Robust assessment of railway vehicle safety risks in operation using a proposed data-driven wheel profile generation approach: Design of computer experiments and surrogate models

Worldwide objectives for railway vehicles are increased capacity, faster travels and higher levels of safety. In the vehicle-track complex system, assessing and controlling the interactions between the wheels and the rail track is crucial to these goals. Wheel profiles are specifically designed to steer the vehicle and avoid derailment. Maintenance standards and train operating companies establish safe envelopes for wheel profile geometric parameters. A design of experiments is conducted to model relationships between allowable wheel parameters and expected vehicle safety risks, which is supported by condition monitoring data from operation. Such a robust assessment is missing in the literature. The applied methods consist of: (i) selection of predictors and pre-processing, based on literature, standards and a purely data-driven approach to generate wheel profiles; (ii) space-filling design, using Latin hypercube sampling; (iii) obtaining vehicle responses and post-processing, using a multibody dynamics commercial software and according to standards; (iv) surrogate modelling, using Gaussian processes and linear models; (v) sensitivity analysis, through Sobol indices; (vi) safety assessment, analysing response surfaces. Wheels with large flange height and thickness result in higher flange climb derailment risks. The proposed approach allows quantifying this risk as a function of profile parameters and mitigate it through maintenance actions.