A decoupling method for wheel-rail contact forces of heavy-duty trains based on big data and neural networks

To accurately identify and predict wheel-rail contact forces during the operation of heavy-duty trains, this paper proposes a novel method of decoupling and predicting these forces based on big data related to non-contact strain at the wheel periphery and a neural network. Initially, a refined finite element numerical simulation model was developed to replicate the wheelset-rail contact of heavy-duty trains. The subsequent analysis of strain sensitivity at the wheel periphery determined the optimal strain collection radius and the layout of collection points on the train wheels. The dynamic response of the wheel-rail contact area under various operating conditions of trains was investigated by analyzing extensive strain data generated at the strain collection points across multiple numerical simulation models, along with the vertical and lateral forces at the contact spots. The results revealed the variation patterns of response in key aspects such as strain and contact forces in the contact spot area when trains operating at a constant speed are accelerated. Following the establishment of a comprehensive data set of strain and contact forces, a neural network model correlating wheel-rail contact forces with non-contact strain at the wheel periphery was established, through a training process to address the relationship between contact forces and non-contact position strain. This model enables the real-time decoupling and accurate prediction of contact forces at the contact spot for trains. This neural network model exhibits superiority in both computational efficiency and prediction accuracy, proving effective in guiding the decoupling identification of wheel-rail contact forces for heavy-duty trains and supporting further engineering applications, such as solving the train derailment coefficient.