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

net/publication/245545699

Analysis of structural failures

Article in Proceedings of the Institution of Civil Engineers · January 1977

DOI: 10.1680/iicep.1977.3259

CITATIONS READS

48 7,511

1 author:

David Ian Blockley

University of Bristol
246 PUBLICATIONS 2,545 CITATIONS

SEE PROFILE

All content following this page was uploaded by David Ian Blockley on 26 August 2016.

The user has requested enhancement of the downloaded file.

Proc. Insrn Ciu. Engrs, Part I, 1977, 62, Feb., 51-74

Analysis of structural failures

D. 1. BLOCKLEY, BEng, PhD, MIStructE, MICE*

In Part 1 a classification of basic types of structural failure is presented. This classifica-

tion is expanded intoa set of parameter statementswhich could be assessed subjectively
in a prediction process. This process is intended to account for a structure failing due
to causes other than stochastic variations in load and strength. The parameters are
assessed for 23 major structural accidents and one existing structure, andare analysed
using a simple numerical interpretation. The accidents are ranked in their order of
inevitability. Human errors of one formor another proved to be the dominant reasons
for the failures considered. A simplified form of the proposed procedurefor predicting
the likelihood of structural accidents is outlined in Part 11. This is then applied to the
24 accident parameter assessments made in Part I. The concept of fuzzy sets is used.

Part I
There has been in recent years an upsurge of interest amongst engineers in
matters related to structural accidents. Reports of inquiries into recent acci-
dents have become compulsive reading, whilst at the same time the redrafting
of codes of practice into the limit state format has stimulated inquiry into the
use of probability theory to determine suitable partial factors. An increasing
concern about the way actual structures behave rather than idealized theoretical
models or isolated laboratory tests on physical models or elements of structures
is another aspect of this interest.
2. One of themajor difficulties in discussing this subject is the problem
which has existed since Galileo, the communicationsgap that often exists
between the practising engineer andthe research worker. Many engineers
believe that, for instance, probability theory will be of little help in understanding
the basic causes of structural failures. Slightly altering the values used for
safety factors (bethey total or partial) will be of little consequence in thisregard.
I t is difficult, on the other hand, to wrap up problems such as poor site control,
errors of judgement, and pressures due to shortage of time and money, into
mathematical models or rules of procedure to be written into codes of practice.
3. Probability theory is a mathematics of uncertainty. Its use requires a
different way of thinking about mathematics than the traditional mathematics
most engineers were taught. Many engineers cannot see any reason why they
should bother. Most research workers in the field of structural safety would
agree that probability theory is an essential tool in helping to predict the actual
behaviour of a structure.
4. The purpose of this Paper is to review and classify somestructural
failures and to consider ways in which predictions of the likelihood of possible
future structural accidents can be performed.
Ordinary meeting, 5.30 p.m., 15 March, 1977. Written discussion closes 15 April, 1977, for
publication in Proceedings, Part 1 .
Lecturer in Civil Engineering, University of Bristol.
51
BLOCKLEY
Design process
5. Inorderto discuss a classification of structural accidents it may be
instructive first briefly to examine the design process in a very general overall
way. Fig. 1 shows a simplified decision tree demonstrating the overall decision
routes facing the structural designer. This is intended to be a conceptual model
made at the design stage of the general decisions yet to be faced. The first two
decision paths to be taken (those of alternative structural forms) are the de-
signer’s choice using mainly professional information. Thethirdpath,the
construction process, will be partially under the control of the contractor and
designer; and the final path, the use of the structure, will be largely the conse-
quence of these earlier decisions:
6. Thereare obviously many overall design solutions to any problem. A
single storey shed may, for instance, be a concrete portal, a steel portal ora steel
truss: this is the first decision path. There is a multitude of detailed design
decisions concerned with the spacing of frames, the lateral bracing, design of
joints, etc., which are represented by a single line in Fig. 1 : the second decision
path. If anyone of these paths is taken,costs will be incurred. It is the
designer’s task to minimize these costs. Probability theory can help with these
decisi0ns.l The designer has to make these decisions so that it is a virtual
certainty that the structure will be constructed successfully and will be used
successfully. However, there is always a small, but finite, chance or probability
that one of the events listed in Table 1 will occur and the structure will be
damaged or collapse, and a cost penalty incurred.
7. The central problem of the designer is to design his structure economically
and aesthetically so that the probability of each of these unacceptable events or
accidents is acceptably low. In order to make his decisions the designer has
certain tools at his disposal. These tools are basically his knowledge of struc-
tural theory, the research and development information available and his
professional knowledge and experience. However, he is also subject to con-
straints such as government regulations. Having chosen the overall type of

Table 1. Some causes of structural failure

Limit states
Overload: geophysical, dead, wind, earthquake, etc.;
man-made, imposed, etc.
Understrength: structure, materials,
instability
Movement: foundation settlement, creep, shrinkage, etc.;
Deterioration: cracking, fatigue, corrosion, erosion, etc.
Random hazards
Fire
Floods
Explosions: accidental, sabotage
Earthquake
Vehicle impact
Human based errors
Design error: mistake, misunderstanding of structure behaviour
Construction error: mistake, bad practice, poor communications

52
 A N A L Y S I S OF S T R U C T U R A LF A I L U R E S : I

Alternative
detailed solutions
-11'. Constructlon
& Use

(Cost) Success
(Benefit)

(cost) (cost)

