See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/329307904

Factors of safety and probabilities of failure in geotechnical engineering:

What do we mean?

Article in Civil Engineering/Siviele Ingenieurswese · April 2018

CITATIONS READS

3 1,938

1 author:

Charles MacRobert
Stellenbosch University
32 PUBLICATIONS 85 CITATIONS

SEE PROFILE

All content following this page was uploaded by Charles MacRobert on 30 November 2018.

The user has requested enhancement of the downloaded file.

Factors of safety and probabilities of failure in
geotechnical engineering: What do we mean?
INTRODUCTION Vick (2002) and the USBR-USACE (2015),
Dr Charles MacRobert
Geotechnical engineers assign factors showing order of magnitude differences Lecturer Geotechnical Engineering
of safety or probabilities of failures to for the low probability terms. Hubble University of the Witwatersrand
geotechnical structures. It is often neces- (2010) suggested some terms that can be Charles.MacRobert@wits.ac.za
sary to describe to others what we mean associated with factors of safety when
by these factors of safety or probabilities considering river bank stability (Table 2).
of failure in verbal terms. A study of terms A study of engineering judgement within that outcome occurring?” Table 3 shows
used in various fields (Reagan et al 1989) geotechnical engineering is under way at the that, for each synonyms pair, failure
identified four stems used to describe University of the Witwatersrand. This study modes were compared within form, and
the likelihood of an occurrence: (1) pos- is seeking to determine whether the often geotechnical structures were compared
sible (almost impossible, possible, very tacit nature of engineering judgement can be between forms. Questions were presented
possible); (2) probable (very improbable, codified. As part of the project, two studies randomly on each form.
improbable, probable, very probable); have been conducted looking at the verbal The questionnaire was distributed at
(3) likely (very unlikely, unlikely, likely, expressions associated with probabilities of the 9th South African Young Geotechnical
very likely); and (4) chance (very low failure and factors of safety. Results from Engineers (SAYGE) Conference and at the
chance, low chance, medium chance, even these studies are presented and discussed. SAICE Geotechnical Division AGM, both
chance, high chance, very high chance). held in 2017. Fifty-two forms were returned
The term “certain” can also be included as STUDY 1: VERBAL TERMS ASSOCIATED at the SAYGE Conference and seven forms
an expression of possible. With so many WITH PROBABILITIES OF FAILURE were returned at the AGM. Five forms
possible terms, the exact meaning of a Two questionnaires were developed to filled in incorrectly were not used. Six
phrase can be misinterpreted and there- assess whether different numerical values forms with some blank numerical values
fore requires clarification. would be associated with different verbal were used by ignoring the blank values.
Numerous studies have been carried terms if the verbal terms were used to By calculating the median response
out to establish what numerical values describe different geotechnical structures for each question on form A and form B,
of probability are associated with these and failure modes. The questionnaire was and then comparing the equivalent
expressions, and whether these differ based on six pairs of verbal terms found to responses between forms (e.g. a tailings
between experts and lay persons. Most of be synonyms by Reagan et al (1989). Two dam – almost improbable – piping,
these studies showed considerable agree- geotechnical structures were considered versus an earth dam – almost improb-
ment and illustrated synonymity between (tailings dams and earth dams) and two able – piping) it can be shown that the
various terms (Reagan et al 1989). Fewer failure modes were considered (piping degree of correlation between the two
studies have been carried out specifically and slope instability). Questions posed to forms is 0.996 (Figure 1). This suggests
with engineering practitioners, and respondents were structured as follows: “If that the geotechnical structure associ-
even fewer studies have been carried out someone told you a geotechnical structure ated with a verbal term had little bearing
looking at factors of safety. has a verbal expression to fail by failure on the numerical value. Consequently,
Table 1 gives verbal terms associated mode, what decimal between 0 and 1
with probabilities of failure suggested by would best represent the probability of Table 2 V
 erbal terms associated with factors
of safety (Hubble 2010)
Table 1 Verbal terms associated with probabilities of failure Probable bank
Factor of safety
USBR-USACE (2015) stability condition
Descriptor Vick (2002) probability
probability Greater than 2.00 Stable
Virtually certain 0.999 0.99 1.51–2.00 Probably stable
Very likely 0.99 0.9 1.31–1.50 Moderately stable
Likely 0.9 – 1.10–1.30 Conditionally stable
Neutral 0.5 – 1.00–1.10 Critically stable

