proceedings
of International Conference on Thin-Wall Structures, Wiley, New York.
Langdon-Thomas, G J and Ramachandran, G (1970), Improving the effectiveness of the
fire check door, Fire International, 27, 73–80.
Ramachandran,G (1995), Probability-based building design for fire safety, Part 1, Fire
Technology, 31, 3, 265–275; Part 2, Fire Technology, 31, 4, 355–368.
Ramachandran, G (1998a), Probabilistic Evaluation of Structural Fire Protection – A
Simplified Guide, Fire Note 8, Building Research Establishment, Fire Research Station,
Borehamwood.
Ramachandran, G (1998b), Reliability of fire protection systems, Advances in Reliability
Technology Symposium, Manchester, UK, April.
Ramachandran, G (1998c), Reliability of structural fire protection, Book of Abstracts,
Annual Conference on Fire Research, NISTIR 6242, National Institute of Standards and
Technology, Gaithersburg, MD.
Ramachandran, G (2003), Probabilistic models for fire resistance evaluation, Proceedings of
the Conference on Designing Structures for Fire, Society of Fire Protection Engineers and
Structural Engineering Institute, Baltimore, MD.
Ramachandran, G (2008), Enhanced structural fire protection for a tall building against
a catastrophic fire – probabilistic evaluation of performance and economic value,
Proceedings of the 7th International Conference on Performance Based Codes and Fire Safety
Design Methods, SPFE, Auckland.
8 Performance and reliability of
detection, alarm and suppression
8.1 Detection
8.1.1 Human detection

As defined in Section 3.7.3.2, the total time taken by an occupant or group of
occupants to evacuate a building involved in a fire is the sum of three periods
demarcated sequentially by four critical events: ignition, discovery or detection
of the fire; commencement of evacuation and reaching a safe place. The first
period is the elapsed time from the start of ignition to perceive or discover the
existence of the fire. This time period can be long if the building is not equipped
with automatic fire detection systems. In such a building, the occupants have to
rely on fire cues such as unusual smells or noises, e.g. breaking glass, which are
generally ambiguous or misinterpreted (Canter, 1985).
As mentioned in Section 3.3.2, the fire brigades in the UK provide estimates
of the fire discovery time according to four categories – discovered at ignition,
discovered under 5 minutes after ignition, discovered between 5 and 30 minutes
after ignition and discovered more than 30 minutes after ignition. While the
discovery time is zero for the first category, average values of 2, 17, and 45 minutes
can be used for human discovery times for the second, third and fourth categories
if an automatic detection system is not installed. Based on the number of fires
for these four categories for the 14-year period 1978–1991, the average human
discovery time was 11 and 12 minutes for single and multiple-occupancy dwellings.
The statistics for 1978–1991 were used in estimating the relationship between
the fatality rate per fire and discovery time discussed in Section 3.3.3 – see
Equation (3.46). In this investigation, the first discovery time category, discovered
at ignition, was not included, since the fatality rate for this category was higher
than that for the second category. Excluding the first category, the average
discovery time based on the other three categories was estimated to be 13 and
18 minutes for single and multiple-occupancy dwellings; these correspond to the
overall fatality rate of 0.012 for both the types of dwellings.
In occupancies such as industrial premises, office buildings and department
stores, a fire occurring during the day will be discovered soon after the start of
ignition, since occupants and staff will be generally present. Fires occurring in
Performance and reliability of detection, alarm and supression 269
these premises in storage areas and during night times may take longer times to
be discovered if automatic detection systems are not installed. The discovery time
may be reduced if the premises are patrolled by security personnel.
In hospitals, detection of a fire takes place relatively early in fire development
as compared with other occupancies – see Ramachandran (1990). This is likely
to be due to the more general spread of people throughout the building and the
fact that there is always somebody awake, on duty. According to an analysis of fire
statistics compiled by the National Health Service Estates, UK (1996), most fires
(60 per cent) in hospitals are discovered by employees, 7 per cent by patients and
about 5 per cent by visitors and passers-by. About 26 per cent of fires are detected
by smoke detectors and 2 per cent by heat detectors. The vast majority (84 per
cent) of fires in hospitals are detected within 5 minutes. In patient care areas,
about 90 per cent of fires are detected within 5 minutes, about 32 per cent by
smoke detectors, apparently due to a higher coverage by these automatic devices.
Where ignition is ‘deliberate’, the patients themselves may raise the alarm. In non-
patient care areas, only 73 per cent of fires are detected within 5 minutes, about
18 per cent by smoke detectors. This may be due to the lower number of staff and
lower coverage by automatic detectors in some of these areas of hospitals.
While smoke and toxic gases pose a greater threat than fire to occupants who
are remote from the place of fire origin, fire itself is a major threat to occupants
in its immediate vicinity, who could be affected seriously by heat and flame. Even
if a fire is discovered soon after ignition starts, it may be too late for people in the
room of fire origin to attempt any fire fighting or escape. According to UK fire
statistics, most of the casualties in single and multiple-occupancy dwellings were
found in the room of fire origin. Figures for dwellings reveal that a high percentage
(62 per cent) of fatal casualties in the room of origin were caused by burns apart
from gas or smoke.
In cases where anyone is present at the ignition and detection stage they are
nearly always incapacitated in some way by alcohol, mental or physical retardation,
sleep or a combination of these conditions – see Ramachandran (1985). Those
present, whose responses are not thus impaired, are usually arsonists. Studies of
actual fires suggest that initial response to a fire may be triggered by any of the
following four senses – sight, hearing, smell and touch. Sight is the most common,
followed by sound, with smell and touch much less often. Visual detection of
smoke and less frequently of flames predominates in domestic fires.
8.1.2 Automatic detection
8.1.2.1 Performance
Mathematical (deterministic) models have been developed to calculate the
response time of an automatic heat or smoke detector under given conditions of
ceiling height, detector spacing and fire/smoke intensity (total heat/smoke release
rate). The parameters of these models are generally estimated with the aid of
data provided by standard tests and research experiments with detectors, carried
out under known and controlled conditions. However, satisfactory operation
of a detector head in an actual fire occurring in a room and time of operation
are random variables affected by uncertainties caused by several factors. These
factors include the location of the seat of a fire in relation to the location of the
detector head, the rate of growth of the combustion product, heat or smoke
and environmental conditions such as humidity, temperature and ventilation
prevailing in the room of fire origin.
Whether or not a smoke detector will respond depends on a number of
factors which include smoke aerosol characteristics, aerosol transport, detector
aerodynamics and sensor response. Smoke aerosol characteristics at the point
of generation are a function of the fuel composition and the combustion state
(smouldering or flaming) and include particle size and distribution, composition,
colour and refractive index. Once smoke reaches the detector, the response
depends on the aerodynamic characteristics of the detector and the type of sensor,
ionisation or photoelectric.
According to a Swedish study (Bengtson and Laufke, 1979/80), operating times
for heat detectors range from 2 minutes in ‘extra high hazard’ occupancies such
as a plastic goods factory, to about 20 minutes for ‘light hazard’ such as flats and
other residential premises. The operating times of smoke detectors range from 0.5
minutes for ‘extra high hazard’ to 2.25 minutes for ‘light hazard’ involving wood
materials and to 0.75 minutes for polystyrene (light hazard).
In a series of tests carried out by the Fire Research Station, UK in 1970, heat
detectors operated between 1 min 16 sec and 3 min 58 sec of ignition, ionisation
chamber detectors operated between 1 min 5 sec and 4 min 30 sec, while optical
detectors took over 3 minutes to operate. Infrared detectors operated in about 3
minutes and laser beam detectors took about 5 minutes to operate, if well above
a fire.
According to Fire Statistics United Kingdom 1991, published by the Home
Office, the proportion of fires discovered in less than 5 minutes in dwellings was 69
per cent for fires discovered by smoke detectors. In other words, the probability of
a smoke detector operating within 5 minutes is 0.69. In other occupied buildings,
the probability of a smoke detector operating within 5 minutes is 0.78.
Statistical data are lacking for evaluating the probability of a detector system
operating in a real fire. A value of 0.8 for this probability has been assumed by
Helzer et al. (1979) in a study concerned with the assessment of the economic
value of different strategies for reducing upholstered furniture fire losses. Apart
from other reasons, a detector head would fail to operate if the heat or smoke
generated by a fire is insufficient to activate the system.
8.1.2.2 Effectiveness
Detectors do not actively take part in fire fighting. However, by detecting and
informing the occupants of a building about the existence of a fire when the fire is
in its early stages of growth, detectors would enable the early commencement of
fire fighting by first-aid means such as buckets of water or sand and portable fire
extinguishers and/or fire brigade. Consequently, the fire can be controlled and
extinguished quickly before it spreads and causes extensive property damage.
According to Fire Statistics United Kingdom 1991, among fires in occupied
buildings detected by smoke detectors, 67 per cent of fires are confined to items
first ignited and 0.2 per cent spread beyond the building of fire origin. If smoke
detectors are not installed in these buildings, only 36 per cent of fires would be
confined to items first ignited and 2.5 per cent would spread beyond the building.
For dwellings, the probability of a fire being confined to the item first ignited is
0.68 if a smoke detector has been installed and it operates and 0.41 if a smoke
detector has not been installed. According to an analysis of statistics for the period
1960–1967 compiled by Cerberus, a Swiss manufacturer of ionisation detectors,
the average fire loss in buildings in Switzerland protected by Cerberus detector
systems was only one-third of the average loss in buildings without these systems.
Statistics in the UK have shown that fire brigades are able to control and
extinguish fires quicker in buildings protected by automatic fire detection systems
than fires in buildings without these systems. The fire brigade control time is
reduced by about half a minute for every minute of early arrival of the brigade at
the fire scene (Ramachandran, 1992). This reduction would vary from one type
of building to another. The saving in control time due to detectors, together with
the reduced time in detecting a fire, would considerably reduce the total duration
of burning and area damage. The area damage will be further reduced, though not
significantly, if the detector system is directly connected to the fire brigade – see
Figure 3.11.
Figure 3.11 is based on the exponential model (Equation (3.42)) discussed in
Section 3.3.2. This figure is an example (textile industry) showing the sizes of the
fire in terms of area damage (m2) at the times of fire brigade arrival and control
for three cases – detector connected to the brigade, detector not connected to
the brigade and detector not installed. The figure reveals a significant reduction
in area (property) damage due to automatic detectors. The figure also shows that,
in the absence of fire brigade intervention and attack, a fire in a textile industry
building can burn for more than 54 minutes with a damage exceeding 140m2.
The important time in a fire situation is the first five minutes after the start
of ignition when the occupants are attempting to escape. Early detection of a
fire would enable the early commencement of evacuation which would increase
the chance of occupants reaching a safe place before the escape routes become
untenable due to heat, smoke or toxic gases. According to the study discussed
in Section 3.3.3, and the parameter λ in Equation (3.46), for every minute saved
in detection time (and hence in evacuation time), the fatality rate per fire in
dwellings would be reduced by about 0.0007, i.e. 7 deaths per 10,000 fires.
Automatic detectors would reduce the fire discovery time in dwellings by about
14 minutes, assuming that they operate, on average, in one minute after the start
of ignition. Consequently, if all the dwellings had been protected by automatic
detectors, the fatality rate per fire would have reduced to about 0.002, i.e. 2 deaths
per 1000 fires from 0.012, i.e. 12 deaths per 1000 fires, which was the average life
risk level that prevailed during 1978–1991. This denotes a saving of 10 deaths per
1000 fires. With about 55,000 fires per year in dwellings during that period, 550
lives could have been saved every year if, particularly, smoke detectors had been
installed which operated, on average, within one minute. According to a US study
(Bukowski et al., 1987), detectors would reduce the fatality rate per fire in single
and two-family dwellings by a factor of two, from 0.0085 to 0.0043.
8.1.2.3 Reliability
Although automatic detectors are only required to operate in a proportion of
fires not discovered by occupants, e.g. less than one-third of fires in hospitals, the
system should be available for operation ‘on demand’ when a fire occurs. If the
system is maintained in good working condition by frequent routine checks, it will
be ‘available’ most of the time for operating on demand and raising an alarm when
a fire generating sufficient heat or smoke breaks out. Unrevealed mechanical
faults or defects will remain unrectified if the system is not checked frequently.
Failure of electrical mains will lead to unavailability but this incidence only occurs
rarely. The probability of a detector operating on demand whenever it is required
to act is the first aspect of reliability of an automatic fire detector (AFD) system.
The second aspect of reliability of an AFD system is concerned with the system
not triggering an alarm in a non-fire situation. Examples of such situations are
dust, debris and insects in the sensing chamber and system fault. The system
should not also trigger alarms in fires which have no potential to spread beyond
the point of ignition. Examples of such fires are cooking smoke, bathroom water
vapours and cigarette smoke. The two types of situations mentioned above are
generally called false or nuisance alarms. Such alarms cause wastage of time and
money, particularly to fire brigades whose response to genuine fires may be delayed
due to unnecessary c all-outs.
Major causes of failure of AFD systems to detect genuine fires are:
• mechanical faults;
• malfunction;
• power surges;
• power failure;
• mechanical damage or abuse after installation; and
• accumulation of dirt and dust.
For estimating the relative frequencies of failure due to different causes, ‘global
statistics’ at the national level do not appear to be available for any country.
Since 1994, fire brigades in the UK provide in their fire reports (FDRI)
information on whether a detection system was installed in the building involved
in a fire, and if so, whether the system operated or not and on the reasons for
non-operation. But the Home Office only processes data provided by a sample
of fires to produce national statistics. Hence, it is difficult to use the fire statistics
compiled by the Home Office to estimate reliably the probability of a detector
operating in a genuine fire and the frequencies of non-operation due to different
causes. The reports on fire incidents in UK hospitals compiled and analysed by
National Health Service Estates do not contain information on fires in which
AFD systems failed to operate and the causes of such failures.
As mentioned earlier, major causes of false or nuisance alarms are signals from
cooking smoke, e.g. burnt toast, steam from boiling water in an electric kettle,
bathroom water vapours, cigarette smoke, dust, debris and insects in the sensing
chamber, system fault and lack of maintenance. Fry and Eveleigh (1975) analysed
data collected in a survey on detector actuations carried out by UK fire brigades
in 1968. They estimated a ratio of 11:1 between false and genuine calls to the
fire brigades. They attended 11 false alarms to every genuine fire attended and
extinguished by them. The ratio was 11:1 for heat detectors and 14:1 for smoke
detectors. Mechanical and electrical faults, especially defective wiring of heads,
accounted for 46 per cent of false calls. Ambient conditions, especially extraneous
heat and smoke, accounted for 26 per cent of false calls. According to Davies
(1984), 95 detector systems of a Swiss manufacturer gave 1329 false calls as
opposed to 85 genuine alarms (a ratio of 16:1).
Reasons for nuisance alarms arising from smoke detectors in homes have
been identified by the National Smoke Detector Project carried out by the US
Consumer Product Safety Commission – see the Commission’s final report (1993)
on the first study, ‘Smoke Detector Operability Survey – Report on Findings’. As
pointed out in this report, power sources for a high percentage of smoke detectors
in homes are intentionally disconnected because of nuisance alarms. The report
also suggested several potential solutions to address this problem. Repeated false
alarms for an organisation may cause a fire brigade to cancel connection facilities,
thus exposing the organisation to increased fire risks.
In 1995, the Home Office, UK, introduced a reporting form FDR3, in which
fire brigades were asked to furnish information on false alarms attended by
them. In this form, false alarms are classified according to three main categories:
malicious, good intent and due to apparatus. The third category has been
further classified into the following four sub-categories – dust/thrips, system
fault, unsuitable equipment or positioning and ‘other’. The breakdown figures
for these four sub-categories were furnished by 23 brigades in 1995 and 50
brigades in 1996. According to an analysis of these figures by Ramachandran
(unpublished), 22 per cent of false alarms were due to dust, insects etc., 37 per
cent due to system fault, 3 per cent due to unsuitable positioning and 38 per
cent due to ‘other’.
As mentioned earlier, statistical data are lacking for evaluating the ‘on demand’
probability of a detector operating in a genuine fire. In estimating this probability,
it is not appropriate to include small fires in which the heat or smoke generated
is not sufficient to activate the system. This success probability is as high as 0.95
according to some manufacturers, provided, of course, the system is maintained
all the time in a satisfactory working condition. The failure probability is 0.05.
Manufacturers usually use the exponential probability distribution to calculate
the failure rate of their product per hour by carrying out laboratory tests and then
convert the results to estimate the failure rate per year. Most of the manufacturers
are reluctant to disclose the data and method adopted by them in the calculation
of the failure rate.
If an item has an exponential lifetime probability distribution, 63 per cent of
the specimens of the items would fail within a time t when tested in a laboratory;
37 per cent would survive beyond the time t. Test data would provide an estimate
of t which is known in reliability theory as mean time to failure (MTTF) for a
non-repairable item or mean time between failure (MTBF) for a repairable or
replaceable item. MTTF or MTBF is the average time likely to lapse before the
item fails after it is installed or replaced after it had failed. The reciprocal (1/t),
where t is MTTF or MTBF, is the failure rate which is usually expressed as failure
rate per year.
The Weibull probability distribution can be applied to estimate the failure rate
of an AFD system for detecting genuine fires. The failure rate for this distribution
is not a constant but a function of time since the system has been installed
or replaced. Formulae have been derived to estimate the parameters of this
distribution and MTTF or MTBF. To estimate these parameters, data are required
on the date of installation or replacement of each of the AFDS systems installed
in an occupancy and the dates on which the system has failed. If such data are
available, the Weibull distribution can be fitted to the data and the parameters
estimated to assess whether the age of the system is a factor affecting the failure
rate.
For some types of buildings, information may be available on the dates of
occurrence of false alarms and types of fire detection systems producing these
alarms. The Weibull distribution can be fitted to time periods between successive
dates of false alarms to estimate the failure rate – see, for example, Peacock and
Sutcliffe (1982). Failure, in this context, is the failure of an AFD system to block
the communication of information about the false alarm to occupants of a building,
control panels or to the fire brigade. A conventional heat or smoke detector does
not have the capability to discriminate between genuine fires and false alarms.
But such a capability has been incorporated in some more ‘modern’ computer
controlled AFD systems such as analogue addressable systems, which have been
developed during the past two decades. However, the reliability of these modern
systems depends on the data collected by the sensors and the method (algorithm)
adopted in the computer software (program) installed in the system.
In an addressable system, signals from each detector and each call point are
individually identified at the control panel. Each circuit is a form of simple data
communication rather than simply an electrical circuit. Within the software of an
addressable system, the device identity can be converted into a pre-programmed
location, which is then displayed on some form of text such as an LCD or vacuum
fluorescent display. The control panel of a conventional heat or smoke detector
system cannot identify the detector head from which a signal is communicated to
a panel.
Among the conventional types, heat detectors are generally the most reliable
in terms of component failure, since these devices respond directly to the presence
of heat by a physical change in the detector operating elements. Heat detectors
do not react to smouldering fires which do not generate heat. A heat detector
does not have an open measurement chamber and is only affected by increases
in temperature. The fire will only be detected when the temperature in the room
of the fire origin has reached a threshold level. This simplicity leads to very few
unwanted alarms. Heat detectors require practically no maintenance.
Since an optical type smoke detector has an open measurement chamber, it is
susceptible to dirt and dust accumulation which will lead to lower sensitivity. An
ion type smoke detector also has an open measurement chamber which must be
cleaned as often as necessary. Otherwise the chamber will become dirty or dusty,
leading to an increase in the sensitivity of the detector and unwanted alarms.
Addressable detection systems can be tested and checked by a printout of the
measurement data in the central control unit. Detectors with measurement data
which have changed beyond a given limit must be cleaned. The others will operate
normally and can be used without any further inspection until the next routine
check. For the reasons mentioned above, service and maintenance are simplified
substantially in the case of addressable detectors, meaning far greater reliability
than conventional systems.
The four major components of an AFD system are:
• detector heads;
• zone control panels;
• central control panel; and
• connection to fire brigade.
Each group of detector heads located in a specific place, e.g. ward in a hospital,
are generally connected to a particular zone panel. The zone panels are connected
to a central panel which is connected to the fire brigade.
Gupta (1984/85) estimated the hardware failure rates of components of an
automatic fire detection (AFD) system at a psychiatric hospital by analysing the
component structure and configuration of the system. The system comprised a
distributed system of ionisation type smoke detectors, break-glass units and heat
detectors. All of them were connected into various zone panels, which in turn
were connected into a central control unit and three repeater panels. The control
unit carried a fire area identity annunciation and was the means for the receipt
of the alarm for activating the hospital’s audible fire alarm and the transmission
of a fire brigade call-out signal. The failure rates of electronic components were
obtained from Military Standardization (1974).
For ionisation smoke detectors, Gupta estimated the total mode failure rate
as 0.057 faults per year of which 0.04 faults per year were of the safe type and
0.017 faults per year were dangerous. Since these figures applied to a first class
environment, factors such as air speed and humidity were taken into account
and the failure rate was assessed to be 0.46 faults per year, which was eight times
greater than for a first class environment. For break-glass unit, the total relevant
failure rate was assessed to be 0.032 faults per year of which 0.018 faults per year
were of the safe type and 0.014 faults per year were dangerous. The figures did
not include the spurious alarm rate due to misuse of the system. For control units,
Gupta estimated the total failure rate for unrevealed dangerous type as 0.06 faults
per year – 0.042 faults per year for indicator control module and 0.018 faults per
year for monitor unit.
Studies by Finucane and Pinkney (1988) showed overall failure rates for control
units varied from 0.25 faults per year, up to one fault per year, with an unrevealed
fail-to-danger rate of typically 0.1 faults per year. According to these authors the
overall failure rates for all types of detectors were 0.1 faults per year with fail-to-
alarm failure rates varying from 0.01 faults per year to 0.1 faults per year. For all
types of detectors, Appleby and Ellwood (1989) estimated a failure rate of 0.02 per
year. According to them, the failure rates of addressable systems can be expected
to be considerably less than 0.02 per year.
Failure rates for electrical components and sub-components are available from
data banks such as those maintained by UK Atomic Energy Authority (UKAEA)
and AEA Technology and Reliability Analysis Center, New York. These data can
provide some indication of the failure rates of AFD system components.
For all types of detectors, false alarm rate per detector per year has been
estimated in some studies:
• 0.025 by Finucane and Pinkney (1988);

• 0.025 by Appleby and Ellwood (1989);
• 0.092 by Peacock and Sutcliffe (1982) for Lincolnshire hospitals;
• 0.044 by Bukowski and Istvan (1980) for USA hospitals;
• 0.06 quoted by Caffolla (1997) for St Paul’s hospital, Vancouver;
• 0.047 by Japanese Association of Fire Science and Engineering (1982), for
Japanese health care facilities; and
• 0.1 by Pearce (1986) for UK health care premises.
Gupta (1984/85) divided false alarms into four categories:
1 failure of equipment;
2 ‘non-fire’ disturbances;
3 ‘external’ effects;
4 ‘unknown’ reason for alarm.
The second category included causes such as cigarette smoke, steam, dust and
smoke/vapour from cooking. The third category included human error, water from
leaks, power supply interruption or surge, electrical interference such as arcing or
switching and birds, animals and insects.
Gupta analysed data collected from various sites on time periods between
successive events for the categories mentioned above except the first. He fitted
the Weibull probability distribution to these data in order to understand the
statistical behaviour of the events. Parameters of the distribution were estimated
using maximum likelihood and least square methods. The mean of the ‘scale’
parameter, interval between successive events, varied between 10 and 42 days for
the second category (‘non-fire’), 16 and 40 days for the third category (‘external
effects’) and 14 and 60 days for the fourth category (‘unknown’). The value of the
‘shape’ parameter in all the three categories was less than 1.0 (except in ‘external
effects’ for one site) which indicated a decreasing failure rate for the occurrence
of the events.
On the other hand, the mean duration between two successive false alarms
at various sites varied between 5 and 15 days. This duration had little bearing on
the type of site, according to results presented in the paper. Little variance in the
‘shape’ parameter observed for different sites within a given category indicated a
common cause of false alarm. The explanation for this phenomenon was perhaps
the attribution to the increasing effectiveness, efficiency and better maintenance
policies of the fire officers of sites.
For any type or group of buildings, statistical data may be available for
estimating the overall reliability of the whole of an AFD system in detecting a
genuine fire or suppressing a false alarm. But the overall reliability of an AFD
system for a particular building within a type or group depends on the reliabilities
of the components and sub-components constituting the system and the types of
connections between them. Estimation of the system reliability of an AFD system
for a particular building is a complex problem on which practically no research
studies have been carried out so far. In developing a model for this problem, the
following general points may be considered.
In the case of detecting a genuine fire, the four major components of an
AFD system, mentioned earlier, are in a ‘series’ arrangement in the context
of reliability theory. Failure of any of the components to transmit information
about the existence of a fire to the next component in the arrangement will lead
to system failure For a series arrangement, the overall reliability of a system is
the product of the reliabilities of the components. Failure of a detector head
to operate when a fire occurs in the area protected by it may be considered
to be system failure. However, the fire may be detected by any other head in
the vicinity of the fire area or in the place considered. Hence, detector heads
in a specified place may be considered to be in a ‘parallel’ arrangement. For a
parallel arrangement the overall unreliability of a system is the product of the
unreliabilities of the components. Unreliability is the probabilistic counterpart
of reliability. But failure of one or more heads in a place will lead to a delay in
commencing the evacuation and fire fighting and increase in life risk. The system
is, therefore, a mixture of series and parallel arrangements of main components
and sub-components.
In the case of false alarms, none of the detector heads in a place should
respond to a signal from a non-fire situation for the system to succeed in not
communicating the signal. Hence, the heads in a specified area may be considered
to be in a series arrangement. Even if such a signal is transmitted to a zone panel,
further communication of the signal can be blocked by this panel or by the central
panel. The system would succeed if any of the four major components suppress
information about a false alarm. Hence, the components can be regarded as being
in a parallel arrangement.
Apart from the main components mentioned above, there are sub-components
of an AFD system such as cables/wires, power supply and discriminator software
(for an analogue addressable system).
Practically, any level of sensitivity to fire, in particular to signals of combustion,
can be achieved with existing technology. However, if sensitivity is set at a high
level, a detector may pick up signals given by spurious fires from sources such as
cigarette smoking and cooking which are normal activities. On the other hand,
a low level of sensitivity can increase the risk of genuine fires being undetected.
Some studies, e.g. Cholin (1975), have suggested that the gap between
reliability in detection and rate of unwanted and false alarms may be closed
to some extent by cross-zoning of detectors where the activation of the alarm
is delayed until a second detector is activated. Another approach suggested by
Custer and Bright (1974) is the use of multi-mode detectors requiring signals from
several fire signatures before a fire alarm is initiated.
The technique known as ‘coincidence’ or ‘double knock’ will automatically
withhold or limit a fire signal given by a detector until the presence of the fire
is confirmed by the response of a second detector in an independent circuit at a
different location. ‘Gating’ is another technique for automatically withholding a
fire signal given by a single detector until the presence of fire is confirmed by its
second response within a pre-determined time period. Techniques such as those
mentioned above might reduce the frequency of false alarms but may, at the same
time, cause undue delays in the operation of a detector in a genuine fire.
Advanced computer-controlled addressable detection systems such as analogue
and multi-state types can provide improved capability for discriminating between
a genuine fire and a false or unwanted signal from a non-fire source or a small fire
with a local influence such as a puff of cigar or pipe smoke. An analogue system
gives a ‘pre-alarm warning’ if the signal from a detector exceeds a certain threshold
level and a ‘fire warning’ if the signal exceeds a higher threshold level. At a very
low threshold level, a fault signal is given. The signal level represents the amount
of heat, smoke or flame that is being sensed. In a multi-state system, each detector
is capable of transmitting several states such as fault, normal, pre-warning and fire.
It is necessary to design a cost-effective AFD system which is capable of
achieving the right balance between detection of genuine fires and suppression
of false alarms. An acceptable level needs to be determined for the ratio between
number of false alarms and number of fires attended by fire brigades.
8.2 Alarm
8.2.1 Conventional alarms