k
Fig. 1. Overall project decision tree
.I
structure using his professional knowledge, the designer has to make his detailed
design decisions on the basis of a theoretical model coupled with research and
development information, codes of practice, etc. The classical way of avoiding
the unpleasant events labelled as limit states in Table 1 is to use the well-known
working stress concept. Ultimate load and now limit state concepts are more
modern approaches. Here the decisions are based upon a complex procedural
model (cpm) which is usually the combination of elastic theory or plastic theory,
with research and development information often distilled by research workers
and various committees into recommended procedures.
8. Certain of these procedural models are good in that they are well proven
by analysis and laboratory testing, usually on elements of the final structure, and
designers can use these procedures with some confidence. Much of the dif-
ference between the performance of the structural element in the laboratory and
in the structure is due to the random nature of the parameters describing the
system. Here then the probability of an extremely high value of load or an
extremely low value of strength causing collapse can be calculated with some
confidence if the variabilities of the various parameters are known. This has
been done, for example on steel portal frames and concrete slabs.lVa
9. In contrast the behaviour of a foundation on a compressible soil or the
cumulative damage of fatigue loading cannot be predicted at all accurately. In
this situation there are two types of uncertainty: the system itselfis poorly
understood; and there is a high random variability of the parameters describing
the system. In trying to understand the behaviour of this sort of system
probability theory could well be used much more in the basic research work.
10. However, designing to the limit states of Table 1 is only part of the
problem. The precautions to be taken by the designer against such random
hazards asfire and earthquakes obviously depend on the likelihood of occurrence
of the hazard and its consequences. Thus fire is a major hazard to all buildings
but earthquakes in Britain for most structures can be neglected. However, in
the case of nuclear reactors the consequences of failure due to an earthquake
53
BLOCKLEY
could be enormous and the possibility of earthquake loading should be con-
sidered.
11. The designer tries to ensure that the third groupof events in Table 1, the
human based errors, do not occur. He does this by good professional practice
and communication. He does not take the possibility of a failure due to this
sort of reason directly into accountin his design calculations. However, in
solving a particular problem he may recognize that certain structural solutions
may be more susceptible to this type of error than others although this is a
factor difficult to quantify and would rarely change a major design decision.
12. Under normal circumstances the designer cannot make his design de-
cisions with detailed knowledge of the contractors’plans, but of course the good
designer is always aware of possible construction problems. Humanerror
during the construction phase can, for the most part, be prevented by good
communications between all parties concerned and well-defined responsibilities
under the contract.
Classification of failures
13. With this discussion in mind and based on a reading of reports of 23
structural failure^^-^^ and more general reference^,^*-^^ a classification of basic
types of structural failure was formulated. It must be appreciated that any one
accident will in general involve a number of basic types. Errors may occur at
any stage in the decision route (Fig. 1) and will generally result in one of the un-
desirable events listed in Table 1.
14. The basic types proposed are as follows.
(a) structures, the behaviour of which are reasonably well understood by
the designers (and consequently the calculation procedural models
are good), but which fail because a random extremely high value of
load or extremely low value of strength occurs (e.g. excessive wind
load, imposed load, inadequate beam strength);
(6) structures which fail due tobeing overloaded or tobeing under strength
(as (a)), but where the behaviour of the structureis poorly understood
by the designer and the system errors in the calculation procedural
models are aslarge as the random errorsin the parameters describing
the model; the designer here is aware of the difficulties (e.g. foundation
movements, creep, shrinkage, cracking, cumulative damage, dura-
bility generally);
(c) structural failures where some independent random hazard is the cause,
e.g. earthquake, fire, floods, explosion, vehicle impact; the incidence
of this type can be obtained statistically;
( d ) failures which occur because the designers do not allow for some basic
mode of behaviour inadequately understood by existing technology
(this mode of behaviour has probably never before been critical with
the type of structure under consideration; a basic structural parameter
may have been changed so much from previous applications that the
new behaviour becomes critical, or alternatively, the structure may be
entirely of a new type or involve new materials or techniques; it is
possible, however, that someinformation concerning the problem
may be available from otherdisciplines or fromspecialist researchers,
and this will be information which has notgenerally been absorbed by
the profession);
54
 A N A L Y S I S O F S T R U C T U R A LF A I L U R E S : I
(e) failures which occur because the designer fails to allow for some basic
mode of behaviour well understood by existing technology;
(f)failures which occurthrough an error duringconstruction; these
would be the result of poor site control, poor inspection procedures,
poor site management, poor communications leading to errors of
judgement, the wrong people taking decisions without adequate con-
sultation etc., and may also occur through a lack of appreciation of
critical factors and particularly through poor communications
between designers and constructors;
(g) failures which occur in a deteriorating climate surrounding the whole
project; this climate is defined by a series of circumstances and
pressures on the personnel involved; pressures may be of a financial,
political or industrial nature, and may lead directly to a shortage of
time or money with the consequent increased likelihood of errors
during both design and construction processes; they may also result
in rapidly deteriorating relationships between those involved in the
project ;
(h) failures which occur because of a misuse of a structure or because the
owners of the structure have not realized the critical nature of certain
factors during the use of a structure; associated failures are those
where alterations to the structure are improperly done.
15. Although it- would be interesting to classify all structural failures into
these basic types, the information gained would not be directly useful. A more
detailed examination of each type is required and statements which will be useful
not only in studying past failures but in predicting the likelihood of future
failures need to be formulated. This checklist of failure parameter statements,
now presented, is based on thebasic failure types proposed and may be used in a
prediction process.36 A total of 23 past structural failures will be assessed, for
calibration purposes, using these parameters. These assessments will also
enable certain conclusions to be drawn regarding the major reasons for past
failure. It will be appreciated that these assessments can only be made sub-
jectively and thus, for this exercise, the parameters were formulated as statements
and the assessor (the Author in all cases but one) was asked to assess his degree
of confidence in the truth of and the importance of the statements.

