Accurate probabilistic risk assessment of nuclear power plants through parallel computing

Abstract

Nowadays, the “small event tree / large fault tree” is the most used method for the probabilistic risk assessment (PRA) of nuclear power plants. A nuclear power plant is a complex system and it is composed by many large subsystems with different purposes, including safety. The fault tree of any subsystem is large with up to thousands basic events and thousands gates. For the PRA different event trees are considered. Each event tree represents the paths from a possible initiating event to a final outcome. In addition, common cause failure analysis, dynamic fault tree, and dynamic event tree analysis are also used. While the theory of all these analysis tools is simple when there are used together the solution of the problem is a challenge and its numerical simulation takes long time. In this study, the state of the art in parallel computing is used to achieve reasonable computational time and great accuracy. The numerical simulation of each analysis was reshaped to work on a general purpose graphic processing unit (GPGPU), which are single instruction multiple data (SIMD) processors. The results are parallel algorithms that outperform the current state of the art and solve large fault tree problems within seconds rather than minutes or hours.