The alarm, when raised, is a signal for occupants to evacuate or to be alert in
preparation for evacuation. The alarm should result in the fire service being
summoned, so that they can start fire fighting and if necessary assist in evacuation.
The most common form of fire alarm sounder used is the electric bell. Its sound
is able to carry through a building. It can ring as an intermittent pulse or a
continuous sound signifying respectively alert and evacuation. Where a large site
is involved, it may be necessary to use a siren. Where there will be difficulty in
hearing an alarm due to disabled occupants or noisy equipment, then a visual
signal, for example, flashing lights, will be necessary. The alarm may be required
to wake sleeping occupants. There have been a number of fires where lives have
been lost through the lack of an alarm system.
In buildings which have resident staff, visitors, patients or inmates, the fire alarm
is required to alert the staff and to commence a pre-arranged fire evacuation plan.
Low-note buzzers or electronic sounders can be particularly useful in situations
where it is necessary to warn staff, rather than the occupants, of fire. In shops and
shopping complexes, the alarm can take the form of a coded message to alert staff
to the existence of a fire and its location. In hospital premises and prisons, it is
normal to evacuate initially the zone of fire origin as part of a staged evacuation
plan. Visitors are usually asked to leave the building while other occupants make
their own way or are assisted to a safe refuge. Conventionally, in these instances,
fire alarm bells ringing continuously for evacuation and intermittently for alert are
used. Coded messages are also sometimes used.
The fire alarm system can also be used to activate other systems such as door
hold-open devices, motorised ventilation dampers/controls, powered smoke
extraction systems, fire extinguishing systems and staff call systems. Self-contained
smoke detectors with integral alarms could provide an earlier warning and so
significantly reduce the number of deaths.
However, a conventional fire alarm described above or even an automatic
detector can be insufficient in being able to convey an effective warning of fire
(Tong and Canter, 1985). A delay in the accurate perception of fire threat would
also lengthen the ‘recognition time’, the second component period, denoted as B,
of the total evacuation time, H, discussed in Section 3.7.3.2. During this period,
from perception to commencement of evacuation, people try to gain enough
information to confirm the existence of a fire. Even direct information of fire
(such as smoke) may be insufficient to motivate evacuation due to inaccurate
perceptions of the rate of fire growth (Canter et al., 1988).
8.2.2 Informative fire warning systems

For the reasons mentioned above, it is necessary to communicate to the occupants
of a building, timely, adequate and convincing information about the existence,
location and spread of fire. Such information can be provided by computer-
controlled informative fire warning systems (IFWS) which have been developed
over the last few years in the UK and in other countries. Ramachandran (1991)
reviewed the research carried out in the UK on IFWS and the basic features
necessary in these systems for motivating rapid and safe evacuation.
The Fire Research Station (FRS), UK, developed the prototype of an IFWS
known as BRESENS (Pigott, 1983). In systems such as BRESENS, simple
electronic design should ensure that no switch, sounder or sensor will give rise
to a fire alarm when one of these or the associated wiring develops an open or
short circuit condition. The outputs of fire ‘sensors’ in the prototype equipment
installed at FRS demonstrated the theoretical possibility of tailoring the output for
an approximate sensitivity to the ratio between false alarms and genuine fires. It is
possible to calibrate the sensor outputs and then compensate them for drift with
time. The amount of compensation necessary can itself be used as a measure of
probable serviceability, permitting demand maintenance.
The FRS specified its prototype equipment with liquid crystal 16-character
text displays, with display units placed at strategic points throughout a building.
These displays were programmed to show the time and date in normal use and,
when relevant, to indicate that an alarm test was being carried out. During a fire
alarm the system can display the word ‘Fire’ with two levels of detail – the general
area of the fire for display everywhere, and with a room number for those in the
immediate area of fire origin.
To improve the performance and quality of a system like BRESENS, under
contract from FRS, Surrey University commenced, in 1981, an investigation
to determine the nature and effective methods of presentation of appropriate
information through automatic informative fire warning (IFW) systems. A series
of studies was carried out. A summary of methods, results and conclusions relating
to these studies are contained in Ramachandran (1991).
To appreciate the need for IFW systems, it was necessary to assess the
deficiencies of responses to conventional fire alarm systems. For this purpose, a
street survey (Tong and Canter, 1985) of randomly selected individuals and fire
drills in three buildings were carried out. Results revealed that a conventional
alarm provokes deficient responses such that it may be ignored or it may simply be
the first alerting signal in a complex process whereby a need is created to explore
the reason for the alarm sounding. An alarm will be ignored if it is interpreted
as a circuit malfunction or system test. As pointed out earlier by Tong (1983),
inadequate behavioural responses of occupants in many cases were due to a
failure of conventional alarm systems to provide a clear and unambiguous fire
warning.
Certain psychological criteria need to be met if effective fire alarm systems
are to be manufactured and installed – see Tong (1983) and Canter et al.
(1988). These include a clear meaning of a fire alarm and a valid indication
of the presence of a fire and its location. Also, building occupants should be
provided with information on the most appropriate response to an alarm and on
the escape routes available.
Case studies (Tong, 1983), based on training and fire drills, identified the
following problems in the provision of information by existing means. A visual
indication about the zone of alarm origin by some form of indication equipment
was highly centralised in some cases where no information was directly available
to building occupants about the location of the fire. Information had to be
obtained from the point at which the zone indicators were installed by telephone,
personal attendance or radio paging devices. Problems were encountered in all
three methods. Other problems included inadequate amount of information and
locationally imprecise information.
The weaknesses of existing fire alarms and communication systems are being
rectified by advanced IFW systems that improve the input of information to
the central control equipment and the output of information from the central
control. Information is taken directly from each individual ‘addressable’ detector
and messages are displayed that enable people to distinguish fire alarms from
other alarms. To significantly reduce delays in evacuation, IFW systems should
be backed up by efficient and organised management. Large delays can occur, for
example, from a lack of exits of sufficient width or lack of provision of equipment,
e.g. wheelchairs for facilitating the evacuation of disabled people.
The messages to be displayed visually on an IFW system should be determined
with due regard to an optimum balance between urgency and detail. To tackle
this problem, it would be useful to consider the recommendations of Canter et
al. (1988) for message length, use of abbreviations, message specificity, message
formulation, type of information and message format. In a care establishment,
for example, the precise location of the fire (e.g. ‘linen store’) should be displayed
only on the affected floor while a less specific message (e.g. ‘Ward G7’) is displayed
on a non-affected floor. Location addresses should first refer to the most general
area (e.g. block or wing) followed by a more specific address (e.g. floor number)
and finally by the most precise address (e.g. ward or room). Location addresses
should take priority over the inclusion of any other type of information. Updated
information on fire/smoke spread should be displayed along with, but separate
from, the address of the fire source.
Instructions displayed on an IFW system should only be targeted at the
general population of building occupants. Instructions to individuals should be
communicated by other means. If information is displayed on a VDU-like node,
the use of grouping as a means of emphasising specific information is preferable to
organising text in a long string. More detailed and specific messages would only
be appropriate for buildings such as hospitals where there is a stable population
and not for sport areas, shopping centres and hotels where the population is
more transient and will only be able to react to very simple instructions.
Under a contract from the Fire Research Station, Technica carried out an
investigation to explore the possibility of using microprocessor technology more
sophisticated than an IFW system such as BRESENS (16-character LCD display) –
see Bellamy (1989). Technica performed two laboratory experiments by presenting
to subjects (members of the public) different warning displays (lists of choices) for
information acquisition, message identification and action decision. The subjects’
responses were recorded automatically on a computer. Five building types were
included – residential block, hotel, hospital, department store and office building.
The complete details including statistical design and analysis of these experiments
are described in the final report (Technica Ltd, 1990). Ramachandran (1991)
presented a summary of this report.
The first experiment was based on a (3 × 2 × 2 × 2) factorial design with three
modes each with high and low quality, two levels of building familiarity and two
levels of threat. The three modes were:
1 Graphic – computer-generated colour mimics based on building floor plans
and elevations; isometric (i.e. not perspective)
High quality – 3-dimensional
Low quality – 2-dimensional
2 Visual display of text messages
High quality – computer VDU
Low quality – LCD (BRESENS)
3 Auditory (speech)
High quality – computer*-generated
Low quality – fire bell
*Commodore Amiga PC model A1000
Subjects familiar with building plans were considered ‘trained’; otherwise

they were ‘untrained’. ‘High threat’ referred to a fire located on the ground floor
directly below the specified location of the subject. A ‘low threat’ fire was a fire
located on the third floor above the location of the subject. The 24 combinations
provided by the factors and their levels were ‘replicated’ by using 48 males and
48 females (four subjects to each combination) with repeated messages for each
subject across building types. Each subject was tested on one mode, one level of
familiarity and one level of threat.
The results of the first experiment are given in Table 8.1. As shown in this table,
the three component response times were analysed separately. The ranks of each of
the six modes for the mean of each of the response times are also shown in Table
8.1, together with the ranks of the probability (%) for the three cases indicated in
columns 2 to 4. The results are shown in brackets. For each of the modes, the mean
of the ranks over the six attributes (columns 2 to 7) is shown in the last column.
The results in Table 8.1 reveal that 3D graphic and Amiga speech are the best
modes. With these two modes, the second experiment (7 × 2 factorial design)
was performed with seven multi-mode combinations, specified in Table 8.2,
constituting the main factor and familiarity as the second factor. ‘IFWS speech’
referred to modified computer (Amiga) generated speech while ‘speech’ was the
unmodified mode. ‘Alarm’ referred to computer (Amiga) generated alarm sound.
The subjects were 56 males and 56 females with four males and four females for
each of the 14 factorial combinations.
The overall mean for the total response time for all factorial combinations
was 43 seconds in the first experiment and 38 seconds in the second. The second
experiment (not the first) revealed that mode had a significant effect on the total
time. The longest total response time of 50 seconds was for the combination
modified 3D/IFWS speech/alarm and the shortest time of 31 seconds was for 3D/
speech.
The main implication of the first experiment was that IFW systems could
produce as much as a six-fold increase in the probability (proportion) of occupants
evacuating immediately when compared with a conventional fire warning. The
evacuation initiation probability could be further enhanced by adopting the
mode combination 3D/IFWS speech. Familiarity with a building increased the
Table 8.1 Experiment 1: Summary of results and overall rankings (best to worst) for modes (ranks are shown in parenthesis)
Dependent variable
Genuine fire
(location correct) Genuine fire
warning warning Immediate Warning Warning Action decision
interpretation interpretation evacuation acquisition time interpretation time time Mean of
Mode % % % (secs) (secs) (secs) ranks
Graphic 3D 73 (1.5) 81 (1) 64 (1) 18 (5) 14 (1) 12 (3) 2.1
Graphic 2D 73 (1.5) 75 (2) 45 (2.5) 19 (6) 16 (3) 14 (5) 3.3
BRESENS 56 (4) 61 (4) 44 (4) 12 (3.5) 18 (4.5) 12 (3) 3.8
Amiga Text 48 (5) 52 (5) 28 (5) 10 (2) 24 (6) 16 (6) 4.8
Fire bell 5 (6) 13 (6) 11 (6) 6 (1) 18 (4.5) 12 (3) 4.4
Amiga speech 70 (3) 72 (3) 45 (2.5) 12 (3.5) 15 (2) 11 (1) 2.5
Table 8.2 Experiment 2: Summary of results and overall rankings (best to worst) for modes (ranks are shown in parenthesis)
Dependent variable
Genuine fire
(location correct) Interpretation as a
warning genuine fire Immediate Warning Warning Action decision
interpretation warning evacuation acquisition time interpretation time time Mean of
Mode % % % (secs) (secs) (secs) ranks
Fire bell 3 (7) 16 (7) 8 (7) 6 (1.5) 16 (6) 10 (1.5) 4.3
3D/IFWS 91 (1) 92 (1.5) 77 (1) 13 (4) 10 (1.5) 11 (3) 2.0
speech
BRESENS/ 88 (3) 89 (3) 63 (2) 18 (5) 12 (4) 12 (5) 3.7
IFWS speech
Modified 3D/ 70 (5) 77 (5) 42 (5) 20 (6.5) 14 (5) 15 (7) 5.6
IFWS speech/
alarm
3D/IFWS 84 (4) 84 (4) 50 (4) 20 (6.5) 11 (3) 12 (5) 4.4
speech/alarm
3D/speech 89 (2) 92 (1.5) 58 (3) 11 (3) 10 (1.5) 10 (1.5) 2.1
Amaiga alarn 6 (6) 33 (6) 23 (6) 6 (1.5) 18 (7) 12 (5) 5.2

probability of interpreting a warning as a genuine fire emergency but such a correct
interpretation did not guarantee immediate evacuation, which depended largely
on the mode of warning.
Conventional fire detectors would reduce the delay in discovering a fire
but may not reduce the ‘recognition time’ from discovery to commencement
of evacuation. IFW systems have the potential to reduce both the two time
components and hence to reduce the total time before the commencement of
evacuation. Consequentially, IFW systems have the potential to reduce life risk
significantly – see Ramachandran (1993).
8.3 Suppression
8.3.1 First-aid fire fighting

If a fire is discovered soon after ignition and its size is small, the occupants of a
building attempt to fight and extinguish the fire using first-aid methods. Such an
initial attack on a fire usually involves ‘sundry means’ such as buckets of water or
sand, smothering, garden hose, immersion and beating. According to a sample of
fires in dwellings analysed by Ramachandran et al. (1972), ‘sundry means’ were
used by occupants in about 44 per cent of the fires. About 43 per cent of these
were successfully extinguished; fire brigades had to extinguish only the remaining
57 per cent. Occupants used portable fire extinguishers of the types such as dry
powder, water and carbon dioxide in only about 3 per cent of the dwelling fires.
Only 27 per cent of these fires were extinguished, with the fire brigades putting
out the remaining 73 per cent.
If a fire is attacked by the occupants with first-aid methods, even if the fire
is not extinguished, the fire brigade can be expected to bring such a fire under
control quicker than a fire which is not attacked by first-aid methods. According
to the analysis of Ramachandran et al. (1972), the mean (average) ‘control time’
of fire brigades was 6.7 minutes for all fires with initial attack by first-aid methods
and 9 minutes for all fires with no initial attack. The mean control time was 6.5
minutes for all ‘sundry means’ and 8.9 minutes for all portable fire extinguishers.
The analysis discussed above did cast some doubts about the effectiveness
of portable fire extinguishers. It is possible that the extinguishers in dwellings
were located at considerable distance from places of fire origin, e.g. in cars or
garages. An analysis by Sime et al. (1981) indicated that people have inadequate
knowledge of the location of extinguishers. According to Chandler (1978), the
success rate of extinguishers used in hospital fires was less than the success rate of
other methods. This conclusion was confirmed by Canter (1985) who suggested
that people (especially staff in hospitals, hotels, etc) should be made aware of the
location of extinguishers and trained in the use and capabilities of different types
and sizes of extinguishers.
It should be pointed out, however, that a number of small fires extinguished by
portable fire extinguishers were not reported to the fire brigade. The percentage of
such fires was 70 per cent according to the UK Fire Extinguishing Trade Association
(FETA) – see Fire Prevention, March 1990. The percentage of fires extinguished
by portable fire extinguishers was 80 per cent in 2002, according to a later (March
2003) report of FETA. This report on a survey into portable fire extinguishers
and their use contains a detailed analysis of the effectiveness of extinguishers in
the UK and in five other EU member countries – Austria, Belgium, Germany,
France and Netherlands. Across these six countries, it was found that in 81.5
per cent of incidents, fires were successfully put out by extinguishers. This report
(March 2003) of FETA provides some guidance and recommendations on the
siting (location), maintenance and the use of different types of extinguishers. Fire
brigades should be called quickly even if an attempt is made by occupants to fight
the fire by fi
rst-aid methods.
8.3.2 Sprinklers
8.3.2.1 Performance
In essence, an automatic sprinkler system is a fire-fighting system designed to be
operated by the fire itself, so as to dispense water in the area where it is needed to
ensure rapid suppression of the fire with minimum damage to property. The salient
feature of the system is an adequate water supply which can be pumped through
a network of pipes, usually at ceiling level, to a series of sensitive ‘sprinkler heads’
which are designed to respond to the thermal conditions created by the fire. Thus,
only those heads which have been affected by the fire will operate and allow water
to flow from them to be distributed in the form of a spray onto the fire below.
Sprinklers are generally required to operate at an average temperature of 68°C but
there are special requirements for certain occupancies.
All sprinkler systems can be categorised as one of four basic types; they differ
in terms of how the water is put into the area of the fire. Wet pipe systems and dry
pipe systems use automatic sprinklers, while deluge systems, instead of automatic
sprinklers, use open sprinklers. The fourth type is similar to a deluge system,
except that automatic sprinklers are used.
Several factors cause uncertainties in the activation and operating times of
sprinklers in actual fires, although deterministic models using experimental data
have been developed to estimate the response times of different types of sprinklers.
Based on factors such as rate of temperature rise, height of upper fire surface above
the floor and height of the premises, Bengtson and Laufke (1979/80) estimated
sprinkler operation times varying from 2.5 minutes for ‘extra high hazard’
occupancies to 16.8 minutes for ‘light hazard’.
According to statistics on actual fires attended by UK fire brigades, in a
sprinklered building, there is a 55 per cent chance that a fire may not produce
sufficient heat to activate the system such that it is either self-extinguished or
extinguished by first-aid methods. In the remaining 45 per cent of fires requiring
sprinkler intervention, the system operates in 87 per cent of cases and does
not operate in 13 per cent of cases. According to an investigation carried out
by the Fire Research Station, UK, quoted by Rogers (1977), one-third of fires
in sprinklered buildings are extinguished by the system and are not reported to
the brigade. Hence, fire brigades attend only two-thirds of fires in sprinklered
buildings such that, on the whole, sprinkler intervention is required in 63 per cent
(= (2/3 × 0.45) + 0.33) of fires. Also, sprinklers operate in 59 per cent (= (2/3
× 0.45 × 0.87) + 0.33) of fires. Sprinklers, therefore, operate in 94 per cent (=
0.59/0.63) of fires in which their action is required.
Rutstein and Cooke (1979) estimated, for various types of occupancies in the
UK, the percentages of fires in which sprinklers operate satisfactorily, which range
from 92 per cent to 97 per cent. Based on data for a hundred years, Marryatt (1988)
estimated for Australia and New Zealand a success rate of over 99 per cent for
sprinklers. The success rate for sprinklers in the USA was about 96 per cent for the
period 1897–1964 according to the National Fire Protection Association (NFPA),
85 per cent for the period 1970–1972 according to the Factory Mutual Research
Corporation (FMRC) and 95 per cent for the period 1966–1970 according to the
US Navy. The figures mentioned above were quoted in a study by Miller (1974)
who also estimated success rates of 86 per cent for wet systems, 83 per cent for dry
systems and 63 per cent for deluge systems.
The reasons for sprinkler failure (non-operation) in the UK were investigated
by Nash and Young (1991). The main causes were shut valves (55 per cent),
system fault (7 per cent) due to problems in design or manufacture and other and
unknown causes (38 per cent). According to the NFPA investigation mentioned
above, out of the 4 per cent of failures, 36 per cent were due to system shut down;
of these, 85 per cent could probably be attributable to human error. Sprinkler stop
valves were shut in one out of every 74 fires, water supply inadequate in one out of
every 276 fires and pipework blocked in one out of every 550 fires.
8.3.2.2 Effectiveness – property protection

The effectiveness of a sprinkler system in controlling fire spread or extinguishing a
fire can be assessed in terms of number of heads operating. According to an analysis
by Baldwin and North (1971), 75 per cent of fires are controlled or extinguished
by four heads or less, 80 per cent by five heads or less and 98 per cent by 35 heads
or less. The corresponding figures for Australia and New Zealand (Marryatt, 1974)
are 90, 92 and 99 per cent respectively. According to Rees (1991), 69 per cent of
fires are controlled by 5 heads or less, 83 per cent by 10 heads or less and 94 per
cent by 25 heads or less; these results are based on data published by the FMRC
for the years 1978–1987. The figures mentioned above demonstrate that only
sufficient sprinkler heads required to control a fire will activate, thus reducing the
amount of water damage and fire loss.
Consequences of water damage and accidental leakage have been used as
arguments against the installation of sprinklers in certain areas, e.g. computer
centres, art galleries and libraries. Case studies of some individual fires reveal that
losses due to water damage were not appreciable and the chance of a leakage
occurring is very small. Additional loss due to water damage is likely to be smaller
than that which would result from further fire spread in the absence of sprinklers.
Effectiveness of sprinklers in reducing property damage has been well
documented, discussed and established in fire protection literature. Some of these
studies were mentioned in Chapter 3 (Tables 3.12, 3.13 and 3.14) in regard to
financial loss and in Section 3.3.1 and Chapter 5 (Table 5.7 and Figures 5.1 and
5.2) in regard to extent of fire spread and area damage. Fire loss data compiled by
FMRC for the period 1980–1989, for a wide range of occupancies, indicate that
the average fire loss for an unsprinklered building is approximately 4½ times the
average loss for an adequately sprinklered building (Rees, 1991). An analysis of
NFPA data for the period 1980–1990 has shown that the reduction in average loss
per fire ranges from 43 per cent for stores and offices to 74 per cent for educational
establishments (Hall, 1992).
Based on Home Office statistics for the years 1981 to 1987, Beever (1991)
found that the probability of fire damage exceeding a given area is very much
smaller with sprinkler protection than without. According to research carried out
by Morgan and Hansell (1984/85), one in ten fires in offices will exceed 16m2 in a
sprinklered building but 47m2 in a building without sprinklers.
As discussed in Section 3.3.2, with reference to the exponential model of fire
growth, sprinklers would reduce the rate of fire growth in a fire developing beyond
the stage of ‘established burning’; this action would increase the ‘doubling time’
of the fire. Sprinklers would also reduce the probability of flashover occurring in a
fire in a compartment – see Section 5.3 and Table 5.6. Melinek (1993a) assessed
the effectiveness of sprinklers in reducing fire severity, expressed in terms of
area damage. He estimated that sprinklers would reduce the probability of fire
size in industrial and commercial buildings reaching 100m2 by a factor of five.
He also found that damage to the structure of a building would be reduced by a
factor of 2.5.
8.3.2.3 Effectiveness – life safety

Although sprinklers have been designed primarily for reducing property damage
in industrial and commercial buildings, they have the potential to reduce life risk
in non-industrial buildings, particularly those with large numbers of people at
risk. Such non-industrial buildings include retail shops and department stores,
office buildings, cinemas, theatres, clubs, pubs, restaurants, hotels, hospitals, flats
and apartment buildings. These buildings may, perhaps, require the installation
of fast response sprinkler systems which operate and detect fires quicker than
those designed for industrial properties. Such systems may provide extra time,
particularly for occupants not in the room of fire origin, to escape to safe places
within or outside the building of fire origin.
Even if a fire is discovered soon after the start of ignition by automatic detection
or sprinkler systems, it may be too late for people in the room of fire origin to
attempt any fire fighting or escape. According to UK fire statistics, most of the
fatal and non-fatal casualties in dwellings were found in the room of fire origin.
Statistics reveal that the fatality rate per fire, for fires discovered at ignition, is
higher than the fatality rate for fires not discovered at ignition, but discovered
within five minutes after the start of ignition – see Ramachandran (1993). Fast-
response domestic sprinklers can, perhaps, save some of the casualties in the room
of fire origin by extinguishing the fire quickly. According to the specifications
of the NFPA, USA, residential sprinklers should act five times quicker than
commercial and industrial sprinklers.
Sufficient statistical data are not available for estimating the number
of lives that could be saved by installing sprinklers in large non-industrial
buildings. However, tentative results are available from some statistical studies.
Comprehensive data for Australia and New Zealand for a 100-year period up
to 1986 show that, during that period, there were 11 deaths in 9,022 fires in
sprinklered buildings (Marryatt, 1988). This represents a fatality rate of 0.0012
per fire. According to Hall (1992), if sprinklers are installed in one- and two-
family houses in the USA, 63 to 69 per cent reduction in death rates per thousand
fires can be achieved. Under some assumptions Melinek (1993b) estimated that,
if all the buildings in the UK were sprinklered, the number of fatal casualties
in fires would be reduced by half and the number of non-fatal casualties by
about 20 per cent. He has also shown that sprinklers can significantly reduce the
number of multi-casualty fires.
Ramachandran (1993) attempted to estimate the probable reduction in life
risk if sprinklers are installed in all single and multiple occupancy dwellings in the
UK. For this analysis, he expanded the regression model in Equation (3.46), by
considering also the parameter K given by
K = λ (B + E – F) (8.1)
where, as defined in Section 3.7.2.2, B is the ‘recognition time’ and E the ‘design
evacuation time’. The time period F is the time taken by a combustion product,
e.g. smoke, to travel from the place of fire origin and produce an untenable (lethal)
condition on an escape route.
By retarding the rate of fire growth, sprinklers would increase the value of F in
addition to decreasing the value of D in Equation (3.46) by acting as an automatic
detection system. Ramachandran assumed that sprinklers would reduce the
average value of the discovery time D to 3 minutes from 15.5 minutes with a saving
of 12.5 minutes and increase the average value of F by 4 minutes. He also assumed
that sprinklers will not affect the values of B and E. Under the above assumptions,
with a total saving of 16.5 minutes and λ = 0.0007, applying Equations (3.46) and
(8.1), Ramachandran estimated that sprinklers would reduce the fatality rate per
fire in dwellings to 0.0009 from the level of 0.0124 for unsprinklered dwellings
revealed by fire statistics. The fatality rate of 0.0009 for sprinklered dwellings is
not significantly different from the rate of 0.0012 estimated by Marryatt (1988) for
all sprinklered buildings.
In a report (unpublished) to the Fire Research Station, UK, Ramachandran
(1999) investigated the effectiveness of sprinklers in reducing life risk in non-
industrial buildings. According to his results, reproduced in Table 8.3, sprinklers
would reduce the probability of flashover in fires in the five occupancies considered
Table 8.3 Effectiveness of sprinklers – non-industrial buildings
Probability of confinement to Probability of flashover Probability of spreading beyond Average area damage
item first ignited room
Occupancy (m2)
type Sprinklered Unsprinklered Sprinklered Unsprinklered Sprinklered Unsprinklered Sprinklered Unsprinklered
Office buildings 0.58 0.37 0.25 0.53 0.03 0.15 5.69 13.22
Hotels 0.85 0.30 0.07 0.50 Less than 1% 0.14 2.75 21.29
Pubs, clubs 0.59 0.26 0.26 0.62 0.07 0.17 8.95 23.98
restaurants
Hospitals 0.87 0.54 0.12 0.30 0.06 0.04 0.57 7.59
Flats 0.90 0.53 0.04 0.37 0.01 0.07 1.85 2.91
by a factor of two or more. Consequently, sprinklers considerably reduce the
probability of a fire in these buildings (except hospitals) spreading beyond the
room of origin. Sprinklers also significantly reduce the average area expected to be
damaged in fires in non-industrial buildings.
Some communities in the USA have adopted ordinances which promote the
use of residential sprinklers. The leading ones are Scottsdale, Arizona; Prince
George’s and Montgomery Counties, Maryland; Greenburgh, New York; and
Cobb County, Georgia. According to a recent investigation (Butry et al. 2007)
in benefit-cost analysis of residential sprinkler systems, in terms of fire-risk
mitigation, multipurpose network systems achieve greater cost-effectiveness over
alternative systems.
8.3.2.4 Reliability
As discussed in Section 8.3.2.1 there is a small chance that, for various reasons, a
sprinkler system may not operate when it is required to act in an actual fire. This
probability for failing to operate ‘on demand’ can vary from 15 per cent to 1 per
cent depending on the type of the system and the manufacturer. Combining all the
data discussed in Section 8.3.2.1, the following estimates appear to be reasonable
for conditional probabilities (percentages) if failure occurs:
• 55 per cent due to system shut off for maintenance;