A check list for assessment of likelihood of structural accident

16. A check list of statements has been devised, and the procedure is as fol-
lows. Assess (a) the degree of confidence in the truth of and (b) the importance
of the following statements. Choose any twoof the following descriptions:
l-very high confidence: A-very low importance
2-high confidence : B-low importance
3-medium confiaence: C-medium importance
4-low confidence: D-high importance
5-very low confidence: E-very high importance
] ( a ) The loads assumed in the design calculations are a good and/or a safe
representation of the loads the structure will actually experience.
(b) Any variabilities in the values assumed forthe parameters used to
describe the strength of the structure have been well catered for.
55
BLOC KLEY
2(u) Assuming that the design calculations have covered all possible failure
modes for the structure, the calculation procedural model is a good '
representation of the way the structure will behave if constructed to
plan.
(b) The quantity and quality of research and development available to the
designer is sufficient.
3(u) The information availableregarding the likelihood of such random
hazardsasearthquakes, fire, floods,explosions, vehicle impact is
sufficient.
(b) The structure is not sensitive to the phenomena of 3(u).
4(u) The materials to be used in the structure are yell tried and tested by
use in previous structures.
(b) There are no possible effects which could occur in the material which
have not been adequately catered for in the design calculations.
(c) The formof structure hasbeen well tried and tested by its use in previous
structures.
( d ) There is no step changein the values of the basic parameters describing
the structural form from thosevalues adopted in previous structures.
(e) There is no possible danger of a mode of behaviour of the structure,
inadequately understood through existing technology and which has
never before been critical with this structural form, now becoming
critical.
(f)There is noinformationconcerningthe materials orthestructure
which is available in other disciplines and which could havebeen used
in this design calculation.
5(u) There are no errors in the calculation procedural model or possible
modes of behaviour which are well known throughexisting technology
but which have been missed by the designers.
(b) Assuming the design is based uponagoodcalculationprocedural
model the likelihood of calculation errors is negligible.
( c ) The designers are adequately experienced in this type of work.
( d ) The personnel available for site supervision are adequately experienced.
(e) The design specifications are good.
6(a) The construction methods to be used are well tried and tested (including
off-site fabrication).
(b) The structure is not sensitive to erection procedures.
(c) The likelihood of construction errorsis negligible.
( d ) The contractor is adequately experienced in this type of work.
(e) The contractor has personnel available for site work and supervision
capable of appreciating the detailed technical problemsassociated
with the design.
7(u) The contractual arrangements areperfectly normal.
(6) The general climate surrounding the project design and construction is
perfect under each of the following headings: financial, industrial,
political, professional.
8 The structure is not sensitive to the way it is used.
56
 A N A L Y S I S O F S T R U C T U R A LF A I L U R E S : I
17. In parameter l(a) it is presumed that the designer always tries to choose
a representation of the actual loadson his structure which is conservative and safe.
A critical form of loading is more likely to be missed if the representation of the
actual loads is based on a poor model. For example, the use of equivalent
static loads for dynamic loading situations could easily lead to trouble where
unusual circumstances produced resonance or large dynamic magnification. In
parameter l(b) the aim is to cover adequately statistical variations in strength
values which comply with the specifications and which are catered for generally
by the use of appropriate safety factors. Parameters 2(a) and (b)are included to
question the degree of confidence of the assessor in the model used for the design
decisions; that is effectively the degree of confidence in the use of elastic and/or
plastic theory, research and development information and codes of practice.
This can be contrasted with %e) which deals with the possibility of a completely
new form of structural behaviour taking over, such as occurred with the famous
Tacoma Narrows Bridge.7 Parameter 5(a) deals with the calculation procedural
model but concentrates on the possibility of human error. Parameters 5(b)-(e)
are also aimed at the possibility of human error. Parameter 5(b) asks effectively
if the designers are working in good conditions, with good communications
between the various members of the design team and good management pro-
cedures. Are the calculations numerically complicated and how reliable are the
computer programs used? Parameters 5(6), ( d ) and (e) are intended to cover
design office to resident engineer (if any) communication and communications
between resident engineers and contractor, both on the site and through the
specification.
18. Relevant subsidiary questions to 6(c) would be, for example, does the
contractor have a good safety record,good management procedures, good
labour relations with no evidence of slack site control? Has a low tender bid
produced excessive pressures (financial or shortage of time)? Parameter 7(a)is
intended to cover the contractual arrangements and whether the responsibilities
of the various parties to theagreement have been adequately defined. P ~ g s l e y ~ ~
has discussed those factors covered by parameter 7(b)under the heading 'engi-
neering climate'. This parameter is intended to check whether there are
excessive pressures on theproject from international, nationalor local, financial,
political or industrial conditions. Parameter 8 is included because for some
special structures the user may need some educating with regard to critical
factors in the structure.
Analysis of 23 structural failures
19. The parametersoutlinedabove were formulated for the purpose of
assessing the likelihood of a structural accident in the future. In this section the
values of assessments made by the Author on 23 failure^^-^' are presented
(Table 2). Assessments for parameter 4(f) were not made. The analysis was
purposely limited to a number of accidents where detailed inquiries into the
nature of the failure were reported. Thus reports in engineering news journals
were not considered. It is appreciated that this selection of accidents is not a
random sample. The assessments are entirely the subjective opinion of the
Author, based upon a reading of the references. It is therefore quite possible
that engineers more qualified to consider any particular failure would make
slightly different assessments. However, it is felt that general conclusions may
be drawn from the results presented.
57
BLOCKLEY

l
-W l
__
NU

-W -n

t
 A N A L Y S I SO FS T R U C T U R A LF A I L U R E S : I

I I
e-
IsI l
v.c1
Is I -n I
I
-I

jma
I
BLOCKLEY

l
I
I
I
I

-n
 A N A L Y S I S OF S T R U C T U R A LF A I L U R E S : I
20. Certain difficulties arisein assessing statementsformulated forfuture
projects when considering events of the past. This is particularly highlighted
in parameters 5(d) and qe) when trying to assess the competence of the personnel
running the site and inspecting the works. In assessing a future project one can
only make a judgement on the basis of past performance; when assessing a
failure one can assess actual performance whcre this is discussed in the report.
In some accidents, e.g. Kings Bridge, the designers were not represented during
fabrication or construction under parameter5(d) and inspection was done by the
client. The assessment made hereis based upon theinspection done on behalf of
or by the client. Parameters 5(4, 6(c)and 6(e)are often difficult to separate (as
for theaccidents at Listowel and Aldershot), but it is assumed from the way the
structure failed that all three aspects were deficient. In several of the accidents
the design work was done by the contractoreither as partof a package deal (e.g.
first Quebec Bridge) or because the failure was part of some supporting false-
work (e.g. Lodden Bridge). The assessments for these failures were based on
the performance of the designers who designed that part of the structure which
failed.
21. The failure of the PointPleasant Bridgeg indicates the difficultyof
preventing some failures. Built to specification, this American suspension
bridge was completed in 1928 and failed in 1967. The cause of the failure was a
fracture in an eyebar link resulting from a crack which had grown through stress
corrosion and corrosion fatigue. The crack was impossible to detect without
disassembly of the joint.
22. The failure of the Bedford Town Hall1* is an unusual exampleof a design
failure. The design calculations were hopelessly incorrect and were performed
by an engineer working in isolation. Any one of a number of mistakes in-
cluded in parameter 5(a) could have caused failure. In most other accidents
the mistakes were much more difficult to isolate. Thus there was a high con-
centration of errors in one parameter.
23. It is impossible, due to limitation of space, to discuss fully the reasons
for each of the assessments in Table 2. Only a brief mention can be made, as
an illustration, of the assessments for the Quebec Bridge.

