AP-R617-20 Bridge Assessment Beyond The AS5100 Deterministic Methodology PDF
AP-R617-20 Bridge Assessment Beyond The AS5100 Deterministic Methodology PDF
AP-R617-20 Bridge Assessment Beyond The AS5100 Deterministic Methodology PDF
AP-R617-20
Prepared by Publisher
Mayer M Melhem, Colin C. Caprani, Mark G Stewart, Shaohua Austroads Ltd.
Zhang Level 9, 287 Elizabeth Street
Sydney NSW 2000 Australia
Project Manager Phone: +61 2 8265 3300
[email protected]
Andy Ng www.austroads.com.au
© Austroads 2020 | This work is copyright. Apart from any use as permitted under the Copyright Act 1968, no part may be reproduced by
any process without the prior written permission of Austroads.
Acknowledgements
The valuable input and feedback from the participating local and international review committee members is greatly appreciated. Thanks
to the Victorian Department of Transport and Queensland Department of Transport and Main Roads in providing the necessary
engineering drawings and other relevant data. Contributions from Lachlan Twigg, Mohd Shihabuddin Khan, Pabita Sijapati, and Dr Lizi
Sironic towards this project are also greatly acknowledged.
This report has been prepared for Austroads as part of its work to promote improved Australian and New Zealand transport outcomes by
providing expert technical input on road and road transport issues.
Individual road agencies will determine their response to this report following consideration of their legislative or administrative
arrangements, available funding, as well as local circumstances and priorities.
Austroads believes this publication to be correct at the time of printing and does not accept responsibility for any consequences arising from
the use of information herein. Readers should rely on their own skill and judgement to apply information to particular issues.
Bridge Assessment Beyond the AS 5100 Deterministic Methodology
Summary
With more than 50,000 bridges across Australia and New Zealand, road freight continually increasing, and
maintenance budgets under pressure, bridge asset agencies are having to implement different hierarchical
strategies beyond the current deterministic methodology found in AS 5100 to safely manage heavy vehicle
freight access to their bridge networks. From international practice, the most holistic approach that provides
objective and quantifiable measures of safety on a continuum is the use of probability-based bridge
assessment (PBBA) based on structural reliability theory. This study demonstrates the practical application
of PBBA for the Australian context, deriving a set of pertinent results.
A probabilistic framework for bridge assessment was developed using state-of-the-art structural reliability
approaches as appropriate to the Australian/New Zealand context. This approach is now commonly used
internationally. The framework follows a similar procedure to the deterministic flowchart presented in
AS 5100.7 (2017), but instead considers the random nature and uncertainty in member capacity and load
effects.
The framework was demonstrated for a set of existing Australian highway bridges to infer on the
contemporary code-implied levels of safety (Step 1), and the current safety under as-of-right road freight
access (Step 2). The bridge eras range from 1950s to present day, are simply supported, made of concrete
(reinforced or prestressed), and comprise two design lanes. The ultimate limit states of bending and shear of
the critical bridge deck components (girder, beam, plank, etc.) were assessed. The selected road freight for
Step 2 was the Higher Mass Limit (HML) network, which includes general access vehicles (GAV), at various
traffic volumes and flows.
The results from the application of the PBBA framework showed a safety margin available that lower-tier
assessment approaches did not reveal. This was particularly notable for more recently designed bridges. For
example, for ultimate shear, about 75% of bridges that failed a deterministic assessment were shown to have
a reliability higher than the acceptable level of safety/risk.
The findings demonstrate the benefit of using a structural reliability-based framework as the higher-tier
bridge safety assessments for Australia. In particular, the acceptable level of safety may vary across
individual bridges based on the needs of the bridge asset owner, public safety, and the economy. Thus,
PBBA frameworks can provide more fit-for-purpose assessments, better utilising bridge assets, and thereby
facilitating an optimally productive road transport network.
Glossary
ABDC Austroads Bridge Design Code. Code of practice for the design of bridges in Australia
from 1992 to 2004
Accompanying lane factors Reduction factors adopted in design to consider multiple vehicle presence on bridge
deck
AS 5100 Australian Standard 5100. Documentation used for the design of bridges in Australia
since 2004
Bias Ratio between mean value and characteristic (nominal) value of a random variable
Block Maximum Method Procedure for establishing the probabilistic distribution of a stochastic variable
Block Size Part of the block maximum method. Selected time period where maximum value is
extracted as a realisation of a stochastic variable
BTLS Bridge Traffic Load Simulation. In-house software that generates artificial traffic
streams on bridge spans for a defined measurement length and converts to the
realisations of traffic load effects random variable
CDF Cumulative Distribution Function. Describes the probability of having less than or
equal to a given value of a random variable
Characteristic value Value from a random variable that has a defined chance of being exceeded. Selected
for use in design
Coefficient of variation Ratio between standard deviation and mean value of a random variable
Correlation Coefficient Numerical measure describes the linear dependency between to random variables.
(Pearson's linear) Range from -1 to 1
Curve fitting Method for establishing most appropriate probability distribution by using generated
realisations as data points and fitting to a line on a probability paper
Deterministic method Approach that takes resistance and loading as single point-estimates
Directional cosines Outputs from FORM that are used for sensitivity analysis
Dynamic load allowance Amplification factor adopted in design to consider the effects on moving loads on
bridges
Economic days Day when traffic flows are at the most common form during the year (i.e. weekdays
and not during annual vacation periods, ~ 250 days per year)
FERUM Finite element reliability using MATLAB. Software containing structural reliability
methods
FORM First order reliability method. Quantifies the structural reliability index
γ Loading safety factor in limit state design
GEV probability distribution Generalized Extreme Value probability distribution. A type of probability distribution
commonly used for expressing uncertainty in maximum values. Described by location,
scale and shape parameter. Depending on the shape parameter, it is further divided
into Gumbel, Freichet, and Weibull probability distributions
Gumbel probability A type of GEV probability distribution where the scale parameter is equal to zero
distribution
Histogram Bar charts where each bar presents the relative frequency of the observed
realisations in a predefined interval (also called bins)
HML Higher mass limit. Type of classification of heavy vehicles in Australia with permitted
greater axle loads than general access vehicles (GAV), but restricted network access.
Independent When there is no correlation between two random variables. The realisations of one
variable does not affect the realisations of another
Influence lines Graphs describing the change in a load effect at a certain point of structure as a unit
load traverses the structure.
Inverse standard normal Provides the variate value for a corresponding probability for normal distribution
probability distribution function with mean of 0 and standard deviation of 1.
function
Level of Safety The probability that failure to a specific limit state does not occur
Limit State Function Equation describing the boundary between failure and non-failure of given mode of
failure.
Lognormal probability A type of probability distribution commonly used for expressing uncertainty in values
distribution that cannot be zero.
Maximum likelihood Method for establishing most appropriate probability distribution
estimation
MCS Monte Carlo Simulation. Procedure for generating unique realisations of a random
variable by using random number generation.
Mean A measure of central tendency in observed data
Measurement length Part of the block maximum method. Duration of data collection period for generating
realisations of a stochastic variable
MS18 Traffic load model adopted in Australian bridge design from 1947 to 1976
NAASRA BDC National Association of Australian State Road Authorities Bridge Design Code.
Documentation used for the design of bridges in Australia from 1959 to 1992.
Normal probability The common type of probability distribution commonly used for expressing uncertainty
distribution
PBBA Probability based bridge assessment. Approach that adopts structural reliability theory
infer directly on the level of safety for bridges
PDF Probability Distribution/Density Function. Describes the probability observing a given
value of a random variable
ϕ Capacity reduction factors in limit state design
Probabilistic method Approach that takes resistance and loading as random variables
Probability distribution Used in statistics for expressing the uncertainty of given variable
Probability of failure The chance that a specific mode of failure occurs
Probability paper Plots the relative cumulative frequency of observed data points, with probabilities
scaled to specific distribution type
Random Variable A quantity that is associated with some degree of uncertainty
Realisation An observation or data point from a random variable
Reference Period The unit of time associated with the probabilistic distribution for a stochastic variable
Reliability calibration Use of structural reliability theory to establish safety factors so such that
designs/assessment meet a predefined level of safety
Sensitivity analysis Method for checking the robustness and validate assumptions made in a set of results
SM1600 Traffic load model adopted in Australian bridge design since early 2000s
Smoothed bootstrapping A Monte Carlo simulation approach where values are randomly selected from the
predefined set and then further randomized
SORM Second order reliability method. Quantifies the structural reliability index using a
quadratic approximation to the limit state function at the design point.
Standard deviation A measure of spread in observed data
Standard external variate Variate value for a corresponding probability for Gumbel distribution function with
mean of 0 and standard deviation of 1
Standard normal probability A normal distribution with mean of 0 and standard deviation of 1
distribution
Stationarity Describes when two probability distribution have not changed over time
Statistical inference Establishing the most appropriate probability distribution for a set of realisations
Statistical parameters Describes probabilistic distributions. Common types include the mean, standard
deviation.
Stochastic variable Random variables with observations/realisations changing with time
Structural Reliability Index A measure of the level of safety in the structure
(β)
Structural reliability method Approach to quantify the probability of failure / structural reliability index
T44 Traffic load model adopted in Australian bridge design from 1976 to early 2000s
Contents
Summary ..................................................................................................................................................... i
Glossary .................................................................................................................................................... ii
1. Introduction ............................................................................................................................................ 1
1.1 Background ...................................................................................................................................... 1
1.2 Purpose ............................................................................................................................................ 2
1.3 Scope ............................................................................................................................................... 3
1.4 Outline .............................................................................................................................................. 3
2. Probability-Based Bridge Assessments.............................................................................................. 4
2.1 Overview .......................................................................................................................................... 4
2.1.1 Comparison to Deterministic Approach ............................................................................... 4
2.1.2 Probability of Failure and Reliability Index β ....................................................................... 5
2.1.3 Comparison to Codes of Practice........................................................................................ 6
2.1.4 Acceptable Level of Safety .................................................................................................. 6
2.2 International Adoption ...................................................................................................................... 6
2.2.1 United Kingdom ................................................................................................................... 7
2.2.2 United States ....................................................................................................................... 7
2.2.3 Europe ................................................................................................................................. 7
2.2.4 Australia/New Zealand ........................................................................................................ 7
3. Framework for Bridge Assessment Beyond the Deterministic Method ......................................... 10
3.1 Overview ........................................................................................................................................ 10
3.2 Framework Demonstration ............................................................................................................. 11
3.2.1 Representative Bridges Subset ......................................................................................... 11
3.2.2 Modes of Failure (Limit States) ......................................................................................... 13
3.2.3 Traffic Considerations ....................................................................................................... 13
4. Framework Components ..................................................................................................................... 16
4.1 Limit State Function........................................................................................................................ 16
4.1.1 Reference Period............................................................................................................... 16
4.2 Capacity Uncertainty ...................................................................................................................... 17
4.2.1 Function of Random Variables .......................................................................................... 17
4.2.2 Monte Carlo Simulation ..................................................................................................... 18
4.3 Loading Uncertainty ....................................................................................................................... 20
4.3.1 Uncertainty in Permanent Loads ....................................................................................... 20
4.3.2 Uncertainty in Maximum Transient Loads ......................................................................... 21
4.4 Other Uncertainties ........................................................................................................................ 23
4.4.1 Model Errors ...................................................................................................................... 23
4.4.2 Dynamic Load Allowance .................................................................................................. 23
4.5 Bridge Assessment and the Reliability Index ................................................................................. 24
4.5.1 Structural Reliability Index ................................................................................................. 24
4.5.2 Structural Reliability Method.............................................................................................. 24
Tables
Table 2-1: Annual target probability of failures (reliability index) based on economic optimisation. ............ 6
Table 3-1: Bridge structural families to be assessed in estimated chronological order of construction. ... 12
Table 4-1: Uncertainty or probabilistic distributions (expressed as bias and CoV = coefficient of
variation) for section properties used as input parameters for the ultimate bending and
ultimate shear capacity models (Appendix G). ......................................................................... 18
Table 4-2: Correlations between non-independent random variables taken in Monte Carlo
Simulation.................................................................................................................................. 20
Table 4-3: Recommended uncertainty or probability distributions (expressed as bias and
CoV = coefficient of variation) for permanent loading. All are normal (Gaussian)
distributed. ................................................................................................................................ 20
Table 5-1: Input parameters for shear model extracted from T44 designed I-girder. ................................ 27
Table 5-2: Characteristic shear capacity (as per AS 5100.5 model) and associated probability
distribution as per Monte Carlo simulation (all sections in Appendix G). .................................. 28
Table 5-3: Characteristic shear force (as per TMR grillage model) and associated probability
distribution. Traffic load includes dynamic allowance and multiple vehicle presence
factors (all sections in Appendix H)........................................................................................... 28
Table 5-4: Fitted probability distribution (uncertainty) for annual maximum shear force due to
different higher mass limit (HML) network at different flows for example girder (all
sections in Appendix H). ........................................................................................................... 31
Table 5-5: Traffic load factors adopted in deterministic analysis. .............................................................. 32
Table 5-6: Step 1 ULS shear structural reliability results for T44 designed I-girder (all sections
in Appendix J). .......................................................................................................................... 34
Table 5-7: Step 2 ULS shear structural reliability results for T44 designed I-girder (all sections
in Appendix K). .......................................................................................................................... 34
Table 5-8: Step 1 ULS shear relative importance (sensitivity analysis) for T44 designed I-girder
(all sections in Appendix J). ...................................................................................................... 35
Table 5-9: Step 2 ULS shear relative importance (sensitivity analysis) for T44 designed I-girder
for all flows (all sections in Appendix K).................................................................................... 35
Table B-1: Annual target reliability index based on economic optimisation. .............................................. 63
Table B-2: Annual target reliability index based on life safety (LQI = life quality index). ............................ 64
Table D-1: Breakdown of 25,000 vehicles per day in bi-directional traffic flow with uniform pattern
(Flat-Bi) across two traffic lanes. ............................................................................................... 71
Table D-2: Breakdown of 25,000 vehicles per day in uni-directional traffic flow with uniform pattern
(Flat-Uni) across two traffic lanes. ............................................................................................ 72
Table D-3: Breakdown of 25,000 vehicles per day in bi-directional traffic flow with peaked pattern
(Low-Bi) across two traffic lanes. .............................................................................................. 73
Table D-4: Breakdown of 25,000 vehicles per day in uni-directional traffic flow with peaked pattern
(Low-Uni) across two traffic lanes. ............................................................................................ 74
Table D-5: Breakdown of 50,000 vehicles per day in bi-directional traffic flow with peaked pattern
(High-Bi) across two traffic lanes. ............................................................................................. 75
Table D-6: Breakdown of 50,000 vehicles per day in uni-directional traffic flow with peaked pattern
(High-Uni) across two traffic lanes. ........................................................................................... 76
Table F-1: Breakdown of Heavy Vehicles based on 6-months WIM Data on West Gate Bridge............... 83
Table F-2: Variable-bandwidth kernels density adopted for artificial traffic simulation. ................................ 84
Table G-1: Characteristic bending capacity (as per AS 5100.5 model) and associated probability
distribution as per correlated Monte Carlo simulation for all sections. ..................................... 91
Table G-2: Characteristic shear force (as per AS 5100.5 model with adaptation for assessment)
and associated probability distribution as per correlated Monte Carlo simulation for
all sections. ............................................................................................................................... 92
Table H-1: Finite element model used and components in these models for the different bridge
families. ................................................................................................................................... 113
Table H-2: Characteristic bending moment (in kNm) at midpsan. Traffic load is from traffic load
model and includes dynamic allowance and multiple vehicle presence factors as
stipulated by respective design code. ..................................................................................... 116
Table H-3: Characteristic shear force (in kN) at the distance of do from face of support. Traffic
load is from traffic load model and includes dynamic allowance and multiple vehicle
presence factors as stipulated by respective design code. .................................................... 117
Table H-4: Fitted probability distribution (uncertainty) for annual maximum bending moment
due to different higher mass limit (HML) network at different flows (Appendix D).
