On deriving Bayesian Network formulation incorporating enhanced split fraction for a novel seismic probabilistic safety assessment of nuclear systems

The seismic correlation of component responses and capacities significantly affects the risk of nuclear power plants (NPPs). Despite several studies addressing this seismic correlation, challenges with accuracy and efficiency in calculating system failure probability remain. Therefore, this paper introduces an enhanced method for calculating system failure probability by integrating the split fraction concept with the Bayesian Network (BN). It extends the existing split fraction concept and formulates an equation incorporating the nonlinear relationship between the split fraction and the component response correlation coefficients. Because the improved split fraction concept is difficult to implement using the existing fault tree method, it is developed based on the BN. Consequently, the proposed method is tested on three examples, including actual NPP. The results show that the method produces system failure probability with greater accuracy than existing methods and reduces the computational cost compared to the sampling-based method having high accuracy. Finally, relying solely on the linear relationship between the split fraction and component response correlation coefficients leads to errors ranging from approximately 3.7% to 39%. However, the nonlinear relationship has errors ranging from approximately 0.77% to 3%, demonstrating that its accuracy is improved over the linear one.