Railway Track Fault Classification for Predictive Maintenance Using Deep Learning Algorithms Based on a 2D Train‐Track‐Bridge Interaction Model

ABSTRACT In the era of smart infrastructure, ensuring the safety and reliability of railway systems through advanced fault detection has become critical. This study proposes a deep learning–based framework integrated with a 2D train–track–bridge interaction simulation model for automated classification of railway track superstructure faults. Vertical acceleration of the train vehicle and the bogie is computed dynamically by MATLAB‐based simulations, and the results are evaluated by continuous wavelet transform (CWT) and short time Fourier transform (STFT) to extract unique time‐frequency signatures corresponding to the fault. Rail corrugation, ballast discontinuity, and a healthy condition are three class problems with two fault types. The Class 4 German spectral irregularity track function is used to mimic the real track behavior; 51,000 labeled samples are created, which include 33,000 labeled as healthy, 6000 labeled under corrugation, and 12,000 labeled as ballast. All these signals are utilized to train and assess the performance of deep learning models, including LSTM, 1D‐CNN, GRU, and MLP. Of all models, the best performance was achieved by the LSTM with an accuracy of 98% and an F1 score of 97%, which cross‐checked the precision and recall on all classes. The results demonstrate the efficacy of integrating numerical simulations and deep learning for automated, data‐centric fault diagnosis in railway systems, which hold great promise for advanced predictive maintenance and improved safety during operations.