Application of Subset Simulation Methods to Reliability Benchmark Problems

Abstract

This paper presents the reliability analysis of three benchmark problems using three variants of Subset Simulation. The original version of Subset Simulation, SubSim/MCMC, employs a Markov chain Monte Carlo (MCMC) method to simulate samples conditional on intermediate failure events; it is a general method that is applicable to all the benchmark problems. A later version of Subset Simulation, Sub- Sim/Splitting, is applicable to first-passage problems involving deterministic causal dynamical systems; it uses splitting of excitation time histories rather than MCMC to generate the conditional samples. The latest version, SubSim/Hybrid, combines the advantages of MCMC and splitting and is also applicable to firstpassage problems. Results show that all three Subset Simulation methods are effective in high-dimensional problems and that some computational efficiency can be gained by adopting the splitting and hybrid strategies when calculating the reliability for the first-passage benchmark problems.