Download as pdf
Download as pdf
You are on page 1of 6
Annual paar farseeaae RELIABILITY and MAINTAINABILITY Symposium 1993 PROCEEDINGS Atlanta, Georgia USA 1993 January (25)26-28 A Monte Carlo Simulation to Determine Minimal Cut Sets and System Reliability John Y. Lin + University of Houston * Houston Charles E, Donaghey + University of Houston + Houston, Key Words: Minimal cut set, Minimal path set, System reliability, Monte Carlo simulation. SUMMARY & CONCLUSIONS ‘This paper describes an approach for determining the ‘minimal cut sets and the system reliability using Monte Carlo simulation from the reliability block diagram and the life distribution of each component, First, the minimal path sets are determined by tracing through the system from the input ‘components to the output components of the reliability diagram in a random manner. Using Monte Carlo simulation ‘and the minimal path sets as the criteria for system failure, the system fails when all minimal path sets are broken. ‘The ‘components failed prior to the system failure constitute a cut set. The frequencies of the minimal cut sets are tallied during, the simulation runs to show the distribution of the frequencies Of the minimal cut sets. In addition, the system mean time to failure is calculated, the system failtime distribution is tabulated, and the cumulative failure rates are determined. 1 INTRODUCTION “The evaluation of system reliability usually is based on the physical structure in the form of a reliability block diagram to describe a network of components. There are many available methods to determine system reliability by analyzing the reliability block diagram of the system, ‘Two of the methods are the cut set method and path set method. A cut set is the set of system components which, when all fail, cause system failure, The minimum subset of these cut sets are known as ‘minimal cut sets in which the system fails if and only if all components of one of the minimal cut sets fail. The path set ‘method is essentially the complement of the cut set method. ‘Any component failure in each of the minimal path sets causes that path set to fail. For the system to fail ll ofthe minimal path sets must fail. References [1] and (3] required the user to provide the path sets, and it is important to provide all the path sets or the resulting cut sets will be incorrect. This paper first uses the Monte Carlo method to determine the minimal path sets, Uien uses the minimal path seis to simulate system failures, again using a Monte Carlo method, the minimal cut sets and system reliability at various times are determined. 2. DETERMINING MINIMAL CUT SETS AND SYSTEM RELIABILITY BY MINCUT MINCUT uses a Monte Carlo simulation development program modified from the program SIMPAK by Donaghey {2} to determine minimal cut sets and system reliability of a network of components, MINCUT is written in C to be run on IBM PC or other compatible machines and should run on other systems since standard C routines are used. After the ‘minimal path sets have been determined, the individual failure times of each of the components are scheduled, Random values from their life distributions are used. The most imminent event is assumed to have occurred and the simulated time is advanced (0 the point of occurrence of the event. This cycle is repeated until the system fails as determined by the minimal path sets. 2.1 Data Requirements For each component, the user is asked to provide the ‘component life distibution, its parameter values, and the successor components of the reliability block diagram. A successor of -1 indicates that the component is an output ‘component, Five types of distributions can be used in MINCUT and the user specifies the particular distribution by the letter (either in upper or lower case) N for normal distribution; E for exponential distribution; R for erlang dis- tribution; L for lognormal distribution; U for uniform distribution and W for weibul distribution. The program allows either interactive or file data input. EXAMPLE. Figure 1 A Simple Network 0199-144X90/63.00 © 1993 IEEE 246 1993 PROCEEDINGS Annual RELIABILITY AND MAINTAINABILITY Symposium ‘The file input format of the system in Figure 1 is shown below. 12 E23 n101 1.53 e43 -1 Note: The first line indicates that the input components are 1 and 2 © The second line indicates that component 1 has an exponential life distribution with a mean of 2 and a successor component of 3. ‘The third line indicates that component 2 has a normal life distribution with a mean of 10.1 and a standard deviation of 1.5 (notice the parameter values can be either integers or real numbers) and a successor component of 3 The last line indicates that component 3 has an exponential life distribution with a mean of 4.3, and a successor of -1 indicates that component 3 is an output component. For interactive input, the user enters component life distributions, their parameters, and the suecessor components atthe prompt. 2.2 Determining Minimal Path Sets ‘The minimal path sets of a system can be determined by ‘Monte Carlo simulation once the reliability block diagram is given, Here the reliability block diagram is presented by the input components and the associated successors for each component. A random path is traced from one of the input components, through the successors specified by each component, until one of the output components is reached. For multipie suecessors, one is chosen at random. If the randomly selected successor component is one of the ‘componens inthe current path set, a loop has formed and the output component will never be reached. As the loop formation is recognized, anosher successor is randomly selected and checked again for loop formation until no loop has formed. ‘A successor of -1 indicates an output component and a ‘completed path set. Since this random process does not guarantee unique path sets, each path set is tested with known path sets for redundancy using one of the simplification thcorems of Boolean algebra (AB + ABC = AB, ABC + AC AC). If an appropriate match is found, the path set that ‘contains fewer number of components is retained and the other path set is redundant and is ignored. kis crucial to have the complete set of minimal path sets to generate the correct system reliability and minimal cut sets Since the minimal path sets are used as the criteria for system failure. To ensure the completeness of minimal path sets, a counter is used for the number of iterations elapsed without large. On the other hand, when randomly exploring the paths, the program need not know the explored and the unexplored components, the number of variables to keep track of is thus fewer. 