Iso TR 12845-2010

TECHNICAL ISO/TR
REPORT 12845
First edition
2010-08-01
Selected illustrations of fractional

factorial screening experiments
Illustrations choisies de plans d'expériences factoriels fractionnaires
Reference number
ISO/TR 12845:2010(E)
--`,,```,,,,````-`-`,,`,,`,`,,`--
© ISO 2010
Provided by IHS
No reproduction or networking permitted without license from IHS Not for Resale
ISO/TR 12845:2010(E)
PDF disclaimer
This PDF file may contain embedded typefaces. In accordance with Adobe's licensing policy, this file may be printed or viewed but
shall not be edited unless the typefaces which are embedded are licensed to and installed on the computer performing the editing. In
downloading this file, parties accept therein the responsibility of not infringing Adobe's licensing policy. The ISO Central Secretariat
accepts no liability in this area.
Adobe is a trademark of Adobe Systems Incorporated.
Details of the software products used to create this PDF file can be found in the General Info relative to the file; the PDF-creation
parameters were optimized for printing. Every care has been taken to ensure that the file is suitable for use by ISO member bodies. In
the unlikely event that a problem relating to it is found, please inform the Central Secretariat at the address given below.
COPYRIGHT PROTECTED DOCUMENT

© ISO 2010
All rights reserved. Unless otherwise specified, no part of this publication may be reproduced or utilized in any form or by any means,
electronic or mechanical, including photocopying and microfilm, without permission in writing from either ISO at the address below or
ISO's member body in the country of the requester.
ISO copyright office
Case postale 56 • CH-1211 Geneva 20
Tel. + 41 22 749 01 11
Fax + 41 22 749 09 47
E-mail [email protected]
Web www.iso.org
Published in Switzerland
--`,,```,,,,````-`-`,,`,,`,`,,`---
ii © ISO 2010 – All rights reserved

Provided by IHS
ISO/TR 12845:2010(E)
Contents Page
Foreword ............................................................................................................................................................iv
Introduction.........................................................................................................................................................v
1 Scope ......................................................................................................................................................1
2 Normative references............................................................................................................................1
3 Terms and definitions ...........................................................................................................................1
4 Symbols and abbreviated terms ..........................................................................................................4
5 Generic description of fractional factorial designs ...........................................................................5
5.1 Overview of the structure of the examples in Annexes A to F .........................................................5
5.2 Overall objective(s) of the experiment (Step 1) ..................................................................................5
5.3 Response variable(s) (Step 2) ..............................................................................................................5
5.4 Factors affecting the response(s) (Step 3) .........................................................................................5
5.5 Select a fractional factorial design (Step 4)........................................................................................6
5.6 Analyse the results — Numerical summaries and graphical displays (Step 5) ..............................7
5.7 Present the findings (Step 6)................................................................................................................8
5.8 Perform confirmation runs (Step 7) .....................................................................................................8
6 Description of Annexes A to F .............................................................................................................8
6.1 Comparing and contrasting the examples..........................................................................................8
6.2 Experiment summaries .........................................................................................................................8
Annex A (informative) Direct mail marketing campaign ...............................................................................10
--`,,```,,,,````-`-`,,`,,`,`,,`---
Annex B (informative) Optimizing a polymer emulsion ................................................................................20
Annex C (informative) Insight into PVC foam formulations .........................................................................34
Annex D (informative) Process validation for an insulin product................................................................47
Annex E (informative) Washing machine experiment ...................................................................................60
Annex F (informative) Aggregated shipworm bacterium..............................................................................72
Bibliography......................................................................................................................................................80
© ISO 2010 – All rights reserved iii

Provided by IHS
ISO/TR 12845:2010(E)
Foreword
ISO (the International Organization for Standardization) is a worldwide federation of national standards bodies
(ISO member bodies). The work of preparing International Standards is normally carried out through ISO
technical committees. Each member body interested in a subject for which a technical committee has been
established has the right to be represented on that committee. International organizations, governmental and
non-governmental, in liaison with ISO, also take part in the work. ISO collaborates closely with the
International Electrotechnical Commission (IEC) on all matters of electrotechnical standardization.
International Standards are drafted in accordance with the rules given in the ISO/IEC Directives, Part 2.
The main task of technical committees is to prepare International Standards. Draft International Standards
adopted by the technical committees are circulated to the member bodies for voting. Publication as an
International Standard requires approval by at least 75 % of the member bodies casting a vote.
In exceptional circumstances, when a technical committee has collected data of a different kind from that
which is normally published as an International Standard (“state of the art”, for example), it may decide by a
simple majority vote of its participating members to publish a Technical Report. A Technical Report is entirely
informative in nature and does not have to be reviewed until the data it provides are considered to be no
longer valid or useful.
Attention is drawn to the possibility that some of the elements of this document may be the subject of patent
rights. ISO shall not be held responsible for identifying any or all such patent rights.
ISO/TR 12845 was prepared by Technical Committee ISO/TC 69, Applications of statistical methods,
Subcommittee SC 7, Applications of statistical and related techniques for the implementation of Six Sigma.
--`,,```,,,,````-`-`,,`,,`,`,,`---
iv © ISO 2010 – All rights reserved

Provided by IHS
ISO/TR 12845:2010(E)
Introduction
The Six Sigma 1 ) and international statistical standards communities share a philosophy of continuous
improvement and many analytical tools. The Six Sigma community tends to adopt a pragmatic approach
driven by time and resource constraints. The statistical standards community arrives at rigorous documents
through long-term international consensus. The disparities in time pressures, mathematical rigor, and
statistical software usage have inhibited exchanges, synergy and mutual appreciation between the two groups.
The present document takes one specific statistical tool (two-level fractional factorial design) and develops the
topic somewhat generically (in the spirit of International Standards) but then illustrates it through the use of six
detailed and distinct applications. The generic description focuses on the commonalities across the designs.
The annexes containing the six illustrations follow the basic framework but also identify the nuances and
peculiarities in the specific applications. Each example offers at least one “wrinkle” to the problem, which is
generally the case for real Six Sigma applications. It is thus hoped that practitioners can identify with at least
one of the six examples, if only to remind them of the basic material on fractional factorial designs that was
encountered during their Six Sigma training. Each of the six examples is developed and analysed using
statistical software of current vintage. The explanations throughout are devoid of mathematical detail – such
material can be readily obtained from the many design and analysis of experiments textbooks (such as those
given in the Bibliography).
1) Six Sigma is a trade mark of Motorola, Inc.
--`,,```,,,,````-`-`,,`,,`,`,,`---
© ISO 2010 – All rights reserved v

Provided by IHS
--`,,```,,,,````-`-`,,`,,`,`,,`---
Provided by IHS
TECHNICAL REPORT ISO/TR 12845:2010(E)
Selected illustrations of fractional factorial screening

experiments
1 Scope
This Technical Report describes the steps necessary to use and to analyse two-level fractional factorial
designs through illustration with six distinct applications of this methodology.
NOTE 1 Each of these six illustrations is similar in that resource constraints precluded the possibility of naively running
full factorial designs. Other commonalities among the six examples are noted [e.g. study objective, two levels for factors,
response variable(s), factors affecting the response]. On the other hand, the individual illustrations have some salient
features that are distinct.
NOTE 2 The examples suggest the spectrum of possibilities both in application area and in choice of fractional factorial
designs. Fractional factorial designs can be used to identify important factors for subsequent investigation (screening
design) and can in some cases provide a viable understanding of the process under study. Fractional factorial designs
include screening designs and designs that have been popularized by Genichi Taguchi.
NOTE 3 Fractional factorial experiments are sometimes employed by individuals (so-called “black belts” or “green
belts”) associated with Six Sigma methods. Six Sigma methods are concerned with problem solving and continuous
improvement. A fractional factorial experiment can be a cost-effective tool for obtaining timely improvements of processes
and products. Detailed discussions and treatment of other tools employed by Six Sigma practitioners can be identified in
various ISO/TC 69/SC 7 documents.
2 Normative references
The following referenced documents are indispensable for the application of this document. For dated
references, only the edition cited applies. For undated references, the latest edition of the referenced
document (including any amendments) applies.
ISO 3534-1, Statistics — Vocabulary and symbols — Part 1: General statistical terms and terms used in
probability
ISO 3534-2, Statistics — Vocabulary and symbols — Part 2: Applied statistics
ISO 3534-3, Statistics — Vocabulary and symbols — Part 3: Design of experiments
3 Terms and definitions

For the purposes of this document, the terms and definitions given in ISO 3534-1, ISO 3534-2, ISO 3534-3
and the following apply.
3.1
analysis of variance
ANOVA
technique which subdivides the total variation of a response variable into components associated with defined
--`,,```,,,,````-`-`,,`,,`,`,,`---
sources of variation
NOTE Adapted from ISO 3534-3:—2), definition 3.3.8. (The notes and example are not included here.)
2) To be published. (Revision of ISO 3534-3:1999)
© ISO 2010 – All rights reserved 1

Provided by IHS
ISO/TR 12845:2010(E)
3.2
binomial distribution
discrete distribution having the probability mass function
n!
P ( X = x) = p x (1 − p ) n − x
x !( n − x )!
where x = 0, 1, …, n and with indexing parameters n = 1, 2, …, and 0 < p < 1.
NOTE Adapted from ISO 3534-1:2006, definition 2.46. (The example and notes are not included here.)
3.3
block
collection of experimental units
NOTE Adapted from ISO 3534-3:—, 3.1.23. (The notes are not included here.)
3.4
centre point
vector of factor level settings of the form (a1, a2, ..., ak), where all ai equal 0, as notation for the coded levels of
the factors
NOTE Adapted from ISO 3534-3:—, 3.1.38. (The note and example are not included here.)
3.5
design matrix
matrix with rows representing individual treatments (possibly transformed according to the assumed model)
which can be extended by deduced levels of other functions of factor levels (interactions, quadratic terms,
etc.) but are dependent upon the assumed model
NOTE Adapted from ISO 3534-3:—, definition 3.2.24. (The notes are not included here.)
3.6
full factorial experiment
factorial experiment
designed experiment consisting of all possible treatments formed from two or more factors, each being
studied at two or more levels
3.7
interaction
combination of two or more factors
3.8
factor level
setting, value or assignment of a factor in accordance with the design region
NOTE Adapted from ISO 3534-3:—, definition 3.1.10. (The notes and examples are not included here.)
3.9
normal distribution
Gaussian distribution
continuous distribution having the probability density function
--`,,```,,,,````-`-`,,`,,`,`,,`---
2 © ISO 2010 – All rights reserved

Provided by IHS
ISO/TR 12845:2010(E)
( x −µ) 2
−
1 2σ 2
f ( x) = e
σ 2π
where −∞ < x < ∞ and with parameters −∞ < µ < ∞ and σ > 0
NOTE Adapted from ISO 3534-1:2006, definition 2.50. (The notes are not included here.)
3.10
predictor variable
factor
variable that can contribute to the explanation of the outcome of an experiment
NOTE 1 The extent to which a given predictor variable can be controlled dictates its potential role in a designed
experiment. Predictor variables can be controllable (fixed), modifiable (controllable only for short duration or at
considerable expense) or uncontrollable (random).
NOTE 2 A predictor variable can include a random element in it or it can, for example, be from a set of qualitative
classes which can be observed or assigned without random error.
NOTE 3 The term predictor variable is typically used in contexts involving the mathematical relationship among the
response variable and predictor variable(s) or functions of predictor variables. The term “factor” tends to be used
operationally as a means to assess the response variable as particular factors vary.
NOTE 4 A factor may be associated with the creation of blocks.
NOTE 5 “Independent variable” is not recommended as a synonym due to potential confusion with “independence”
(see ISO 3534-1:2006, 2.4). Other terms sometimes substituted for predictor variable include “input variable”, “descriptor
variable” and “explanatory variable”.
[ISO 3534-3:—, definition 3.1.4]
3.11
randomization
strategy in which each experimental unit has an equal chance of being assigned a particular treatment
3.12
replication
〈experiment〉 multiple occurrences of a given treatment combination or setting of predictor variables
--`,,```,,,,````-`-`,,`,,`,`,,`---
3.13
repetition
performance of an experiment more than once for a given set of predictor variables not necessarily with a
complete set-up of the predictor variables
3.14
split-plot design
experimental design in which the group of experimental units (“plot”) to which the same factor level assigned
to the principal factor is subdivided (“split”) so as to study one or more additional principal factors within each
level of that factor
NOTE Adapted from ISO 3534-3:—, definition 3.2.16. (The example and notes have not been included here.)
3.15
design resolution
〈design of experiments; fractional factorials〉 length of the shortest word in the defining relation

Provided by IHS
ISO/TR 12845:2010(E)
NOTE 1 Design resolution indicates the extent of aliasing among main effects and two-way and higher-order
interactions.
NOTE 2 The design resolution describes the aliasing in a particular experimental design. The numerical length is
generally given by upper case Roman numerals. The three most common practical situations are resolutions III, IV and V.
⎯ For a resolution III design, main effects are not aliased with other main effects. This observation can be made by
examining the expressions in the defining relation. For example, I = ABD = BCE = ACDE includes the expressions I,
ABD, BCE and ACDE and counting the number of letters for each term (1, 3, 3 and 4, respectively for this example).
The shortest length of these aside from 1 (corresponding to I) is 3, which is known as the length of the shortest
“character string”. At least one main effect is aliased with a two-way interaction. For example, for I = ABD it would be
the case that A is confounded with BD, B is confounded with AD and D is confounded with AB.
⎯ For a resolution IV design, main effects are not aliased with other main effects or with any two-way interactions. At
least one two-way interaction is aliased with another two-way interaction. For example, for the defining relation
I = ABCE = BCDF = ADEF includes the expressions I, ABCE, BCDF and ADEF and counting the number of letters for
each term (1, 4, 4 and 4, respectively). The smallest value aside from 1 (corresponding to I) is 4, which is known as
the length of the shortest “character string” for this design. For example, for I = ABCE, it would be the case that AB is
confounded with CE, AC with BE, AE with BC, while A is confounded with BCE, B is confounded with ACE, C is
confounded with ABE and E is confounded with ABC.
⎯ For a resolution V design, main effects and two-way interactions are not aliased with any other main effects or with
any other two-way interactions. For example, with I = ABCDE, it is clear that each main effect is confounded with a
four-way interaction (A is confounded with BCDE) and each two-way interaction is confounded with a three-way
interaction (AB is confounded with CDE).
NOTE 3 The higher the resolution, the more effects (main or interactions) can be estimated unambiguously, provided
the higher-order interactions are negligible. Given a choice of two potential designs involving the same number of factors
and experimental units, the design with the higher resolution should be selected. Fortunately, for most cases of k and p of
practical interest, the most appropriate defining relations are recorded.
NOTE 4 Full factorial designs have no confounding. For most practical purposes, a resolution V is excellent and a
resolution IV design may be adequate. Resolution III designs are useful as economical screening designs.
[ISO 3534-3:—, definition 3.2.6]
4 Symbols and abbreviated terms
--`,,```,,,,````-`-`,,`,,`,`,,`---
The main symbols and abbreviated terms used in this Technical Report are given below.
Y Response variable
A, B, C, D Factors
AB, AC, AD, BC, BD, CD Two-way interactions
ABC, ACD, BCD Three-way interactions
ABCD Four-way interactions
+1, −1 High and low settings
24 Four factors each at two levels
σ Standard deviation

Provided by IHS
ISO/TR 12845:2010(E)
5 Generic description of fractional factorial designs
5.1 Overview of the structure of the examples in Annexes A to F
This Technical Report provides general guidelines on the design, conduct and analyses of two-level fractional
factorial designs and illustrates the steps with six distinct applications given in the annexes. Each of the six
examples in Annexes A to F follows the basic structure as given in Table 1.
The steps given in Table 1 apply to design and analysis of screening experiments in general. Each of the
seven steps is explained below.
Table 1 — Basic steps in experimental design
Step 1 State the overall objective(s) of the experiment

Step 2 Describe the response variable(s)
Step 3 List the factors that might affect the response(s)
Step 4 Select a fractional factorial design
Step 5 Analyse the results – numerical summaries and graphical displays
Step 6 Present the findings
Step 7 Perform a confirmation run
5.2 Overall objective(s) of the experiment (Step 1)
Experiments are conducted for a variety of reasons. The primary motivation for the experiment should be
clearly stated and agreed to by all parties involved in the design, conduct, analysis and implications of the
experimental effort. There may be secondary objectives that could be addressed with this experiment but, in
general, the focus is on identifying a subset of the factors that impact the response variable.
The ultimate outcome of the experiment might be to take immediate action on factor levels or to obtain a
predictive model, both of which dictate some elements of the analyses. Further, the experiment could be
recognized as the first step in an ongoing sequence of experiments to determine improvements to the product
--`,,```,,,,````-`-`,,`,,`,`,,`---
or process.
5.3 Response variable(s) (Step 2)
Associated with the objective of an experiment is a measurable outcome or performance measure. A

response of interest could involve maximization (larger is better), minimization (smaller is better) or meet a
target value (be close to a specified value). The response variable (denoted here by the variable Y) should be
intimately (if not directly) related to the objective of the experiment. For some situations, there may be multiple
characteristics of interest to be considered, although there is typically a primary response variable associated
with the experiment. In other cases, multiple responses must be considered.
5.4 Factors affecting the response(s) (Step 3)
The response variable very likely depends in some unknown way on a variety of conditions. These could be
set in the course of generating a response variable outcome such as a temperature or a time setting. These
conditions are presumed to relate to controllable factors that may be continuous (temperature, concentration)
or discrete (two assembly lines A or B, two vendors, two packaging styles, and so forth). For the fractional
factorial experiments considered in this Technical Report, the experimental design process is simplified by
selecting two levels for each factor to be varied in the experiment.
⎯ For discrete factors, with only two possible settings, the levels are just these two settings.

