Hindawi Publishing CorporationAdvances in Fuzzy SystemsVolume 2012, Article ID 957697, 12 pagesdoi:10.1155/2012/957697
Research Article
A Hybrid Approach to Failure Analysis Using StochasticPetri Nets and Ranking Generalized Fuzzy Numbers
Abolfazl Doostparast Torshizi and Jamshid Parvizian
Department of Industrial Engineering, Isfahan University of Technology, Isfahan 84156-83111, Iran
Correspondence should be addressed to Abolfazl Doostparast Torshizi, [email protected]
Received 25 April 2012; Accepted 4 September 2012
Academic Editor: Zeng-Guang Hou
Copyright © 2012 A. Doostparast Torshizi and J. Parvizian. This is an open access article distributed under the Creative CommonsAttribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work isproperly cited.
We present a novel failure analysis approach combining structural properties of stochastic Petri Nets and flexibility of fuzzy logic.Firstly, we develop a powerful fuzzy ranking technique. We analyze major drawbacks of existing ranking techniques. Then wedemonstrate the capabilities of the presented algorithm to overcome such drawbacks. The approach considers weight, spread, anddifference of x coordinate of the center of gravity (COG) point of each fuzzy number and is able to deal with a wide variety of fuzzynumbers. Using this technique, we utilize isomorphism between stochastic Petri Nets and their corresponding Markov chains andpresent a failure analysis algorithm incorporating some critical factors. This algorithm can be implemented in diverse industrialapplications.
1. Introduction
Failure can be defined as any unwanted deviation from thedesired predetermined plan, which may lead to any kindof human injuries or damages of machines. In order toperform corrective actions, prioritizing failures is necessary.One of the most popular risk analysis procedures aimingto prioritizing failure states is Failure Mode and EffectAnalysis (FMEA) which is based on three factors of severity,detectability, and occurrence of failure. This method wasintroduced in 1960s. Its wide applications are documented byBraglia et al. [1] and Stamatis [2]. Process of calculating riskpriority numbers in this method consists of multiplication ofthe three-mentioned risk factors. It is apparent that variouscombinations of risk factors may lead to a constant riskpriority number. This is the most critical challenge FMEAfaces. However, there are many other weaknesses in thismethod which makes it impractical in real world problems[3, 4].
Up to now, many different approaches have been pre-sented to overcome the limitations of FMEA. These methodsrely on diverse techniques such as grey theory [4], BayesianNets [5], Monte Carlo simulation [6], Markov models [7],and fuzzy logic [8]. Among all these techniques, fuzzy logic
has been extensively applied in risk analysis. Fuzzy logic iscapable of handling vagueness of human suggestions; so thatit is useful in prioritizing risky behaviors of the systems.
Fuzzy risk analysis is a new field firstly introducedby Schmucker [9]. This method is similar to traditionalFMEA; however, risk parameters are fuzzy numbers. Fuzzyarithmetic is used to obtain fuzzy risk priority numbers. Thismethod is rather versatile although suffers from ignoringmany critical risk parameters. Using fuzzy risk analysisand fuzzy arithmetic, many different researches have beenperformed so far. Among them, combination of fuzzy riskanalysis and ranking methods are extensively studied. Basedon Schmucker [9], a fuzzy risk analysis method usingsignal/noise ratio was presented by Chen and Wang [10], inwhich fuzzy risk priority numbers were ranked by this ratio.Different kinds of fuzzy numbers such as generalized onesand their application in safety analysis are investigated by S.M. Chen and J. H. Chen [11, 12]. Application of similaritymeasures between fuzzy numbers is investigated in [11–13].The fuzzy risk analysis method proposed by Schmucker [9] iswidely utilized in so many researches performed so far. Fuzzyrisk analysis methods are based on either ranking techniquesor similarity measures, with differences only in prioritizationstyle. In ranking based methods fuzzy risk priority numbers
2 Advances in Fuzzy Systems
(FRPN) are ranked by different ranking techniques; whilein similarity based techniques, the same FRPN are rankedby measuring the similarity level between FRPN and somepredetermined linguistic variables.
Failure analysis techniques traditionally rely on staticevaluation of systems, which fails in tracking risky behaviors.Hence, implementing dynamic tools to enable auditors tomonitor systems dynamically seems beneficial. One of themost common tools in modeling dynamic systems are PetriNets (PN) [21, 22]. PN are powerful modeling methodswhich can simulate behaviors of systems dynamically in sucha way that different states of the system can be visualized.There are many different versions of Petri Nets includingtimed, stochastic, and colored. In stochastic Petri Netsfiring of transitions depend on a stochastic variable withexponential distribution.
In this paper, we introduce a novel failure analysismethod in which the Petri Net model of the system isconstructed. This model includes different potential risks.According to isomorphism between live bounded stochasticPetri Nets and Markov chains, fuzzy steady state probabilitiesare calculated in the form of generalized trapezoidal fuzzynumbers. In these fuzzy probabilities, the weight of eachfuzzy number is a combination of several new risk factors.Finally, these fuzzy probabilities are ranked by a newpowerful fuzzy ranking method.
The rest of the paper is structured as follows. In Section 2,the bases of Stochastic Petri Nets (SPN) are given. SPNwith fuzzy parameters and their application in safety analysisof the systems are summarized in Section 3. In Section 4,a novel ranking method of generalized fuzzy numbers ispresented and compared with other methods in the field.In Section 5, we deal with a risk prioritizing method usingSPN and generalized fuzzy numbers which opens a newfield in risk prioritizing techniques literature. In Section 6,an illustrative example is presented to clarify the methodsproposed in the paper. The paper is concluded in Section 7.