Quebec Bridge
24. The Quebec Bridge was a steel cantilever structure over the St Lawrence
River which failed during construction in 1907. A second bridge suffered the
failure of a suspended span in 1916, but was finally completed in 1918.
25. An error in the calculation of the dead load on the first Quebec Bridge
resulted in dead load stresses of the order of 7% in excess of those calculated.
This error has been included in the assessment of parameter l(a) as it concerns
loading rather than structural response. The state of knowledge around the
turn of the century with regard to the behaviour of latticed columns was very
uncertain and the research information available to the designers was very
limited. No proof testing was done on the made-up column sections for the
first bridge and higher permissible stresses than were usual in bridge construction
were adopted. The importance of the end details and splices and the effect of
lack of straightness in the columns were not fully appreciated. Consequently
the assessments made for parameters 2(a), 2(b), 4(e) and 5(a) were all serious.
The chord which failed in the Quebec Bridge, when compared to five other
61
BLOCKLEY
American cantilever bridges of the time, had considerably less horizontal
stiffness ( l / r ) , less lattice area, less rivet area and less splice plate area in propor-
tion to the size of the members and theassessment of parameter 4(d) reflects this.
26. Parameters 5(b)-(d) are concerned with the other conditions surrounding
the design of the bridge. Besides the dead load calculation error other minor
calculation errors occurred. Theodore Cooper, the highly respected but aging
consultant, was not responsible for theactual design (which was in fact done by
the contractors), but because he checked all the calculations a feeling of false
security amongst the design staff was engendered. Theactual designers’
experience was based on small bridge practice. The site supervision was in-
adequate and there was a confusion of responsibility between the chief engineer
of the Quebec Bridge Company and Cooper, the consultant. Cooper rarely
visited the site, but advised frequently. The specifications were written by an
inexperienced engineer and there was nothing to indicate that the bridge was an
exceptional structure. The specifications were infact based on small bridge
design with tolerances which were too tight.
27. Parameters 6(u)-(e) deal with the construction of the bridge. As this
was a cantilever structure it was obviously at a very sensitive stageduring
erection. There was evidence of problems due to waviness in the members of
some columns, though this mayhave been dueto overtight tolerances. At
least one member was damagedduringtransportation. The contractor’s
experience was mainly in small bridges, andthoughthe organization was

Table 3. Results of simple analysis: failures

Simple analysis
Accident
2x Order W Order
Quebec 2 2.52 1 1.45 1
Bedford 2.68 2 1.68 2
Second Narrows 3.12 3 1.90 3
Heron Road 3.68 4 2.04 4
Point Pleasant 3-96 5 2.78 5
Mt Gambier 4.60 6 3-16 6
Camden 4.76 7 3.27 7
Tacoma 5.08 8 3.91 9
Aroyo Sec0 5.12 9 3-81 8
Lodden 5.60 l0 412 10
Ferrybridge 5.92 11 4.34 13
Transocean III 5.96 12 4.13 11
Ardeer 5.96 12 4.36 14
Ilford 6.16 14 4-28 12
Listowel 7.00 15 4.99 15
Stepney 7.72 16 5.94 17
Aldershot 7.96 17 6.32 19
Sea Gem 8.12 18 6-30 18
Ronan Point 8.68 19 5-89 16
Kings Bridge 8-68 19 6.77 20
Westgate 11.32 21 9.23 21
Quebec 1 12.04 22 9-29 22
Tay 12.96 23 10-46 23
Example 2.76 0.71
 ANALYSIS O F STRUCTURAL
FAILURES: I
generally efficient there was a lack of appreciation of the magnitude of the job.
The contractual arrangements were unusual for a large structure and thegeneral
financial climate surrounding the Quebec Bridge Company was uncertain and
‘seriously interfered with carrying out of the ~ndertaking’.~There was also
evidence of some political pressure when the Dominion government came to be
involved, in that it desired that the bridge be opened in time for the Quebec
Tercentenary in 1908.
28. In contrast to the fist Quebec Bridge, the second bridgea was very care-
fully researched and designed. It was extremely unfortunate that a failure in
one of the castings during the erection of a suspended span caused the loss of
that span.
Interpretation of theassessments
29. The assessments in Table 2 are presented in a form that makes it difficult,
to draw conclusions about the relative inevitability of each failure and to
identify the dominant reasons for the accidents. It must be appreciated that
these are subjective assessments and should be interpreted quite differently from
the results of objective experiments. There is obviously much more uncertainty
in them and they depend crucially on the person making the assessment. They
are an expression of engineering judgement. However, in order to reach con-
clusions from past failures and to produce predictive methods to help avoid
future problems, two numerical interpretations of the assessments will be used.

Table 4. Results of simple analysis: parameters

Order
Parameter T Size
No. I Brief description

1 Design error 15.48

2 Construction error 11.88
3 Contractor’sstaff site control 11.76
4 Designer’s site staff} 11.68
5 R & D information 10.88 ’
6 Calculation procedural model 10.68
7 Unknown material effects 9-32
8 New structural behaviour 6.52
9 Overload 652
10 Specifications 5.76
11 Step change in structuralform 5.36
12 Contractual arrangements 5.32
13 Use of structure 5.24
14 Materials well tested
15
16
General
climate }
Sensitivity to erection
4.52
4.46
17 Calculation errors 3.80
18 Form of structure well tried 3-24
19 Construction methodswell tried 2.80
20 Contractor’s experience 2-56
21 Designer’s experience 2-52
22 Strength variability 2.48
23 Sensitivity to random hazards 1.92
24 Random hazards 1.28
B L 0 C KLEY
In this section a simple scale 0-+1 will be used and in Part I1 use ismade of
fuzzy
30. Each assessment in Table 1 was given numerical values as follows:
l-very high confidence = 0.2: A-very low importance = 0.2
2-high confidence = 0.4: l3-10~ importance = 0.4
3-medium confidence = 0.6: C-medium importance = 0.6
4-low confidence = 0.8:D-high importance = 0-8
5-very low confidence = 1 : E-very high importance = 1
For each accident and for each parameter the two assessments were multiplied
together to give a value X . The summations of all parameter values of X for
each accident are given in Table 3 and the summations over all accidents for
each parameter are given in Table 4. In Table 3 summations of xa are also
given for comparison, as this summation tends to repress thelower values of X
and give predominance to the higher values. The accidents are listed in the
order of value of these totals, as these are a crude measure of the degree of
likelihood (in these cases inevitability!) of a structure having an accident. The
fact that the two Quebec Bridge failures are numbers 1 and 23 in the list bears
this out, because the number of factors contributing to thefirst Quebec failure
is far in excess of the unfortunate second Quebec failure.
31. Table 3 also includes a total for an exampleof a structure which has not
failed. This estimate was made by a consulting engineer in practice and serves
as a comparison with the totals gained by the failures. It is, in fact, for a four
storey office block of steel-concrete composite construction, and the score is a
composite of many small assessments. The total z x a demonstrates this, as the
influence of assessments less than 1 is considerablyreduced.However these
simple totals are poor indicators of the degree of proneness of the structure to
failure (Part 11).
32. Table 4 shows perhaps the most interesting table of results. It must be
remembered when interpreting this table that the sample of failures taken is not
random and includes only failures important enough tomerit individual reports
of inquiry. Thus, it is not surprising that the scores obtained for parameters
3(a) and (b), therandomhazards,are very low. Arandomsample of all
failures would producea very high score for these parametersdue to the known
high incidence of fires, floods, explosions, etc. Also similarly it might be ex-
pected that parameter 2(u) would have an increased score if serviceability of
partial failuresof buildings due to foundation settlement and creep and shrinkage
cracking of concrete were included. However, Table 4 does have an importance
with respect to failures of a major type and shows the predominance of human
error in causing these structural failures. Design errors varying from simple
second year undergraduate mistakes, such as the Second Narrows Bridge, to a
fundamentalmisunderstanding of overall structuralbehaviour,suchas at
Ronan Point and Aldershot, are the predominant total. Inadequate site control
and checking by the contractor and the clients’ representatives are also major
parameters. The lack of enough research and development information is an
inevitably high total in a list of accident parameters and reinforces the case for
an increasedexpenditure in this field. The uncertainty surrounding the cal-
culations one can perform in making structural design decisions, whether due to
loads, materialsor theapplication of theory, is also obviously important (param-
eters 4(b), qe), l@)).
64
 ANALYSIS O F STRUCTURAL F A I L U R E S : I I
33. These results, of course, may well be affected by the inclusion in the
sample of 23 failures of structures of similar type. Failures of falsework, box
girders, structures of high alumina concrete or brittle steel, represent over half
of the sample. It is interesting to note that two-thirds of the bridges but only
one-quarter of the other structures in the sample failed during construction and
erection.