Provided by IHS
ISO/TR 12845:2010(E)
⎯ For continuous factors, discretion can be used in choosing the two specific values. In some cases, one
setting could be the historical value and the other a proposed value. In other cases, the two settings could
be nominal adjustments from the historical setting. In any event, the settings should be sufficiently far
removed to have an opportunity to reveal an impact subject to the inherent uncertainty, while not being so
disparate that the settings are unreasonable from a practical, safety or sensibility standpoint. The setting
of levels of continuous factors benefits from an expert collaborating in the experiment.
There may be additional factors that could impact that response variable but may be deemed much less
important than the chosen factors or are too difficult or expensive to control. Finally, it is the case that the
factors are to be set independently from each other. It could, however, be discovered that the factors interact
(the setting of one factor impacts how a second factor affects the response).
5.5 Select a fractional factorial design (Step 4)
It is usual that the number of factors thought to be important to experiment with is large, e.g. 10 or more.
Designing an experiment to investigate all possible factors and their combinations would lead to excessively
large experiments. For example, in studying only eight factors at two levels each, the number of experimental
runs is 256. For three levels, this would be 6 561 runs! For most organizations, this would be a very
extravagant use of resources. Thus, the need is there to design experiments using fewer runs but still giving
the experimenter all of the important experimental results. Although this gives full information about the effects
of the factors, it also delivers information about higher-order factor combinations that are of minimal practical
utility. Consequently, a balance is sought to reduce the experimental effort without foregoing the information
about main factors of interest by designing a fractional experiment.
The examples provided in the annexes use 6 to 15 factors with a maximum of only 40 runs to identify the
critical factors.
A classical design consists of 16 runs obtained by considering all combinations of four factors with two
possible levels. See ISO/TR 29901 for a description. Table 3 provides the basic layout in a standard order for
ease of understanding. Each row of the table represents one set of experimental conditions that when run will
produce a value of the response variable Y. The four factors are designated as A, B, C and D. For an
individual factor, the level “−1” is the “low” setting, or one of the two levels if the factor is categorical. The level
“+1” is the “high” setting, or the other level of the categorical factor. The column “Y ” is a placeholder for the
response value once a run has occurred.
This will provide unambiguous information about the following factors and their interactions, namely:
A, B, C, D, AB, AC, AD, BC, BD, CD, ABC, ABD, ACD, BCD and ABCD
For most practical applications, the higher-order interactions, e.g. ABC, are often taken as unimportant and
can therefore, in a design, be discounted. This, therefore, becomes the way in which larger numbers of factors
can be incorporated into designs that are intended for full factor combinations, albeit for a smaller number of
factors.
It is imperative that the experimenter be aware of the exact pattern of main effects and the associated two-
way interactions created by this discounting of higher-order interactions. A careless selection may inhibit the
clear identification of the effects (be they main or two-way interactions). These aspects can be seen in the six
different examples contained in the annexes.
Table 2 provides a useful summary of the shortcomings of certain designs. The table describes the three most
common design resolutions. The examples in the annexes (see Table 4) refer to these resolutions.
6
--`,,```,,,,````-`-`,,`,,`,`,,`---
© ISO 2010 – All rights reserved
Provided by IHS
ISO/TR 12845:2010(E)
Table 2 — Selected design resolutions
Resolution Description
No main effects are aliased with any other main effect, but at least one main effect is aliased
III
with two-factor interactions.
No main effects are aliased with any other main effect or two-factor interactions, but at least
IV one two-factor interaction is aliased with other two-factor interactions and at least one main
effect is aliased with three-factor interactions.
No main effects or two-factor interactions are aliased with any other main effect or two-factor
V interactions, but at least one two-factor interaction is aliased with three-factor interactions
and at least one main effect is aliased with four-factor interactions.
The design choice ultimately rests with the aforementioned considerations as well as resource constraints.
Guidance of a general nature can be gleaned from the reference texts in the Bibliography. Increasingly,
software packages contain useful guides to design selection and algorithms for optimal design where the
problem is well defined.
Table 3 — Layout of a generic 24 full factorial design
Row
A B C D Y Run order
number
1 −1 −1 −1 −1 y1 6
2 +1 −1 −1 −1 y2 14
3 −1 +1 −1 −1 y3 4
4 +1 +1 −1 −1 y4 11
5 −1 −1 +1 −1 y5 9
6 +1 −1 +1 −1 y6 2
7 −1 +1 +1 −1 y7 3
8 +1 +1 +1 −1 y8 1
9 −1 −1 −1 +1 y9 8
10 +1 −1 −1 +1 y10 13
11 −1 +1 −1 +1 y11 7
--`,,```,,,,````-`-`,,`,,`,`,,`---
12 +1 +1 −1 +1 y12 10
13 −1 −1 +1 +1 y13 15
14 +1 −1 +1 +1 y14 16
15 −1 +1 +1 +1 y15 5
16 +1 +1 +1 +1 y16 12
5.6 Analyse the results — Numerical summaries and graphical displays (Step 5)
At the completion of the conduct of the experiment, the yi values would be replaced by the actual observed
responses. Many statistical software packages exist to produce output to aid in the understanding of the
results of the experiment. Of immediate concern is the determination of the impact of the factors individually
on the response variable. By the nature of fractional factorial designs, it is hoped, if not presumed, that the
main effects that are estimated are indeed attributable to the main factor although it is recognized that aliasing
could be taking place.

Provided by IHS
ISO/TR 12845:2010(E)
Much of the analysis that is conducted for 24 full factorials (see ISO/TR 29901) is applicable here with the
proviso that aliasing patterns must be recognized and properly evaluated. In particular, main effects plots,
interaction plots, Pareto diagrams, normal and/or half-normal probability plots are relevant instruments of
analysis. Although the emphasis of the analysis is likely to be graphical in its nature, additional supporting
documentation, e.g. effects table, should be given.
5.7 Present the findings (Step 6)
Almost certainly, the conclusions from a screening experiment will constitute a description of the next phase of
experimentation. Design of experiments is inherently sequential in nature and a screening experiment is
frequently the first phase of investigation and is often followed by subsequent experiments with fewer factors
and, sometimes, more levels. In some cases, the screening experiment supports a predictive model that can
suggest alternative settings for the important factors. Although the process has not been fully optimized,
superior settings may, in the meantime, have been identified.
Further, the experiment may reveal important interactions which may be, unfortunately, aliased with others. An
example of this can be found in Annex C.
5.8 Perform confirmation runs (Step 7)
Confirming experiments are considered to be good “practice” in that they should provide the experimenter
assurance that
a) the original experiment was correctly conducted,
b) the findings still hold true at a later point in time, and
c) there were no apparent “lurking” variables that were ignored in the original experimental design.
This being said, there are circumstances that preclude a confirmation run, e.g. time pressures due to
commercial issues or matters of cost (resources) or management acceptance of the risk of a faulty screening
experiment.
6 Description of Annexes A to F
6.1 Comparing and contrasting the examples
Six distinct examples of fractional factorial designs are illustrated in Annexes A to F. Each example follows the
general template given in Table 1.
6.2 Experiment summaries
Table 4 summarizes the six examples detailed in the annexes and indicates aspects of the analyses that were
unique to that experiment.
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
Table 4 — Experiment summaries found in annexes
Annex Experiment Business area Resolution Problem-specific aspects
A Direct mail campaign Marketing IV Proportion response variable, standard errors

based on binomial distribution.
B Polymer emulsion Chemical IV Blocking, response transformation, two
optimization engineering responses.
C PVC foam formulation Chemical III Analysis of three responses with different levels
engineering of success, due to severe confounding of
effects.
D Insulin process Pharmaceuticals IV Eight responses, blocking, factor levels
validation expressed as a range, interest in no effect as a
result.
E Washing machine Mechanical Not Taguchi orthogonal arrays, inner/outer arrays.
robustness engineering applicable
F Aggregated shipworm Bioengineering III, IV Plackett-Burman design.
bacterium
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
Annex A
(informative)
Direct mail marketing campaign
A.1 General
Mail material for a direct marketing campaign can be sent in a variety of packaging styles and with various
textual and graphical cues to encourage a high response rate. Historically, the company had experimented
with a small subset of their campaigns in which one aspect of the package would be varied. The company was
receptive to a more sophisticated approach than their previous “one factor at a time” style. Management and
the marketing firm that implemented the mailings in consultation with a statistician identified seven items that
could be varied in the package sent to potential customers and decided on the specific fractional factorial
design as described in the next subclauses.
A.2 Overall objective for the experiment

The objective of the experiment was to maximize magazine subscription response rate based on a direct
mailing campaign3).
A.3 Description of the process

A typical mailing involves solicitations to 400 000 potential subscribers. The actual package mailed can be
varied regarding outer envelope design and other features that had previously enhanced response rate. For
this study, the parent company opted to devote 40 000 packages to the experiment. The mailing itself is
subcontracted to a third party that specializes in assembling these packages.
A.4 Response variable
A.4.1 Choice of variable
The response variable is the number of people who subscribed and paid (either by credit card or cheque).
A.4.2 Measurement of the response variable

--`,,```,,,,````-`-`,,`,,`,`,,`---
Exactly 2 500 mailings went out for each combination of package contents. Subscribers were tracked to
determine which package of materials they had been sent.
3) This example has been adapted from the article “Using a Fractional Factorial Design to Increase Direct Mail
Response,” by J. Ledolter and A. J. Swersey (2006), Quality Engineering, 18, pp. 469-475. Used with permission from the
publisher, the American Society of Quality, Milwaukee, Wisconsin.

Provided by IHS
ISO/TR 12845:2010(E)
A.4.3 Relationship of the response variable to the objective of the experiment
A.4.3.1 General
The publisher expected a response rate of about 2 %. Further, the publisher wanted to be fairly certain to be
able to recognize a 0,5 % increase and at least a decent chance of detecting a 0,25 % increase in response
rate. The response rate is the critical factor in addressing the objective of the experiment as the package
variations have roughly the same preparation cost. Of secondary interest is the proportion of positive
responders who pay immediately as well, which avoids a later billing expense.
A.4.3.2 Description of each factor (continuous/discrete) to be varied
The factors chosen to be varied within the experiment were determined jointly by the publisher, the marketing
firm involved in preparing the mailing and the statistical consultant. The factors are described as follows:
⎯ presence/absence of an “Act now to respond/pay today” insert, a separate colour enclosure with the
encouragement presented in a bold font, a copy of the cover of the next issue and a description of its
contents; this insert also encourages prospective subscribers to pay at the same time as responding, to
preclude a follow-up mailing requesting payment for those who indeed “acted now”;
⎯ additional option of paying by credit card (rather than only by personal cheque);
⎯ offer wording strength; an additional phrase encouraging the potential subscriber to send in the reply card
and payment immediately in order to receive the next issue “hot off the press”;
⎯ inclusion or non-inclusion of a bumper sticker;
⎯ a guarantee that provides the opportunity to cancel the subscription and recuperate money for the issues
not yet received (one option) or for the full subscription price (other option);
⎯ testimonials that are inserts consisting of positive quotes from subscribers and possibly some celebrities;
⎯ presence/absence of a phrase on the outer envelope with mild profanity (“ballsy” versus “gutsy”).
Each of these factors is categorical with two levels. Some incremental costs to the mailing are incurred with
the variables that involve inserts.
A.4.3.3 Selection of levels (related to size of effect to be determined)
The factors used in the experiment and their associated levels are given in Table A.1.
Table A.1 — Factors
Factor Level 1 (−1) Level 2 (+1)
A: Act now No “Act now” insert “Act now” insert

B: Credit card No credit card Credit card
C: Offer hardness Hard offer Harder offer
D: Bumper sticker No bumper sticker Bumper sticker
E: Guarantee Guarantee partial refund Guarantee full refund
F: Testimonial No “Testimonial” insert “Testimonial” insert
G: Profanity “Gutsy” “Ballsy”
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
A.4.3.4 Other factors noted but not incorporated due to issues with controllability or relevance
No other factors were identified for the experiment.
A.4.4 Fractional factorial design
A.4.4.1 Choice of specific design
There are a multitude of possible designs that allow the investigation of the seven factors, each having two
levels. The possibilities depend on the number of combinations to be considered which in turn impacts the
number of runs considered. The publisher was willing to allocate a total of 40 000 mailings for the entire
experiment. The designs considered are given in Table A.2.
Table A.2 — Design choices
Number Fraction
Resolution Confounding patterns Comments
of runs size
None – All effects can be estimated Benefits from estimating higher-order

128 Full factorial Full
interactions are minimal
Main effects confounded with Less costly than a full factorial but
six-way interactions the additional benefits are minimal
64 Half fraction VII Two-way interactions not
confounded with other two-way
interactions
Main effects confounded with Slightly better confounding structure
three-way interactions than the 16-run case but at twice the
Quarter level of effort
32 IV Some two-way interactions
fraction
interactions
Main effects confounded with three- Excellent trade off between level of
way interactions effort and capability to estimate main
One-eighth effects free of two-way interactions
16 IV Some two-way interactions
fraction
interactions
Main effects confounded with two- Requires a follow-up experiment to
One- way interactions separate main effects from
8 sixteenth III confounding with two-way
Two-way interactions confounded
fraction interactions
with several other two-way
interactions
The two most viable designs were the 32-run one-quarter fraction and the 16-run one-eighth fraction, both of
which are resolution IV (namely, main effects can be estimated to be free of two-way interactions, but there
are some two-way interactions confounded with other two-way interactions). The larger experiments are
simply too big for the benefits accrued (ability to estimate higher-order interactions which are unlikely to be
significant). The small 8-run experiment would require a follow up experiment in order to resolve the two-way
interactions. The 16-run experiment was chosen since it was felt that, with appropriate labelling of factors, the
confounding structure could be accommodated. In particular, the factors A, B, C and D are set as if they would
constitute a 24 full factorial design, in which all 16 combinations of the two levels of these factors are
considered. The levels of the three additional factors are set according to E = ABC, F = BCD and G = ACD.
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
Based on subject matter expertise in direct mailing, the experimenters believed that factors E and F were
unrelated to each other and to the other factors (i.e. there should be no two-way interactions involving E and
F). Factor G conceivably could entice the recipient to open the envelope, but once the contents were in hand,
interaction between the slang expression on the envelope would not interact with the contents. In short, it was
assumed, at the design stage, that all of the following two-way interactions would be negligible: AE, AF, AG,
BE, BF, BG, CE, CF, CG, DE, DF, DG, EF, EG, FG.
The full confounding (alias) structure is given in Table A.3. Each of the main effects and plausible two-way
interactions (in bold typeface in Table A.3) can be unambiguously estimated under the assumptions given in
the previous paragraph.
Table A.3 — Confounding structure for the fractional factorial design
Generators: E = ABC, F = BCD, G = ACD
Defining relation: I = ABCE = BCDF = ACDG = ADEF = BDEG = ABFG = CEFG
Complete alias structure:

A = BCE = ABCDF = CDG = DEF = ABDEG = BFG = ACEFG
B = ACE = CDF = ABCDG = ABDEF = DEG = AFG = BCEFG
C = ABE = BDF = ADG = ACDEF = BCDEG = ABCFG = EFG
D = ABCDE = BCF = ACG = AEF = BEG = ABDFG = CDEFG
E = ABC = BCDEF = ACDEG = ADF = BDG = ABEFG = CFG
F = ABCEF = BCD = ACDFG = ADE = BDEFG = ABG = CEG
G = ABCEG = BCDFG = ACD = ADEFG = BDE = ABF = CEF
AB = CE = ACDF = BCDG = BDEF = ADEG = FG = ABCEFG
AC = BE = ABDF = DG = CDEF = ABCDEG = BCFG = AEFG
AD = BCDE = ABCF = CG = EF = ABEG = BDFG = ACDEFG
AE = BC = ABCDEF = CDEG = DF = ABDG = BEFG = ACFG
AF = BCEF = ABCD = CDFG = DE = ABDEFG = BG = ACEG
AG = BCEG = ABCDFG = CD =DEFG = ABDE = BF = ACEF
BD = ACDE = CF = ABCG = ABEF = EG = ADFG = BCDEFG
ABD = CDE = ACF = BCG = BEF = AEG = DFG = ABCDEFG
Alias structure restricted to two-way interactions (main effects are clear of two-way interactions):
--`,,```,,,,````-`-`,,`,,`,`,,`---
AB = CE == FG
AC = BE = DG
AD = CG = EF
AE = BC = DF
AF = DE = BG
AG = CD = BF
BD = CF = EG
As can be seen from Table A.3, each main effect and those two-way interactions likely to be of interest are
confounded only with non-interesting two-way or higher interactions.

Provided by IHS
ISO/TR 12845:2010(E)
A.4.4.2 Design matrix (standard order and run order)
The design selected was the 16-run, one-eighth fractional factorial design, with seven factors (A, B, C, D, E, F
and G). Table A.4 shows the design.
Table A.4 — Design matrix
Row
A B C D E (= ABC) F (= BCD) G (= ACD)
number
1 −1 −1 −1 −1 −1 −1 −1
2 1 −1 −1 −1 1 −1 1
3 −1 1 −1 −1 1 1 −1
4 1 1 −1 −1 −1 1 1
5 −1 −1 1 −1 1 1 1
6 1 −1 1 −1 −1 1 −1
7 −1 1 1 −1 −1 −1 1
8 1 1 1 −1 1 −1 −1
9 −1 −1 −1 1 −1 1 1
10 1 −1 −1 1 1 1 −1
11 −1 1 −1 1 1 −1 1
12 1 1 −1 1 −1 −1 −1
13 −1 −1 1 1 1 −1 −1
14 1 −1 1 1 −1 −1 1
15 −1 1 1 1 −1 1 −1
16 1 1 1 1 1 1 1
Randomization was used in selecting 2 500 addresses out of the 40 000 addresses constituting the sampling
frame. Considerable effort was expended to ensure that the third-party provider correctly assembled the
mailed packages. There was no need to randomize the order with respect to the design matrix.
--`,,```,,,,````-`-`,,`,,`,`,,`---
A.4.4.3 Centre points
Centre points were not selected for this experiment because the variables are categorical. Moreover, the
response rate can be assessed as a binomial random variable with a nominal response rate of 2 % so that an
independent estimate of the uncertainty can be calculated.
A.4.4.4 Replication and repetition
As mentioned in A.4.4.2, the mailings were sent to 2 500 potential subscribers. This experiment had neither
replication nor repetition.