• 17.5 per cent due to shut sprinkler stop valves;
• 7.5 per cent due to defective system;
• 5 per cent due to inadequate water supply;
• 2.5 percent due to blocked pipe work; and
• 12.5 per cent due to other causes.
Consider now, as an example, a reliability of 0.92 for sprinkler operation when

a fire occurs. The unreliability or failure probability is 0.08. If, on average, 200 fires
occur every year in a group of buildings with 2000 sprinkler systems, the overall
annual failure rate per system is likely to be 0.008 (= (200/2000) × 0.08). The
mean time between failures (MTBF) for each system is hence, approximately, 125
years (= 1/0.008). Based on the conditional probabilities estimated earlier, the
annual failure rates per system for different causes are likely to be:
• 0.0044 due to system shut off for maintenance;

• 0.0014 due to shut sprinkler stop valves;
• 0.0006 due to defective system;
• 0.0004 due to inadequate water supply;
• 0.0002 due to blocked pipe; and
• 0.001 due to other causes.
The reciprocals of these annual failure rates are the corresponding MTBFs in
years.
The probabilities discussed above provide some ‘global’ estimates for the
reliability of a sprinkler system installed in a group of buildings. This reliability
can be improved by controlling (reducing) or eliminating the unreliabilities
associated with some or all the causes of failure. The system reliability of
sprinklers installed in a particular building within the group would, however,
depend on the reliabilities for the different component parts of the system.
The main components are sprinkler heads (considered as a group), pipe work
and water supply from public mains or elevated private reservoirs, gravity
tanks etc. The reliabilities (success probabilities) for these three components
may be denoted by RH, Rp and Rw respectively. One minus the reliability is the
unreliability or probability of failure.
The failure of any of the three main components would result in the system
failing to operate when a fire occurs. For such a ‘series’ arrangement, according to
reliability theory, the system reliability, Rs, is given by the product
Rs = RH . Rp . Rw (8.2)
A group of sprinkler heads located in a specific area may be considered to be

in ‘parallel’ arrangement, since the failure of one head to operate will not lead to
system failure; another head located close to the failed head may operate. Hence,
the reliability, RH, of a group of sprinkler heads may be estimated by
RH = 1 – (1 – RH1) (1 – RH2) (1 – RH3) … (8.3)
where RH1, RH2, RH3, … are the reliabilities of individual heads. Equation
(8.3) follows from the reliability theorem that, for a parallel arrangement, the
unreliability of a system is the product of the unreliabilities of the components.
It should be noted, however, that failure of one or more heads would reduce the
effectiveness of the system in controlling fire spread.
The model discussed above is a simple framework for evaluating the reliability
of a sprinkler system for a particular building, based on the reliabilities of its
components. This framework needs to be expanded to consider other components
and sub-components such as water pressure. This is a complex problem on which
practically no research has been carried out so far.
Satisfactory operation to release water is the primary function of a sprinkler
system. Raising an alarm is the secondary function. The alarm may not sound in
the event of a fire but this will not affect the primary function. Due to some causes,
a sprinkler head may operate in a non-fire situation, raise an alarm and discharge
water until the time the sprinkler stop valve is shut. Such a ‘false’ operation of
sprinklers is a rare occurrence.
Blockages of a sprinkler system can occur due to debris entering the pipe work.
This problem applies mainly to systems being installed in new buildings where
sections of the system are erected during the early stages of construction when
building materials and debris are widespread on the site. Poor maintenance can
cause the failure of a sprinkler system to operate when a fire occurs.
References
Appleby, D and Ellwood, SH (1989), A fire detection system using distributed processing,
Proceedings of the 9th International Conference on Automatic Fire Detection, NIST,
Gaithersburg, MD
Baldwin, R and North, M A (1971), The Number of Sprinkler Heads Opening in Fires, Fire
Beever, P (1991), How fire safety engineering can improve safety, Institution of Fire
Engineers Meeting, Torquay, UK, October.
Bellamy, L (1989), Informative fire warning systems: a study of their effectiveness, Fire
Surveyor, August, 17–23.
Bengtson, S and Laufke, H (1979/80), Methods of estimation of fire frequences, personal
Bukowski, R W et al. (1987), Hazard 1 Vol 1. Fire Hazard Assessment Method. Report NBSIR
87-3602, National Bureau of Standards, Gaithersburg, MD.
Bukowski, R W and Istvan, S M (1980), A Survey of Field Experience with Smoke Detectors
in Health Facilities, NBSIR 80-2130, National Bureau of Standards, Gaithesburg, MD.
Butry D T, Brown M H, and Fuller S K (2007), Benefit-Cost Analysis of Residential Fire
Sprinkler Systems, Report NSTIR 7451, National Institute of Standards and Technology,
Office of Applied Economics, Building and Fire Research Laboratory, Gaithersburg,
MD.
Cafolla, D (1997), The impact of unwanted fire alarms on the provision of fire safety in
healthcare facilities, Dissertation for Master of Science in Fire Safety Engineering,
University of Ulster.
Canter, D (1985), Studies of Human Behaviour in Fires: Empirical Results and Their Implications
for Education and Design. Fire Research Station, Borehamwood.
Canter, D, Powell, J and Booker, K (1988), Psychological Aspects of Informative Fire Warning
Systems, Fire Research Station, Borehamwood.
Chandler, S E (1978), Some Trends in Hospital Fire Statistics, Current Paper CP 67/78,
Building Research Establishment, Watford.
Cholin, R R (1975), Reappraising early warning detection, Fire Journal, 69, 2, 54–58.
Custer, R L P and Bright, R G (1974), Fire Detection: The State of the Art, Technical Note
839, National Bureau of Standards, Gaithersburg, MD.
Davies, D (1984), Means of cutting down false alarms in automatic systems, Fire, 77, 9–14.
Finucane, M and Pinkney, D (1988), Reliability of Fire Protection and Detection Systems,
Report SRD R431. Safety and Reliability Directorate, Atomic Energy Authority,
Warrington.
Fry, J F and Eveleigh, C (1975), The Behaviour of Automatic Fire Detection Systems, Current
Paper CP 32/75, Building Research Establishment, Watford.
Gupta, Y P (1984/85), Automatic fire detection systems: aspects of reliability, capability
and selection criteria, Fire Safety Journal, 8, 105–117.
Hall, J R (1992), US Experience with Sprinklers: Who Has Them, How Well Do They Work,
National Fire Protection Association, Quincy, MA.
Helzer, S G, Buchbinder, B and Offensend, F L (1979), Decision Analysis of Strategies for
Reducing Upholstered Furniture Fire Losses, NBS Technical Note 1101. National Bureau
of Standards, Gaithersburg, MD.
Japanese Association of Fire Science and Engineering (1982). Bulletin, 32, 6.
Marryatt, H W (1974). The Australian experience with automatic sprinklers. Australian
Fire Protection Association Journal, September, 2–7.
Marryatt, H W (1988), Fire: A Century of Automatic Sprinkler Protection in Australia and
New Zealand, 1886–1986, Australian Fire Protection Association, Boxhill, Victoria.
Melinek, S J (1993a), Effectiveness of sprinklers in reducing fire severity, Fire Safety Journal,
21, 299–311.
Melinek, S J (1993b), Potential value of sprinklers in reducing fire casualties, Fire Safety
Journal, 21, 275–287.
Military Standardization MIL-HDBK.217B (1974), Reliability Prediction of Electronic
Equipment, Naval Publications and Forms Centre, Philadelphia, PA.
Miller, M J (1974), Reliability of fire protection systems, Chemical Engineering Progress, 70,
4, 62–67.
Morgan, H P and Hansell, G O (1984/85), Fire sizes and sprinkler effectiveness in offices:
implications for smoke control design, Fire Safety Journal, 8, 187–198.
Nash, P and Young, R A (1991), Automatic Sprinkler Systems for Fire Protection, Paramount
Publishing, London.
National Health Service (NHS) Estates (1996), Fire Practice Note 9 – NHS Healthcare Fire
Statistics, Her Majesty’s Stationery Office, London.
Peacock, S T and Sutcliffe, N (1982). Report on False Alarms for Fire Alarm Systems in the
Lincolnshire Area Health Authority, University of Bradford, Bradford.
Pearce, N (1986), Fire alarm systems in healthcare premises, Fire Surveyor, April.
Pigott, B B (1983), Computer-based analogue fire detection systems, Fire Engineers Journal,
43, 128, 23–26.
in the United Kingdom, Fire Safety: Science and Engineering. ASTM STP 882,
American Society for Testing and Materials, Philadelphia, PA.
Ramachandran, G (1990), Human behaviour in fires, Fire Technology, 26, 2, 149–155.
Ramachandran, G (1991), Informative fire warning systems, Fire Technology, 27, 1, 66–81.
Ramachandran, G (1992), Statistically-Determined Fire Growth Rates for a Range of Scenarios:
Part 1: An Analysis of Summary Data. Part 2: Effectiveness of Fire Protection Measures –
Probabilistic Evaluation, unpublished report to the Fire Research Station, Borehamwood.
Ramachandran, G (1993), Early detection of fire and life risk, Fire Engineers Journal,
December, 33–37.
Ramachandran, G (1999), Reliability and Effectiveness of Sprinkler Systems for Life Safety,
unpublished report to the Fire Research Station, Borehamwood.
Ramachandran, G, Nash, P and Benson, S P (1972), The Use of Fire Extinguishers in
Dwellings, Fire Research Note 915, Fire Research Station, Borehamwood.
Rees, G (1991), Automatic sprinklers; their value and latest developments, Fire Surveyor,
October, 9–13.
Rogers, F E (1977), Fire Losses and the Effect of Sprinkler Protection of Buildings in a Variety of
Industries and Trades, Current Paper CP 9/77, Building Research Establishment, Watford.
Rutein, R and Cooke, R A (1979), The Value of Fire Protection in Buildings, Report 16/78.
Scientific Advisory Branch, Home Office, London.
Sime, J, Canter D and Breaux, J (1981), University team studies use and success of
extinguishers, Fire, 73, 909, 509–512.
Tong, D A (1983), Human response to fire alarms, Fire Surveyor, 12, 4, 4–8.
Tong, D and Canter, D (1985), Informative warnings: in situ evaluations of fire alarms, Fire
Safety Journal, 9, 267–279.
US Consumer Product Safety Commission (1993), Smoke Detector Operability Survey:
Report on Findings, US Government Printing Office, Washington, DC.
9 Performance and reliability of human
response and evacuation
9.1 Recognition
As defined in Section 3.7.2.2, the total time taken by an occupant or group of
occupants, say, on a particular floor of a building to reach a safe place within
or outside the building is the sum of three time periods. In sequential order, the
first period is the time taken to discover a fire by occupants or detect the fire by
automatic detection systems after the start of the fire. The second period is the
‘recognition time’ or ‘gathering phase’ which has been discussed in several research
studies concerned with human behaviour in fires. The third is the evacuation
time (Section 9.3) relating to the period from the commencement of evacuation
to reaching a safe place. The actions of the occupants during the three periods
mentioned above would depend, to some extent, on whether they are in the room
of fire origin or in some other room – see Figures 9.1 and 9.2 reproduced from
Ramachandran (1993a).
Consider the second period which is the subject matter of this section. This
period is the elapsed time from detection or discovering fire (Figure 9.1) or receiving
fire information (Figure 9.2) to commencing evacuation (Figures 9.1 and 9.2). It
is not possible for occupants to put into motion any actions to cope with the
fire until someone has identified that the fire is present and has acted upon that
identification by informing others. As pointed out in several studies on human
behaviour in fires, the early stage of fire recognition is typically characterised by
ambiguity – see Canter (1980). It is clear that early acceptance of the fact that
the unusual circumstances present constitute a fire of some severity is frequently
delayed to a dangerous extent. Once a fire has been recognised as such, there
is then the possibility for a range of actions, including first-aid fire fighting and
commencement of evacuation (Section 9.3).
Statistics on fires attended by fire brigades provide some information on the first
period of total evacuation time, the fire discovery time by occupants. Laboratory
experiments with different types of automatic detection systems can provide
estimates for this period, if any such systems have been installed in a building. But
neither statistics on fires nor experiments can provide any estimates for the second
period, recognition time, for any type of occupancy. Some data for this period for
a few occupancy types can be obtained from case studies discussed in the papers of
Working
Discussing Pre-ignition activities
Eating
Detection or discovering fire
Fire Brigade Communications to Other occupants
First aid
Decide on action
Fire fighting
Casualty No casualty
Prepare for evacuation
Commence evacuation
Route Escape Escape Route

blocked route route blocked
Retreat Retreat
Reach Reach
place of place of
Casualty No casualty safety safety Casualty No casualty
Figure 9.1 Occupants’ response – room of fire origin
some authors on human behaviour in fires – see Canter (1980). For several types
of occupancies, on average, the recognition time is likely to be about 2 minutes –
see Canter (1980) and Ramachandran (1993a, 1993b).
To collect satisfactory data on recognition time it would be necessary to
carry out evacuation exercises or computer simulated laboratory experiments
with ‘subjects’ (members of the public) as discussed with reference to the results
produced in Tables 8.1 and 8.2.
Sime (1991) defined recognition time as ‘time to start to move’ from the onset
of an alarm or discovery of a fire by someone in a building or device such as a
smoke alarm. A delay in this time exceeding two minutes is a fundamental problem
which characterises large-scale fire disasters involving injury, including those at
the Bradford City Football Club ground (1985), Kings Cross Underground Station
(1987), Isle of Man Summerland Recreation Complex (1973), Kentucky Beverley
Hills Supper Club (1977) and Woolworths (1979).
9.2 Response
Reviewing research in the UK on the human aspect of fires in buildings,
Ramachandran (1985) discussed occupants’ response to learning about a fire. As
pointed out by him, initial response may be triggered by four senses: sight, hearing,
Performance and reliability of human response and evacuation 297
Working
Pre-fire information activities Discussing
Eating
Occupant
Receive fire information Ignore
Other sources
e.g. alarm
Investigate Withdraw
and wait
Prepare for evacuation
Commence evacuation
Route Escape Escape Route

blocked route route blocked
Retreat Retreat
Reach Reach
place of place of
Casualty No casualty safety safety Casualty No casualty
Figure 9.2 Occupants’ response – room other than room of fire origin
smell and touch. Studies of actual fires suggest that sight is the most common,
followed by sound, with smell and touch much less often. Visual detection of
smoke, and less frequently of flames, predominates in domestic fires. Several
group-residential buildings in which fires were detected by sight had audible
alarm systems installed and in others flames rather than smoke were seen. All
group-residential buildings in which fires were heard had audible alarm systems,
though in one case the sound of a bang preceded the sounding of the alarm bell.
Sometimes, the alarm bell is ignored due to the ‘normality’ of false alarm sounding.
In a domestic fire detected by smell, a sleeping occupant may be awakened by a
choking sensation from smoke emanating from downstairs.
Responses of those discovering a fire independently of the first person,
simultaneously or after, are less well documented. The discovery is usually by sight
or sound as it is if knowledge of the fire is received via the initial respondent. A
noise stemming from a fire may be ignored if it is thought to be attributable to
some other source, e.g. slamming the door.
In general terms, occupants of residential buildings respond to real fire situations
by one or more of the following actions:
• seeking to confirm the existence of the fire;

• fighting the fi
re;
• telling others about the fire;
• escaping from the fire; and
• assisting others to e scape.
Their choice of these reactions or the priority they place on each varies
according to their perception of danger, responsibility and location. In cases
where an individual is incapacitated mentally, physically, by sleep or by alcohol,
the ability to respond normally is often dulled or frustrated as a consequence of
impaired s enses.
If a fire occurs in a hotel, the initial human response is usually to seek it
out. Hotel staff (particularly management) seem more likely to carry out this
investigation than their guests. In domestic fires it is the man who confronts the
fire and delegates the job of phoning the fire brigade to his wife or others. A
nursing auxiliary’s response to seeing a glow in an occupied ward of a hospital is
usually to enter the ward and attempt to remove the blazing object, e.g. a c hair.
People do not often fight a fire immediately, perhaps because of more pressing
priorities such as calling the fire brigade and commencing evacuation. Those
whose initial response is to fight the fire are in the minority. Hotel staff and those
with similar responsibilities in halls of residence operate fire extinguishers only as a
secondary response. Where domestic fires evoked the fire-fighting response, it was
usually an immediate one, using a garden hose or water supply with easy access
rather than an extinguisher which, if available, may be located at a distance, say,
in the g arage.
Escaping from the fire seems to be the most common response. Generally, those
threatened by a fire elsewhere seem more willing to stay in their room and undergo
the risk of leaping from windows. In domestic fires, some occupants insist on
returning to their bedrooms to retrieve their cherished belongings, even though
such an action would put them at risk of inhaling smoke and dying eventually.
Mentally handicapped people and people who have been drinking heavily do
not respond positively and quickly to escaping from a fire and hence are likely to
sustain fatal or non-fatal injuries. In some hotel fires, guests following a signalled
escape route faltered at a bolted emergency exit, unable to work out how to
operate the lock.
Telling others about a fire can be for the purpose of warning potential victims
or informing the fire brigade directly or by delegation. Warning potential victims
can be from a neighbour or passer-by who noticed smoke coming from a w indow.
Those with responsibility for people in normal day-to-day life seem to apply
that role equally in fire situations to assisting them to escape. People with such a
responsibility include duty staff in a hospital, the husband of a bed-bound woman
and the staff of a h
otel.
9.3 Evacuation
Once an occupant or group of occupants of a building recognise the existence of a
fire, they will decide to commence evacuation and escape to a safe place within or
outside the building. This course of action has to be achieved before the available
escape routes become untenable due to the build-up of excessive or lethal levels
of heat, smoke or toxic gases. The occupants should move quickly to the safe
place from the places in the building where they were located at the time when
the existence of the fire was recognised. This period of occupants’ movements
is the evacuation time defined in Section 9.1 and Section 3.7.2.2 as the third
component period of the total evacuation time. For a successful evacuation, the
total time, which is the sum of discovery or detection time, recognition time and
evacuation time, should be less than the time taken by a combustion product, e.g.
smoke, to travel from the place of fire origin and produce an untenable condition
on the escape r oute.
The mobility of occupants during evacuation depends on a wide range of
ambulancy from mobile to totally immobile. Considering a hospital, for example,
some patients may only need minimum assistance but the non-ambulant may
require maximum assistance in preparing for evacuation and during evacuation.
Factors contributing to non-ambulancy include physiological deficiencies, whether
through mental capability limitations or temporary reduction of ability because
of sleep, drugs or alcohol. A factor enhancing the escape potential is training,
through fire drills, in the use of first-aid fire fighting devices such as portable fire
extinguishers and in following safe escape routes. Training and publicity through
sign boards are particularly important if routes to be used normally are blocked
temporarily or permanently for some reason.
Evacuation of disabled people from fire has been the subject of intensive code
activities and research during the past three decades.
National concern for life safety of people with disabilities was first marked
by a seminar, ‘Fire Safety for the Handicapped’ held in Edinburgh in 1975 – see
Marchant (1975). The 1985 edition of the Life Safety Code of the National Fire
Protection Association, USA, focused on life safety for people with disabilities in
situations where many such people were found and where there was a record of
serious life loss from fire. Special requirements for disabled people were introduced
in the 1985 edition of the National Building Code of Canada. Based on some
substantial early effort of the Home Office, the British Standards Institution, in
1988, issued BS 5588, Part 8, Code of Practice for Means of Escape for Disabled
Persons, which included extensive guidance on the use of elevators (lifts) for
egress during fires.
Reviewing the above-mentioned code activities concerned with the evacuation
of disabled people, Pauls and Juillet (1993) raised significant questions about the
adequacy of building codes and fire codes which existed at that time to deal with
the life safety of people with disabilities. The authors concluded that the codes
were deficient in relation not only to people with disabilities but also to all building
users. Other important research studies on the topic of evacuation of disabled
people include those of Dunlop (1993), Rubadiri et al. (1993), Shields et al. (1996,
1998), Walsh (1998) and Yoshimura (1998).
While fire is a major threat to occupants in its immediate vicinity, it is generally
smoke and toxic gases which pose the greatest threat to occupants who are
remote from the fire. Hence, as suggested in Figures 9.1 and 9.2, the following
two categories should be distinguished for evaluating the evacuation success of
occupants of a building:
1 occupants in the room of fire origin;

2 occupants in other rooms.
The second category can be further sub-divided into
1 occupants in other rooms in the floor of fire origin;

2 occupants on other floors.
Occupants on floors above the floor of fire origin are at greater risk than those
on floors below the fire floor.
Occupants in the room of fire origin may become aware of the fire immediately
and directly after it started, whereas those in other rooms may receive information
about the fire either directly from those in the room of fire origin or indirectly
through fire alarm or other communication systems. If the room of fire origin
is unoccupied or unprotected by automatic fire detectors or sprinklers, the fire
discovery time will be long and highly variable. If the room is occupied, someone
in the room will discover the fire, perhaps, earlier than a detector system, if it has
been installed. Fire detectors are only designed to reduce the delay in discovering
a fire, particularly in an unoccupied area. On the other hand, computer based
informative fire warning systems (Section 8.1.2) are capable of reducing the delays
in both discovery time and recognition time.
As discussed above, for evaluating the evacuation or egress success of occupants
of a building involved in a fire, it is important to consider the location of the
occupants in relation to the location of the fire, mobility of the occupants, training
through fire drills, delay in response or recognition and whether effective detection
and other communication systems and suppression systems such as sprinklers have
been installed or not. The behaviour of occupants will be significantly influenced
by whether they are alone or with a group. Generally, group reactions will exhibit
a greater inertia than individuals. Other factors include the size, density and
distribution of occupants. People with mobility handicaps should have their work
spaces on the ground floor, if possible.
The last, but not the least, important factor affecting escape potential is the
escape system designed for a building including the number, location and capacity
of the escape route network and continuing ease of use of the routes throughout the
duration of the fire. The clarity and simplicity of the egress system and dead-end
corridors could also affect the number of persons ultimately escaping. Uncertainty
in the successful evacuation of a building can also be caused by a variety of escape
routes available to the occupants and the need to choose the safest r oute.
As fire, smoke and toxic products continue to spread, the number of routes
available for escaping will diminish and vary with each group of the escaping
population. Continuing degradation of the environment and reduction in
the choice of escape routes could change the mode of escape for groups of the
population not yet evacuated. Hence, in parallel to the movement of people
involved in evacuation, it is necessary to analyse the uncertainties in the
development of fire and combustion products.
9.4 Design evacuation time

It is necessary to evaluate the evacuation time for determining an appropriate
design for a given building with a specified number of occupants and their mobility
characteristics. The design parameters include number of floors, number of rooms
on each floor and number, location and capacity (widths etc) of the escape routes.
The values of these parameter depend on the design value estimated for the
evacuation t ime.
The time it takes an occupant to move from their starting point to a location of
safety is simply a function of travel speed and distance.
Travel time (s) = distance (m) / travel speed (m/s)
Distance is a function of the escape route selected by the occupant. This
selection process is affected by the occupant’s familiarity with the building layout,
the availability of the escape route, segments (corridors, staircases etc) on the
escape route and the degree of difficulty in moving through each s egment.
Travel speed is affected by the occupant’s mobility and by the mobility of any
accompanying occupants; by crowding, light levels, presence of smoke, quality of
the floor and wall surfaces (e.g. roughness, unevenness etc), width and riser height
of stairs, width of doorways and corridors etc. The travel speed is also affected by
training through fire drills.
Some of the above-mentioned factors are occupant characteristics; others
are building characteristics. The designer or engineer must deal with each factor
explicitly or be able to justify why a factor is not relevant to the a nalysis.
A range of values need to be calculated to predict the movement component of
total evacuation time. These include clearing time for each segment of the escape
route, time for a person to travel the longest (most remote) route and location of
individuals so that exposure levels can be calculated. The calculation method chosen
will depend to a great extent on the values needed for the evaluation of a design.
To estimate the evacuation time, information is needed on the following people
movement characteristics:
• Speed: rate of travel along a corridor, ramp, stair;

• Floor: number of persons passing through a particular segment of the escape
route per unit of time (e.g. persons/sec passing through a doorway or corridor);
• Specific flow: flow per unit of width of the egress segment (e.g. persons/sec/m
of corridor width).
Most of the information on people movement has been collected in fire d rills.
The parameters have been investigated in many research studies for the
movement of able-bodied and disabled people on stairs, in corridors and through
doorways. Calculation methods have been suggested for estimating the evacuation
times for specific types of buildings – see, for example, Pauls (1987) for tall
buildings. Watts (1987) discussed computer models for evacuation a nalysis.
Speed has been shown to be a function of the density of the occupant flow
type of egress component and mobility capabilities of the individual (Nelson and
Mowrer, 2002).
For a density greater than 0.55 pers/m2:
υ = k – a k D (9.1)
where
υ = speed (m/s)
a = constant, 0.266m2/pers
k = velocity factor (m/s)
D = density of occupant flow (pers/m2).
For a density less than 0.55 pers/m2, too few other people are present to impede
the walking speed of an individual. In this case, maximum walking velocities for
level walkways and stairways are
υ = 0.85k (9.2)
The velocity factor, k, varies from 1.00 m/s to 1.40 m/s.

