Sunteți pe pagina 1din 10

Proc. rnstn Civ. Engrs, Part 2, 1975, 59, Dec.

, 659-668

7825 Predicting the likelihood of


structural accidents

The Paper outlines a possible approach to the problem of predicting the likelihood of
a structure failing due to causes other than the stochastic variations in loads and
strengths. The concept of fuzzy sets is used in the formulation of the method.

Notation
A, B fuzzysets
F fuzzifier
K kernel
of
fuzzifier F
n defined by pr= 10-"
PI size
of parameter i
PT total effectof allparameters
pi probability
of failure
R, S fuzzyrelations
U, W elementsof U, W
U, W universes of discourse
W C weightingofparameter i
X, y elementsof X, Y
X, Y fuzzysets
pa membership level
of A
pJut element of a fuzzy set where ut is a value and pt is a membership level
AvB union of stts A and B
A n B intersection of sets A and B
a V b max (a, b)
a A b min (a, 6 )
= is
defined to be

Introduction
Much research work has been done on stochastic analyses ofengineering
structures. As has been pointed out by Pugsley,l over the past century the
success of the designers of engineering structures regarding the safety, if not
always the economy, of their structures has been considerable. This has been
largely due to anincreased understanding of the way structures behave under
extreme conditions and it is from this base that much of the present work
on stochastic analyses has developed. Structural failures due solely to over-
loading or to structural weakness under normal loadings are now rare.

Written discussion closes 16 February, 1976, for publication in Proceedings, Part 2.


* Lecturer in Civil Engineering, University of Bristol.
659
Downloaded by [ UNIVERSITY OF EXETER] on [12/09/16]. Copyright © ICE Publishing, all rights reserved.
BLOCKLEY
2. A large proportion of structural accidents, whether involving partial or
total collapse, occur either because an error is made in thedesign or construc-
tion of the structure or because several unfortunate circumstances just happen
to combine. Examples of errors vary from those of design concept (such as
an insufficient consideration of the lateral stability of a building) to construc-
tion errors (such as misplaced reinforcement in a concrete slab). Pugsleyl
discussed the likelihood of such errors causing accidents under the headingof
‘engineering climatology’. He attempted to isolatethe pressures on the
designers and contractors in orderto predict whether or not the structureswill
be prone to an accident.a He discussed the pressures under the headings of
political, financial, scientific, professional and industrial climate,l and identi-
fied eight important parameter?
(U) new or unusual materials
(b) new or unusual methods of construction
( c ) new or unusual types of structure
(d) experience and organization of design and construction team
(e) research and development background
cf> industrial climate
(g)financial climate
(h) political climate
3. Attempts have been made to formulate ways of assessing the magnitude
h e nature of the parameters means that they
of these p a r a r n e t e r ~ . ~ . ~ Tvery
can be assessed only subjectively. However, in common with earlier attempts
at subjective asses~ment,~ the meanings of linguistic values such as small or
large are open to wide individual interpretation. This has been an overriding
difficulty and has prevented any formaluse of subjective assessment in engin-
eering analysis.
4. However, all engineers recognize the importance of engineering judge-
ment based on sound engineering experience. These qualities are used every
day by engineers but only in an informal way. For instance, they are used in
the setting up of analytical models of a structureto be designed, but not in the
actual analysis. Aformulation of the design process has been suggested5
which uses decision theory and involves the subjective assessment of the likeli-
hood of structural accidents as well as a stochastic treatment of improbable
overloading or weakness. In this Paper the problem of subjectively assessing
the likelihood of structural accidents is treated using the notion of fuzzy sets.

