Solving Large Fault Trees with Importance Sampling and Tree Search

Fault tree analysis has been widely used in various areas for reliability analysis. A major task in fault tree analysis is to quantify the probability of the top event in the fault tree. Various algorithms based on minimal cut sets, binary decision diagrams, and Monte Carlo simulation have been developed. For fault trees that are large, have complex logic structures, and for which the top event probabilities are small, these algorithms are inefficient. This paper proposes a new algorithm to address this challenge. The new algorithm is based on importance sampling and Monte Carlo tree search. Compared with existing algorithms based on Monte Carlo simulation, the new algorithm can generate sample combinations of event states that cause the top event more efficiently, and therefore improves the computational efficiency of top event probability quantification for large, complex fault trees for which the top event probabilities are small. A preliminary case study shows that the new algorithm can be about ten times faster than an algorithm based on Monte Carlo simulation. This approach becomes very important if the top event has major impacts on an organization, such as loss of life and/or the organization's reputation, though the probability of the top event is low.