At lower densities, people have a greater freedom to move at their own pace.
As the crowd density increases, they become more controlled by others in the
moving stream. At a density of about 1.88 persons per m2, the combination of
closeness of individuals and speed of movement are indicated to be the maximum.
Such a high density is not comfortable. Higher densities not only slow the flow
but also can reduce movement to a shuffling gait and, in the extreme, a crushing
condition. People movement is faster for level segments of an escape route than
stairs. It is faster for stairs with lower risers and deeper treads than stairs with
higher risers and shallower t reads.
The speed correlations presented in Equations (9.1) and (9.2) principally relate
to adult, mobile individuals. Proulx (1995) indicates that the mean speed on stairs
for children under 6 and the elderly was approximately 0.45 m/s in unannounced
drills in multi-storey apartment buildings. The speed for an ‘encumbered’ adult
is 0.22–0.79 m/s, also appreciably less than the maximum speed expressed in
Equation (9.2). (An encumbered adult is an individual carrying packages, luggage
or a child.)
Mean velocities for impaired individuals were estimated by Shields et al. (1996).
These velocities depend on whether an individual uses an electric wheelchair,
manual wheelchair, crutches, walking stick, walking frame/walker or rollator. For
level walkways, the mean speed ranges from 0.51 m/s for walking frame/walker to
0.94 m/s for crutches. For stairs-down, the mean speed is 0.22 m/s for crutches and
0.32 m/s for walking stick. For stairs-up, the mean speed is 0.22 m/s for crutches
and 0.34 m/s for walking stick. For people with no disability, the means speeds are
1.24 m/s for level walkways and 0.70 m/s for stairs, up or d
own.
Typical densities of people range from 1.0 to 2.0 persons per m2 – see Fruin
(1987), Pauls (1995), Predtechenskii and Milinskii (1978) and Frantzich (1996).
The specific flow is the product of the density and the s peed:
Fs = Dυ (9.3)
where Fs is the specific flow, persons/s/m. Expressions for Fs as a function of density

can be obtained by substituting in Equation (9.3), the relationship for speed, υ,
given in Equation (9.1) or (9.2).
The width referenced in the units for the specific flow, Equation (9.3), relates
to the ‘effective width’ as defined by Pauls (1995). The concept of effective width
is based on the observation that people do not generally occupy the entire width
of an egress segment, staying a small distance away from the walls or edge of the
segment. Nelson and Mowrer (2002) refer to this small distance as a ‘boundary
layer’.
Maximum specific flow is achieved at a density of
Dmax = ½ a (9.4)
where a is the constant defined with reference to Equation (9.1). Because a is

independent of the type of egress segment, according to this correlation, the specific
flow is maximised at the same density for all types of segments. Predtechenskii
and Milinskii (1978) provide results from their data indicating differences in
the density at which the specific flow is maximised for different types of egress
segments or components.
The ‘specific flow’ provides a measurement of the flow capability of an egress
component on a per unit basis (e.g. per metre). In an evaluation of an egress
component or multiple egress components the total flow can be calculated and
related to the affected population:
Flow capacity = Fc = Fs . We (9.5)
where Fs is given by Equation (9.3) and We is the effective width. The evacuation
time for a populated area through one exit element is the population, P, divided
by the flow capacity of the exit element, plus the travel time through the exit
element.
The method (Equations (9.1) to (9.5)) described above was used by Nelson
and Mowrer (2002) to obtain a first order approximation of the egress time in
buildings. The method involves determining the maximum flow rate for each of
the egress components in the egress system. Proulx (1995) described in detail
building evacuation models, including simple first-approximation methods and
movement assumptions used in these calculations. She highlighted the effective-
width model for evacuation flow, especially in relation to prediction formulae for
the total evacuation time of large buildings.
Melinek and Booth (1975) carried out an analysis of evacuation times and of
the movement of crowds in buildings. Using published data on crowd flows along
corridors, the authors estimated that the normal capacity of corridors was about
1.5 people per metre per second; this value can be exceeded for a large rush of
people. At low crowd densities (under one person per square metre), the speed
of individuals is equal to their free walking velocity. The flow rate is a maximum
when the crowd density is between one and five persons per square metre. For
densities greater than five persons per square metre, congestion exists and the flow
rate decreases as the crowd density increases.
Also, for movement up or down staircases or both, as the crowd density
increases, the flow reaches a maximum and then decreases. There is no reduction
of flow for bends in the staircase. The capacity of a staircase decreases if a large
distance has to be traversed and reduces markedly by two-way flow, which should
be avoided where possible. Velocities of 1.3m per second along corridors and 0.5m
per second on stairs (along the line of stairs) can be assumed for unimpeded flow.
Free flow at the head of the staircase can be assisted by a suitable system of barriers.
For a multi-storey building with n floors, Melinek and Booth (1975) derived the
following formula for the minimum total evacuation time, Tr, for the population
of floor r and a bove:
 n 
r  (
T =  ∑ Q  / N ′ b
i  )
r −1
+ rt
s
(9.6)
i=r
where
Qi = population of floor i.
br = staircase width between floor r – 1 and r.
N´ = rate of flow of people per unit width down the stairs
ts = time for a member of an unimpeded crowd to descend one s torey.
The minimum total evacuation time for the whole building is equal to the
highest of the n values of Tr (r = 1 to n). If there are several staircases, br will be
the total width of the staircases considered. Using the formula in Equation (9.6),
evacuation times were calculated assuming that the populations and staircase
widths were the same for all floors and N´ = 1.1 persons per metre per second.
In most of the eleven office buildings considered in the study, the observed
evacuation times were greater than the calculated values, the average difference
being about 2 minutes. The calculated times ranged from 1.8 min for a six-floor
building to 9.9 min for a 25-floor building.
In a multi-storey building designed for ‘total evacuation’, the escape stairs should
have the capacity to allow the whole building to be evacuated simultaneously. For
large buildings, over 30m high, complete evacuation would be lengthy and difficult
and, hence, the fire safety codes envisage evacuation of only part of the building,
usually the fire floor and the floor above. This is referred to as ‘phased evacuation’
which is applicable to a large building in which every floor is a floor with fire resistant
compartments and which is fitted with an appropriate fire warning system.
According to means of escape provisions in fire safety codes, building occupants
should be able to reach the entrance of an escape route, such as a protected
staircase, within a reasonable period of time in the event of a fire. This period
known as the ‘design evacuation time’, is used to determine the number and
widths of escape routes required to facilitate an unimpeded flow of people. For this
purpose, the following formula has been suggested in British Standard BS 5588:
M = 200b + 50 (b – 0.3) (n – 1) (9.7)
where M is the maximum number of people who can enter a staircase of width b
(in metres) serving n floors.
Equation (9.7) was derived by substituting the value of u given b y
u = 50 (b – 0.3) (9.8)
in the following equation which assumes a uniform population d

ensity:
M = N b t + (n – 1) u (9.9)
where the flow rate N is approximately 80 persons per metre width per minute. The
parameter, u, is the number of people who can be accommodated on the stairway
between one storey and the next. The parameter t is the design evacuation time,
usually taken as 2.5 minutes, which is considered to be an acceptable period within
which all the occupants of a floor should be able to enter a protected staircase
after leaving their places of occupation. For buildings with two or more staircases,
it is assumed that any one staircase may be unusable in a fi
re.
For buildings without compartmental floors between storeys, the maximum
number of people who can enter a staircase is recommended to be 222b. For a
building designed for phased evacuation, British fire safety codes recommend that
the minimum width of a stair should be based on the formula [(P × 10) – 100]
mm, where P is equal to the number of people on the most heavily occupied storey.
For the parameter u, the following general formula can be u sed:
u = ( 5.2b + 4b2 )ρ (9.10)
where ρ is the density of people per m2 in the staircase, a calculation which may
be determined with due regard to the characteristics of the evacuating population
and the presence of active fire protection systems such as detectors and sprinklers.
Melinek and Booth (1975) assumed that ρ = 3.5 but Fruin (1971) suggested a
lower density of 1.5 to avoid psychological discomfort and to minimise the risk of
panic over prolonged delays. The absence of a b2 term in Equation (9.7) implies
that for wide staircase widths it makes less allowance for the standing capacity of
the landing.
In sequential order, the design evacuation time, t defined in Equation (9.9), is
the third component period of the total evacuation time, H, defined in Section
3.7.2.2. It is inappropriate to assign for E an arbitrary value such as 2.5 minutes
without taking into consideration the first two component periods, discovery or
detection time D and recognition time B. In determining the value of E, it is also
necessary to consider the duration of two other phases of evacuation movement
of occupants – movement for down staircases and movement through exits. The
evacuation time E is the sum of the three phases of occupants, movement – time
(t) to enter a staircase, movement down the staircase and movement through an
exit. The value of E may be adjusted to take account of the presence or absence
of active fire protection measures such as detectors, communication systems,
sprinklers, emergency lighting and smoke ventilation s ystems.
It is also necessary that the total evacuation time H (= D + B + E) is less than
the time, F, taken by a combustion product to travel from the place of origin and
produce an untenable condition on any segment or component of an egress route.
The deterministic design criterion is H is less than F. To satisfy this condition,
the approach based on partial safety factors, discussed in Section 3.7.2, may be
adopted – see Equation (3.91).
The partial safety factors approach is only a semi-probabilistic approach based
on the expert judgement of a fire safety engineer and the quality of information
available to them for estimating the values of the partial safety factors αf and αh.
A more rigorous probabilistic method is the ‘Beta method’, discussed in
Section 3.7.4, which takes into account the uncertainties in the evaluation of
F and H. In this method, the deterministic design criterion is modified to the
probabilistic criterion specified in Equation (3.77) or (3.78). To apply this method,
it is necessary to determine the probability distributions of H and F, together
with their parameters such as mean and standard deviation. This might be
possible by performing simulations for several scenarios, based on mathematical
(deterministic) or computer models for evacuation and for spread of heat, smoke
and other toxic products. Probability distributions can also provide more realistic
values for the partial safety factors αf and αh in Equation (3.91). This method
based on the ‘design point’ was described in detail by Ramachandran (1998) with
reference to the determination of fire resistance for a c ompartment.
As discussed in Section 3.7.3, a normal probability distribution can be assumed
for the total evacuation time H. This assumption may not be realistic for buildings
with a mixture of able-bodied and disabled occupants. For such occupancies the
distribution may be long-tailed such as log normal. An appropriate distribution
for H can be determined, if possible, by collecting and analysing necessary data
provided by case studies, fire drills, evacuation exercises and computer simulations
of evacuation models. Such a method can also be adopted to determine the
probability distribution of F for any combustion product, although a normal
distribution has been assumed in Section 3.7.3.
Egress failure would occur with a probability Q if H > F. If this undesirable
event occurs, one or more people may die due to visual obscuration caused by
smoke, incapacitation and other reasons. As discussed by Ramachandran (1993a),
the probability of death in a fire or fatality rate per fire, Pd, is the product of the
probability of egress failure, Q, and the conditional probability, Pp, of death if
egress failure o ccurs:
Pd = Q × Pp (9.11)
As illustrated by Ramachandran, it would be possible to estimate the values of
Pd and Pp for any type of occupancy by analysing data on real fires attended by the
fire brigades.
By specifying an acceptable level for the product Pd, the value of Q, probability of
egress failure, can be adjusted to take account of the presence or absence of active
fire protection measures such as detectors, communication systems, sprinklers and
smoke ventilation systems. It should be mentioned again in this context that, in
addition to acting as a detection system, sprinklers, even if they fail to extinguish a
fire, would retard the rate of growth of heat and smoke, thus increasing the value
of F and providing more time for the occupants to escape to a safe p lace.
By following the above probabilistic method, the design value of the evacuation
time E can be determined for any acceptable levels for the fatality rate, Pd, and
probability of egress failure, Q. This procedure is explained in Table 9.1, reproduced
Table 9.1 Design evacuation time and fatality rate for single and multiple occupancy
dwellings
E H θ β Probability of Probability of Fatality rate per
success failure (Q) fire (Pd )
(min) (min)
2 14 1.07 0.32 0.6255 0.3745 0.0097
3 15 1.00 0.00 0.5000 0.5000 0.0130
4 16 0.94 –0.29 0.3859 0.6141 0.0160
5 17 0.88 –0.60 0.2743 0.7257 0.0189
6 18 0.83 –0.87 0.1922 0.8078 0.0210
7 19 0.79 –1.10 0.1357 0.8643 0.0225
8 20 0.75 –1.33 0.0918 0.9082 0.0236
9 21 0.71 –1.58 0.0571 0.9429 0.0245
10 22 0.68 –1.76 0.0392 0.9608 0.0250
Pd = Q × Pp; Pp = 0.026
Note: H=D+B+E
D = 10 min
B = 2 min
F = 15 min
θ= F/H
β= (θ – 1) / c (θ2 + 1)½ ; c = 0.15
2 2 1/2
θ = 1 + βc(2 − β c )
2 2
1−β c
H and F have normal probability distributions.
from Ramachandran (1993a). In this table, the parameter E is the average or
expected time for occupants to travel from the places where they are located
and reach the entrance to a protected staircase. It does not include the duration
concerned with the occupants’ movement down the staircases and movement
through e xits.
Hasofer and Odigie (2001) developed a stochastic model for evaluating the
interaction between the spread of untenable conditions and occupant egress.
The authors measure safety by the expected number of deaths. The building is
represented by a network for modelling fire spread and by another network for
modelling occupant egress. A major innovation is the introduction of the concept
of discrete hazard function. It allows the interaction between the various factors
involved in the spread of untenable conditions and occupant egress to be taken
into a ccount.
References
Canter, D (1980), Fires and Human Behaviour, Wiley, Chichester.
Dunlop, K (1993), Real fire emergency evacuation of disabled people, Proceedings of
the CIB W14 International Symposium on Fire Safety Engineering. University of Ulster,
Jordanstown.
Frantzich, H (1996), Study of Movement on Stairs During Evacuation Using Video Analysing
Techniques, Lund Institute of Technology, Lund, Sweden.
Fruin, J J (1971), Pedestrian Planning and Design, Metropolitan Association of Urban
Designers and Environmental Planners, New Y ork.
Fruin, J J (1987), Pedestrian Planning and Design, revised edition, Elevator World Educational
Services Division, Mobile, AL.
Hasofer, A M and Odigie, D O (2001), Stochastic modeling for occupant safety in a building
fire, Fire Safety Journal, 36, 269–289.
Marchant, E W (1975), Fire safety for the handicapped, Proceedings of Seminar on Fire Safety
for the Handicapped, University of E dinburgh, Edinburgh.
Melinek, S J and Booth, S (1975), An Analysis of Evacuation Times and the Movement of
Crowds in Buildings, Current Paper CP96/75, Building Research Establishment, Fire
Nelson, H E and Mowrer, F W (2002). Emergency movement, Section 3, Chapter 14, SFPE
Handbook of Fire Protection Engineering, 3rd edn, National Fire Protection Association,
Quincy, MA.
Pauls, J (1987), Calculating evacuation times for tall buildings, Fire Safety Journal, 12, 213–
236.
Pauls, J (1995), Movement of people, Section 3, Chapter 13, SFPE Handbook of Fire
Protection Engineering, 2nd edn, National Fire Protection Association, Quincy, MA.
Pauls, J and Juillet, E (1993), Life safety of people with disabilities, Proceedings of the CIB W14
International Symposium on Fire Safety Engineering, University of Ulster, Jordanstown.
Predtechenskii, V and Milinskii, A (1978). Planning for Foot Traffic Flow in Buildings,
Amerina Publishing, New Delhi.
Proulx, G (1995). Evacuation times and movement times in apartment buildings. Fire
Safety Journal, 24, 229–246.
in the United Kingdom, Fire Safety: Science and Engineering, ASTM STP 882,
American Society of Testing and Materials, Philadelphia, PA.
Ramachandran, G (1993a), Probabilistic evaluation of design evacuation time, Proceedings
of the CIB W14 International Symposium on Fire Safety Engineering, University of Ulster,
Jordanstown
Ramachandran, G (1993b), Early detection of fire and life risk, Fire Engineers Journal,
December, 33–37.
Ramachandran, G (1998), Probabilistic Evaluation of Structural Fire Protection – A
Simplified Guide, Fire Note 8, Building Research Establishment, Fire Research Station,
Borehamwood.
Rubadiri, L, Roberts J P and Ndumu, D T (1993), Towards a coherent approach to
engineering fire safety for disabled people, Proceedings of the CIB W14 International
Symposium on Fire Safety Engineering, University of Ulster, Jordanstown.
Shields, T J, Dunlop, K and Silcock, G W H (1996). Escape of Disabled People From Fire. A
Measurement and Classification of Capability for Assessing Escape Risk. BRE Report 301.
Building Research Establishment, Fire Research Station, Borehamwood.
Shields, T J, Smyth, B, Boyce, K E and Silcock, G W H (1998), Evacuation behaviours
of occupants with learning difficulties in residential homes, Human Behaviour in Fire:
Proceedings of the First International Symposium, University of Ulster, Jordanstown.
Sime, J D (1991), Human Behaviour in Fires, Publication 2/91, Fire Research and
Development Group, Home Office, London.
Walsh, C J (1998), A rational fire safety engineering approach to the protection of people
with disabilities in or near buildings during a fire or fire related incident, Human
Behaviour in Fire. Proceedings of the First International Symposium, University of Ulster,
Jordanstown.
Watts, J M (1987), Computer models for evacuation analysis, Fire Safety Journal, 12, 237–
245.
Yoshimura, H (1998), Sounding out the disabled in the lower extremities on their escape
behaviour in building fire for safer fire escape design, Human Behaviour in Fire: Proceedings
of the First International Symposium, University of Ulster, Jordanstown.
10 Performance and effectiveness of fire
service intervention
10.1 Introduction
The demand for fire protection required by a community takes two forms: a
potential demand and a realised demand. The first form is concerned with ‘hazards’
which have the potential to initiate the occurrence of fires and cause damage to life
and property. When fires actually occur, the potential demand is converted into
realised demand. Fire prevention activities and protection measures, passive and
active, are aimed at reducing the adverse effects arising from potential demand.
The fire services are also involved in these activities to some extent but their
main responsibility is to meet the realised demand effectively by putting out fires,
rescuing people and carrying out salvage operations.
Three factors primarily determine the level of resources needed by a fire
department to cope with the realised demand. Firstly, the types of hazards involved
(e.g. apartment houses, industrial properties) determine the types and numbers of
fire fighting equipments needed, location of fire stations, with due regard to their
proximity to areas with high risk in terms of frequency of fires and damage and
number of personnel that should be assigned to engines, ladders and special units.
Secondly, the geography of the area protected affects the travel time required by
fire fighting equipments to reach various sites of fires. Extra protection may be
needed for isolated locations, e.g. ridges, or where access may become difficult
or restricted, e.g. tunnels, mountain roads. Thirdly, peak-period alarm rates
determine the extra personnel and equipment needed to handle the simultaneous
demand for service.
Subject to the factors mentioned above and economic and other constraints,
a fire department has to consider alternative deployment policies and identify
a policy or set of policies which will ensure ‘optimum’ performance in terms of
reduction in damage to life and property. The central parameter in obtaining an
optimum solution to this problem is the travel time to fire incidents in a region,
which depends on the region’s area, alarm rate, number of fire stations and
other characteristics. Travel time is the major component of attendance time or
response time. Several ‘descriptive models’ are available for estimating the travel
time directly or indirectly by converting results based on models for travel distance
– see Walker et al. (1979).
Performance and effectiveness of fire service intervention 311
Models discussed in this chapter are concerned with the evaluation of the
relationship between attendance time and damage to property and life. Other
aspects discussed relate to percentage of fires requiring fire brigade intervention,
methods of extinction, control time and its interaction with attendance time and
the effectiveness of fire brigade intervention.
10.2 Attendance time and travel t ime
10.2.1 Definition
Attendance time is the elapsed time from the moment a fire brigade is called
until fire fighting personnel and equipment arrive at the fire scene. After arrival,
a certain time is required for preparing and positioning equipment at the scene
before commencing fire fighting. This period, ‘set up time’, is not included in the
attendance time defined in the United Kingdom fire statistics; it is part of the
‘control time’ elapsing from arrival until the fire is brought under control by the
fire brigade. In United States fire statistics, the ‘set up time’ is included in the
‘response time’ which is the elapsed time from the moment the fire department is
notified until a fire company is on the scene and is ready to o perate.
An important component of the attendance time, or response time, is the ‘travel
time’, the time between the start of a fire unit from its quarters and its arrival at
the fire scene. An analysis of fire station location should focus on changes in travel
time but attendance time, or response time, is often the most important measure
since changes in fire cover policy might affect other components, ‘dispatching
time’ and ‘turnout time’. The first of these two components is the time between
the receipt of an alarm and the dispatch of a unit. The second is the time required
for the unit to leave its quarters and start moving once it has been dispatched.
Although our discussion in this section concentrates on travel time, in most
cases, attendance time or response time can be substituted for travel time and the
statements will remain correct.
10.2.2 Travel time vector

Different types of units (engines and ladders, for example) perform different
functions at a fire. Hence, it is necessary to distinguish travel times for each type
of equipment. In addition, since two units of the same type working together may
be able to take some action that neither could take alone, the travel time of each
arriving piece of equipment is important. Therefore, in evaluating alternative
deployment policies, it would be useful to consider the list or vector of travel
times for engines, ladders and other fire fighting units that would respond to each
incident under each policy.
The travel time vector for engines, for instance, gives the time of arrival (relative
to the time of departure from the fire station) of the first engine, second engine
etc. It is, however, difficult in practice to use individual travel or response time to
compare the attractiveness of alternative deployment policies. For example, the
travel time for the first engine under one policy may be longer than the travel time
for the first engine under another policy. On the other hand, the travel time for
the second engine under the first policy may be shorter than the travel time for
the second engine under the second policy. One way of comparing policies is to
calculate the average of the travel times of the first and second engines. From a
fire cover point of view, the first engine response time is more important than the
second engine response t ime.
10.2.3 Travel time measures

A wide variety of travel time measures may be used in different situations.
One measure is based on the frequency distribution of first-engine (or ladder)
travel time to incident locations throughout a city or in different regions of a
city. Consider, for example, a long travel time of 4 minutes. The frequency or
probability of exceeding this time may be 10 per cent under one deployment policy
while it may be 20 per cent under another policy. The first policy would certainly
be preferred. Another approach would be to compare alternative policies on the
basis of how well they satisfy requirements for maximum acceptable travel times
for different incident locations.
Another set of travel time measures relates to average travel times to groups
of incidents or incident locations. Incident locations can be grouped, for
example, by region of the city or company response area. The average travel
time to incidents can be found by weighting the travel time to each location by
the structural fire rate at that location (and not by alarm rate which includes
false alarms).
10.3 Travel distance
10.3.1 Methods of estimation

Travel time of an engine (or ladder) generally increases with distance travelled.
In many situations it may be inconvenient, impossible or even unnecessary to
measure or estimate travel times directly. In an analysis of deployment policies,
travel distances can be used directly or can be converted to travel times. An
advantage of direct use of distances is that they are more stable than times. They
do not change with weather, time of day etc. Moreover, there is likely to be broad
agreement on distances and how they are measured.
Consider the distance dij between two specific points i and j that have grid
coordinates (xi, yi) and (xj, yj) respectively. If the streets are laid out as a rectangular
grid, then
dij = | xi – xj | + | yi – yj |
If the travel is ‘as the crow flies’ in a straight line connecting the two points,
then
dij = [(xi – xj)2 + (yi – yj)2]½
In practice, neither of the two conditions mentioned above may hold exactly
and hence an approximation is needed. In many cities throughout the USA the
following formula has provided good estimates of the travel distance (Walker et
al., 1979):
dij = 1.15 [(xi – xj)2 + (yi – yj)2]½
Another approach to estimating travel distances was used in The New York
City – Rand Institute’s Fire Operations Simulation Model. In the simulation, the
distance between two points was computed on the following basis: If the distance
travelled is short, the fire company is presumed to travel on a right-angle grid
of streets oriented at a specified angle with respect to the coordinate axes. If
the distance travelled is long, it is assumed that a straight line is followed. If the
distances are of intermediate length, a combination of right-angle and straight-
line distances is used. Carter (1974) explained the details of these c omputations.
The third possible approach is based on a network model which will provide
estimates of travel distances with the greatest accuracy if the model has a ‘node’
for every intersection and an ‘arc’ for every street segment. One-way streets could
be modelled explicitly in the network by using directed arcs. Parks, lakes, railroads,
highways and other barriers to direct travel are automatically taken into account,
since permissible paths in the network would avoid these obstacles. The US Bureau
of the Census has created a file of data for computerised map generation that can
be used to construct such a network. It exists for several Standard Metropolitan
Statistical Areas (SMSA) in the USA and is called the Dual Independent Map
Encoding (DIME) file. In the UK, some of the Geographic Information System
(GIS) computer packages can enable the creation of networks to evaluate travel
distances. Efficient algorithms are available to compute the length of the shortest
path between every pair of points in a network – see, for example, Dreyfus (1969),
Dreyfus and Law (1977).
Although a network can be used to estimate travel distances which can be
converted to times, the result will not necessarily be the same as when using a
network to estimate travel time directly. Timed measurements take into account
hills and traffic lights and rarely the convenient set of rectilinear streets. Hence,
the shortest-distance path between two points may not be the fastest path. For
example, part of the trip could be made on a limited-access divided highway which
may make the trip longer but faster. When a network is used to estimate travel
times, the computer program assumes that the fire companies follow the fastest
route, even if some other route would have a shorter distance. For the reasons
mentioned above, very specific and accurate timed measurements over the station
grounds are made in the UK instead of a generalised fire m odel.
The network method is particularly useful in cities that are irregularly shaped,
divided by having a limited number of crossover points, or that contain large
areas (parks or airports, for example) through which fire companies cannot
travel. However, many cities in the USA are sufficiently regular that much
simpler methods are appropriate for estimating travel times by conversion from
travel distances. Such a conversion would be easy if fire companies travelled at a
constant speed but this is not supported by data collected in several cities. These
data show that average speed is higher on longer trips and an appropriate time–
distance relation should reflect this.
10.3.2 Time–distance relationship

A statistical analysis (Walker et al., 1979) of empirical data, collected in four cities
(Denver, Trenton, Wilmington and Yonkers) revealed that travel time and travel
distance were related according to a curve. The individual travel time functions
for the four cities were remarkably similar to each other. This was the reason for
combining the data obtained in the four cities and estimating the parameters of a
single function. These data were collected during experiments in which selected
units measured their travel times with stop watches and their travel distances
with odometers. An analysis of the data revealed that a square-root relationship
is applicable for distances up to 0.38 miles and a linear relationship for greater
distances.
The results of an analysis in a range of US cities (Walker et al., 1979) did
not indicate any practical differences between travel speeds under conditions of
daylight and darkness. While speeds were somewhat slower during rush hours,
they were closer to speeds at other hours than anticipated. The reduction in
average speed (about 20 per cent) was greatest during the 8am to 9am period.
The results of an additional analysis indicated clearly that response speed did not
change markedly by time of day.
10.3.3 Square-root law

Suppose we wish to compute the average travel distance of engine companies to
fires in a particular region of a city. We will be computing this average over all
the engines and alarm boxes in the region. Let D denote the distance travelled
by an available engine to a fire scene and E(D) the average (or expected) travel
distance. According to square-root law (Walker et al., 1979)
1
D1 = E(D ) = k( A / N ) 2 (10.1)
where k is a constant of proportionality, A is the area of the region and N is the

number of firehouses in the region that have engines available to respond. This
inverse relationship between E(D) and N implies, as one would expect, that the
average travel distance would decrease if the number of firehouses in the region
is increased.
In deriving Equation (10.1) it has been assumed that each firehouse has
a company available and ready to respond when an alarm is received. But the
number of companies available will vary from time to time and hence, it is not
realistic to use the number of firehouses as the value for N if Equation (10.1) is
used to estimate the average travel distance over a period of time. However, as
discussed by Walker et al. (1979), if the number of available engines in a region is
not too small (say, more than 2.0), a reasonable approximation is to replace N in
Equation (10.1) by the expected (long-run average) number of firehouses having
at least one engine company available.
The square-root law is also based on the following assumptions:
• Alarms are distributed randomly but with uniform probability density

throughout the region of interest.
• Firehouses are spread either in a regular pattern or randomly throughout the
region.
• Boundary effects are insignificant.
• Fire companies travel either on a straight line between two points or on a
dense rectangular grid of streets.
In the real world, none of these assumptions is strictly true. Hence, the
validity of the square-root law for any region needs to be examined by performing
computer simulations. In the simulation experiment, conditions such as alarm
rates and number of active ladder companies should be varied over a broad
range, reflecting real situations, although the number of engine companies may
be fixed at an appropriate level. Alarm locations and patterns should be so
chosen that they imitate reality. The locations of engine and ladder companies
may be fixed at their actual locations adding, if required, new locations at
appropriate spots.
Walker et al. (1979) described a simulation experiment performed for the Bronx
region of New York City. (The reader should refer to this book for details.) The
results of this exercise supported the validity of the square-root law and showed
a general consistency between the parameters for engines and ladders. Hence,
the analysis was repeated, with the data grouped from various simulations and for
engines and ladders.
Walker et al. (1979) also examined the results from two independent studies:
one to determine fire station locations in Bristol, England (Hogg, 1968) and the
other to determine locations for ambulances in a suburban county near Washington
DC (Berlin and Liebman, 1974). In Hogg’s study neither the site locations nor the
fires were evenly distributed spatially, both being more dense in the centre of the
rectangular region under study. In her analysis, variations in the alarm pattern
by time of day were considered but possible unavailability of fire companies was
ignored. She calculated travel times from knowledge of the distances involved and
from estimates of travel speed. Berlin and Liebman used a ‘set-covering model’ and
computed the relationship between the maximum response time and the number
of locations occupied. Their computations were based on a linear relationship
between time and distance. Both the studies mentioned above confirm the validity
of the square-root model in terms of travel or response time but the results can be
converted to travel distances.
10.3.4 Average travel times in a region
The average travel time in a region is the average of many point-to-point travel
times. The frequency of each potential trip is determined by the frequency of alarms
at each alarm box, the dispatching rules used to assign fire companies to respond and
the availability of the fire companies assigned to respond. Using data on actual travel
times for a series of alarms, the average travel time for a region can be estimated by
dividing the sum of travel times for all trips by the number of trips. The computation
should include repetitions of the same trip, company m in travelling to alarm box b,
as many times as that run was made during the period of interest.
An alternative approach is to use other historical data and convert travel
distance to time for each possible trip by applying one of the methods discussed
earlier. This estimate can be combined with information on the fraction of all trips
made by all first arriving companies that involve company m in responding to box
b. Then, by considering all the m to b combinations, the average regional travel
time can be e stimated.
Both the two methods mentioned above involve tedious computations. A
simple approximation is obtained by combining the square-root law for average
distances (e.g. Equation (10.1)) with a function for converting travel distances
into travel times based on an analysis of data such as the one mentioned in
Section 10.3.2. With the resulting function, average travel times can be estimated
for regions where comparatively little is known about the details of travel patterns,
alarm distributions etc.
10.4 Attendance time and fire damage
10.4.1 Introduction
Fire brigades are required to provide fire cover to different categories of fire risk by
meeting certain standards, specified in terms of attendance or response time. In
the United Kingdom, for example, there are four risk categories: A, B, C, D. These
correspond approximately to commercial and industrial city complexes, centres of
large towns, built-up areas of towns and rural areas. Exceptionally high risks and
remote rural areas are treated as special cases. For each risk category there is a
recommended first attendance. For A category it is two pumps (first attendance
fire appliance) within a maximum period of five minutes and one further pump
within eight minutes. It is one pump within five minutes plus a second within 8
minutes for B category, one in 8 to 10 minutes for C and one in 20 minutes for
D. For known small fires, such as those on waste ground or in derelict buildings,
brigades use their discretion.
The safety levels provided by standards such as those mentioned above are
practically unknown quantities. It would be useful to evaluate these levels in terms
of life loss and property damage in order to assess changes in these levels due to a
re-location of fire stations or re-allocation of fire fighting equipments, personnel
and other resources to existing stations. For this purpose, it is necessary to evaluate
the relationship between the attendance or response time and damage due to fires,
life and property. This is a difficult task, particularly due to the random nature of
fire spread and interactions between various factors. However, some researchers
have attempted to model such a relationship as described briefly in the next two
subsections.
10.4.2 Attendance time and fire loss