Data (represented as ratio of simulated load effect with design traffic load model l
oad effect) fits as GEV distribution with = shape parameter, = scale parameter,
u = location parameter (see Appendix A). .............................................................................. 234
Table H-5: Fitted probability distribution (uncertainty) for annual maximum shear force due to
different higher mass limit (HML) network at different flows (Appendix D). Data
(represented as ratio of simulated load effect with design traffic load model load effect)
fits as GEV distribution with = shape parameter, = scale parameter, u = location
parameter (see Appendix A). .................................................................................................. 236
Table I-1: AS 5100.5 Bending Capacity Model Error. ............................................................................. 238
Table I-2: AS 5100.5 Shear Capacity Model Error. ................................................................................. 239
Table I-3: Loading Model Error................................................................................................................ 239
Table J-1: Results for Step 1 (Bending). .................................................................................................. 240
Table J-2: Results for Step 1 (Bending). .................................................................................................. 241
Table J-3: Results for Step 1 (Shear)....................................................................................................... 243
Table J-4: Results for Step 1 (Shear)....................................................................................................... 244
Table K-1: Results for Step 2 (Bending). .................................................................................................. 246
Table K-2: Results for Step 2 (Bending) – Flat Bi..................................................................................... 256
Table K-3: Results for Step 2 (Bending) – Flat Uni. ................................................................................. 258
Table K-4: Results for Step 2 (Bending) – Low Bi. ................................................................................... 260
Table K-5: Results for Step 2 (Bending) – Low Uni. ................................................................................. 262
Table K-6: Results for Step 2 (Bending) – High Bi. .................................................................................. 264
Table K-7: Results for Step 2 (Bending) – High Uni. ................................................................................ 266
Table K-8: Results for Step 2 (Shear)....................................................................................................... 268
Table K-9: Results for Step 2 (Shear) – Flat Bi. ....................................................................................... 278
Table K-10: Results for Step 2 (Shear) – Flat Uni. ..................................................................................... 280
Table K-11: Results for Step 2 (Shear) – Low Bi........................................................................................ 282
Table K-12: Results for Step 2 (Shear) – Low Uni. .................................................................................... 284
Table K-13: Results for Step 2 (Shear) – High Bi. ...................................................................................... 286
Table K-14: Results for Step 2 (Shear) – High Uni..................................................................................... 288
Figures
Figure 1-1: Plot of rating factor outputs vs structural reliability index illustrating the margin the
probability methods provide in the Australian context (βT = target reliability index). .................... 2
Figure 2-1: Probabilistic-based bridge assessment (PBBA) procedure. .......................................................... 4
Figure 2-2: Probability distributions (uncertainty) of load and resistance......................................................... 5
Figure 2-3: Probabilistic-based bridge assessment adopted in DRD............................................................... 8
Figure 2-4: Verification method using probability-based methods adopted in ABCB. ..................................... 9
Figure 3-1: Proposed higher-order assessment within the Australian bridge assessment flowchart. ........... 10
Figure 3-2: Traffic flow rates, hourly variation, and truck proportions considered in the demonstration
of the framework. ......................................................................................................................... 14
Figure 4-1: Sample realisations from the joint distribution of annual maximum static load effect
and dynamic amplification factors, showing the marginal histograms, for the TMR
T44 I-girder bridge bending moment using bi-directional high flow. ............................................ 24
Figure 5-1: Cross-section of T44 designed I-girder bridge deck: the outer girder is assessed
for shear. ...................................................................................................................................... 26
Figure 5-2: Engineering drawings of outer girder for T44 designed I-girder bridge deck............................... 27
Figure 5-3: Histogram and Gumbel probability paper plot of shear capacity realisations for
T44 designed I-girder (all sections in Appendix G). ..................................................................... 28
Figure 5-4: Influence lines (kN shear/kN axle) for shear force at the critical location on the
outer girder of TMR T44 designed I-girder bridge deck for vehicles in two different
traffic lanes (Lane 1 and Lane 2). ................................................................................................ 29
Figure 5-5: Histogram and Gumbel probability paper plot of maximum annual shear force
realisations for T44 designed I-girder under various traffic flows (all sections
in Appendix H). ............................................................................................................................. 30
Figure 6-1: Safety of ‘as-built’ bridge families as inferred to historical code for ultimate bending. ................ 36
Figure 6-2: Safety of ‘as-built’ bridge families as inferred to historical code for ultimate shear. ...................... 37
Figure 6-3: Relationship between deterministic and probabilistic measures of safety for ‘as-built’
bridge families as inferred from the historical code for (a) ultimate bending and
(b) ultimate shear. ........................................................................................................................ 38
Figure 6-4: Relative importance for each random variable that comprise limit state function in
the reliability index results (Step 1) for ultimate bending. ............................................................ 39
Figure 6-5: Relative importance for each random variable that comprise limit state function in
the reliability index results (Step 1) for ultimate shear. ................................................................ 40
Figure 6-6: Range in safety of ‘as-built’ bridge families under different HML traffic flows for
ultimate bending. .......................................................................................................................... 42
Figure 6-7: Range in safety of ‘as-built’ bridge families under different HML traffic flows for
ultimate shear. .............................................................................................................................. 43
Figure 6-8: Relationship between deterministic and probabilistic measures of safety for
‘as-built’ bridge families to selected HML flows for (a) ultimate bending and
(b) ultimate shear. ........................................................................................................................ 44
Figure 6-9: Relative importance for each random variable that comprise the limit state function
in the reliability index results (Step 2) for ultimate bending for High Bi (others shown
in Appendix K). ............................................................................................................................. 45
Figure 6-10: Relative importance for each random variable that comprise the limit state function
in the reliability index results (Step 2) for ultimate shear for High Bi (others shown in
Appendix K). ................................................................................................................................. 46
Figure A-1: Visualisation of uncertainty and fitting tails to probability distribution. ......................................... 56
Figure A-2: Relationship between reliability index (β) and the probability of failure (pf). ................................ 59
Figure C-1: Breakdown of Victorian Road Bridges by Material Type. ............................................................ 66
Figure C-2: Bridge Structural Forms for Common Materials........................................................................... 67
Figure C-3: Number of traffic lanes for concrete bridge structural forms. ....................................................... 67
Figure C-4: Span lengths for concrete bridge structural forms. ...................................................................... 68
Figure C-5: Precast Tee Slabs considered in this analysis............................................................................. 69
Figure E-1: Detailed Procedure for Accomplishing Step 1. ............................................................................ 77
Figure E-2: Distribution of Australia WIM data on triaxle mass suggests a CoV of 0.167 at the legal load ... 80
Figure F-1: Detailed Procedure for Accomplishing Step 2. ............................................................................ 82
Figure H-7: Influence lines of PSC plank bridge family for each traffic lane (Lane 1 and Lane 2)
at critical location for bending moment and shear force based on 1 kN 2.3 m wide
traversing bogey using grillage/finite element model. ................................................................120
Figure H-8: Influence lines of MS18 I-girder bridge family for each traffic lane (Lane 1 and Lane 2)
at critical location for bending moment and shear force based on 1 kN 2.3 m wide
traversing bogey using grillage/finite element model. ................................................................122
Figure H-9: Influence lines of TMR I-girder for each traffic lane (Lane 1 and Lane 2) at critical
location for bending moment and shear force based on 1 kN 2.3 m wide traversing
bogey using grillage/finite element model. .................................................................................124
Figure H-10: Influence lines of Precast Tee slab bridge family for each traffic lane (Lane 1 and
Lane 2) at critical location for bending moment and shear force based on 1 kN 2.3 m
wide traversing bogey using grillage/finite element model. .......................................................124
Figure H-11: Influence lines of VR Super-T girder for each traffic lane (Lane 1 and Lane 2) at
critical location for bending moment and shear force based on 1 kN 2.3 m wide
traversing bogey using grillage/finite element model. ................................................................126
Figure H-12: Influence lines of Super-T girder bridge family for each traffic lane (Lane 1 and
Lane 2) at critical location for bending moment and shear force based on 1 kN 2.3 m
wide traversing bogey using grillage/finite element model. .......................................................127
Figure H-13: Histogram and Gumbel probability paper plot of maximum annual bending moment
realisations for MS18 designed U-slab under various traffic flows. Data represented
as ratio of simulated load effect with design traffic load model load effect. ...............................131
Figure H-14: Histogram and Gumbel probability paper plot of maximum annual bending moment
realisations for MS18 designed PSC Planks under various traffic flows. Data
represented as ratio of simulated load effect with design traffic load model load effect. ..........139
Figure H-15: Histogram and Gumbel probability paper plot of maximum annual bending moment
realisations for MS18 designed I-girders under various traffic flows. Data represented
as ratio of simulated load effect with design traffic load model load effect. ...............................147
Figure H-16: Histogram and Gumbel probability paper plot of maximum annual bending
moment realisations for T44 designed I-girder under various traffic flows. Data
represented as ratio of simulated load effect with design traffic load model load effect. ..........155
Figure H-17: Histogram and Gumbel probability paper plot of maximum annual bending moment
realisations for T44 designed Precast Tee slabs under various traffic flows. Data
represented as ratio of simulated load effect with design traffic load model load effect. ..........157
Figure H-18: Histogram and Gumbel probability paper plot of maximum annual bending moment
realisations for T44 designed Super-T girders under various traffic flows. Data
represented as ratio of simulated load effect with design traffic load model load effect. ..........165
Figure H-19: Histogram and Gumbel probability paper plot of maximum annual bending moment
realisations for SM1600 design Super-T girders under various traffic flows. Data
represented as ratio of simulated load effect with design traffic load model load effect. ..........167
Figure H-20: Histogram and Gumbel probability paper plot of maximum annual shear force
realisations for MS18 designed U-slab under various traffic flows. Data represented
as ratio of simulated load effect with design traffic load model load effect. ...............................182
Figure H-21: Histogram and Gumbel probability paper plot of maximum annual shear force
realisations for MS18 designed PSC Planks under various traffic flows. Data
represented as ratio of simulated load effect with design traffic load model load effect. ..........190
Figure H-22: Histogram and Gumbel probability paper plot of maximum annual shear force
realisations for MS18 designed I-girders under various traffic flows. Data represented
as ratio of simulated load effect with design traffic load model load effect. ...............................198
Figure H-23: Histogram and Gumbel probability paper plot of maximum annual shear force
realisations for T44 designed I-girder under various traffic flows. Data represented
as ratio of simulated load effect with design traffic load model load effect. ...............................206
Figure H-24: Histogram and Gumbel probability paper plot of maximum annual shear force
realisations for T44 designed Precast Tee under various traffic flows. Data
represented as ratio of simulated load effect with design traffic load model load effect. ..........208
Figure H-25: Histogram and Gumbel probability paper plot of maximum annual shear force
realisations for T44 designed Super-T girders under various traffic flows. Data
represented as ratio of simulated load effect with design traffic load model load effect. ..........216
Figure H-26: Histogram and Gumbel probability paper plot of maximum annual bending moment
realisations for SM1600 design Super-T girders under various traffic flows. Data
represented as ratio of simulated load effect with design traffic load model load effect. ..........218
Figure K-1: Step 2 annual structural reliability indices for MS18 designed U slabs for different
HML flows at ultimate bending. ..................................................................................................247
Figure K-2: Step 2 annual structural reliability indices for MS18 designed PSC planks for different
HML flows at ultimate bending. ..................................................................................................247
Figure K-3: Step 2 annual structural reliability indices for MS18 designed I-girders for different
HML flows at ultimate bending. ..................................................................................................247
Figure K-4: Step 2 annual structural reliability indices for T44 designed Precast Tee Slabs for
different HML flows at ultimate bending. ....................................................................................248
Figure K-5: Step 2 annual structural reliability indices for Case Studies for different HML flows
at ultimate bending. ....................................................................................................................248
Figure K-6: Step 2 annual structural reliability indices for SM1600 designed Super-T girders for
different HML flows at ultimate bending. ....................................................................................249
Figure K-7: Relative importance for each random variable that comprise limit state function
in the reliability index results (Step 2) at ultimate bending for Flat Bi. .......................................250
Figure K-8: Relative importance for each random variable that comprise limit state function
in the reliability index results (Step 2) at ultimate bending for Flat Uni. .....................................251
Figure K-9: Relative importance for each random variable that comprise limit state function
in the reliability index results (Step 2) at ultimate bending for Low Bi. ......................................252
Figure K-10: Relative importance for each random variable that comprise limit state function
in the reliability index results (Step 2) at ultimate bending for Low Uni. ...................................253
Figure K-11: Relative importance for each random variable that comprise limit state function
in the reliability index results (Step 2) at ultimate bending for High Bi. ......................................254
Figure K-12: Relative importance for each random variable that comprise limit state function
in the reliability index results (Step 2) at ultimate bending for High Uni. ...................................255
Figure K-13: Step 2 annual structural reliability indices for MS18 designed U slabs for different
HML flows at ultimate shear. ......................................................................................................269
Figure K-14: Step 2 annual structural reliability indices for MS18 designed PSC planks for different
HML flows at ultimate shear. ......................................................................................................269
Figure K-15: Step 2 annual structural reliability indices for MS18 designed I-girders for different
HML flows at ultimate shear. ......................................................................................................269
Figure K-16: Step 2 annual structural reliability indices for T44 designed Precast Tee Slabs for
different HML flows at ultimate shear. ........................................................................................270
Figure K-17: Step 2 annual structural reliability indices for Case Studies for different HML flows
at ultimate shear. ........................................................................................................................270
Figure K-18: Step 2 annual structural reliability indices for SM1600 designed Super-T girders
for different HML flows at ultimate shear. ..................................................................................271
Figure K-19: Relative importance for each random variable that comprise limit state function
in the reliability index results (Step 2) at ultimate shear for Flat Bi. ...........................................272
Figure K-20: Relative importance for each random variable that comprise limit state function
in the reliability index results (Step 2) at ultimate shear for Flat Uni. .........................................273
Figure K-21: Relative importance for each random variable that comprise limit state function
in the reliability index results (Step 2) at ultimate shear for Low Bi. ..........................................274
Figure K-22: Relative importance for each random variable that comprise limit state function
in the reliability index results (Step 2) at ultimate shear for Low Uni. ........................................275
Figure K-23: Relative importance for each random variable that comprise limit state function
in the reliability index results (Step 2) at ultimate shear for High Bi...........................................276
Figure K-24: Relative importance for each random variable that comprise limit state function
in the reliability index results (Step 2) at ultimate shear for High Uni. .......................................277
1. Introduction
1.1 Background
There are about 33,500 road bridges across Australia, and around 17,000 in New Zealand, of which around
70% are more than 40 years old (Ng et al., 2017). Despite the increased need for greater investment in
bridge infrastructure, maintenance budgets are under pressure across the two countries in response to
limited government funding. Concurrently, the freight industry is demanding increased access to the road
network, with road freight expected to double from 2008 levels by 2030 (BITRE, 2014). It is estimated that
75% of existing road bridges (excluding culverts) were designed to traffic load models with magnitudes lower
than present legal vehicle limits (Austroads, 2018c), that is, general access vehicles (GAV). These
competing requirements can be effectively managed through objective bridge capacity assessments against
a quantifiable benchmark of acceptable safety to the public. Only in this way can the existing bridge assets
be utilised to the fullest safe extent possible.
The Oxford English Dictionary defines safety as the condition of being protected from or unlikely to cause
danger, risk, or injury. From this definition, the likelihood (or, more accurately, the probability) of danger or
injury is integral to the concept of safety. Indeed, this is long recognized as Mayer (1926) states “The safety
of the building construction is a matter of calculating probabilities”. In structural engineering, the probability of
failure is calculated, where failure is defined against a specified performance criterion or limit state. Hence,
the level of safety, or reliability, for a structure can be inferred through its probability of failure.
Currently, bridge safety assessments in Australian and New Zealand practice are based upon the load rating
equation. Guidance on this mostly deterministic approach is found in the Australian Standard AS 5100:
Bridge Design – Part 7: Bridge Assessment. Factored values of available capacity for traffic load effects are
compared with load effects for a nominated vehicle desiring bridge access. Ideally, these safety factors ( for
resistance and for loading) are established by calibration to a benchmark acceptable level of safety using
probability approaches such as structural reliability theory (Nowak and Collins, 2012). In this way, the load
rating equation informs safety in a binary fashion. That is, the approach answers the question: “Does the
assessed bridge meet the benchmark for safety?”
The load rating equation in AS 5100 adopts safety factors for design principles selected for a 100-year
design life, with some safety factors (such as traffic loading) selected based on engineering judgement
(Heywood et al., 2000). It is therefore recognised that the current methodology may be overly conservative
as the benchmark safety levels for existing structures should be lower than notional levels for new structures
(Frangopol, 1999). Thus, “higher-tier” assessments are necessary for selected structures. While AS 5100.7
permits higher-tier bridge assessments, little guidance is provided on such methods beyond the current
approach. As a result, bridge asset agencies are having to implement different hierarchical strategies beyond
the AS 5100 methodology to safely manage heavy vehicle freight access to their bridges (Austroads, 2018c).
Recently, Austroads (2018b), investigated available options and methodologies for such higher-tier
assessment of existing bridges. In most cases, these are simply refinements to the load rating equation
inputs to provide more accurate results. However, upon comparison with international practice, the report
recommended further investigations in the transferability and practical applications in the use of probability-
based bridge assessment (PBBA) in the Australian/New Zealand context. PBBA adopts structural reliability
theory to inform on safety as a quantifiable objective continuum measure – the safety spectrum. In other
words, the probabilistic approach can answer the question: “How safe is the assessed bridge?” against
which the asset owner can address the question “Is it safe enough?”. The Australian Standards General
Principles on Reliability for Structures (AS 5104-2017) and Basis for design of structure – Assessment of
existing structure (AS ISO 13822-2005) do provide guidance on the probability-based assessment of
structures. Indeed, these codes can be applied to bridges. Hence, within the suite of Australian Standards,
there is scope for probability-based bridge assessments.
The quantified safety measure is then compared to an appropriate and acceptable benchmark level which
may vary across failure modes and individual bridges, as based on the needs of the bridge asset owner,
public safety, and the economy. Hence, the probabilistic approach has the potential of making individual
bridge assessments more fit-for-purpose.
1.2 Purpose
The purpose of this project is to adopt state-of-the-art structural reliability theory and develop a probability-
based bridge assessment (PBBA) framework for Australian and New Zealand road agencies and
practitioners as a higher-tier assessment methodology for bridge structural elements that are deemed
insufficient through deterministic methods. Such reliability-based probabilistic assessments are already
adopted in international countries’ bridge assessment practice (O'Connor et al., 2009, Wisniewski et al.,
2009). Limited adoption is currently seen in the Australian/New Zealand context.
Specifically, this project aims to demonstrate the benefits in the use of the more advanced probability-based
bridge assessment over conventional deterministic methods, for problematic cases, where the additional
expenditure is deemed worthwhile. This is best illustrated through plots like that shown in Figure 1-1.
Figure 1-1: Plot of rating factor outputs vs structural reliability index illustrating the margin the probability
methods provide in the Australian context (βT = target reliability index).
Benefit of PBBA
In these plots, the AS 5100 deterministic (rating factor) outcome for one bridge structural element is
compared with the probabilistic (structural reliability index) outcome for the same structural element. Each
bridge structural element is a point on the graph. A point in the green region represents where both
assessment approaches indicate the structural element is safe, whereas the red region represents when
both approaches indicate the structural element is unsafe. The top left corner in yellow indicates where
AS 5100 deems the element unsafe, but PBBA deems it safe. The bottom right corner in yellow indicates
where AS 5100 deems the element safe, but PBBA deems it unsafe.
As shown in Figure 1-1, the results often indicate for certain bridge structural elements the probabilistic
assessment can reveal an available margin in safety (i.e., reserve capacity) otherwise not revealed by the
(necessarily) conservative AS 5100 deterministic method.
1.3 Scope
The project will demonstrate the developed probability-based bridge assessment framework through
implementation on a subset of the Australian bridge inventory for two key ultimate limit states. This
implementation is done through two steps:
• Step 1: Determining the safety levels of ‘as-built’ bridges under their design traffic load models as found
in the historical codes (Code-Implied Safety).
• Step 2: Determining the safety levels of the same bridges under current as-of-right freight vehicles and
traffic flow (Current Safety).
The outputs from this project then provide a snapshot of safety levels and margins available to a proportion
of the Australian bridge inventory through the probabilistic methods. The project does not deliver answers on
the safety of all bridges in Australia/New Zealand. Rather, the project’s main aim is to illustrate the benefits of
adopting probability-based assessments as a feasible higher-tier method beyond the current deterministic
methods (Figure 1-1). Levels of safety for other bridges may be obtained in subsequent work using the
framework demonstrated in the current project.
1.4 Outline
This report is purposely detailed and self-contained. As such, it may appear complex for those not familiar
with structural reliability theory and probability-based bridge assessment. For these readers, it is
recommended to examine the listed references, particularly the relevant textbooks, and the glossary located
at the beginning of the report. Nevertheless, Sections 1 and 2 of this report, complemented by Appendix A,
provide the relevant information detailing the background of the proposed framework concepts. Sections 3
and 4 dwell more on the framework and its demonstration, with an example provided in Section 5. Finally,
Sections 6 and 7 provides the results, discussions, and conclusions. The proceeding Appendices present all
results, with further details for those interested in adopting the proposed probability-based bridge
assessment framework.
2.1 Overview
Probability-based bridge assessment (PBBA) adopts structural reliability theory to directly assess the levels
of safety for existing bridge structural components (O'Connor et al., 2009). The level of safety (reliability) is
expressed as the probability that failure to a specific limit state (strength, deflection, cracking, and so on)
does not occur. The framework presented here is a mixture of Level II and III reliability methods (Madsen et
al., 2006).
Although referred to here as a deterministic approach, the rating factor methodology of AS 5100.7 uses
partial factors for load and resistance. Notionally, these partial factors are tailored to achieve a specific level
of safety as is later explained. Such methods are termed Level I reliability code formats (Madsen et al.,
2006). Nevertheless, as will be seen later, the AS 5100.7 factors have not been reliability calibrated and
hence the term ‘deterministic’ is adopted to describe the approach which obscures its intended probabilistic
basis.
The PBBA procedure in comparison to the conventional deterministic approach is shown in Figure 2-1. The
same initial step is taken in both approaches, the outputs of which provide information for decision-making in
the PBBA framework. It is important to note that:
• The conventional Rating Factor approach is always first conducted;
• The higher-tier PBBA is only conducted if it is deemed likely to be of benefit, in either:
– Finding that the component or limit state actually has acceptable safety, or;
– Guiding the design of safety-maximal cost-optimal rehabilitation measures.
If it is determined that PBBA is cost-effective (usually if the structural deficiency is not major or if there is a
special loading scenario), then only those critical components failing the traditional assessment are further
evaluated. In this way, PBBA makes the most value out of all available data and minimises the cost of the
assessment itself.
Contrary to what some may think, absolute safety (zero risk or zero probability of failure) does not exist.
Structural capacity or resistance (R) and applied loading or load effects (S) are uncertain. Hence, there is
always a chance that loading exceeds the capacity. In PBBA, these uncertainties in R and S are established
and commonly represented through probability distributions (Figure 2-2) described by parameters such as
the bias (λ = characteristic value divided by the mean) and coefficient of variation (CoV = standard deviation
divided by the mean). A review of probability distributions is found in Appendix A. Note, uncertainties in the
design practice and construction practices (or human error) are considered to be removed by quality
control/quality assurance schemes.
Once the uncertainties are established, the probability of failure (the chance loading is greater than capacity)
is found using structural reliability methods. A range of textbooks are available with worked examples for
these structural reliability methods (Ang, 1984, Stewart and Melchers, 1997, Haldar and Mahadevan, 2000,
Schneider, 2006, Lemaire et al., 2009, Cremona, 2012, Nowak and Collins, 2012, FIB CEB-FIP, 2018,
Melchers and Beck, 2018).
fs(x), fR(x)
Capacity (R)
x
Failure region
There are several commercially available software packages with structural reliability methods built in. The
following freely available programs are employed in this work:
• Finite Element Reliability Using Matlab (FERUM) (https://www.sigma-clermont.fr/en/ferum)
• Framework for Uncertainty Quantification (UQLab) (https://www.uqlab.com/)
The structural reliability index (β) is often used as a convenient and normative measure for expressing the
probability of failure, and thus the level of safety, of the bridge structural component. More details can be
found in Appendix A.
Since the 1970s, codes of practice are themselves written to achieve a certain level of safety based on
structural reliability methods. Philosophically then, there is no difference to the basis of PBBA and a code of
practice. Instead:
• PBBA is the development or writing of a bridge or component-specific code of practice.
• In PBBA there is no compromise on safety; Indeed, the level of safety is explicitly known.
The acceptability of the calculated level of safety in a PBBA is defined by a predefined benchmark or the
target reliability index (βT). The optimum value for the target reliability index depends on a range of factors.
These include the structure type, the limit state function or type of failure (serviceability/ultimate,
brittle/ductile), the expected cost of failure, cost of increasing safety, and existing levels of safety
(Vrouwenvelder, 2002, Sykora et al., 2017).
The Australian Standard on General Principles on Reliability for Structures, AS 5104 (2017), (adopted from
ISO 2394-2015) provides “tentative” advice for annual target reliability indices for ultimate limit states based
on economic (cost-benefit) optimisation acceptance criteria (Table 2-1). Further details can be found in
Appendix B.
Table 2-1: Annual target probability of failures (reliability index) based on economic optimisation.
Consequences of Failure
Relative Costs of Safety Measures
Class 2 (Minor) Class 3 (Moderate) Class 4 (Large)
Large 1.0×10-3 ( = 3.1) 4.8×10-4 (β = 3.3) 1.1×10-4 (β = 3.7)
Medium 1.1×10-4 (β = 3.7) 1.3×10-5 (β = 4.2) 5.4×10-6 (β = 4.4)
Small 1.3×10-5 (β = 4.2) 5.4×10-6 (β = 4.4) 1.3×10-6 (β = 4.7)
Source: AS 5104-2017.
From Table 2-1 the target reliability index is selected on the relative costs of safety measures and
consequence of failure. The relative cost is the cost needed to reduce the risk. Meanwhile, the consequence
of failure relates to loss of life and direct and indirect failure costs. Ultimately, the setting of the target
reliability index is dependent on the ‘risk-appetite’ of the asset owner considering the relative cost of safety
measures and consequences of failure for the limit state of interest.
The choice in the acceptable level of safety is a significant benefit over the conventional deterministic
method where the target reliability index is set uniformly through the resistance and loading safety factors ϕ
and γ. This is what allows PBBA to provide tailored bridge assessments.
The UK bridge assessment code of practice BD (79/13-2013) allows the use of full probabilistic reliability
analysis (PBBA) methods with the agreement of the relevant authorities and organisations. The standards
recommend specialist knowledge and verification of the analysis by an independent organisation.
Bridge assessments in the United States are carried out according to the AASHTO (2018). There are three
methods of assessment, increasing in complexity. The third level of assessment uses PBBA. First order
reliability method (FORM) and probability distributions presented in the design code are recommended.
2.2.3 Europe
The Eurocode suite of structural design codes is now the legal document in many European countries and
commonly used for bridge assessments. Since the Eurocodes allow for full probabilistic designs EN (1990-
2002) Section 3.5(5) based on a target reliability index, bridge-specific assessments are feasible in the
present framework.
The final report from European Cooperation in Science and Technology (COST) Action 345 summarises a
detailed research project in Europe which aimed to provide procedures required for the assessment of
highway structures (Cost Action 345). In particular, Chapter 5 of the report presents on procedure for
probability-based bridge assessments. The chapter lists how uncertainties are modelled, discusses on target
reliability indices, and lists the increasing levels of structural reliability analysis.
PBBA is being used in some European countries such as Poland, Ireland, and Denmark, among others.
Recently, the uptake has been significantly increasing since the completion of COST Action TU 1406 (2019).
The report presents on how PBBA is adopted to simulate different scenarios based on various limit states
and reduce the computational costs.
The International Federation for Structural Concrete fib Bulletin 86 (FIB CEB-FIP, 2018) presents a review of
procedures to adopt PBBA. This work is going towards the work of fib Model Code 2020. Both Model Code
2020 and the new generation of Eurocodes will detail assessment existing structures, based on a PBBA
procedure.
