Dynamic performance reliability analysis of mixed passenger and freight railway turnouts based on neural networks

Mixed passenger and freight railway turnouts must meet competing demands: high-speed stability for passenger trains and tolerance of heavy axle loads for freight traffic. However, their coupled dynamic behavior and reliability remain insufficiently studied. This work introduces an efficient reliability analysis framework that combines back-propagation (BP) neural networks with Monte Carlo simulation (MCS), enhanced through latinized partially stratified sampling (LPSS). The framework enables accurate assessment of low-probability failure events while reducing the computational cost of high-dimensional implicit limit state functions. The findings reveal distinct passenger and freight response patterns within turnout zones and show that the proposed BP–MCS–LPSS method significantly outperforms conventional techniques.