2.3 Determining Minimal Cut sets ‘Once the minimal path sets are known, Monte Carlo simulation is used again to simulate the system failure time. ‘AL each simulation run, after each component failure, the remaining working components are checked against the ‘minimal path sets {0 see if the system is functioning. If the system fails, the failed components constitute a cut set. Minimal cut sets are determined by employing the same reduction scheme used to determine the minimal path sets. When the frequencies of the minimal cut sets are tallied, it shows that the distribution of the minimal cut sets is not ‘uniform, some cut sets occur often while others occur rarely. It is therefore advisable to conduct many runs in order to generate the complete minimal cut sets. 2. Determining System Life Distribution and Reliabitity In tabulating the system failures, the class width (the difference between the lowest value of one class and the highest value of the next class) is determined first by sampling the first 200 system failtimes (if the number of iterations is less than 200, all failure times are sampled,) and the range of the sampled failtimes divided by 10 (10 is used arbitrarily) is the class ‘width, Two additional classes are added: > the maximum value of the sampled failtimes, and << the minimum value of the sampled failtimes for the nonsampled failure times that fall outside the range determined by the sampled failure times. ‘The average of all the failure times (system MTF) is also computed. On the column to the right of the failure distribution ratios, the system reliability at the various time intervals is displayed. ‘The cumulative failtime distribution at time t is equivalent to the system unreliability at time t, thus this value subtracted from 1 gives the system reliability at time t. The user can interpolate the system reliability at times within the intervals. ‘AL this point, the user is asked whether to continue the simulation run, and if the user wishes to continue, the user is asked again to provide the number of iterations. Since the values of all relevant variables will carry over to the additional iterations, the user can run additional 38 until the user is satisfied with the results, EXAMPLE 1 ‘The single-bridge network shown below has the following component failure distributions. any changes in the contents or the number of the minimal path Comp. Distr Parameters Successors sets. Any changes in minimal path sets resct the counter 10 Normal Meaasi0, 81 zero. When the counter reaches 500 (arbitrarily set in the pro- eon siaid a gram), the iteration process to find the minimal path sets is 2 Exponential Mean:15 35 stopped. This exhaustive exploration scheme is obviously less efficient than systematically exploring all possible paths. 3 Legnormel ea ele 5a However, in computer applications, the number of variables 4 Exponential Mean:17 7 needed to keep track of the status of the components, even for ‘a moderately large numbers of components, becomes very 5 Erlang Mean:12, K:3 a 1993 PROCEEDINGS Annual RELIABILITY AND MAINTAINABILITY Symposium 247 aps Pigwe 2 Te Single Bdge Newek of Example 1 hollow abe chow hs oapuc of MINCUT plying tense. ‘fe dsutton apd syste elabitym varows ues. TABLE 1 Ouputot MINCUT Pattie 4, Saleh Raiitiy Lots taine 235 098m 099 2S claims e595 sm 0089 595 clime c8k ROT 84 ehitime <1085 178 Oae 108s fbi <3 BAB 02M 12° fide <1578 se Os 1576 efinine <182t Suse oe 1221 etanine <2065 28% 0083 2086 =2556 093% © 0 The reliability curve is plotted in Figure 3. ‘The system reliability at time Cis the reliability reading at time t. From Figure 3, the system reliability at time unit 10 is about 0.58. 248 Reliability 0 10 20 Time Figure 3. The System Reliability Curve Obtained in Example 1 EXAMPLE 2 ‘This example is taken from Bellmore and Jensen (1} in order to verify that the cut sets obtained by MINCUT are correct and to show that the results approach "steady state” values with an increase in number of iterations. 6 7 Figure 4 Reliability Network of Example 2 The minimal cut sets for the network shown in Figure 4 at 1000 iterations are shown below: cut sets: (1000 iterations) 1, 6 : 12.661% 1,5: 13.591% 4,8: 14.02% 7: 12.089% 2,7, 10: 4.864% 25,9: 4.649% 3,8, 10 : 5.079% 3,5,9 5.794% 3,7, 10 4.22% 2,8, 10: 5.579% 2,6,9: 5.794% 3,6,9 : 5.293% 4,6,9, 10 : 1.574% 1,7, 9, 10 : 1.645% 1, 8,9, 10: 1.288% ‘The value of MTTF is 2.5314 4, 5,9, 10: 1.86% The resulting 16 minimal cut sets agree with the results in Bellmore and Jensen (2]. Furthermore, MINCUT showed 1993 PROCEEDINGS Annual RELIABILITY AND MAINTAINABILITY Symposium varied frequencies of the cut sets, indicating that some cut sets are more prone to cause system failures than others. 25 Limitations It is important to note that when determining system reliability, each minimal cut set is independent of others. ‘There are no partial system failures allowed, either the system is totally functional in which all components of at least one ‘minimal path set are functioning, or the system fails as all ‘minimal path sets have at least one failure. No repair of the failed component is allowed until the system has failed. REFERENCES R.N. Allan, R, N. Billinton, Reliability Evaluation of Engineering Systems: Concepts and Techniques, London: Pitman Advanced Publishing Program (1983). 2. C. E. Donaghey, Digital Simulation Course Notes, University of Houston Industrial Engineering Department, 1989 Jan, 3. D. Magee, A. Refsum. “RESIN, A Desktop-Computer Program for Finding Cut-Sets", [EEE Trans. on Reliability, Vol. R-30, 1981 Dec., pp. 407-410. BIOGRAPHIES Jobo ¥. Lin 5419 Bent Bough ‘Houston, Texas 77088 USA, John Y. Lin was bom in 1965 and has seceived an MS in Industrial Enginesring from the University of Houston in May 1992. Charles B. Donaghey. Ph.D. Industrial Engineering Dept. University of Houston Houston, Texas 77204-4812 USA Cares E, Donaghy is Professor and Chairman of Industrial Engineeing, Univenity of Houston, He har over 25 yours experince in teaching and reseuich in the tea of simulation and probability theory. 1993 PROCEEDINGS Annual RELIABILITY AND MAINTAINABILITY Symposium 249

You might also like