Conclusions
34. A checklist of parameterstatements which enable an independent
observer of a structural project to make an assessment of the degree of proneness
of the structure to an accident can be formulated.
35. Such a proposed checklist, when used to assess parameters for 23 major
structural accidents, shows that failures are due to a variety of causes and
combinations of circumstances. However, humanerror in using existing
technology is the predominant overall factor in the accidents considered.
36. Insufficient research and development information and the resulting
uncertainty surrounding design and construction decisions is also a major
factor in the failures considered.

Part II
In Part I a discussion of the formulation of suitable parameters with which to
assess the likelihood of structural failure is presented. Assessments of proposed
parameters for 23 structural accidents have been made (Table 2) and analysed
using a simple numerical interpretation. In this section an analysis of the
assessments is made using fuzzy sets.36

Notation
A fuzzy linguistic variable 'confidence'
B weighted version of A
C. D. F fuzzv.~sets
E[ 1 ' expected value of [ ]
H a transfer matrix
K kernel of a fuzzifier
M norm
n defined bypf=l0-"; an element on N
N fuzzy universe (1+10)
PC probability of failure
P fuzzy sets of all accident parameters
4 Ebl
Q a non-fuzzy set containingq
R fuzzy set
T worst possible P
U fuzzy universe (+l)
a weighting factor
PA membership level in A
r,(u)lut element of a fuzzy set; U,is the element, p&) is the membership level
CUD union of sets C and D
CnD intersection of sets C and D
- is defined to be
65
 BLOCKLEY
38. InPart I an assessment of the degree of confidence in a parameter
statement, such as for example ‘ A l o w confidence’, was interpreted as a single
digit of 0.8. In this section each of the assessments 1-5 is considered as a fuzzy
linguistic variable. Such a variable is a fuzzy set of discrete (for, this purpose)
elements chosen on the scale 0 to 1. Each discrete element has a certain mem-
bership level in the fuzzy set of p which is also measured on a scale 0 to 1. Thus
the set A of elements U in the total set U is m discrete values

where = should be interpreted as ‘is defined to be’, I is a delimiter, and the

summation is taken as theunion rather than thearithmetic sum. For example if
U = 0+0-1+0.2+0.3+0-4... l
then, for this Paper, the linguistic variables used are:
A1 = very high confidence= 1~0+0~7~0~1+0~1~0~2
Aa = high confidence = 0510.1 +0.910.2+0*910.3+0-510.4
+ +
A3 = medium confidence = 0-110.3 0.7(0.4+ 1\050-710.6+0.110.7
A4 = low confidence = 0~5~0~6+0~9~0~7+0~9~0~8+0~5~0~9
+
A5 = very low confidence = 0.1 10.8 0*710.9+111
These sets are shown in Fig. 2. By drawing smooth curves through the points,
the sets could be made continuous.
39. The method used herein is a simplified version of that originally pro-
posed.36 The assessment of the degree of importance of a parameter was
originally interpreted also as a fuzzy set. A simpler procedure is used here
where the degree of importance is interpreted as a simple scalar multiplier or
weighting factor a, so that the fuzzy set A becomes a set

when both assessments are. taken into account. The values of a used in this
section are those defined in Part I. Thus:
a1 = A
= very low importance = 0.2
aa = B = low importance = 0.4
a3 = C = mediumimportance = 0.6
a4 = D = high importance = 0.8
a5 = E = very high importance = 1

~ Fuzzy set analysis

40. Each of the 24 proposed parameters (5 16) is subjectively assessed using a
linguistic variable to describe the size of the parameter and a scalar weighting to
modify the size of the variable according to its importance in relation to the
failure. Thus for each failure, 24 fuzzy sets B, j = 1, . . .24 are formulated.
41. The total or overall effect of these parameters needs to be calculated so
that a judgement concerning the degree of likelihood of a structural accident
may be assessed. In fact there are two ways of calculating this effect. First,
one would like to knowthe worst possible combination of circumstances
(parameters). For example, it would be useful to know if it is likely that ex-
treme financial pressure (parameter 7(6)) and extreme uncertainty in the con-
66
 A N A L Y S I SO FS T R U C T U R A LF A I L U R E S : II

Element value Y

Fig. 2. Fuzzy sets for the variable 'confidence'

struction process (parameter 6(a)) will both be present, and whether they will
combine to form the worst possible set of circumstances. This is obtained by
taking the largest intersection of sets Bp Secondly, one would like to know the
overall picture when the effects of all the parameters are included. This is
obtained by taking the union of the sets B,.