Provided by IHS
ISO/TR 12845:2010(E)
A.4.5 Analysis of results
A.4.5.1 Data acquired through the experiment and analysis considerations
The results from the experiment are given in Table A.5. They were keyed into a software programme called
JMP4).
Table A.5 — Experiment results
Response
Row rate
A B C D E F G
number
%
1 −1 −1 −1 −1 −1 −1 −1 2,08
2 1 -1 −1 −1 1 −1 1 2,76
3 −1 1 −1 −1 1 1 −1 2,36
4 1 1 −1 −1 −1 1 1 3,04
5 −1 −1 1 −1 1 1 1 2,36
6 1 −1 1 −1 −1 1 −1 2,52
7 −1 1 1 −1 −1 −1 1 2,64
8 1 1 1 −1 1 −1 −1 2,64
9 −1 −1 −1 1 −1 1 1 2,40
10 1 −1 −1 1 1 1 −1 2,52
11 −1 1 −1 1 1 −1 1 3,24
12 1 1 −1 1 −1 −1 −1 2,12
13 −1 −1 1 1 1 −1 −1 2,12
14 1 −1 1 1 −1 −1 1 3,12
15 −1 1 1 1 −1 1 −1 1,96
16 1 1 1 1 1 1 1 3,20
The “Fit Model” in JMP generated the following results:
Singularity details
AE = BC = DF
AC = BE = DG
AG = BF = CD
AF = BG = DE
--`,,```,,,,````-`-`,,`,,`,`,,`---
AB = CE = FG
BD = CF = EG
AD = CG = EF
4) JMP is the trade name of a product supplied by SAS Inc. This information is given for the convenience of users of this
Technical Report and does not constitute an endorsement by ISO of the product named. Equivalent products may be used
if they can be shown to lead to the same results.

Provided by IHS
ISO/TR 12845:2010(E)
The confounding structure of the fractional factorial design is provided; it can be verified that it agrees with
what was provided in Table A.4. This information is necessary to understand the parameter estimate results
given in Table A.6.
Table A.6 — Parameter estimates
Biased
Term Estimate Standard error t ratio Prob > |t|
or zeroed
Intercept — 2,567 5 0,052 5 48,90 0,013 0
A — 0,172 5 0,052 5 3,29 0,188 1
B — 0,082 5 0,052 5 1,57 0,360 8
C — 0,002 5 0,052 5 0,05 0,969 7
D — 0,017 5 0,052 5 0,33 0,795 2
E — 0,082 5 0,052 5 1,57 0,360 8
F — −0,022 5 0,052 5 −0,43 0,742 2
G — 0,277 5 0,052 5 5,29 0,119 0
AB Biased −0,072 5 0,052 5 −1,38 0,399 0
AC Biased 0,127 5 0,052 5 2,43 0,248 7
AD Biased −0,017 5 0,052 5 −0,33 0,795 2
AE Biased −0,042 5 0,052 5 −0,81 0,566 8
AF Biased 0,102 5 0,052 5 1,95 0,301 3
AG Biased 0,012 5 0,052 5 0,24 0,851 2
BC Zeroed 0 0 — —
BD Biased −0,037 5 0,052 5 −0,71 0,605 1
BE Zeroed 0 0 — —
BF Zeroed 0 0 — —
BG Zeroed 0 0 — —
CD Zeroed 0 0 — —
CE Zeroed 0 0 — —
CF Zeroed 0 0 — —
--`,,```,,,,````-`-`,,`,,`,`,,`---
CG Zeroed 0 0 — —
DE Zeroed 0 0 — —
DF Zeroed 0 0 — —
DG Zeroed 0 0 — —
EF Zeroed 0 0 — —
EG Zeroed 0 0 — —
FG Zeroed 0 0 — —
With 16 combinations of the factors, 15 parameters are estimated — the intercept, seven main effects and
seven sets of two-way interactions. At this point, the statistical package has not been “informed” of the
nature of the response, so that the calculations proceed as if the response were a single value rather than
the proportion of positive responses out of 2 500 attempts. Each effect can be estimated by the average
response rate for the effect at the high level minus the response rate for the effect at its low level. Since the
overall response rate is 2,57 %, the standard error of each of these estimated effects is
4 ( 2,57 )(100 − 2,57 ) 40 000 = 0,158 %. Two-way interactions can also be estimated but as has been noted,
they are confounded with other two-way interactions.

Provided by IHS
ISO/TR 12845:2010(E)
Table A.7 — Estimation of effects
Terma Coefficient Effect Standard error z ratio Prob > |z|

A 0,172 5 0,345 0,158 2,18 0,029
B 0,082 5 0,165 0,158 1,04 0,297
C 0,002 5 0,005 0,158 0,03 0,975
D 0,017 5 0,035 0,158 0,22 0,825
E 0,082 5 0,165 0,158 1,04 0,297
F −0,022 5 −0,045 0,158 −0,28 0,776
G 0,277 5 0,555 0,158 3,51 0,000
AB + CE + FG −0,072 5 −0,145 0,158 −0,92 0,359
AC + BE + DG 0,127 5 0,255 0,158 1,61 0,107
AD + CG + EF −0,017 5 −0,035 0,158 −0,22 0,825
AE + BC + DF −0,042 5 −0,085 0,158 −0,54 0,591
AF + DE + BG 0,102 5 0,205 0,158 1,30 0,195
AG + CD + BF 0,012 5 0,025 0,158 0,16 0,874
BD + CF + EG −0,037 5 −0,075 0,158 −0,47 0,636
a The most plausible of the interactions in each of the three term sums is in bold.
In Table A.7, the z ratio is the effect divided by the corrected standard error for each term as computed in
Table A.6. The expression “Prob > |z|” refers to the probability that a standard normal variable is greater than z
or less than −z.
The results in Table A.7 indicate the significance of factors A and G and a marginal AC interaction. The
coefficients were generated in JMP and the results stored in a data sheet. The z ratio and Prob > |z| were
calculated using the formula feature.
In addition, the probability plot in Figure A.1 indicates potentially significant factors. However, the significance
--`,,```,,,,````-`-`,,`,,`,`,,`---
is based on a separate calculation that is not pertinent to the binomial response.
Y
6
G
4
A
2
-2
-4
-6
-3 -2 -1 0 1 2 3 X
Key
X normal quantile
Y t ratio
The blue line is Lenth's PSE (pseudo standard error), from the estimates population.
The red line is the RMSE, i.e. root mean squared error, from the residual.
Figure A.1 — Probability plot

Provided by IHS
ISO/TR 12845:2010(E)
A.4.5.2 Pareto chart of effects
The relative size of the effects can be seen in the Pareto chart shown in Figure A.2.
Term t Ratio
G 5,285 714
A 3,285 714
A*C 2,428 571
A*F 1,952 381
E 1,571 429
B 1,571 429
A*B -1,380 952
A*E -0,809 524
B*D -0,714 286
F -0,428 571
A*D -0,333 333
D 0,333 333
A*G 0,238 095
C 0,047 619
Figure A.2 — Pareto chart of estimates
A.4.5.3 Estimation of experimental error and standard error of effects
Each effect is estimated by the average response rate for the effect at the high level minus the response rate
for the effect at its low level. Since the overall response rate is 2,57 %, the standard error of each estimated
effect is √[4(2,57)(100 −2,57)/ 40 000] = 0,158 %.
A.4.5.4 Effects plots
Figure A.3 shows the main effects of the factors. The relative slopes of the lines are consistent with the
significance of the factors G and A.
±0,667 076
3
2,567 5
2,5
--`,,```,,,,````-`-`,,`,,`,`,,`---
y
2
1,5
1,5
1
1
1,5
1,5
1,5
1
1
1,5
1
1,5
0
1
0
0
0,5
0,5
0
0,5
0,5
0,5
0
0,5
0,5
-1,5
-1,5
-1,5
-1,5
-1,5
-1
-1
-1,5
-1
-1
-1
-1,5
-1
-0,5
-0,5
-1
-0,5
-0,5
-0,5
-0,5
-0,5
0 0 0 0 0 0 0
A B C D E F G
Figure A.3 — Effects plots

Provided by IHS
ISO/TR 12845:2010(E)
A.4.5.5 Interaction plots
Figure A.4 shows the interaction plots between the factors. The most noteworthy interaction is that between
factors A (act now) and C (offer hardness).
3,25 1
1 1 1 1 1 -1
2,75
A -1 -1 -1
-1
A
y
-1
2,25
1,75
3,25
1
2,75 -1 1 1 1 1 -1
B -1 -1 -1
y
-1
B
2,25
1,75
3,25
1 -1
2,75 -1 -1 -1 -1 -1
C
C
y
2,25
1,75
3,25
-1 -1
2,75 -1 -1 -1 -1
D
y
D
2,25
1,75
3,25
1 1
2,75 1 1 1 1 -1
-1 -1 -1 -1 E -1
E
y
2,25
1,75
--`,,```,,,,````-`-`,,`,,`,`,,`---
3,25
1 -1
2,75 -1 -1 -1 -1
-1 F
F
y
2,25
1,75
3,25 1 1
1 1 1 1
2,75 G
-1 -1 -1
G
y
2,25 -1 -1 -1
1,75
0
0
0
0
1
1
-1
1
-1
-1
-1
1
-1
-1
0
-1
Figure A.4 — Interaction profiles
The interaction plots affirm the presence of an AC interaction. It appears that the “Act now” insert in
conjunction with the harder offer wording was synergistically beneficial.
A.4.6 Presentation of findings
The experiment confirmed the benefits of using the word “ballsy” rather than the less vitriolic word “gutsy”. The
“Act now” insert was supported, particularly when combined with the hard phrasing of the offer. The payment
option, although not significant as a factor, led to a higher level of immediate payment. Finally, the bumper
sticker and testimonial factors did not generate significant benefits, while incurring some costs. Likewise, the
stronger guarantee did not generate appreciable improvement in response while having the possibility of
costing the publisher additional money. Thus, it was decided not to include these options in the future.
A.4.7 Confirmation results
Subsequent mailings confirmed the benefits of offering the credit-card-payment option in addition to the
personal-cheque option.

Provided by IHS
ISO/TR 12845:2010(E)
Annex B
(informative)
Optimizing a polymer emulsion
B.1 General
Overprint varnishes (OPV) and flexographic inks in the graphic arts business make use of a polymer emulsion
that provides flexibility, water resistance and gloss. Production of this emulsion involves the importation of a
resin from abroad before manufacturing. Production operators experience difficulties meeting lot-to-lot
requirements for emulsion viscosity and particle size, as well as keeping production costs low, since a
competitor is able to source this resin locally at lower prices. Documentation on the existing process was
limited as the product was acquired during a merger of companies.
B.1.1 Overall objective for the experiment
The goal of this experimental work was to quickly identify a solution to the reaction problem by producing a
polymer emulsion with specific viscosity and particle size characteristics while maintaining other product
properties and characteristics5).
B.1.2 Description of the process
Seven reaction parameters were manipulated to understand how these changes influence viscosity and
particle size of the polymer. The chemists wanted to identify the primary drivers (main effects) in this emulsion
reaction and to control (block) for variability due to operator-to-operator and vessel-to-vessel differences. At
this time, quantifying interactions is not important, but they must not interfere with the determination of main
effects.
B.1.3 Response variable
B.1.3.1 Choice of response variables
--`,,```,,,,````-`-`,,`,,`,`,,`---
Viscosity and particle size were the primary responses investigated in the experiment.
B.1.3.2 Measurement of the response variables
Viscosity was measured in centipoise (cps) using a Brookfield viscometer, using spindle #3 at 60 r/min.
Particle size measurements were measured in nanometres (nm) using a Brookhaven BI 90 particle sizer.
B.1.3.3 Relationship of the response variables to the objective of the experiment
Viscosity and particle size are primary monitoring parameters used during production of the flexographic
emulsion. Viscosity is monitored to assure end-use flow in printing conditions. Particle size impacts the clarity
of the final coating on a printed surface.
5) This example has been kindly donated by Rohm and Haas (Research Center, Pennsylvania), a wholly owned
subsidiary of the Dow Chemical Company.

Provided by IHS
ISO/TR 12845:2010(E)
B.1.4 Factors affecting the response
B.1.4.1 Description of each factor (continuous/discrete) to be varied
The factors chosen to be varied within the experiment and their units of measurement are described, as
follows:
⎯ resin level (%),
⎯ kettle initiator (%),
⎯ co-feed initiator (%),
⎯ process temperature (°C),
⎯ sodium sulfate level (%),
⎯ feed time (minutes),
⎯ Disponil6) level (%).
Each of these factors is continuous but for the purpose of this preliminary experiment, only two levels of each
factor were used.
B.1.4.2 Selection of levels (related to size of effect to be determined)
The factors used in the experiment and their associated levels are given in Table B.1.
Table B.1 — Controlled factors

A: Resin level 26,50 32,50
B: Kettle initiator 0,48 0,72
C: Co-feed initiator 0,48 0,72
D: Process temperature 80,00 86,00
E: Sodium sulfate level 0,00 0,50
F: Feed time 144,00 216,00
G: Disponil level 0,48 0,96
B.1.4.3 Other factors, external to the experiment, that needed to be accounted for
Variability due to both operator-to-operator and vessel-to-vessel differences was believed to be non-negligible
and needed to be isolated. Thus, the design was run in four blocks, where each block included experiments
run under similar conditions, a given single operator and a given single vessel.
The associated levels for the blocking factors are given in Table B.2.
6) Disponil is the trade name of a product supplied by Cognis. This information is given for the convenience of users of
this Technical Report and does not constitute an endorsement by ISO of this product.
--`,,```,,,,````-`-`,,`,,`,`

Provided by IHS
ISO/TR 12845:2010(E)
Table B.2 — Blocked factors
Operator Tara Aaron

Vessel type Small hood Large hood
B.1.5 Fractional factorial design
B.1.5.1 Choice of specific design
The primary limitation in this experiment was time. A workable solution to the reaction problems was needed
quickly to secure customers and turn them away from the competition. A maximum of 16 experiments could
be accomplished in the available time. It was agreed that estimation of the interactions would be ignored at
this stage and that, if needed, a follow-up experiment to refine the recommendation would be performed.
The experimental design chosen to fit all of these requirements is a blocked fractional factorial design of
7 −3
resolution IV in 16 runs, denoted 2IV .
The full confounding (alias) structure is given in Table B.3.
Each of the main effects is confounded with a few three-term interactions, which the experimenter was willing
to assume to be negligible. Aside from the block effects, only four “clusters” of two-term interactions can be
uniquely estimated, a limitation acceptable to the experimenter, given the context explained above.
Table B.3 — Confounding structure for the blocked fractional factorial design
Factor generators: E = ABC, F = ABD, G = ACD

Block generators: 1 = AB, 2 = AC
Defining relation: I = ABCE = ABDF = ACDG = AEFG = BCFG = BDEG = CDEF
Complete alias structure:
[Estimated terms] Aliased terms
[A] A + BCE + BDF + CDG + EFG
[B] B + ACE + ADF + CFG + DEG
[C] C + ABE + ADG + BFG + DEF
[D] D + ABF + ACG + BEG + CEF
[E] E + ABC + AFG + BDG + CDF
[F] F + ABD + AEG + BCG + CDE
[G] G + ACD + AEF + BCF + BDE
[AD] AD + BF + CG
[AF] AF + BD + EG
[AG] AG + CD + EF
[BG] BG + CF + DE
[ABG] ABG + ACF + ADE + BCD + BEF + CEG + DFG
[Intercept] Intercept
--`,,```,,,,````-`-`,,`,,`,`,,`---
Block generator – These effects are lost to blocks

[1] = AB + CE + DF + ACFG + ADEG + BCDG + BEFG
[2] = AC + BE + DG + ABFG + ADEF + BCDF + CEFG
[3] = AE + BC + FG + ABDG + ACDF + BDEF + CDEG

Provided by IHS
ISO/TR 12845:2010(E)
B.1.5.2 Design matrix
The design selected was the 16-run one-eighth fractional factorial design with seven factors (A, B, C, D, E, F
and G). Table B.4 shows the design.
Table B.4 — Design matrix
Row
Blocked factors A B C D E (= ABC) F (= ABD) G (= ACD)
number
1 Tara/Large hood 26,5 0,72 0,72 80 0 216 0,96

2 Tara/Large hood 32,5 0,48 0,48 86 0,5 144 0,48
3 Tara/Large hood 26,5 0,72 0,72 86 0 144 0,48
4 Tara/Large hood 32,5 0,48 0,48 80 0,5 216 0,96
5 Aaron/Small hood 26,5 0,48 0,72 80 0,5 144 0,96
6 Aaron/Small hood 26,5 0,48 0,72 86 0,5 216 0,48
7 Aaron/Small hood 32,5 0,72 0,48 86 0 216 0,48
8 Aaron/Small hood 32,5 0,72 0,48 80 0 144 0,96
9 Tara/Small hood 32,5 0,48 0,72 80 0 216 0,48
10 Tara/Small hood 26,5 0,72 0,48 86 0,5 144 0,96
11 Tara/Small hood 32,5 0,48 0,72 86 0 144 0,96
12 Tara/Small hood 26,5 0,72 0,48 80 0,5 216 0,48
13 Aaron/Large hood 32,5 0,72 0,72 86 0,5 216 0,96
14 Aaron/Large hood 26,5 0,48 0,48 80 0 144 0,48
15 Aaron/Large hood 26,5 0,48 0,48 86 0 216 0,96
16 Aaron/Large hood 32,5 0,72 0,72 80 0,5 144 0,48
Randomization was used between each block and then within each block.
B.1.5.3 Centre points
Centre points were not selected for this experiment because of the time constraints.
B.1.5.4 Replication and repetition
Replication and repetition were not selected for this experiment because of the time constraints.
B.1.6 Analysis of results
B.1.6.1 Data acquired through the experiment and analysis considerations

--`,,```,,,,````-`-`,,`,,`,`,,`---
The results from the experiment for both viscosity and particle size are given in Table B.5. They were keyed
into a software programme called Design-Expert7).
7) Design-Expert is the trade name of a product supplied by Stat-Ease, Inc. This information is given for the convenience
of users of this Technical Report and does not constitute an endorsement by ISO of the product named. Equivalent
products may be used if they can be shown to lead to the same results.

