Wheel polygonization is an important failure mode for high-speed trains and causes huge maintenance costs, however, the studies on its reliability and risk are rare. First, failure effects analysis via dynamical simulations and tests indicates that high-order polygonization induces large wheel-rail forces and vehicle vibrations, which is quite detrimental to reliability and safety. Then, correlation analysis demonstrates that wheel polygonization is affected by season, wheel diameter, vehicle type and historical incidence rate. Next, a Bayesian network topology is designed based on related factors in sequential wheel operation process, and a risk assessment model based on an array of Bayesian networks is developed to produce the probability distribution of wheel polygonization over different severities. Further, the model is trained through a two-step scheme based on historical measurement data, including partially missing data. Finally, the proposed model is validated to effectively assess polygonization risks and detect high-risk wheels. Its application to risk-based maintenance can support the decision-making of wheel reprofiling, reduce failure impacts on reliability, and save maintenance costs.

No abstract

No abstract