Intersection
42. There are 24 parameter sets B, describing a possible failure. Let 241r
be
a combination of r of the numbers 1,2, 3 . . .24. Let ik be the kth number in
241,.
43. The worst combination of the sets B, is thus the fuzzy set:
P I = max,[
r A -B1,]
k= 1
. . . . . .
and the maximum set is that set which has the most concentrated membership
levels towards the value of the element of the set of 1.
44. The intersection used here is the soft version37 and corresponds to a soft
interpretation of the word 'and'. The operation used to obtain the intersection
of two sets is the algebraic product. Thus if pC(u)and pD(u) are themembership
levels of element U in sets C and D then
pCAD(u) = p d u ) .pD(U)
67
BLOCKLEY
Union
45. The union of the 24 sets B, is the set which includes them all and is

46. The union used here is the soft version37 and corresponds to a soft
interpretation of the word 'or'.The operation used to obtain this is the
algebraic sum. For the two sets C and D
= pc(u) + p D ( u ) - p c ( ~.)p d u )
~ C V ~ U ) . . . . (5)
where =, + and - are interpreted in the normal way as arithmetic symbols.
47. It is assumed that a structural analysis of the structure under the normal
assumptions (i.e. not consideringthosefactors covered by the parameters
discussed here) can be performed and an expected value of the probability of
failure (pr) in some critical limit state estimated. If pr=lO-", then let the
expected value of II be E[n]=g, where N is a set of numbers n i.e. N = 1 2 3 + + +
. . . + 1 0 and q is a non-fuzzy set Q in N so that p(n)Jq=1 or Q = 119.
48. Q was a set calculated ignoring all the parameter assessments. The
procedure is now that of changing this estimate of the probability of failure in
the light of the assessments. In fact Q will be fuzzified by PIand P2.
49. In order tocarry out this operation a fuzzifier with kernel K(n) has tobe

Table 5. Fuzzy set analysis: Membership values of largest intersection PI

Element value
Accident
0.9 1
Tay 0.08 1
Quebec 1 0.24 1
Quebec 2 0.7 1
Tacoma 0.34 1
Kings Bridge 0.17 1
Point Pleasant 0.49 1
Westgate 0-08 1
Second Narrows 0.7 1
Heron Road 0.7 1
Lodden 0.34 1
Aroyo Sec0 0.34 1
Listowel 0.24 1
Aldershot 0.17 1
Bedford 0.7 1
Ronan Point 0.34 1
Camden 0.49 1
Stepney 034 1
Ilford 0.49 1
Ferrybridge 0.34 1
Mt Gambier 0-49 1
Sea Gem 0.24 1
Ardeer 0-34 1
Transocean ZZI 0.49 1

Example = 0 ~ 0 8 ~ 0 ~ 3 + 0 ~ 5 6 ~ 0 ~ 4 + 0 ~ 8 ~ 0 ~ 5 + 0 - 5 6 ~ 0 6 + 0 ~ 0 8 ~ 0 - 7
68
 A N A L Y S I S OFSTRUCTURALFAILURES: II
formed, and this is used to fuzzify Q to give
Q' W p&). . . . . . . (6 )
and because Q is a non-fuzzy set, Q' =K(q).
50. The fuzzifier K(n) is obtained by the fuzzy composition3' of P (a matrix
1, U) and H (a matrix U, n).
K(n) = PoH = 2 max [min pP(u), p& n)lln
N U
. (7) -
where H is a square transfer matrix converting the set P into a fuzzifier K@).
It has dimensions (2q+ 1) and is chosen so that U = 0, n = 2q; U = 0.5, n = q ;
U = 1, n = 0 are along theleading diagonal.
51. The values of the elements of the matrixcan only be obtained by a
calibration with previous failures and existingstructures. For thepresent
purpose H is assumed as a unit matrix. In this case Q' ismerely the set P
continuously transferred from U ( b 1 ) to N ( 1 b l ) .
52. The assumption that H is a unit matrix provides therefore a fuzzy set Q'
which is independent of q. In other words, the degree of likelihood of a struc-
tural accident is the same whatever the expected value of the probability of
failurecalculatedignoring the parameters. Any objectionstothis unlikely
result can be overcome by including in H values other than on the leading
diagonal. However a consideration of past failures is insensitive to the unit

Table 6. Fuzzy set analysis: membership values of soft union PS

Element value
Accident - ---- _.

0 0.1 0-2 0.3 0-4 0.5 0.6 0.7 0-8 0.9

- - - - - -- -- -
Tay 0.59 0.83 0.92 0.93 0.98 1.0 0.99 1.0 1.0 1e o

Quebec 1 0.79 0.83 0.84 085 093 097 1.0 1.0 1.0 1.o
Quebec 2 0.99 0.96 0.36 - - - 0 4 0.72 0.75 0.82
Tacoma 0.98 0.94 0.43 0.25 0.60 0.80 0.78 091 0.93 0.99
Kings Bridge 0.93 0.92 0.86 0.87 0.99 1.0 0.98 0.93 0.94 1.o
Point Pleasant 099 0.97 0.84 0.77 0.46 - 0.50 0.90 0.92 0.96
Westgate 0.87 0.82 0.62 0.56 0.50 0.40 0.85 097 0.99 1.o
Second Narrows 0.99 0.96 0.35 0.08 056 0.80 0.78 091 0.91 0.85
Heron Road 0.99 0.94 0.32 0.28 0.97 1.0 0.97 0.28 0.10 0.70
Lodden 097 0.96 0-93 0.91 0-84 0.88 0.84 0.91
0.93 0.99
Aroyo Sec0 0.98 0.94 0.55 0.41 0.59 064 0.74 0.91
0.93 0.99
Listowel 0.93 0.88 0.59 047 028 - 0-78 0.96 097 l .o
Aldershot 0.950.90 0.65 0.58 0.65 0.68 080 094 0.96 1*o
Bedford 0-990.96 0-36 - - - - 017 087
Ronan Point 0.890.82 0.47 0.58 0.98 1.0 0.99 0.98
0.98 1-0
Camden 0.98 0.95 0.68 0-57 0.59 0.60 0.65 0.74 0.79 0.98
Stepney 096 0.90 0.26 0.10 0.70 1.0 0.98 1.0 1.0 1
- - - 0.76 0-96 0.97
n o