Provided by IHS
ISO/TR 12845:2010(E)
Table B.5 — Experiment results
Particle
Row Blocked Viscosity ln(viscosity)
A B C D E (= ABC) F (= ABD) G (= ACD) size
number factors
cps cps nm
Tara/
1 26,5 0,72 0,72 80 0 216 0,96 385 5,953 2 97
Large hood
Tara/
2 32,5 0,48 0,48 86 0,5 144 0,48 1 185 7,077 5 103
Large hood
Tara/
3 26,5 0,72 0,72 86 0 144 0,48 445 6,098 1 97
Large hood
Tara/
4 32,5 0,48 0,48 80 0,5 216 0,96 3 075 8,031 1 99
Large hood
Aaron/
5 26,5 0,48 0,72 80 0,5 144 0,96 252 5,529 4 116
Small hood
Aaron/
6 26,5 0,48 0,72 86 0,5 216 0,48 694 6,542 5 96
Small hood
Aaron/
7 32,5 0,72 0,48 86 0 216 0,48 11 700 9,367 3 102
Small hood
Aaron/
8 32,5 0,72 0,48 80 0 144 0,96 1 946 7,573 5 100
Small hood
Tara/
9 32,5 0,48 0,72 80 0 216 0,48 3 300 8,101 7 108
Small hood
--`,,```,,,,````-`-`,,`,,`,`,,`---
Tara/
10 26,5 0,72 0,48 86 0,5 144 0,96 612 6,416 7 98
Small hood
Tara/
11 32,5 0,48 0,72 86 0 144 0,96 3 840 8,253 2 98
Small hood
Tara/
12 26,5 0,72 0,48 80 0,5 216 0,48 1 282 7,156 2 100
Small hood
Aaron/
13 32,5 0,72 0,72 86 0,5 216 0,96 4 740 8,463 8 100
Large hood
Aaron/
14 26,5 0,48 0,48 80 0 144 0,48 187 5,231 1 141
Large hood
Aaron/
15 26,5 0,48 0,48 86 0 216 0,96 388 5,961 0 96
Large hood
Aaron/
16 32,5 0,72 0,72 80 0,5 144 0,48 3 061 8,026 5 93
Large hood
B.1.6.2 Emulsion viscosity analysis
Viscosity measurements often span several orders of magnitude. In this situation, a transformation of the data
is useful for stabilizing the variance of the residuals. Here, the response “viscosity” has been transformed
using natural logarithms and is subsequently denoted “ln(viscosity)”.
Normal probability plots of the estimate effects in the log-transformed viscosity data indicate that resin level,
kettle initiator and feed time are important. These plots are shown in Figure B.1.

Provided by IHS
ISO/TR 12845:2010(E)
99
97
95 A
90
85 F
80
B
70
60
--`,,```,,,,````-`-`,,`,,`,`,,`---
40
20
0,00 0,50 1,00 1,50 2,00 X

Key
X effect
Y half-normal % probability
A resin level
B kettle initiator
F feed time
Figure B.1 — Probability plot for ln(viscosity)
A linear model containing the resin level, kettle-initiator level and feed-time effects explains 93 % of the
observed variability in the ln(viscosity) data. This model is significant and all terms in this linear model are
significant. The analysis of variance is shown in Table B.6.
Table B.6 — Analysis of variance for ln(viscosity)
Sum Degrees
Source Mean square F value Prob > F
of squares of freedom
Block 1,19 3 0,40

Model 18,99 3 6,33 38,91 < 0,000 1 significant
A 16,01 1 16,01 98,46 < 0,000 1
B 1,17 1 1,17 7,20 0,025 1
F 1,80 1 1,80 11,08 0,008 8
Residual 1,46 9 0,16
Cor Totala 21,64 15

a Cor Total = Totals of all the information corrected for the mean.

Provided by IHS
ISO/TR 12845:2010(E)
Model coefficient estimates, their standard errors and 95 % confidence intervals (CI) for the true coefficient
are given in Table B.7. This information does not apply for blocks since block effects are random by definition.
The purpose of blocking is solely to increase the power of detecting experimental factor effects.
Table B.7 — Estimation of effects
Coefficient Degrees 95 % CI 95 % CI
Factor Standard error
estimate of freedom Low High
Intercept 7,11 1 0,10 6,88 7,34

Tara/Large hood −0,32 3
Aaron/Large hood 0,14
Aaron/Small hood 0,37
Tara/Small hood −0,19
A: Resin level 1,00 1 0,10 0,77 1,23
B: Kettle initiator 0,27 1 0,10 0,042 0,50
F: Feed time 0,34 1 0,10 0,11 0,56
The final equation in terms of coded factors is:
ln(viscosity) = 7,11 + 1,00A + 0,27B + 0,34F
The predicted effects are shown in Figure B.2. To produce these plots, the factors not shown are set at their
mid-level values.
The slopes of the lines are consistent with the significance of the factors.
26 --`,,```,,,,````-`-`,,`,,`,`,,`---
Provided by IHS
ISO/TR 12845:2010(E)
Y Y
9,367 9,367
8,333 8,333
7,299 7,299
6,265 6,265
5,231 5,231
26,50 28,00 29,50 31,00 32,50 X1 0,48 0,54 0,60 0,66 0,72 X2
a) A: Resin level b) B: Kettle initiator
Y
9,367
8,333
--`,,```,,,,````-`-`,,`,,`,`,,`---
7,299
6,265
5,231
144,00 162,00 180,00 198,00 216,00 X3
c) F: Feed time
Actual factors
Factor
Key Figure B.2 a) Figure B.2 b) Figure B.2 c)
X1 A: resin level A: resin level: — 29,50 % 29,50 %
X2 B: kettle initiator B: kettle initiator: 0,60 % — 0,56 %
X3 F: feed time C: co-feed initiator: 0,60 % 0,60 % 0,60 %
Y ln(viscosity) D: process temperature: 83,00 °C 83,00 °C 83,00 °C
E: sodium sulfate level: 0,25 % 0,25 % 0,25 %
F: feed time: 180,00 min 180,00 min —
G: Disponil level: 0,72 % 0,72 % 0,72 %
Figure B.2 — One-factor plots of A, B and F — Effects on viscosity plotted in transformed scale

Provided by IHS
ISO/TR 12845:2010(E)
The predicted effects can also be plotted in the original viscosity scale. They are represented in Figure B.3. It
can be seen that the logarithmic transformation of viscosity allows for a simple way to fit nonlinear effects, as
indicated in the plot of the effect of A.
Y Y
11 700,00 11 700,00
8 821,75 8 821,75
5 943,50 5 943,50
3 065,25 3 065,25
167,00 167,00
26,50 28,00 29,50 31,00 32,50 X1 0,48 0,54 0,60 0,66 0,72 X2
a) A: Resin level b) B: Kettle initiator
Y
11 700,00
8 821,75
5 943,50
3 065,25
167,00
144,00 162,00 180,00 198,00 216,00 X3
c) F: Feed time
Key
X1 A: resin
X2 B: kettle initiator
X3 F: feed time
Y viscosity
Figure B.3 — One-factor plots for A, B and F — Effects on viscosity plotted in original scale
B.1.6.3 Particle size analysis
The results from the analysis of the particle size data are not as apparent. The normal probability plots, shown
in Figure B.4, suggest that kettle initiator (B) and process temperature (D) are important, as well as two
apparent interactions, AF and AD. These interactions are not directly interpretable as they represent
confounded clusters of two-factor interactions.
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
Y2
Y1
99
99
95
97 90
AF
95 AF 80
70 AD
90
85 50
AD
80
B 30
70
D 20
60 D
10
40
5 B
20
1
0
0,00 2,44 4,88 7,31 9,75 X -8,75 -4,13 0,50 5,13 9,75 X
a) Half-normal plot b) Normal plot
Key
X effect
Y1 half-normal % probability
Y2 normal % probability
A resin level
B kettle initiator
D process temperature
E sodium sulfate level
F feed time
Figure B.4 — Probability plots for particle size
A linear model containing kettle initiator (B), process temperature (D) and two sets of interactions explains
81 % of the observed variability in particle size. This model is marginally significant, as are the effects of B and
D. The analysis of variance is shown in Table B.8. This model is not good enough to be used as a basis to get
predicted particle sizes, but the information obtained about the effects of kettle initiator and process
temperature are valuable and may be used to improve the production process.
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
Table B.8 — Analysis of variance for particle size
Sum Degrees
Block 161,00 3 53,67

Model 1 483,00 6 247,17 4,15 0,053 4
A 90,25 1 90,25 1,52 0,264 2
B 306,25 1 306,25 5,15 0,063 8
D 256,00 1 256,00 4,30 0,083 4
F 144,00 1 144,00 2,42 0,170 8
AD 306,25 1 306,25 5,15 0,063 8
AF 380,25 1 380,25 6,39 0,044 8
Residual 357,00 6 59,50
Cor Totala 2 001,00 15

a Cor Total = Totals of all the information corrected for the mean.
Model coefficient estimates, their standard errors and 95 % confidence intervals for the true coefficient are
given in Table B.9. This information does not apply for blocks since block effects are random by definition. The
purpose of blocking is solely to increase the power of detecting experimental factor effects.
Table B.9 — Estimation of effects
Coefficient Degrees 95 % CIa 95 % CI

Factor Standard error VIFb
Intercept 102,75 1 1,93 98,03 107,47

Tara/Large hood −3,75 3
Aaron/Small hood 0,75
Tara/Small hood −1,75
Aaron/Large hood 4,75
A: Resin level −2,38 1 1,93 −7,09 2,34 1,00
B: Kettle initiator −4,38 1 1,93 −9,09 0,34 1,00
D: Process temp. −4,00 1 1,93 −8,72 0,72 1,00
F: Feed time −3,00 1 1,93 −7,72 1,72 1,00
AD 4,37 1 1,93 −0,34 9,09 1,00
AF 4,88 1 1,93 0,16 9,59 1,00
a CI = Confidence interval.
b VIF = Variation inflation factor.
The final equation in terms of coded factors is:
particle size = 102,75 −2,38A −4,38B −4,00D −3,00F +4,37AD +4,88AF
Note that the term 4,88AF could also be 4,88BD. The predicted effects are shown in Figure B.5. To produce
these plots, all factors not shown are set at their mid-levels.
30
--`,,```,,,,````-`-`,,`,,`,`,,`---
Provided by IHS
ISO/TR 12845:2010(E)
Y Y
141 141
129 129
117 117
105 105
93 93
0,48 0,54 0,60 0,66 0,72 X1 80,00 81,50 83,00 84,50 86,00 X2
a) B: Kettle initiator b) D: Process temperature
Actual factors
Factor
Key Figure B.5 a) Figure B.5 b)
X1 B: kettle initiator A: resin level: 29,50 % 29,50 %
X2 D: process temperature B: kettle initiator: — 0,65 %
Y particle size C: co-feed initiator: 0,60 % 0,60 %
D: process temperature: 83,00 °C —
E: sodium sulfate level: 0,25 % 0,25 %
F: feed time: 180,00 min 180,00 min
G: Disponil level: 0,72 % 0,72 %
Figure B.5 — One-factor plots for B and D — Effects on particle size
B.1.6.4 Optimization recommendations
To get good flexibility and gloss properties, this polymer requires viscosity to be in the range 450 cps to
650 cps with a desired target of 550 cps and particle size to be in the range 90 µm to 100 µm with a desired
target of 95 µm. Using the software, desirability step functions were defined for each response, as
represented in Figure B.5.
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
Figure B.6 — Desirability curves (step functions) for in(viscosity) and particle size
The result of the optimization search process is a mapping of the overall joint desirability. Joint desirability
ranges between 0 and 1 are obtained. A value of 0 is obtained if at least one of the response ranges is not
achievable. A value of 1 is obtained if, and only if, all ranges are achieved and the exact target values are
achieved. Table B.10 presents the results of eight optimization searches showing the levels of each process
variable and the corresponding joint desirability values.
Table B.10 — Actual process conditions and joint desirability values obtained in eight optimization
searches
--`,,```,,,,````-`-`,,`,,`,`,,`---
Process Sodium
Resin Kettle Co-feed Feed Disponil Particle
a tempera- sulfate Viscosity Desi-
Run level initiator initiator time levela size
ture levela rability
% % % °C % minutes % cps nm
1 26,81 0,56 0,66 85,75 0,44 199,40 0,56 550 95,000 5 1

2 26,79 0,61 0,52 85,92 0,01 188,91 0,86 550 94,999 2 1
3 26,71 0,67 0,62 85,97 0,35 177,62 0,62 550 95,001 1
4 26,52 0,62 0,68 85,16 0,45 195,33 0,92 550 94,999 1 1
5 26,72 0,53 0,50 85,32 0,31 209,21 0,63 550 94,998 1 1
6 26,72 0,67 0,70 86,00 0,01 177,42 0,95 550 94,999 9 1
7 26,64 0,72 0,50 86,00 0,42 167,39 0,73 550 95,148 0,996
8 26,77 0,50 0,56 84,35 0,02 216,00 0,96 550 97,592 5 0,912
a Has no effect on optimization results.
B.1.7 Presentation of findings
This study indicates that emulsion viscosity is positively influenced by resin level, reaction feed time and kettle
initiator level. Emulsion particle size is negatively influenced by kettle initiator level and process temperature.
Preliminary use of the models developed here suggests that targeted values of emulsion viscosity and
emulsion particle size can be achieved.

Provided by IHS
ISO/TR 12845:2010(E)
Variability due to operator differences and reactor hood difference was initially thought to be large and
potential lurking variables. Blocking techniques were used to absorb this variation and improve the power of
detecting small effects in the experiment. After completing the experiment, operator and reactor hood
differences were found to be small and not considered in later experimentation. This was an unexpected but
valuable piece of information. When differences in operator technique or reactor vessels exist, blocking serves
as an active method of statistical control on the magnitude of the error term.
The results of the optimization exercise reported above will be validated. This involves running additional
formulations at the recommended location and assessing the quality of the predictions relative to observed
results. Alternatively, a smaller number of factors can be chosen for investigation in an experimental region
defined by a reduced set of experimental ranges. The latter approach would yield better information about the
aptness of the first order planar model fit to the data. If interactions are important in this system, an
experimental design in a new region could help quantify the effects of the interaction and obtain better
recommendations for validation runs.
--`,,```,,,,````-`-`,,`,,`,`,,`---
B.1.8 Confirmation results
A set of confirmation experiments were completed to validate the results of the factorial experiment. The
scale-up team confirmed the utility of the optimization results and quickly scaled an initial product to
commercial quantities. Similar work was performed successfully by the scale-up team on other members of
the flexographic emulsions. The family of flexographic emulsions described here were successfully transferred
to the North American production facilities. The product line was successful in expanding its market leading
position in spite of higher raw material prices.

Provided by IHS
ISO/TR 12845:2010(E)
Annex C
(informative)
Insight into PVC foam formulations8)
C.1 General
This study discusses the formulation technology involved in the manufacturing of PVC foams. Typically, a
PVC formulation will involve about a dozen ingredients (also referred to as additives) and will follow a complex
manufacturing process. The number of characteristics measured on the product is large, on the order of about
35 responses. Some of these responses are highly correlated. Much is known about the way in which
individual additives function, but a holistic approach is required to better understand the way a formulation
functions as a whole. A fractional factorial experiment will support this holistic approach in a minimum number
of runs which will allow economical use of laboratory time and equipment.
C.2 Description of the process

All formulations required to execute the experimental design were prepared using a high-speed mixer whose
speed was varied to obtain a constant load on the motor. Process aid and blowing agents were added to the
mixer at 120 °C, shortly before discharge to the cooler. This protocol is representative of most industrial
practices. Samples were then tested using a variety of equipment as listed in C.3.2.
C.3 Response variable
C.3.1 Choice of response variables
Fifteen characteristics known to characterize foam formulations were measured in this experiment, thus
maximizing the amount of information obtained from each formulation. In this example, the three most
important responses for the analyst to study, namely fusion torque, hot expansion ratio and elongation at
break, will be studied. Large values are desirable for all three properties.
C.3.2 Measurement of the response variables
A variety of laboratory equipment was used to generate the responses, including a torque rheometer, a
capillary rheometer, a two-roll mill and a single screw extruder, all under fixed operating conditions.
C.4 Factors affecting the response
C.4.1 Description of each factor (continuous/discrete) to be varied
The study concentrates on nine factors related to the lubrication, filler, processing aid, nucleating and blowing
--`,,```,,,,````-`-`,,`,,`,`,,`---
agent components of a basis formulation. These factors are presented in Table C.1. Quantities are expressed
in phr, or parts per hundred of PVC resin.
8) This example has been kindly donated by Rohm and Haas (European Laboratories), a wholly owned subsidiary of the
Dow Chemical Company.