Denmark
Significant cost-savings through adopting PBBA on highway bridges have been observed in Denmark
(Enevoldsen, 2011). This motivated the Danish Road Directorate (DRD) to publish a practical guideline on
PBBA to be performed on other bridges where suitable (Danish Road Directorate, 2004). A flowchart of the
procedure adopted for classifying bridge reliability by DRD is shown in Figure 2-3.
In Denmark all bridges are first evaluated using deterministic analysis based on separate assessment
guidelines by DRD. When any particular structural component fails the deterministic assessment, the PBBA
seen in Figure 2-3 is performed if deemed beneficial (Figure 2-1).
PBBA is used seldomly in Australian and New Zealand practice. New South Wales (RMS) used structural
reliability for the assessment of Iron Cove Bridge, resulting in favourable outcomes for heavy vehicle access
(Austroads, 2018b). The literature presents some other additional examples where PBBA have been
beneficial in Australian (Reid, 2004, Shah et al., 2009) and New Zealand bridges (Waldin et al., 2018).
The Australian Standard AS 5100 does not give specific guidance on the target level of safety (or reliability
index) that is achieved by assessments to the code. However, AS 5104-2017 and the Australian Standard on
Basis for design of structure – Assessment of existing structures (AS ISO 13822-2005) do provide guidance
on the probability-based assessment of structures. Indeed, these codes can be applied to bridges. Hence,
within the suite of Australian Standards, there is scope for probability-based bridge assessments.
Probability-based structural assessments are not a new concept in Australia. For example, the Australian
Building Code Board (ABCB) has published an informative Structural Reliability Handbook for assessing the
safety performance of buildings (ABCB, 2015). An overview of the procedure is shown in Figure 2-4.
Refer to sourced document for Chapters and Sections. Note DRD refers to the reliability index as the “safety index”.
3.1 Overview
Figure 3-1 presents the proposed higher tier assessment as a probability-based bridge assessment (PBBA)
of bridge structural elements and its incorporation into the existing AS 5100 deterministic methodology. The
probability-based bridge assessment framework adopts structural reliability theory to calculate the annual
structural reliability index (β) – the actual level of safety of the structural element for an assessed bridge.
Figure 3-1: Proposed higher-order assessment within the Australian bridge assessment flowchart.
Clause
Data collection
10.2
Clause Section
Bridge condition Limit state function
10.3 4.1
Clause
Field geometry
10.4
Clause
Fatigue
13
Yes No Yes No
The framework workflow has some steps corresponding to the AS 5100.7 deterministic method, but instead
now considering the variable and uncertain nature of capacity and loading as discussed in the subsequent
sections of this document (as indicated in Figure 3-1). In the deterministic method, “Adequate Capacity” is
when the rating factor (RF) is greater than 1. In the proposed probabilistic methods “Acceptable Safety” is
when β is greater than the prescribed target reliability index (Appendix B).
PBBA is not a single approach, but instead a suite of approaches of increasing complexities (Melchers,
1999). Hence, a feedback loop exists in Figure 3-1 for when a lower level of PBBA complexity indicates the
bridge structural element is still unsafe, and that further analysis could be beneficial. Details on these higher-
order PBBA assessments can be found in the literature. Nevertheless, it may be found that strengthening
measures are required, but the outputs of the PBBA typically inform more targeted strengthening measures
(reliability-optimal) that are often cheaper than the a priori measures (cost-optimal) (O'Connor et al., 2009).
The question “PBBA Beneficial?” is included in the higher-order assessment flowchart to highlight that the
PBBA itself is carried out only on the critical bridge components, and upon the discretion of the engineer or
road authority in situations where the additional cost of the higher-tier assessment itself is deemed
worthwhile through the avoidance or reduced cost of the final rehabilitation measures.
The framework shares similarities to international adoptions, such as those found in the Danish Road
Directorate Report 291 (Danish Road Directorate, 2004) shown in Figure 2-3.
Note, specific elements of bridge condition and fatigue presented in the AS 5100 deterministic methodology
are not considered for demonstration, but can be easily be incorporated into a PBBA if the information is
known (Crespo-Minguillón and Casas, 1998).
The demonstration of the framework is conducted on a selected subset of already existing or designed
bridges in the Australian inventory. Clause 10.7 of AS 5100.7 lists critical sections to analyse and so for the
demonstration of the framework only the components of the bridge superstructure (i.e. beams, girders) are
considered. The methods presented here may be easily applied to other sections.
The representative bridges subset is further divided by bridge families. These families are sets of bridges
that have the same material type, structural form, number of traffic lanes, and design code, but vary in
section depths and span length.
The selected bridge families for the bridge inventory subset are summarised in Table 3-1.
Historically, standard engineering drawings were developed to a specific code (and thus load model) and
used for the construction of bridges in Australia. These drawings provided the precise span and specified
exactly the material properties and amount of any active, passive, and stirrup reinforcement. These drawings
allow an easy definition of a bridge family. Nowadays, bridge designs tend to be more individualised. For
example, the reinforcement arrangement may be different for the same structural form and span. Hence
families are not as easily classified. Instead, appropriate lower and upper spans of the different standardised
sections can be selected to define the bridge family. These lower and upper spans are determined based on
the relevant design drawings and the relevant literature.
Table 3-1: Bridge structural families to be assessed in estimated chronological order of construction.
Design Code
Era Structural Form Victorian Stock Sections and Spans
(Load Model)
(15’) 4.57 m
(20’) 6.10 m
High-strength RC U-
1953 – 1976 NAASRA (MS18) 6.2% (25’) 7.62 m
slabs
(30’) 9.14 m
(35’) 10.67 m
(15’) 4.57 m
(20’) 6.10 m
1953 – 1976 NAASRA (MS18) PSC Precast Planks 7.4% (25’) 7.62 m
(30’) 9.14 m
(35’) 10.67 m
(35’) 10.67 m
(40’) 12.19 m
1953 – 1976 NAASRA (MS18) PSC I-girders 9.9% (45’) 13.72 m
(50’) 15.24 m
(60’) 18.29 m
1976 – 1992 NAASRA (T44) PSC I-girders From TMR 19 m
(450 mm) 11 m
(450 mm) 13 m
PSC Precast Tee
1992 – 2004 ABDC (T44) 7.5 % (550 mm) 14 m
Slabs
(650 mm) 16 m
(750 mm) 19 m
1992 – 2004 ABDC (T44) PSC Super-T Girders From VicRoads 26.5 m
(T1) 15 m
(T1) 20 m
(T2) 17 m
(T2) 23 m
(T3) 18 m
2004 – 2017 AS 5100 (SM1600) PSC Super-T Girders* 3.0%
(T3) 27 m
(T4) 22 m
(T4) 32 m
(T5) 28 m
(T5) 37 m
As justified in Appendix C, the most prominent bridges in the state of Victoria are concrete with two lanes of
traffic. The selected families make up approximate 40% of all Victorian bridges. In addition to these families,
two additional individual bridges are included in order to capture further Australian historical design codes
and load models that have been used in bridge engineering practice. More details on historical design codes
and load models can be found in the literature (O'Connor and Shaw, 2000, Austroads, 2014, Austroads,
2018a). In total, thirty-two bridges are selected for demonstration of the PBBA framework. Since around 50%
of the existing bridges in Australia have been designed to MS18 (Austroads, 2018a), almost 50% of the
bridges (15 out of 32) in this study are designed to MS18.
MS18 designed I-girders, HSRC U-slabs, and PSC precast planks families are defined by the Victorian
standard design drawings to MS18, each at five spans. These drawings have been provided by the
Department of Transport (former VicRoads) and range from 13 m (45’) long and 20 m (65’) long for I-girders,
and 4.57 m (15’) to 10.5 m (35’) for the others. The family of T44 (ABDC) designed precast Tee-slabs are
based on the lower and upper spans for four standardised depths (450 mm to 750 mm) based on the design
drawings from the Department of Transport and Rapattoni and Wells (1991) with an additional 160 mm
overlay slab. Meanwhile, the family of SM1600 designed Super-T girders are based on the recommended
lower and upper spans for each of the five standard depths (T1 to T5) as per the National Precast Concrete
Association Australia (2002) with a 180 mm overlay slab. For the additional bridges, drawings for the T44
(NAASRA) designed I-girders were provided by the Queensland Department of Transport and Main Roads
(TMR) and the T44 (ABDC) designed Super-T drawing was provided by the Victorian Department of
Transport.
It is important to note that the Super-T girders have been designed precisely to the specifications of AS 5100
2004 by the project team. Unlike the other bridge families, the SM1600 designed Super-T girder family are
not already built structures. Moreover, practical detailing and construction staging is not considered. Thus,
the considered Super-T girders are the slimmest possible complying designs meeting the code requirements
for the considered limit states. Thus, these designs do not contain the conservatisms that may be introduced
intentionally by a designer or indirectly through design for other limit states. The girder designed is one of five
Super-T girders in bridge deck considered typical of a two traffic lane carriageway.
In the reference vehicle technique currently adopted by road agencies for heavy vehicle access (Austroads,
2018c), ultimate bending, ultimate shear, and reaction at the critical locations are typically considered. In this
work, reaction force is not considered since it is not a mode of failure for the superstructure (only a proxy for
substructure failure) and substructure failure is so strongly site-specific that the necessary assumptions
would render the results meaningless for the vast majority of cases. Hence, the modes of failure for
demonstration purposes are ultimate bending and ultimate shear. Note, ultimate shear encapsulates three
modes (web-shear, flexural-shear, and web-crush). Of course, the proposed framework can be adapted for
assessment of other modes of failure such as serviceability or fatigue (Crespo-Minguillón and Casas, 1998).
Traffic Loading
Throughout the study, two traffic lanes are considered. To avoid the influence of the choice of design lane
width, two corresponding design lanes are therefore considered. Designs for two traffic lanes but with a
carriageway width that requires 3 or more design lanes are therefore more conservative than the results here
suggest. Nevertheless, the PBBA framework applied to bridges of any number of traffic or design lanes.
Step 1 in the implementation adopts the relevant design traffic load model (MS18, T44, SM1600), across
both design lanes, considering accompanying lane factors and dynamic load allowances as per the clauses
of the relevant code/standards (NAASRA, ABDC, AS 5100).
Step 2 uses traffic streams from as-of-right freight vehicles established from weigh-in-motion (WIM) data.
While there are a number of different freight networks in Australia, the Higher Mass Limit (HML) network is
selected for Step 2. Hence, the investigated current as-of-right freight vehicles fleet are heavy vehicles (mass
greater than 4.5 tonnes) approved to access Higher Mass Limit (HML) networks. Note, HML networks
includes General Access Vehicles (GAV). In Victoria, an accurate piezoelectric pavement weigh-in-motion
(WIM) station is used on the West Gate Bridge which is part of the HML network. Data from this WIM station
is used for Step 2. Step 2 also uses a probabilistic formulation for the dynamic load allowance after Caprani
(2013), which is explained later.
Traffic Flows
For traffic flows, three different arrangements as presented in Lipari et al. (2017) are considered for Step 2
for the demonstration. Hourly changes in flow and truck proportions are based on data from German
motorways (Geistefeldt and Hohmann, 2013). These are deemed a suitable representation of Australian/New
Zealand traffic flows. The hourly flows are plotted in Figure 3-2 and detailed in Appendix D. Two annual
average daily traffic scenarios (across both lanes) are considered:
• 25,000 vehicles per day (across both lanes) in two daily patterns: uniform (Flat) and peaked (Low)
• 50,000 vehicles per day (across both lanes) in one daily pattern (High)
Figure 3-2: Traffic flow rates, hourly variation, and truck proportions considered in the demonstration of the
framework.
3500
Total
3000
Trucks
2500
Flow (veh/h)
2000
1500
1000
500
0
0 3 6 9 12 15 18 21 24
Hour
(a) 25,000 vehicles per day (Flat). Data points can be found in Appendix D.
3500
Total
3000
Trucks
2500
Flow (veh/h)
2000
1500
1000
500
0
0 3 6 9 12 15 18 21 24
Hour
(b) 25,000 vehicles per day – morning peak (Low). Data points can be found in Appendix D.
3500
Total
3000
Trucks
2500
Flow (veh/h)
2000
1500
1000
500
0
0 3 6 9 12 15 18 21 24
Hour
(c) 50,000 vehicles per day (High). Data points can be found in Appendix D.
A daily flow of 50,000 vehicles per day implies that the road is congested throughout the day-time. For the
lower flow of 25,000 vehicles per day, the two patterns represent a strong morning peak rush hour, and a flat
daily pattern.
Both uni-direction lane flows (Uni) and bi-directional lane flows (Bi) are considered for the demonstration. For
uni-directional flows, the traffic flows of trucks are divided as 80% in the slow lane (deemed Lane 1, assumed
speed = 80 km/h), and 20% in the overtaking lane (deemed Lane 2, speed = 100 km/h). This proportionality
of trucks is similar to other studies (Fwa and Li, 1995, Caprani et al., 2016). The remaining vehicles (cars)
are divided equally across the two traffic lanes. In this way, there are generally more vehicles in the slow
lane (Lane 1) compared to the fast lane (Lane 2). Meanwhile for bi-directional lane flows, the traffic flows of
vehicles (cars and trucks) are divided equally across both traffic lanes, with the same proportion of trucks as
indicated in Figure 3-2.
With these considerations, there is a total of 6 different traffic flow considerations, referred to in this report as:
Flat-Uni, Flat-Bi, Low-Uni, Low-Bi, High-Uni, High-Bi. The hourly breakdown of trucks and cars for these six
traffic flows are shown in Appendix D. Indeed, these traffic flows are simply for demonstration of the
framework. Other more appropriate flows to any particular bridge or network can be adopted in future
applications of the framework.
4. Framework Components
where
self-weight load effect of bridge deck component (beam; girder; plank; etc.)
Si =
(probabilistic)
Notice in Equation 1, variables are considered to be probabilistic. That is, they are random or uncertain
variables. As per structural reliability theory, these uncertainties must be quantified and expressed as
probability distributions in a probability-based bridge assessment (see Appendix A). Moreover, the random
variables in the limit state function are considered to be independent to each other. Further variables can be
considered as necessary for a particular case, potentially including correlations between loading and
resistance variables.
Uncertainty in the maximum traffic load effect varies depending on the exposure duration of interest. In other
words, the uncertainty in daily maximum traffic load effects are different to yearly maximum traffic load
effects. This means SLL is a stochastic variable defined against a given reference period (see Appendix A).
The selected reference period is therefore the same reference period for the calculated structural reliability
index β, and thus the level of safety.
For bridge assessment, quantification of the design-life (100-year) level of safety is not as useful as
quantifying the annual level of safety. An annual reference period is therefore selected and is commonly
used as a reference period. This means the uncertainty for SLL must be the annual maximum traffic load
effect, so that an annual structural reliability index β, is calculated. Assuming stationarity in loading and
resistance, this annual reliability can be converted to 5- or 10-year levels of safety as may be necessary in
an assessment situation.
The assumption that loading and resistance do not change with time (stationarity) can be relaxed, and a
time-dependent reliability assessment can be conducted incorporating degradation and loading increases as
may be necessary for particular bridge assessments, as described in Appendix A.
Clauses 10.3 and 10.4 of AS 5100.7 discuss the management of bridge condition and the need for
measuring field geometry. However, Clause 10.5 notes that in the absence of this information the bridge
components are to be considered in ‘as-built’ condition.
For this framework, it will be assumed bridge structures are in ‘as-built’ condition. However, it is possible if
desired to introduce bridge condition and deterioration processes into a more sophisticated framework using
time-dependent reliability methods. Such methods are discussed in Appendix A.
One way to establish the uncertainty of the section capacity (R) is through Equation 2 (Nowak et al., 1994,
ABCB, 2015):
R = Km Kf Ks Rn 2
where
Whilst some guidelines give suggested K values (JCSS, 2000, Danish Road Directorate, 2004), they may not
be applicable to the Australian context, and so an alternative approach is preferable.
An alternative and more complex approach to establish the probabilistic distribution, and thus the
uncertainty, of the section capacity is to use Monte Carlo simulation (MCS) (Rubinstein and Kroese, 2016,
Haldar and Mahadevan, 2000). This is adopted for Step 1 and Step 2 in the demonstration of the framework.
Further details are found in 0 and Appendix F. Unlike the deterministic approach, the capacity reduction
factor (ϕ) is not included in establishing the section capacity random variable in the probability approach as
this is the means of managing safety in a deterministic setting.
In MCS, the selected input variables for the relevant structural capacity models (usually the design
provisions) are considered random. Then, for the simulation, the structural capacity model is applied N
number of times, with each simulation selecting one realisation from the appropriate probability distribution of
the input variable. This provides N realisations of the section capacity. A probability model for the section
capacity is then developed using appropriate fitting techniques (Ang, 1984), as discussed in Appendix A.
Given ultimate bending and ultimate shear are considered for demonstration, then the bending provisions
and the recently introduced modified compression field theory (MCFT) based shear provisions in AS 5100.5
2017 (+ Amdt. 1) are used to determine structural capacity. These are summarised in Appendix G. Note, that
MCFT-based provisions require adaptation for assessment which is not clearly indicated within the clauses
of the code (Caprani and Melhem, 2019). Moreover, as seen in Appendix G, ultimate shear encapsulates
three modes; web-shear, flexural-shear, and web-crush.
The uncertainty of these input variables is represented by relevant probability distribution described by
statistical parameters such as the bias and coefficient of variation (see Appendix A). Based on a literature
review, Table 4-1 presents the probability distributions for the input parameters of the considered structural
capacity models. These probability distributions describe various uncertainties such as material quality (for
material properties) or fabrication process (for geometric properties). Notice that not all variables are
considered to be random or uncertain by the literature, and instead some are treated as deterministic.
Table 4-1: Uncertainty or probabilistic distributions (expressed as bias and CoV = coefficient of variation) for
section properties used as input parameters for the ultimate bending and ultimate shear capacity
models (Appendix G).
Effective shear width bv Shear 1.01 0.04 Ref. A 1.01 0.04 Ref. A
Rebar cross-sectional area As Both 1.00 0.02 Ref. C 1.00 0.02 Ref. C
Stirrup-to-spacing ratio Asvs Shear 1.00 0.02 Ref. C 1.00 0.02 Ref. C
Concrete compressive strength f’c Both 1.03 0.18 Ref. B Varies Ref. A
Steel rebar yield strength fsy Both 1.18 0.12 Ref. B 1.15 0.05 Ref. A
Steel stirrups yield strength fsys Shear 1.18 0.12 Ref. B 1.15 0.05 Ref. A
Prestress strand yield strength fpy Both 1.04 0.03 Ref. E 1.05 0.02 Ref. A
Elastic modulus for concrete Ec Shear 1.00 0.08 Ref. C 1.00 0.08 Ref. C
Elastic modulus for steel Es Both 1.00 0.04 Ref. C 1.00 0.04 Ref. C
Elastic modulus for prestress Ep Both 1.00 0.20 Ref. C 1.00 0.20 Ref. C
Sources: A) Foster et al. (2016); B) Pham (1985); C) Wiśniewski et al. (2012); D) Mathieu (1991); E) Ellingwood (1980).
All random variables are considered normal, expect for f’c, which is taken as log-normal.
The latest known Australian publication for which statistical analysis of these properties was performed is the
work by Foster et al. (2016), where mainly material properties were investigated. Prior to this, the next
Australian publication is the 1980s research by Leicester et al. (1985) and Pham (1985) who again
considered only the material properties. There appears to be a gap in the statistical analysis for Australian
material properties between the 1980s and 2010s. Moreover, there has been only specific geometric
statistical analysis between this time, with Brameld et al. (1986) specifically looking at prestressed concrete I-
girders. As a result, it becomes difficult to capture the uncertainty for each individual bridge family era shown
in Table 3-1, particularly without Australian (nor international) statistics earlier than the 1980s. To address
this, as shown in Table 4-1, two groups are created – ‘historical’ (before 2000) and ‘recent’ (after 2000).
Bridge families are then assigned these uncertainties for their section properties accordingly. That is:
• Historical Bridges: All bridges designed to MS18, and Precast Tee Slab and I-Girder designed to T44;
• Recent Bridges: All Super-T girder bridges (designed to T44 or SM1600).
The choice of 2000 is one of judgement as an approximate average between 1980 and 2010s, and
corresponds roughly to the introduction of AS 5100 (2004). For other section properties where no relevant
Australian statistics could be found, the most relevant international statistics were used instead. All random
variables shown in Table 4-1 are considered to be normally-distributed, except for the compressive strength
of concrete which is taken as log-normal (Foster et al., 2016). In addition, statistics for the ‘recent’ compressive
strength of concrete varies with different concrete strength and can be found in Foster et al. (2016).
Correlations
Correlation describes the dependencies between pairs of random variables. Correlations between the
material properties (eg. strength and elasticity) must be considered as part of the Monte Carlo simulations
(MCS). Typically, Pearson’s linear correlation coefficient, ρ, is used to describe the linear relationships.
Suggested correlation coefficients between the assumed non-independent random variables listed in
Table 4-1 used in this study are presented in Table 4-2.
Table 4-2: Correlations between non-independent random variables taken in Monte Carlo Simulation.
Conventionally, finite element models such as grillage models are constructed to provide the characteristic
load effects from permanent and transient loads. These are then multiplied by load safety factors (γ). In the
probabilistic framework, these load effects are described as random variables. Similar to resistance, load
factors are not included in the probability distributions for load effect in the probability approach.
To establish the probability distribution of the permanent load, the characteristic values from grillage models
can be converted to probability models based on appropriate statistical basis (bias and coefficients of
variation) from the literature. Recommendations for probabilistic distribution of permanent loads by Rakoczy
and Nowak (2013) presented in Table 4-3 are generally adopted in the literature (Caprani et al., 2017, Estes
and Frangopol, 2005).
Table 4-3: Recommended uncertainty or probability distributions (expressed as bias and CoV = coefficient of
variation) for permanent loading. All are normal (Gaussian) distributed.