Fuzzy sets
5. The basic ideas of fuzzy sets are outlined and some of the fundamental
definitions which are used are nowgiven.Zadeh6e7gives a more complete
discussion.
6 . The approach using the concept of fuzzy sets is different from conven-
tionalquantitativetechniques of system analysis. The analysis of systems
using fuzzy sets is based on the contention that conventional techniques are
unsuitable for dealingwith humanistic systems or any system whose complex-
ity is comparable to that of humanistic systems. As the complexity of systems
increases, ability to make precise and yet significant statements about their
660
Downloaded by [ UNIVERSITY OF EXETER] on [12/09/16]. Copyright © ICE Publishing, all rights reserved.
PREDICTINGTHELIKELIHOOD O F STRUCTURALACCIDENTS
behaviour diminishes. Most of the techniques used at present for the analysis
of humanistic systems are adaptations of methods that have been developed
over a long periodfor dealing with physical systems whichare governed by the
laws of mechanics and so on.
7. The key elements in human thinking, it is assumed, are not numbers but
labels of fuzzy sets, i.e. classesof objects in which the transition frommember-
ship to non-membershipis gradualratherthanabrupt. The ability of a
human brain to summarize information from themass of data which impinges
on it is not well understood, but a logic with fuzzy truths, fuzzy connectives
and fuzzy rules of inference may well play a basic role.
8. The main features are the use of so-called linguistic variables, charac-
terization of simple relationships between variables byfuzzy conditional
statements, and characterization of complex relations by fuzzy algorithms.
9. A linguistic variable isdefined as a variable whosevalues are words,
phrases or sentences in a language. Thus, if tall, not tall, very tall, very very
tall and so on arevalues of height then height is a linguistic variable.
10. Fuzzy conditional statements are expressions of the form‘If A then B,’
where A and B are fuzzy assignments, e.g. ‘If X is small then y is large.’ Fuzzy
algorithms are ordered sequences of instructions whichmay contain fuzzy
assignment and conditional statements.

Definitions
11. No attempt is made here to justify the following expressions but
merely some of those given by Zadeh6n7are repeated.
12. A fuzzy subset of a universe ofdiscourse U is characterized by a member-
shipfunction pA:U+[O, l ] which associates with each element U of U a
number pA(u)in the interval [0, l ] which is the gradeof membership of U in U.
The set A may be thus represented by m discrete values of U together with
membership values p
m
A = 2
i= 1
pi[ut . . . . . - * (1)

where = should be interpreted as ‘is defined to be’ and I is a delimiter and the
summation is taken as the union rather than the arithmetic sum.
+
13. For example if U = 1 + 2 + 3 . . . 10, then a fuzzy subset small may
be expressed as
small = ll1+0.912+0.613+0.414
14. The union of fuzzy sets A and B corresponds to the connective ‘or’
and is
A U B = A + B = J(pA(u)vpB(u))lU . . . . (2)
U
where the symbol Q v b = max(a, b).
15. The intersection of A and B corresponds to the connective ‘and’ and
is
A nB = jbA(u) A pB(u))1u . . . . . (3)
U
where the symbol a h b = min(a, b).
661
Downloaded by [ UNIVERSITY OF EXETER] on [12/09/16]. Copyright © ICE Publishing, all rights reserved.
BLOCKLEY
16. The cartesian product A X B, if B is a fuzzy subset of a different uni-
verse W, is
AxB =
uxw
I p.+(~)A p ~ ( w ) l ~ , . . . . . (4)

17. A fuzzy relation R from X to Y is a fuzzy subset of the cartesian pro-


duct X x Y
R= I
Vdx, y)l(x, v)
XXY
. . . . .(5)
18. If R is a relation fromX to Y and S is a relation from Y to Z then the
composition of R and S is a fuzzy relation denoted by
RoS = I
xxz
V
b
(PR(& Y ) A p&, z))l(X, 2) ' (6)

19. A fuzzifier F is characterized by its kernel K ( y ) which is itself a fuzzy


set
F(A; K) =
U
l pA(y)K(y) . . . . . . (7)

This has the effect of increasing the fuzziness of a fuzzy set.


20. Values of linguistic variables suchas small may be modifiedby linguistic
hedges or operators, e.g. small, quite small and very small. Very is defined by
very X = x2 . . . . . . * (8)
very
very X = (very X)" . . . . . * (9)
21. Other powers of dilution and concentration of X maybeused. For
example, Zadehep7 defines
plus X = X145 . . . . . . . (10)
minus X = x O . " ~ . . . . . (11) 9

These linguistichedges may be combined.


