A machine-learning-based approach for railway excitation force prediction

Purpose Vibrations induced by external loads play a critical role in the performance and safety of high-speed train bogies. Accurate knowledge of the dynamic forces acting on bogie frames is essential for predicting structural responses, enhancing numerical modelling and planning maintenance more effectively. This study aims to develop and evaluate a comprehensive model-based framework for predicting excitation forces on railway bogies, addressing the challenges posed by forces that are difficult or impossible to measure directly. Design/methodology/approach The proposed framework integrates multi-body dynamics (MBD) simulations, structural finite element (FE) modelling and machine learning (ML) to estimate the forces acting on the bogie frame of a high-speed train. Firstly, an MBD model of a Chinese high-speed train was established and validated against in-service measurement data, from which realistic time-domain loads acting on the bogie frame can be obtained. Separately, modal dynamic simulations of the bogie frame's FE model were performed with stochastic loading to extract corresponding accelerations over a broad range of dynamic behaviour. These were employed to train an ML model to learn the inverse mapping from structural response to applied forces. For validation, the MBD-derived forces were applied to the FE model to obtain corresponding accelerations, which were then used to assess the ML model’s ability to reconstruct the original forces. Findings The approach can successfully predict nonlinear excitation forces acting on the bogie frame. Based on modal system responses to randomized force inputs, the entire parameter space can be represented, and the trained ML model demonstrates a strong capability to estimate dynamic loads from validated MBD simulations. Through appropriate training, the method exhibits robustness against noise and sensor placement and opens new opportunities for improving the analysis of track–vehicle interaction and the dynamic modelling of bogies. Research limitations/implications The approach depends on the accuracy of the validated MBD and FE models, meaning modelling assumptions and simplifications may introduce errors in the predicted forces. High-fidelity in-service measurements are required for model validation but are not always available. Purely simulation-based models enable the prediction of forces and load distributions at the bogie, but the results are strongly model-dependent. Even if the models are validated against reference data, they only reflect an idealized operating condition, and uncertainties in measurement parameters, model parameters, damping behaviour or contact models can significantly affect the accuracy of force predictions. Originality/value This research introduces a novel, integrated framework for indirect bogie force estimation that enhances both modelling accuracy and practical diagnostic capability in railway engineering. By integrating numerical simulations, in-service measurements and ML, the study advances current methodologies for analysing high-speed railway vehicles. The approach offers valuable potential for refining vehicle models, guiding maintenance strategies and informing future research on data-driven structural force prediction.