In the Untied Kingdom, for example, Hogg (1973) carried out a study of fire losses
in relation to attendance time. She modelled a fire as growing in stages: confined
to object first ignited, confined to room, beyond room and beyond building. Fire
spread from stage to stage was governed by transition probabilities as functions of
(dependent on) time. The starting probabilities at discovery, together with the
transition probabilities of fire spread, generated the probability that a fire would
be in a given stage according to the discovery to arrival time. For each stage, i.e.
extent of spread category, the observed mean area burned was obtained according
to whether or not the fire was fought before the arrival of the fire brigade. Hogg’s
model took account of the first pump at each station and the first at each fire and
provided output in terms of arrival times of the first pump.
Hogg estimated the relationship between attendance time and fire loss for four
major types of occupancies, each subdivided into two categories: fires fought before
the arrival of the fire brigade and fires not so fought. Using loss per unit area, Hogg
converted area damage to financial loss. As one might expect, the losses in the
first category were much smaller than those in the second. The overall results
obtained by Hogg are given in Table 10.1. In the USA a similar study was carried
out by Halpern, Isherwood and Wand (1979).
Maclean (1979) attempted to evaluate the loss-attendance time relationship
for nine occupancy groups by applying ‘longitudinal’ analysis and comparing the
size of a fire at one known time with the size of the same fire at a different time. He
used area damage to describe fire size. Converting area damage to financial loss,
Maclean estimated that an increase in attendance time of one minute beyond the
level at that time would result in an overall increase of £1570 per fire (at 1978
prices) to the total loss to the e conomy.
10.4.3 Attendance time and life loss

In Section 3.3.3, the relationship between fatality rate and delays in discovering
fires was estimated with the aid of fire statistics published by the Home Office,
UK. Based on this relationship, a parameter λ was estimated to have values of
0.0008 and 0.0006 for single-occupancy and multiple-occupancy dwellings.
This parameter, measuring the average increase in fatality rate per minute can,
perhaps, be regarded as applicable to the period following the discovery of a fire
until the arrival of the fire brigade at the fire scene. Under this assumption, it can
be approximately estimated that one additional life per thousand fires would be
lost for every minute of delay in attendance or response t ime.
Table 10.1 Direct loss in relation to attendance time – non-dwellings, UK
(£, 1967 values)
Attendance time Single storey Multi-storey
(mins) 21,806 fires 40,419 fires
1 3,420 4,878
2 3,865 5,187
3 4,312 5,489
4 4,764 5,784
5 5,226 6,073
6 5,700 6,355
7 6,189 6,632
8 6,696 6,903
9 7,224 7,170
10 7,776 7,431
11 8,354 7,687
12 8,960 7,939
13 9,599 8,187
14 10,271 8,432
15 10,981 8,672
16 11,730 8,909
17 12,521 9,143
18 13,358 9,374
19 14,244 9,601
20 15,181 9,827
The value of λ can be expected to increase with time since the start of the fire,
although it has been assumed as a constant independent of time in the analysis
mentioned above. Hogg (1973) attempted to evaluate this relationship for the
period from discovery of fire to the arrival of the brigade at the fire scene. She
estimated, for this period, that the increase in the probability of one or more
deaths for every minute delay in arriving at the scene would be 0.015 on average
over a duration of 21 minutes. According to this result, the average value of λ
would increase 15 times, from 0.001 to 0.015, from the first period up to discovery
of fire to the second period after discovery. It is difficult to explain the reason
for such a significant increase in λ from one period to a subsequent period but it
should be pointed out that Hogg’s results were based on a much smaller (7,818)
number of fi res.
Using a ‘square-root law’, Hogg (1973) estimated that if the number of fire
stations covering the UK in 1967 had been double the actual number of stations,
the fatalities predicted would have been fewer by only 51, a fall from 584 to 533,
while a fourfold increase in the number of stations would probably have saved a
further 37 lives. She concluded that any change in fire cover would only have
a marginal effect on the number of lives lost. It ought to be mentioned in this
connection that the increasing installation of smoke alarms in dwellings in the UK
may affect the attendance time/life loss relationship.
Corman et al. (1976) investigated the effects on fire casualties of reducing
response distances of fire companies. According to their analyses, based on a
‘pairing scheme’ of fatal and non-fatal fires, none of the means of the differences
of response distance for fatal pairs was significantly different from zero. For injury
data, however, the mean of engine distance difference was significantly non-zero
at the 87 per cent probability level and for ladders it was significantly non-zero at
the 92 per cent level. The important conclusion of this study was that the effect
of fire company response distance (for average distances typical of New York City)
on fire casualties was very small compared with the effects of other factors, such as
time of day, season, type of building construction and floor of fire o rigin.
10.5 Fire fi
ghting
10.5.1 Fire brigade intervention

Some fires in buildings burn out themselves for a variety of reasons or are
extinguished by first-aid methods such as portable fire extinguishers, buckets of
water or sand and smothering. Fire incidence data contain information on some
of these ‘small’ fires which were reported to the fire brigade but were out on (or
before) the arrival of the brigade. In sprinklered buildings, some ‘small’ fires do not
produce sufficient heat to activate these fire protection systems, such that they are
either self-extinguished or extinguished by fi
rst-aid means.
According to UK fire statistics, about 55 per cent of the fires in sprinklered
buildings attended by fire brigades or reported to them are ‘small’ fires as defined
in the previous paragraph. The remaining 45 per cent are ‘big’ fires requiring
intervention by sprinklers, with sprinklers operating in 87 per cent of these cases
(39 per cent of all reported fires) and not operating in 13 per cent of the cases (6
per cent of all fires). Some of the fires in which sprinklers operate are extinguished
by the system before the arrival of the brigade and some are extinguished by the
brigade. The Home Office Research and Planning Unit has conducted surveys in
which a question on first-aid fire fighting was asked. They produced figures of the
order of 1 in 10 fires being reported to the fire service.
Fire brigades deal with ‘big’ fires which are neither self-extinguished nor
extinguished by first-aid means or sprinklers. Based on UK fire statistics for the
period 1984 to 1986, the proportion of fires extinguished by fire brigades is shown in
Table 10.2 for seven types of occupancies and the two cases, sprinklered and non-
sprinklered buildings. Also shown in this table are proportions of fires extinguished
by sprinklers and proportions of small fires self-extinguished or extinguished by
first-aid methods. In the sprinkler case, the figures include estimated numbers
Table 10.2 Percentage of small fires and fires extinguished by sprinklers and by fire brigade
Sprinklered building Non-sprinklered building
Extinguished by
Extinguished
Small Fire Small by fire
Occupancy fires Sprinklers brigade fires brigade
Textile industry 37 39 24 33 67
Chemical and allied 37 44 19 52 48
industries
Paper, printing and 37 42 21 47 53
publishing industries
Timber. furniture, etc. 37 38 25 13 87
industries
Retail distributive trade 37 42 21 21 79
Wholesale distributive 37 39 24 17 83
trade
Office buildings 37 49 14 28 72
of fires which are neither reported to the fire brigade nor any insurance claims
are made. Such ‘unreported fires’ extinguished by sprinklers, according to an
unpublished research study quoted by Rogers (1977), constitute about one-third
of fires in sprinklered buildings. Fire brigades only attend two-thirds of fires in
2
these buildings. Also, in this case, 4 per cent (= 6 × ) of reported fires in which
3
sprinklers did not operate have been included in fires extinguished by the brigade.
10.5.2 Methods of extinction

The growth of a fought fire depends not only on the burning characteristics of
the fire and fire fighting before the arrival of the brigade, but also on the method
used by the fire brigade to extinguish the fire. These methods include hose reels,
main jets and other means. According to UK fire statistics for 1991, fire fighting by
brigades is not required in 29 per cent of the fires. In 71 per cent of fires requiring
fire brigade action, no hose reels or main jets are used in 17 per cent of the fires,
only hose reels in 68 per cent of fires, 1 or 2 main jets in 12 per cent of fires, 3
or 4 main jets in 2 per cent of fires and 5 or more jets in 1 per cent of fires. This
distribution of number of fires according to hose reels and main jets indicates, to
some extent, the sizes of fires which have to be tackled by the fire brigades. Until
1965, in the absence of data on financial losses, fires extinguished by 5 or more jets
by the fire brigades in the UK were regarded as large fi res.
However, there is a counter-acting factor affecting the sizes of fires at the time
of extinction. The sizes reflected in the extent of fire spread or damage also depend
on the number of jets used and the times at which the jets are applied. A simulation
method, based on data for a large number of fires, can be applied to estimate the
general relationship between average spread or damage and the arrival times of
pumps. Rutstein (1975) attempted this method of analysis and found that, on
average, non-dwelling fires were growing at the rate of about 10 per cent per
minute before the arrival of the first pump. Once fire fighting commenced, the
average rate of spread was about 1–2 per cent per minute before the arrival of the
second pump and about 0.25 to 0.5 per cent per minute before the arrival of the
third pump. These results indicate that fire damage depends mainly on the arrival
time of the first pump.
10.5.3 Control time

As discussed in the previous section, the time taken by a fire brigade to bring a
fire under control and extinguish it depends, to some extent, on the number of
hose reels or main jets used and the build-up of jets over the fire fighting period.
Control time also depends on the size of the fire at the time of arrival of the
brigade at the fire scene. This size is related to the elapsed period TA , from the
start of the fire until fire brigade arrival. A fire discovered or detected soon after
ignition, and reported to the fire brigade, will be in its early stage of growth when
fire fighting commences and hence can be controlled q uickly.
A reduction in TA will, therefore, shorten the control time, TB, as well,
thus reducing the total fire duration (T) and damage to an appreciable extent,
according to the exponential model in Equation (3.42). The relationship
between TA and TB may be assumed to be approximately linear and can be
estimated with the aid of raw data on individual fires, such as those available in
the UK. Ramachandran (1992) carried out such an analysis and obtained the
results shown in Table 10.3.
The regression coefficient given in Table 10.3 indicates the additional control
time (mins) required for every additional delay of one minute in arriving at the
fire scene due to delays in discovery of fire or calling the fire brigade or attendance
(response). This coefficient also indicates the extent to which the control time
(mins) will be reduced for every minute of early arrival of the Fire and Rescue
Service at the fire scene due to early detection of fire or early attendance. The
coefficient was statistically significant in most of the cases. Figures such as those
in Table 10.3 can be used in conjunction with an exponential (or any other)
model of fire growth to estimate the additional damage likely to occur due to
delays in discovering fires or responding to them. Long delays due to these
factors would increase the chance of a fire spreading beyond the room of origin,
particularly in a building without sprinklers. Figures 10.1 and 10.2 are examples
showing the extent to which damage and probability of spread would increase if
the fire brigade commences its attack when a fire is in its second or third stage
of growth. Such figures were produced by Ramachandran (1992) for some types
of industrial occupancies and scenarios by performing simulations based on an
exponential model of fire growth (Equation (3.42)) in real fires attended by fire
brigades.
Table 10.3 Control time (minutes) and arrival time (minutes): regression coefficients
Industry Area of fire origin
Production Storage Other
areas
Food, drink, tobacco 0.458 0.359 0.405
Chemicals and allied industries 0.417 0.560 0.370
Metal manufacture 0.334 0.325 0.118
Mechanical, instrument and electrical 0.336 0.367 0.218
engineering
Textiles 0.439 0.198 0.117
Clothing, footwear, leather and fur 0.318 0.172 0.351
Timber, furniture, etc. industries 0.249 0.311 0.299
Paper, printing and publishing industries 0.465 0.440 0.187
10.5.4 Number of jets

According to an analysis carried out by the Fire Research Station, UK on behalf
of the Swedish Fire Protection Association (Bengtson and Hagglund, 1986) the
number of hose reel jets (Jh) required to fight a fire of size Af square metres is given
by the following equation:
Jh = 2.71 + 4 loge Af (10.2)
In this analysis, five hose reel jets were considered to be equivalent to one main
jet in regard to discharge rate of water. Hence, in terms of main jets (J), Equation
(10.2) may be rewritten as
J = 0.54 + 0.8 loge Af (10.3)
If, for example, Af is equal to 80m2, Jh = 20 or J = 4. In Equations (10.2) and

(10.3) Af denotes damage in terms of fire area and not total area including smoke
and water d amage.
Data for all fires were utilised in deriving Equations (10.2) and (10.3) and
hence this result was weighted heavily in the direction of small fires. It is unwise
to apply this result to large fires for which Thomas (1959) obtained the following
equation based on 48 fires ranging in ground plan area from 28m2 to 56,000 m 2:
J = 0.1 A
where A is the area in square feet, or
J = 0.33 A (10.4)
Non-sprinkled building Intervention by fire brigade
Occupancy type Retail distributive trade
Area of fire origin Assembly areas
Calculated time Probability of

Spread Extinction Damage
Intervention stage Elapsed Cumulative Extinction
stage by SM/FB m2 Intervention Extinction
(mts) (mts) (conditional)
1 SM 0 0* 1.15 0.00 0.00 0.00
1 FB 9.4 9.4 1.63 1.00 0.33 0.33
2 FB 24.0 33.4 3.99 1.00 0.27 0.27
1 3 FB 52.4 85.8 27.96 1.00 0.27 0.27
4 FB 42.0 127.8 133.29 1.00 0.13 0.13
Average 26.49
damage
1 SM 0 0 1.15 0.00 0.00 0.00
2 SM 13.5 13.5* 1.67 0.00 0.00 0.00
2 FB 29.4 42.9 4.99 1.00 0.40 0.40
2 3 FB 57.8 100.7 42.85 1.00 0.40 0.40
4 FB 47.4 148.1 249.88 1.00 0.20 0.20
Average 69.11
damage
1 SM 0 0 1.15 0.00 0.00 0.00
2 SM 16.2 16.2 1.67 0.00 0.00 0.00
3 SM 81.8 95.3* 15.98 0.00 0.00 0.00
3 3 FB 90.5 185.8 463.10 1.00 0.67 0.67
4 FB 80.1 265.9 9114.87 1.00 0.33 0.33
Average 3316.18
damage
* Time of commencement of fire brigade attack Stages
SM Small fire 1: Confinement to item first ignited
FB Fire brigade 2: Spreading beyond item first ignited but confinement to contents of
room of fire origin (structure not involved
3: Spreading beyond item first ignited and other contents but confinement
to room and involvement of structure
4: Spreading beyond room of origin
Figure 10.1 Effectiveness of early intervention by fire brigade – example based on UK fire statistics, non-sprinklered buildings
Sprinkled building Intervention by fire brigade
Occupancy type Retail distributive trade
Area of fire origin Assembly areas
Calculated time Probability of

Spread Extinction Damage
Intervention stage Elapsed Cumulative Extinction
stage by SP/FB m2 Intervention Extinction
(mts) (mts) (conditional)
1 SP 0 0* 1.80 0.70 1.00 0.70
1 FB 21.5 21.5 4.60 0.30 0.31 0.09
2 FB 15.0 36.5 9.36 0.30 0.44 0.13
1 3 FB 23.2 59.7 27.96† 0.30 0.19 0.06
4 FB 26.4 86.1 133.29† 0.30 0.06 0.02
Average 7.14
damage
1 SP 0 0 1.67 0.79 0.89 0.70
2 SP 4.6 4.6* 2.67 0.79 0.11 0.09
2 FB 16.8 21.4 5.90 0.21 0.64 0.13
2 3 FB 25.0 46.4 19.20 0.21 0.27 0.06
4 FB 28.3 74.6 94.32 0.21 0.09 0.02
Average 5.21
damage
1 SP 0 0 1.67 0.92 0.87 0.80
2 SP 4.6 4.6 2.67 0.92 0.11 0.10
3 SP 17.1 21.7* 14.97 0.92 0.02 0.02
3 3 FB 31.9 53.6 67.47 0.08 0.75 0.06
4 FB 35.1 86.7 420.06 0.08 0.25 0.02
Average 14.35
damage
* Time of commencement of fire brigade attack Stages
† Figure for non-sprinled building used as estimate 1: Confinement to item first ignited
SP Sprinklers 2: Spreading beyond item first ignited but confinement to contents of
FB Fire brigade room of fire origin (structure not involved
3: Spreading beyond item first ignited and other contents but confinement
to room and involvement of structure
4: Spreading beyond room of origin
Figure 10.2 Effectiveness of early intervention by fire brigade – example based on UK fire statistics, sprinklered buildings
where A is the area in square metres. This sample was biased heavily towards
large fires over 200 m2. Equation (10.4) corresponds to the use of one jet per
10m circumference to the fire. Equations (10.3) and (10.4) give approximately the
same number of jets for 200m2 but the first equation underestimates the number
for fires of larger s izes.
According to Thomas (1959), with A in square feet, the average control time
T4 (minutes) is given b y
T4 = A
which is equivalent t o
T4 = 3.3 A (10.5)
with A in square metres. Reviewing data on 134 fires attended by fire brigades
which were published by Illinois Institute of Technology Research Institute (Labes,
1968), Baldwin (1970) obtained the following relationship between control time
and fire area in square metres:
T4 = 1.66 A0.559 (10.6)
Equations (10.3) or (10.4) and (10.5) or (10.6) provide an estimate of the

number of jets that would be required to control a fire of a certain area within a
certain time. For example, if the fire area is 100m², according to Equations (10.4)
and (10.5), 3.3 or 4 jets would be required to control the fire within 33 m
inutes.
References
Baldwin, R (1970), The Use of Water in the Extinction of Fires by Brigades, Fire Research Note
803, Fire Research Station, Borehamwood.
Bengtson, S and Hagglund, B (1986), The use of a zone model in fire engineering application,
Fire Safety Science: Proceedings of the First International Symposium. Hemisphere Publishing
Corporation, New York.
Berlin, G and Liebman, J (1974), Mathematical analysis of emergency ambulance location,
Socio-Economic Planning Sciences, 8, 323–328.
Carter, G (1974), Simulation of Fire Department Operations: Program Description, Report
R-1188/2-HUD/NYC, Rand Corporation, Santa M onica.
Corman, H, Ignall, E J, Rider, K L and Stevenson, A (1976), Fire casualties and their
relation to fire company response distance and demographic factors, Fire Technology,
12, 193–203.
Dreyfus, S (1969), An appraisal of some shortest path algorithms, Operations Research, 17,
3, 395–412.
Dreyfus, S and Law, A (1977), The Art and Theory of Dynamic Programming, Academic
Press, New York.
Halpern, J, Isherwood, G and Wand, Y (1979), Response times and fire property losses in
single and double family dwelling units, INFOR, 17, 373–379.
Hogg, J M (1968), The siting of fire stations, Operational Research Quarterly, 19, 275–287.
Hogg, J M (1973), Losses in Relation to the Fire Brigade’s Attendance Time, Fire Research
Report 5/73, Scientific Advisory Branch, Home Office, L ondon.
Labes, W G (1968), Fire Department Operation Analysis, Final Report. Illinois Institute of
Technology Research Institute, Office of Civil Defense, Washington, DC.
Maclean, A D (1979), Fire Losses – Towards a Loss-Attendance Relationship, Fire Research
Report 17/79, Scientific Advisory Branch, Home Office, L ondon.
Ramachandran, G (1992), Statistically Determined Fire Growth Rates for a range of scenarios:
Part 1: An Analysis of Summary Data. Part 2: Effectiveness of Fire Protection Measures –
Probabilistic Evaluation, upublished report to the Fire Research Station, Borehamwood.
Rutstein, R (1975), Methods of Planning Fire Cover Using Cost Effectiveness Criteria, Fire
Research Report 7/75, Scientific Advisory Branch, Home Office, L ondon.
Thomas, P H (1959), Use of water in the extinction of large fires, Institution of Fire Engineers
Quarterly, 19, 35, 130–132.
Walker, W E, Chaiken, J M, Ignall, E J (Editors), (1979), Fire Department Deployment
Analysis: A Public Policy Analysis Case Study, The Rand Fire Project, Rand Corporation,
North Holland, NY.
11 Whole project analysis
11.1 Introduction
This chapter describes the application of full quantitative fire risk assessment. The
process is called ‘full’ since the analysis quantifies the probabilistic and physically
deterministic processes of fire events for a particular building. Chapter 3 provides
detailed information on the methods for undertaking these two key processes and
Chapter 4 contains information to help inform the judgement of the acceptability
of any fire risk predicted.
In contrast, some methods in Chapter 3 concentrate on purely statistical
methods, where the consequences of fire events are contained in the data. This
has the advantage that all the consequences implicit in the analysis are from actual
events. The disadvantage is that it implicitly assumes that any particular data set is
from a homogenous set of buildings and that they are similar to the building being
assessed. For that reason, purely probabilistic methods tend to be used to inform
decisions concerning large sets of buildings such as those concerning regulation or
insurance portfolios.
Full quantitative fire risk assessment allows the performance and reliability of
fire precautions and the physical nature of a building, its fire hazards and occupants
to be explicitly taken into account in the fire safety assessment of the building
(Charters 2000). It can be used to quantify levels of fire risk to life, property,
business and/or the environment and the uncertainties affecting risk in terms of
probabilities (Charters 2000).
Full quantitative fire risk assessment is usually most useful in quantifying the
levels of risk for an individual building either during design and/or o ccupation:
• in demonstrating equivalent life safety to a code-compliant building or

satisfaction of fire risk criteria;
• for cost-benefit analysis of property protection/business continuity; a nd/or
• in assessing the environmental impact of large fires.
Therefore, circumstances where full quantitative fire risk assessment would be

useful include where life safety is affected by buildings:
• containing very large numbers of people who are unfamiliar with the building
• containing large numbers of sleeping occupants
• containing large numbers of vulnerable people
• with very high fire growth rates
• with restricted means of escape; and/or
• with other unusual features which could be significantly detrimental to fire
safety.
Examples where full quantitative fire risk assessment would be useful for life
safety in buildings and structures may include the larger assembly buildings, hotels,
hospitals, industrial and commercial premises, tunnels, offshore installations and
ships.
• Property protection/business continuity is affected by buildings w

ith:
• a high v alue;
• high value operations; and/or
• high value images.
Premises where full quantitative fire risk assessment would be useful for
property protection/business continuity may include certain landmark,
headquarters, leisure, assembly and heritage buildings as well as some
industrial, commercial and transport premises (Charters and Wu 2002).
• Environmental protection is affected by buildings with a high potential impact
on the environment. Premises where full quantitative fire risk assessment
would be useful may include industrial and transport premises particularly
where hazardous goods may be present in significant quantities (Charters and
McGrail 2002).
Fire risk assessment is essential where deterministic fire safety engineering

cannot adequately address all the fire scenarios of concern. This tends to occur
when the consequences in terms of life safety, loss prevention and/or environmental
protection of a fire may be intolerable.
Factors that might indicate where quantitative fire risk assessment would be
essential include:
• When the reliability and performance of protection systems is critical. Fire

risk assessment should be used to assess the defence in depth of a design that
relies heavily on a single fire safety system.
• When the variability of and uncertainties governing input parameters have
significant impact on the results. Fire risk assessment should be used where
there are significant variations in variables like the number of people, their
characteristics, fire growth rates etc and deterministic analysis shows that
credible combinations of the variables are not acceptably safe (Charters,
McGrail, Fajemirokum and Wang 2002).
• When a wide range of fire scenarios is deemed to be necessary. Fire risk
assessment should be used in complex, high value, mission critical, buildings
where high levels of defence in depth (multiple fire safety systems) are required.
Whole project analysis 329
11.2 Risk parameters
Since fire risk assessment is one of the tools of risk management, it is clear that
there is a need to measure risk on the basis that if it isn’t measured, it can’t be
managed. A measure of risk is often called a risk parameter. There are several
ways of measuring any particular risk and the optimum risk parameter is therefore
related closely to the type of risk being analysed and the decision being taken. For
example:
• If the objective is life safety, and there is concern about injuries (or fatalities)
to people in a building, the risk parameter could be measured in terms of the
number of injuries (or fatalities) per building year.
• If the objective centres on property protection and/or business continuity,
the risk parameter could be measured or monitored in terms of the ‘financial
value of losses per year’.
In general terms, the risk parameter is measured in ‘outcomes per unit of

activity’, where the ‘outcome’ is the unwanted event and the ‘unit of activity’ is
a measure of time, number or frequency of occurrences, etc. The unit of activity
can have a significant impact on the perceived level of fire risk. For example, risk
parameters using ‘per building year’ may be appropriate if society’s experience or
perception of those risks is linked to a specific building or building type. However,
‘per building year’ as a measure of activity, will naturally be biased against large or
high occupancy buildings. If the unit of activity is ‘per area year’ or ‘per occupant
year’, this may produce a less biased comparison between buildings of different
sizes.
11.3 The full quantification fire risk assessment process

Risk assessment is the appraisal of one or more risk parameters. Risk assessment
can be used to implement a safety objective more cost-effectively by directing
spending to address the most significant safety issue(s).
Having identified a suitable risk parameter, the full quantitative risk assessment
process can begin. The full quantitative fire risk assessment process involves four
main tasks (see Figure 11.1):
1 Hazard identification – to identify hazards that could give rise to incidents

or events of concern.
2 Frequency analysis – to estimate how often such events are likely to o ccur.
3 Consequence analysis – to estimate the potential severity of such e vents.
4 Risk evaluation – to decide what, if anything, should be done to address the
levels of risk predicted.
Hazard
identification
Frequency Consequence
analysis analysis
Risk Risk
evaluation reduction
Risk NO
acceptable
YES
END
Figure 11.1 Typical quantified fire risk assessment process
11.3.1 Hazard identification processes

For fire safety, events of concern might be those that could lead to the outcome of
concern (death/injury, asset damage and/or business interruption etc). There are
many forms of hazard identification and these include:
• checklist type approaches (see Section 2.1);

