System reliability evaluation using Monte Carlo & support vector machine

Reliability evaluation of real engineering systems is often performed using simulation tools. In general, to determine the state of the system (operating or failed) as a function of its components, it is necessary to evaluate a function that is called system function or structure function (SF). In a Monte Carlo simulation, system reliability is evaluated by generating at random several systems states and evaluating the SF. Since a large number of SF evaluations are required, it could be necessary to substitute those evaluations with a fast, approximated algorithm. Several approaches have been used to address the definition of these approximated algorithms. This paper deals with the feasibility of using a support vector machine (SVM) to build empirical models for use in reliability evaluation. The approach takes advantage of the speed of SVM in the numerous model calculations typically required to perform a Monte Carlo reliability evaluation. The main idea is to develop an estimation algorithm, by training a model on a restricted data set, and replace SF evaluation by a simpler calculation, which provides reasonably accurate model outputs. In the example presented, related to an electric power system, the SVM built from a very small fraction of the total state space, produces very close reliability estimation.