2. Stochastic Petri Nets
SPN are a family member of Petri Nets in which firing ratesare exponentially distributed. Before introducing SPN, wepresent the definition of Petri Nets introduced by Petri [23].Petri Net (PN) Z = (P,T , I ,O,m) is a five-tuple, where
(1) P = {p1, p2, . . . , pn},n > 0 is a finite set of placespictured by circles,
(2) T = {t1, t2, . . . , ts}, s > 0 is a finite set of transitionspictured by bars,
(3) I = P × T → N is an input function that definesthe set of directed arcs from P to T where N ={0, 1, 2, . . .},
(4) O = T×P → N is an output function that defines theset of directed arcs from T to P,
(5) m : P → N is a marking whose ith component repre-sents the number of tokens in the ith place. An initialmarking is denoted by m0. The tokens are pictured bydots.
PN are able to model systems dynamically. One of thecrucial drawbacks in utilizing ordinary PN is their inability inhandling some important factors such as time and vagueness.To solve such problems various kinds of nets such as timed,fuzzy, and stochastic PN have been developed.
In SPN, the set of firing rates Λ = (λ1, λ2, . . . , λs) areexponentially distributed, such that each transition can befired only after an exponentially distributed time delay withparameter 1/λ elapses.
An important aspect of stochastic PN is their iso-morphism with Markov chains. It has been proved thatlive and bounded SPN are isomorphic to continuous-timeMarkov chains [24]. This important property makes SPNanalysis straightforward. In SPN, each marking of the netis equivalent to the states of its corresponding Markovchain. Therefore, some important factors, like steady stateprobabilities, are easily computed. Here, we present theformal definition of SPN.
A SPN Z = (P,T , I ,O,m0,Λ) is a six-tuple, where
(1) P = {p1, p2, . . . , pn},n > 0 is a finite set of places,
(2) T = {t1, t2, . . . , ts}, s > 0 is a finite set of transitionswith P ∪ T /=∅, and P ∩ T = ∅,
(3) I = P × T → N is an input function that definesthe set of directed arcs from P to T where N ={0, 1, 2, . . .},
(4) O = T × P → N is an output function that definesthe set of directed arcs from T to P,
(5) m : P → N is a marking whose ith component repre-sents the number of tokens in the ith place. An initialmarking is denoted by m0, and
(6) Λ : T → R+ is a firing function whose ith componentrepresents the firing rate of the ith transition whereλi denotes the firing rate of ti and R+ is the set of allpossible real values.
Firing rules in SPN are simply similar to ordinary PNwith a difference only in firing times of transitions. In SPN,when a transition is enabled, all tokens in the upstreamplaces remain in their places until the firing time of thecorresponding transition elapses; then the tokens depositedin upward places are removed and added to all downwardplaces of that transition.
In order to construct the corresponding Markov chain ofSPN, the reachability graph of the net must be constructedfirst. Then using (1), the steady state probabilities can becalculated,
ΠQ = 0,s∑
i=0
πi = 1 (1)
in which πi is the probability of being in state Mi; andΠ = (π1,π2, . . . ,πs). From steady state distribution Π, this ispossible to predict the performance of the system. For moredetails the reader is referred to Bause and Kritzinger [25].
Advances in Fuzzy Systems 3
X
1
˜AW
˜A
Figure 1: A generalized trapezoidal fuzzy number.
3. Stochastic Petri Nets with Fuzzy Parameters
Safety is the knowledge of utilizing recorded data to predictpotential failure in future. In normal conditions, failuresoccur rarely. Thus, expertise in predicting the frequency ofdifferent failure states is necessary. Suggestions by expertsare usually vague and do not include specific numericalvalues and are given often in linguistic variables. Classicalmathematics faces difficulties in handling nondeterministicvalues. Fuzzy set theory is a powerful method for handlingvague conditions.
In this paper, according to vagueness of experts’ knowl-edge in predicting failure rates of different risky states, wepropose a new approach to determine fuzzy steady stateprobabilities based on generalized fuzzy numbers.
As explained earlier, SPN are isomorphic to continuous-time Markov chains; thus steady state probabilities canbe computed by dominating rules on Markov chains. Weconsider the failure rates of each potential risk in the systemas a fuzzy number. Let us introduce some notations used inthe rest of the paper; for more information, see [26].
Definition 1. A fuzzy number is a convex normalized fuzzyset M of the real line R such that
(1) there exists exactly one x0 ∈ R with μm(x0) = 1 (x0 iscalled the mean value of M);
(2) μm(x0) is piecewise continuous.
For the sake of computational efficiency, some specialforms of fuzzy numbers with triangular or trapezoidalmembership functions are used. Sometimes, a more generalform of fuzzy numbers is needed. For instance, in thispaper, generalized fuzzy numbers are used with the extrafeature of weighted membership functions. Their generalityis according to the weight of the mean value or thesupremum of their membership function. In Figure 1 ageneralized trapezoidal fuzzy number is depicted.
Definition 2. An alpha cut (α-cut) of a fuzzy number A, if itis a subset of the set Ω, is defined as:
A(α) ={x ∈ Ω | A(α) ≥ α
}0 < α ≤ 1. (2)
On the other hand, each fuzzy number can be represented byits alpha cut. For example, in a fuzzy number Q we have
Q(α) = [q1(α), q2(α)
](3)
in which values of q1(α) and q2(α) are the lower and upperbounds of this alpha cut, respectively.
Fuzzy arithmetic offers two concepts: extension principleand operations between alpha cuts. In this paper, we rely onthe second concept since it can be adapted by the extensionprinciple; in addition, incorporating alpha cuts is easier.
Let us consider two fuzzy numbers A and B and theiralpha cuts A(α) = [a1(α), a2(α)] and B(α) = [b1(α), b2(α)],respectively. Operations between fuzzy numbers in theframework of alpha cuts are
A(α) + B(α) = [a1(α) + b1(α), a2(α) + b2(α)],
A(α)− B(α) = [a1(α)− b2(α), a2(α)− b1(α)],
A(α) · B(α) = [c(α),d(α)]
(4)
in which,
c(α)=min{a1(α)b1(α), a1(α)b2(α), a2(α)b1(α), a2(α)b2(α)},d(α)=max{a1(α)b1(α), a1(α)b2(α), a2(α)b1(α), a2(α)b2(α)}.