• structured brainstorms (see matrix methods in Section 2.2).
In fact, a highly effective way of assessing risks is to undertake a structured

brainstorm process such as HAZID (HAZard IDentification) or HAZOP (HAZard
and OPerability) studies and then identify the core events of concern using full
quantitative risk a ssessment.
11.3.2 Frequency analysis

In order to estimate the frequency of fire events, historical data is collected on
how often fires occur in different building types. Because serious fires in buildings
are relatively infrequent, there is arguably insufficient information to provide a
meaningful direct estimate of the frequency of severe fires. If all the data is broken
down into sub-events that could lead to the severe event of concern, this can be
reconstructed using fault and event trees to provide a prediction for the frequency
of rare, but severe fire e vents.
Frequency analysis is addressed in detail in Chapter 3 .
11.3.3 Consequence analysis

There are several ways of estimating the severity of the consequences of an event
of concern – by historical data, simple analytical methods, computational methods
and/or small/full-scale tests and experiments. Each approach has its advantages
and d isadvantages.
Historical data gives details of known past outcomes, but this may not always
be complete or even relevant in the context of current or future practices. Simple
analytical methods can be used to estimate the severity of event outcomes cost-
effectively, but the predictions are only as good as the model, its data and its
application including any assumptions (implicit or explicit). Computer models can
be used to estimate the severity of outcomes more closely, but such analysis can
be expensive and time consuming, limiting the range of events being quantified.
Full-scale testing can give the most accurate assessment of the level of fire hazard
and the response of a building and its occupants, but again it is usually highly
costly and time consuming as separate experiments are needed for fire hazards and
occupant r esponse.
Recently, significant advances have been made in the prediction of fire
consequences using computer models. Models have been developed and improved
to give more accurate and cost-effective prediction of fire consequences and again
great care is needed in their application and a detailed understanding of the
physical and computational basis of the model is necessary for competent u se.
Consequence analysis is addressed in detail in Section 3 .6.
11.3.4 Risk evaluation

The level of risk can be calculated by combining the frequency and consequence
analysis for the outcomes of events. Once the level of risk has been quantified, it
is necessary to decide whether or not it is acceptable, and if not, what should be
done to reduce it. The two main approaches to risk evaluation a re:
1 Risk comparison – where the risk of an activity is compared with the risks
from relevant prescriptive standards for life safety. For risk comparison the
Health and Safety Executive and Health and Safety Commission have also
proposed risk acceptance criteria. For example, the maximum tolerable risk to
an individual member of the public is a 10–4 probability of death per year. This
equates to a one in 10,000 chance of an individual dying in one y ear.
2 An economic approach – where the cost-effectiveness of risk reduction can
be assessed. This approach is usually used for property protection or business
continuity risk assessments.
Risk evaluation is addressed in detail in Chapter 4. If the risks are not
acceptable, then risk reduction is required.
11.3.5 Risk reduction

Overall, several approaches to risk reduction can be employed (see Section 11.1
for risk management). In terms of the full quantitative fire risk assessment process,
there are three main approaches:
1 Reduce the hazard – less material, lower hazard material, control of ignition
sources, e tc.
2 Reduce the frequency – by reducing the hazard or introducing additional
mitigation measures or improving the reliability of existing measures to reduce
the frequency of severe events of concern.
3 Reduce the consequences – again by reducing the hazard or introducing
additional mitigation measures to reduce the severity of event o utcomes.
The full quantitative fire risk analysis can then be modified in line with the
proposed changes and the predicted levels of fire risk e valuated.
11.4 Examples
The first example fire risk assessment was to satisfy the Building Regulations for a
novel design. A quantitative probabilistic method was used to satisfy acceptance
criteria and demonstrate equivalency with similar buildings designed to the
prescriptive guidance. The second example is a financial risk assessment of the
risk-cost benefit of the installation of sprinklers in s chools.
Building Regulations
The objectives of the Building Regulations 2000, with respect to fire safety, are set
out in five functional requirements under Part B:
1 To ensure satisfactory provision of means of giving an alarm of fire and a

satisfactory standard of means of escape for persons in the event of a fire in
a b uilding.
2 That fire spread of the internal linings of buildings is inhibited.
3 To ensure the stability of buildings in the event of fire; to ensure that there is
a sufficient degree of fire separation within buildings and between adjoining
buildings; and to inhibit the unseen spread of fire and smoke in concealed
spaces in buildings.
4 That external walls and roofs have adequate resistance to the spread of fire
over the external envelope, and that spread of fire from one building to
another is restricted.
5 To ensure satisfactory access for fire appliances to buildings and the provision
of facilities in buildings to assist fire fighters in the saving of life of people in
and around b uildings.
Guidance on how to comply with Part B of the Building Regulations can be

found in Approved Document B. This guidance is intended for some of the more
common building situations. However, it recognises that there are alternative ways
of complying with the requirements. The means for developing alternative design
solutions is commonly known as fire safety engineering. Paragraph 0.11 goes on to
state that a fire engineering approach may be the only practical way to achieving a
satisfactory standard of fire safety in some large and complex buildings. Paragraph
0.13 indicates that factors that should be taken into account include:
• the anticipated probability of a fire occurring; and

• the anticipated fire severity.
Furthermore, alternative solutions are accepted on the basis of equivalency,

that is t o:
… demonstrate that a building, as designed, presents no greater risk to

occupants than a similar type of building designed in accordance with well
established codes.
Therefore, fire risk and fire risk assessment can play a major role in the design
of new buildings. Further information on fire engineering can be found in BS
7974: 2001 ‘Code of practice on the application of fire engineering principles to
the design of buildings’. Part 7 of BS 7974 contains guidance on the application of
probabilistic fire risk assessment.
BS 7974 Code of practice on the application of fire engineering principles

to the design of buildings
This Code of Practice was published under the Fire Standards Policy Committee
and is published as part of the PD 7974 series. Other parts published or about to
be published are as follows:
• Part 0: Introduction to fire safety engineering

• Part 1: Initiation and development of fire within the enclosure of o rigin
• Part 2: Spread of smoke and toxic gases within and beyond the enclosure of
origin
• Part 3: Structural response and fire spread beyond the enclosure of o rigin
• Part 4: Detection of fire and activation of fire protection systems
• Part 5 Fire service intervention
• Part 6: E
vacuation
• Part 7: Probabilistic risk assessment.
This Code of Practice provides guidance on the use of BS 7974:2001 and
is a framework for an engineering approach to the achievement of fire safety
in buildings. It gives guidance on the application of scientific and engineering
principles to the protection of people and property from fire. It also gives a
structured approach to assessing the effectiveness of the total fire safety system in
achieving the design objectives.
It provides guidance on the design and assessment of fire safety measures in
buildings. It provides some alternative approaches to existing codes and guides
for fire safety and also allows the effect of departures from more prescriptive codes
to be evaluated. It recognises that a range of alternative and complementary fire
protection strategies can achieve the design objectives.
This Code of Practice is intended to provide a framework for a flexible but
formalised approach to fire safety design that can also be readily assessed by the
statutory authorities. It is intended that this Code of Practice, when used by
persons suitably qualified and experienced in fire safety engineering, will provide
a means of establishing acceptable levels of fire safety economically and without
imposing unnecessary constraints on other aspects of building d esign.
BS 7974 can be used to define one or more fire safety design issues to be
addressed using fire safety engineering. The appropriate Published Documents
(PDs) can then be used to specific acceptance criteria and/or undertake detailed
analysis.
A fire safety engineering (FSE) approach that takes into account the total fire
safety package can often provide a more fundamental and economical solution
than more prescriptive approaches to fire safety. It may, in some cases, be the only
viable means of achieving a satisfactory standard of fire safety in some large and
complex b uildings.
Fire safety engineering can have many benefits. The use of BS 7974 will
facilitate the practice of fire safety engineering and in particular it w
ill:
• provide the designer with a disciplined approach to fire safety d esign;

• allow the safety levels for alternative designs to be c ompared;
• provide a base for selection of appropriate fire protection s ystems;
• provide opportunities for innovative design; and
• provide information on the management of fire safety for a b uilding.
Fire is an extremely complex phenomenon and there are still gaps in the
available knowledge. When used by a suitably qualified person, experienced in fire
safety engineering, this series of documents will provide a means of establishing
acceptable levels of fire safety economically and without imposing unnecessary
constraints on aspects of building design.
Part 7: Probabilistic risk assessment
This PD provides guidance on the application of probabilistic risk assessment

for fire safety engineering in buildings. This approach can be used to show how
regulatory, insurance or other requirements can be satisfied. Probabilistic risk
assessment, like fire safety engineering in buildings, is a developing field. As with
all engineering and risk disciplines, models and data can never fully describe
actual circumstances and so judgement is required in assessing whether a design
is acceptable. This judgement should be based on the best and most appropriate
facts and evidence available.
This PD may be applied to the design of new buildings and the appraisal of
existing buildings. Probabilistic risk assessment may be used in conjunction with
the other PDs (see Figure 11.1) and other guidance documents. It may also be
used to justify approaches that differ from those in other guidance d ocuments.
This PD provides guidance on probabilistic risk analysis in support of BS 7974,
‘Application of fire safety engineering principles to the design of buildings – Code
of Practice’. It sets out the general principles and techniques of analysis that can
be used in fire safety engineering. This PD also outlines the circumstances where
this approach is appropriate and gives examples illustrating their u se.
This PD also includes data for probabilistic risk assessment and criteria for
assessment. The data included is based on fire statistics, building characteristics
and reliability of fire protection systems. The criteria included cover life safety and
property protection, both in absolute and comparative terms.
This PD does not contain guidance on techniques for hazard identification or
qualitative risk analysis.
Probabilistic risk assessment of fire in buildings (with the exception of nuclear,
chemical process, offshore and transport) is not widely used and so a discussion of
possible future developments is included.
11.4.1 Example 1 Application of full quantitative fire risk assessment

of a shopping centre design
The Fields shopping centre is a multi-purpose retail/leisure complex in Denmark
(Charters, Paveley and Steffensen 2001). The podium is approximately 200m by
150m with linked adjacent blocks. The retail and leisure complex can be divided
into the following functional a reas:
• car park
• retail levels
• offices
• leisure facilities
• conference centre
• cinema complex.
A shopping and leisure complex such as Fields is not specifically addressed

in the Danish Building Regulations, so a performance-based fire engineering
approach has been adopted. Innovatively for a commercial development, a
quantified fire risk assessment was undertaken in parallel to ensure that there was
sufficient ‘defence in depth’ and this provided a powerful insight into the fire safety
of the c omplex.
11.4.1.2 The fire s trategy

A fire strategy was developed for the shopping and leisure complex using a
performance-based fire engineering approach and incorporating international
practice for shopping centres and the Danish Building R egulations.
Some of the fire safety objectives for the project were:
• safe and efficient evacuation of high populations;

• minimise property loss and business interruption;
• maintain open routes and shop fronts;
• provide flexibility for varied tenant use.
The fire safety measures for the building were designed to provide an integrated
package. With fire safety systems such as sprinklers, smoke control to protect
the mall, automatic fire detection and a voice alarm system, the mall forms an
evacuation route through which people can escape from the fire (see Figure 11.3).
Each area of the complex is within their own fire/smoke compartment,
although they will share escape routes, including stairs. Evacuation of the complex
is phased, evacuating each functional area individually. The shopping centre area
is s ub-divided into a number of evacuation zones.
The fire strategy was developed and agreed with the local authorities and was
used as the basis of the building design and formed part of the Building Regulations
submission. It is also be used for the future development and management of the
centre.
The provision of automatic sprinklers, smoke control to protect the malls
and large shops, automatic fire detection and a voice alarm system, allows the
alternative design to incorporate:
• extended travel distances within malls;

• travel distances in shops to be measured to the exit to the m all;
• shops not fire separated from the mall;
• fire resistance levels will be reduced in specific areas; a nd
• open stairs at the main entry/exit point from the mall areas will be used for
egress direct to the outside.
11.4.1.3 Quantified fire risk assessment

One of the more innovative aspects of the project was that a fire risk analysis was
required to be undertaken in parallel with the deterministic fire engineering. This
was in accordance with the principles of the Draft Common Guidelines for the
Øresund region, which includes acceptance criteria.
The risk analysis examined the overall ‘defence in depth’ of the proposed fire
safety systems and determined the level of risk that occupants were exposed to.
This included sensitivity analysis and incorporated the performance, reliability
and the effect of the failure of key fire safety measures (such as smoke control).
The quantified fire risk assessment was undertaken using SAFiRE (Simplified
Analytical Fire Risk Evaluation). SAFiRE combines consequence and frequency
analysis in the same way as traditional quantified fire risk assessment (developed
in the nuclear industry). Frequency analysis is based on statistical analysis, simple
fault and event trees and Monte Carlo analysis of highly variable parameters.
Consequence analysis is based on statistical analysis and simplified models for time
to untenability and time for evacuation. This method is used for the design process
because it is significantly quicker than traditional quantitative risk assessment.
SAFiRE takes into account the regulatory paradigm. That is, it addresses the
same fire safety objectives as Building Regulations and assesses adequacy using the
same variables, e.g. exit width, travel distance, number of people, compartment
size, e tc.
The method is also consistent with the principles of traditional quantitative
risk assessment. That is, it uses deterministic models based on theoretical and
empirical physical models to predict the consequences of fires and probability
theory to predict the likelihood of fires and their outcomes. There is also a balance
between the way that deterministic and probabilistic processes are m odelled.
Figure 11.2 shows a typical example of an F-n curve for life safety. The further
to the upper right the higher the risk and the further to the lower left the lower
1.E-05
Frequency of fatalities (people/occupant.year)
1.E-06
1.E-07
1.E-08
1.E-09
1.E-10
1 10 100
Number of fatalities
Figure 11.2 Typical output from SAFiRE

the risk. The individual curves show relatively high frequency/low consequence
fire events on the upper left and relatively low frequency/high consequence events
on the lower right.
The risk analysis showed that the overall risk to occupants satisfied the
acceptance criteria and is the same or less than that for a department store or
similar public building designed to meet the Danish Building Regulations 1 995.
11.4.1 4 Conclusion
The quantified fire risk assessment showed that, for certain parts of the
development, relatively simple changes to the design could significantly improve
life safety. This powerful combination of deterministic fire engineering and
probabilistic risk assessment provided a more rounded view of fire safety in the
development and may be a taste of things to come in countries with risk-informed
functional fire safety regulations.
11.4.2 Example 2: Life safety and asset protection in schools

A study of the cost/benefits of sprinklers in schools was undertaken using full
quantitative fire risk assessment (Charters, Salisbury and Wu 2004). The objective
of the study was to assess whether sprinkler systems in new and refurbished schools
are:
• a c ost-effective alternative to other fire precautions for life safety; a nd/or

• represent good value for money in terms of protection of p roperty.
The study assessed the risk-cost/benefits in four schools of varying size and
extent of new build/refurbishment. The types of school w ere:
1 refurbishment;
2 small all new build;
3 new build extension;
4 large new build.
The quantified fire risk assessments were undertaken for all four schools,
occupied and unoccupied and with and without sprinkler systems.
Table 11.1 shows the results of the risk–cost benefit analysis.
The net present value (NPV) is the total (aggregate) present values of the
annual benefits over the 30 years discounted at 6 per cent minus the capital c ost.
In the analysis, the cost of allocating temporary accommodation was
incorporated based upon £260 per 30 pupils (including toilets).
Table 11.1 Risk–cost benefit analysis
Average NPV with
annual temporary
benefits with accommoda-
NPV temporary tion included
Capital cost Average accommoda-
of sprinkler annual (after 30 tion included (after 30
School system (£) benefits (£) years at 6%) (£) years at 6%)
Refurbish- 53,350 2,566 +22,145 3,427 +10,300
ment
Small new 58,580 3,825 +10,060 5,074 –7,130
build
New build 50,550 6,194 –30,580 6,808 –39,030
extension
Large new 39,640 2,293 +12,210 3,173 +94
build
11.4.2.1 C
onclusion
The results for the small new build and new build extension schools indicated
relatively robust investments with pay-back periods of 22 years and 10 years
respectively.
Refurbishment and large new build showed positive NPVs and as such the
installation of sprinklers does not necessarily represent a cost-effective investment;
this is probably due to the relatively high level of compartmentation present within
the layout of the schools, coupled with the relatively low number of classrooms
that are likely to be affected by any one fire.
It is interesting to note that no allowance could be made in either analysis for
any excess or reduction in premium for the installation of sprinklers with respect
to insurance provisions. The wider decision-making process did, however, take
into account the social and educational benefits of fewer schools being lost due
to fire.
References
Charters D (2000), What does quantified fire risk assessment need to do to become an
integral part of design decision-making, SFPE International Conference on Performance
Based Codes, San Francisco, CA.
Charters D (2004), A review of fire risk assessment methods, Proceedings of Interflam ’04,
Interscience Communications, Cambridge.
Charters D and McGrail D (2002), Assessment of the environmental sustainability of
different performance based fire safety designs, Proceedings of the 4th International
Conference on Performance-Based Codes and Fire Safety Design Methods, SFPE, Melbourne.
Charters D and Wu S (2002), The application of ‘simplified’ quantitative fire risk assessment
to major transport infrastructure, SFPE Symposium on Risk, New O rleans.
Charters D, Paveley J and Steffensen F (2001), Quantified fire risk assessment in the
design of a major multi-occupancy building, Proceedings of Interflam ’01, Interscience
Communications, Cambridge.
Charters D, McGrail D, Fajemirokum N and Wang Y (2002), Analysis of the number
of occupants, fire growth, detection times and pre-movement times for probabilistic
risk assessment, Proceedings of the 7th International Symposium on Fire Safety Strategy,
International Society for Fire Safety Science,Worcester, M
A.
Charters D, Salisbury M and Wu S (2004), The development of risk-informed performance-
based codes, Proceedings of the 5th International Conference on Performance-Based Codes
and Fire Safety Design Methods, SFPE, Luxembourg.
12 Interactions
12.1 Introduction
For over a century, regulatory control for fire safety has mainly been achieved
through a framework of prescriptive rules for passive fire protection measures.
These rules are highly empirical and depend heavily on simple standard fire tests
and Codes of Practice. The rules are modified largely on the basis of experience.
The rules do not take sufficient account of the effectiveness of active measures
such as sprinklers, detectors and smoke ventilation s ystems.
A life safety level implicit in the fire regulations or any other level acceptable
to the society can be achieved through combinations of passive and active fire
protection measures, which are appropriate to the hazard involved, having due
regard to practicability. Any combination considered for the fire safety design of
a building must demonstrate an acceptable performance reflected in fatal and
non-fatal casualties likely to be sustained if a fire breaks out in the building. The
performance depends on the successful operation, reliability and effectiveness of
the passive and active fire protection measures included in the c ombination.
Combinations of safety measures produce interactions between measures
in their joint performance to provide a prescribed level of life and property
protection. A balanced fire safety system would recognise these interactions and
permit adjustments in requirements. This is commonly referred to as ‘trade off’ or
‘equivalence’. Among combinations providing equivalent safety, a property owner
may select one which is the most cost-effective. Fire safety measures also have
interactions and trade-offs with fire brigade operations.
Evaluating the interactions mentioned above is a complex problem requiring
the application of statistical/probabilistic methods, deterministic models and fire
safety engineering techniques. Sufficient statistical and experimental data are not
available at present for evaluating all the interactions.
12.2 Passive and active fire protection
12.2.1 Sprinklers and passive fire protection
Although there is a small chance, about 5 per cent, of not operating in a fire, if
maintained in a satisfactory working condition, sprinklers have a high potential to
extinguish a fire before it spreads beyond the object first ignited. If a fire is not thus
confined to the object first ignited, sprinklers can prevent a fire from involving all
the objects in a room. For the reasons mentioned above, sprinklers would reduce
the probability of ‘flashover’ in the room (or compartment) of fire origin which,
in turn, would reduce the probability of fire spreading beyond the room to other
parts of the building.
It is, therefore, reasonable to relax requirements specified in fire regulations,
codes and standards for buildings which have full sprinkler protection. Sprinklered
properties can be permitted to have one or more of the following c oncessions:
a. larger building size and height;

b. larger compartments;
c. increased travel time and distance to the entrance of a protected s taircase;
d. reduced widths for exits and staircases;
e. fewer staircases;
f. reduced fire resistance.
Some of the concessions mentioned above have already been incorporated in

some form or other in the fire safety codes of many countries.
12.2.1.2 Sprinklers and building or compartment size

The ‘power’ relationship between area damaged, D(A), and total floor area,
A, discussed in Section 3.3.1 (Equation (3.41)) can be applied to evaluate the
interaction between sprinklers and building or compartment size. Figures 12.1 and
12.2 are examples.
According to Figure 12.1, if a damage of 2300m2 is acceptable for a building
of total floor area 10,000m2 without sprinklers, the building size can be increased
to 33,000m2 if it is provided with sprinklers or to a slightly smaller size if the
reliability of sprinklers is less than 100 per cent, say, 95 per cent. Likewise, as
revealed by Figure 12.2, if a damage of 153m2 is acceptable, the size of a sprinklered
compartment can be increased from 500m2 to 4,000m2 or slightly less depending
on the reliability of the system.
12.2.1.3 Sprinklers and travel distance

Sprinklers have the potential to reduce significantly the number of fatal and non-
fatal casualties likely to be sustained if a fire breaks out in a large building occupied
Interactions 343
1000
Nonsprinklered building
Damage (m2)
Sprinklered building
100
10
1
10 25 50 100 250 500 1000 2500 5000 10,000 33,000
Total floor area of building (m2)
Figure 12.1 Damage and building size
by several people. Hence, for any floor of a sprinklered building, the maximum
travel distance to the entrance of a staircase can be increased up to a limit such
that life risk does not exceed an acceptable level.
An increase in travel distance will enable the construction of a smaller number
of staircases but life risk might increase due to a reduction in the number of
staircases. Sufficient data are not available, at present, for assessing the increase
in life risk due to an increase in travel distance or reduction in the number of
staircases for large sprinklered buildings. Hence, such a relaxation may not be
justified at present but may be considered if smoke ventilation systems are also
installed in addition to sprinklers.
1000
Damage (m2)
100
Nonsprinklered compartment
Sprinklered compartment
10
1
1 2.5 5.0 10 25 50 100 250 500 1000 2500 4000
Total floor area of compartment (m2)
Figure 12.2 Damage and compartment size
12.2.1.4 Sprinklers and staircase width

An increase in travel distance implies an increase in the ‘design evacuation time’,
t, for which 2.5 minutes has currently been recommended in UK fire regulations.
With this value for t, the width b required for a staircase is determined according
to the following equation:
M = Nbt + (n – 1) 50 (b – 0.3) (12.1)

Interactions 345
where M is the maximum number of people who can enter a staircase of width
b (in metres) serving n floors. The flow rate, N, is approximately 80 persons per
metre width per m inute.
Consider, for example, a six-storey building with 690 people designed for total
(simultaneous) evacuation. With M = 690, t = 2.5, and n = 6, calculations based
on Equation (12.1) show that the total width of a stair should be at least 1.7m.
Two 1.7m stairs would be necessary to meet the needs of ‘discounting’ of one stair
assumed to be not available due to fire or smoke.
If the building is sprinklered, the design evacuation time, t, may be increased to
say, 4 minutes such that, using Equation (12.1), b can be reduced to 1.34 m for M
= 690 and n = 6. Two 1.34 m stairs would be necessary to allow for ‘discounting’
of one stair.
A reduction in the total width of staircases for a sprinklered building will
produce a gain in rentable floor space but this action might increase life risk. This
increase in life risk may be compensated by an increase in life safety provided by
sprinklers.
12.2.1.5 Sprinklers and fire resistance

Sprinklers have the potential to reduce the severity likely to be attained in an
actual fire occurring in a compartment. Hence, there is a clear case for reducing
the fire resistance required for a sprinklered compartment from a level required
for a compartment without sprinklers. This reduction, for any level specified for
the probability of compartment failure, can be evaluated by applying the Beta
method discussed in Section 3.7.4. For any specified level for β , the value of mean
severity, µ s, in Equation (3.108) for a sprinklered compartment will be less than
that for an unsprinklered c ompartment.
For a compartment of any given size, the mean severity, µ s , in units of time
can be realistically assumed to be proportional to the logarithm of area damage
expected in a fire. This assumption is based on the exponential model discussed
in Section 3.3.2 – see Equation (3.42). This relationship between area damage
and severity can be estimated by carrying out some further research and statistical
analysis of data – see Ramachandran (1990). Sprinklers would reduce the area
expected to be damaged in a fire and, hence, would reduce the mean severity, µ s.
In the following, simpler, semi-probabilistic approach (Ramachandran, 1993),
total fire safety may be defined by probability C:
C = A*B (12.2)
where
A is the probability of ‘flashover’
B is the conditional probability of ‘structural failure’ given fl
ashover.
Probability of flashover, A, can be estimated from statistical data on area

damage – see Section 5.3.
‘Compartment failure’ would occur if the design fire resistance period of a
structural element is exceeded by the severity actually attained in a fire. The
probability of this event, B, depends empirically on the fractile value, F, of design
fire load density, L, selected for a compartment:
B = 1 – F (12.3)
Probability of structural failure, B, is 0.2 if F = 0.8.