Details on the grillage and finite element models developed as part of the framework demonstration to obtain
the characteristic value, and thus the probability distribution, are found in Appendix H. The models are
simply-supported with no skew or cross-fall. An exception is the T44 designed I-girder bridge deck, in which
TMR provided the grillage model for the bridge deck.
If a statistical description or basis of the maximum loading is known or assumed, the uncertainty or
probability distribution of the annual maximum traffic load can be established from the characteristic value
obtained by a grillage model for a specific vehicle traversing the bridge.
By definition, the characteristic value is associated with a probability of exceedance in a reference period ().
This is the probability that values of random variable exceed the characteristic values during the reference
period.
When Australian limit state design codes were drafted in the late 1980s, it was agreed that “ultimate loads
have only 5% chance of occurring during the design life of 100 years” (Lay, 1980, Fenwick, 1985). This is in
contrast to international practice where only design (nominal) loads are specified to have a 5% probability of
exceedance, and ultimate loads are determined from these using reliability-calibrated partial load factors.
Nevertheless, this statement continues to remain in Clauses of the Australian limit state design codes in the
present day with AS 5100.1 (2017) Clause 8.3.1 (i) stating:
“An ultimate design action is an action that has a 5% probability of being exceeded during
the design life”
Thus, for concordance with AS 5100, the probability of exceedance associated with ultimate loads (taken as
characteristic loads) is taken as 5%. From this probability of exceedance, and further knowing the coefficient
of variation and distribution type, the bias can be established. Melchers and Beck (2018) presents biases for
different probabilities, coefficients of variation, and distribution types. Maximum traffic load effects are
commonly assumed in the literature to follow a Gumbel distribution (OBrien et al., 2015). As explained by
Caprani et al. (2017), the bias for the Gumbel distribution from a given probability of occurrence and
coefficient of variation is shown in Equation 5:
1+r
()= 5
1+rsp ()
where
= bias
−1
r = ( - )
CoV√6
probability the characteristic value (traffic load model stipulated in the code) exceeding
p() =
within the reference period () (probability of exceedance)
The international literature recommends a coefficient of variation of 0.18 for bridges, including dynamic
allowance (Rakoczy and Nowak, 2013). As presented in 0, this is slightly conservative to the observed
values in Australia.
Recall, an annual reference period is selected for the reliability assessment. Hence, the probability of
exceedance in Equation 5 must be converted from the design-life to an annual basis. Details on this
conversion are found in 0. Based on the annual probability of exceedance, a coefficient of variation of 0.18,
and assuming peak load effects follows a Gumbel distribution, the bias for annual peak traffic load effect for
Step 1, as calculated by Equation 5, is equal to 0.50 (ratio of mean to characteristic).
The calculated bias of 0.50 is applicable to the limit state design codes for which the characteristic value is
defined. However, with the more historical codes, not based on limit state design, it is not known if the load
effects are the mean or characteristic value. Consequently, the probability of exceedance is not known.
Because latter codes are usually tested to at least give the same safety as older codes, it is a reasonable
assumption to take these historic load effects as also characteristic and the probability of exceedance
definition the same as in the limit state design codes. Following this assumption then, the same bias of 0.5
and coefficient of variation of 0.18 applies for the considered historic codes.
When a statistical basis for a loading is not known, real traffic data can be collected and processed, either
through Weigh-in-Motion (WIM) or through the resulting load effects using Structural Health Monitoring
(SHM). The probability distribution can then be obtained either by direct application of the recorded data or
simulations of the summarised traffic data on necessary bridge influence lines/surfaces. In this way, the
uncertainty of vehicle loading, including overloading, is captured. Examples of these different approaches
are found in the existing literature (Nowak, 1993, Caprani et al., 2008, Enright and O'Brien, 2013, Eamon et
al., 2016).
Simulations are particularly useful as they can provide a longer period of data (measurement length, T),
compared to the direct application of the recorded traffic data. Simulation is adopted in Step 2 of the
framework demonstration using the available WIM data from the West Gate Bridge provided by the Victorian
Department of Transport.
The in-house software Bridge Traffic Load Simulation (BTLS) simulates artificial traffic streams crossing
bridges with pre-defined properties (span length, number of lanes, etc) using Monte Carlo simulation, and
calculates resulting load effects at the critical locations of the bridge structural element using appropriate
influence lines or surfaces. Specific details on this procedure is found in Appendix F. For the demonstration,
traffic is simulated on each of the two lanes for the bridges’ families with spans presented in Table 3-1 for the
six traffic flows considered (Flat-Uni, Flat-Bi, Low-Uni, Low-Bi, High-Uni, High-Bi) presented in Appendix D.
The traffic lanes are taken as the design lanes used in finite element models from Step 1.
For generating the vehicle in the artificial traffic stream, the Monte Carlo simulation adopted is known as a
smoothed bootstrapping using variable-bandwidth kernel density estimators (Enright, 2010, Scott, 2015). In
this method, a vehicle is randomly selected from a garage of measured vehicles (bootstrapping), and then
randomised (smoothing) using deviates (kernels). Thus, a very large number of artificial heavy vehicles in the
traffic stream can be generated, including overloaded vehicles, while maintaining correlations and
relationships within the vehicle characteristics (e.g. correlation between axle weights and gross vehicle
mass). Details on the creation of the “garage” are found in Appendix F.
The simulation is performed for 25,000 ‘economic’ days (representing 100 calendar years) in BTLS using an
appropriate headway model. Details can be found in Appendix F. Using the Block Maximum method (see
Appendix A) realisations of the maximum load effect each year are collated and the probability distribution for
annual maximum traffic load effect is established using appropriate fitting techniques (see Appendix A).
Since maximum values are selected, it is expected that probability distributions fit a General Extreme Value
(GEV) probability distribution (OBrien et al., 2015). Rather than using the conventional bias and coefficient of
variation (CoV), this probability distribution type is better described by its parameters: shape (), scale (),
and location (u).
An important source of uncertainty that must be captured is the uncertainty inherent in the models employed
for quantifying section capacity and load effects, known as epistemic uncertainty (Der Kiureghian and
Ditlevsen, 2009, Holický et al., 2016). This is separate to the inherent statistical uncertainty in the input
parameters used for the models to quantify capacity and loading, known as aleatoric uncertainty. The
epistemic model uncertainty is more commonly referred to as the model error (ω).
As per the suggested limit state function in Equation 1, two model errors are considered:
• ωR: Error associated with the model adopted to quantify structural capacity, R (further divided into the
bending capacity model error, and shear capacity model error – the two modes of failures being
investigated).
• ωS: Error associated with the model adopted to establish the bridge load effects, S (loading model error).
Until recently, there appears to be no data or statistical parameters described in previous Australian-based
reliability calibration studies for model error (Foster et al., 2016), particularly for the recently-introduced
MCFT-based shear provision. The recommended capacity and loading model errors can be found in
Appendix I detailed in the literature (JCSS, 2000, Danish Road Directorate, 2004, Foster et al., 2016,
Melhem et al., 2018).
Since Step 1 is interested in code-implied safety, the dynamic load allowance (DLA) is not considered to be
uncertain for Step 1. Instead, the DLA is taken as the stipulated values based on the relevant code/standard.
For Step 2, a probabilistic model for DLA is used. As Caprani (2017) explains, it is known that the dynamic
amplification factor (DAF = 1 + DLA) decreases with increasing static loading, such as single very heavy
trucks, or multiple vehicle presences. Thus, there is a negative correlation between the distributions of DAF
and static loading. Two aspects are then necessary to be determined:
• The distribution of the DAF itself;
• The degree of correlation between the DAF model and the static loading.
The probabilistic model proposed by Caprani (2013) is adopted for the distribution of DAF. In this model, the
dynamic amplification factor (DAF = 1 + DLA) is taken as the GEV distribution for minima, with parameters
as = -0.597, = 0.176, and u = 1.294. This distribution corresponds to a 20 m span, considering 95% of
DAFs are higher than 1.05 and a mean DAF of 1.26, as consistent with the literature (Caprani, 2013, Caprani,
2017). Until there is more information available about annual maximum distributions, the distribution of DAF is
conservatively taken here as that which appears to be closer to the distribution of daily maximum DAF.
In the model proposed by Caprani (2013) the correlation is effectively -1.0. However, to add a further degree
of conservatism, and to admit some scatter in the DAF values at high levels of static loading, in this work a
correlation coefficient of -0.5 is used. This is until better models of annual maximum dynamic and static load
effects, such as that in Caprani et al. (2012), are available for the Australasian context. An example set of
realisations of annual maximum static load effect and DAF using the model is shown in Figure 4-1.
Figure 4-1: Sample realisations from the joint distribution of annual maximum static load effect and dynamic
amplification factors, showing the marginal histograms, for the TMR T44 I-girder bridge bending
moment using bi-directional high flow.
Recall, the structural reliability index (β) is often used as a more convenient and normative probability of
failure, and thus the level of safety. This is defined in Equation 6 and detailed in Appendix A.
β = -Ф-1[ Pr (g < 0) ] 6
where
β = reliability index
The structural reliability index (β) is calculated using structural reliability methods (Appendix A). For linear
limit state functions, the First Order Reliability Method (FORM) is readily applicable with the Second Order
Reliability Method (SORM) used as a verification on the linearity of the limit state function. Monte Carlo
simulation (MCS) is also common and does not rely on the limit state surface. These structural reliability
methods are available in the software packages FERUM and UQLab.
4.7 Acceptability
Recall, the acceptability of the calculated level of safety is addressed through comparison with a predefined
benchmark or the target reliability index (βT) as presented in Table 2-1 from AS 5104 (2017).
The target reliability index is usually the implicit level of safety from previous generation of codes of practice.
However, this is not known for Australia/New Zealand. Step 1 in the application of the probability-based
bridge assessment (PBBA) framework can infer this unknown level, and comparison with international
benchmarks can further inform on the implicit historical level of safety.
An annual target of βT = 4.2 is selected for ultimate bending, while an annual target of βT = 4.4 is selected for
ultimate shear as the international benchmark. The justification is discussed in Appendix B. Essentially, the
choice is made based on absence of other considerations, by being slightly conservative, and noting that
brittle structural elements expected to fail without warning (i.e. ultimate shear) should have higher target
reliability indices (JCSS, 2000) than ductile failures.
Ultimately, the setting of the target reliability index as part of the PBBA is dependent on the ‘risk-appetite’ of
the asset owner considering the relative cost of safety measures and consequences of failure for the mode
of failure of interest. This is further discussed in Appendix B.
For example, when significant control measures are put in place resulting in significant expenditures, but the
consequences of failure to society are relatively low, the annual target reliability indices may be around 3.1.
Conversely, if there are no control measures, and much doubt in the costs and consequences of failure, an
annual target reliability index as high as 5.0 may be selected.
5. Example Application
To explain how results for Step 1 and Step 2 are obtained, consider the following example as the T44 (1976)
designed I-girder bridge deck shown in Figure 5-1 provided from TMR. The example here describes the
procedure undertaken for ultimate shear of the first outer girder. The girder is taken to be simply supported
as per the provided grillage model. The set of results for the other bridge family sections, can be found in
Appendix J for Step 1 and Appendix K for Step 2.
Figure 5-1: Cross-section of T44 designed I-girder bridge deck: the outer girder is assessed for shear.
Using the values shown in Table 5-1 as characteristic values, the characteristic shear capacity of the section
is found to be 1166.3 kN following AS 5100.5 provisions for assessment. By definition, this does not include
the capacity reduction factor (ϕ). Notice, as shown in Appendix G, this requires a triple iteration process, and
not simply following the design provisions. The capacity analysis includes the 160 mm overlay slab and 50
mm in-situ concrete flange thickening shown in Figure 5-1.
Using the probabilistic distributions (Table 4-1) and correlations (Table 4-2), the Monte Carlo simulation is
conducted using N = 106 runs. The histogram for the shear capacity of this section and the realisations
plotted on probability paper (Gumbel) are shown in Figure 5-3. Details describing these plots are given in
Appendix A. The lower tail (50% of data) is fitted, as indicated in red. A Normal distribution was found to be a
suitable fit to this lower tail, with a bias and coefficient variation shown in Table 5-2. The resistances for other
sections are found similarly, and descried in Appendix G.
Table 5-1: Input parameters for shear model extracted from T44 designed I-girder.
Figure 5-2: Engineering drawings of outer girder for T44 designed I-girder bridge deck.
Table 5-2: Characteristic shear capacity (as per AS 5100.5 model) and associated probability distribution as
per Monte Carlo simulation (all sections in Appendix G).
Distribution
Example Section Characteristic Shear Capacity (kN)
Type Bias CoV
T44 designed I-girder 1166.27 Normal 1.09 0.07
Figure 5-3: Histogram and Gumbel probability paper plot of shear capacity realisations for T44 designed I-girder
(all sections in Appendix G).
Note: Fitting is done on the lower tail (red shading) and the distribution is indicated by the red line.
In Step 1, the level of safety for the subset of the ‘as-built’ bridges families presented in Table 3-1 under their
design loads is obtained.
A grillage model for the bridge deck was provided by TMR. Using this model, the maximum characteristic
shear force from the four different loading components at the critical section being assessed for shear
(distance of do for the outer girder) was extracted from the grillage model and summarised in Table 5-3. This
includes the load effect due to the traffic load model (T44) as per the requirements of Step 1. For the other
bridge elements’ critical sections see Appendix H. By definition, characteristic values do not apply the
loading safety factors, γ. However, in Step 1 traffic loads do include dynamic allowance and multiple
presence factors stipulated in the design code (NAASRA). Also shown in Table 5-3 are the corresponding
probabilistic distributions, recalled from Section 4.3.1.
Table 5-3: Characteristic shear force (as per TMR grillage model) and associated probability distribution.
Traffic load includes dynamic allowance and multiple vehicle presence factors (all sections in
Appendix H).
Characteristic Distribution
Example Section Load Component
Shear Force (kN) Type Bias CoV
Self-weight 165.54 Normal 1.03 0.08
Slab load 60.61 Normal 1.05 0.10
T44 designed I-girder
Superimposed dead load 23.65 Normal 1.00 0.25
Traffic load (T44) 264.02 Gumbel 0.5 0.18
In Step 2, the probability-based bridge assessment framework is implemented to quantify the level of safety
for the subset of the ‘as-built’ bridges families presented in Table 3-1 under current ‘as-of-right’ freight
vehicles access regimes. The permanent loading and their probability distribution are the same as that of
Step 1.
To construct the influence lines necessary to convert loads to load effect, the grillage model provided from
TMR is used and a 1 kN 2.3 m wide bogey is set to travel along the bridge deck on one of the two traffic
lanes. The shear force at the critical location of the outer girder (distance of do) is then extracted as the
bogey traverses. The process is repeated for the second traffic lane as defined by the provided grillage
model. The resulting influence lines are shown in Figure 5-4. Notice, when the bogey is in Lane 2, the shear
force at the outer girder is not as significant compared to when the bogey is in Lane 1 due to the transverse
stiffness of the deck. For the other sections, the most critical structural element is considered. Influence lines
for both bending moment and shear force for the other bridges are found in Appendix H.
Figure 5-4: Influence lines (kN shear/kN axle) for shear force at the critical location on the outer girder of TMR
T44 designed I-girder bridge deck for vehicles in two different traffic lanes (Lane 1 and Lane 2).
The in-house software BTLS simulates the traffic streams for the six different flows as previously discussed
(and detailed in Appendix D). Bridge load effects are calculated using the influence lines shown in Figure 5-4
from the simulated vehicles from the “garage” of 1 million trucks. The annual maximums are extracted (Block
Maximum method with =1 year) for 100 years (25,000 economic days) and fitted to a Generalized Extreme
Value (GEV) distribution. More details on GEV are found in Appendix A. Figure 5-5 illustrates the fitting of
these realisations for all six traffic flows.
In plots shown in Figure 5-5, the data is expressed as the ratio between the static load effect due to the
simulated traffic and characteristic load effect due to the traffic load model (T44 in this case). In this way, the
relative difference from the design load effect and current ‘as-of-right’ freight load effect is depicted. The
statistical parameters for the fitted GEV distribution are shown in Table 5-4 for the six different flows. Based
on the location parameters (a measure of central tendency, similar to the mean), it appears that shear load
effects from T44 are 50% lower than the load effects for current traffic for this example at the six flows
considered. Parameters and plotted fits for other bridges for all six traffic flows are found in Appendix H.
Figure 5-5: Histogram and Gumbel probability paper plot of maximum annual shear force realisations for T44
designed I-girder under various traffic flows (all sections in Appendix H).
Data represented as ratio of simulated load effect with design traffic load model load effect.
Data represented as ratio of simulated load effect with design traffic load model load effect.
Table 5-4: Fitted probability distribution (uncertainty) for annual maximum shear force due to different higher
mass limit (HML) network at different flows for example girder (all sections in Appendix H).
where
gs = load factor for superimposed dead load (found in AS 5100.7 – 2.0 adopted)
Q = traffic load factor (found in AS 5100.7 Table 12.2(B) dependent on vehicle - Table 5-5)
characteristic traffic load including the dynamic allowance and multiple vehicle presence
∗
S𝐿𝐿 =
(using AS 5100)
Thus, for this example, using the presented characteristic values, and associated safety factors from AS
5100.7, the rating factor for T44 is:
Since RF < 1, then the section is deemed unsafe. As per Figure 3-1, the higher-order assessment is
necessary (if deemed beneficial).
where
self-weight load effect of bridge deck component (beam; girder; plank; etc.)
Si =
(probabilistic)
Using the software FERUM and UQLab, first order reliability method (FORM) is used to quantify the
structural reliability index shown in Equation 9 and detailed in Appendix A.
where
β = reliability index
pf = probability of failure
For Step 1 in this example application, R is seen in Table 5-2 and the components of S are seen in Table 5-
3. The model error for loading (ωS) is taken as per the recommendations of JCSS for all sections (JCSS,
2000). For the capacity model error (ωR), the procedure detailed by Melhem et al. (2018), is followed. Both
are summarised in Appendix I. All variables in the limit state function shown in Equation 8 are considered to
be independent.
The FORM results for this example are shown in Table 5-6 and Appendix J for all sections. A detailed
procedure in how FORM calculates the structural reliability index can be found in text books (Nowak and
Collins, 2012). Second order reliability method (SORM) on UQLab is also performed on the limit state
function to assess any nonlinearity in the limit state function and results shown in Table 5-6 for this example
as well as Appendix J for all sections. The results are consistent. Table 5-6 also presents the suggested
target reliability. The calculated structural reliability index is larger than the annual target reliability index, thus
the section code-implied safety is deemed acceptable for ultimate shear following the probabilistic approach.
Table 5-6: Step 1 ULS shear structural reliability results for T44 designed I-girder (all sections in Appendix J).
For Step 2 in this example, R is seen in Table 5-2. The permanent loading uses the characteristic values and
the probabilistic distribution in Table 5-3. As for the traffic loading, the GEV parameters presented in
Table 5-4 are used for each of the six flows. The same model errors are adopted (Appendix I) and DLA is
now taken as a random variable as described in Section 4.4.2. Again, all variables in the limit state function
shown in Equation 8 are considered to be independent.
The FORM results (using UQLab) for this example are shown in Table 5-7 for the six different traffic flows.
Summarised results for other sections are seen in Appendix K. Table 5-7 also presents one possible target
reliability index. The calculated structural reliability index is lower than the annual target reliability index, thus
the existing safety of the girder for ultimate shear may be considered unacceptable following the probabilistic
approach for the types of traffic flows considered in the analysis. The results are reasonably consistent
across the different flow type. The most critical traffic flow (lowest annual reliability index) is High-Uni
(β = 3.84). It’s important to note that the target reliability index can be altered in accordance with AS 5104 to
reflect social and economic considerations as outlined in Section 4.7.
Table 5-7: Step 2 ULS shear structural reliability results for T44 designed I-girder (all sections in Appendix K).
Table 5-8 presents the importance coefficients from the Step 1 results for the example section. For the other
sections, refer to Appendix J. From these results, the annual probability distribution for maximum annual
traffic load effect is generally what most influences the β values shown in Table 5-6. This is followed by the
loading model error, and then the capacity probability distribution. Recall, DLA is not considered a random
variable in Step 1. Rather, DLA is the code stipulated deterministic value.
The high importance coefficients for these random variables indicate that more rigorous data is necessary for
when establishing their probabilistic distribution for confidence in the results. For traffic loading, the methods
adopted for Step 2 are one example of more rigorous data. Moreover, since these are the most sensitive
variables, any assumptions made when establishing the probability distribution must be validated for
robustness in the final results.
Table 5-8: Step 1 ULS shear relative importance (sensitivity analysis) for T44 designed I-girder (all sections in
Appendix J).
Table 5-9 presents the importance coefficients from the Step 2 results for the example section. For the other
sections, refer to Appendix K. The results are generally consistent across the six flows. The additional
complexity in the analysis has now reduced the influence of traffic loading, with the loading model error now
influencing the β values most, as shown in Table 5-7, followed by the DLA. This suggests that with further
research into the actual loading on the bridge (most notably DLA), thereby facilitating a reduced loading
model uncertainty estimate, the measured safety of the section could be increased.
Table 5-9: Step 2 ULS shear relative importance (sensitivity analysis) for T44 designed I-girder for all flows (all
sections in Appendix K).
5.6 Conclusions
The example application of the PBBA framework to a specific bridge component (girder shear) aims to
demonstrate the holistic nature of probability-based structural safety assessment. Specifically, for the
considered bridge, it is found that:
• In Step 1, the as-designed girder has significant safety reserves, β = 6.25 > 4.4. In this analysis, the most
important parameters were the live load, the resistance, and the loading model error.
• In Step 2, for a range of traffic flows, the in-service reliability is β = 3.84 to 4.04. While these values are
less than the nominal target reliability index of 4.4, specific social and economic factors may allow a
target reliability index lower than this, in accordance with AS 5104. Finally, in this analysis, the most
important parameters are the loading model error, DLA, the resistance, and the resistance model error.
With further refinement of these variables through site inspection, material samples, and loading
measurements, the model errors could be reduced, revealing a further reserve of safety.
Figure 6-1 and Figure 6-2 plot the annual structural reliability indices for all the bridge families listed in
Table 3-1 as per Step 1 (design load as the traffic load applied to the bridge deck) for ultimate bending and
ultimate shear, respectively. For values refer to Appendix J.