(5)
The dividing operation among fuzzy numbers is defined as
A(α)
B(α)= [a1(α), a2(α)] ·
[1
b2(α),
1b1(α)
]. (6)
4. Ranking Method of GeneralizedFuzzy Numbers
Application of fuzzy sets theory in reliability and safetyengineering has been an active field of research in recentyears. One of the most alluring fuzzy techniques is rankingof fuzzy numbers. Chen and Wang [10] proposed a fuzzy riskanalysis method based on signal/noise ratio. For more details,reader is referred to [11–13].
In this section, we propose a novel approach for rankinggeneralized trapezoidal fuzzy numbers. This approach will beused for risk analysis in the following sections. The algorithmis as follows.
Step 1. Consider the generalized fuzzy number Ai =(ai1, ai2, ai3, ai4;wA). Use (7) for standardizing the general-ized fuzzy number
Ai∗ =(ai1k
,ai2k
,ai3k
,ai4k
;wAi∗
),
k = maxi j(⌈∣∣∣ai j
∣∣∣⌉
, 1).
(7)
4 Advances in Fuzzy Systems
Table 1: Comparison results of Figure 2.
Ranking method A B
S. M. Chen and J. H. Chen [11] 0.1375 0.1375The proposed method 1.68486 0.30964
Step 2. Calculate the center of gravity point of the standardfuzzy number, (7),
yAi∗ =
⎧⎪⎪⎪⎨⎪⎪⎪⎩
((ai3 − ai2)/(ai4 − ai1) + 2)wAi∗
6ai1 /= ai4,
wAi∗
2, ai1 = ai4,
xAi∗ =yAi∗ (ai3 + ai2) + (ai4 + ai1)
(wAi∗ − yAi∗
)
2wAi∗.
(8)
Step 3. Calculate the standard deviation of each standardizedfuzzy number,
STDAi∗ =
√√√√∑4
j=1
(a∗i j − xAi∗
)2
4− 1.
(9)
It is apparent that the interval of the obtained standarddeviation is [0, 1.1547]. Variation of a crisp value is zeroand variation of the generalized fuzzy number (−1, −1, −1,−1:w) is 1.1547.
Step 4. Calculate the difference of x coordinate of each fuzzynumber from the least value,
α = x∗ − x∗min
x∗max − x∗minif x∗min /= x∗max (10)
in which
x∗max = Max{xA1∗ , xA2∗ , . . . , xAn∗
},
x∗min = Min{xA1∗ , xA2∗ , . . . , xAn∗
},
α = 0 if x∗min = x∗max /= 0,
α = 0.5 if x∗min = x∗max = 0.
(11)
Step 5. Calculate the ranking value of each standard general-ized trapezoidal fuzzy number,
RankAi∗ =xAi∗
(wAi∗ + 1
)+ α
1 + STDAi∗. (12)
Notation 1. The proposed ranking method concentrates oncrisp value of the fuzzy number in contrast to its deviation.
Example 3. In order to demonstrate the capabilities of themethod we have compared the ranking output with sixother important and common ranking techniques. Thiscomparison is performed for two groups of fuzzy numbers,Figures 2 and 3.
The ranking is shown and compared for the first groupin Table 1. This is clear that S. M. Chen and J. H. Chen [11]
1
0.8
0.2
0.2 0.4 0.6 0.8 1
˜B
˜A
Figure 2: Group 1: Two generalized triangular fuzzy numbers.Both deviations and (mean value × weight) are the same for bothnumbers.
method fails to rank the fuzzy numbers A and B. This is thecase whenever the deviations and (mean value × weight) arethe same for fuzzy numbers, [27].
It is noteworthy to mention that in our proposedmethod, crisp value of the fuzzy number has priority over itsdeviation and spread. Therefore, the number A has priorityover the other fuzzy numbers.
The second group, Figure 3, consists of eight sets of fuzzynumbers with different shapes and deviations suitable toevaluate our ranking algorithm [12]. Table 2 gives the resultof ranking for this group, ranked using the proposed methodand seven different algorithms presented in the literature.The highlighted items indicate invalid ranking by differentalgorithms. Cheng’s method [14] and Chu’s method [15]cannot rank fuzzy numbers of the second and third sets.Murakami’s technique [16] gives the same result for the twomembers of set 3. Yager’s method [17], as one of the mostcommon ranking techniques, gives the same result for set 2,set 3, and set 4. This shows that this algorithm is not capableof ranking fuzzy numbers in general. All these algorithms failin ranking of fuzzy numbers in the set 5. S. J. Chen’s and S. M.Chen’s method [18] ranks the numbers correctly; however,the ranking scores are very close to each other. For sets 2 and6, Lee and Chen’ method [19] ranks the fuzzy numbers in anincorrect order.
As noted before, the main focus of the method proposedin the current paper is on crisp values of fuzzy numbers; thatis why S. M. Chen’s and J. H. Chen’s method [11] has a validbut different ranking result. The latter method puts priorityon variation and spread of numbers instead of their crispvalues.
According to the results presented in Table 2, Figure 5shows the percentage of correct answers of the comparedmethods. It can be seen that only the proposed method andthe approach proposed by S. J. Chen and S. M. Chen [18] hassuccessfully solved the entire fuzzy sets of Figure 3.
Example 4. One of the latest fuzzy risk analysis approaches,based on ranking fuzzy numbers, is presented in [20], inwhich the ranking technique is based on the areas between
Advances in Fuzzy Systems 5
Ta
ble
2:C
ompa
riso
nre
sult
sof
Figu
re3.
Ran
kin
gm
eth
odSe
t1
Set
2Se
t3
Set
4Se
t5
Set
6Se
t7
Set
8A
BA
BA
BA
BA
BA
BA
BA
BC
Ch
eng’
sm
eth
od[1
4]0.
5831
0.70
710.
5831
0.58
310.
5831
0.58
310.
461
0.58
310.
4243
∗0.
5831
0.58
310.
7673
0.72
410.
680.