Provided by IHS
ISO/TR 12845:2010(E)
Table C.1 — Factors
Component type Component Parts per hundred of PVC resin
Nucleating agent Calcium stearate 0,60

Lubricant Partially saponified montanic acid ester 0,50
Lubricant Oxidized polyethylene wax 0,40
Lubricant Dicarboxylic acid ester 0,60
Lubricant Hydroxy stearic acid 0,40
Filler Ground Ca carbonate 3,00
Processing aid Paraloid K-4009) 5,50
Blowing agent Sodium bicarbonate 1,50
Blowing agent Azo dicarbonamide 0,20
C.4.2 Selection of levels (related to size of effect to be determined)
The factors used in the experiment and their associated levels are given in Table C.2.
Table C.2 — Controlled factors
Label in the analysis Factor Level 1 (−1) Level 2 (+1)
A CaSt Calcium stearate 0,2 1,0

B OPWax Partially saponified montanic acid ester 0,3 0,7
C AC680A10) Oxidized polyethylene wax 0,2 0,6
D G60 Dicarboxylic acid ester wax 0,2 1,0

E G21 Hydroxy stearic acid 0,1 0,7
F 95T Calcium carbonate 1,0 5,0
G K400 Paraloid K-400 4,0 7,0
H BIN Sodium bicarbonate 1,2 1,8
J EPE Azo dicarbonamide 0,15 0,25
--`,,```,,,,````-`-`,,`,,`,`,,`---
C.4.3 Other factors, external to the experiment, that need to be accounted for
Other components in a typical PVC formulation such as resin, stabilizer, impact modifier and pigment were
deemed less important regarding their impact on the responses and were fixed as constants in this
experiment.
9) Paraloid K-400 is the trade name of a product supplied by Rohm and Haas. This information is given for the
convenience of users of this Technical Report and does not constitute an endorsement by ISO of the product named.
10) A-C 680A is the trade name of a product supplied by RheochemTM. This information is given for the convenience of
users of this Technical Report and does not constitute an endorsement by ISO of the product named.

Provided by IHS
ISO/TR 12845:2010(E)
C.5 Fractional factorial design
C.5.1 Choice of specific design
The primary limitation in this experiment was the time required to obtain all 15 of the responses because of
the need to use sophisticated, expensive equipment that were also needed to run other applications in the
laboratory. With nine factors to be investigated, running a resolution III design appeared to be the only viable
strategy to outweigh the all too common one-factor-at-a-time approach which would have otherwise been
9 −5
adopted. Sixteen experiments were proposed in a resolution III design denoted 2III .
It was agreed that estimation of the interactions would be ignored at this stage. The experimenters were
hunting for large effects, highly significant, that would guarantee an impact on the responses across widely
different manufacturing environments, where processing conditions might differ from those in the laboratory.
Each of the main effects is confounded with two-term and three-term interactions.
C.5.2 Design matrix (standard order and run order)
The design selected was the 16-run one-eighth fractional factorial design with nine factors (A, B, C, D, E, F, G,
H, J). Randomization was used to set the order of the experiments. Table C.3 shows the design. The layout is
given in standard (std) order.
Table C.3 — Design matrix
Std A: CaSt B: OPWax C: AC680A D: G60 E: G21 F: 95T G: K400 H: BIN J: EPE
1 0,2 0,3 0,2 0,2 0,1 1 4 1,2 0,25

2 1 0,3 0,2 0,2 0,7 1 7 1,8 0,15
3 0,2 0,7 0,2 0,2 0,7 5 4 1,8 0,15
4 1 0,7 0,2 0,2 0,1 5 7 1,2 0,25
5 0,2 0,3 0,6 0,2 0,7 5 7 1,2 0,15
6 1 0,3 0,6 0,2 0,1 5 4 1,8 0,25
7 0,2 0,7 0,6 0,2 0,1 1 7 1,8 0,25
8 1 0,7 0,6 0,2 0,7 1 4 1,2 0,15
9 0,2 0,3 0,2 1 0,1 5 7 1,8 0,15
10 1 0,3 0,2 1 0,7 5 4 1,2 0,25
11 0,2 0,7 0,2 1 0,7 1 7 1,2 0,25
12 1 0,7 0,2 1 0,1 1 4 1,8 0,15
13 0,2 0,3 0,6 1 0,7 1 4 1,8 0,25
14 1 0,3 0,6 1 0,1 1 7 1,2 0,15
15 0,2 0,7 0,6 1 0,1 5 4 1,2 0,15
16 1 0,7 0,6 1 0,7 5 7 1,8 0,25
17 0,6 0,5 0,4 0,6 0,4 3 5,5 1,5 0,2
18 0,6 0,5 0,4 0,6 0,4 3 5,5 1,5 0,2
19 0,6 0,5 0,4 0,6 0,4 3 5,5 1,5 0,2
36
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
C.5.3 Centre points
Three centre points were selected for this experiment (std 17 to std 19 in Table C.3), a maximum given the
experimental circumstances described. Caution should be exerted when the number of centre points is small,
(typically less than five) because of the small degrees of freedom used in the estimation of the pure error
variance. The corresponding estimate (usually labelled MSE, or Mean Square Error) of the pure error variance
should be carefully examined by the scientist to judge if it is larger or smaller than might reasonably be
expected.
C.5.4 Replication and repetition
Replication (other than for the centre points) and repetition were not selected for this experiment because of
the time constraints.
C.6 Analysis of results
C.6.1 Data acquired through the experiment and analysis considerations
The results of the experiment for the three responses of interest are shown in Table C.4.
Table C.4 — Experiment results
Fusion Hot Elongation

Std A B C D E F G H J torque expansion at break
N⋅m ratio %
1 0,2 0,3 0,2 0,2 0,1 1 4 1,2 0,25 56 2,98 472

2 1 0,3 0,2 0,2 0,7 1 7 1,8 0,15 57,3 3,48 254
3 0,2 0,7 0,2 0,2 0,7 5 4 1,8 0,15 50,2 2,16 454
4 1 0,7 0,2 0,2 0,1 5 7 1,2 0,25 58 3,66 337
5 0,2 0,3 0,6 0,2 0,7 5 7 1,2 0,15 52,7 2,44 537
--`,,```,,,,````-`-`,,`,,`,`,,`---
6 1 0,3 0,6 0,2 0,1 5 4 1,8 0,25 52 3,28 517

7 0,2 0,7 0,6 0,2 0,1 1 7 1,8 0,25 58,8 3,27 400
8 1 0,7 0,6 0,2 0,7 1 4 1,2 0,15 48,4 1,74 228
9 0,2 0,3 0,2 1 0,1 5 7 1,8 0,15 57,3 3,71 447
10 1 0,3 0,2 1 0,7 5 4 1,2 0,25 46,4 2,62 531
11 0,2 0,7 0,2 1 0,7 1 7 1,2 0,25 50,9 2,35 450
12 1 0,7 0,2 1 0,1 1 4 1,8 0,15 52,1 3,36 399
13 0,2 0,3 0,6 1 0,7 1 4 1,8 0,25 45,7 1,97 366
14 1 0,3 0,6 1 0,1 1 7 1,2 0,15 57,6 3,48 485
15 0,2 0,7 0,6 1 0,1 5 4 1,2 0,15 49,4 2,5 532
16 1 0,7 0,6 1 0,7 5 7 1,8 0,25 51,6 2,68 270
17 0,6 0,5 0,4 0,6 0,4 3 5,5 1,5 0,2 53,4 2,94 410
18 0,6 0,5 0,4 0,6 0,4 3 5,5 1,5 0,2 53,1 2,98 414
19 0,6 0,5 0,4 0,6 0,4 3 5,5 1,5 0,2 54,6 2.98 472

Provided by IHS
ISO/TR 12845:2010(E)
The results were keyed into a software programme called Design-Expert11). The analyses for each one of
these three responses (fusion torque, hot expansion and elongation at break) are shown in C.6.2, C.6.3 and
C.6.4, respectively.
C.6.2 Analysis of fusion torque
The normal probability plot in Figure C.1 indicates three factors (E, D, G) as highly significant. R-square for
this model is 89 %. The analysis of variance, shown in Table C.5, shows no significant curvature nor lack of fit.
The coefficient estimates for the model are shown in Table C.6. Plots of effects are displayed in Figure C.2.
99
97
G
95
90
E
85
D
80
70
60
40
20
0,00 1,37 2,75 4,12 5,50 X

Key
X effect
--`,,```,,,,````-`-`,,`,,`,`,,`---
D G21
E 95T
G BIN
Figure C.1 — Half-normal probability plot for fusion torque
11) Design-Expert is the trade name of a product supplied by Stat-Ease, Inc. This information is given for the convenience
of users of this Technical Report and does not constitute an endorsement by ISO of the product named. Equivalent
products may be used if they can be shown to lead to the same results.

Provided by IHS
ISO/TR 12845:2010(E)
Table C.5 — ANOVA table for fusion torque
ANOVA for selected factorial model

Analysis of variance table (Partial sum of squares)
Sum Degrees
Model 242,61 3 80,87 36,88 < 0,000 1

D 31,36 1 31,36 14,30 0,002 0
E 90,25 1 90,25 41,16 < 0,000 1
G 121,00 1 121,00 55,18 < 0,000 1
Curvature 2,16 1 2,16 0,99 0,337 6
Residual 30,70 14 2,19

Lack of fit 29,44 12 2,45 3,89 0,222 3
Pure error 1,26 2 0,63
Cor Totala 275,47 18

a Cor Total = Totals for all the information corrected for the mean.
Table C.6 — Estimation of effects for fusion torque

Factor Standard error VIFb
Intercept 52,78 1 0,37 51,98 53,57

D: G60 −1,40 1 0,37 −2,19 −0,61 1,00
E: G21 −2,38 1 0,37 −3,17 −1,58 1,00
G: K400 2,75 1 0,37 1,96 3,54 1,00
Centre point 0,92 1 0,93 −1,07 2,92 1,00
a CI = Confidence interval.
--`,,```,,,,````-`-`,,`,,`,`,,`---
b VIF = Variation inflation factor.

Provided by IHS
ISO/TR 12845:2010(E)
Y Y Y
58,80 58,80 58,80
55,52 55,52 55,52
52,25 52,25 52,25
48,98 48,98 48,98
45,70 45,70 45,70
0,20 0,40 0,60 0,80 1,00 X1 0,10 0,25 0,40 0,55 0,70 X2 4,00 4,75 5,50 6,25 7,00 X3
Actual factors
Factor
Key Figure C.2 a) Figure C.2 b) Figure C.2 c)
X1 D: G60 A: CaSt: 0,60 % 0,60 % 0,60 %
X2 E: G21 B: OPWax: 0,50 % 0,50 % 0,50 %
X3 G: K400 C: AC680A: 0,40 % 0,40 % 0,40 %
Y fusion torque D: G60: — 0,60 % 0,60 %
x design point E: G21: 0,40 % — 0,40 %

F: 95T: 3,00 % 3,00 % 3,00 %
G: K400: 5,50 % 5,50 % —
H: BIN: 1,50 % 1,50 % 1,50 %
J: EPE: 0,20 % 0,20 % 0,20 %
Figure C.2 — One-factor plots — Effect of factors D, E and G on fusion torque

--`,,```,,,,````-`-`,,`,,`,`,,`---
C.6.3 Analysis of the hot expansion ratio
The normal probability plot in Figure C.3 indicates six main effects (aliased), A, B, C, E, G and H, which are
highly significant and one weak interaction term, AC (aliased), which is significant to a lesser extent. R-square
for the model including all these seven terms is 99 %, perhaps suspiciously high. The analysis of variance,
shown in Table C.7, shows no significant curvature, nor lack of fit. The coefficient estimates in the
corresponding model are shown in Table C.8. The interaction between A and C, shown in Figure C.4,
indicates a weak effect, thus was ignored, taking into account both practical significance and the risk of wrong
interpretation due to aliasing. Removing AC from the model does not change the coefficients of the remaining
terms, and has a minimum impact on R-square.

Provided by IHS
ISO/TR 12845:2010(E)
99
97
E
95
90 G
85
C
80
A
--`,,```,,,,````-`-`,,`,,`,`,,`---
70 B
H
60 AC
40
20
0,00 0,21 0,43 0,64 0,85 X

Key
X effect
A CaSt
B OPWax
C AC680A
E G21
F 95T
G K400
H BIN
J EPE
Figure C.3 — Half-normal probability plot for the hot expansion ratio

Provided by IHS
ISO/TR 12845:2010(E)
Table C.7 — ANOVA table for the hot expansion ratio
ANOVA for selected factorial model

Analysis of variance table (Partial sum of squares)
Sum Degrees
Model 5,87 7 0,84 125,05 < 0,000 1
A 0,53 1 0,53 79,52 < 0,000 1
< 0,000 1
--`,,```,,,,````-`-`,,`,,`,`,,`---
B 0,31 1 0,31 46,79
C 0,55 1 0,55 81,71 < 0,000 1
E 2,89 1 2,89 432,24 < 0,000 1
G 1,24 1 1,24 185,51 < 0,000 1
H 0,29 1 0,29 42,71 < 0,000 1
AC 0,053 1 0,053 7,89 0,018 5
Curvature 0,032 1 0,032 4,70 0,055 3
Residual 0,067 10 6,702 × 10−3

Lack of fit 0,066 8 8,244 × 10−3 15,46 0,062 2
Pure error 1,067 × 10−3 2 5,333 × 10−4
Cor Totala 5,96 18

a Cor Total = Totals for all the information corrected for the mean.
Table C.8 — Estimation of effects for hot expansion ratio

Factor Standard error
Intercept 2,86 1 0,020 2,81 2,90

A: CaSt 0,18 1 0,020 0,14 0,23
B: OPWax −0,14 1 0,020 −0,19 −0,094
C: AC680A −0,19 1 0,020 −0,23 −0,14
E: G21 −0,43 1 0,020 −0,47 −0,38
G: K400 0,28 1 0,020 0,23 0,32
H: BIN 0,13 1 0,020 0,088 0,18
AC −0,057 1 0,020 −0,10 −0,012
Centre point 0,11 1 0,052 −3,093 × 10−3 0,23
a Confidence interval.

Provided by IHS
ISO/TR 12845:2010(E)
3,71
3,22
2,73
2,23
1,74
0,20 0,40 0,60 0,80 1,00 X

Key
X A: CaSt
Y hot expansion ratio with the top line at C = 0,2 parts per hundred resin of AC680A and the bottom line at C = 0,6 parts
per hundred resin of AC680A
z design point
C = 0,200
S C = 0,600
Actual factors
B OPWax = 0,50
D G60 = 0,60
E G21 = 0,40
--`,,```,,,,````-`-`,,`,,`,`,,`---
F 95T = 3,00
G K400 = 5,50
H BIN = 1,50
J EPE = 0,20
Figure C.4 — Effect of interaction between A (CaSt) and C (AC680A) on the hot expansion ratio

Provided by IHS
ISO/TR 12845:2010(E)
Y Y Y
3,71 3,71 3,71
3,22 3,22 3,22

2 2 2
2,73 2,73 2,73
2,23 2,23 2,23
1,74 1,74 1,74
0,20 0,40 0,60 0,80 1,00 X1 0,20 0,30 0,40 0,50 0,60 X2 0,30 0,40 0,50 0,60 0,70 X3
Y Y Y
3,71 3,71 3,71
3,22 3,22 3,22

2
2 2
2,73 2,73 2,73
2,23 2,23 2,23
1,74 1,74 1,74
0,10 0,25 0,40 0,55 0,70 X4 4,00 4,75 5,50 6,25 7,00 X5 1,20 1,35 1,50 1,65 1,80 X6
Key
X1 A: CaSt
X2 C: AC680A
X3 B: OnWax
X4 E: G21
X5 G: K400
X6 H: BIN
Y hot expansion ratio
Figure C.5 — Effect plots for the hot expansion ratio
C.6.4 Elongation at break
The normal probability plot in Figure C.6 indicates that there are possibly up to 12 significant effects (aliased)
that, when included in the model, return an R-squared value of 98 %. The standard deviation, or RMSE, of
22,29 was believed by the experimenters to underestimate the noise in the system. In addition, some of the
factors were affecting the response in an unexplainable direction. (The ANOVA table for elongation at break is
--`,,```,,,,````-`-`,,`,,`,`,,`---
presented in Table C.9.)
This might be the combined consequence of too few replicates and too many confounded effects in the
estimation of the model coefficients, which obscures the practical interpretation of the results. It was decided
to drop this response from further consideration.

