Reliability Evaluation Of Complex Systems Based On Monte Carlo Simulation

The reliability evaluation of complex systems, due to their complex structures, alarge number of components, the interdependence between components etc, is adifficult and focused problem. The main purpose of complex systems reliabilityanalysis is to evaluate the systems reliability using simple and effective methods andthen find optimal solutions of improving the systems reliability. This paper proposessome methods to evaluate the reliability of two-state or multi-state complex systemswhose corresponding components are independent or not.This paper proposes a CA-SVM based method to evaluate the reliability oftwo-state complex systems whose corresponding components are independent. Theproposed method uses the trained support vector machine (SVM) to judge theconnectivity of the corresponding network of each Monte Carlo simulation (MCS)iteration, largely improving the computational efficiency compared with thetraditional method using cellular automata (CA) directly. When training the SVM, CAis selected to establish the training data points instead of minimal path/cut based orDFS/BFS based procedures because of its high efficiency. Before the training, theSVM parameters should be set first. The proposed method uses particle swarmoptimization (PSO), due to its faster search speed and better overall, to set theseparameters instead of the grid search algorithm. The given example illustrates that theproposed method is exact and efficient.Common cause failures (CCFs) exist in the real-life engineering systems widely.The existence of CCFs makes the assumption that each component is independentinvalid, increases the joint failure probability of systems, and decreases the systemsreliability. This paper only considers the influence of common cause failures, whichare originated from the components with propagated failures having selective effect,on the systems reliability.When evaluating the reliability of two-state complex systems in consideration ofcomponents with propagated failures having selective effect, the traditional fault treeor generate function based methods can lead to the "combinatorial explosion" problemand time-consuming computation. To overcome the drawbacks, this paper proposes aMCS based method. The proposed method uses CA, which has the advantage of parallel computing, to judge the connectivity of the corresponding network of thesystem, so it can evaluate the complex systems reliability in consideration of commoncause failures and imperfect protections efficiently. This paper also extends the aboveissue to multi-state complex systems and proposes a method to evaluate themulti-state complex systems reliability in consideration of common cause failures andimperfect protections. The proposed method combines MCS and the proposedalgorithm to find multi-state minimal path vectors. Compared with the traditionalmethods, the proposed method is out of the limitations that systems should be directlydecomposed into series and parallel structures and the quantities of components withpropagated failures having selective effect or imperfect protection groups should notbe too large.This paper also considers the influence of maintainability on the reliability oftwo-state systems with common cause failures and proposes MCS based algorithms tocompute the systems reliability, availability and maintenance characteristics.Compared with the traditional Markov chain based methods, the proposed method isout of the limitations that the life and maintenance time distributions of componentsshould be exponential distributions and won’t lead to "combinatorial explosion"brought about by the increasing components states, so the proposed method hasbroader applications.