7275
0.74
62C
huan
dTs
ao’s
met
hod
[15]
0.15
0.25
0.15
0.15
0.15
0.15
0.12
0.15
0.15
∗−0
.15
0.15
0.28
70.
2619
0.22
810.
2624
0.27
84
Mu
raka
mie
tal
.’sm
eth
od[1
6]0.
30.
50.
30.
4176
0.3
0.3
0.23
330.
30.
4167
∗−0
.30.
30.
60.
50.
440.
5333
0.52
5
Yage
r’s
met
hod
[17]
0.3
0.5
0.3
0.3
0.3
0.3
0.3
0.3
0.3
∗−0
.30.
30.
60.
50.
440.
5333
0.52
5S.
J.C
hen
and
S.M
.Ch
en’s
met
hod
[18]
0.44
560.
4884
0.42
390.
4456
0.44
560.
4728
0.35
650.
4456
0.42
40.
860.
4456
0.74
730.
4128
0.40
050.
3719
0.41
550.
3979
S.M
.Ch
enan
dJ.
H.C
hen
’sm
eth
od[1
1]0.
2579
0.42
980.
2537
0.25
790.
2579
0.27
740.
2063
0.25
790.
2537
1−0
.258
0.25
790.
4428
0.40
430.
3354
0.40
790.
4196
Lee
and
Ch
en’s
met
hod
[19]
0.60
976
0.70
223
0.61
531
0.60
976
0.60
980.
6299
0.50
980.
6097
0.61
530.
880.
5244
0.52
230.
7152
0.69
680.
6507
0.68
960.
6695
Th
epr
opos
edm
eth
od0.
5157
81.
7192
50.
5073
70.
5157
70.
5158
0.55
470.
4641
90.
5157
0.50
733
−0.5
158
1.37
541.
6923
0.76
920.
6521
1.60
531.
462
6 Advances in Fuzzy Systems
˜B˜A
0.1 0.3 0.5 0.7 0.9X
1
Set 1
˜B = (0.3, 0.5, 0.5, 0.7; 1)
˜A = (0.1, 0.3, 0.3, 0.5; 1)
˜B
˜A
X
1
0.1 0.3 0.5 0.7 0.9
Set 2
˜B = (0.1, 0.3, 0.3, 0.5; 1)
˜A = (0.1, 0.2, 0.4, 0.5; 1)
˜B
˜A
X
1
0.1 0.3 0.5 0.7 0.9
Set 3
˜B = (0.2, 0.3, 0.3, 0.4; 1)
˜A = (0.1, 0.3, 0.3, 0.5; 1)
˜B
˜A
1
0.1 0.3 0.5 0.7 0.9
X
Set 4
˜B =˜A = (0.1, 0.3, 0.3, 0.5; 0.8)
(0.1, 0.3, 0.3, 0.5; 1)
˜B˜A1
X
0.1 0.3 0.5 0.7 0.9
Set 5
˜B =˜A = (0.1, 0.2, 0.4, 0.5; 1)
(1, 1, 1, 1; 1)
˜B˜A 1
X
(0.1, 0.3, 0.3, 0.5; 1)
Set 6
˜B =˜A = (−0.5,−0.3,−0.3,−0.1; 1)
˜B˜A
1
X
0.1 0.3 0.5 0.7 0.9
Set 7
˜B = (0.1, 0.6, 0.6, 0.8; 1)
˜A = (0.3, 0.5, 0.5, 1; 1)
˜B
˜A1
X
0.1 0.3 0.5 0.7 0.9
Set 8
˜B = (0.2, 0.5, 0.5, 0.9; 1)
˜A = (0, 0.4, 0.6, 0.8; 1)
˜C
˜C
= (0.1, 0.6, .0.7, 0.8; 1)
Figure 3: Group 2: seven sets of fuzzy numbers used to test the ranking algorithm [12].
the left and the right parts of the membership function ofeach fuzzy number with 1 and −1 as possible infimum andsupremum values of fuzzy numbers. However this approachis not able to deal with symmetric fuzzy numbers. Insuch cases, the result of ranking for each symmetric fuzzynumber will be zero. For example, consider two symmetric
generalized fuzzy numbers A = (−0.8, 0, 0, 0.8; 0.5) and B =(−0.5, 0, 0, 0.5; 0.7) as depicted in Figure 4. Table 3 comparesthe results using Chen and Sanguansat method [20] and thecurrent algorithm. It is clear that the Chen and Sanguansatmethod [20] is not able to rank symmetric generalized fuzzynumbers.
Advances in Fuzzy Systems 7
˜B
W
˜A
−1 −0.8 −0.5 0.5 0.8 1X
Figure 4: Two symmetric fuzzy numbers.
0102030405060708090
100
(%)
Ch
eng’
s m
eth
od [
14]
Yage
r’s
met
hod
s [1
7]
Lee
and
Ch
en’s
met
hod
[19
]
Th
e pr
opos
ed m
eth
od
Chu
an
d Ts
ao’s
met
hod
[15
]
Mu
raka
mi e
t al
’s m
eth
od [
16]
S. J.
Ch
en a
nd
S. M
. Ch
en’s
met
hod
[18
]
S. M
. Ch
en a
nd
J. H
. Ch
en’s
met
hod
[12
]
Figure 5: Percentage of correctness of the compared methodaccording to Table 2.
Table 3: Ranking of numbers given in Figure 4.
Ranking method A B
Chen and Sanguansat [20] 0 0Current paper 0.3024 0.3464
5. Risk Prioritizing
Modeling complex systems using ordinary Petri Nets is adaunting task, since implementation of some concepts suchas time, possibility, and probability is not considered inthe initial definition of ordinary PN. Generally, analysisof complex systems includes two kinds of uncertainties[18]: stochastic situations and fuzzy states. For stochasticsituations the behavior of system parameters is describedby probability distribution functions. In other words, thiskind of uncertainty models randomness. On the other hand,uncertainty in fuzzy form models the level of measure-ment accuracy using linguistic structures and insufficientinformation. There are many inaccuracy sources in systems,such as inaccurate internal operations. In some cases,uncertainty is the result of both randomness and inaccuracy,simultaneously. In stochastic PN, where time is the only
existing stochastic parameter, system delays can be describedby probability functions. It is worth to note that duringsystem analysis, existing uncertainties may be hidden inthe final results. Therefore, utilizing fuzzy set theory is animportant alternative to overcome this drawback.