Equally likely – 0.5 0.91–0.99 Unstable

Unlikely 0.1 – 0.71–0.90 Very unstable

Very unlikely 0.01 0.1 0.50–0.70 Highly unstable

Virtually impossible 0.001 0.01 Less than 0.50 Extremely unstable

Civil Engineering April 2018 45

Table 3 Verbal-numerical probability questionnaires

Pair Form A Form B

a tailings dam – almost improbable – piping an earth dam – almost improbable – piping
1
a tailings dam – very improbable – slope instability an earth dam – very improbable – slope instability

a tailings dam – very unlikely – piping an earth dam – very unlikely – piping
2
a tailings dam – very low chance – slope instability an earth dam – very low chance – slope instability

a tailings dam – unlikely – piping an earth dam – unlikely – piping

3
a tailings dam – low chance – slope instability an earth dam – low chance – slope instability

an earth dam – medium chance – piping a tailings dam – medium chance – piping
4
an earth dam – even chance – slope instability a tailings dam – even chance – slope instability

an earth dam – probable – piping a tailings dam – probable – piping

5
an earth dam – likely – slope instability a tailings dam – likely – slope instability

an earth dam – very high chance – piping a tailings dam – very high chance – piping
6
an earth dam – almost certain – slope instability a tailings dam – almost certain – slope instability

further comparison is based purely on

100 100
verbal terms and no distinction is made
on question phrasing.
75 75
To generate Figure 2, the numerical
Responses %

Responses %
values associated with each verbal expres-
50 50
sion were divided into bins in increments
of 0.05 and counted. The vertical scale
25 25
is therefore the frequency of responses
within each 0.05 increment expressed as a
0 0
percentage of total responses. It is evident 0
0.25 0.50 0.75 1 0 0.25 0.50 0.75 1
from this chart that there is strong cor- (a) Numerical probability (b) Numerical probability
relation between synonymous terms, Almost certain Very high chance Likely Probable
suggesting that failure modes had little
bearing on assigned probabilities. 100 100
Table 4 summarises median values
obtained in this study and compares them 75 75
Responses %

Responses %

to median values suggested by Reagan et

al (1989) for undergraduate psychology 50 50
students. The degree of correlation be-
tween these two studies is 0.994. Table 4 25 25
also shows the correlation between terms,
calculated by comparing the number of 0 0
terms in each bin, showing strong correla- 0.250 0.50 0.75 1 0 0.25 0.50 0.75 1
(c) Numerical probability (d) Numerical probability
tion between synonymous terms, despite
Even chance Medium chance Unlikely Low chance

1 100 100
Median

75 75
Responses %

Responses %
Form B

50 50
0.5

Line of unity 25 25

0 0
0 0
0.25 0.50 0.75 1 0 0.25 0.50 0.75 1
0 0.5 1 (e) Numerical probability (f) Numerical probability
Form A Very low chance Very unlikely Almost improbable Very improbable

Figure 1 Between form comparisons Figure 2 Numerical values associated with verbal terms for probability of failure

46 April 2018 Civil Engineering

different failure modes. One difference Table 4 Comparison between terms
between the current study and that of Reagan et al
Reagan et al (1989) is that the terms “al- Current study
(1989)
Term Verbal term
most improbable” and “very improbable”
Median Median Correlation Correlation
are seen to be synonymous with “very low
chance” and “very unlikely”.
1 Almost improbable 0.02 0.10
It is well known that people find as- 1 vs 2
signing low probabilities difficult (Vick 0.990
2 Very improbable 0.05 0.10
2002). However, it would be expected that 2 vs 3
0.897
engineers, who design so that the prob- 3 Very low chance 0.10 0.10
3 vs 4
ability of failure is extremely low, would
0.959
be better able to differentiate low prob- 4 Very unlikely 0.10 0.10
4 vs 5
abilities and associated terms. Whilst on 0.857
the whole it would appear that engineers 5 Low chance 0.20 0.20
5 vs 6
assign similar numerical values of prob- 0.885
6 Unlikely 0.15 0.20
ability to lay persons, some respondents 6 vs 7
in the current study suggested extremely 0.124
7 Even chance 0.50 0.50
small values of probabilities for the low 7 vs 8
0.931
probability terms. Thirty percent of the 8 Medium chance 0.50 0.50
8 vs 9
respondents suggested numerical values
0.234
of probability less than 0.05. To illustrate 9 Likely 0.70 0.70
9 vs 10
these responses, numerical values were 0.691
divided into order of magnitude bins 10 Probable 0.70 0.60
10 vs 11
and plotted to a log scale (Figure 3). This 0.292
11 Almost certain 0.90 0.90
shows that, even when extremely small 11 vs 12
numerical values of probability are given, 0.882
12 Very high chance 0.90 0.85
there is synonymity between terms (see