Figure 6-1: Safety of ‘as-built’ bridge families as inferred to historical code for ultimate bending.
Figure 6-2: Safety of ‘as-built’ bridge families as inferred to historical code for ultimate shear.
FORM results from both FERUM and UQLab software are plotted and are identical for all sections in both
ultimate bending and ultimate shear, thus validating the results. SORM results (shown in Appendix J) are
also reasonably consistent, indicating that the linearity assumption of FORM is reasonable.
For both bending and shear, the selected sections surpass the target reliability index, with some sections
significantly surpassing the international benchmark level of safety.
Figure 6-3 plots the deterministic rating factor (using the characteristic values and AS 5100.7 safety factors ϕ
and γ) against the probabilistic annual structural reliability index (using probability distributions) for bridge
families listed in Table 3-1 as per Step 1 (design load model as the traffic load applied) for both ultimate
bending and ultimate shear.
Figure 6-3: Relationship between deterministic and probabilistic measures of safety for ‘as-built’ bridge families
as inferred from the historical code for (a) ultimate bending and (b) ultimate shear.
(a) (b)
Notice for some sections RF < 1, but β > βT. Hence, for bending and shear, 100% of bridges that failed a
deterministic assessment (RF < 1) have a reliability index higher than the acceptable level of risk. This
illustrates PBBA can reveal margins of safety otherwise not indicated by the conventional AS 5100
deterministic method.
Figure 6-4 and Figure 6-5 depict the relative importance (2 from FORM) of the random variables in the limit
state function (Equation 8) on the annual structural reliability indices results for all the bridge families listed in
Table 3-1 as per Step 1 (design load as the traffic load applied to the bridge deck) for ultimate bending and
ultimate shear, respectively. For values refer to Appendix J.
These results indicate the probabilistic distribution for maximum annual traffic load effect, is generally the
most influential variable in the reliability result, followed by the model error for loading, and then the capacity.
Exceptions to this include the T44 designed Super-T in shear, MS18 design U-slabs in bending, and 35 ft
PSC plank in bending, where the capacity is more influential on the results.
These coefficients give important information on where further research and site measurements would be
most beneficial in reducing the uncertainties in the model error and random variable descriptions. For
example, the widespread use of WIM data collection would allow far more certain load models and thus have
a large impact on the determined levels of safety.
Figure 6-4: Relative importance for each random variable that comprise limit state function in the reliability
index results (Step 1) for ultimate bending.
Figure 6-5: Relative importance for each random variable that comprise limit state function in the reliability
index results (Step 1) for ultimate shear.
6.1.4 Discussion
Traffic loading safety factors in the AS 5100 load rating equation are selected from engineering judgement,
rather than through structural reliability calibration (Heywood et al., 2000). As a result, the benchmark safety
level being compared against in current deterministic bridge assessments was heretofore unknown. Since
Step 1 indicates the level of safety achieved by the historical design codes, these results infer a benchmark
safety level.
The high reliability indices suggest all designs are significantly safe against their contemporaneous load
model, and historical design codes provided considerable capacity reserve. This explains why some bridges
can accommodate access of present-day heavy vehicles greater than the traffic load models used for their
design, particularly when in some cases a dogmatic application of the rating factor suggests otherwise
(sometimes referred to as the “plausibility gap”).
Results are generally consistent across the eras considered, ranging from β = 6 to 9 suggesting new
generations of design codes (NAASRA, ABDC, AS 5100) were drafted to ensure backwards compatibility in
the designs, rather than calibrated to an explicit safety target.
It is observed that for ultimate shear, some sections have a rating factor less than 1, and are thus deemed
unsafe as per deterministic approach. The majority of these sections are the SM1600 designed Super-T
girders. Recall, unlike the other sections which are existing ‘as-built’ bridges, the Super-T girders have been
designed by the project team to the minimum requirements of AS 5100 2004, reflecting the many bridges
built in the period 2004-17. It would be expected that these rating factors should fall close to 1. However, the
same girders are then assessed to AS 5100 2017. The shear resistance model used in the two codes is
different, with 2017 adopting MCFT-based shear provisions compared to the empirical based 2004 shear
provisions. The 2004 provisions have been shown to significantly over-estimate the real shear capacity
(Collins & Michaels, 2015) and although MCFT is a more accurate theory, the capacity reduction factor
remains unaltered. Since the deterministic approach stipulates the calculation of the characteristic resistance
using these design provisions, then the characteristic capacity is conservative, and the rating factors result in
being less than one for the Super-Ts. Meanwhile, in the probabilistic approach, an appropriate model error is
considered, allowing the true level of safety to be determined.
The sensitivity results indicate further investigations are necessary on the assumptions made for the
maximum annual traffic load effect probability distribution. This is achieved through Step 2.
Figure 6-6 and Figure 6-7 plot the highest and lowest annual structural reliability indices for all the bridge
families listed in Table 3-1 as per Step 2 (simulated HML traffic as the traffic load applied to the bridge deck
at six different flows) for ultimate bending and ultimate shear, respectively. For values refer to Appendix K.
Only the traffic flows that provide the greatest and lowest safety are presented, indicated by the text. The
safety of other traffic flows then falls between this range for each individual bridge. Plots for the other flows
for each section are found in Appendix K.
Almost all MS18 designed bridges in bending do not meet the specified target reliability index. The level of
safety tends to increase with greater design loading for bending. These significant trends are also observed
for shear.
Figure 6-6: Range in safety of ‘as-built’ bridge families under different HML traffic flows for ultimate bending.
Figure 6-7: Range in safety of ‘as-built’ bridge families under different HML traffic flows for ultimate shear.
Figure 6-8 plots the deterministic rating factor (using the characteristic values, the design load model, and
AS 5100.7 safety factors ϕ and γ) against the probabilistic annual structural reliability index (using probability
distributions) for bridge families listed in Table 3-1 as per Step 2 (simulated HML traffic as the traffic load
applied to the bridge deck at six different flows) for both ultimate bending and ultimate shear. For each
section, all six flows are plotted.
Figure 6-8: Relationship between deterministic and probabilistic measures of safety for ‘as-built’ bridge families
to selected HML flows for (a) ultimate bending and (b) ultimate shear.
(a) (b)
A distinct separation between the different traffic load models (MS18, T44, SM1600) is observed for bending.
Notice for sections where design loads are lower than present day traffic, RF > 1 (under design load) and
β < βT (under traffic loads). For shear, the results are more scattered. However, for ultimate shear 75% of
bridges that failed a deterministic assessment (RF < 1) are shown to have a reliability index higher than the
acceptable level of risk – for these bridges a margin of safety is still available.
Figure 6-9 and Figure 6-10 depict the relative importance (2 from FORM) of the random variables in the limit
state function (Equation 8) on the annual structural reliability indices results for all the bridge families listed in
Table 3-1 as per Step 2 (simulated HML traffic as the traffic load applied to the bridge) for ultimate bending
and ultimate shear, respectively. Results are generally consistent across the six different flows (see
Appendix K). Thus, only High-Bi flows results are presented here.
Figure 6-9: Relative importance for each random variable that comprise the limit state function in the reliability
index results (Step 2) for ultimate bending for High Bi (others shown in Appendix K).
Figure 6-10:Relative importance for each random variable that comprise the limit state function in the reliability
index results (Step 2) for ultimate shear for High Bi (others shown in Appendix K).
The influence of the traffic load on the level of safety has now decreased from Step 1, with the model error
being the most significant variable. This means that further analysis of loading would be valuable, the
increased certainty of which would allow a reduction in the model error, and hence potentially revealing a
further safety margin.
6.2.4 Discussion
The increase in the safety as the design loading increases from MS18 to SM1600 is expected. The lack of
significant trends observed for shear suggests that the notional models may not represent shear forces in the
bridge decks adequately.
While MS18 bridges at the flows considered may have not passed the suggested benchmarks, the setting of
the acceptable level of safety is at the discretion of the asset owner. Hence, the acceptability of risk is not
ruled out and these bridges can still be considered as safe depending on socio-economic factors.
The sensitivity analysis indicates further analysis is necessary to reduce the uncertainty in DLA, in the
capacity, and the model error for loading. For loading model error, further research could facilitate a
reduction in the epistemic uncertainty in the model error variable. Meanwhile for capacity, further statistical
analysis on the input parameters and model error would also help reduce uncertainty.
7. Conclusions
In Step 1, the traffic loads applied to bridges were the design traffic load model of the bridge’s era. This
indicates the level of safety implied by the historical codes. The results show the annual structural reliability
indices are generally larger than stipulated target reliability indices that define the acceptability of safety/risk.
When compared to the deterministic method results, the probabilistic method reveals a margin of safety
available that the otherwise conservative deterministic approach does not illuminate, most noticeably for
shear.
In Step 2, real traffic streams from a HML network are simulated on the bridge decks to infer the current level
of safety. It is observed the level of safety decreased with lower design traffic load models, especially for
bending. In ultimate bending, if the rating factor was less than one the reliability indices tended to be lower
than acceptable levels. However, for ultimate shear, 75% of bridges that failed a deterministic assessment
(RF < 1) were shown to have a reliability index higher than the acceptable level of risk – for these bridges a
margin of safety is still available.
7.2 Limitations
The work presented here demonstrates how the mature field of structural reliability theory can be adapted to
the Australasian context for improved bridge safety and cost-optimal decisions. Necessarily, the results
presented here are limited to a subset of Australian bridges. Specific limitations are:
• It is considered throughout that maintenance and inspection regimes are such that bridge performance
can be assumed to be as per the pristine as-built condition.
• Only simply-supported superstructure components under traffic loading for ultimate bending and shear
have been examined.
• Only two-lane non-skewed bridges are considered, for bi- and uni-directional traffic flows.
• Only dead, superimposed dead, and traffic live loads, including dynamic actions, are considered.
• Only HML as-of-right traffic flows are used, based on the weight and composition data from one WIM site
in Victoria.
• Time-dependent factors such as traffic growth and material degradation are neglected.
• Specific choices are made in the implementation of structural reliability theory, which are detailed
throughout the work, and these may not be the appropriate choices in all cases.
It is important to note that the limitations noted here relate only to the demonstration of the PBBA framework
and are not limitations of the methodology itself. For example, for longer span and landmark structures, the
PBBA framework still applies, and can be supplemented with bridge- and component-specific material test
data. Indeed, globally, structural reliability is the holistic framework in which all structural engineering safety
decision-making takes place.
References
AASHTO 2018, Manual for bridge evaluation, 3rd ed. Washington, DC, USA: American Association of State
Highway and Transportation Officals.
ABCB 2015, Structural Reliability: Handbook. Canberra, ACT: Australian Building Codes Board.
Akgul F & Frangopol DM 2004, 'Time-dependent interaction between load rating and reliability of
deteriorating bridges', Engineering Structures, 26, 1751-1765.
Ang AH-S and Tang HW 1984, Probability Concepts in Engineering Planning and Design, John Wiley &
Sons Inc.
Austroads 2014, Review of AS 5100.7 Rating of Existing Bridges and the Bridge Assessment Group
Guidelines, Austroads, Sydney, NSW.
Austroads 2018a, Guide to Bridge Technology Part 1: Introduction and Bridge Performance, Austroads,
Sydney, NSW.
Austroads 2018b, Higher Order Bridge Assessment in Australia, Austroads, Sydney, NSW.
Austroads 2018c, Implementation of a nationally consistent framework for the assessment of bridges in
Australasia, Austroads, Sydney, NSW.
Bentz EC, Vecchio FJ & Collins MP 2006, 'Simplified modified compression field theory for calculating shear
strength of reinforced concrete elements', ACI Materials Journal, 103, 614.
Brameld G, Vecchio M & Muhammed F, 'The Geometric Variability of Prestressed Concrete Bridge Beams',
Australian Road Research Board Proceedings, 1986.
Bureau of Infrastructure, Transport and Regional Economics (BITRE) 2014, Freightline 1 - Australian freight
transport overview, Canberra: Australian Government.
Caprani C, 'Dynamic load allowance: a synthesis of the state of the art', Austroads Bridge Conference, 10th,
2017, Melbourne, Victoria, Australia.
Caprani CC & Melhem MM 2019, 'On the use of MCT per AS 5100.5 for the assessment of shear capacities
of existing structures', Australian Journal of Structural Engineering, Under publication.
Caprani CC 2013, 'Lifetime highway bridge traffic load effect from a combination of traffic states allowing for
dynamic amplification', Journal of Bridge Engineering, 18, 901-909.
Caprani CC, Gonzlaez A, Rattigan PH & O'Brien EJ 2012, 'Assessment dynamic ratio for traffic loading on
highway bridges', Structure and Infrastructure Engineering, 8, 295-304.
Caprani CC, Melhem MM & Siamphukdee K 2017, 'Reliability analysis of a Super-T prestressed concrete
girder at serviceability limit state to AS 5100: 2017', Australian Journal of Structural Engineering, 18, 60-
72.
Caprani CC, O'Brien EJ & Lipari A 2016, 'Long-span bridge traffic loading based on multi-lane traffic micro-
simulation', Engineering Structures, 115, 207-219.
Caprani CC, O'Brien EJ & McLachlan GJ 2008, 'Characteristic traffic load effects from a mixture of loading
events on short to medium span bridges', Structural safety, 30, 394-404.
Collins MP & Mitchell D 2014, 'Shear design and evaluation of concrete structures', Concrete in Australia.
COST, Action 345 Procedure required for the assessment of highway structures - Final report, Europe, The
European Cooperation in Science and Technology (COST).
Cremona C 2012, Structural performance: Probability-based assessment, John Wiley & Sons.
Crespo-Minguillon C & Casas JR 1997, 'A comprehensive traffic load model for bridge safety checking',
Structural Safety, 19, 339-359.
Crespo-Minguillon C & Casas JR 1998, 'Fatigue reliability analysis of prestressed concrete bridges', Journal
of Structural Engineering, 124, 1458-1466.
Danish Road Directorate 2004, Reliability-Based Classification of The Load Carrying Capacity of Existing
Bridges, Danish Road Directorate (DRD), Denmark.
Der Kiureghian A & Ditlevsen O 2009, 'Aleatory or epistemic? Does it matter?' Structural Safety, 31, 105-
112.
Eamon CD, Kamjoo V & Shinki K 2016, 'Design live-load factor calibration for Michigan highway bridges',
Journal of Bridge Engineering, 21, 04016014.
Ellingwood B 1980, Development of a probability based load criterion for American National Standard A58:
Building code requirements for minimum design loads in buildings and other structures, US Department
of Commerce, National Bureau of Standards.
Enevoldsen I 2011, 'Practical implementation of probability based assessment methods for bridges',
Structure and Infrastructure Engineering, 7, 535-549.
Enright B 2010, Simulation of traffic loading on highway bridges, PhD, University College Dublin.
Enright B & O'Brien EJ 2013, 'Monte Carlo simulation of extreme traffic loading on short and medium span
bridges', Structure and Infrastructure Engineering, 9, 1267-1282.
Estes AC & Frangopol DM 2005, 'Load rating versus reliability analysis', Journal of Structural Engineering,
131, 843-847.
Faber MH & Sorensen JD 2002, Reliability based code calibration, The Joint Committee on Structural Safety,
Zurich, Switzerland.
Fenwick J 1985, Design loads in the NAASRA Bridge Code, Institution of Engineers (Australia) Civ Eng
Trans.
FIB CEB-FIP 2018, Safety and performance concept. Reliability assessment of concrete structures: Guide to
good practice, The International Federation for Structural Concrete fib.
Forbes C, Evans M, Hastings N & Peacock B 2011, Statistical distributions, John Wiley & Sons.
Foster SJ, Stewart MG, Loo M, Ahammed M & Sirivivatnanon V 2016, 'Calibration of Australian Standard
AS 3600 Concrete Structures: part I statistical analysis of material properties and model error', Australian
Journal of Structural Engineering, 17, 242-253.
Fwa T & Li S 1995, 'Estimation of lane distribution of truck traffic for pavement design', Journal of
transportation engineering, 121, 241-248.
Geistefeldt J & Hohmann S 2013, 'Assessment of Design Hours for Freeways', Transportation Research
Record, 2395, 49-56.
Hajializadeh D, Stewart MG, Enright B & O'Brien 2016, 'Spatial time-dependent reliability analysis of
reinforced concrete slab bridges subject to realistic traffic loading', Structure and Infrastructure
Engineering, 12, 1137-1152.
Haldar A & Mahadevan S 2000, Reliability assessment using stochastic finite element analysis, John Wiley &
Sons.
Heywood R, Gordon R & Boully G 2000, 'Australia’s Bridge Design Load Model: Planning for an Efficient
Road Transport Industry', Transportation research record, 1696, 1-7.
Holicky M, Retief JV & Sykora M 2016, 'Assessment of model uncertainties for structural resistance',
Probabilistic Engineering Mechanics, 45, 188-197.
Lawrence SJ & Stewart MG 2009, 'Reliability-based calibration of the capacity reduction factor for design of
masonry in compression to AS3700', Australian Journal of Structural Engineering, 9, 97-110.
Lay MG 1980, Load combinations-a calibration oriented approach to more rational design rules.
Leicester R, Pham L & Kleeman P 1985, Use of reliability concepts in the conversion of codes to limit states
design, Institution of Engineers (Australia) Civ Eng Trans.
Lemaire M, Chateaunerf A & Mitteau J-C 2009, Structural reliability, Wiley Online Library.
Lipari A, Caprani CC & O'Brien EJ 2017, 'A methodology for calculating congested traffic characteristic
loading on long-span bridges using site-specific data', Computers & Structures, 190, 1-12.
Marsh JG 1985, Some aspects of the review of the National Assoication of Australian State Road Authority
Bridge Design Code (NAASRA BDC), Institution of Engineers (Australia) Civ Eng Trans.
Mathieu H 1991, 'Reliability problems associated with prestress', Reliability of concrete structures, 99-163.
Mayer M 1926, Die Sicherheit der Bauwerke und ihre Berechnung nach Grenzkräften anstatt nach
zulässigen Spannungen, J. Springer.
Melchers RE & Beck AT 2018, Structural reliability analysis and prediction, John Wiley & Sons.
Melhem MM, Caprani C & STEWART MG 2018, 'Model Error For Australian Code Shear Capacity Of
Concrete Structures', in Wang CM, Ho JCM & Kitipornchai S (eds.) 25th Australasian Conference on
Mechanics of Structures and Materials (ACMSM25). Brisbane, Australia.
Munter S & Lume E 2017, Guide to Historical Reinforcement, Steel Reinforcement Institute of Australia
(SRIA).
Myers RH, Montgomery DC & Anderson-Cook, CM 2016, Response surface methodology: process and
product optimization using designed experiments, John Wiley & Sons.
NAASRA 1976, NAASRA Bridge Design Code, National Association of Australian State Road Authorities
(NAASRA).
Ng A, Rooke A & Page T 2017, 'Austroads Project TP1952 – Higher order assessments for existing bridges',
Austroads Bridge Conference 2017, Melbourne, Australia: Austroads.
Nowak AS 1993, 'Live load model for highway bridges', Structural safety, 13, 53-66.
Nowak AS, Yamani AS & Tabsh SW 1994, 'Probabilistic models for resistance of concrete bridge girders',
Structural Journal, 91, 269-276.
OBrien EJ & Caprani CC 2005, 'Headway modelling for traffic load assessment of short to medium span
bridges', The Structural Engineer, 83, 33-36.
OBrien EJ, Schmidt F, Hajializadeh D, Zhou XY, Enright B, Caprani CC, Wilson S & Sheils E 2015, 'A review
of probabilistic methods of assessment of load effects in bridges', Structural safety, 53, 44-56.
O'Connor A, Pedersen C, Gustavsson L & Enevoldsen I 2009, 'Probability-based assessment and optimised
maintenance management of a large riveted truss railway bridge,' Structural Engineering International,
19, 375-382.
O'Connor C & Shaw P 2000, Bridge loads: An international perspective, CRC Press.
O’Connor A. & Enevoldsen I 2008, 'Probability based modelling and assessment of an existing post-
tensioned concrete slab bridge', Engineering Structures, 30, 1408-1416.
Pham L 1985, Reliability analyses of reinforced concrete and composite column sections under concentric
loads, Institution of Engineers (Australia) Civ Eng Trans.
Pham L 1985b, 'Load Combinations and Probabilistic Load Models for Limit States Codes', Civil Engineering
Transactions, IEAust, CE 27 (1): 62–67.
Rakoczy AM & Nowak AS 2013, 'Reliability-based sensitivity analysis for prestressed concrete girder
bridges', PCI journal, 58.
Rapattoni F, Wells E & Gran P 1994, 'New developments in bridge superstructures', Bridges-Essential To
Our Economy; Proceedings Of The Austroads Bridges Conference, February 16-18, 1994, Melbourne,
Australia.
Rapattoni F, & Wells J 1991, 'New bridge superstructure using PSC T-slabs', Bridges: Part of the Transport
System. Proceedings of the Austroads Bridges Conference, Queensland University of Technology,
Queensland Department of Transport, and Australian Road Research Board, 1991.
Reid S 2004, 'Reliability-Based Load-Rating of Existing Bridges', Austroads Bridge Conference, 5th, 2004,
Hobart, Tasmania, Australia.
Ribeiro A, Calixto J & Diniz S 2016, 'Assessment of epistemic uncertainties in the shear strength of slender
reinforced concrete beams', Engineering Structures, 116, 140-147.
Roscoe K, Diermanse F & Vrouwenvelder T 2015, 'System reliability with correlated components: Accuracy
of the Equivalent Planes method', Structural Safety, 57, 53-64.
Rubinstein RY & Kroese DP 2016, Simulation and the Monte Carlo method, John Wiley & Sons.
Scott DW 2015, Multivariate density estimation: theory, practice, and visualization, John Wiley & Sons.
Shah PM, Stewart MG & Fok H 2009, 'Reliability assessment of a typical steel truss bridge', ustroads Bridge
Conference, 7th, 2009, Auckland, New Zealand.
Stewart M & Melchers RE 1997, Probabilistic risk assessment of engineering systems, Springer.