Although the dominant paradigm in describing uncer-tainties of models is stochastic modeling based upon prob-ability, using such models is only appropriate for describingstochastic states among entire uncertain situations. This ismore important when considering inaccuracy of some datawhich are not instinctively statistical [28].
Here, a comprehensive approach to prioritize differentrisk states of the system is presented. In this approach, byutilizing probability determination via steady state probabil-ities of stochastic PN, fuzzy probabilities of occurrence ofeach failure are calculated. In our approach, parameters ofthe exponential distribution are deemed to be generalizedfuzzy numbers. Eventually, values of the resulting fuzzyprobabilities are ranked by the novel ranking methodpresented in Section 3. In the following, we present theprocess of prioritization of failure modes. Our approach hasthree stages where each stage consists of several substages.This approach extends the method proposed in [29] to a riskanalysis method.
Stage 1. (1) Modeling the desired system using Petri Netsand determining the entire failure modes and allocatingexponential firing times to considered transitions.
(2) Constructing reachability graph of the net anddetermining all states.
(3) Incorporating (1) in order to calculate the steadystate probabilities of the system parametrically based onexponential rates of system transitions.
Stage 2. (1) Conversion of parametric probabilities of (3) inStage 1 to triangular fuzzy numbers considering each param-eter.
(2) Determination of fuzzy probabilities based on alphacuts of each fuzzy number using (4) to (6).
(3) Calculation of each probabilistic value (πi) anddetermination of maximum and minimum values of thisprobability (α = 0). When α = 0, each probability πi must bein interval [0, 1] to be feasible; thus next steps are proceededif this condition does not hold.
(4) Operations among fuzzy numbers using alpha cutsdepend on minimum and maximum operators. This willprovide a larger interval during the calculation. Theoretically,α = 0 cut of a fuzzy number gives the largest interval ofthe number. Since our aim is to find the fuzzy probabilityvalues, the largest possible value of fuzzy numbers mustbe constrained to [0, 1]. Therefore, our aim is to find theshortest alpha:
Min (Z) = α
St. π+i (α) ≤ 1
π−i (α) ≥ 0
π−i (α) ≤ π+i (α)
0 ≤ α ≤ 1.
(13)
8 Advances in Fuzzy Systems
Table 4: Classification of safety severity of each failure mode.
State Sj1
Very dangerous, without warning and periodicalinspection
1
Very dangerous, without periodical inspection 0.9Very dangerous with automatic warning system 0.8Dangerous, without warning and periodicalinspection
0.7
Dangerous, with periodical inspection or warningsystem
0.6
Average danger, without warning system andperiodical inspection
0.5
Average danger with warning system andperiodical inspection
0.4
Low danger, without warning system orperiodical inspection
0.3
Low danger, with warning system or periodicalinspection
0.2
Without any important risk 0.1No safety risk 0
Table 5: Classification operational dependability between failuremodes.
State Sj2
Very high, many numbers of machines malfunctioned 1Very high, loss of quality in many of machines 0.9High, an entire machine is off, a bottleneck is made forsome sets of machines
0.8
High, an entire machine is off 0.7Average to high, the initial performance of machine islost but some tasks are possible to perform
0.6
Average, the initial performance of machine is lost butsome tasks are possible to perform
0.5
Average, machine loses its functionality in some specifictasks
0.4
Low, quality decrease in secondary functions ofmachines
0.3
Low, machine loses a little part of its functionality 0.2Very low 0.1No dependability 0
Stage 3. (1) Converting each resulting failure probability toa generalized fuzzy number. This conversion is performedby combining severity index of each failure mode withfuzzy probabilities calculated in Stage 2. In this research, wehave considered five critical factors to determine occurrenceweight of each failure mode. These factors are maintenancecosts, operational dependability, safety, failure detectionmethods, and repairing time.
In order to obtain these factors easily, we have providedsome linguistic variables with their corresponding weightspresented in Tables 4, 5, 6, 7, and 8. Finally, these criticalfactors must be combined to get the severity factor of eachfailure mode. This process includes a multiplication of thesevariables as
Sj = Sj1 × Sj2 × Sj3 × Sj4 × Sj5 (14)
Table 6: Classification of detection and identification of failuremodes.
State Sj3
Uncertainty 1
Very unlikely 0.9
unlikely 0.8
Very low 0.7
low 0.6
Average 0.5
Average to high 0.4
high 0.3
Very high 0.2
Nearly definite 0.1
Completely definite 0
Table 7: Classification of repairing costs of failure modes.
State Sj4
Is not worth to fix 1
Hardly worth to fix 0.9
Extreme 0.8
High 0.7
Average high 0.6
Average 0.5
low 0.4
Average low 0.3
Low 0.2
Low importance 0.1
No cost 0
Table 8: Classification of repairing time of a failure mode.
State Sj5
Very time consuming, no worth to fix 1
Long fixing time, hardly worth to fix 0.9
Very time consuming to fix 0.8
Long fixing time 0.7
Average to high fixing time 0.6
Average fixing time 0.5
Short fixing time 0.4
Fairly short fixing time 0.3
Very short fixing time 0.2
Fixing time not very important 0.1
Fixing time negligible 0
in which Si j represents the value of ith factor from fivecritical factors influencing failure j and Sj represents the finalseverity measure of the failure j.
(2) Ranking each generalized fuzzy probability providedin (1) in Stage 3 using the ranking technique proposed inSection 3.
Advances in Fuzzy Systems 9
Table 9: Description of Places of Figure 7.
Place Description
P1 Work-piece ready
P2 Robot in progress
P3 Machine in process
P4 Robot in repair
P5 Machine in repair
P6 Robot idle
P7 Machine idle
Outgoing conveyor
Machine
Incoming conveyor
Figure 6: A flexible manufacturing cell.