L is one of the parameters in a formula for estimating fire severity in a
compartment of given dimensions, ventilation factor and thermal inertia. This
formula is based on the concept of equivalent time of fire exposure in which fire
resistance required is equated to fire severity.
If Lo is the design fire load density for a non-sprinklered compartment
corresponding to a fractile Fo of the load density d istribution:
Bo = 1 – Fo (12.4)
For a sprinklered compartment:
Bs = K * Bo; K = Ao/As (12.5)
Fs = 1 – Bs (12.6)
where As is the probability of flashover for a compartment without sprinklers and

As the probability of flashover for a compartment with s prinklers.
The design fire load density Ls, for a sprinklered compartment, would correspond
to the fractile Fs. Ls will be less than Lo since Fs will be less than Fo. If K = 3.5, as
on average for many occupancy types, and F0 = 0.8, Bo = 0.2 such that Bs = 0.7
and Fs = 0 .3.
The sprinkler factor S is defined as the ratio
S = Ls/Lo (12.7)
The value of S is about 0.6 for office buildings, retail premises and hotels.
Hence, the fire resistance of a sprinklered compartment of these occupancies
can be about 60 per cent of the resistance specified for a compartment without
sprinklers – see Ramachandran (1993).
12.2.1.6 Sprinklers and smoke ventilation systems

The interaction between sprinklers and smoke ventilation systems has not, so
far, been clearly evaluated by fire scientists and engineers. This interaction arises
from the fact that water sprays from activated sprinklers remove buoyancy from
combustion products and generate air currents which counter the outflow from
the vents, and additionally, transport combustion products to the floor. Use of
Interactions 347
vents may reduce the total water demand by the sprinkler system but this benefit
due to vents has not yet been clearly established. Sufficient statistical information
needs to be collected to assess the effects of the interactions on damage to life and
property and the performance of the fire service.
It is arguable whether a vent should operate before the operation of a sprinkler
or not. There are indications from current research that the effect of venting on
the opening of the first sprinklers and their capacity to control the fire is likely to be
small. There are also indications that the earlier vents are opened, the more likely
they would be effective in preventing smoke-logging of a sprinklered building. The
controversy between sprinklers and vents can, perhaps, be resolved by deciding,
for any type of building considered, whether property protection or life safety is
the main objective. In the initial stage of fire growth, a vent should operate before
a sprinkler if life safety is the dominant objective, e.g. hotels, shopping centres,
office buildings. In industrial buildings, the first sprinkler may operate before the
opening of any vent (Ramachandran 1998).
12.2.1.7 Sprinklers and automatic fire detectors

For discovering the existence of a fire, sprinklers installed in industrial and
commercial buildings are generally less sensitive in the sense that their operating
times are longer than those of automatic detectors. Normally, it would take about
6 minutes before the heat generated by a fire is sufficient to activate a sprinkler
head, whereas a detector is designed to operate in 2 or 3 minutes after the start
of the ignition. Consequently, if both of these systems are installed in a building,
early discovery of a fire by the detector is likely to be followed by first-aid fire
fighting such that the sprinkler system is not brought into a ction.
12.2.1.8 Automatic fire detectors and means of escape facilities

By discovering the existence of a fire very soon after the start of the ignition,
automatic detection systems provide extra time for the escape of occupants of
a building. Consequently, detectors have a great potential for reducing life risk.
Hence, for a building fitted with detectors, particularly advanced types such as
computer-controlled addressable systems, design evacuation time and travel
distance to the entrance of a staircase can be increased or the total width of stairs
reduced.
This concession appears to exist in some form in the fire regulations of
some countries. In the UK, for example, a building can be designed for ‘phased
evacuation’ if it has been fitted with an appropriate fire warning system. This
would enable narrower stairs to be built than would be the case if the building had
to be designed for total (simultaneous) evacuation.
12.3 Smoke movement and evacuation of occupants
As defined in Sections 3.7.3, 3.7.4 and 9.3, the evacuation time of occupants of
a building is the third main component period of the total evacuation time. The
first and second component periods are concerned with the detection or discovery
time of a fire and recognition time (Section 9.1). For a successful evacuation, the
total evacuation time should be less than the time taken by a combustion product
to travel from the place of fire origin and produce an untenable condition on any
segment or component of an egress route. Otherwise, egress failure would occur,
with consequences leading to fatal or non-fatal injuries to one or more occupants.
The probability of occurrence of egress failure should be reduced by reducing
the total evacuation time and the actual evacuation time, which is the time taken
by an occupant or group of occupants to leave their places of occupation in the
building and reach safe places inside or outside the building. To achieve this
objective, for any target level specified for the probability of egress failure, a semi-
probabilistic approach (safety factors) was recommended in Section 3.7.3 and a
probabilistic approach (Beta method) in Section 3.7.4. Following one of these
methods, appropriate design values can be determined for the total evacuation
time and the actual evacuation time. To achieve these design values, the building
may be equipped with necessary detection and communication systems and
means of escape facilities which include staircases, emergency lighting and smoke
ventilation systems.
As discussed in Section 9.3, smoke is the major combustion product threatening
occupants, particularly those who are not in the immediate vicinity of a fire. The
time taken by smoke to produce an untenable condition on an escape route
depends on the rate (as a function of time) at which large quantities of smoke
are produced and the location of the escape route with reference to the location
of the fire. Egress failure or success also depends on the location of an occupant
or group of occupants with reference to the location of the escape route. It also
depends on the mobility conditions (disabled etc) of the occupants, as discussed
in Section 9 .3.
Sufficient statistical data are not available, at present, for evaluating
quantitatively the interaction between smoke movement and evacuation of
occupants. Simulations based on deterministic models may provide necessary
data on smoke movement. Likewise, data on the movement of occupants can be
obtained by performing simulations, based on computer models, for evacuation. It
may not be necessary to carry out evacuation exercises or drills which are costly
and time c onsuming.
12.4 Fire protection measures and fire brigade
12.4.1 Building design and fire brigade

One of the objectives of providing fire resistant compartments to a building is
to ensure that a fire is contained within the compartment of origin, for a length
Interactions 349
of time which is sufficient for the fire brigade to arrive at the fire scene and
commence fire fighting, before the fire spreads beyond the compartment. The fire
resistance of a compartment should, therefore, be determined with due regard to
the attendance or response time of the brigade, which is a function of the travel
distance or time from a fire station to the building.
Fire spread beyond a compartment depends on the size of the compartment,
apart from its fire resistance and other factors such as fire load and area of
ventilation. Fire spread in a building also depends on the design and size of the
building, particularly the building height. Compartments should not, ideally,
extend to more than two or three floors in low-rise buildings and not more than
one floor in high-rise buildings. Each floor (including basements) in a high-rise
building should be a fire-resistant compartment. Fire brigades usually find it
difficult to fight fires in high-rise buildings and basements.
The factors mentioned above should be taken into account for specifying or
relaxing requirements in fire regulations for size and fire resistance of compartments
and for size and height of buildings. These factors have been considered to
some extent in fire regulations but their effects have not, so far, been assessed
quantitatively with due regard to interactions.
12.4.2 Sprinklers and fire brigade

Sprinklers extinguish several fires which are not even reported to the fire brigades.
Some fires to which the brigades are called are also put out by the system before the
brigade arrives at the premises involved. If they fail to extinguish a fire, sprinklers
would slow down the rate of fire growth and restrict fire spread until the arrival
of the brigade.
The performance characteristics of sprinklers mentioned above, together with
their capability for giving early fire warning, would enable a brigade to attend a
smaller number of fires in sprinklered premises and to control and extinguish fires
quickly in these premises. Statistical data are available in the UK for estimating
the reduction in brigade control time due to sprinklers. For the reasons mentioned
above, sprinklers would reduce considerably the time (and cost) spent by fire
brigades in responding to fire incidents.
The interaction between sprinkler performance and fire brigade operation
might affect the number of fire stations required to cover a particular city or
area. In the USA, for example, a fire station in Fresno had to be closed down
after almost all buildings in two districts in this city were fitted with complete
automatic sprinkler protection in 1970. For providing protection to life, sprinklers
are required to be installed in apartment buildings in many cities in the USA and
Canada. Scotsdale in Arizona is an example where sprinklers have been installed
in residential buildings.
12.4.3 Automatic detection systems and fire brigade
Early discovery or detection of a fire would enable the commencement of fire
fighting by first-aid methods and/or fire brigade when the fire is small in size. This
would reduce the time required by the brigade to bring the fire under control.
Statistical studies indicate that, on average, the control time will be reduced by
half a minute for every minute of early arrival of the brigade at the fire scene (see
Table 10.2). The ‘attendance time’ is the period from the time when information
about the fire is received by a fire brigade, to the arrival of the brigade at the scene
of fire.
Since automatic fire detection systems have the potential to reduce, indirectly,
the average control time for fires in an area, the average attendance time for
this area can be increased up to a limit, such that risk to life and property does
not exceed an acceptable level. An increase in attendance time would lead to a
reduction in the number of fire stations required for an a rea.
The number of fire stations required for an area protected (or not) by detectors
depends on the relationship between ‘service time’ and fire damage (to property and
life). Attempts have been made to establish this relationship, which can provide
inputs for cost-benefit comparisons and trade-offs between the effectiveness of
detector systems and fire b rigades.
Automatic fire detection systems are known to produce false alarms by
operating when there are no real fires. False alarms cause wastage of time and
money for a fire brigade.
12.4.4 Smoke ventilation systems and fire brigade

One of the objectives of installing smoke ventilation systems is to facilitate fire
brigades by enabling fire fighters to enter a building involved in a fire and locate
easily the seat of the fire. The extent to which this object is achieved can be
assessed by comparing the average control time of fire brigades for fires in buildings
with ventilation systems with that for fires in buildings without these systems.
Data available at present are not sufficient for carrying out this investigation.
However, tentative conclusions can be drawn by analysing data on fires
collected by leading manufacturers of smoke ventilation systems. These data can
be combined with those contained in reports on fires furnished by fire b rigades.
Such combined data were analysed some years ago using Fire Case Histories
published by Colt International in October 1975 – see Ramachandran (1998). These
case histories related to 73 fires which occurred during 1957 to 1974. Fire brigade
reports were identified for 55 of these fires which occurred in the UK. According
to the information contained in these reports, the average control time was 126
minutes for 19 fires in buildings without ventilation systems or sprinklers. The
average control time was 57 minutes in 26 fires in buildings with only ventilation
systems and 64 minutes in 7 fires in buildings with both vents and sprinklers.
The figures mentioned above, although based on small samples of data,
indicate that fire brigades will be able to bring under control fires in buildings
Interactions 351
with ventilation systems quicker than fires in buildings not equipped with these
systems.
References
Ramachandran, G (1990), Probability-based fire safety code, Journal of Fire Protection
Ramachandran, G (1993), Fire resistance periods for structural elements – the sprinkler
factor, Proceedings of the CIB W14 International Symposium on Fire Safety Engineering,
University of Ulster, Jordanstown.
Ramachandran, G (1998). The Economics of Fire Protection. E & F N Spon, L
ondon.
13 Combining data from various
sources – Bayesian t echniques
13.1 Introduction
The results of a risk assessment are sensitive to the probabilities attached to
parameters such as the probability of a fire starting and probable loss if a fire
occurs. Sufficient information to estimate these quantities for a particular building
is unlikely to be available. In such cases, a ‘prior’ or initial assessment of the
probabilities can be made by consulting an expert such as a fire safety engineer or
by analysing statistical data for a group of buildings with similar fire r isk.
A prior assessment of the probability estimated for any parameter can be revised
in the light of expert opinion and/or data which may be available for the building
considered. These data may be provided by a quantitative assessment of fire risk
in that building and a sample of fires which occurred in the building. Bayesian
techniques can be applied to revise a ‘prior’ estimate of the probability and obtain
a ‘posterior’ estimate (Ramachandran 1998).
Bayesian statistical techniques provide a mechanism for combining subjective
and objective information from two or more sources: national data, sample data,
expert opinion, experimental results and simulations based on deterministic (zone
or field) m
odels.
13.2 Bayes’ t heorem

According to Bayes’ theorem (La Valle, 1970), the ‘posterior’ probability, P(Hi/A),
of occurrence of event Hi given observation A is proportional to the ‘prior’
probability, P(Hi), times the likelihood of A, given Hi, P(A/Hi):
P (Hi / A ) ∝ P ( A / Hi ).P (Hi ) (13.1)
If H1, H2, …, Hp are a set of (exhaustive, mutually exclusive and independent)

events,
p
∑ P (H ) = 1
j=1
j (13.2)
Combining data from various sources – Bayesian techniques 353
To replace the proportionate sign, ∝ , in Equation (13.1) by an equal to sign,
the equation is ‘normalised’ with respect to the set of events in Equation (13.2).
This process is achieved by dividing the right-hand side of Equation (13.1) b y
p
∑ P( A / H ).P(H )
j=1
j j
(13.3)
We can therefore write

P ( A / Hi ).P (Hi )
P (H i / A ) = p (13.4)
∑ P( A / Hj ).P(Hj )
j=1
13.3 Probability of fire recurrence
13.3.1 Example 1
For a particular industrial building of a specified type and size, a power function
(Section 3.3.1) based on national fire statistics provided a certain estimate for
the annual probability of fire occurrence. This ‘global’ figure was adjusted for the
building considered by a statistician who carried out a risk assessment and took
into account the human (e.g. careless disposal of cigarettes) and non-human (e.g.
faulty electrical appliances) sources of ignition actually present in the different
parts of building. The adjustment followed the method described in Section 5.1.2
and Equation (5.2) and produced an estimate of 0.45 for the annual probability of
fire occurrence for the property considered.
During a recent five-year period, four fires occurred in the building considered,
indicating an annual probability of 0.8 for fire occurrence. This figure is
considerably higher than 0.45. Statistically, however, with a higher variance, the
estimated figure of 0.8 provided by a small sample is likely to be less reliable than
the adjusted estimate of 0.45 based on a much larger sample provided by the
national fire statistics. It would be appropriate, however, to combine both the
above estimates using Bayes’ theorem and obtain a composite estimate for the
probability. The following procedure may be a dopted.
The figure of 0.45 may be considered as the prior probability P(H1) for the
occurrence of a fire during a year with P(H2) = 0.55 for the probability of non-
occurrence, with only two events p = 2 in Equations (13.2) to (13.4). If the
information provided by the recent fires in the building is denoted by the letter A,
P(A/H1) = 0.8 and P(A/H2) = 0.2. Then applying Equation (13.4):
0.8× 0.45
P (H1 / A ) = = 0.77 (13.5)
( 0.8× 0.45) + ( 0.2× 0.55)
The ‘prior’ probability of 0.45 has been revised upwards to the ‘posterior’
probability of 0.77 to give weight to the data on fire occurrence in the property
considered. The ‘posterior’ probability for non-occurrence of a fire, P(H2/A), is
hence 0.23 which is lower than the prior. However, according to the fire safety
engineer who carried out a risk assessment using his own method, the annual
probability of fire occurrence was only 0.4. In this case, with P(A/H1) = 0.4 and
P(A/H2) = 0.6, a calculation as in Equation (13.5) would show that the prior
probably 0.45 for fire occurrence needs to be revised downwards to the posterior
probability of 0.35; the revised posterior probability for n
on-occurrence is 0 .65.
It will be apparent that a value for P(A/H1) exceeding 0.5 will produce a
posterior probability P(H1/A) greater than the prior while a value for P(A/H1) less
than 0.5 will produce a posterior less than the prior. A calculation would also show
that P(H1/A) and P(H1) would be equal to 0.45 for the example considered, if
P(A/H1) = 0.5. In this case, there is no need to revise the initial (prior) assessment
of the probabilities P(H1) and P(H2) which are 0.45 and 0.55 for the example.
The prior annual probability of 0.45 and the revised annual probability of 0.77
(Equation (13.5)) for fire occurrence may be considered as ‘objective’ estimates
based on statistics from real fires, while the value of 0.4 for this probability
estimated by the fire safety engineer, although based on expert opinion, is a
subjective assessment, not based sufficiently on facts and figures. Instead of the
‘objective’ prior probability of 0.45, the revised ‘objective’ probability of 0.77 may
be considered as the prior since it has taken into account the recent data on fire
occurrence in the property considered. In this case, P(H1) = 0.77, with P(H2) =
0.23. This prior may be combined with the subjective estimate 0.4 to give weight
to the opinion of the fire safety engineer. Combining 0.77 and 0.4, with P(A/H1)
= 0.4 and P(A/H2) = 0.6, a calculation as in Equation (13.5) would show that a
more realistic estimate of annual probability of fire occurrence, P(H1/A), is 0.69.
The corresponding annual probability of non-occurrence of fire, P(H2/A), is 0.31.
These estimates are based on information from three sources: global statistics,
data for the property considered and expert opinion.
13.3.2 Example 2
Bayesian techniques can be applied to the probability of financial loss or area
damage exceeding a specific level. According to national fire statistics, for example,
the probability of loss in a fire in an industrial building exceeding £500,000 may be
0.05. The probability of area damage exceeding 100m2 may be 0.08. These national
estimates can be used as prior probabilities for any particular industrial building
and revised on the basis of expert opinion, a sample of fires in that building and
estimates provided by deterministic models.
Consider, as another example, the probability of a fire spreading beyond
the room of origin in an industrial building. This prior probability, according
to national statistics, may be 0.07 if the room has no sprinklers and 0.02 if it
has sprinklers. According to a deterministic model, however, the probability for
this undesirable event may be 0.6 and 0.3 in the unsprinklered and sprinklered
cases (respectively). An application of Bayes’ theorem would provide ‘posterior’
probabilities of 0.101 if unsprinklered and 0.009 if sprinklered.
13.3.3 Example 3
One may be interested to estimate the probability that a certain number of fires
would occur in a particular type of building during a given period of time, say a year.
It is reasonable to assume that the fires would occur independently of one another
and with a constant tendency to occur. Under this assumption, the frequency of
fire occurrence is likely to follow the Poisson distribution (Ramachandran 1998).
Several factors can cause uncertainties in the value of the parameter θ of
the Poisson distribution providing estimates of the probabilities of different
frequencies of fire occurrence during a given period. The probabilities and hence
the value of the parameter are likely to vary from one period to another. The
value of the parameter needs to be revised or updated continuously as more and
more information becomes available on the frequencies of fire occurrence and
the factors affecting them. This task can be accomplished easily by applying, as
discussed in the Sections 13.3.1 and 13.3.2, the discrete form of Bayes’ theorem
presented in Equation (13.4). In this equation, the letter Hi will be replaced by
θi denoting a particular value for the Poisson parameter θ. P(θi) is the probability
that one assigned, prior to obtaining the new information A, to θi being the
correct value. P(A/ θi) is the probability that the new information A would have
been observed given that θi is true. P(θi /A) is the posterior probability that θi is the
correct value after obtaining the new information A. The parameter θ may have p
different values θj with j = 1, 2, …, p.
A more complex problem is concerned with the estimation of the posterior
probability distribution of fire occurrence given that the prior distribution is
Poisson. This problem was discussed by Johansson (2003) who considered the
Gamma distribution which is a ‘conjugate’ distribution of Poisson. If one considers
the Gamma distribution for the probabilities of frequencies of fire occurrence,
the resulting posterior distribution will also be a Gamma distribution but with
different parameter values.
Another example referring to the probability distribution of occurrence of an
event is concerned with uncertain situations involving only two possible, mutually
exclusive events such as a sprinkler system extinguishing or not extinguishing a
fire or an automatic fire detector operating or not operating when a fire occurs. For
such cases, the binomial distribution can be used to denote the probability, p, of a
sprinkler extinguishing a fire or of a detector operating in a fire. Such a situation
was studied by Sui and Apostolakis (1988) who used Bayes’ theorem to combine
indirect evidence with direct evidence for the demand availability of a sprinkler
system.
The Beta distribution is the ‘conjugate’ distribution that is applicable when the
probability of obtaining the evidence, P(A/p), can be calculated using a binomial
distribution (Ang and Tang 1984). Using a prior distribution in the form of Beta
distribution, one can simplify the use of Bayes’ theorem considerably. This problem
was discussed in detail by Johansson (2003). He showed how Bayes’ theorem
can be used to update the parameters of the prior Beta distribution to provide
estimates of posterior values of the parameters.
Johansson (2003) also discussed the application of a multi-nomial distribution
for situations involving more than two possible, mutually exclusive events. He
used the Dirichlet distribution as a ‘conjugate’ distribution for the multi-nomial
distribution. A Dirichlet distribution with only two parameters is the same as
a Beta distribution. Thus following the method developed by him for the Beta
distribution, Johansson discussed the calculation of posterior values of the
parameters of the Dirichlet distribution. He has also proposed a Bayesian network
model for the continual updating of fire risk measurement.
13.4 Probable loss in a fire

The probable or expected loss in a fire, whenever it occurs, can be estimated
by identifying and using the probability distribution of loss and its parameters
– such as mean and standard deviation. Research studies have shown that this
distribution can be Pareto or log normal, as discussed in Section 3.3.4. One of
these two distributions can be identified as a ‘prior’ distribution for any occupancy
type by collecting and analysing past data on fire losses occurring on buildings of
this type. Estimation of the ‘posterior distribution’ when the ‘prior distribution’ is
Pareto or log normal is a complex statistical problem.
Statistical techniques are available for ascertaining whether the probability
distribution of fire loss for any occupancy type is Pareto or log normal or any other
distribution such as Weibull. However, none of these distributions can be expected
to fully account for the uncertainties involved in the development and spread of a
fire in a building and the physical or financial damage sustained.
To achieve a substantial reduction in the uncertainties it would be worthwhile
developing a composite model which incorporates all suggested plausible
probability distributions for fire loss. Such a model was attempted by Shpilberg
(1974) using weights for the distributions before observing the data by a prior
assumption regarding the likelihood of any one of the postulated distributions being
the ‘true’ one. After observing the data, the composite model weighs the different
distributions by a posterior probability that is a function of the prior assumption
and the data, thereby modifying prior knowledge by observed information.
13.4.1 A heavy goods vehicle (HGV) fire in a tunnel

Carvel et al. (1999) considered HGV fires in tunnels from a probabilistic viewpoint.
The authors described the effect of forced longitudinal ventilation on fire size by
defining the coefficient k, based on the heat release rate (HRR):
HRR of fire with forced ventilation
k=
HRR of fire with naturral ventilation
Bayes’ theorem was expressed as follows:
P (I / k = k i ).P (k = k i )
P (k = k i / I ) = n
(13.6)
∑ P (I / k = k ).P (k = k )
i=1
i i
where
P(k = ki) is the prior estimate of the probability that k = ki
P(I/k = ki) is the likelihood of evidence I given k = ki and
P(k = ki/I) is the posterior probability, that is the probability of hypothesis ki updated
in the light of evidence, I.
In the Bayesian method, it is necessary to consider a range of hypotheses, k1, k2,

…, kn only one if which can be true for a specific situation.
A ‘Delphi’ like panel of eight independent fire engineering and firefighting
experts were presented with a description of the tunnel and the HGV. The experts
were asked to provide their estimates of the HRR for two scenarios, a time t after
ignition (growth phase) and after a further period of time when the fire becomes
fully developed. The estimates collected from the experts were converted into
values of k and collated to give a ‘prior’ probability distribution for k in order to
keep to a minimum the errors in individual judgements. Originally, it was hoped
that each expert would be able to provide estimates of the likelihood values as well
as prior estimates. However, finally the principal author of this study estimated the
likelihood values as he was considered to be independent of all the prior estimates
and also very familiar with all of the evidence data.
For updating the prior probabilities as estimated above, results of four
experiments providing values of k were considered. These experiments w ere:
• Hammerfest HGV test

• Hammerfest wooden crib test
• Buxton simulation of HGV test and
• Buxton wooden crib test.
The estimates of the likelihoods for the first test in the growth phase were applied to
the prior probabilities using Bayes’ formula (Equation (13.6)). The series of posterior
probabilities produced were used as the new prior probabilities and the likelihoods
for the second test were applied. This process was repeated for the fourth and third
tests. It was observed that the third test was not appropriate for the calculation
of k values in the growth phase of a HGV fire. These data were discounted and
the posterior probabilities recalculated. The same process was repeated for k values
of fully involved HGV fires. The expectations of k for each airflow velocity was
calculated from the posterior probabilities using the Bayesian estimator:
E (k ) = ∑ k iP (k = k i / I ) (13.7)
The Bayesian analysis revealed that if an HGV fire in a tunnel is subjected to a

2ms-1 ventilation airflow, then the fire may increase in severity seven times faster
than under natural conditions and the maximum HRR may be up to four times
greater. If a 10ms-1 airflow is used, it may grow up to 22 times faster and ultimately
be eight or nine times more severe, and so on. It would be interesting to compare
these predictions with estimates made from the data directly.
References
Ang A H S, and Tang W H, (1984), Probability Concepts in Engineering, Planning and Design,
Volume 2 – Decision, Risk and Reliability, Wiley, New Y ork.
Carvel, R O, Beard, A N and Jowitt P W (1999), The effect of forced longitudinal
ventilation on a HGV fire in a tunnel, Proceedings of the International Conference on
Tunnel Fires, Independent Technical Conferences, Lyon.
Johansson, H (2003), Decision Analysis in Fire Safety Engineering – Analysing Investments
in Fire Safety, PhD Thesis, Lund University, Institute of Technology, Department of
Fire Safety Engineering.
La Valle, I H, (1970), An Introduction to Probability, Decision and Inference, Holt, Rinehart
and Winston, New Y ork.
Shpilberg, D C (1974), Risk Insurance and Fire Protection: A Systems Approach. Part 1:
Modelling the Probability Distribution of Fire Loss Amount. Technical Report, FMRC,
Serial No. 22-431, RC74-TP-23, Factory Mutual Research Corporation, Norwood, MA.
Sui, N and Apostolakis, G (1988), Uncertainty, data and expert opinions in the assessment
of the unavailability of suppression systems, Fire Technology, 24, 2, 138–162.
Index
absorbing boundary 119 As Low As Reasonably Practicable see