Stewart MG, Foster S, Ahammed M & Sirivivatnanon V 2016, 'Calibration of Australian Standard AS 3600
concrete structures part II: reliability indices and changes to capacity reduction factors', Australian Journal
of Structural Engineering, 17, 254-266.
Stewart MG & Mueller J 2014, 'Terrorism risks for bridges in a multi-hazard environment', International
Journal of Protective Structures, 5, 275-289.
Stewart MG & Suo Q 2009, 'Extent of spatially variable corrosion damage as an indicator of strength and
time-dependent reliability of RC beams', Engineering Structures, 31, 198-207.
Sykora M, Diamantidis D, Holicky M & Jung K 2017, 'Target reliability for existing structures considering
economic and societal aspects', Structure and Infrastructure Engineering, 13, 181-194.
Tu 1406 2019, 'Quality specifications for roadway bridges, standardization at a European level (BridgeSpec)',
The European Cooperation in Science and Technology (COST).
Vrouwenvelder A 2002, Reliability based code calibration. The use of the JCSS Probabilistic Model Code.
Waldin J, Mchaffie B & Wong V, 'How strong is your bridge? Optimising New Zealand's longest road bridge',
in Powers N, Frangopol D, Al-Mahaidi R & Caprani C (eds) International Conference of Bridge
Maintenance, Safety and Management, 2018 Melbourne, Australia, Taylor & Francis Group, 2643 - 2650.
Wang C, Zhang H & Li Q 2017, 'Time-dependent reliability assessment of aging series systems subjected to
non-stationary loads', Structure and Infrastructure Engineering, 13, 1513-1522.
Wisniewski D, Casas JR, Henriques AA & Cruz PJ 2009, 'Probability-Based Assessment of Existing
Concrete Bridges—Stochastic Resistance Models and Applications', Structural Engineering International,
19, 203-210.
Wisniewski DF, Cruz PJ, Henriques AAR & Simoes RA 2012, 'Probabilistic models for mechanical properties
of concrete, reinforcing steel and pre-stressing steel', Structure and Infrastructure Engineering, 8, 111-
123.
Standards
BD 79/13-2013. Design manual for roads and bridges: the management of sub-standard highway structures:
volume 3: section 4: assessment.
g=R-S A-1
where
Note, Equation A-1 can be expressed for each different modes of failure in consideration for a structural
member (e.g. bending moment, shear force, stresses, deflection, and so forth).
Distribution functions can be described by probability density functions (PDF) or cumulative distribution
functions (CDF).
Probability density functions inform on the number of data observations (also known as realisations) of a
given value x (known as the variate) out of the total number of realisations. That is, the PDF informs on the
probability of observing the given variate x, or a normalisation of the relative frequency of the realisations.
A cumulative distribution function (CDF) is simply the integral of the PDF, informing on the probability of
having less than the variate x.
Distribution Types
One of the most commonly used distributions in engineering problems is the Normal distribution, well
recognised by its symmetrical ‘bell-shaped’ curve – Figure 2-2. However, there are many different
probabilistic distributions types that may better represent the data uncertainty (Forbes et al., 2011):
• Log-normal distributions are commonly used for variables that cannot be negative, for example
compressive strength of concrete (Foster et al., 2016).
• Generalized Extreme Value (GEV) distributions are good for depicting the uncertainty of the maximum
values, for example, the annual load effects from traffic loading (OBrien et al., 2015). These are further
divided into Gumbel, Fréchet, and Weibull distributions.
Statistical Parameters
Probability distribution functions are described through statistical parameters depending on the function type.
Typical parameters are the mean (a measure of central tendency) and standard deviation (a measure of data
spread). In structural reliability, it is common to express these probability distributions through a bias and
coefficient of variation where:
• The bias is the ratio of mean divided by the characteristic value adopted in design.
• The coefficient of variation is the ratio of standard deviation and the mean.
Depending on the function type, alternative parameters specific to a distribution type may be used. For
example, GEV distributions are described using shape (), scale (), and location (u) parameters. When the
shape parameter is zero, the GEV distribution is known as a Gumbel distribution.
Distribution fitting (more commonly known as statistical inference) aims to deduce the most appropriate
distribution type and statistical parameter for a set of realisations. Ideally, the fitting is done for the entire set
of realisations. Otherwise, at least the lower tail of resistance variables and upper for tail for loading variables
should be fitted, as this corresponds to the failure region (Figure 2-2). Examples of statistical inference
methods include maximum likelihood estimation or curve fitting. Details can be found in relevant textbooks.
Visualisation
The set of realisations may be visualised on a histogram or probability paper (Figure A-1).
Histograms are bar charts where each bar presents the relative frequency of the observed realisations in a
predefined interval (also called bins). These can be normalised and the inferred probability distribution
overlayed to visualise the goodness-of-fit. This is depicted on the left of Figure A-1.
Probability paper plots the relative cumulative frequency of the observed realisation (probability of observing
less than the given value). However, the vertical axis is scaled depending on probability paper type. On
Gumbel paper (as shown on the right of Figure A-1), the probabilities are scaled using the standard extreme
variate (s) which is shown in Equation A-2.
Similar to histograms, plots of the set of realisations on Gumbel probability paper can be overlayed with the
inferred probability distribution for visualising the goodness-of-fit (Figure A-1.)
Gumbel probability paper is commonly used in structural reliability theory instead of Normal probability paper
as it extends the tails of the realisations and inferred probability distribution. As shown in Figure 2-2, it is the
tails that are of interest for quantifying the probability of failure.
A.2.3 Correlations
In statistics, correlations describe how a pair of random variables are dependent on each other. Typically,
Pearson’s linear correlation coefficient is used to describe the linear inter-relationships. These correlations
must be known in the structural reliability analysis. If there is no correlation between the two variables, then
they are deemed statistically independent.
For transient loads (such as traffic loading) not only are the individual realisations of interest, but also their
chronological sequence. This is referred to as a stochastic variable. For example, a new traffic load effect
realisation is observed at each traffic loading event on the bridge. Hence, traffic load effect is a stochastic
variable.
The uncertainty in stochastic variables such as traffic load is established through various methods (OBrien et
al., 2015). The most common approach is for the realisations to be the maximum load effect within a
selected reference period () from a data set collected over some time (the measurement length, T). This is
because bridges are designed for maximum loads, rather than average point-in-time loads. This approach is
known as the Block Maximum method with the selected reference period () known as the block size. The
block size must be large enough so at least one cyclic or periodic behaviour is observed. Hence, = 1 day or
1 year is typical. Meanwhile, the measurement length must be sufficiently long to provide realisations so to
establish the probability distribution of the daily/annual maximum traffic loads.
Extrapolation
After establishing the daily or annual maximum traffic load probability distribution it may be of interest to
establish annual or design-life probability distribution to quantify the annual or design-life reliability indices.
Assuming loading events are independent and probability distributions are identically distributed (known as
stationarity), then the annual maximum load effect probability distribution is simply the product of each daily
probability distribution for the number of days in the year.
A common assumption has been to represent ‘economic days’ in a year (Caprani et al., 2008). Around 250
such days per year is normally reasonable (5 days for 50 working weeks in a year). The design-life load
effect probability distribution is the product of each annual probability distribution for the number of years in
the design-life (100 years for AS 5100).
The assumption of stationarity however does not properly reflect reality since bridge traffic volumes typically
increase with time. The annual maximum load effect probability distribution for each year is likely to be
different, and not identically distributed, to the following year. If the reliability assessment is conducted for a
short reference period, it is fair to assume stationarity for daily load effects. That is, the traffic volumes are
not increasing each day and thus daily maximum load effects can be assumed to be identically distributed.
To consider non-stationary processes, time-dependent reliability methods are required
pf = Pr (g < 0) A-3
where
There exists are a range of tools available for quantifying this probability of failure. These include:
• First order reliability methods (FORM);
• Second order reliability methods (SORM);
• Simulation-based (for example Monte Carlo Simulation (MCS)).
A range of textbooks are available with worked examples for these and various other structural reliability
methods (Ang, 1984, Stewart and Melchers, 1997, Haldar and Mahadevan, 2000, Schneider, 2006, Lemaire
et al., 2009, Cremona, 2012, Nowak and Collins, 2012, FIB CEB-FIP, 2018, Melchers and Beck, 2018).
If the probability distribution of a random variable such as traffic loading is for a time , then the calculated
probability of failure is for the same selected reference period ()
β = -Ф-1(pf) A-4
where
β = reliability index
Figure A-2: Relationship between reliability index (β) and the probability of failure (pf).
Notice, the relationship is not linear, but logarithmic (similar to the Richter scale used for measuring the
severity of earthquakes). The higher the reliability index, the safer is the component being assessed.
The acceptability of this level of safety is compared with a target reliability index (see Appendix B).
The directional cosines ( values) are outputs from FORM analysis that describe how strong each random
variable (v) in a limit state function contributed to the quantified structural reliability index. The squaring of
each value normalises these terms as shown in Equation A-5.
𝑛
∑ 𝛼𝑣2 = 1 A-5
𝑣=1
where
The 2 terms act as a form of sensitivity analysis for the reliability index, and are often called the importance
coefficients (also referred to as influence coefficients) (Roscoe et al., 2015). The larger the importance
coefficient for a random variable, the greater the relative importance or influence the uncertainty of the
variable has on the reliability index result, compared to the other variables in the limit state function.
Variables with very low influence can be treated as deterministic in the limit state function. More rigorous
data collection may be necessary for variables with very high influence on the reliability index. This is to
reduce their uncertainty and converge towards a more confident value of the structural reliability index, the
probability of failure, and thus the level of safety.
where
t = time
Since the reliability assessment in this work is conducted for a short reference period (annual) and bridge
capacities are assumed to be ‘as-built’ time-dependent reliability methods are not required. Example of time-
dependent reliability for bridges can be found in the literature (Akgül and Frangopol, 2004, Stewart and Suo,
2009, Hajializadeh et al., 2016, Wang et al., 2017).
A.7 Applications
A prominent application for structural reliability theory is in the calibration of capacity safety factors (ϕ) and
loading safety factors (γ). These factors are reliability-calibrated so new structures being designed, and
existing structures being assessed, meet an acceptable uniform benchmark of safety (the prescribed target
reliability index (βT)). Discussion on the choice for βT is found in Appendix B.
For example, the partial factors of safety for steel and concrete in the Danish Structural Codes were
historically implied to be 1.32 and 1.58, respectively. Using a target reliability index, βT = 4.79 (pf ≈ 1×10-6),
Sørensen (2002) established the reliability-calibrated factors to be 1.23 and 1.49. In the final code these
were given as 1.3 and 1.5, and so designs using these values actually achieve higher safety than the target
level of safety (βT).
In Australia, AS 5104-2017 and AS ISO 13822-2005 give the general reliability principles. The target
reliabilities presented in AS 5104-2017 in fact have been used for calibration of the Australian concrete
structures design AS 3600 (Stewart et al., 2016) and masonry structures design code AS 3700 (Lawrence
and Stewart, 2009). However, the Australian Standard AS 5100 for bridges does not give specific guidance
on the benchmark level of safety (or target reliability index) for its safety factors ϕ and γ. A review of the
code’s historical developments suggests these factors may be selected based on ensuring the “status quo”
(Marsh, 1985), or, as is the case for traffic loading safety factors, through engineering judgement (Heywood
et al., 2000). Hence, it appears the current Australian bridge design code is not calibrated to a prescribed
target reliability (Caprani et al., 2017). This is notably different to other recent international codes of practice
(Faber and Sørensen, 2002). As a result, the benchmark safety levels implied by the historical design codes
of Australian bridges is unknown.
In assessment, the use of reliability-calibrated factors is also known as a semi-probabilistic approach for
assessment. It does not inform on the actual level of safety for the structure. Rather, the method informs
whether the structure meets the specified benchmark. When the safety factors are not reliability-calibrated
and selected arbitrarily, as is the case for AS 5100, the approach is considered a deterministic approach.
The calibrated safety factors in assessment codes of existing structures are based on uniform notional
benchmark levels of safety. However, structural members have different benchmark levels of safety. Hence,
this may result in conservative conclusions on safety for an individual bridge. The direct use of structural
reliability theory on the individual structural member for quantifying safety is more beneficial. Probability-
based bridge assessment (also known as PBBA) directly adopts structural reliability theory to assess actual
levels of safety of existing bridge structural members as measured by the structural reliability index.
Subsequently, the acceptability of the safety/risk is informed by comparison to a prescribed benchmark (the
target reliability index) selected based on the needs of the bridge asset owner, public safety, and the
economy. This makes the safety assessment more fit-for-purpose. Significant cost-savings through adopting
PBBA on highway bridges have been observed internationally (Enevoldsen, 2011). As such, PBBA is
explored further as part of this work.
It is important to note that target reliability indices vary from country to country; some are based on past
experience (i.e. based on reliability of existing codes) or international and Australian standards based also
on past experience and judgements about risk acceptability (ISO 2394 and AS 5104). Target reliability
indices tend to be set for individual members or components. Hence, the target reliability provides guidance
on acceptable levels of safety but should not be considered as an absolute level of acceptable safety/risk.
Also note that calculated reliability indices are influenced by model error which may be obtained from
inadequate data, and calculated reliability indices may change if new performance data becomes available
and new statistics of model error are derived.
It is recognised that target reliability indices are compared with “notional" probabilities of failure. Note, target
reliability indices used for assessment are usually lower than those used for design and code calibration. For
existing structures, achieving an acceptable level of safety may require greater costs than to achieve the
same level of safety in a new design. Additionally, the required remaining service life of existing structures
are less than new designs.
The Australian Standard General Principles on Reliability for Structures AS 5104 2017 (adopted from ISO
2394-2015) provides “tentative” advice for target reliability indices based on a one year reference period and
ultimate limit states for (i) economic (cost-benefit) optimisation (Table B-1) and (ii) life safety (or Life Quality
Index LQI) acceptance criteria (Table B-2). In both cases the costs of safety measures are considered when
selecting target reliability indices. The Australian Standard Basis for Design of Structures – Assessment of
Existing Structures ISO AS 13822-2005, the Australian Building Code Board (ABCB) structural reliability
handbook for structural buildings also provides advice on target reliabilities. However, the costs of safety
measures are not considered in this standard. Hence, Sykora et al. (2017) conclude that ISO 2394-2015, AS
5104-2017 and JCSS (2000) “seem to provide a more appropriate reliability differentiation for existing
structures”. Hence, advice from AS 5104-2017 is used herein to suggest what might be target reliability
indices for existing bridges in Australia.
Consequences of Failure
Relative Costs of Safety Measures
Class 2 (Minor) Class 3 (Moderate) Class 4 (Large)
Large 1.0×10-3 (β =3.1) 4.8×10-4 (β =3.3) 1.1×10-4 (β =3.7)
Medium 1.1×10-4 (β =3.7) 1.3×10-5 (β =4.2) 5.4×10-6 (β =4.4)
Small 1.3×10-5 (β =4.2) 5.4×10-6 (β =4.4) 1.3×10-6 (β =4.7)
Source: AS 5104-2017.
Table B-1 shows consequences of failure which is related to loss of life and failure costs, presented in three
consequence classes, defined as:
Class 2: Material losses and functionality losses of significance for owners and operators but with little or no
societal impact. Expected number of fatalities fewer than 5. Examples of structures: low-rise buildings where
only a few people are present, minor wind turbines, stables, etc. For bridges these might be short to medium
spanning bridges that are not part of the HPFV network. The Joint Committee on Structural Safety
Probabilistic Model Code (JCSS, 2000) suggests that the ratio of failure costs to construction costs (CF/CC) is
approximately 2.
Class 3: Material losses and functionality losses of societal significance, causing regional disruptions and
delays in important societal services over several weeks. Expected number of fatalities fewer than 50.
Examples of structures: most residential buildings, tunnels, typical offshore facilities, larger and/or hazardous
industrial facilities. For bridges, these might be medium spanning bridges that form part of the HPFV
network. This relates to CF/CC = 2 to 5.
Class 4: Disastrous events causing severe societal service disruptions and delays at a national scale over
periods in the order of months. Expected number of fatalities fewer than 500. Examples of structures: high-
rise buildings, grandstands, major long-spanning bridges and tunnels, dikes, dams, smaller offshore
faculties, pipelines, refineries, chemical plants, etc. This relates to CF/CC = 5 to 10.
In the case of bridge collapse, not all vehicle occupants on it will be killed. For example, the collapse of the
10-lane, 14-span, 580 m I35W Bridge in Minneapolis in 2007 killed 13 people, but 111 vehicles were on the
bridge at the time of collapse. A bridge collapse over the Arkansas River in 2002 killed 14 people when 11
vehicles, of the many that were on the bridge, plunged into the river. The unexpectedly high survival rates
arise not only because the bridge only partially collapses but also because a car is designed to crumple on
impact and thus absorb energy. Stewart and Mueller (2014) estimate that the typical number of fatalities from
structural collapse of a typical highway bridge is about 20. On the other hand, failure of one member on a
bridge may not cause collapse, but redistribution of loads and localised damage – resulting in minimal loss of
life or modest repair costs. If this were the case, then a Class 2 (Minor) consequence may be appropriate,
but to be conservative, the consequence may be taken to be Class 3 (Moderate).
For the failure cost (CF), Stewart and Mueller 2014 considered highway bridges are two to four lanes wide
with spans of 30 to 50 m crossing rivers, roads, and railway lines. Replacement cost may be set at $20
million and is double the initial construction cost (CC = $10 million). Traffic diversion and associated user
delay costs for a bridge under reconstruction can total $430,000 per day, which, even in the case of a rapid
bridge replacement in the city of Oklahoma of only 46 days, amounts to nearly $20 million. In addition to the
economic cost of traffic diversion, there are other social and economic costs to a community. These are
harder to quantify but may be in the order of tens to hundreds of millions of dollars because, although the
loss of one bridge will not isolate a community, it will generally cause considerable inconvenience and
disruption. A loss of $100 million is assumed and assume that the expected number of fatalities is 20, at a
cost of $130 million based on value of statistical life of $6.5 million. The total losses for a damaged bridge
including both the loss of life and economic considerations is CF = $250 million, leading to CF/CC = 25.
However, the bridge families considered in this study mostly span less than 35 m. Thus, CF/CC are likely be
to be lower for these bridges.
The term ‘relative costs of safety measures’ referred to in Table B-1 is the cost needed to reduce the risk and
its effect on the target reliability index. JCSS (2000) notes that for “most typical design situations” the relative
cost of safety measures is medium and consequences of failure are Class 3. However, “for existing
structures the costs of achieving a higher reliability level are usually high compared to structures under
design. For this reason, the target level for existing structures usually should be lower”. In this case it may be
reasonable to assume that the relative cost of safety measures is large for existing bridges. For example, this
might be bridge replacement if an existing bridge fails a safety assessment. On the other hand, the relative
costs of safety measures may be lower (e.g. Medium) if the decision is to restrict vehicle loads rather than
recommend full bridge replacement.
Table B-2: Annual target reliability index based on life safety (LQI = life quality index).
Source: AS 5104-2017
Note that Table B-1 is based on a normal design life. JCSS (2000) states that “For structures designed for
short service life or otherwise rapid obsolescence (say less than 10 years) the beta-values can be lowered
by one or half a class.” If the policy horizon for asset owners is ten years, then a lower target reliability index
may be permitted. JCSS (2000) also notes that “The values given relate to the structural system or in
approximation to the dominant failure mode or structural component dominating system failure. Therefore,
structures with multiple, equally important failure modes should be designed for a higher level of reliability”. If
the bridge has more than one dominant failure mode (i.e., several structural elements can fail during the
same failure event), then a higher target reliability index may be selected.
B.4 Summary
The discussion above provides some guidance on how risk acceptance criteria (target reliability indices) may
be set for a highway bridge. Ultimately, the setting of the target reliability index is dependent on the ‘risk-
appetite’ of the asset owner considering the relative cost of safety measures and consequences of failure for
the mode of failure of interest. For example, when significant control measures are put in place resulting in
significant expenditures, but the consequences of failure to society are relatively low, then using the
principles of AS 5104-2017 the ultimate limit state target reliability indices may around 3.1. This may be
applicable in the assessment of a highway bridge not in the HPFV network subject to a well controlled
oversize overmass vehicle. Alternatively, if the asset owner decides that a medium consequence class and
the relative cost of safety measures is moderate, appropriate for highway bridges, the annual βT is taken as
4.2 for ductile members and sections, and 4.4 for brittle failure. Finally, if there is much doubt in the costs
and consequence of failure, or little control on loading, a target reliability index as high as 5.0 may be
selected.
Steel material road bridges are seen mainly in designations SR, SU, and SP. However, these steel bridges
are not as prevalent compared to the concrete bridges. As such, the bridge inventory subset will not be
considering steel bridges, only concrete road bridges.
Figure C-3: Number of traffic lanes for concrete bridge structural forms.
From this plot, some general families can be observed. For example, there are a significant number of SI
(mainly I-girders) between 10.5 m (35’) long and 18 m (60’) long, correlating to historical Victorian standard
design drawings to MS18 traffic load model. Similarly, there are also an elevated number of bridges of the
form SE (mainly planks), SN and SO (mainly U-slabs) at the spans 4.5 m (15’), 6 m (20’), 7.5 m (25’), 9 m
(30’), and 10.5 m (35’) correlating to their standard design drawings to MS18 traffic load model. Meanwhile,
AS (mainly Super-Ts) range between 18 m to 35 m, with peaks responding to the current recommended
lower and upper spans for the different standardised sections (T1 to T5; National Precast Concrete
Association Australia, 2002). However, AS does also have some variation, mostly attributed to the broad
designation as it includes both Super-Ts and T-roughs.
In contrast, other forms have much variation, are rather unique, and hence cannot be easily classified into a
bridge family. This is the case of SB (mainly box girders) and SS (mainly flat slabs). As such, these forms will
not be considered in the subset.
