Parallel fault tree analysis for accurate reliability of complex systems

Abstract

Fault tree analysis is one of the methods for the probabilistic risk assessment of components and subsystems of nuclear power plants. The algorithms that solve a fault tree have been until now serial. Instead, this study presents new algorithms that handle and solve a fault tree by taking advantage of the new state of the art in parallel computing: general purpose graphic processor unit (GPGPU). The subsystems of nuclear power plants are the target of this study. However, the method can be used on many others, complex, engineering systems. The different, developed, parallel algorithms are: one builder, which assembles the topology matrix of the fault tree and leads the computation of the three, developed, new solvers. A bottom-up solver, a cut sets solver, and a Monte Carlo simulation solver. The probability of the top event, and the probabilities of each cut sets are computed. The results shows that, given the same investment, a GPU can handle larger fault trees than a CPU implementation. The developed solvers are the foundation of the next generation parallel algorithms for the tree-based analysis of complex systems.