Provided by IHS
ISO/TR 12845:2010(E)
99
97
A
95
90
F
85
80 AB
70 B
E
60
H
AD
40 AE
AG
20 G
D
AH
0
0,00 19,91 39,81 59,72 79,63 X

Key
X effect
A CaSt
B OPWax
C AC680A
D G60
E G21
F 95T
G K400
H BIN
J EPE
Figure C.6 — Half-normal probability plot for elongation at break
Table C.9 — ANOVA table for elongation at break
Standard deviation 22,29 R-squared 0,984 2

Mean 419,74 Adjusted R-squared 0,946 4
Coefficient of variation 5,31 Predicted R-squared 0,952 0
PRESSa 7 585,11 Adequate precision 15,934

a PRESS = Predicted Residual Error Sum of Squares.
45
--`,,```,,,,````-`-`,,`,,`,`,,`---
Provided by IHS
ISO/TR 12845:2010(E)
C.7 Presentation of findings

In both responses for which results were obtained, factor E and factor G turned out to be the most significant
effects. Thus, Paraloid K400 and hydroxy stearic acid play a dominant role in determining the properties of the
formulation. More K400 and less hydroxy stearic acid are needed to maximize the fusion torque and hot
expansion ratio. The effects produced by other factors are of a lesser magnitude. No evidence was found to
suggest that either K400 or hydroxy stearic acid had any significant interactions, either together or with other
factors.
C.8 Confirmation results

Though it would have been desirable to confirm the results with a few additional runs, in this application the
primary objective was not to check the accuracy of the model predictions but rather to identify major players
affecting all or most responses studied, knowing that more experimenting was going to be needed at the
customer sites. Thus, no confirmation runs were obtained. Given the general significance of factors E and G, it
was safe to recommend their use as variables in experiments on an industrial scale and only consider the
other factors where very specific modifications to the formulation behaviour were required.
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
Annex D
(informative)
Process validation for an insulin product
D.1 General
Both healthcare products and drugs must show stability under specified storing conditions. The exact rules are
specified by regulatory bodies such as the Food and Drug Administration (FDA) in the USA. In this example,
aspects of stability of an insulin product are the responses that are investigated, but the experiment is also a
process validation study in the sense that stability of the drug for varying process parameters is being
investigated.
D.2 Description of the process

In order to understand the experiment, it is sufficient to realize that it is a complicated chemical process where
relevant process parameters are concentrations of chemicals, temperatures, filtration times, etc.
D.3 Response variables
D.3.1 Choice of variables
No single response variable reflects the stability or the degradation of the product. All response variables,
except two, were differences between responses immediately after production and responses after storage at
4 °C for 18 months or after storage at 25 °C for 12 months. Only data from the storage at 4 °C for 18 months
will be considered in this example.
Eight response variables were chosen for analysis. Table D.1 gives an overview of the response variables
analysed. The first two, zinc and pH, are preparations of the product. The next one, labelled “HPLC
assay U/ml”, expresses the potency of the drug, while the remaining five variables are related to degradation
of the product.
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
Table D.1 — Response variables considered
Response Descriptiona Name in analysis
Zinc (µg/ml) Change 0-18 months 4 °C Zinc018

pH Change 0-18 months 4 °C ph018
HPLCb assay U/ml Change 0-18 months 4 °C HPLC018
Higher molecular weight products Change 0-18 months 4 °C HMWP018

Dissolved insulin, total Change 0-18 months 4 °C disol018
Other insulin-related products (%) Change 0-18 months 4 °C Other018
Desamido insulin 18 months at 4 °C A21des18
(A21, acid) HPCEc (%)
Desamido insulin 18 months at 4 °C B3des18
(B3, neutral) HPCE (%)
a See D.3.1 for expanded descriptions.
b HPLC = High-performance liquid chromatography.
c HPCE = High-performance capillary electrophoresis.
D.3.2 Measurement of the response variable
The variables were measured right after production and again after storage for the stipulated amount of time.
The response variables that were analysed were the differences of the values obtained after storage and the
initial values measured right after production. Exceptions were the desamido values (the last two variables in
Table D.1), where no initial values were measured because they were very small.
The measurements were obtained in two laboratories, and the fact that two laboratories were used makes it
natural to run the experiment in two blocks corresponding to laboratories.
D.3.3 Relationship of the response variables to the objective of the experiment
The objective of the experiment is to study stability of an insulin product. In this light, one may say that the
third variable of Table D.1, “HPLC assay U/ml”, is the most important aspect of stability, namely the potency of
the drug after storage, whereas the last five variables of Table D.1 are used to study more subtle aspects of
degradation.
D.4 Factors affecting the response
D.4.1 Description of each factor (continuous/discrete) to be varied
The factors of the experiment and their associated levels are given in Table D.2. The first six, denoted by the
capital letters A to F, are process parameters that describe process conditions at different stages of production.
Factors G and H describe the adjustments of the product after production, i.e. adjustment of the zinc level and
the pH level. Note that zinc and pH are both factors of the experiment and response variables. The factors are
continuous but for each factor, two levels are chosen.
D.4.2 Selection of levels (related to size of effect to be determined)
The eight factors used in the experiment and their associated levels are given in Table D.2.
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
Table D.2 — Factors
Factor Low level High level
A Time2(IV ins.sol.+ins/prot.sol.)+filtering (−1) 70 ± 5 + 30 ± 5 min (1) 55 ± 5 + 70 ± 5 min

B Temp2(ins/prot.sol.) (−1) 20 °C ± 2 °C (1) 27 °C ± 2 °C
C Time4(VII ins.sol.) (−1) 30 ± 5 min (1) 100 ± 5 min
D Temp4(VII ins.sol.) (−1) 5 °C ± 3 °C (1) 18 °C ± 3 °C
E Temp2(IV ins.sol.) (−1) 5 °C ± 3 °C (1) 18 °C ± 3 °C
F pH2(ins/prot.sol.) (−1) 2,65 ± 0,03 (1) 3,25 ± 0,03
G Zincprep. (−1) 25,0 ± 2,0 µg/ml (1) 31,0 ± 2,0 µg/ml
H pHcryst, via mix = pHprep. (−1) 7,15 ± 0,03 (1) 7,40 ± 0,03
D.4.3 Other factors noted but not incorporated due to issues with controllability or relevance
No other factors were identified for the experiment.
D.5 Fractional factorial design
D.5.1 Choice of specific design
This design was constructed by embedding the factors E, F, G and H in the complete 24 factorial design
defined by the factors A, B, C and D. The following definitions (aliasing) of the factors E, F, G and H in terms of
factors A, B, C and D were used:
E = BCD
F = ACD
G = ABD
--`,,```,,,,````-`-`,,`,,`,`,,`---
H = ABC
In addition, the experiment was run in two blocks, corresponding to the laboratories R1 and R2 where the
measurements were made. The blocks were defined as
Block = ABCD
The resulting design with levels coded as −1 (low level) and 1 (high level) is shown in Table D.3.

Provided by IHS
ISO/TR 12845:2010(E)
Table D.3 — Design matrix
Experimental design
Std A B C D E F G H Block/Number Test number
1 −1 −1 −1 −1 −1 −1 −1 −1 R2 / 4 8
2 1 −1 −1 −1 −1 1 1 1 R1 / 2 3
3 −1 1 −1 −1 1 −1 1 1 R1 / 8 15
4 1 1 −1 −1 1 1 −1 −1 R2 / 7 14
5 −1 −1 1 −1 1 1 −1 1 R1 / 1 1
6 1 −1 1 −1 1 −1 1 −1 R2 / 1 2
7 −1 1 1 −1 −1 1 1 −1 R2 / 2 4
8 1 1 1 −1 −1 −1 −1 1 R1 / 4 7
9 −1 −1 −1 1 1 1 1 −1 R1 / 5 9
10 1 −1 −1 1 1 −1 −1 1 R2 / 6 12
11 −1 1 −1 1 −1 1 −1 1 R2 / 3 6
12 1 1 −1 1 −1 −1 1 −1 R1 / 7 13
13 −1 −1 1 1 −1 −1 1 1 R2 / 5 10
14 1 −1 1 1 −1 1 −1 −1 R1 / 3 5
15 −1 1 1 1 1 −1 −1 −1 R1 / 6 11
16 1 1 1 1 1 1 1 1 R2 / 8 16
Although complete information about the settings of the factors in each run of the experiment is available from
Table D.2 and Table D.3, it is advisable to write out detailed instructions about the settings of the factors in
each run to the people who are running the experiment. An example of a description of a single experiment
(run) from the design is given in Table D.4.
Table D.4 — Experiment (run) number 6 to be carried out in laboratory R212)
Process parameter Level in experiment
A Time2(IV ins.sol.+ins/prot.sol.)+filtering (1) 55 ± 5 + 70 ± 5 min

B Temp2(ins/prot.sol.) (−1) 20 °C ± 2 °C
C Time4(VII ins.sol.) (−1) 30 ± 5 min
D Temp4(VII ins.sol.) (1) 18 °C ± 3 °C
E Temp2(IV ins.sol.) (1) 18 °C ± 3 °C
F pH2(ins/prot.sol.) (−1) 2,65 ± 0,03
G Zincprep (−1) 25,0 ± 2,0 µg/ml
H pHcryst, via mix = pHprep (1) 7,40 ± 0,03
12) Experiment (run) number 6 in laboratory R2 corresponds to number 10 in the standard order.
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
Generators: E = BCD, F = ACD, G = ABD, H = ABC
Block generator: Blocks = ABCD
Defining relation: I = BCDE = ACDF = ABDG = ABCH = ABEF = ACEG = ABEH

= BCFG = BDFH = CDGH = DEFG = CEFH = AFGH = BEGH
= ACDEFGH
The length of the shortest word in the defining relation is four and this shows that the design is of resolution IV.
The complete alias structure for each factor contains 16 terms. The one for factor A, for example, is obtained
from the defining relation by multiplying each term by A and using the rule that AA cancels:
A = ABCDE = CDF = BDG = BCH = BEF = CEG = BEH

= ABCFG = ABDFH = ACDGH = ADEFG = ACEFH = FGH = ABEGH
= CDEFGH
The complete alias structure for blocks is obtained by multiplying each term in the defining relation by ABCD
and recognizing that the terms AA, BB, CC and DD cancel:
Blocks = AE = BF = CG = DH = CDEF = BDEG = CDEH

= ADFG = ACFH = ABGH = ABCEFG = ABDEFH = BCDFGH = ACDEGH
= BEFGH
The alias structure restricted to two-way interactions is given in Table D.5.
Table D.5 — Alias structure of the two-factor interactions
Blocks = J = AE = BF = CG = DH
AB = EF = DG = CH
AC = DF = EG = BH
BC = DE = FG = AH
AD = CF = BG = EH
BD = CE = AG = FH
CD = BE = AF = GH
This means that main effects can be estimated to be free from two-factor interactions and that the
(8 × 7)/2 = 28 two-factor interactions are aliased in seven groups of four two-factor interactions. If specific two-
actor interactions were expected to be active, it is important at this stage to inspect the alias structure to make
sure that supposedly active interactions are not aliased. In this case, there were no such expectations.
--`,,```,,,,````-`-`,,`,,`,`,,`---
D.5.2 Design matrix (standard order and run order)
The design selected was the 16-run one-sixteenth fractional factorial design with eight factors (A, B, C, D, E, F,
G and H). The design is shown in Table D.3 in standard order, i.e. in the sense that the full 24 design in the
defining factors A, B, C and D is given in standard order.

Provided by IHS
ISO/TR 12845:2010(E)
The runs were randomized within blocks. The number of each run within the specific block is shown in the
“Block/Number” column of Table D.3.
D.5.3 Centre points
Centre points were not selected for this experiment.
D.5.4 Replication and repetition
Replication and repetition were not selected for this experiment.
D.6 Analysis of results
D.6.1 Data acquired through the experiment and analysis considerations
The results from the experiment are given in Table D.6. The first five columns are also the first five columns of
the design matrix of Table D.3. The following eight columns are the response variables recorded. The names
are the ones given in the last column of Table D.1. The final column gives the run number within block
(laboratory), “r_w_b”. In order to save space, the remaining factors are not listed, but they are defined in terms
of the factors A, B, C and D as explained in D.5.1 as E = BCD, F = ACD, G = ABD, H = ABC, and block = ABCD.
The analysis was performed using the software R13). With 16 runs, it is possible to estimate 15 parameters in
addition to the grand mean. The chosen parameters are the main effects of the eight factors A to H, the block
factor J and an arbitrarily chosen two-factor interaction in each of the remaining six sets of aliased two-factor
interactions in Table D.5. If a two-factor interaction should turn out to be significant, the alias pattern in
Table D.5 as well as other significant factors will be considered before any conclusions are made.
We have a saturated model with no estimate for error so the analysis will be based on normal plots of effect
supplemented with Lenth plots.
--`,,```,,,,````-`-`,,`,,`,`,,`---
13) R is a free software environment for statistical computing and graphics. It compiles and runs on a wide variety of UNIX
platforms, Windows and MacOS; see http://www.r-project.org/. This information is given for the convenience of users of
this Technical Report and does not constitute an endorsement by ISO of this product.

Provided by IHS
ISO/TR 12845:2010(E)
Table D.6 — Experimental results
Std A B C D Zinc018 Ph018 HPLC018 HMWP018 Disol018 Other018 A21des18 B3des18 r_w_ba
1 −1 −1 −1 −1 −5,5 −0,02 −0,9 0, 2 3,2 0,3 0,5 0,7 4

2 1 −1 −1 −1 −5,6 0,00 −0,8 0,3 2,8 0,5 0,6 0,7 2
3 −1 1 −1 −1 −3,1 −0,03 −0,5 0,2 4,4 0,6 0,6 0,8 8
4 1 1 −1 −1 −2,2 0,01 −0,4 0,2 3,4 0,4 0,6 0,5 7
5 −1 −1 1 −1 0,3 −0,01 −0,9 0,3 1,1 0,4 0,4 0,6 1
6 1 −1 1 −1 −1,7 0,01 −0,1 0,2 3,0 0,5 0,5 0,4 1
7 −1 1 1 −1 −2,8 0,05 −0,7 0,1 3,1 0,5 0,6 0,5 2
8 1 1 1 −1 0,1 −0,08 −0,2 0,2 0,7 0,6 0,7 0,6 4
--`,,```,,,,````-`-`,,`,,`,`,,`---
9 −1 −1 −1 1 −2,0 0,02 −0,2 0,1 0,7 0,6 0,5 0,4 5
10 1 −1 −1 1 −0,4 0,00 0,6 0,3 −0,3 0,6 0,6 0,8 6
11 −1 1 −1 1 −1,3 0,02 −0,7 0,4 0,7 0,8 0,6 0,9 3
12 1 1 −1 1 −5,4 0,00 −0,1 0,1 4,4 0,8 0,7 0,6 7
13 −1 −1 1 1 −5,2 0,01 −0,7 0,2 0,7 0,6 0,5 0,6 5
14 1 −1 1 1 −2,4 0,03 −1,6 0,4 1,3 0,7 0,6 0,8 3
15 −1 1 1 1 −0,4 0,00 0,2 0,2 1,5 0,4 0,6 0,4 6
16 1 1 1 1 −6,4 0,03 −0,1 0,3 4,0 0,6 0,7 0,8 8
a r_w_b = run number within block (laboratory).

Provided by IHS
ISO/TR 12845:2010(E)
zinc018 ph018
Y Y
3 3
2 2
E F
1 C 1
0 0
-1 -1
CD
-2 G -2
-3 -3
-2 -1 0 1 X -0,02 -0,01 0,00 0,01 0,02 0,03 X
a) Response variable zinc018 b) Response variable ph018
HPLC018 HMWP018
Y Y
3 3
2 2
1 1
0 0
-1 -1
-2 -2
-3 -3
-0,4 -0,2 0,0 0,2 0,4 X -0,05 0,00 0,05 X
--`,,```,,,,````-`-`,,`,,`,`,,`---
c) Response variable HPLC018 d) Response variable HMWP018
Key
X estimated effects
Y normal fractile
Figure D.1 — Normal plots of estimated effects for the response variables zinc018, ph018,
HPLC018 and HMWP018
Estimated effects are along the horizontal axis. The points indicated by arrows and labelled are not indicative
of significance. They are labelled for discussion in the text.

Provided by IHS
ISO/TR 12845:2010(E)
disol018 Other018
Y Y
3 3
2 2
1 1
0 0
-1 -1
-2 -2
-3 -3
-1,0 -0,5 0,0 0,5 1,0 1,5 X -0,05 0,00 0,05 0,10 0,15 X
a) Response variable diso018 b) Response variable other018
A21des18 B3des18
Y Y
3 3
2 2
B H
1 A 1 AD
0 0
-1 -1
-2 -2
-3 -3
0,00 0,05 0,10 X -0,05 0,00 0,05 0,10 0,15 X
c) Response variable A21des18 d) Response variable B3des18
Key
X estimated effects
Y normal fractile
Figure D.2 — Normal plots of estimated effects for the response variables disol018, other018,
A21des18 and B3des18
Estimated effects are along the horizontal axis. Estimated effects are along the horizontal axis. The points
indicated by arrows and labelled are not indicative of significance. They are labelled for discussion in the text.
55
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
zinc018
Y Y
ph018
4
ME 0,03
ME
2
0,01
0
X X
-0,01
-2
ME ME
-4 -0,03
A B C D E F G H J AB AC BC AD BD CD A B C D E F G H J AB AC BC AD BD CD
a) Response variable zinc018 b) Response variable ph018
HPLC018 HMWP018
Y ME
Y ME
0,15
0,5
0,05
--`,,```,,,,````-`-`,,`,,`,`,,`---
0,0
X -0,05
X
-0,5
-0,15
ME ME
c) Response variable HPLC018 d) Response variable HMWP018
Key
X factors
Y estimated effects
ME margin of error
SME simultaneous margin of error
Figure D.3 — Lenth plots of estimated effects for the response variables zinc018, ph018, HPLC018
and HMWP018

Provided by IHS
ISO/TR 12845:2010(E)
Y ME Y
2
0,3
ME
1 0,2
0,1
0
X
-1 X
-0,1
-2 ME -0,2
A B C D E F G H J AB AC BC AD BD CD ME
-0,3 A B C D E F G H J AB AC BC AD BD CD
a) Response variable disoI018 b) Response variable other018
Y
Y SME
0,3
ME
0,10
0,2
ME 0,1
0,05
0
0 -0,1
X
X
-0,2
-0,05 ME ME
-0,3
c) Response variable A21des18 d) Response variable B3des18
Key
X factors
Y estimated effects
ME margin of error
SME simultaneous margin of error
Figure D.4 — Lenth plots of estimated effects for the response variables disol018, other018,
A21des18 and B3des18