Civil Engineering April 2018 47

Table 5 Comparison between terms for low probabilities of failure also Table 5). The geometric mean of the
Term Verbal term Geometric mean Correlation Correlation numerical responses for the low prob-
ability terms shows closer agreement to
1 Almost improbable 0.04 1 vs 2
the terms suggested by Vick (2002) and
2 Very improbable 0.04 0.979 2 vs 3 the USBR-USACE (2015). There is strong
0.991 correlation between all these terms, and
3 Very low chance 0.06 3 vs 4
it is difficult to distinguish if they can be
4 Very unlikely 0.05 0.996 4 vs 5 considered distinct verbal expressions for
5 Low chance 0.13 5 vs 6 0.857 distinct numerical probabilities of failure.

6 Unlikely 0.08 0.975

STUDY 2: VERBAL TERMS ASSOCIATED
WITH FACTORS OF SAFETY
At the end of 2017, seven students and fifty
100
practitioners of geotechnical engineering
participated in a study to predict the factor
75 of safety of a tailings dam. Participants
were not informed that the exercise was
Responses %

based on a relatively unknown case study

50 of a tailings dam that had failed. The
failure resulted when the facility was 15 m
above the starter wall, by rotational failure
25
through highly plastic material, that had
been placed by spigotting behind an 8 m
0 high starter wall (Blight 2010).
–7 –6 –5 –4 –3 –2 –1 0 The exercise required respondents to
Numerical probability 10^ predict a factor of safety and suggest a
(a) Unlikely Low chance
verbal expression (same expressions as
Study 1 listed randomly) to describe the
100 likelihood of failure at three points during
the exercise. The first was after partici-
pants had considered a largely qualitative
75
description of the facility before failure;
Responses %

the second was after respondents had

50 sketched a likely cross-section and pro-
vided material parameters; and the third
after a factor of safety had been calculated
25 from a predefined limit equilibrium sta-
bility model in which respondents could
only change material strength parameters
0
–7 –6 –5 –4 –3 –2 –1 0 (respondents were not required to use
Numerical probability 10^ the actual calculated factor of safety).
(b) Very unlikely Very low chance Whilst a considerable amount of data was
captured in this exercise, consideration
100 is given only to the predicted factors of
safety and associated verbal terms.
Figure 4 illustrates all suggested factors
75
of safety and associated verbal terms, with
Responses %

no distinction as to which stage of the ex-

50 ercise the respondent was answering. The
area under all the curves in all the charts
sums to 100%. Fifty-two percent of the
25 responses suggested that the probability
of failure was in the range of unlikely to
very improbable, while forty-eight percent
0
–7 –6 –5 –4 –3 –2 –1 0 suggested there was a medium chance to
Numerical probability 10^ almost certain probability of failure. It is
(c) Very improbable Almost improbable evident from the plots that there is similar
synonymity between terms, as was evident
Figure 3 Numerical values associated with verbal terms for low probabilities of failure in Study 1. However, as the responses

48 April 2018 Civil Engineering

were not evenly distributed between the
10 10
different terms, no statistical test of cor-
relation was undertaken.

Responses %

Responses %
COMPARING THE STUDIES
5 5
Various attempts have been made to relate
factors of safety to probabilities of failure.
The key problem in doing this is quan-
tifying the variability due to the input
0 0
parameters (load and resistance) and due 0.5 0.8
1.0 1.3 1.5 1.8 2.0 0.5 0.8 1.0 1.3 1.5 1.8 2.0
to the model used to analyse the problem. (a) Factor of safety (b) Factor of safety
A complete picture, however, also needs Almost certain Very high chance Likely Probable
to include the manner in which the
structure is constructed and maintained. 10 10
Lambe et al (1988) suggested four curves,
based largely on engineering judgement,
Responses %

Responses %
to relate factors of safety to probabilities
of failure for slopes: 5 5
I. Design by qualified engineer, full-time
construction supervision and con-
tinuous maintenance by trained crew.
II. Design by qualified engineer, part-time 0 0
construction supervision and routine 1.0 1.3 1.5 1.8 2.0
0.5 0.8 0.5 0.8 1.0 1.3 1.5 1.8 2.0
(c) Factor of safety (d) Factor of safety
maintenance.
Even chance Medium chance Unlikely Low chance
III. Approximate design using inferred
parameters, informal construction
supervision and maintenance limited 10 10

to emergency repairs.
IV. No rational design, no construction
Responses %

Responses %

supervision and no maintenance.