6. An Illustrative Example
In this section the approaches proposed in the paper areapplied to a flexible manufacturing (FM) cell, adopted from[30]. This cell is of course adapted with the nature offailure analysis.The cell, Figure 6, has one incoming and oneoutgoing conveyor, one robotic arm, and one processingmachine. Work-pieces enter the cell by an incoming conveyorand the robot (R) loads them to the machine (M). Sincewe want to consider different potential failure modes ofthe system, a breakdown loop is considered for the robotand processing machine which consists of a breakdown anda repair transition with their corresponding rates. Whenprocessing on M is over, it is unloaded by R and the work-piece will go to the outgoing conveyor. Now suppose that
(1) the processing machine can have failure modes. Mtakes two time units to breakdown and a quarter timeunit to be repaired. Therefore, the average failure andrepair rates are 0.5 and 4, respectively;
(2) robot loading and taking work-piece rate is 45 perunit time. Also its unloading rate plus average rate ofM processing is 8 per unit time. Robot is not failurefree; hence, breakdown and repair rates, for the robot,are 0.4 and 5, respectively;
(3) time delays considered in this example are entirelyexponential.
Now, the problem is to find fuzzy risk probabilitiesbased on fuzzy steady state probabilities and risk parametersintroduced earlier; then prioritizing them on the basis of
p6
p7
p1 p2 p3
p4 p5
t1t2
t3
t4 t5 t6 t7
Figure 7: The Petri Net model of a flexible manufacturing cell.
the presented ranking technique. The Petri Net model ofthe system is live and bounded. Therefore, it is isomorphicwith its corresponding Markov chain and the analysis can beperformed by analysis of the Markov chain.
In order to find fuzzy probabilities, we have to delineatethe reachability and the Markov chain of the correspondingstochastic Petri Net of the FM cell.
The description of places and transitions is representedin Tables 9 and 10, respectively. Transitions firing rates aredisplayed in Table 11.
In order to apply the proposed method, transitionfiring rate must be converted to fuzzy form and theircorresponding alpha cuts must be obtained. The results ofthis transformation are shown in Table 12.
Based on Figure 8, parametric steady state probabilitiesare calculated using (1),
(π0,π1,π2,π3,π4)
×
⎡⎢⎢⎢⎢⎢⎣
−λ1 λ1 0 0 00 −λ2 − λ4 λ4 λ2 00 λ5 −λ5 0 0λ3 0 0 −λ3 − λ6 λ6
0 0 0 λ7 −λ7
⎤⎥⎥⎥⎥⎥⎦= 0,
π0 + π1 + π2 + π3 + π4 = 1.
(15)
The resulting parametric steady state probabilities are
Π =
⎡⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎢⎣
π0 = λ2λ3λ5λ7
λ
π1 = λ1λ3λ5λ7
λ
π2 = λ1λ3λ4λ7
λ
π3 = λ1λ2λ5λ7
λ
π4 = λ1λ2λ5λ6
λ
⎤⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎥⎦
, (16)
10 Advances in Fuzzy Systems
m0 = (1000011)
m2 = (0001001)
m1 = (0100001)
m4 = (0000100)
m3 = (0010000)
Figure 8: Markov chain and reachability graph of the modeled system.
where λ = λ2λ3λ5λ7 + λ1λ3λ5λ7 + λ1λ3λ4λ7 + λ1λ2λ5λ7 +λ1λ2λ5λ6. After obtaining parametric steady state proba-bilities, they have to be converted to their correspondingfuzzy form. By applying fuzzified transition firing rates insteady state probabilities, fuzzy steady state probabilities arecalculated. Therefore, the alpha cut representations of fuzzysteady state probabilities are
π0=[
10α4 + 175α3 + 1100α2 + 2975α + 2940131α4 − 2964.5α3 + 25107α2 − 94312.5α + 132577.5
;
10α4 − 255α3 + 2390α2 − 9795α + 14850131α4 + 1916.5α3 + 9464α2 + 25268.9α + 22760.5
],
π1=[
10α4 + 175α3 + 1100α2 + 2975α + 2940131α4 − 2964.5α3 + 25107α2 − 94312.5α + 132577.5
;
10α4 − 255α3 + 2390α2 − 9795α + 14850131α4 + 1916.5α3 + 9464α2 + 25268.9α + 22760.5
],
π2=[
α4 + 16.5α3 + 96.5α2 + 241.5α + 220.5131α4 − 2964.5α3 + 25107α2 − 94312.5α + 132577.5
;
α4 + 24.5α3 + 219.5α2 − 857.5α + 1237.4131α4 + 1916.5α3 + 9464α2 + 25268.9α + 22760.5
],
π3=[
100α4 + 1400α3 + 7325α2 + 16975α + 14700131α4 − 2964.5α3 + 25107α2 − 94312.5α + 132577.5
;
100α4 − 2200α3 + 18125α2 − 66275α + 90750131α4 + 1916.5α3 + 9464α2 + 25268.9α + 22760.5
],
π4=[
10α4 + 150α3 + 842.5α2 + 2102.4α + 1960131α4 − 2964.5α3 + 25107α2 − 94312.5α + 132577.5
;
10α4 − 230α3 + 1982.5α2 − 7590α + 10890131α4 + 1916.5α3 + 9464α2 + 25268.9α + 22760.5
].
(17)
Since the value of fuzzy probabilities must be in interval[0, 1], we have to observe if they are out of this interval.When α = 0 all the minimum and maximum bounds of eachfuzzy number must be in [0, 1]; however it is apparent thatπ+
3 (α) exceeds 1. On the other hand, all π−i (α) are positive.
Table 10: Description of transitions in Figure 7.
