Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Standard view
Full view
of .
Look up keyword
Like this
0 of .
Results for:
No results containing your search query
P. 1
A Theory of Independent Fuzzy Probability for System Reliability

A Theory of Independent Fuzzy Probability for System Reliability

Ratings: (0)|Views: 62 |Likes:
Published by ngelectronic5226

More info:

Published by: ngelectronic5226 on Sep 06, 2010
Copyright:Attribution Non-commercial


Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less





A Theory of Independent Fuzzy Probabilityfor System Reliability
James Dunyak, Ihab W. Saad, and Donald Wunsch,
Senior Member, IEEE 
Fuzzy fault trees provide a powerful and compu-tationally efficient technique for developing fuzzy probabilitiesbased on independent inputs. The probability of any event thatcan be described in terms of a sequence of independent unions,intersections, and complements may be calculated by a fuzzy faulttree. Unfortunately, fuzzy fault trees do not provide a completetheory: many events of substantial practical interest cannot bedescribed only by independent operations. Thus, the standardfuzzy extension (based on fuzzy fault trees) is not completesince not all events are assigned a fuzzy probability. Othercomplete extensions have been proposed, but these extensionsare not consistent with the calculations from fuzzy fault trees.In this paper, we propose a new extension of crisp probabilitytheory. Our model is based on
independent inputs, each witha fuzzy probability. The elements of our sample space describeexactly which of the
input events did and did not occur. Ourextension is complete since a fuzzy probability is assigned toevery subset of the sample space. Our extension is also consistentwith all calculations that can be arranged as a fault tree. Ourapproach allows the reliability analyst to develop complete andconsistent fuzzy reliability models from existing crisp reliabilitymodels. This allows a comprehensive analysis of the system.Computational algorithms are provided both to extend existingmodels and develop new models. The technique is demonstratedon a reliability model of a three-stage industrial process.
 Index Terms—
Fuzzy fault trees, fuzzy probability, fuzzy sets,independence.
I. I
ANY system reliability models require (as input) theprobabilities of a number of independent events. Oftenthese probabilities can be estimated from data or theory, butsometimes choosing probabilities for input is difficult. Thiswork is part of an ongoing study in high-consequence suretyanalysis. Many of the factors of interest come from tradi-tionally nonmathematical areas of research such as estimatingthe probability of a terrorist attack, compliance with safetypractices, or a flawed design of a safety system. Other factorsare too expensive or dangerous to measure experimentally.Instead, expert opinion is used to provide these probabilities,but these estimates are rarely precise. Fuzzy sets and possi-bility theory provide a tool for describing and analyzing theseuncertain quantities.
Manuscript received December 1, 1997; revised June 29, 1998. This work was supported in part by the Idaho National Engineering and EnvironmentalLaboratory.J. Dunyak is with the Department of Mathematics, Texas Tech University,Lubbock, TX 79409 USA.I. W. Saad and D. Wunsch are with the Department of Electrical Engineer-ing, Texas Tech University, Lubbock, TX 79409 USA.Publisher Item Identifier S 1063-6706(99)02796-4.
An uncertain parameter may be assigned a fuzzymembership function : [0, 1], which is the mem-bership function of a fuzzy set . Then the possibility thatis in a set is designated by andThis is the sense in which we describe uncertainty in theprobability of an event .In this paper, is a fuzzy set describing uncertainty inthe crisp probability and : [0, 1] [0, 1]. Fuzzyfault trees provide a method for developing fuzzy probabilitiesbased on independent fuzzy inputs [1]. The probabilityof any event that can be described in terms of a sequenceof independent unions, intersections, and complements maybe calculated by a fuzzy fault tree. Unfortunately, we showbelow that some events of substantial practical interest cannotbe described only by independent operations: Standard fuzzyfault trees do not provide a complete theory. Thus, the standardfuzzy extension (based on fuzzy fault trees) is not completesince not all events are assigned a fuzzy probability. Zadehproposed another extension which is complete [11], but hisextension is shown (in our context) to be inconsistent with thecalculations from fuzzy fault trees. For recent developments of fuzzy probability, see [2]–[5]. Walley and de Cooman, in par-ticular, discuss completeness and consistency in a more generalsetting. See also Cai’s work on system failure engineering [6]and other discussions of fuzzy fault trees [7]–[10].Here we develop a new extension of crisp probability theorybased on independent inputs, each with a fuzzy probability.The elements of our sample space describe exactly which of the input events did and did not occur. This extension willbe shown to be both complete and consistent.II. I