Ilford 0.97 0.91 0.28 1.o

Ferrybridge 098 0.92 0-29 0.15 0.83 1.0 0.94 0.93 0.94 0.99
Mt Gambier 0.99 0.94 0.32 0.1 9 0.91 1.0 0-95 0.77 0.77 0.95
Sea Gem 0.96 0.88 0.25 0.06 0-42 0.60 0-91 0.99 1.0 1-0
Ardeer 0.97 0.93 0.54 0-43 0.73 0.84 0.90 097 0.98 099
Transocean I11 0.97 0.91 0.28 0.10 0.58 076 0.96 1.0 1.0 0.99
Example 1.0 1.0 0.94 0.92 0.94 0.96 0.81 0.15 - -
-- -
B L 0 C KLEY
matrix assumption because of the high membership levels in the sets P at and
near the element value of 1. In order to arrive at suitable values for H it is
obvious that considerable calibration, not only with failures but with existing
structures, needs to be performed. This is also coupled with the difficulty of
estimating q.
53. The fuzzy sets Q are expressions of the likelihood of a structural accident.
Using these as part of the evidence, a decision may have to be made by an
assessor whether to recommend the delaying of a project whilst certain matters
are reinvestigated or whether to allow events to proceed normally. For ex-
ample, more time may need to be allocated (at presumably greater expense)
because the nature of somestage of construction is more critical than was
originally allowed for in the estimates. These decisions, which on the face of it
increase the cost of a project, may have to be taken to avoid the possible astro-
nomical costs of failure. The assessor needs some quantitative information to
reinforce his arguments and these are provided in the form of the fuzzy sets Q‘.
However in order to interpret the sets Q’ into a simple fuzzy variable such as
‘very high confidence in the safety of this structure’, resortmay have to be made
to themathematics of pattern recognition. Nevertheless, a simple interpretation
would be to choose the linguistic variable F describing the sets Q‘ as follows.
54. Take F as that set which has the maximum value of the norm of its
intersection with Q‘. The norm of the set A with m elements of U in U is a
scalar quantity
m
M= ,I1 p?(u)
I=
. . . . . . * (8)

Thus theset F’ describing the stateof the structureis the set F for which M is a
maximum where

where R=(F,nQ’) and is the ‘hard’ intersection i.e.

Fuzzy analysis of accidents

55. Tables 5 and 6 show the sets PI,the soft intersection, and Pz,the soft
union, for the 24 structural failures and the example described in 0 31. As
would be expected the membership levels of the elements of the set Pz at and
close to 1 for each accident are very high. In comparison, the fuzzy set P2 for
the example structure has high membership levels only for the elements of the
set close to zero, and for those close to 1 the membership levels are zero. Thus
these patterns reflect the severity of the conditions (described by the parameters)
surrounding the failures. The fact that there were many factors contributing
to thefirst Quebec Bridge failure is reflected by high membership levels from 0.5
to 1. The high membership level at element value 1 for the second Quebec
Bridge failure quickly tails to zero at value 0.5, and then at very low element
values themembership values are high. This reflects thefact thatthe con-
ditions surrounding the bridge were fairly well controlled. In fact, nearly all
parameters were assessed as ‘very high confidence’ values resulting in the high
membership levels at the low end of the set Pa with only two parameters con-
70
 A N A L Y S I S OF S T R U C T U R A LF A I L U R E S : II

Table 7. Summary of accident parameter assessments

T
~~

Fuzzy set analysis

Accident Simple analysis
(Part I) Intersection
- Norm
Order Order V a l u e Order Order
- --
Quebec 2 2.52 1 1.45 1 2-23 l 3 1
Bedford 2-68 2 1.68 2 1.78 2 1
Second Narrows 3-12 3 1.90 3 2.55 5 1
Heron Road 3.68 4 2.04 4 1-50 1 1
Point Pleasant 3.96 5 2.78 5 2-77 7 5
Mt Gambier 4.60 6 3.16 6 2.49 4 5
Camden 4.76 7 3.27 7 2-58 6 5
Tacoma 5.08 8 3-91 9 2.83 8 10
Aroyo sec0 5.12 9 3.81 8 2.83 8 10
Lodden 5-60 10 4.12 10 2.83 8 10
Ferrybridge 5-92 11 4-34 13 2.83 8 10
Transocean 111 5-96 12 4-13 11 2.83 8 5
Ardeer 5.96 12 4-36 14 2.83 8 10
Ilford 6.16 14 4.28 12 2-85 14 5
Listowel 7.00 15 4.99 15 2.85 14 17
Stepney 7.72 16 5-94 17 2.85 14 10
Aldershot 7.96 17 6.32 19 2.85 14 20
Sea Gem 8.12 18 6.30 18 2-85 14 17
Ronan Point 8.68 19 5.89 16 2.85 14 10
Kings Bridge 8.68 19 6.77 20 2.85 14 20
Westgate 11.32 21 9.23 21 2.85 14 22
Quebec 1 12.04 22 9.29 22 2.85 14 17
Tay 12.96 23 10.46 23 2.85 14 22
Example 2-76 0.71 0
-
71
BLOCKLEY
tributing to the membershipvalues near to 1. The sets P2 for three failures and
for the example are shown graphically in Fig. 3 and present a pictorial view of the
situations. If, as discussed in the $0 50-52, the matrix H is a unit matrix and
q = 5, the set Qf2 (on N) has the same membershiplevels as P2 (on U). For any
value of q the continuous graphical representation of Pa is the same as that of
Q‘2 with the X axis altered from U((k1) to N(l(k1).
56. In order to obtaina ranking for the23 accidents in order of severity, the
norm M for the intersectionof each set Pzwith the set T2 representing the worst
possible assessment of the 24 parameters was calculated. T2 represents the
union of 24 assessments of ‘very low confidence’ in the truth of a parameter
statement and a ‘very high importance’ (A5, a5 or 5, E: see 8 16). Ta is calcu-
lated to be the set
T2 = 0.9210*8+110-9+1)1
57. The values of the norm so calculated are shown in Table 7 together with
the results of the simple analysis from Part I. It will be seen that the order of
ranking in the twois similar.
58. In contrast, Table 5 shows the sets P1,the maximum intersections for
eachaccident andforthe example. Themaximumintersections are easily
calculated for the failures, as they are directly the number of assessments made
as Ab,a5or 5, E, ($ 16). This is because every failure has at leastone assessment
at this maximum level. Also the worst possible set of assessments, 24 of ‘very
low confidence’ and ‘very high importance’, would result in a worst possible
intersection set of T1= 111. The norms of all accidents sets in Table 5 inter-
sected with T1 are simply therefore 1.
59. A ranking of the failures according to the number of maximum assess-
ments is also shown in Table 7. This is probably the bestway to rank the
failures as it reflects the worst possible set of conditions on the basis of the
assessments.
60. The contrast between the three methods is perhaps best shownup by
assessments of the failure of Ronan Point. The simple summations and the
fuzzy union reflect the large number of assessments made in the medium and
low confidence range, with only three assessments at very IOW confidence and
very high importance (c.f. Tay Bridge with seven). The worst intersections
reflect only these three worst assessments. The maximum intersections for three
failures and the example structure are shown in Fig. 3. Again this figure gives a
pictorial representation of the seriousness of the situation.