Transition Description
T1 Robot taking part
T2 Machine is processing
T3 Processing is finished and robot is unloading
T4 Robot breakdown
T5 Robot in repair
T6 Machine breakdown
T7 Machine in repair
Therefore, the LP model of the problem, using (13), is givenas:
Min Z = α
121α4 + 2171.5α3 + 7074α2 + 35063.9α + 7910.5 ≥ 0
130α4 + 1941α3 + 9244.5α2 + 26126.4α + 21523 ≥ 0
31α4 + 4116.5α3 − 8661α2 + 91543.9α− 67990 ≥ 0
121α4 + 2146.5α3 + 7481.5α2 + 32858.9α + 11870.5 ≥ 0
0 ≤ α ≤ 1.(18)
Using LP software or spreadsheets like Excel, we can find theoptimal value of α as 0.779. Thus the feasible fuzzy steadystate probabilities will be as given in, Table 13.
In the next stage, the obtained fuzzy probabilities mustbe converted to generalized fuzzy numbers. According toFigure 8, we have two risky states in our model, m2 and m4,so we just deal with them. We call m2 and m4 as risky states s1
and s2, respectively. Based on our intuition, we consider riskfactors for both s1 and s2, as in Table 14.
The weights of both risky states are
S1 = 0.6× 0.8× 0.2× 0.8× 0.7 = 0.05376,
S2 = 0.8× 0.8× 0.2× 0.7× 0.7 = 0.06272.(19)
Finally, the resulting generalized fuzzy probability num-bers are (Figures 9–10):
π2 = (0.0065, 0.0097, 0.0097, 0.0144; 0.05376),
π4 = (0.0573, 0.085, 0.085, 0.0124; 0.06272).(20)
Advances in Fuzzy Systems 11
Table 11: Transition firing rates of Figure 7.
Transition Description Rate
T1 Taking part by R 45
T2 R loading 45
T3 M process finished and R is unloading 8
T4 R breakdown 0.4
T5 R repaired 5
T6 M breakdown 0.5
T7 M repaired 4
Table 12: Fuzzified transitions firing rates with their alpha cuts.
Fuzzified rates Alpha-cut representation
λ1 = (35, 45, 55) λ1 = (35 + 10α; 55− 10α)
λ2 = (35, 45, 55) λ2 = (35 + 10α; 55− 10α)
λ3 = (7, 8, 9) λ3 = (7 + α; 9− α)
λ4 = (0.3, 0.4, 0.5) λ4 = (0.3 + 0.1α; 0.5− 0.1α)
λ5 = (4, 5, 6) λ5 = (4 + α; 6− α)
λ6 = (0.4, 0.5, 0.6) λ6 = (0.4 + 0.1α; 0.6− 0.1α)
λ7 = (3, 4, 5) λ6 = (3 + α; 5− α)
Table 13: Fuzzy steady state probabilities.
πi Fuzzy number
π0 [0.082, 0.121, 0.174]
π1 [0.082, 0.121, 0.174]
π2 [0.0065, 0.0097, 0.0141]
π3 [0.453, 0.68, 1]
π4 [0.0573, 0.085, 0.124]
Table 14: The considered risk factors for s1 and s2.
Si j Risk value
S11 0.6
S12 0.8
S13 0.2
S14 0.8
S15 0.7
S21 0.8
S22 0.8
S23 0.2
S24 0.7
S25 0.7
Now, we incorporate the proposed ranking method andrank these two fuzzy probability numbers. Their rankingscore for the failure modes are (R(π2) = 0.010713), R(π4) =1.0185). Hence, s2 has priority to perform corrective actionsover s1 because its fuzzy risk probability has a higher rankingscore according to the ranking algorithm introduced inSection 4.
W
0.06272π4
0.05 0.1
X
Figure 9: Fuzzy probability number π4.
W
0.05376π2
0.01 0.02X
Figure 10: Fuzzy probability number π2.
7. Conclusions
In this study we proposed a hybrid approach utilizing iso-morphism between stochastic Petri Nets and Markov chains,and also a novel fuzzy ranking method. This approach isgeneral and many different risk factors in the systems, whichshould be considered, are studied, for the first time. Anothercontribution of this paper is the application of fuzzy logicin determining steady state probabilities of systems andincorporating them in risk analysis.
The proposed methodology can be useful in most reallife applications such as industrial systems; however, animportant issue regarding this methodology is complexity.Although this method is quite efficient in dealing withsmall or medium sized Petri Nets but it would be hard toimplement it on more complex nets. Therefore, presentingnew approaches to improve efficiency of the prosed methodin order to handle large scale problems can be an appropriatetopic for future studies.
References
[1] M. Braglia, M. Frosolini, and R. Montanari, “Fuzzy criticallyassessment for failure modes and effect analysis,” InternationalJournal of Quality & Reliability Management, vol. 20, no. 4, pp.503–524, 2003.
[2] D. H. Stamatis, Failure Mode and Effect Analysis- FMEA FromTheory to Execution, ASQC Quality Press, Milwaukee, Wis,USA, 1995.
12 Advances in Fuzzy Systems
[3] K. Xu, L. C. Tang, M. Xie, S. L. Ho, and M. L. Zhu,“Fuzzy assessment of FMEA for engine systems,” ReliabilityEngineering and System Safety, vol. 75, no. 1, pp. 17–29, 2002.
[4] C. L. Chang, C. C. Wei, and Y. H. Lee, “Failure modeand effects analysis using fuzzy method and grey theory,”Kybernetes, vol. 28, no. 9, pp. 1072–1080, 1999.
[5] B. H. Lee, “Using Bayes belief networks in industrial FMEAmodeling and analysis,” in Proceedings of the InternationalSymposium on Product Quality and Integrity, pp. 7–15, January2001.
[6] M. Bevilacqua, “Monte Carlo simulation approach for amodified FMECA in a power plant,” Quality and ReliabilityEngineering International, vol. 17, no. 9, pp. 3–13, 2000.
[7] A. D. Dominguez-Garcia, J. G. Kassakian, and J. E. Schindall,“Reliability evaluation of the power supply of an electricalpower net for safety-relevant applications,” Reliability Engi-neering and System Safety, vol. 91, no. 5, pp. 505–514, 2006.