absorbing state 122 ALARP principle
absorption law 81 assumptions, assessment of implications
acceptance criteria 175–94; absolute 153
181–8 ; comparative 177–80; atria, design fire size 241–2
financial objectives 188–93; attendance time: definition 311, 350; and
individual risk levels 182; multiple- fire damage 316–19; and fire loss 317,
death fires 185–8; non-financial 318; and fire spread 257; and life loss
objectives 193; in qualitative risk 317–19
assessment 177; in semi-quantitative audit 13
risk assessment177; societal risk levels automatic fire detector (AFD) systems
182–5; types 176–7 250–2, 269–78; addressable 274, 275,
accident frequency categories, look-up 278; analogue 278; causes of failure
matrix 47 272; components 275 ; components
accident probability, look-up matrix 54 failure rates 275–6); cross-zoning
accident severity, look-up matrices 47, 54 of detectors 278; detector heads,
accident trigger probability, look-up ‘series’ vs ‘parallel’ arrangement 277;
matrix 46 effectiveness 270–2; failure rates
ADAC fire risk assessment 15, 62–3 273–4, 275–6; false alarms 272, 273,
AFD systems see automatic fire detector 276; and fire brigade 350; and means
(AFD) systems of escape facilities 347; multi-mode
alarms 278–85; conventional 278–9; detectors 278; multi-state 278;
ignored 297; informative fire operating times 251, 270 ; operating
warning (IFW) systems 279–85, times design values 250, 251;
300; laboratory experiments 281–5; performance 269–70; reliability 272–8;
psychological criteria 280 and sprinklers 347
ALARP principle 9, 176, 177 availability 84–5
ambiguity 295
AND gates 78, 79, 94–5 base events 78–9
area damage: acceptable upper limit Bayes’ theorem 352–3
for 189–90; and building size 342, Bayesian techniques 352–7
343; and compartment size 342, 344; beam deflection, effect on fire-resisting
correlation with heat output 239, 248; wall 264, 265
correlation with injuries 138, 139; benchmark, code-compliant building as
flashover probability estimation 220–2; 180
frequency distribution 115; input Bernoulli probability 128
variables affecting 134; probability Beta distribution 355, 356
distributions 112–14, 242; probability Beta method 155, 162–9, 260, 306;
of exceeding specific level 354 bivariate approach 164–6; log normal
360 Index
safety index 168–9; probabilistic design combustion product time, design value 161
criterion 162–3, 250, 306; safety factor combustion toxicology 233
in 166–8 Comparison of Risk Indices by Simulation
BFSM 135 Procedures (CRISP) 134–5
bias 137; assessment of implications 153 compartment size, and flashover
binomial distribution 355 probability 224–5
Boolean manipulation 80–2; normal compartmentation performance and
arithmetic vs 80–1 reliability 258–65; probabilistic
boundary layer 303 approach 258–61; reliability approach
brainstorming, keys to 50 261–5
break-glass units, failure rates 275–6 Competent Authority (CA) 23
BRESENS 279, 281 complement: of event 72, 81; of set 75
British Standards Institution: British compound events 72
Standard BS 5588 159, 299, 305; computational fluid dynamics 151
British Standard BS 7974 333–5; data conditional probability 73
sources and quality guidance 153 conductive heat transfer 146–7
building design, and fire brigade 348–9 confidence limits 106
building evacuation see evacuation conflagration, probability 267
building fire safety, not addressing all consequence analysis 136–51, 331;
aspects 30–1 disasters and near misses 138–9;
Building Fire Simulation Model (BFSM) example 337; experiments 139–40,
135 141; fire tests 139–40; historical data
Building Regulations 10–11, 28, 332–3 137–8; objectives 136–7; see also
building structure, performance and modelling
reliability 265–7 consequence aversion 8
bus, engine compartment fire 140, 141
consequence probability distribution 138
bus garages, fires in 90–4, 192–4
consequential loss, target levels 190–1
business continuity, buildings where
conservation of energy 144–5
affected 6, 328
conservation of mass 143–4
business risk: look-up matrix 56; total 49
conservation of species 144
Buxton simulation of HGV test 357
Buxton wooden crib test 357 control time 285, 311, 321–2, 350; and
arrival time 321–2; and fire area 325
casualties: causes of 233; location of control units, failure rates 275–6
230–2, 269; rate per fire 234–5 control volume models 141–2; application
CEA see European Insurance Committee in fire safety 148; application to fires
certification 25, 85 142–8 ; assumptions 142; conservation
characteristic values 156–60 equations 143–5; mass and heat
Chebyshev inequality 156 transfer sub-models 145–8; source
checklists 17, 35–6, 38 terms 145)
chemical plant installations 21–2 convection, fire spread by 255, 257
CIB Design Guide 259, 263 convective heat transfer 146
CIBSE TM19 140 cooking appliances, misuse 233
CIMAH regulations 22 corporate governance 8–9; principles 8
coded messages 279 corporate liability 9
Codes of Practice 333–5, 341 cost of asset damage, look-up matrix 54
coincidence technique 278 cost-benefit analysis 92, 192–3, 338–9
collapse, of structural barriers 191–2, 255, cost per life saved 180
264–5 cost of performance penalties or
collinearity 110 operational loss, look-up matrix 55
COMAH regulations 22 CRISP see Comparison of Risk Indices by
combinations 74 Simulation Procedures
combustible materials 35–6; hazard level cut sets 82–3; minimal 82–3, 95
determination factors 37 cyclohexane 21
Index 361
Danish Building Regulations 335, 336, 338 equivalence 11, 341
De Morgan’s laws 82 equivalent time of fire exposure 157, 259
Dean Schedule 61 escape stairs, capacity 304
decaying period 202 escape system 300–1
defence in depth 2, 6–7, 336, 337 established burning, time before 225
Delphi 17–18, 60 European Insurance Committee (CEA) 59
density of occupant flow 302 European Tunnel Directive 26–7
depth-first search of a graph 126 evacuation 298–301; design model
design: objectives 4; parameters 4 159–60; phased 304–5, 347; smoke
design base accident (DBA) analysis 20–1 movement and 348
design evacuation time (E) 159, 249–50 evacuation analysis 149, 302; discrete
design fire size 189–90, 214–15, 239–53; 151, 152
current concept 240–7; definition 239; evacuation time; design (E) 159, 299,
probability–time based 247–53 301–8, 344–5; design value 161, 250;;
design parameters 301 and fatality rate 307–8; minimum 304;
design point 162, 261, 306 total (H) 159, 249–50, 299, 306
design process, adding value to 30 event space 77
design values 160–2 event space method 126–7
detection: automatic see automatic fire event tree analysis 88–94; bus garage
detector (AFD) systems; human 268–9 example 90–4; in flashover probability
detection time (D) 159, 249 estimation 218, 219–20
deterministic approaches 4–5 event trees 16–17, 88–9, 122–5;
deterministic processes 71 construction 89
Dirichlet distribution 356 events: algebra of 77–83; base 78–9;
disabled people, evacuation 299 combinations of 78; complementary
disasters, insight from 138–9 72, 81; compound 72; experiments
discrete hazard function 308 and 77; frequency estimation 91;
dispatching time 311 identifying 91; independent 73, 89;
disutility function 191 mutually exclusive 72, 89; see also top
doors: closing devices 261–2; fire events
resistance 260 EXCLUSIVE OR gate 79
double knock technique 278 experiments 71–2; and events 77; insight
doubling times 105, 288; paper, printing from 139–40, 141
and publishing industries 206, 207, explosion 265, 266
210, 212 exponential distribution 84, 111, 116
Dual Independent Map Encoding (DIME) exponential models 101–5, 119, 201–2,
file 313 203–8; applications 213–15; in
ductility, of load-bearing capacity 259 flashover probability estimation 216–19
extreme order statistics 116
early detection, economic value 104 extreme order theory 116
early intervention, effectiveness 323–4 extreme value distributions 116–17
‘effective’ fire risk 59 extreme value theory 71, 261, 265
effective heat of combustion 216, 243
effective width 303 factorial expressions 74
egress, deterministic calculation 4 Factory Mutual Overall Rating 111
egress failure, probability of 306–7 FAFR 235–6
electrical components, failure rates 276 false alarms 272, 273; categories 276;
emergency equivalent network 128 classification 273; rates 276
energy, conservation of 144–5 Farmer curve see Fn curves
engineering principles, as basis for models Fatal Accident Frequency Rate (FAFR)
28–30 235–6
entrainment in plumes 146 fatality rates; automatic detectors and
environmental protection, buildings where 271–2; design evacuation time and
affected 6, 328 307–8; discovery times and 107–9,
362 Index
268, 271; probability distributions fire load 213
114–15 fire load densities 157, 213, 244, 259, 346
fault tree analysis 94–8; algebra of events fire losses: attendance time and 317, 318;
in 79–80; hospital wards example averages per fire 117; costs 50; factors
96–8; notation 77; process 81 affecting 111; probability distribution
fault trees 17, 78, 94; general form 95 for 111–12, 356; residential 228;
Fields shopping centre 335–8 survivor probability distribution 118;
financial loss; expected value 228–30; see also financial loss
probability distribution 228; fire occurrence frequency 355–6
probability of exceeding specific level Fire Offices’ Committee 61
354; see also fire losses Fire Operations Simulation Model 313
fire alarms see alarms fire outbreak chances see frequency of
fire brigades see fire services ignition
Fire Case Histories 350 Fire Precautions Act (1971) 25
fire damage measurement 225–36; extent fire prevention activities 310
of fire spread 226–8; financial loss fire protection systems: and fire brigade
228–30; floor area damaged 226–8; 348–51; operation times, design
total performance 236; see also area values 249, 250; see also automatic
damage; fire spread; life loss fire detector (AFD) systems; sprinkler
fire damage risk: by occupancy 100; for systems; ventilation systems
industrial buildings 101, 102 fire recurrence, probability of 353–6
fire disasters: delay in time to start to fire regulations, existing buildings 11
move 296; learning lessons from 7 Fire Research Station (FRS) 134, 214,
fire discovery times: classification 103, 279–80, 286, 322
268; and fatality rates 107–9, 268, 271 fire resistance of compartment 156–8;
fire drills 301 probabilistic approach 258–61;
fire engineering 3–4, 11; application to reliability approach 261–5; sprinklers
design of buildings 333–4; definition and 345–6
3; objectives 3 fire risk assessment: adoption examples
Fire Engineering Design Brief (FEDB) 34 18–27; development background 7–18;
fire events: not recreating exact 31; factors indicating need 6–7; general
trends 50 principles 27–31; uses 5
fire extinguishers, portable 285–6 Fire Risk Assessment by Simulation (Fire
Fire Extinguishing Trade Association sim) 135
(FETA) 285–6 fire risk assessment methods see qualitative
fire fighting: and arrival time 321–2; fire risk assessment methods;
control time 285, 311, 321–2, 350; quantitative fire risk assessment
effectiveness of early intervention methods; semi-quantitative fire risk
323–4; extinction methods 320–1; assessment methods
and fire area 325; fire brigade fire risk levels, by building type 178
intervention 319–20; and fire growth fire safety, see building fire safety
rate control 205; number of jets 322, fire safety advisors, role 12–13
325 Fire Safety Evaluation System (FSES)
fire growth: model types 201–2; stages 61–2
120, 202; see also exponential models fire safety engineering see fire engineering
fire growth parameter/rate 102; by areas fire safety guidance, inadequacy 3
of fire origin 204, 205; by occupancy fire safety management 12–13
105; expected/average value 206–8, Fire Safety Order: assessments under 35;
211, 249; maximum 208–9, 210, 211; guidance 35
standard deviation 208, 211, 213; in fire safety policy 12; structure 12–13
underground facilities 211–13 fire safety strategies 62
Fire HAZard ANalysis (HAZAN) 15 fire services: deployment policies 310;
fire hazards see hazards effectiveness of early intervention
fire insurance 190 323–4; fire protection measures and
Index 363
348–51; intervention by 319–20; floor area 88; normalised per building
resource levels needed 310; see also 88; particular buildings 197–9; power
attendance time; fire fighting; travel function model 98–101, 196; socio-
distance; travel time economic factors 199–200; weather
fire severity: coefficient of variation 157, conditions and 200
158; design value 160; maximum, FSES 61–2
estimation 259; probability distribution fuel types, and frequency of fire occurrence
260; standard deviation 157, 158; 199
statistical distribution estimation 134 full quantitative fire risk assessment 327–
fire spread: beyond building of origin 39; examples 332–9; process 329–32;
267; beyond room of origin 255–67; risk parameters 329; where essential
equivalent network for 128–30; extent 328; where useful 327–8
classification 226; extent measurement furnishings, ignition 233–4
226–8; factors affecting 110–11; fire
brigade attendance time and 257; Gamma distribution 355
from material to material in room gathering phase see recognition time
215; STM approach for probability gating technique 278
evaluation 124–5; stochastic nature generalised least squares method 116–17
117–19; see also building structure; Geographic Information System (GIS)
compartmentation packages 313
fire spread matrix 127 GOFA 17
Fire Standards Policy Committee 333 Gretener method 58–60, 65
Fire Statistics United Kingdom 234 group reactions 300
fire tenability 141, 149–50
fire tests, insight from 139–40 Hammerfest HGV test 357
fire zones 62 Hammerfest wooden crib test 357
FIRECODE HTM 86 14, 15, 40–2, 64; HAZAN 15
assessment sheets 41, 42 hazard frequency, look-up matrices 46, 54
firemen, FAFR for 235–6 hazard identification processes 330
first-aid fire fighting 285–6, 319 hazards: definition 9; from building fires
flashover 119, 202, 215, 220; heat output 150; high 42–3; low 42; normal 42
rate required for 217, 220 HAZID 18, 51, 330
flashover probability 125, 215–25, 255, HAZOP 18, 21, 43, 330
289–91, 345; compartment size and Health and Safety at Work Act (1974)
224–5; event probability tree model 179
218, 219–20; exponential model of Health and Safety Executive (HSE),
fire growth 216–19; hotel bedroom Offshore Safety Division 23
222–3; by occupancy type and area of heat, tenability to 150
fire origin 221; probability distribution heat detectors: operating times 251, 270;
of area damage 220–2; time before reliability 274–5
established burning 225 heat output: correlation with area damage
Flixborough disaster 21 239, 248; exponential growth 203, 248;
flow capacity 303 fire growth rate estimation in terms of
flows through openings 146 213–14; rate of 102–3, 145, 216, 243,
F-n curves 19–20, 337–8 248; for shopping centres 244–5
f(N) relationship 182, 235 heat release rate (HRR) 356–7
fractional effective dose model 233 heat transfer models 145–8
fractional effective doses 150 heavy goods vehicle (HGV) fire in tunnel
frequency analysis 330–1, 337 356–7
frequency–consequence curves see F-n heirarchical approach 61
curves historical data 137–8
frequency of ignition 86–7, 196–200; by hose reel jets, number of 322, 325
building group 87; fuel factors 199; hospitals: detection of fires in wards
global estimation 196–7; normalised by 96–8; fire discovery times 269; fire risk
364 Index
assessment 14, 40, 60–1, 66–7; passive causes of casualties 233; location of
fire protection measures 129–30; fire casualties 230–2, 269; materials
points scheme applicability 63; see also first ignited 233–4; nature of injuries
FIRECODE HTM 86 232–3
hotel bedrooms: design fire size 241–2; fire life risk: FAFR measurement of 235–6;
growth analysis 222–3 look-up matrix 55; total 48–9
hotels: area damage in fires 246; design life safety: acceptance criteria 180;
fire sizes 247; response to fire 298 buildings where affected 5–6;
human response 296–8 effectiveness of sprinklers in 288–91
hyperthermia 150 log normal distribution 111, 112, 113, 242,
hypoxia 150 356
logic trees 87–98; see also event trees; fault
idempotent laws 81 trees
ignition, frequency of see frequency of logit model 110–11
ignition lower confidence limit 106
ignition sources 36–7; hazard level
determination factors 37–8; human major accidents, definition 23
196; non-human 196–7; spinning and Markov model 120–5
doubling industry 198 mass, conservation of 143–4
incapacity, effects on response 298 mass flow rate of fuel 145
incompleteness, problems of 154 mass loss model 233
independent events 73, 89 mass loss rate of fuel 133, 216–17, 220
individual risk 181–2; levels 182 mass transfer models 145–8
informative fire warning (IFW) systems matches, ignition of 233
279–85, 300; laboratory experiments material first ignited, and fire spread
281–5 probability 257–8
infra-red detectors 251, 270 matrix methods 15–16, 43–58; case
initiating events 88–9 study 51–6; flow chart 44; followed
injuries: correlation with area damaged by quantitative fire risk assessment 50;
138, 139; nature of 232–3 ‘look-up’ matrices 46–8; methodology
Insurance Services Office (ISO) 61 43–8; quantification for 48–9; record
integrity 259 sheet 45, 50; workshops 49–50
interactions 341–51; fire protection maximum likelihood method 116–17
measures and fire brigade 348–51 ; mean time between failures (MTBF) 84,
passive and active fire protection 342– 274, 291
7; smoke movement and evacuation of mean time to failure (MTTF) 274
occupants 348 median 156
intersection: operation 80; set 76 Mercantile Schedule 61
intolerable risks 176 minimal cut sets 82–3, 95
ionisation chamber detectors: failure rates mixing between layers 146
275; operation times 251, 270 mobility of occupants 299
mode 156
jets, number of 322, 325 modelling, in consequence analysis
140–51; approaches 140–1; complex
King’s Cross underground fire 7, 8, 24, computational analysis 151; see also
138–9, 296 control volume models
modelling, evacuation analysis 149, 302;
large fire, definition 112 discrete 151, 152
laser beam detectors, operation times 251 modelling, fire tenability 141, 149–50
Latin hypercube sampling 154–5 Mont Blanc fire 26
least squares method 106 Monte Carlo analysis 131–4, 154–5; see
legislaton 10–12 also simulation models
life loss 230–6; attendance time and multicollinearity 205–6
317–19; casualty rate per fire 234–5; multi-nomial distribution 356
Index 365
multiple-death fires, acceptance criteria phased evacuation 304–5, 347
185–8 phases 120, 124
Piper Alpha disaster 8, 22–3
National Bureau of Standards (NBA), plumes: entrainment in 146; schematic
USA see National Institute of Standards 147
and Technology points schemes 15, 58–67; for fire
National Fire Protection Association safety in hospitals 60–1, 66–7;
(NFPA) 135, 287; Life Safety Code Gretener method 58–60, 65; merits
299 and demerits 63–4; risk assessment
National Institute of Standards and schedules 61–2; for road tunnels 15,
Technology (NIST), USA 62, 252 62–3
National Smoke Detector Project 273 Poisson distribution 114, 115, 234, 355
natural fabrics 233–4 portable heating appliances 233
near misses, insight from 138–9 posterior probability 352
negative binomial distribution 115 power functions 98–101
Net Present Value (NPV) 338–9 power stations, design fire size 244
network models 125–30; for travel prior probability 352
distances 313–14 probabilistic methods 5–7, 16;
NFPA 101 Life Safety Code 61–2 application guidelines 334–5; see also
NFPA 101A 62 event tree model; fault tree analysis
nodal events 88–9 probabilistic safety analysis (PSA) 21
normal distribution 111; see also standard probabilities 72, 78; adding 72–3;
normal distribution conditional 73; of failure 84;
NOT logic 83 multiplying 73; posterior 352; prior
nuclear installations 19–21; fire risk 352
assessments 199; risk quantification probability distributions 111–15; density
182 function 111–12
Nuclear Installations Act (1965) 20–3 probability of fire starting see frequency of
Nuclear Site Licences 20 ignition
nuisance alarms 272, 273 probability model 72
null set 75, 77–8 probability theory 19, 71–4
process industries, societal risk;
offices: area damage in fires in 219, 245; acceptability targets 183–5
design fire size 241, 245, 247; flashover property damage, target level 189–90
probability estimation 218–20 property protection, buildings where
offshore installations 8, 22–4 affected 6, 328
Offshore Installations (Safety Case)
Regulations 23–4 Qualitative Design Review (QDR) 34
optical detectors, operation times 251, 270 qualitative fire risk assessment methods
OR gates 78, 79, 94–5 13–15, 34–43; structured 14–15, 40–
outcomes 89; severity estimation 91 2; summary report 38–9; unstructured
14, 35–40
paper, printing and publishing industries, quantitative fire risk assessment methods
fire growth rates 206–8, 209–13 17–18, 70–169; following matrix
parent distribution 111 methods 50; full 17–18; theoretical
Pareto distribution 111, 113–14, 119–20, basis 71–87; see also consequence
242, 356 analysis; full quantitative fire risk
Pareto parameters, for underground assessment; logic trees; Monte Carlo
facilities 243 analysis; probabilistic methods;
partial safety factors 250, 259, 306 statistical models; stochastic models;
passive fire protection, and active fire uncertainty, dealing with
protection 342–7
percolation theory 136 radiative heat transfer 146
permutations 74 rail Safety Risk Framework 51–6
366 Index
Railway Inspectorate 25 schools, life safety and asset protection in
railway stations: fire safety risk assessment 338–9
51–6; overall risk ratings 55–6; risk scientific principles, as basis for models
ranking spreadsheet 56 28–30
Railways (Safety Case) Regulations 25 SCOPES 61
random processes 71 self-termination 202
random variables, examples 131 semi-quantitative fire risk assessment
random walks 119 methods 15–16, 43–67; see also matrix
ratio estimate 208 methods; points schemes
realms 120; for residential occupancies sensitivity analysis 151–4
122, 135 service time 350
recognition time (B) 159, 249, 279, 295–6 set-covering model 315
regression analysis 105–11; multiple set theory 74–7
109–11 set up time 311
regulatory paradigm, consistency with 28 sets: complement 75; intersection 76;
reliability 83–4, 292; of building 265–7; of members 75; null 75, 77–8; union 76;
compartment 261–5; definitions 261–2, universal 75, 77
266; factors influencing 85–6; general severe accident analysis 20
data on 93–4; of smoke detectors 275; severity of fire see fire severity
of sprinkler systems 291–2 SEVESO II Directive 22
residential fire loss 228 shopping centre design 335–8; fire strategy
residual error 106, 110 336; quantified fire risk assessment
response, human 296–8 336–8
response time 311 shops: unsprinklered 241; see also retail
retail premises: area damage in fires in premises
240–1; design fire sizes 247; fire load simulation models 130–6; continuous
density 244 131; discrete 131; examples 134–6;
risk acceptability 175; see also acceptance merits and demerits 136; Monte Carlo
criteria 131–4, 154–5
risk assessment see fire risk assessment ‘small’ fires 319
risk assessment schedules 61–2 smoke: rate of growth of 214; as threat to
risk categories 316 occupants 231–2, 299–300
risk classification, look-up matrix 48 smoke detectors: factors affecting response
risk comparison 331 270; with integral alarms 279;
risk evaluation 331–2 operating times 251, 270; power source
risk management 9–10 disconnection 273; reliability 275
risk parameters 329 smoke movement: for early period of fire
risk premiums 101 248; and evacuation of occupants 348;
risk reduction 332 prediction 148; simulation 134
risk tolerability, look-up matrix 48 smoke ventilation systems see ventilation
risk of undesirable event, definition 2 systems
smokers’ materials: careless disposal 233;
safety, erring on side of 153 fire risk assessment 197
safety assessment principles (SAPs) 21 societal risk 182; levels 182–5
safety cases 18; definition 20 socio-economic factors, and frequency of
safety factors 155–62; characteristic values fire occurrence 199–200
156–60; design values 160–2; partial space heating, incidents with 233
160–2; see also Beta method species, conservation of 144
safety index 164, 165, 169, 260; log specific flow 301, 303
normal 168–9 speed 301–3
safety margin 164 sprinkler factor 245, 346
safety reports, contents 22 sprinkler systems 252, 286; and automatic
SAFiRE 337 fire detectors 347; and building or
schedule rating 61–2 compartment size 342, 343, 344; in
Index 367
bus garages 91–4, 192–4; and damage suppression 285–92; first-aid fire fighting
extent 101, 228, 229; effectiveness 285–6, 319
in life safety 288–91; effectivness suppression size 228
in property protection 287–8; and Swedish Fire Protection Association 322
fire brigade 349; and fire resistance synthetic fabrics 233–4
345–6; non-operation 86, 287, 291; system reliability see reliability
operation times 252, 286; operation
times design values 250, 252; and T2 curve 103, 201–2
passive fire protection 342–7; Tauern fire 26
percentages of fires extinguished Technica 281
by 319–20; performance 286–7; Technical Assessment Guides 21
probability of extinguishing fire 205; Technical Report ISO/TR 13387-2:
reliability 291–2; residential 289, 1999(E) 239
291, 349; in retail premises 241; in telecommunication facilities, fire risk
schools 338–9; reduced probability assessment 61
of flashover 290; and staircase width textile industry: fire spread extent 226–8;
344–5; in textile industry 103–4; and fire studies 103–4, 112–13
travel distance 342–3; types 286; and textiles, ignition 233–4
ventilation systems 253, 346–7 thermal failure 156, 255; see also
square-root law 314–15 compartmentation
stability, of load-bearing capacity 259 thermal inertia 259, 346
staircases: minimum width 305; thermal insulation 259
movement on 304; sprinkler systems thermal radiation 150
and 344–5 tolerable risks 176
stakeholders, in fire safety 8–9 top events 78; sum of products form 82
Standard Metropolitan Statistical Areas total evacuation 304–5
(SMSA) 313 toxic gases: tenability to 150; as threat to
standard normal distribution, probabilities occupants 231–2, 299–300
of success and failure 167 trade-off 341
state function 164 transition matrix 121
states 120 transition probabilities 119, 120
station see railway station transport 24–5
statistical analysis: advantages 70; travel distance 312–16; estimation
limitations 70–1 methods 312–14; sprinklers and
statistical models 98–117; exponential 342–3; square-root law 314–15; time–
see exponential models; extreme distance relationship 314
value distributions 116–17; multiple travel speed 301–3
regression analyis 109–11; power travel time 311; average in region 316;
functions 98–101; probability measures 312
distributions 111–15; regression travel time vector 311–12
analysis 105–11 ‘trench effect’ 24
stochastic models 117–30; Markov trials see experiments
120–5; network models 125–30; truth tables 78
random walk 119; state transition tunnels 9, 26–7; heavy goods vehicle
model (STM) 122–5; stochastic (HGV) fires in 356–7; points schemes
nature of fire spread 117–19 for 15, 62–3
strength, of load-bearing capacity 259 turnout time 311
strength variable 155
stress variable 155 unavailability 85
structural collapse 191–2, 255, 264–5 uncertainty: dealing with 151–69; in fire
structural failure: probabilities 125, 346; resistance model parameters 261;
target levels 191–2 see also Beta method; safety factors;
subsets 75 sensitivity analysis
sum of products form 80, 82, 83 uncertainty analysis 154–5
368 Index
under-reporting 137 fire brigade 350–1; operation times
underground facilities: design fire sizes 253; operation times design values
242–5; fire growth rates 211–13 250, 253; and sprinkler systems 253,
union: operation 80; set 76 346–7
unit of activity 329 visibility criteria 150
univariate approach 163–4 visual detection 297
universal set 75, 77 voluntary vs. involuntary risk 181
unreliability 84, 277, 292
untenable conditions, spread of 308 WASH 1400 182
upholstery, ignition 233–4 water damage 287–8
upper confidence limit 106 weather conditions, and frequency of fire
urban fires, simulations 136 occurrence 200
utility/disutility theory techniques 191 Weibull probability distribution 84, 274, 356
Windscale reactor fire 7, 19
‘value of a life’ 188 World Trade Center disaster 266
variance-covariance matrices 133
Venn diagrams 75 y intercept 106
ventilation factor 157, 259
ventilation systems 253; design 241; and zone models see control volume models
Taylor & Francis
eBooks
ORDER
Y
FREE 3 OUR
0 DAY
INSTITU
TIONA
TRIAL T L
ODAY!
FOR LIBRARIES
Over 22,000 eBook titles in the Humanities,
Social Sciences, STM and Law from some of the
world’s leading imprints.
Choose from a range of subject packages or create your own!
Free MARC records

Benefits for
you COUNTER-compliant usage statistics
Flexible purchase and pricing options
Off-site, anytime access via Athens or referring URL

Benefits Print or copy pages or chapters
for your
Full content search
user
Bookmark, highlight and annotate text
Access to thousands of pages of quality research
at the click of a button
For more information, pricing enquiries or to order

a free trial, contact your local online sales team.
UK and Rest of World: [email protected]
US, Canada and Latin America:
[email protected]
www.ebooksubscriptions.com
eBooks
A flexible and dynamic resource for teaching, learning and research.

Quantitative Risk Assessment in Fire Safety

Uploaded by

Copyright:

Available Formats

Quantitative Risk Assessment in Fire Safety

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Quantitative Risk Assessment in Fire Safety

Uploaded by

Copyright:

Available Formats

Quantitative Risk Assessment

2 Qualitative and ­semi-­quantitative risk assessment techniques 34

3 Quantitative risk assessment techniques 70

4 Acceptance criteria 175

6 Design fire size 239

7 Fire spread beyond room of origin 255

8 Performance and reliability of detection, alarm and suppression 268

9 Performance and reliability of human response and ­evacuation 295

10 Performance and effectiveness of fire service i­ntervention 310

11 Whole project ­analysis 327

13 Combining data from various sources – Bayesian ­techniques 352

1.1 Breakdown of different fire risk assessment methods 13

2.1 Typical matrix method record sheet 45

Ganapathy Ramachandran, PhD, DSc, FIFireE, MSFPE, held senior scientific

The authors acknowledge the following permissions to reproduce material in this

• the severity of its outcome (its consequences); and

If we simply consider the potential consequences of fires, then we may not

• Does the fire grow?

In this way we can assess the impact of fire retardant or ­non-­combustible

1.1 Fire engineering

the application of scientific and engineering principles to the protection of

Fire engineering is generally more demanding, technically and in terms of

1.2 Deterministic approaches

1.3 Probabilistic approaches

• select fire scenarios for deterministic analysis; and/or

• in demonstrating equivalent life safety to a ­code-­compliant building or

Therefore, circumstances where fire risk assessment could be useful include

1 Life safety is affected by buildings:

2 Property protection/business continuity is affected by buildings:

3 Environmental protection is affected by buildings with a high potential impact

• When the reliability and performance of protection systems is critical. Fire

1.4 Background to the development of fire risk assessment

• 0.5 to 2 people per 100,000 of population per year; and

• protect stakeholders rights;

Staff expect a safe working environment. This expectation may be balanced

1.4.2 Risk management

• establish a strategic policy (individual policies such as fire safety may be

The remaining business risk can then either be:

1 Transferred: by means such as insurance, or

1.4.3.1 Building regulations

1 to ensure satisfactory provision of means of giving an alarm of fire and a

Guidance on how to comply with building regulations can be found in a variety

• the anticipated probability of a fire occurring; and

These factors clearly imply the use of fire risk assessment.

…demonstrate that a building, as designed, presents no greater risk to

1.4.3.2 Fire regulations for existing buildings

• a fire risk assessment to be undertaken;

1.4.3.3 Other safety legislation

1.4.4 Fire safety management

1.4.4.1 Fire safety policy

1.4.4.2 Management structure

• supervising the ­day-­to-­day upkeep of the policy;

Members of organisations also have general duties and responsibilities and

• understanding the nature of fire hazards;

• assisting management in the interpretation and application of legislation and

Specialist fire safety advisors may be internal or external to an organisation.

1.4.4.3 Audit and risk assessment

1.4.5 Qualitative fire risk assessment methods

Fire risk assessment

Qualitative methods Semi-quantitative methods Quantitative methods

2 Qualitative and semi-quantitative risk assessment techniques 34

9 Performance and reliability of human response and evacuation 295

10 Performance and effectiveness of fire service intervention 310

11 Whole project analysis 327

13 Combining data from various sources – Bayesian techniques 352

In this way we can assess the impact of fire retardant or non-combustible

• in demonstrating equivalent life safety to a code-compliant building or

• supervising the day-to-day upkeep of the policy;

1.4.6 Semi-quantitative fire risk assessment methods

This chapter describes some of the qualitative and semi-quantitative fire

The building was surveyed on a floor-by-floor and occupancy-by-occupancy