Time‐State‐Dependent Reliability Analysis of Wheel‐Rail Contact Parameter Degradation for High‐Speed Trains

Abstract

Wheel‐rail contact parameters are essential to high‐speed‐train stability, but their stochastic degradation complicates reliability analysis. A time‐state‐dependent reliability method is developed for complex thresholds defined implicitly or probabilistically. Different from classical first‐passage‐based time‐dependent reliability analysis, the proposed framework replaces the analytical first‐passage distribution with reliability evaluated at fixed inspection intervals. The method screens features with a stacking feature deep forest, and constructs degradation trajectories for equivalent conicity, tangent gamma, and wheel diameter difference. Model parameters were estimated via Bayesian inference using automatic differentiation and variational inference. The Euler–Maruyama method and Monte Carlo simulation were then used to generate mileage‐based degradation scenarios. The time‐state‐dependent reliability curve was then derived. Numerical simulations illustrate that equivalent conicity increases over time, whereas the contact angle exhibits periodic behaviour. Field measurements further revealed a nonlinear decline in WRCP reliability, with a mean failure time of approximately 300,000 km. The method is applicable to systems with complex thresholds and provides practical guidance for wheelset maintenance scheduling.