--`,,```,,,,````-`-`,,`,,`,`,,`---
57
Provided by IHS
ISO/TR 12845:2010(E)
D.6.2 Analysis of results — Normal plots
The normal plots of the estimated effects for the eight response variables in Table D.1 are collected in
Figures D.3 and D.4.
The estimated effects are average responses at the high level minus the average responses at the low level.
Here, the normal plots are used in the following way. Straight lines are fitted to the points by eye with
emphasis on the central points. Negative estimated effects that lie to the left of the line and positive estimated
effects that lie to the right of the straight line are considered to be active. Only the normal plots for responses
zinc018, A21des18 and B3des18 show estimated effects that may seem active.
In the normal plot for the response zinc018, it is tempting to plot a straight line based on the 11 central
estimates which leads to the conclusion that G, CD, C and E are active factors. But numerically the estimate of
E is not larger than some of the 11 central estimates, so either E is active and even more factors must be
considered as active, or E is not active. The choice was made to not consider E as active. Then, if E does not
deviate too much from the straight line neither does C, thus leaving G and CD as the only potentially active
effects.
For the response A21des18, the evaluation of the normal plot is straightforward. Clearly, the positive
estimated effect A and B are to the right of a straight line defined by the remaining estimates.
For the response B3des18, the situation is similar but not as clear. Active effects, if any, are AD and H, but it is
not obvious that the points deviate significantly from a straight line defined by the remaining estimates.
The Lenth plots in the following subclause will be used to support the tentative conclusions already made.
D.6.3 Analysis of results — Lenth plots
The Lenth plot is a graphical illustration of Lenth's procedure to decide which factors are active. Based on the
estimated effects an estimated standard error is calculated. Two multiples of the estimated standard error are
calculated: the margin of error (ME) and the simultaneous margin of error (SME). If we entertain the
hypothesis before the experiment that a certain effect is active, an appropriate test is to consider it active if the
absolute value of its estimate exceeds the ME. For screening purposes, the SME is used. A factor is
considered active if the absolute value of its effect estimate exceeds the SME. The Lenth plot illustrates this
procedure by placing the factors along the horizontal axis and representing the estimated effects by vertical
lines. Horizontal lines represent the ME and the SME, so it is easy to see which effects exceed one or the
other. Note that the SME is always larger than the ME, so the SME may not show in all Lenth plots.
The Lenth plots of the estimated effects for the eight response variables in Table D.1 are plotted in Figure D.3
and Figure D.4.
For the response zinc018, the conclusion is that no factors are active. This conclusion is consistent with the
discussion of the normal plot. It is indeed possible to fit a straight line that is close to all points.
For the response ph018, the factor F exceeds the ME line, so if that had been a specific hypothesis before the
experiment, F would have been considered to be an active factor. But it was not the case, so it has to exceed
the SME line to be considered active.
The response A21des18 is the only variable with an estimated effect, B, exceeding the SME.
D.6.4 Presentation of results
The results are summarized in Table D.7. As already explained in D.6.2 and D.6.3, A21des18 is the only
variable showing an active effect. The experiment has indicated that the process is stable within the process
window considered.
58 --`,,```,,,,````-`-`,,`,,`,`,,`---
Provided by IHS
ISO/TR 12845:2010(E)
Table D.7 — Summary of results
Response Analysis Comments
Zinc (µg/ml) Change 0-18 months 4 °C no effects detected

pH Change 0-18 months 4 °C no effects detected
Higher molecular weight products Change 0-18 months 4 °C no effects detected
Dissolved insulin, total Change 0-18 months 4 °C no effects detected
HPLC assay U/ml Change 0-18 months 4 °C no effects detected
Other insulin-related prod. (%) Change 0-18 months 4 °C no effects detected
Desamido insulin 18 months at 4 °C Temp2 effects (factor B)
(A21, acid) HPCE (%)
Desamido insulin 18 months at 4 °C no effects detected
(B3, neutral) HPCE (%)
D.6.5 Confirmation results
In this case, no confirmation experiment or single confirmation run was performed.
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
Annex E
(informative)
Washing machine experiment
E.1 General
Washing machines are usually repairable items and consumers want the interval between repairs to be as
long as possible. The service organizations, whose job it would be to provide assistance in any repair, are
also interested parties. The fewer times a washing machine breaks down, the lower the associated costs to
the service organization.
During the development programme for a new washing machine, the design engineers test different
subsystems of the new machine and record, amongst other things, the number of cycles to failure for the
subsystem. Design engineers attempt to design subsystems and components that will perform well in spite of
many different usage conditions, such as the water type, that the machines will encounter during their lives.
E.2 Overall objective for the experiment

A washing machine manufacturer, developing a new range of machines, was interested to know if certain
design changes would lead to greater reliability. The following paragraphs describe an experiment performed
by the design engineers to better understand which design of tub and which certain operating conditions are
most robust against some of the usage factors the washing machines are expected to face. A Taguchi design
with inner (for the design variables) and outer (for the usage factors) arrays was chosen.
E.3 Response variable

--`,,```,,,,````-`-`,,`,,`,`,,`---
E.3.1 Choice of variable
The response (Y ) chosen for the experiment was the “mean cycles between failure” (MCBF).
A signal-to-noise ratio (“biggest is best”), ∆b, was chosen to analyse the data.
⎡1 ⎛ 1 ⎞⎤
∆b = −10log10 ⎢
⎢⎣ N
∑ ⎜ ⎟⎥
⎜ Y 2 ⎟⎥
⎝ i ⎠⎦
where
Yi are the observations;
N is the number of observations.
The greatest MCBF is indicated when this expression is maximized. An analysis of the results of the
experiment shows that the levels of the design factors which maximize the signal-to-noise ratio indicate a
design that gives the greatest reliability in spite of the behaviour of the noise factors.

Provided by IHS
ISO/TR 12845:2010(E)
E.3.2 Measurement of the response variable
Six washing machines were built for each arrangement of the design factor levels (runs). They were all run in
a laboratory and continuously cycled to failure. Each machine had a counter fitted to it to record the number of
cycles. For each run, the MCBF was calculated and used in the analysis of the experiment. Depending on the
nature of the failure, a washing machine would be repaired and then cycled to failure again. In this way,
several machines were repaired several times during the experiment.
E.3.3 Relationship of the response variable to the objective of the experiment
The MCBF is directly linked to the objective of the experiment, i.e. to determine the design considerations that
provide a greater number of cycles between failure with least variability once subjected to usage factors.
E.4 Factors affecting the response
E.4.1 Description of each factor (discrete) to be varied
The factors chosen to be varied within the experiment were determined using the knowledge of the design
engineers and technical staff. These were:
B: Tub type
C: Detergent type
E: Wash temperature
Additionally, the usage factors chosen were:
A: Wash load
D: Water type
F: Wash frequency
All of the chosen factors were treated as discrete variables as indicated in E.4.2.
E.4.2 Selection of levels (related to size of effect to be determined)
The factors used in the experiment and their associated levels are listed in Table E.1.
Table E.1 — Factors
Factor Level 1 Level 2
B Tub type Design A Design B
C Detergent type Liquid Powder
E Wash temperature Low High
A Wash load Half Full
D Water type Soft Hard
F Wash frequency Low High
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
E.4.3 Other factors noted but not incorporated due to issues with controllability or relevance
Other factors that might have been considered but were not included in the experiment are given in Table E.2.
Table E.2 — Excluded factors
Factor Reason for exclusion
External air temperature Air temperature external to the machines was

not regarded as important to the response by
engineers as the internal temperatures were
likely to be much higher.
External humidity Considered irrelevant as the humidity levels
within the machines were likely to be much
more important.
Operator As long as the operator placed the detergent in
the correct location and shut the door properly,
engineers regarded the operation of the
machines as automatic.
E.4.4 Independence of factors
The factors selected for the experiment were regarded as independent of each other.
E.5 Experimental design
E.5.1 Inner array
The design selected for the “inner” array was a Taguchi L8 design giving full resolution. With only three design
factors, this was thought to be appropriate.
Table E.3 shows the design. The numbers “1” and “2” in the first three columns of this table refer to the level
of the factor.
Table E.3 — Inner array
B C E Run
1 1 1 1
1 1 2 2
1 2 1 3
1 2 2 4
2 1 1 5
2 1 2 6
2 2 1 7
2 2 2 8
--`,,```,,,,````-`-`,,`,,`,`,,`---
NOTE The design is printed in “standard order” and not randomized since all runs were conducted concurrently.
Because the design was a full factorial, there were no aliased factors.

Provided by IHS
ISO/TR 12845:2010(E)
E.5.2 Outer array
The usage factors comprised the “outer” array. As there were three usage factors, each at two levels, it was
decided to use a Taguchi L4 design and perform a half fraction since interactions were regarded as
unimportant.
Table E.4 shows the design.
Table E.4 — Outer array
A D F Run
1 1 1 1
1 2 2 2
2 1 2 3
2 2 1 4
E.5.3 Full design
The full design is shown in Table E.5.
Table E.5 — Full design
F 1 2 2 1
D 1 2 1 2
Run B C E A 1 1 2 2
1 1 1 1
2 1 1 2
3 1 2 1
4 1 2 2
--`,,```,,,,````-`-`,,`,,`,`,,`---
5 2 1 1
6 2 1 2
7 2 2 1
8 2 2 2
E.5.4 Centre points
Centre points were not selected for this experiment because two of the factors in the inner array were
regarded as categorical (binary) factors making the creation of centre points very difficult.
E.5.5 Replication and repetition
The washing-machine-development laboratory could not accommodate any more experimental time or space
than that allocated to carry out the 8 × 4 experiment shown above. Therefore, there was no replication of any
of the experimental runs.
For each experimental run, there were eight machines providing repetition.

Provided by IHS
ISO/TR 12845:2010(E)
E.6 Analysis of results
E.6.1 Data acquired through the experiment and analysis considerations
The results from the experiment are given in Table E.6. They were keyed into a software programme called
MINITAB™14). Plots from this programme are shown in Figures E.1 to E.4.
Table E.6 — Experimental results
F 1 2 2 1
D 1 2 1 2
Run B C E A 1 1 2 2
1 1 1 1 278,1 367,2 463,8 310,9

2 1 1 2 342,0 355,1 451,7 253,9
3 1 2 1 393,0 406,2 356,0 349,8
4 1 2 2 421,0 411,9 530,7 377,8
5 2 1 1 431,7 444,8 541,4 433,3
6 2 1 2 581,9 519,1 615,7 462,8
--`,,```,,,,````-`-`,,`,,`,`,,`---
7 2 2 1 465,3 500,6 575,0 422,1

8 2 2 2 686,4 699,6 942,8 643,2
14) MINITAB is the trade name of a product supplied by Minitab Inc. This information is given for the convenience of users
of this Technical Report and does not constitute an endorsement by ISO of the product named.

Provided by IHS
ISO/TR 12845:2010(E)
E.6.2 Analysis of the signal-to-noise ratio
Y
B C
55
54
53
52
51
A B L P
E
55
54
53
52
51
L H
Key
Y mean of signal-to-noise ratios (signal-to-noise: larger is better)
B tub type, i.e. design A or design B

C detergent type, i.e. liquid (L) or powder (P)
E wash temperature, i.e. low (L) or high (H)
Figure E.1 — Main effects plot for signal-to-noise ratios — Data means
Table E.7 — ANOVA — Signal-to-noise ratios
Degrees
Source Seq SSa Adj SSb Adj MSc F value p value
of freedom
B Tub type 1 23,276 7 23,276 7 23,276 7 277,72 0,038
C Detergent type 1 4,783 1 4,783 1 4,783 1 57,07 0,084
E Wash temperature 1 4,209 4 4,209 4 4,209 4 50,22 0,089
BC 1 0,001 4 0,001 4 0,001 4 0,02 0,917
BE 1 1,935 3 1,935 3 1,935 3 23,09 0,131
CE 1 1,474 1 1,474 1 1,474 1 17,59 0,149
Residual error 1 0,083 8 0,083 8 0,083 8
Total 7 35,763 9
a Seq SS = Sequential sums of squares.
b Adj SS = Adjusted sums of squares.
c Adj MS = Adjusted mean square.
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
The analysis indicates that factor B is significant at the 0,05 level with C and E marginally significant at the
0,10 level. As the p values for the interaction terms are all above 0,10, they were regarded as not significant.
Because only one degree of freedom was available to assess the residual error, a probability plot was drawn
of the effects of the factors to better assist the determination of significance. This was done for all of the
responses.
Y
99
95
90
B
80
70
60
50
40
30
20
10
5
1
0,0 0,5 1,0 1,5 2,0 2,5 3,0 3,5 X
Key
X signal-to-noise effect
Y percent
B tub type
Figure E.2 — Normal probability plot of signal-to-noise effect
To improve the model, the analysis was then re-done (see E.6.3), removing the interaction terms from the
model fitted.
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
E.6.3 Analysis of MCBF means
Y
B C
560
520
480
440
400
A B L P
E
560
520
480
440
400
L H
Key
Y mean of means

Figure E.3 — Main effects plot for means — Data means
Table E.8 — ANOVA — MCBF
Degrees
of freedom
B Tub type 1 65 549 65 549 65 549 46,02 0,093
C Detergent type 1 13 778 13 778 13 778 9,67 0,198
E Wash temperature 1 18 925 18 925 18 925 13,29 0,170
BC 1 1 805 1 805 1 805 1,27 0,462
BE 1 9 769 9 769 9 769 6,86 0,232
CE 1 6 821 6 821 6 821 4,79 0,273
Residual error 1 1 424 1 424 1 424
Total 7 118 071
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
Y
99
95
B
80
50
20
5
--`,,```,,,,````-`-`,,`,,`,`,,`---
0,01
0 50 100 150 200 X
Key
X mean effect
Y percent
B tub type
Figure E.4 — Probability plot — MCBF
The analysis indicates that factor B is marginally significant at the 0,10 level. As the p values for the interaction
terms are all above 0,20, they were regarded as not significant. The analysis was then re-done, removing
those interaction terms from the model fitted.

Provided by IHS
ISO/TR 12845:2010(E)
E.6.4 Revised analysis of signal-to-noise ratio
Table E.9 — Revised ANOVA — Signal-to-noise ratios
Degrees
of freedom
B Tub type 1 23,277 23,277 23,276 7 26,64 0,007

C Detergent type 1 4,783 4,783 4,783 1 5,47 0,079
E Wash temperature 1 4,209 4,209 4,209 4 4,82 0,093
Residual error 4 3,495 3,495 0,873 7
Total 7 35,764
The analysis shown in Table E.9 confirms that factor B (tub type) is significant at the 0,05 level with
C (detergent type) and E (wash temperature) marginally significant at the 0,10 level.
E.6.5 Revised analysis of MCBF means
Table E.10 — Revised ANOVA — MCBF
Degrees
of freedom
B Tub type 1 65 549 65 549 65 549 13,29 0,022
C Detergent type 1 13 778 13 778 13 778 2,78 0,171
E Wash temperature 1 18 925 18 925 18 925 3,82 0,122
Residual error 4 19 819 19 819 19 819
Total 7 118 071
The revised analysis in Table E.10 indicates that factor B (tub type) significantly affects the MCBF at the
0,05 level but that factors C (detergent type) and E (wash temperature) are not significant. Thus, the new
design is unaffected by the detergent type used and the wash temperature.
NOTE Although in the usual application of the Taguchi method, the above analysis would be all that is performed, it
would be considered good practice if a standard deviation were calculated per run to assess the variation induced by the
--`,,```,,,,````-`-`,,`,,`,`,,`---
usage factors. This might be regarded as important here, since the signal-to-noise ratio (the larger the better) selected for
this study does not really describe variation. Clause E.7 contains the analysis of ln(S). It shows no factor significantly
affected by the variability.
E.6.6 Presentation of results — Optimization recommendations
One of the objectives for the new washing machine design was to produce a machine that maximized the
MCBF with the least variability when subjected to the usage factors water type, wash load and wash
frequency.

Provided by IHS
ISO/TR 12845:2010(E)
The signal-to-noise ratio used here cannot indicate which factors influence variability. This is only achieved
from the analysis of the standard deviation. Regarding the output in E.7, none of the factors had any effect on
the variability. Therefore, the recommendation was to select the factors to maximize the MCBF.
From the analysis of means, having factor B set to level 2 will bring a higher value of MCBF.
Given that, in practice, the user of the washing machines will choose what temperatures to wash at and what
detergent will be used, the expected MCBF will be 560,4. Given “average” washing demand, this equates to
an approximate mean time between failure of 3,6 years.
The findings of the experiment assisted the engineers in designing a very successful washing machine that
performed much better than its predecessors. The new machine was perceived by customers as very reliable.
E.7 Analysis of ln(S)
Y
B C
4,4
4,3
4,2
4,1
4,0
A B L P
E
4,4
4,3
4,2
4,1
4,0
L H
Key
Y mean of ln(S)

Figure E.5 — Main effects plot — ln(S)
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
Table E.11 — ANOVA — ln(S)
Degrees
of freedom
B Tub type 1 0,114 04 0,114 045 0,114 045 5,92 0,248

C Detergent type 1 0,018 54 0,018 539 0,018 539 0,96 0,506
E Wash temperature 1 0,432 48 0,432 478 0,432 478 22,46 0,132
B*C 1 0,592 55 0,592 551 0,592 551 30,77 0,114
B*E 1 0,001 78 0,001 779 0,001 770 0,09 0,812
C*E 1 0,233 35 0,233 350 0,233 350 12,12 0,178
Residual error 1 0,019 26 0,019 259 0,019 259
Total 7 1,412 00
Y
99
95
90
80
70
60
50
40
30
20
10
1
-0,1 0,0 0,1 0,2 0,3 0,4 0,5 0,6 X
Key
X ln(S) effect
Y percent
Figure E.6 — Normal probability plot – ln(S)
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
Annex F
(informative)
Aggregated shipworm bacterium
F.1 General
The marine shipworm hosts a symbiotic parasite commonly known as the shipworm bacterium (teredinobacter
turnirae) which has been found to have commercial applications in impacting nitrogen levels in media by
producing cellulolytic and proteolytic activities. The bacterium grows in rod-like or clump-like fashions and can
produce extracellular proteolytic activity. A more desirable growth morphology is in the form of aggregates
which, if grown at an exponential rate, can be commercially viable.
F.2 Overall objective for the experiment

The inability of the shipworm bacterium to grow exponentially in aggregates suggests the impedance owing to
a medium component. As there are many such candidates, a screening experiment was sought to identify key
inhibitors, recognizing the possibility of some interactions.
A Plackett-Burman 20-run design was chosen along with its 20-run fold-over as a likely second phase of
experimentation. Eventually, a response surface design was employed using the three most critical
components that impacted growth.
This annex adapts the experiment and results from a study by Ahuja, Ferreira and Moreira [2].
F.3 Response variable
F.3.1 Choice of variable
Cell growth was determined via an optical scanning procedure which converts observed optical density values
to dry cell weight of biomass. Proteolytic activity was determined by a protease assay, resulting in response
variable Y outcomes measured in micrograms (µg) of azocasein digested per hour.
F.3.2 Measurement of the response variable
A rigid protocol was followed for each run of the experiment and measurements were performed as described
in F.3.1 in duplicate.
F.3.3 Relationship of the response variable to the objective of the experiment
Microbial growth proceeds at a rate proportional to the biomass concentration. To capture the limitations to the
growth, the response was determined by measuring growth at 8 h and 36 h.
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
F.4 Factors affecting the response
F.4.1 Description of each factor to be varied
Various medium factors were chosen as candidates that impact growth. The components are given in the form
of their chemical composition or other description, as follows:
1 NaCl
2 KCl
3 MgSO4⋅7H20
4 MgCl2⋅6H20
5 CaCl2⋅2H20
6 HEPES
7 K2HPO4
8 Na2CO3
9 Sodium citrate⋅2H2O
--`,,```,,,,````-`-`,,`,,`,`,,`---
10 Fe2(SO4)3⋅H2O
11 H3BO3
12 MnCl2⋅4H2O
13 ZnSO4⋅7H2O
14 Na2MoO4⋅2H2O
15 CoSO4⋅6,5H2O
16 CuSO4⋅5H2O
17 Sucrose
18 NH4Cl
NOTE Components 1, 6 and 17 were ultimately held constant throughout the experiment and were used for
consistency checks by the original experimenters. Henceforth, these variables are treated as unvaried and not suitable for
further analysis.