Throughout this paper, we use the bar notation toindicate a fuzzy set representing probability of , the notationto indicate the corresponding upper semicontinuousmembership function and : to indicatethe corresponding cuts. A convex fuzzy set has specialstructure: each cut is a closed and convex subset of .We see for a convex fuzzy probability that each cut canbe written as a closed interval with , . Thisassumption of convexity is equivalent to assuming that themembership function has a single mode. Earlier work withindependent fuzzy probabilities relied on this (often quitereasonable) assumption of convexity, but our work will be
1999 IEEE
et al
more general. Following the lead of most fuzzy models, allfuzzy sets here are required to have nonempty cut. Thisproperty follows here from normality.Consider independent events , , , , withestimated fuzzy probabilities , , , , which willbe used in a reliability model. Following the lead of Tanaka
et al.
[1] and many others, our concept of independence iscrisp. We assume the underlying probabilities for , ,, , are independent in the conventional (crisp) sense.Our goal is to build a fuzzy probability theory to describe theprobabilities of various unions, intersections, and complementsof these sets. To this end, we follow the standard approachof Tanaka
et al.
[1] and first build fuzzy intersections of independent events.If events are independent, then for crisp probabilities wehaveandUsing the usual extension principle, we define the fuzzyindependent union and intersection as(1)and(2)Complements of fuzzy probabilities are similarly defined by(3)We then have the following familiar properties:(4)This third formula is DeMorgan’s law and extends in theobvious way to(5)If the fuzzy probabilities are convex, we have the relationshipsbetween endpoints of the -cut intervals(6)and(7)
Fig. 1. A fuzzy fault tree that maintains independence.
Unfortunately, the distributive laws fail. Straightforwardapplication of the above formulas show(8)This formula fails because and are notindependent so (4) cannot be applied.As we see in (8), care must be used in organizing cal-culations to maintain independence. This is usually done bydescribing calculations as a tree structure. This viewpoint wasnaturally assumed in several papers on fuzzy fault trees [1],[12]–[16]. To illustrate this concept, consider the exampletree diagram in Fig. 1. This diagram contains three varietiesof nodes: unions, intersections, and complements. At thenodes, fuzzy input probabilities are combined according to theformulas in (1)–(3). As long as the tree only feeds upward andeach node has only one output, independence is maintained.Because of DeMorgan’s laws in (5), we can develop fault treesusing only unions and intersections (but no complements) oronly intersections and complements (but no unions). Thus,several somewhat different approaches to fault trees are in factequivalent when the standard extensions in (1)–(3) are used.Unfortunately, many problems do not easily fit into astraightforward tree structure, with each node having only oneoutput. In our investigations, certain factors (such as terrorismrisk) influence many different events so that construction of independent trees is problematic. As we will see in the nextsection, other problems also occur.III. C
The representation of some sets can be rearranged to allowuse of (1)(3). For example, in (8), since is not(necessarily) independent of , we could simply define(9)Now and are independent so we can correctly calculateusing (2). Since is independent of ,we can apply (1) to calculate . Unfortunately,unraveling such relationships can be very difficult in complexmodels. Of greater concern is the fact that not all possiblefuzzy probabilities can be calculated by rearranging into acalculation that maintains independence.For example, a listing of all possible independent cal-culations easily shows that may not berearranged to allow calculation by independence formulas.

You're Reading a Free Preview

/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->