Conclusions
61. It is possible to discuss the structural failure parametersin terms of fuzzy
variables and to produce fuzzy statements as to the likelihood of an accident to
any given structure.
62. Fuzzy sets representing 23 structural failures enable a ranking of these
failures in order of inevitability.

References
1. BLOCKLEY D. I. Structural designdecisionsandsafety. Publs. Int. Ass. Brirlpe
Struct. Engng, 1974, 34-11, Sept., 1-18.
72
 A N A L Y S I S OF STRUCTURALFAILURES: II
2. BAKER M. J. The reliability of reinforced concrete Poor slabs in ofice buildings: a
probabilistic study. Report 57, CIRIA, London, March, 1976.
3. Report of the Court of Enquiry upon the circumstances attendingthe fall of a portion
of the Tay Bridge on 28th Dec. 1879. HMSO, London, 1880.
4. THOMAS J. The Tay Bri&e disaster, new light on 1879 tragedy. David & Charles,
1972.
5. Royal Commission on the collapse of the QuebecBrirlpe. Vol. I, 11, 111, Ottowa,
1908.
6. Report of the GovernmentBoard of Engineers. Vols I, 11. Dept. Railways and
Canals, Canada, 1918.
7. AMMAN A.H. et al. The failure of the Tacoma Narrows Bri&e. Federal Works
Agency, Washington DC, March 1941.
8. Report of the Royal Commission into the failure of theKingsBridge. Victoria,
Australia, 1963.
9. Collapse of US 35 Highway Bridge, Point Pleasant, West Virginia, Dec. 15, 1967.
Report No. NTSB-HAR-71-1, National Transportation Safety Board, Washing-
ton DC.
10. Report of the Royal Commission into the failure of West Gate Bridge. Victoria,
Australia, 1971.
11. FREEMAN R. and OTTERJ. R. H. The collapse of the Second Narrows Bridge,
Vancouver. Proc. Znstn Civ. Engrs, 1959, 12,April, N 3 W 1 .
12. HEALTH AND SAFETYEXECUTIVE. Final Report of Advisory Committee on False-
work. HMSO,London,June, 1975.
13. ACRESH. G . Report on the HeronRoadBridge failure. Report to Supervising
Coroner of Ontario, Toronto, Nov., 1966.
14. Collapse of falsework for the viaduct over R. Lodden on 24th Oct., 1972. HMSO,
London, 1973.
15. STATEOF CALIFORNIABUSINES AND TRANSPORTATION AGENCY. Finalreport:
Investigation into collapse of falsework, Aroyo-Sec0 ‘Bridge Road. 07-LA-210.
Department of Public Works, Division of Highways, Jan.,1973.
16. NRC CANADA.Thecollapse of theListowelArena. Div.ofBldgRes.Tech.
Paper 97, Ottowa, May 1960.
17. BUILDING RE~EARcHSTATION.The collapseof aprecast concrete building. HMSO,
London, 1963.
18. CLARKE B. L. CountyofBedford,NewCounty Hall, Report on thestructural
engineering aspects of the original design and recommended remedial measures.
County of Bedford, Feb. 1966.
19. M I N I ~ OF Y HOUSING AND LOCAL GOVERNMEW.Coflapse of flats at Ronan
Point, CanningTown. HMSO,London, 1968.
20. DEPARTMENT OF EDUCATION AND SCIENCE. Report on the col[apseof the roof of the
assembly hall of the Camden School for Girls. HMSO, London, 1973.
21. BATES. C. C. Report on thefailure of roofbeams at Sir John Cuss’s Foundation and
RedCoat ChurchofEnglandSecondary School, Stepney. BuildingResearch
Establishment, Garston, June 1974.
22. MAYO A. P. An investigation of the collapse of a swimming pool roof constructed
with plywood box beams. Building Research Establishment, Garston, April 1975.
23. Report of the committee of inquiry into the collapse of cooling towers at Ferrybridge
Monday 1st December 1965. CEGB, London. 1965.
24. JOHNS P.M. and MOTTRAM K. G. Investigation into the failure of the Mount
Gambier television mast. J. Instn Engrs Aust., 1968, 40, 117-21.
25. MINISTRY OF POWER. Report of the inquiry into the causes of the accident to the
drilling rig Sea Gem. HMSO, London, Oct. 1967.
26. Report of the committee of inquiry into the collapse of the cooling tower at Ardeer
Nylon Works, Ayrshire, on Thursahy27thSeptember1973. ICI Petrochemicals
Division, London, 1973.
73
B L 0 C KLEY
OF ENERGY.Report on the loss of the drilling barge Transocean 111.
27. DEPARTMENT
HMSO, London, 1975.
28. AMERICAN
SOCIETY OF CMLENGINEERS.Structuralfailures:modes,causes,
responsibilities.NationalMeeting on StructuralEngineering,Cleveland,Ohio,
April 1972.
29. FELD J. Lessons from failures of concrete structures. ACI, Detroit, Mich., 1964.
30. HAMMOND R. Engineeringstructural failures. OdhamsPress,London, 1956.
31. MCKAIGT. H. Building failures. McGraw-Hill,NewYork, 1962.
32. MERCHANT W. Threestructuralfailures:casenotes and generalcomments.
Proc. Instn Civ. Engrs, 1967, 36, March, 499-500.
33. PUGSLEY A. G . Theengineeringclimatology of structuralaccidents. Inter-
nationalconference on structural safety and reliability. Washington, 1969,335-
340.
34. P ~ G S L E YA. G. The safety of structures. Edward Amold, London, 1966.
35. Scorn G. Building disasters and failures-+ practical report. Lancaster,Con-
struction Press, 1976.
36. BLOCKLEY D. I. Predicting the likelihood of structural accidents. Proc. Znstn Civ.
Engrs, Part 2, 1975, 59, Dec., 659-668.
37. BELLMAN R. E. and ZADEHL. A.Decision-makingin a fuzzyenvironment.
Mngmt Sci., 1970,17,Dec., B141-Bl64.

74

View publication stats