When the T44 traffic model was introduced, standard design drawings were not updated. Standard sections
for the different forms were still present, for example Types 1 to 4 I-girders. Moreover, significant
development in the structural form for PSC girders in Victoria took place around the same time ABDC was
first released to improve construction efficiency. These include the introduction of T-slabs (6 types;
(Rapattoni and Wells, 1991), bulb T-beams, and later closed-flanged precast Super-T girders (Types 1 to 4;
(Rapattoni et al., 1994). It follows that ABDC (T44) would have been used for the design of these structural
forms, although the standard sections were not published in the code appendices, but some were developed
by the Victorian Department of Transport. According to the Victorian Department of Transport, precast Tee-
slabs designed to T44 (Figure C-5) were most prominent, as such they have been considered in the subset.
As seen in Appendix H of AS 5100.5 (2004), the code and its load model (SM1600) have been used for the
standard I-girders, closed-flanged Super-Ts, and now, popular in new constructions, open-flanged Super-Ts.
While again each design is most likely unique, lower and upper spans of two traffic lane bridges for each
standard girder can be requested to define the family. Now, I-girders have been designed using three
historical codes. However, given the time and budget constraints, I-girders designs to AS 5100 (SM1600) will
not be considered in the bridge inventory subset.
Table D-1: Breakdown of 25,000 vehicles per day in bi-directional traffic flow with uniform pattern (Flat-Bi)
across two traffic lanes.
Total Vehicles Lane 1 Vehicles (80 km/h) Lane 2 Vehicles (80 km/h)
Hour Trucks Cars % Trucks Cars %
Trucks Cars
(50% of Total) (50% of Total) Cars (50% of Total) (50% of Total) Cars
1 150 100 75 50 40 75 50 40
2 150 100 75 50 40 75 50 40
3 150 100 75 50 40 75 50 40
4 150 100 75 50 40 75 50 40
5 150 100 75 50 40 75 50 40
6 350 650 175 325 65 175 325 65
7 420 1080 210 540 72 210 540 72
8 420 1080 210 540 72 210 540 72
9 420 1080 210 540 72 210 540 72
10 420 1080 210 540 72 210 540 72
11 420 1080 210 540 72 210 540 72
12 420 1080 210 540 72 210 540 72
13 420 1080 210 540 72 210 540 72
14 420 1080 210 540 72 210 540 72
15 420 1080 210 540 72 210 540 72
16 420 1080 210 540 72 210 540 72
17 420 1080 210 540 72 210 540 72
18 420 1080 210 540 72 210 540 72
19 420 1080 210 540 72 210 540 72
20 350 650 175 325 65 175 325 65
21 350 650 175 325 65 175 325 65
22 240 260 120 130 52 120 130 52
23 240 260 120 130 52 120 130 52
24 150 100 75 50 40 75 50 40
Table D-2: Breakdown of 25,000 vehicles per day in uni-directional traffic flow with uniform pattern (Flat-Uni)
across two traffic lanes.
Total Vehicles Lane 1 Vehicles (80 km/h) Lane 2 Vehicles (100 km/h)
Hour Trucks Cars % Trucks Cars %
Trucks Cars
(80% of Total) (50% of Total) Cars (20% of Total) (50% of Total) Cars
1 150 100 120 50 29.41 30 50 62.50
2 150 100 120 50 29.41 30 50 62.50
3 150 100 120 50 29.41 30 50 62.50
4 150 100 120 50 29.41 30 50 62.50
5 150 100 120 50 29.41 30 50 62.50
6 350 650 280 325 53.72 70 325 82.28
7 420 1080 336 540 61.64 84 540 86.54
8 420 1080 336 540 61.64 84 540 86.54
9 420 1080 336 540 61.64 84 540 86.54
10 420 1080 336 540 61.64 84 540 86.54
11 420 1080 336 540 61.64 84 540 86.54
12 420 1080 336 540 61.64 84 540 86.54
13 420 1080 336 540 61.64 84 540 86.54
14 420 1080 336 540 61.64 84 540 86.54
15 420 1080 336 540 61.64 84 540 86.54
16 420 1080 336 540 61.64 84 540 86.54
17 420 1080 336 540 61.64 84 540 86.54
18 420 1080 336 540 61.64 84 540 86.54
19 420 1080 336 540 61.64 84 540 86.54
20 350 650 280 325 53.72 70 325 82.28
21 350 650 280 325 53.72 70 325 82.28
22 240 260 192 130 40.37 48 130 73.03
23 240 260 192 130 40.37 48 130 73.03
24 150 100 120 50 29.41 30 50 62.50
Table D-3: Breakdown of 25,000 vehicles per day in bi-directional traffic flow with peaked pattern (Low-Bi)
across two traffic lanes.
Total Vehicles Lane 1 Vehicles (80 km/h) Lane 2 Vehicles (80 km/h)
Hour Trucks Cars Trucks Cars %
Trucks Cars % Cars
(50% of Total) (50% of Total) (50% of Total) (50% of Total) Cars
1 150 100 75 50 40 75 50 40
2 150 100 75 50 40 75 50 40
3 150 100 75 50 40 75 50 40
4 150 100 75 50 40 75 50 40
5 150 100 75 50 40 75 50 40
6 240 260 120 130 52 120 130 52
7 450 1550 225 775 77.5 225 775 77.5
8 450 2550 225 1275 85 225 1275 85
9 450 2050 225 1025 82 225 1025 82
10 420 1080 210 540 72 210 540 72
11 350 650 175 325 65 175 325 65
12 350 650 175 325 65 175 325 65
13 350 650 175 325 65 175 325 65
14 350 650 175 325 65 175 325 65
15 350 650 175 325 65 175 325 65
16 420 1080 210 540 72 210 540 72
17 420 1080 210 540 72 210 540 72
18 420 1080 210 540 72 210 540 72
19 420 1080 210 540 72 210 540 72
20 350 650 175 325 65 175 325 65
21 350 650 175 325 65 175 325 65
22 240 260 120 130 52 120 130 52
23 240 260 120 130 52 120 130 52
24 150 100 75 50 40 75 50 40
Table D-4: Breakdown of 25,000 vehicles per day in uni-directional traffic flow with peaked pattern (Low-Uni)
across two traffic lanes.
Total Vehicles Lane 1 Vehicles (80 km/h) Lane 2 Vehicles (100 km/h)
Hour Trucks Cars % Trucks Cars %
Trucks Cars
(80% of Total) (50% of Total) Cars (20% of Total) (50% of Total) Cars
1 150 100 120 50 29.41 30 50 62.50
2 150 100 120 50 29.41 30 50 62.50
3 150 100 120 50 29.41 30 50 62.50
4 150 100 120 50 29.41 30 50 62.50
5 150 100 120 50 29.41 30 50 62.50
6 240 260 192 130 40.37 48 130 73.03
7 450 1550 360 775 68.28 90 775 89.60
8 450 2550 360 1275 77.98 90 1275 93.41
9 450 2050 360 1025 74.01 90 1025 91.93
10 420 1080 336 540 61.64 84 540 86.54
11 350 650 280 325 53.72 70 325 82.28
12 350 650 280 325 53.72 70 325 82.28
13 350 650 280 325 53.72 70 325 82.28
14 350 650 280 325 53.72 70 325 82.28
15 350 650 280 325 53.72 70 325 82.28
16 420 1080 336 540 61.64 84 540 86.54
17 420 1080 336 540 61.64 84 540 86.54
18 420 1080 336 540 61.64 84 540 86.54
19 420 1080 336 540 61.64 84 540 86.54
20 350 650 280 325 53.72 70 325 82.28
21 350 650 280 325 53.72 70 325 82.28
22 240 260 192 130 40.37 48 130 73.03
23 240 260 192 130 40.37 48 130 73.03
24 150 100 120 50 29.41 30 50 62.50
Table D-5: Breakdown of 50,000 vehicles per day in bi-directional traffic flow with peaked pattern (High-Bi)
across two traffic lanes.
Total Vehicles Lane 1 Vehicles (80 km/h) Lane 2 Vehicles (80 km/h)
Hour Trucks Cars % Trucks Cars %
Trucks Cars
(50% of Total) (50% of Total) Cars (50% of Total) (50% of Total) Cars
1 240 260 120 130 52 120 130 52
2 240 260 120 130 52 120 130 52
3 240 260 120 130 52 120 130 52
4 240 260 120 130 52 120 130 52
5 240 260 120 130 52 120 130 52
6 420 1080 210 540 72 210 540 72
7 450 2050 225 1025 82 225 1025 82
8 450 2550 225 1275 85 225 1275 85
9 450 2550 225 1275 85 225 1275 85
10 450 2550 225 1275 85 225 1275 85
11 450 2550 225 1275 85 225 1275 85
12 450 2550 225 1275 85 225 1275 85
13 450 2550 225 1275 85 225 1275 85
14 450 2550 225 1275 85 225 1275 85
15 450 2550 225 1275 85 225 1275 85
16 450 2550 225 1275 85 225 1275 85
17 450 2550 225 1275 85 225 1275 85
18 450 2550 225 1275 85 225 1275 85
19 450 2550 225 1275 85 225 1275 85
20 450 2050 225 1025 82 225 1025 82
21 450 1550 225 775 77.5 225 775 77.5
22 420 1080 210 540 72 210 540 72
23 350 650 175 325 65 175 325 65
24 240 260 120 130 52 120 130 52
Table D-6: Breakdown of 50,000 vehicles per day in uni-directional traffic flow with peaked pattern (High-Uni)
across two traffic lanes.
Total Vehicles Lane 1 Vehicles (80 km/h) Lane 2 Vehicles (100 km/h)
Hour Trucks Cars % Trucks Cars %
Trucks Cars
(80% of Total) (50% of Total) Cars (20% of Total) (50% of Total) Cars
1 240 260 192 130 40.37 48 130 73.03
2 240 260 192 130 40.37 48 130 73.03
3 240 260 192 130 40.37 48 130 73.03
4 240 260 192 130 40.37 48 130 73.03
5 240 260 192 130 40.37 48 130 73.03
6 420 1080 336 540 61.64 84 540 86.54
7 450 2050 360 1025 74.01 90 1025 91.93
8 450 2550 360 1275 77.98 90 1275 93.41
9 450 2550 360 1275 77.98 90 1275 93.41
10 450 2550 360 1275 77.98 90 1275 93.41
11 450 2550 360 1275 77.98 90 1275 93.41
12 450 2550 360 1275 77.98 90 1275 93.41
13 450 2550 360 1275 77.98 90 1275 93.41
14 450 2550 360 1275 77.98 90 1275 93.41
15 450 2550 360 1275 77.98 90 1275 93.41
16 450 2550 360 1275 77.98 90 1275 93.41
17 450 2550 360 1275 77.98 90 1275 93.41
18 450 2550 360 1275 77.98 90 1275 93.41
19 450 2550 360 1275 77.98 90 1275 93.41
20 450 2050 360 1025 74.01 90 1025 91.93
21 450 1550 360 775 68.28 90 775 89.60
22 420 1080 336 540 61.64 84 540 86.54
23 350 650 280 325 53.72 70 325 82.28
24 240 260 192 130 40.37 48 130 73.03
E.1 Overview
In Step 1, the level of safety for the subset of the ‘as-built’ bridges’ families presented in Table 3-1 under their
design loads is obtained. Super-T girders are designed precisely to the specifications to AS 5100 (2004) and
are the slimmest possible complying designs for the ultimate limit states of bending and shear as per the
code provisions. All other bridges are already built, thus contain some degree of conservatism originally
interpreted by the designer. Figure E-1 summarises the process undertaken for accomplishing Step 1. It
follows the general framework shown in Figure 3-1, but focuses on the load effects due to the design code
traffic load model for SLL in Equation 1. Full results are seen in Appendix J.
• Limit state functions for the critical modes of failure are developed (Equation 1)
• Model errors for capacity and loading are established (Appendix I)
• The First Order Reliability method is used to calculate the structural reliabiltiy index ()
Reliability • Validation of the approach is made using two different types of software and using Second Order
Analysis Reliability method to test the linearity of the limit state function
The engineering drawings provided by the bridge asset jurisdictions and designs of the bridge families listed
in Table 3-1 were studied to establish the material and geometric properties of the critical components being
assessed. The necessary properties established were the input parameters for the selected structural
models (presented in Appendix G) used for calculating the ultimate bending capacity and ultimate shear
capacity.
A list of these input parameters and their associated capacity models are summarised in Table 4-1. Only the
properties at the critical section for bending, and critical section for shear were necessary.
Since all bridge decks are simply-supported, the critical section for ultimate bending is taken as the mid-
span. Meanwhile, the critical section for shear is taken at a distance of do from the face of the support in
accordance to AS 5100.7 (2017) Clause 8.2.3.2, where do is the distance from the top fibre of the section to
the outermost steel reinforcement layer centroid (rebar or prestressed).
If the historical engineering drawings could not provide sufficient information on these properties, then
relevant Australian literature was sought. This included Appendix A of AS 5100.7 (2017), Appendix C of
Austroads (2014), and Munter and Lume (2017). Otherwise, appropriate assumptions were made (for
example, if not indicated, the amount of prestress losses were taken as 30%).
To develop the probability distribution of capacity (R), Equation 2 may be used. Indeed, the international
literature has already published bias and coefficient of variations for their ultimate bending capacity models
(Wisniewski et al., 2009) and ultimate shear capacity models (Nowak et al., 1994, Ribeiro et al., 2016)
through this approach.
However, when the structural capacity models involve multiple random input parameters, as is the case here,
then Monte Carlo simulations are more appropriate for developing the probabilistic distribution of capacity
(Foster et al., 2016). Moreover, Equations 3 and 4 assume random variables are independent. Indeed, there
exists correlation between the material properties (e.g. strength and elasticity). Thus, by using Monte Carlo
simulation (MCS) of the input parameters, the probability distribution of capacity can be properly developed
for the Australian capacity models.
In MCS, a realisation (a single observation from the distribution) of each input parameter is selected at
random, noting the correlation with the other input parameters. Based on a literature review, Table 4-1
presents the probabilistic distributions for the input parameters of the structural capacity models to describe
their uncertainty (material quality, fabrication process, etc.). It is interesting to note here, that although
characteristic values for the resistance are commonly the 5th-percentile of the distribution, the biases
suggest the percentile is in fact lower. The correlation coefficient between the assumed non-independent
random variables is presented in Table 4-2. The set of realisations are then inputted to each of the structural
models, and the capacity realisation is outputted. The capacity reduction factor (ϕ) is not included in the
simulations.
The process is then repeated N times (106 for this work) to obtain N realisations of capacity. Then, using a
suitable distribution fitting approach (Appendix A) the best-fitting distribution type (normal or log-normal) and
its associated statistical descriptors (bias and coefficient of variation) are obtained. The fitting is best
conducted on the lower tail of the distribution as this is within the failure region (Figure 2-2). Summarised
results from the demonstrated bridge structural elements (Table 3.1) are found in Appendix G.
The established probably distributions can then be used for the reliability analysis.
Apart from the respective codes, the details on these traffic load models are also found in Appendix C of
Austroads (2014).
To obtain the bridge load effects, representative bridge decks were modelled using finite element software
for the bridge families at their different spans shown in Table 3-1. As seen in the limit state function (Equation
1), loading (S) is considered to be comprised of four parts; self-weight, slab load (from an overlay slab if
present), superimposed dead load, and maximum traffic load. The characteristic values for these components
are extracted, that is, loading factors of safety γ are not applied. Details are found in Appendix H.
Permanent Loads
For the self-weight load effects, slab load effects, and superimposed dead load effects, the statistics
published by Rakoczy and Nowak (2013) as presented in Table 4-3 are generally well established to use in
reliability analysis of bridges (Caprani et al., 2017). Hence, they are also used in this work to convert
characteristic values to probability distributions.
Traffic Loads
For obtaining the probabilistic distribution of maximum load effect, whilst studies by Pham (1985b) provide
biases and CoV, these are measured for floor loadings, and not vehicles on bridges. For the Australian limit
state design codes (ABDC 1992, AS 5100 2004), as per any limit state design code, the load effects from
these traffic load models should be the characteristic values. By definition, the characteristic value is
associated with a probability of exceedance in the reference period (). According to AS 5100.1 (2018)
Clause 8.3.1 (i) states:
“An ultimate design action is an action that has a 5% probability of being exceeded during
the design life”
From this probability of exceedance, and further knowing the coefficient of variation and distribution type, the
bias can be established. The bias for a Gumbel distribution, given from a given probability of occurrence and
coefficient of variation is shown in Equation 5. This is appropriate for maximum values (see Appendix A).
The international literature recommends a coefficient of variation of 0.18 for bridges including dynamic
allowance (Rakoczy and Nowak, 2013). For Australia, Austroads (2014) presents WIM data for heavy vehicle
mass plotted on normal probability paper (Figure E-2).
Figure E-2: Distribution of Australia WIM data on triaxle mass suggests a CoV of 0.167 at the legal load
Note for normal probability paper, the slope of the fitted line is the standard deviation of the distribution (Ang,
1984). Fitting only the data around the legal limit, (a good representation of the intended design), the
calculated coefficient of variation of 0.167 is not far from the recommended 0.18. As such, to be slightly
conservative as CoV of 0.18 is adopted in this study.
Recall, an annual reference period is selected for the reliability assessment. Hence, the probability of
exceedance in Equation 5 must be for an annual basis. As per AS 5100, the probability of exceedance is 5%
for the design-life. Thus, after 100 years, 95% of peak loading events were less than the ultimate load (the
probability of non-exceedance). Assuming peak loading events each year were independent and identically
distributed (ie. stationary process), the product of the probability that peak loading events were less than the
ultimate load each year out of 100 years is equal to the probability that peak loading events were less than
the ultimate load in 100 years. In other words, the product of 100 annual probabilities of non-exceedance is
equal to the 100-year probability of non-exceedance. This is described in Equation E-1.
[1 − p(1)]
100
= 1 − p(100) E-1
where
probability the characteristic value (traffic load model stipulated in the code) is exceeded
p(1) =
annually
probability the characteristic value (traffic load model stipulated in the code) exceeding
p(100) =
over 100 years
Thus, a 5% chance of exceeding (95% chance of non-exceedance) during the design life of 100 years
(p(100) = 0.05), equates to a 0.05% chance of exceeding annually (p(1) = 0.0005 or 99.95% chance of non-
exceedance). Based on this annual probability of exceedance, a coefficient of variation of 0.18, and
assuming peak load effects follows a Gumbel distribution, the bias for annual peak traffic load effect for Step
1, as calculated by Equation 5, is equal to 0.50.
Using this bias and coefficient of variation, the characteristic values established from the grillage models are
converted to appropriate probability distributions for use in the reliability analysis.
F.1 Overview
In Step 2, the developed reliability-based framework is implemented to quantify the level of safety for the
subset of the ‘as-built’ bridges families presented in Table 3-1 under current ‘as-of-right’ freight vehicles
access regimes (selected as the Higher Mass Limit network). Figure F-1 summarises the process
undertaken for accomplishing Step 2. Notice, the procedure is similar to Step 1, with the traffic loading
random variable instead being for real traffic flows (i.e. multiple vehicles), and not notional load models. That
is, Step 2 relies on the probability distributions for ‘as-of-right’ freight vehicles at different traffic flows. This is
established by simulations of traffic data obtained from relevant HML network weigh-in-motion (WIM) site
(West Gate Bridge). Full results are seen in Appendix K.
• Weigh-in-motion data is processed to establish the freight network "garage" defined as the
dataset of the characteristics and proportions of the different heavy vehicle types present in the
selected freight network
Vehicle Fleet
• Bridge traffic is simulated across the two traffic lanes of the selected bridge families (Table 3.1)
uniformly from the vehicle fleet ("garage"), but considering possible variations in individual vehicle
characteristics (kernels), and appropriate headway/vehicle spacing models
Traffic • Simulations are conducted for different car and truck flow distributions through time, considering
Simulation both two lane uni-directional and two lane bi-directional traffic flows (6 flows in total, Appendix D).
• Influence lines for each traffic lane of the selected bridge families for the two modes of failure are
established to convert simulated traffic loads to bridge traffic load effect (Appendix H)
• Maximum annual load effects from any number of truck loading events are obtained (Block
Loading Maximum Method) and probabilistic characterised
• The framework of Step 1 is used with these new traffic loading distributions but same capacity
distributions (Appendix G) and permanent load distribtion to determine the current levels of safety
under current freight access regimes, adopting model errors seen in Appendix I.
Reliability
Assessment
Moreover, data from WIM informs on the proportion of the different vehicle type configurations (rigid trucks,
semi-trailer, B-Doubles, etc).
In this work, six months of WIM data from the West Gate Bridge (relevant for the HML network) was
collected, cleaned, and processed to obtain a sufficient database of the heavy vehicles and their associated
characteristics.
Over the six months of collected data, approximately 3.2 million heavy vehicles were recorded. The
proportions for the different vehicle configurations based on the data are shown in Table F-1.
Table F-1: Breakdown of Heavy Vehicles based on 6-months WIM Data on West Gate Bridge.
Source: West Gate Bridge WIM data from Victorian Department of Transport.
The database of 3.2 million was used to randomly select a “garage” of 1 million heavy vehicles upon which
the Monte Carlo simulations are based. Random selection means the garage has the same proportions of
vehicle types as the measured vehicles shown in Table F-1. The “garage” is a subset of the WIM database to
reduce computational demands. It is assumed that this garage is an appropriate representation of the heavy
vehicles in the HML networks across Australia.
F.3.1 Methodology
The in-house software Bridge Traffic Load Simulation (BTLS) simulates artificial traffic streams crossing
bridges with pre-defined properties (span length, number of lanes, etc) using Monte Carlo simulation, and
calculates resulting load effects. For this work, traffic is simulated on each of the two lanes for the bridge
families with spans presented in Table 3-1. The traffic lanes are taken as the design lanes used in Step 1.
For generating the vehicle in the artificial traffic stream, the Monte Carlo simulation adopted is known as a
smoothed bootstrap using variable-bandwidth kernel density estimators (Enright, 2010, Scott, 2015). In this
method, a vehicle is randomly selected from the garage of measured vehicles (bootstrapping), and then
randomised (smoothing) using deviates (kernels). Thus, a very large number of artificial heavy vehicles in the
traffic stream, including overloaded vehicles, can be generated, while maintaining intrinsic correlations and
relationships within the vehicle database and characteristics (e.g. correlation between axle weights and
gross vehicle mass).