Provided by IHS
ISO/TR 12845:2010(E)
F.4.2 Selection of levels (related to size of effect to be determined)
The factors used in the experiment and their associated levels are given in Table F.1.
Table F.1 — Factors and levels
--`,,```,,,,````-`-`,,`,,`,`,,`---
Lower level Higher level
Factor
g/l g/l
2 KCl 0,4 1,6
3 MgSO4⋅7H2O 1,9 7,6
4 MgCL2⋅6H2O 1,5 6,0
5 CaCl2⋅2H2O 0,4 1,6
7 K2HPO4 1,53 × 10−2 6,12 × 10−2
8 Na2CO3 1,00 × 10−2 4,00 × 10−2
9 Sodium citrate⋅2H2O 4,56 × 10−3 1,82 × 10−2
10 Fe2(SO4)3⋅H2O 3,14 × 10−3 1,25 × 10−2
11 H3BO3 2,90 × 10−3 1,16 × 10−2
12 MnCl2⋅4H2O 1,80 × 10−3 7,2 × 10−2
13 ZnSO4.7H2O 2,00 × 10−4 8,00 × 10−4
14 Na2MoO4.2H2O 4,00 × 10−5 1,60 × 10−4
15 CoSO4⋅6,5H2O 4,84 × 10−5 1,94 × 10−4
16 CuSO4⋅5H2O 8,00 × 10−5 3,20 × 10−4
18 NH4Cl 1 4
As is apparent from Table F.1, the higher level of each component was set at four times the lower level.
F.4.3 Other factors noted but not incorporated due to issues with controllability or relevance
No other factors were identified for inclusion in the experiment.
F.4.4 Independence of factors
The factors selected for the experiment were considered as independent of each other.
F.5 Experimental design
F.5.1 Design chosen

A 20-run Plackett-Burman design was chosen for the experiment.
NOTE 1 The numbers “−1” and “1” in Tables F.2 and F.3 refer to the lower and higher levels of the factors, respectively,
as defined in Table F.1.
NOTE 2 The design is presented in “standard order” but was run in a randomized order.

Provided by IHS
ISO/TR 12845:2010(E)
Table F.2 — Initial 20-run Plackett-Burman design factor
Run
2 3 4 5 7 8 9 10 11 12 13 14 15 16 18 Y
number
1 1 −1 −1 1 1 1 −1 1 −1 1 −1 −1 −1 −1 1 0,048
2 1 1 −1 −1 1 1 1 −1 1 −1 1 −1 −1 −1 1 0,039
3 −1 1 1 −1 1 1 1 1 −1 1 −1 1 −1 −1 −1 0,039
4 1 −1 1 1 −1 1 1 1 1 −1 1 −1 1 −1 −1 0,029
5 1 1 −1 1 −1 −1 1 1 1 1 −1 1 −1 1 −1 0,044
6 −1 1 1 −1 1 −1 −1 1 1 1 1 −1 1 −1 −1 0,041
7 −1 −1 1 1 1 1 −1 −1 1 1 1 1 −1 1 1 0,046
8 −1 −1 −1 1 −1 1 1 −1 −1 1 1 1 1 −1 −1 0,041
9 −1 −1 −1 −1 1 −1 1 1 −1 −1 1 1 1 1 1 0,026
10 1 −1 −1 −1 1 1 −1 1 1 −1 −1 1 1 1 −1 0,038
11 −1 1 −1 −1 −1 1 1 −1 1 1 −1 −1 1 1 1 0,035
12 1 −1 1 −1 −1 −1 1 1 −1 1 1 −1 −1 1 1 0,031
13 −1 1 −1 1 −1 −1 −1 1 1 −1 1 1 −1 −1 1 0,034
14 1 −1 1 −1 −1 −1 −1 −1 1 1 −1 1 1 −1 1 0,037
15 1 1 −1 1 1 −1 −1 −1 −1 1 1 −1 1 1 −1 0,033
16 1 1 1 −1 −1 1 −1 −1 −1 −1 1 1 −1 1 −1 0,036
17 1 1 1 1 1 −1 1 −1 −1 −1 −1 1 1 −1 1 0,03
18 −1 1 1 1 −1 1 −1 1 −1 −1 −1 −1 1 1 1 0,034
19 −1 −1 1 1 1 −1 1 −1 1 −1 −1 −1 −1 1 −1 0,031
20 −1 −1 −1 −1 −1 −1 −1 −1 −1 −1 −1 −1 −1 −1 −1 0,039
Subsequently, the fold-over to the 20-run Plackett-Burman design of Table F.2 was run and the results were
combined. Table F.3 provides the runs and the corresponding responses (again reporting the average of the
duplicates for the responses).
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
Table F.3 — Foldover of the Plackett-Burman design factor of Table F.2
Run
2 3 4 5 7 8 9 10 11 12 13 14 15 16 18 Y
number
1 −1 1 1 −1 −1 −1 1 −1 1 −1 1 1 1 1 −1 0,044
2 −1 −1 1 1 −1 −1 −1 1 −1 1 −1 1 1 1 −1 0,038
3 1 −1 −1 1 −1 −1 −1 −1 1 −1 1 −1 1 1 1 0,039
4 −1 1 −1 −1 1 −1 −1 −1 −1 1 −1 1 −1 1 1 0,04
5 −1 −1 1 −1 1 1 −1 −1 −1 −1 1 −1 1 −1 1 0,039
6 1 −1 −1 1 −1 1 1 −1 −1 −1 −1 1 −1 1 1 0,039
7 1 1 −1 −1 −1 −1 1 1 −1 −1 −1 −1 1 −1 −1 0,043
8 1 1 1 −1 1 −1 −1 1 1 −1 −1 −1 −1 1 1 0,041
9 1 1 1 1 −1 1 −1 −1 1 1 −1 −1 −1 −1 −1 0,039
10 −1 1 1 1 −1 −1 1 −1 −1 1 1 −1 −1 −1 1 0,043
11 1 −1 1 1 1 −1 −1 1 −1 −1 1 1 −1 −1 −1 0,05
12 −1 1 −1 1 1 1 −1 −1 1 −1 −1 1 1 −1 −1 0,051
13 1 −1 1 −1 1 1 1 −1 −1 1 −1 −1 1 1 −1 0,051
14 −1 1 −1 1 1 1 1 1 −1 −1 1 −1 −1 1 −1 0,052
15 −1 −1 1 −1 −1 1 1 1 1 −1 −1 1 −1 −1 1 0,049
16 −1 −1 −1 1 1 −1 1 1 1 1 −1 −1 1 −1 1 0,052
17 −1 −1 −1 −1 −1 1 −1 1 1 1 1 −1 −1 1 −1 0,048
18 1 −1 −1 −1 1 −1 1 −1 1 1 1 1 −1 −1 −1 0,052
19 1 1 −1 −1 −1 1 −1 1 −1 1 1 1 1 −1 1 0,048
20 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 0,049
F.5.2 Centre points
Centre points were not selected for this experiment as the experiment is a screening experiment and the
relative magnitudes of the various components is of greater interest than their pure statistical significance.
F.5.3 Replication and repetition
Measurements were performed as repeats. The two values obtained are not pure replicates, hence, only the
average of the two values for each run are reported in Tables F.2 and F.3.
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
F.6 Analysis of results
F.6.1 Data acquired through the experiment and analysis considerations
The results from the experiment are given in Table F.4. These results correspond to combining the original
20 runs with the fold-over runs and estimating parameters. The responses were entered into a software
programme called JMP7™15).
Table F.4 — Experimental results — Parameter estimates
Term Estimate Standard error t ratio Prob > |t|
Intercept 0,040 95 0,001 194 34,29 < 0,000 1

×2 −0,000 15 0,001 194 −0,13 0,901 1
×3 −0,000 2 0,001 194 −0,17 0,868 4
×4 −0,001 1 0,001 194 −0,92 0,366 2
×5 0,000 15 0,001 194 0,13 0,901 1
×7 0,001 45 0,001 194 1,21 0,236 6
×8 0,001 55 0,001 194 1,30 0,206 7
×9 0 0,001 194 0,00 1,000 0
×10 0,000 75 0,001 194 0,63 0,536 0
×11 0,000 95 0,001 194 0,80 0,434 2
×12 0,001 8 0,001 194 1,51 0,144 8

--`,,```,,,,````-`-`,,`,,`,`,,`---
×13 0,000 05 0,001 194 0,04 0,967 0
×14 0,000 6 0,001 194 0,50 0,620 0
×15 −0,001 05 0,001 194 −0,88 0,388 1
×16 −0,001 2 0,001 194 −1,00 0,325 1
×18 −0,001 0,001 194 −0,84 0,410 7
The three leading effects are: 7 (K2HPO4), 8 (Na2CO3) and 12 (MnCl2⋅4H2O).
These effects can also be identified from the half-normal probability plot given in Figure F.1.
15) JMP7 is the trade name of a product supplied by SAS Institute Inc. This information is given for the convenience of
users of this Technical Report and does not constitute an endorsement by ISO of the product named.

Provided by IHS
ISO/TR 12845:2010(E)
Y
0,002
x12
x8
0,001 5
x7
0,001
0,000 5
0
0,0 0,5 1,0 1,5 2,0 2,5 3,0 X
Key
X normal quantile
Y estimate
NOTE The blue line is Lenth's PSE, from the estimates population.
Figure F.1 — Half-normal plot
F.6.2 Pareto chart of effects
The same ranking of effects is observed in the Pareto plot given in Figure F.2.
Term Estimate
x12 0,001 800 0
x8 0,001 550 0
x7 0,001 450 0
x16 -0,001 200 0
x4 -0,001 100 0
x15 -0,001 050 0
x18 -0,001 000 0
x11 -0,000 950 0
x10 -0,000 750 0
x14 -0,000 600 0
x3 -0,000 200 0
x5 -0,000 150 0
x2 0,000 150 0
x13 0,000 050 0
x9 0,000 000 0
Figure F.2 — Pareto plot of effects
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
The effects plot given in Figure F.3 provides an indication of the impacts of the three leading effects.
0,055
±0,002 465
0,040 95
0,045
y
0,035
0,025
0
0,5
0,5
1
1
0
0,5
0
1
-0,5
-1
-1
-0,5
-0,5
-1
0 0 0
x7 x8 x12
Figure F.3 — Estimation of the response as a function of the effects ×7, ×8, ×12
Each effect can be estimated by the average response for the effect at the high level minus the average
response for the effect at the low level. The standard error of effects is obtained in the software using the
usual regression framework (there are 24 degrees of freedom for estimating error in the combined analysis).
The objective of the experiment was to identify the leading inhibitors of the aggregate growth which diminishes
the interest in the statistical significance aspects.
F.7 Presentation of findings

Three factors were identified for subsequent use in optimizing the conditions present in the medium for
culturing the bacterium. Estimates of the effects can be carried out individually on each of the 20 original runs
and the 20 fold-over runs. These produced different combinations of leading factors which is attributable to the
presence of interactions. By folding over the original design and combining the results, a resolution IV design
was achieved which gave main effects clear of two-way interactions. A follow-up response surface experiment
built upon the lessons learned in the screening experiment.
F.8 Comments on design choice

As originally described, the experimenters focused on screening 18 factors which were potential inhibitors of
aggregate growth of shipworm bacterium. The 20-run Plackett-Burman design is excellent for this
circumstance. Recognizing the potential impact of aliasing of main effects with two-way interactions, they
chose to augment the original 20 runs with the 20-run fold-over set of runs, yielding a resolution IV design.
In executing the design, the experimenters chose to fix three of the original 18 factors, so that only 15
15-10
variables were actually varied. Under this situation, an alternative design would have been the 32-run 2
fractional factorial design which is resolution IV. Perhaps a greater level of sophistication is necessary to
implement this alternative design.
In any event, to the credit of the experimenters, they were able to shepherd the resources to conduct a 40-run
experiment that ultimately identified three factors contributing to the inhibition of aggregate growth of
shipworm bacterium that subsequently were optimized to obtain improved biomass production. The
Plackett-Burman experiments used are vastly superior to one-factor-at-a-time strategies or other haphazard
approaches.
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
ISO/TR 12845:2010(E)
Bibliography
[1] ISO/TR 29901, Selected illustrations of full factorial experiments with four factors
[2] AHUJA, S.K., FERREIRA, G.M. and MOREIRA, A.R. Application of Plackett-Burman Design and Response
Surface Methodology to Achieve Exponential Growth for Aggregated Shipworm Bacterium.
Biotechnology and Bioengineering, 85(6), 2004, pp. 666-675
[3] BOX, G.E.P., HUNTER, J.S. and HUNTER, W.J. Statistics for Experimenters — Design, Innovation, and
Discovery. New York: Wiley, 2005
[4] DEAN, A. and VOSS, D. Design and Analysis of Experiments. New York: Springer-Verley, 1999
[5] JOHNSON, D.E. and MILLIKEN, G.A. Analysis of Messy Data, Volume 1 — Designed Experiments. New
York: Chapman & Hall, 1992
[6] LEDOLTER, J. and SWERSEY, A.J. Using a Fractional Factorial Design to Increase Direct Mail Response.
Quality Engineering, 18, 2006, pp. 469-475
[7] LENTH, R.S. Quick and easy analysis of unreplicated factorials. Technometrics, 31, 1989, pp. 469-473
[8] OHLERT, G.W. A First Course in Design and Analysis of Experiments. New York: W.H. Freeman and
Company, 2000
[9] WU, C.F.J. and HAMADA, M. Experiments: Planning, Analysis, and Parameter Design Optimization.
New York: Wiley, 2000
--`,,```,,,,````-`-`,,`,,`,`,,`---

Provided by IHS
--`,,```,,,,````-`-`,,`,,`,`,,`---
Provided by IHS
ISO/TR 12845:2010(E)
--`,,```,,,,````-`-`,,`,,`,`,,`---
ICS 03.120.30
Price based on 80 pages

Provided by IHS

Iso TR 12845-2010

Uploaded by

Copyright:

Available Formats

Iso TR 12845-2010

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Iso TR 12845-2010

Uploaded by

Copyright:

Available Formats

TECHNICAL ISO/TR

Selected illustrations of fractional

COPYRIGHT PROTECTED DOCUMENT

ii © ISO 2010 – All rights reserved

© ISO 2010 – All rights reserved iii

iv © ISO 2010 – All rights reserved

1) Six Sigma is a trade mark of Motorola, Inc.

© ISO 2010 – All rights reserved v

Selected illustrations of fractional factorial screening

ISO 3534-2, Statistics — Vocabulary and symbols — Part 2: Applied statistics

ISO 3534-3, Statistics — Vocabulary and symbols — Part 3: Design of experiments

3 Terms and definitions

2) To be published. (Revision of ISO 3534-3:1999)

© ISO 2010 – All rights reserved 1

where x = 0, 1, …, n and with indexing parameters n = 1, 2, …, and 0 < p < 1.

2 © ISO 2010 – All rights reserved

NOTE 4 A factor may be associated with the creation of blocks.

[ISO 3534-3:—, definition 3.1.4]

© ISO 2010 – All rights reserved 3

[ISO 3534-3:—, definition 3.2.6]

4 Symbols and abbreviated terms

AB, AC, AD, BC, BD, CD Two-way interactions

ABC, ACD, BCD Three-way interactions

ABCD Four-way interactions

+1, −1 High and low settings

24 Four factors each at two levels

4 © ISO 2010 – All rights reserved

5 Generic description of fractional factorial designs

5.1 Overview of the structure of the examples in Annexes A to F

Table 1 — Basic steps in experimental design

Step 1 State the overall objective(s) of the experiment

5.2 Overall objective(s) of the experiment (Step 1)

5.3 Response variable(s) (Step 2)

Associated with the objective of an experiment is a measurable outcome or performance measure. A

5.4 Factors affecting the response(s) (Step 3)

© ISO 2010 – All rights reserved 5

5.5 Select a fractional factorial design (Step 4)

Table 2 — Selected design resolutions

Table 3 — Layout of a generic 24 full factorial design

© ISO 2010 – All rights reserved 7

5.7 Present the findings (Step 6)

5.8 Perform confirmation runs (Step 7)

a) the original experiment was correctly conducted,

b) the findings still hold true at a later point in time, and

6.1 Comparing and contrasting the examples

6.2 Experiment summaries

8 © ISO 2010 – All rights reserved

Table 4 — Experiment summaries found in annexes

Annex Experiment Business area Resolution Problem-specific aspects

A Direct mail campaign Marketing IV Proportion response variable, standard errors

© ISO 2010 – All rights reserved 9

Direct mail marketing campaign

A.2 Overall objective for the experiment

A.3 Description of the process

A.4 Response variable

A.4.1 Choice of variable

A.4.2 Measurement of the response variable

10 © ISO 2010 – All rights reserved

A.4.3 Relationship of the response variable to the objective of the experiment