22. The fuzzy conditional statement, 'If A1then B1 else if A2 then B2 ...
else if A,, then B,,' is defined to be
A ~ x B ~ + A ~ x... BA,xB,
~ + . . . . . (12)
Prediction of structural accidents
23. Many structural accidents occur because a major error and/or several
smaller errors combine to eliminate the factors of safety. These errors often
occur because of certain factors inherent to the design or because of external
pressures exerted on the personnel involved. These factors and pressures or
safety parameters are difficult to define and even more difficult to quantify.
However, it is felt that in a given situation it would be possible to isolate
subjectively and define these several parameters andto measure them quantita-
tively using fuzzy linguistic variables.
24. A method is thenrequired whichuses thesevariables to examine
whether or not there is a likely combination of circumstances which would
make a structure accident prone. The variables used in such a method could
be arbitrarily defined as long as they are consistent. The various operations
and manipulations of these variables, which constitute the method,result in a
662
Downloaded by [ UNIVERSITY OF EXETER] on [12/09/16]. Copyright © ICE Publishing, all rights reserved.
PREDICTINGTHELIKELIHOOD OF STRUCTURALACCIDENTS
solution which has also to be interpreted subjectively. However, sucha
solution will indicate in a fuzzy way the likelihood of an accident. In fact
the variables used and the interpretationof the resultcould all be defined by a
suitably experienced committee. All that would be required of the user would
be to state the number and values of the safety parameters.
25. In general terms eleven parametersare considered to beof prime
importanceand these are used to outlineaprocedure.Thisprocedureis
intended only as a possible wayof tackling this problem. The parameters
are as follows.
Materials
1. The degree of confidence in the analytical model used to describe the
behaviour and variability of the parameters (e.g. elasticity).
2. The size of uncatered for effects (e.g. residual stresses).
Type of structure
3. The degree of confidence in the analytical model used to analyse the
structure (e.g. so-called ‘fixed ’ beam column connexions, empirical
research and development information).
4. The complexity of the calculations (arithmetic errors).
Design experience
5. The amount of experience of the design organization in similar types
of structure.

Time
6. The amount of pressure on the designers due to shortage of time.
Construction
7. The degree of confidence in the construction methodsto be used.
8. The amount of experience of the various contractors insimilar types of
structure.

Externals
9. The industrial climate.
10. The financial climate.
11. The political climate.
26. The value of each parametermay be subjectively estimated in two ways
using linguistic variables such as largeand small with suitableoperators,
first in size and second in importance or weighting. The parameter sizes are
then modified so that they act in the same sense(i.e. the larger the more
detrimental) and combined.
27. If a parameter sizeis P, with weighting W, and both are defined as
fuzzy subsets of the interval [0, l ] then the total effect of the parameter PTis
(P1n W,) LJ (Pzn W,) LJ (P3n W,). . .Plln Wll)
or
11
PT = 2 (P,n W,) . . . . . . (13)
i= 1

663
Downloaded by [ UNIVERSITY OF EXETER] on [12/09/16]. Copyright © ICE Publishing, all rights reserved.
BL 0 C KLEY
28. The designers may also be asked to make an a priori judgement of the
degree of confidence they have in the safety of the structure. This could be
done in terms of a linguistic variable-safe-with suitable operators such as
quite, very and not very. In normal circumstances using traditional methods
and factors of safety this judgement will be very safe. The variable very safe
may be defined using an index n on a scale [0,101, where n is related to the
probability of an accident occurring.
29. Equation (13) expresses the total effect of the parameters which affect
the safety of the structure. The problem now becomes one of finding out
whether this total effect is sufficient to alter the a priori judgement to a level
which gives rise to concern. The a priori judgement will actually be fuzzified
in the light of the value of PT. To do this a fuzzifier with a kernel K(n) is
used. The relationship between PT and K(n) may bedefined as(equation
(12))
R = If PT is small then K(n) is small else if PT is medium then K(n) is medium
else if PT is large then K(n) is large (14)
30. The procedure proposed is to find the composition of PT and R, i.e.
PT’R (equation (6)), and from this obtain the kernel K(n) with which to fuzzify
the apriori judgement. The kernel K(n) can be obtained in several ways, the
most conservativeof which isto take themaximum membership levels for each
value of n. The fuzzy statement obtained after the apriori judgement hasbeen
fuzzified is a statement about the likelihood of a structural accident. The
last problem is to identify just when this statement is such that there is cause
for concern. This can be done in severalways. One wayis to choose the
value of n which has thehighest membership level and is the most conservative
if more than one value of n has that membership level. It is then specified
that this value must not be below a certain fixed value. Other ways could be
used, but the crux of the matter is to identify whether the membership levels
of low values of n become significant enough to give rise for concern. This
should be done, of course, bearingin mind the initialdefinition of the linguistic
variable safe.
31. The following numerical example is intended only as an illustration of
the ideas outlined.