The vehicle parameters modified using the kernels are the GVM, axle weights, and axle spacings. The
literature presents two types of such estimators: a normal and triangle kernel. A normal probability
distribution is selected for this work (Figure F-2) with selected parameters for the different vehicle
characteristics shown in Table F-2. A sensitivity analysis has indicated results are generally insensitive to the
choice of the parameters shown in Table F-2.
1.0
Probability Density
0.5
0.0
0.9 1.0 1.1
x
Table F-2: Variable-bandwidth kernels density adopted for artificial traffic simulation.
Headway Model
Over time, BTLS selects the next vehicle (car or heavy vehicle from “garage” modified by the kernels) to
traverse the bridge using an appropriate headway model. Such a model describes the time gaps between
vehicles arriving at the bridge and thus their spacing. The selected headway model is a free-flow model
using a Poisson arrival assumption upon the normalised model of Crespo-Minguillón and Casas (1997). A
minimum gap of 0.1 s and minimum spacing of 1.0 m is enforced to prevent virtual overlaps of the artificially
generated vehicles. This approach is conservative (O'Brien and Caprani, 2005). The choice between a car or
heavy vehicle is based on their predefined proportions in the traffic flows (Appendix D).
Simulation Run
The simulation is run for a defined number of days using the traffic flows presented in Appendix D. In other
words, for this demonstration of the framework implementation, each of the assessed bridges has 6
quantified annual reliability indices for each of the 2 structural mechanisms (bending moment and shear), a
total of 12 per bridge, or 384 in total.
For this work, 25,000 ‘economic’ days (representing 100 calendar years) is selected. Notice, this is longer
than the period of data collection (6 months). There is no limit on the number of days for simulation.
However, the choice of 100 years is to ensure a balance between accuracy and computational/post-
processing effort for establishing the probabilistic distribution.
F.4 Loading
Influence lines (ILs) are graphs describing the change in a load effect at a certain point of structure as a unit
load traverses the structure. Hence, ILs are used to convert the bridge loading events from the artificially
generated traffic of BTLS to the desired loading effect at a certain point of the bridge deck. This conversion is
also accomplished by the in-house software, upon input of relevant influence lines files.
While the Muller-Breslau principle can be adopted for establishing ILs for simple beams (Megson, 2019), the
lateral behaviour across a bridge deck is not captured. Instead, the influence lines of each traffic lane are
established for all bridge families using the grillage/finite element models from Step 1. More details on these
models are found in Appendix H.
As shown in Figure F-3, a 1 kN bogey spaced 2.3 m laterally is applied as the moving load on the traffic lane
of these models. Then, the load effects (bending moment or shear force) at the critical location of the bridge
deck for traffic loading as determined from Step 1 are extracted to construct the influence line for each lane.
These influence lines are presented in Appendix H. Note, the positioning of the traffic lanes and critical
locations vary across the bridge families. As a result, the shapes of these influence lines vary.
Figure F-3: Example of constructing the influence line at a critical location for Lane 1.
Lane 1 (Slow)
3.2 m Lane 2 (Fast)
1kN Bogey 3.2 m
(2.3 m)
Critical Location
Travel
for Load Effect
Direction
During the simulation, the maximum load effect for block sizes of = 1 year is selected from any number of
trucks’ bridge loading events (Block Maximum Method as discussed in Appendix A). These provide the
realisations for the probability distribution of annual maximum traffic loading. Since the simulation is run for
100 years for each bridge, there are then 100 realisations for each of the six flows for each load effect
(bending or shear).
Using fitting approaches described in Appendix A, the appropriate Generalized Extreme Value (GEV)
probability distribution parameters for the annual maximum load effects is established at the six different
flows for each bridge. Summarised results from the demonstrated bridge structural elements (Table 3.1) are
found in Appendix H.
The ultimate capacity in bending is estimated using AS 5100.5 (2017) Clauses 8.1.2 to 8.1.5. In summary, it
applies the following assumptions:
i. plane sections remain plane;
ii. concrete has no tensile strength;
iii. the distribution of compressive stress of the concrete is approximated by a rectangular stress
block, and;
iv. the maximum compressive reinforcement strain does not exceed 0.003.
0.003 α2 f
0.5γdn
γdn
dn Neutral
Axis
dp,i ds
(1-γ) dn (1-0.5γ) dn
ksσsuAs
εsu
kp,iσpu,iAp
εpu,i
In Figure G-1, ks is the number of any passive steel reinforcements, kp,i is the number of prestressed strands
in row i of K rows, As is the area of each passive steel reinforcement, Ap is the area of each prestressed
strand, ds the depth from the top fibre to the centroid of the passive steel reinforcement and dp,i the depth
from the top fibre to the centroid of the row of prestressed strands i. The ultimate stress of the passive steel
reinforcements, σsu, and prestressed strands in each row, σpu,i, is determined from strain compatibility.
Considering the idealised elasto-plastic behaviour of the materials, in general these stresses are:
where
εpe,i = prestrain of the row of prestressed strands (considering prestress losses, ∆P)
dn is the depth from the top fibre to the neutral axis at ultimate strength in bending, and b is the width of the
top fibre. For AS 5100.5, the parameters for the rectangular stress block, α2 and γ (seen in Figure G-1), are 1
– 0.003f’c and 1.05 – 0.007 f’c, respectively. Both parameters for the rectangular stress block have limits of
0.67 to 0.85. Further details on the procedure can be found in Melhem et al. (2018)
Figure G-2 depicts the shear design provisions found in AS 5100.5 for normal density concrete based on the
modified compression field theory. In Figure G-2, f’c is the compressive strength of concrete, bv is the
effective web width, where D is the total depth of the section, Asv/s is the reinforcement per unit length, and
fsy is the yield strength of the steel stirrups.
The prescribed equations for v and kv in AS 5100.5 Clause 8.2.4.2 are similar to those found in the
Canadian and AASHTO standards. The only difference is the use of 32, rather than 35, in the crack spacing
parameter equation. As discussed by Bentz et al. (2006), Collins and Mitchell (2014), and Caprani and
Melhem (2019), there exists a fundamental difference in the consideration of longitudinal strain (x) between
design and assessment:
• In design, the applied shear load is the design shear load (V = V*) as seen in Figure G-2 and the
longitudinal strain directly results.
• In assessment, the applied shear load is the ultimate shear capacity (V = Vu); in other words, the applied
shear is thought of as increasing from zero, until failure of the section at V = Vu, and so the longitudinal
strain must be consistent with this applied shear.
Moreover, flexural-shear, web-shear, and web-crush failures must be considered. A flowchart describing the
process required to adapt the design provisions in AS 5100.5 for shear assessment of concrete structures is
shown in Figure G-3. Further details on this can be found in Caprani and Melhem (2019).
Equation 8.2.4.3/4
Only use term EcAct if section is uncracked (x < 0)
2
0.9T *uh
(V )
2
M * dv + *
− Pv + *
+ 0.5 N − Apt f po
2 Ao
x =
2 ( Es Ast + E p Apt + Ec Act )
−2 10−3 x 3 10−3
Figure G-3: Flowchart adopted for using design provisions of AS 5100.5 for shear assessment.
Table G-1: Characteristic bending capacity (as per AS 5100.5 model) and associated probability distribution as
per correlated Monte Carlo simulation for all sections.
Table G-2: Characteristic shear force (as per AS 5100.5 model with adaptation for assessment) and associated
probability distribution as per correlated Monte Carlo simulation for all sections.
Figure G-4: Histogram and Gumbel probability paper plot of bending capacity realisations for MS18 designed U-
slabs. Fitting is done on the lower tail (red shading) and the best-fit normal distribution is indicated
by the red line.
Figure G-5: Histogram and Gumbel probability paper plot of bending capacity realisations for MS18 designed
PSC Planks. Fitting is done on the lower tail (red shading) and the distribution is indicated by the
red line.
Figure G-6: Histogram and Gumbel probability paper plot of bending capacity realisations for MS18 designed I-
girders. Fitting is done on the lower tail (red shading) and the distribution is indicated by the red
line.
Figure G-7: Histogram and Gumbel probability paper plot of bending capacity realisations for T44 designed I-
girder. Fitting is done on the lower tail (red shading) and the distribution is indicated by the red line.
Figure G-8: Histogram and Gumbel probability paper plot of bending capacity realisations for T44 designed
Precast Tee Slabs. Fitting is done on the lower tail (red shading) and the distribution is indicated by
the red line.
Figure G-9: Histogram and Gumbel probability paper plot of bending capacity realisations for T44 designed
Super-T girders. Fitting is done on the lower tail (red shading) and the distribution is indicated by
the red line.
Figure G-10: Histogram and Gumbel probability paper plot of bending capacity realisations for SM1600
designed Super-T girders. Fitting is done on the lower tail (red shading) and the distribution is
indicated by the red line.
Figure G-11: Histogram and Gumbel probability paper plot of shear capacity realisations for MS18 designed
U-slabs. Fitting is done on the lower tail (red shading) and the distribution is indicated by the red
line.
Figure G-12: Histogram and Gumbel probability paper plot of shear capacity realisations for MS18 designed
PSC Planks. Fitting is done on the lower tail (red shading) and the distribution is indicated by the
red line.
Figure G-13: Histogram and Gumbel probability paper plot of shear capacity realisations for MS18 designed
I-girder. Fitting is done on the lower tail (red shading) and the distribution is indicated by the red
line.
Figure G-14: Histogram and Gumbel probability paper plot of shear capacity realisations for T44 designed I-
girder. Fitting is done on the lower tail (red shading) and the distribution is indicated by the red line.
Figure G-15: Histogram and Gumbel probability paper plot of shear capacity realisations for T44 designed
Precast Tee Slabs. Fitting is done on the lower tail (red shading) and the distribution is indicated by
the red line.
Figure G-16: Histogram and Gumbel probability paper plot of shear capacity realisations for T44 designed
Super-T girders. Fitting is done on the lower tail (red shading) and the distribution is indicated by
the red line.
Figure G-17: Histogram and Gumbel probability paper plot of shear capacity realisations for SM1600
designed Super-T girders. Fitting is done on the lower tail (red shading) and the distribution is
indicated by the red line.
Apart from the T44 designed I-girder (in which a grillage model was provided by TMR), the bridge families
were modelled through finite element software with the same bridge deck cross-section for each family, but
varied in the span, to obtain the different load effect effects. Thus, a given family the bridge decks have the
same number of components (beams, girder, slabs, planks, etc). As shown in Table H-1 either a beam-and-
slab grillage analysis, or surface models with joint elements were adopted depending on the bridge family.
The differences in these models are depicted in Figure H-1 and were simply-supported with no skew or
cross-fall.
Unless specified, superimposed dead load is taken as a surfacing of 65 mm (density of 22 kN/m3) and non-
structural precast concrete edge barriers weighing 6.54 kN/m. In Step 1, for maximum traffic load, the
respective traffic load model is positioned in the ‘worst-case’ scenario (as per the design code) both
longitudinally and transversely for the selected bridge deck component for both bending moment and shear
force. In addition, the load model is factored by the dynamic allowance and multiple vehicle presence factors
as stipulated by the respective design code (eg. a 10% reduction in T44 loading to ABDC since there are two
design lanes, but no reduction for T44 loading to NAASRA).
Table H-1: Finite element model used and components in these models for the different bridge families.
Figure H-1: Finite element models adopted for obtaining characteristic load effects (a) U-slab Surface Model (b)
I-Girder Grillage Model.
(a)
(b)
Figure H-2: Representative Cross-Section for MS18 designed U-Slab Bridge Decks.
Figure H-3: Representative Cross-Section for MS18 designed PSC Planks Bridge Decks.
Figure H-4: Representative Cross-Section for MS18 designed I-girder Bridge Decks.
Table H-2: Characteristic bending moment (in kNm) at midpsan. Traffic load is from traffic load model and
includes dynamic allowance and multiple vehicle presence factors as stipulated by respective
design code.
Table H-3: Characteristic shear force (in kN) at the distance of do from face of support. Traffic load is from
traffic load model and includes dynamic allowance and multiple vehicle presence factors as
stipulated by respective design code.
Figure H-6: Influence lines of U-slab bridge family for each traffic lane (Lane 1 and Lane 2) at critical location for
bending moment and shear force based on 1 kN 2.3 m wide traversing bogey using grillage/finite
element model.
Figure H-7: Influence lines of PSC plank bridge family for each traffic lane (Lane 1 and Lane 2) at critical location
for bending moment and shear force based on 1 kN 2.3 m wide traversing bogey using grillage/finite
element model.
Figure H-8: Influence lines of MS18 I-girder bridge family for each traffic lane (Lane 1 and Lane 2) at critical
location for bending moment and shear force based on 1 kN 2.3 m wide traversing bogey using
grillage/finite element model.
Figure H-9: Influence lines of TMR I-girder for each traffic lane (Lane 1 and Lane 2) at critical location for
bending moment and shear force based on 1 kN 2.3 m wide traversing bogey using grillage/finite
element model.
Figure H-10: Influence lines of Precast Tee slab bridge family for each traffic lane (Lane 1 and Lane 2) at
critical location for bending moment and shear force based on 1 kN 2.3 m wide traversing bogey
using grillage/finite element model.
Figure H-11: Influence lines of VR Super-T girder for each traffic lane (Lane 1 and Lane 2) at critical location
for bending moment and shear force based on 1 kN 2.3 m wide traversing bogey using grillage/finite
element model.
Figure H-12: Influence lines of Super-T girder bridge family for each traffic lane (Lane 1 and Lane 2) at critical
location for bending moment and shear force based on 1 kN 2.3 m wide traversing bogey using
grillage/finite element model.
Figure H-13: Histogram and Gumbel probability paper plot of maximum annual bending moment realisations
for MS18 designed U-slab under various traffic flows. Data represented as ratio of simulated load
effect with design traffic load model load effect.
H.4.2 MS18 designed PSC Planks Maximum Annual Bending Moment Distribution
Figure H-14: Histogram and Gumbel probability paper plot of maximum annual bending moment realisations
for MS18 designed PSC Planks under various traffic flows. Data represented as ratio of simulated
load effect with design traffic load model load effect.
Figure H-15: Histogram and Gumbel probability paper plot of maximum annual bending moment realisations
for MS18 designed I-girders under various traffic flows. Data represented as ratio of simulated load
effect with design traffic load model load effect.
Figure H-16: Histogram and Gumbel probability paper plot of maximum annual bending moment realisations for
T44 designed I-girder under various traffic flows. Data represented as ratio of simulated load effect
with design traffic load model load effect.
H.4.5 T44 designed Precast Tee Slabs Maximum Annual Bending Moment
Distribution
Figure H-17: Histogram and Gumbel probability paper plot of maximum annual bending moment realisations
for T44 designed Precast Tee slabs under various traffic flows. Data represented as ratio of
simulated load effect with design traffic load model load effect.
H.4.6 T44 designed Super-T girders Maximum Annual Bending Moment Distribution
Figure H-18: Histogram and Gumbel probability paper plot of maximum annual bending moment realisations
for T44 designed Super-T girders under various traffic flows. Data represented as ratio of simulated
load effect with design traffic load model load effect.
Figure H-19: Histogram and Gumbel probability paper plot of maximum annual bending moment realisations
for SM1600 design Super-T girders under various traffic flows. Data represented as ratio of
simulated load effect with design traffic load model load effect.
Figure H-20: Histogram and Gumbel probability paper plot of maximum annual shear force realisations for
MS18 designed U-slab under various traffic flows. Data represented as ratio of simulated load effect
with design traffic load model load effect.
H.4.9 MS18 designed PSC Planks Maximum Annual Shear Force Distribution
Figure H-21: Histogram and Gumbel probability paper plot of maximum annual shear force realisations for
MS18 designed PSC Planks under various traffic flows. Data represented as ratio of simulated load
effect with design traffic load model load effect.
Figure H-22: Histogram and Gumbel probability paper plot of maximum annual shear force realisations for
MS18 designed I-girders under various traffic flows. Data represented as ratio of simulated load
effect with design traffic load model load effect.
Figure H-23: Histogram and Gumbel probability paper plot of maximum annual shear force realisations for
T44 designed I-girder under various traffic flows. Data represented as ratio of simulated load effect
with design traffic load model load effect.
H.4.12 T44 designed Precast Tee Slabs Maximum Annual Shear Force Distribution
Figure H-24: Histogram and Gumbel probability paper plot of maximum annual shear force realisations for
T44 designed Precast Tee under various traffic flows. Data represented as ratio of simulated load
effect with design traffic load model load effect.
H.4.13 T44 designed Super-T girders Maximum Annual Shear Force Distribution
Figure H-25: Histogram and Gumbel probability paper plot of maximum annual shear force realisations for
T44 designed Super-T girders under various traffic flows. Data represented as ratio of simulated
load effect with design traffic load model load effect.
H.4.14 SM1600 designed Super-T girders Maximum Annual Shear Force Distribution
Figure H-26: Histogram and Gumbel probability paper plot of maximum annual bending moment realisations
for SM1600 design Super-T girders under various traffic flows. Data represented as ratio of
simulated load effect with design traffic load model load effect.
Table H-4: Fitted probability distribution (uncertainty) for annual maximum bending moment due to different higher mass limit (HML) network at different flows
(Appendix D). Data (represented as ratio of simulated load effect with design traffic load model load effect) fits as GEV distribution with = shape
parameter, = scale parameter, u = location parameter (see Appendix A).
Table H-5: Fitted probability distribution (uncertainty) for annual maximum shear force due to different higher mass limit (HML) network at different flows (Appendix
D). Data (represented as ratio of simulated load effect with design traffic load model load effect) fits as GEV distribution with = shape parameter, = scale
parameter, u = location parameter (see Appendix A).
Probability Distribution
Bridge Family Section
Type Bias CoV
All Normal 1.06 0.05
There is ample scope for further work to better characterise model error for shear by comparison of
experimental data with model predictions as was undertaken for bending capacity.
Without any experimental results, Melhem et al. (2018) proposes a response surface methodology (RSM)
approach (Myers et al., 2016) using a surrogate model (with known model error) to represent the reality. The
software program Response2000 is used as the surrogate for the real shear capacity. This program
conducts the full MCFT analysis (15 equations in an iterative process) and is known to be highly accurate.
Following this approach, response surfaces are constructed for one section for each of the bridge families.
The response surface is a second order polynomial meta-model (Myers et al., 2016), comprising the
important random variables as found by Melhem et al. (2018). The star numerical experimental design is
used to obtain the responses (sample points) and multiple regression techniques are used for fitting the
polynomial.
Using Monte Carlo Simulation on the constructed response surfaces, the model error for the selected
sections are calculated. These are presented in Table I-2 and assumed to be the same for all other sections
within that family.
Probability Distribution
Bridge Family Section
Type Bias CoV
All U-slab bridge decks Normal 1.18 0.10
All PSC planks bridge decks Normal 0.91 0.06
All I-girder bridge decks Normal 1.22 0.11
All Precast Tee Slab bridge decks Log-Normal 1.00 0.06
All Super-T girder bridge decks Normal 1.06 0.07
Probability Distribution
Bridge Family Section
Type Bias CoV
All Normal 1.00 0.10
Figure K-1: Step 2 annual structural reliability indices for MS18 designed U slabs for different HML flows at
ultimate bending.
Figure K-2: Step 2 annual structural reliability indices for MS18 designed PSC planks for different HML flows at
ultimate bending.
Figure K-3: Step 2 annual structural reliability indices for MS18 designed I-girders for different HML flows at
ultimate bending.
Figure K-4: Step 2 annual structural reliability indices for T44 designed Precast Tee Slabs for different HML
flows at ultimate bending.
Figure K-5: Step 2 annual structural reliability indices for Case Studies for different HML flows at ultimate
bending.
Figure K-6: Step 2 annual structural reliability indices for SM1600 designed Super-T girders for different HML
flows at ultimate bending.
Figure K-7: Relative importance for each random variable that comprise limit state function in the reliability
index results (Step 2) at ultimate bending for Flat Bi.
Figure K-8: Relative importance for each random variable that comprise limit state function in the reliability
index results (Step 2) at ultimate bending for Flat Uni.
Figure K-9: Relative importance for each random variable that comprise limit state function in the reliability
index results (Step 2) at ultimate bending for Low Bi.
Figure K-10: Relative importance for each random variable that comprise limit state function in the reliability
index results (Step 2) at ultimate bending for Low Uni.
Figure K-11: Relative importance for each random variable that comprise limit state function in the reliability
index results (Step 2) at ultimate bending for High Bi.
Figure K-12: Relative importance for each random variable that comprise limit state function in the reliability
index results (Step 2) at ultimate bending for High Uni.
Figure K-13: Step 2 annual structural reliability indices for MS18 designed U slabs for different HML flows at
ultimate shear.
Figure K-14: Step 2 annual structural reliability indices for MS18 designed PSC planks for different HML
flows at ultimate shear.
Figure K-15: Step 2 annual structural reliability indices for MS18 designed I-girders for different HML flows at
ultimate shear.
Figure K-16: Step 2 annual structural reliability indices for T44 designed Precast Tee Slabs for different HML
flows at ultimate shear.
Figure K-17: Step 2 annual structural reliability indices for Case Studies for different HML flows at ultimate
shear.
Figure K-18: Step 2 annual structural reliability indices for SM1600 designed Super-T girders for different
HML flows at ultimate shear.
Figure K-19: Relative importance for each random variable that comprise limit state function in the reliability
index results (Step 2) at ultimate shear for Flat Bi.
Figure K-20: Relative importance for each random variable that comprise limit state function in the reliability
index results (Step 2) at ultimate shear for Flat Uni.
Figure K-21: Relative importance for each random variable that comprise limit state function in the reliability
index results (Step 2) at ultimate shear for Low Bi.
Figure K-22: Relative importance for each random variable that comprise limit state function in the reliability
index results (Step 2) at ultimate shear for Low Uni.
Figure K-23: Relative importance for each random variable that comprise limit state function in the reliability
index results (Step 2) at ultimate shear for High Bi.
Figure K-24: Relative importance for each random variable that comprise limit state function in the reliability
index results (Step 2) at ultimate shear for High Uni.