5 5
The curves by Lambe et al (1988) suggest
that for a given factor of safety, the better
the design, construction and supervi-
sion, the lower the probability of failure.
0 0
Gover (2017) suggested three curves 1.0 1.3 1.5 1.8 2.0
0.5 0.8 0.5 0.8
1.0 1.3 1.5 1.8 2.0
relating factors of safety and probabilities (e) Factor of safety (f) Factor of safety
of failure for slopes based on different Very low chance Very unlikely Almost improbable Very improbable
coefficients of variation for the friction
angle (c = 0 analysis) using a first order Figure 4 Verbal expressions associated with factors of safety
reliability method. These curves suggest
that for a given factor of safety, the lower whereas responses in Study 2 were studies. By comparing the results to the
the variability in strength, the lower the made as part of an actual slope sta- various curves, it appears that engineers
probability of failure. bility problem. assume a large amount of variability in
Results from the current study can be NN Study 1 was conducted within the strength, assume that designs are ap-
compared to these various curves to gain broad spectrum of geotechnical en- proximate or perhaps not even rational,
an understanding of the level of variability gineering. Apart from the students in that construction is informal or perhaps
that practitioners tacitly adopt in their Study 2, all participants in this study not supervised and very little mainte-
predictions. This is done by comparing me- were within the tailings sector. nance is carried out. This suggests that
dian and geometrical means of numerical NN Whilst sufficient information was engineers show very little confidence in
probabilities of failure to median factors provided in Study 2 to carry out a their analysis. Whilst the preceding state-
of safety associated with each verbal term. slope stability analysis, participants ments have been written in general terms,
In making this comparison, the following only had 30 to 45 minutes to complete it is accepted that they may only apply
limitations are acknowledged: the entire exercise. within the constraints of the two studies
NN Only a handful of respondents partici- NN The comparison only considers average undertaken.
pated in both studies. Therefore, the responses and hides the underlying
comparison is not necessarily between variability. CONCLUSIONS
similarly minded persons. Figure 5 compares factors of safety to This article showed that geotechnical
NN Responses in Study 1 were somewhat probabilities of failure based on the cur- engineering practitioners and lay
abstracted from a real problem, rent study and those proposed in other persons assign similar numerical values

Civil Engineering April 2018 49

 9th South African Young Geotechnical
1.00 Engineers Conference, SAICE, 93–102.
Median
Geometric mean Hubble, T C T 2010. Improving the stream
of consciousness: A nomenclature
for describing the factor of safety
Probability of failure

Lambe et al in river bank stability analysis.

(1988) Level IV
Ecological Engineering, 36: 1765–1768.
0.10
doi: 10.1016/j.ecoleng.2010.07.001.
Lambe, T W, Silva, F & Lambe, P C 1988.
Expressing the level of stability of a
Gover (2017)
CoV = 15% slope. The Art and science of geotechnical
Lambe et al engineering: at the dawn of the twenty-first
(1988) Level III
century: a volume honoring Ralph B. Peck,
0.01
0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2.0 R B Peck & E J Cording, eds. Prentice Hall,
Factor of safety 558–588.
Reagan, R T, Mosteller, F & Youtz, C 1989.
Figure 5 Comparison of factors of safety and probabilities of failure Quantitative meanings of verbal
probability expressions. Journal of
of probability to verbal expressions of unknowns and therefore show little confi- Applied Psychology, 74(3): 433–442.
probability. The study also showed that dence in their analysis. doi: 10.1037/0021-9010.74.3.433.
practitioners assign similar verbal expres- USBR-USACE 2015. Best Practices in Dam
sions of probability to similar factors of REFERENCES and Levee Safety Risk Analysis, U.S. Bureau
safety. Comparing factors of safety to Blight, G E 2010. Geotechnical engineering of Reclamation. U.S. Army Corps of
probabilities of failure for the same verbal for mine waste storage facilities. Taylor & Engineers.
expression, suggests that practitioners Francis Group, London, UK. Vick, S G 2002. Degrees of belief: Subjective
of geotechnical engineering often think Gover, S M 2017. Linking safety factor probability and engineering judgment.
that they are dealing with significant to probability of failure. Proceedings, ASCE Publications.
Ad.pdf 1 2018/04/06 12:10:24
50 April 2018 Civil Engineering

View publication stats