Numerical example
32. A design forastructure has just been completed. The design team
considers that the parameters in 0 25 are the only important ones and their
first estimatesof the size and weighting of each of them are shown in Table 1.
33. The following variables may be arbitrarily defined on the scale [0,l ]

l
large = 0~1~0~6+0~2[0~7+0~5~0~8+0~9~0~9+1(1
small = 1 [ 0 + 0 ~ 9 ~ 0 ~ 1 + 0 ~ 5 ~ 0 ~ 2 + 0 ~ 2 ( 0 - 3 + 0. ~ 1. ~ 0
(15)
~4
medium = 0 ~ 2 ( 0 ~ 3 + 0 ~ 6 [ 0 ~ 4 + 1 ~ 0 ~ 5 + 0 ~ 6 ~ 0 ~ 6 + 0 ~ 2 ~ 0 ~ 7

34. Thus one can derive for instance


very large = 0.0110.6+ 0.0410.7+ 0.2510.8+ 0.81 10.9+ 111
+
very very large = 0.0610.8 0.6610.9 1 11 +
quite small = ( ~ m a l l ) l ’ ~ ~
= 1~0+0~87(0*1+0~42~0~2+0~13~0*3+0~05~0~4
664
Downloaded by [ UNIVERSITY OF EXETER] on [12/09/16]. Copyright © ICE Publishing, all rights reserved.
PREDICTINGTHELIKELIHOOD OF S T R U C T U R A L A C C I D E N T S
35. As all the definitions are symmetrical about 0.5 one can easily modify
the sense of parameter estimates so that larger is more deterimental. The
estimates of parameters 1, 3, 5 , 7 and 8 have to be so adjusted and their new
values are also shown in Table 1.
36. The set of the intersection of these estimates for parameter 1, for ex-
ample, is shown in Table 2. Table 3 shows the set PTwhich is the union of all
sets like Table 2 for all parameters (equation (13)). PTalready gives an indi-
cation of the likelihood of a structural accident. Large numbers in theright-
hand bottom cornerof the matrix would indicate a high degree of risk.
37. The apriori state of the structure is defined in terms of the safety index
or n where
Pr = 10-" . . . . . . . . (16)
and pr is the probability of an accident occurring.
38. The index n will generally be on thescale [0,101 and one candefine the
likelihood of an accident as a fuzzy set on this scale either subjectively or using
statistical evidence.
39. Thus
very safe = 1~10+0~9~9+0~6~8+0~2~7+0~1[6 . . (17)

Table 1. Numerical example : parameter estimates

Parameter I Size Weighting


1 First
estimate Estimate used
~ ~ ~~~

1 Large Small Large


2 Medium Medium Medium
3 Large Small Very very large
4 Large Large Small
5 Small Large Small
6 Very large Very large Medium
7 Large Small Very large
8 Very small , Very large Medium
9 Quite small Quite small Small
10 Small Small Small
11 Small Small Small

Table 2. Small size n large weighting for parameter 1

Weighting I Size

665
Downloaded by [ UNIVERSITY OF EXETER] on [12/09/16]. Copyright © ICE Publishing, all rights reserved.
BLOCKLEY

? ? ? FF ?
-oooo*oo

X I !

?c??9\
00004

?N"opop
00000

"??N\e\e\eN
00000000 "N"Fu)
00000

lL:l "??c??
00000

?"?c?\e\ePN"""
00000000000
l7 ??'"?c?
00000

NNNNNc?NNNNN
00000000000

???c?" ? ? V ? ?
00000 00000

4\??c?" "c????
00000 00000
NNc?c?"
71
E 00000

??c"?"c???
H0000 OOOOH
:l
I
???c?"
00000

"c?'?Tr?FY-op9\0
oooooooooo*

666
Downloaded by [ UNIVERSITY OF EXETER] on [12/09/16]. Copyright © ICE Publishing, all rights reserved.
P R E D I C T I N GT H EL I K E L I H O O D O F STRUCTURALACCIDENTS
Table 5. Composition of PT and R ; PToR