[8] J. B. Bowles and C. E. Pelaez, “Application of fuzzy logic toreliability engineering,” Proceedings of the IEEE, vol. 83, no. 3,pp. 435–449, 1995.
[9] K. J. Schmucker, Fuzzy Sets, Natural Language Computations,and Risk Analysis, Computer Science Press, Rockville, Md,USA, 1984.
[10] S. M. Chen and C. H. Wang, “Fuzzy risk analysis based onranking fuzzy numbers using α-cuts, belief features and signal/noise ratios,” Expert Systems with Applications, vol. 36, no. 3,pp. 5576–5581, 2009.
[11] S. M. Chen and J. H. Chen, “Fuzzy risk analysis based onranking generalized fuzzy numbers with different heights anddifferent spreads,” Expert Systems with Applications, vol. 36, no.3, pp. 6833–6842, 2009.
[12] S. M. Chen and J. H. Chen, “Fuzzy risk analysis based onsimilarity measures between interval-valued fuzzy numbersand interval-valued fuzzy number arithmetic operators,”Expert Systems with Applications, vol. 36, no. 3, pp. 6309–6317,2009.
[13] S. H. Wei and S. M. Chen, “A new approach for fuzzy riskanalysis based on similarity measures of generalized fuzzynumbers,” Expert Systems with Applications, vol. 36, no. 1, pp.589–598, 2009.
[14] C. H. Cheng, “A new approach for ranking fuzzy numbers bydistance method,” Fuzzy Sets and Systems, vol. 95, no. 3, pp.307–317, 1998.
[15] T. C. Chu and C. T. Tsao, “Ranking fuzzy numbers with an areabetween the centroid point and original point,” Computers andMathematics with Applications, vol. 43, no. 1-2, pp. 111–117,2002.
[16] S. Murakami, S. Maeda, and S. Imamura, “Fuzzy decisionanalysis on the development of centralized regional energycontrol system,” in Proceedings of IFAC Symposium on FuzzyInformatics, Knowledge Representation and Decision Making,pp. 363–368, 1983.
[17] R. R. Yager, “Ranking fuzzy subsets over the unit interval,” inProceedings of 17th IEEE International Conference on Decisionand Control, pp. 1436–1437, New York, NY, USA.
[18] S. J. Chen and S. M. Chen, “Fuzzy risk analysis basedon theranking of generalized trapezoidal fuzzy numbers,” AppliedIntelligence, vol. 26, no. 1, pp. 1–11, 2007.
[19] L. W. Lee and S. M. Chen, “Fuzzy risk analysis based onfuzzy numbers with different shapes and different deviations,”Expert Systems with Applications, vol. 34, no. 4, pp. 2763–2771,2008.
[20] S. M. Chen and K. Sanguansat, “Analyzing fuzzy risk basedon a new fuzzy ranking method between generalized fuzzy
numbers,” Expert Systems with Applications, vol. 38, no. 3, pp.2163–2171, 2011.
[21] T. Murata, “Petri nets: properties, analysis and applications,”Proceedings of the IEEE, vol. 77, no. 4, pp. 541–580, 1989.
[22] D. Vernez, D. Buchs, and G. Pierrehumbert, “Perspectives inthe use of coloured Petri nets for risk analysis and accidentmodelling,” Safety Science, vol. 41, no. 5, pp. 445–463, 2003.
[23] M. Zhou and K. Venkatesh, Modeling, Simulation And Controlof Flexible Manufacturing Systems, A Petri Net Approach, WorldScientific, Singapore, 1999.
[24] M. K. Molly, “Performance analysis using stochastic PetriNets,” IEEE Transactions on Computers, vol. 31, no. 9, pp. 913–917, 1982.
[25] F. Bause and P. S. Kritzinger, Stochastic Petri Nets: An Intro-duction to The Theory, Vieweg, Braunschweig, Germany, 2ndedition, 2002.
[26] H. J. Zimmermann, Fuzzy Set Theory and Its Applications,Kluwer Academic, New York, NY, USA, 3rd edition, 1996.
[27] A. Doostparast Torshizi, Integrated risk analysis of systemsbased on Petri Nets analysis [Dissertation], Isfahan Universityof Technology, 2011.
[28] R. Viertl and D. Hareter, “Fuzzy information and stochastics,”Iranian Journal of Fuzzy Systems, vol. 1, no. 1, pp. 39–52, 2004.
[29] F. Tuysuz and C. Kahraman, “Modeling a flexible manufac-turing cell using stochastic Petri nets with fuzzy parameters,”Expert Systems with Applications, vol. 37, no. 5, pp. 3910–3920,2010.
[30] M. C. Zhou and K. Venkatesh, Modeling, Simulation and Con-trol of Flexible Manufacturing Systems: A Petri Net Approach,World Scientific, Singapore, 1999.
Submit your manuscripts athttp://www.hindawi.com
Computer Games Technology
International Journal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Distributed Sensor Networks
International Journal of
Advances in
FuzzySystems
Hindawi Publishing Corporationhttp://www.hindawi.com
Volume 2014
International Journal of
ReconfigurableComputing
Hindawi Publishing Corporation http://www.hindawi.com Volume 2014
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Applied Computational Intelligence and Soft Computing
Advances in
Artificial Intelligence
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Advances inSoftware EngineeringHindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Electrical and Computer Engineering
Journal of
Journal of
Computer Networks and Communications
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
Advances in
Multimedia
International Journal of
Biomedical Imaging
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
ArtificialNeural Systems
Advances in
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
RoboticsJournal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Computational Intelligence and Neuroscience
Industrial EngineeringJournal of
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Modelling & Simulation in EngineeringHindawi Publishing Corporation http://www.hindawi.com Volume 2014
The Scientific World JournalHindawi Publishing Corporation http://www.hindawi.com Volume 2014
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014
Human-ComputerInteraction
Advances in
Computer EngineeringAdvances in
Hindawi Publishing Corporationhttp://www.hindawi.com Volume 2014