Weighting I index Safety

0
0 1
0.2
0.3
1 1
0.9
0.5
0.2
I 0.6
0.6
0.5
0.2
0.4
0.5
0.6
0.5
0.5
0.5
I 0.6
0.6
0.7 0.2
0.8 05
0.9 0.9
1 1

and the kernel K(n) is


small K(n) = lln+0.51(n-l)+0.21(n-2)
largeK(n) = 0-21n+0.6[(n-1)+0.81(n-2)+ll(n-3)
mediumK(n) = 0.5ln+O.5l(n-l)+O-5[(n-2)+0.5((n-3)
The relationship (14) can then be calculated using equation (12) as shown in
Table 4. The composition of PT and R is shown in Table 5 using equation
(6). This again indicates the likelihood of a structural accident with large
numbers in the right-hand bottom corner of the matrix indicating a high
degree of risk.
40. The last problem is to identify the membership values of K(n) with
which to fuzzify very safe. The most conservative estimate would be to .take
the maximum membership values from each column of Table 5. Thus
K(n) = l[n+0.6[(n-1)+0.8[(n-2)+ll(n-3) . . (19)
but this ignores the weightings.
41. If only maxima from the membership values from weightings in the
range [0.6, l ] are taken
K(n) = lln+0.61(~~-1)+0*61(1~-2)+0*61(n-3) . . (20)
42. Thus using (19) to fuzzify (17) very safe becomes (equation (7))
1(1~10+0~6~9+0~8~8+1[7)+0~9(1~9+0~6~8+0~8~7+1~6)+0~6(1~8+0
+0~8~6+1~5)+0~2(1~7+0~6~6+0~8~5+1~4)+0~1(1[6+0~6[5+0~8[4
= 1[10+1~9+1~8+1~7+1~6+0~82~5+0~28~4+ . 0 ~. 1 ~. 3 . (21)
and similarly using (20) to fuzzify (17) very safe becomes
1 ~ 1 0 + 1 ~ 9 + 1 ~ 8 + 1 ~ 7 + 1 ~ 6 + 0 ~ 5 4 [ 5 + 0 ~ 1 8 ~ 4 .+ 0 .~ 0 (22)
6[3
43. There is a limit of concern below which expressions such as (21) and
(22) should not go before rechecking of the structural solution should be
undertaken. This could be, for instance, that for
n 5 4, p > 0.2 . . . . . . . (23)
667
Downloaded by [ UNIVERSITY OF EXETER] on [12/09/16]. Copyright © ICE Publishing, all rights reserved.
BLOCKLEY
or that the most conservative maximum membership level should not be that
at which
n 1 5 . . . . . . . . (24)
44. Using both (21) and (22) the best conservative estimate of the likeli-
hood of a structural accident for the example is n = 6 or p r = 10-6.

Conclusions
45. Using the concept of fuzzy sets, a method for estimating the likelihood
of occurrence of a structural accident can be derived. Such a method can use
arbitrarily yetpreciselydefinedlinguistic variables which are suitablefor
subjective estimation. The linguistic variables such as verysafe, small and
large and the fuzzifier K(n) could be defined by a suitably experienced com-
mittee, thus enabling wide use of a method such as the one proposed. The
methodproposed may be programmed on a desk top computer. Similar
methods may be used at any stage of, say, the construction programme of a
structure, to try to find out if any particular part
of the construction programme
is accident prone. Other variables than those listed would be used in such an
instance.

References
1 . PUGSLEY A. G. Theengineeringclimatologyofstructuralaccidents. Inter-
national conference on structural safety and reliability, Washington, 1969, 335-
340.
2. PUGSLEY A. G . The prediction of the proneness to structural accidents. Struct.
Engr, 1973, 51, No. 6, June, 195-196.
3. BLOCKLEY D. I. DiscussiononreportonTheprediction of theproneness to
structural accidents. Struct. Engr, 1973, 51, No. 12, Dec., 447.
4. PUGSLEY A. G. et al. Report on structural safety. Struct. Engr, 1955,33, May,
141-149.
5 . BLOCKLEY D. I. Structural designdecisionsandsafety. Publs Int. Ass. Bridge
Struct. Engng, 1974, 34-II, Sept., 1-18.
6. ZADEHL. A. Fuzzy sets. Information & Control, 1965, 8, 338-353.
7. ZADEHL.A. Outline of a new approach to the analysisof complex systems and
decision processes. Trans. Systems, Man & Cybernetics, IEEE, 1973, SMC-3,
28-44.

668
Downloaded by [ UNIVERSITY OF EXETER] on [12/09/16]. Copyright © ICE Publishing, all rights reserved.

S-ar putea să vă placă și