Wheel Defect Detection Using a Hybrid Deep Learning Approach

Defective wheels pose a significant challenge in railway transportation, impacting operational performance and safety. Excessive traction braking forces give rise to deviations from the intended conical tread shape, resulting amplified vibrations noise. Moreover, these contribute accelerated damage of track components. Detecting wheel defects at an early stage is crucial ensure safe comfortable operation, as well minimize maintenance costs. However, presence various vibrations, such those induced by track, motors, other rolling stock subsystems, poses for onboard detection techniques. These create difficulties accurately identifying real-time during activities, often false alarms. This research paper aims address this issue using hybrid deep learning-based approach accurate types accelerometer data. The proposed enhance defect accuracy while considering techniques’ cost-effectiveness efficiency. A realistic simulation model wheelset developed generate comprehensive dataset. To vibration data scenarios, simulated 20 s under different conditions, including one non-faulty scenario six faulty scenarios. simulations are conducted speeds conditions capture wide range operating conditions. Within each iteration, total 200,000 points generated, providing dataset analysis evaluation. generated then utilized train evaluate learning model, employing multi-layer perceptron (MLP) feature extractor multiple machine models (support vector machine, random forest, decision tree, k-nearest neighbors) comparison. results demonstrate that MLP-RF (multi-layer with forest) achieved 99%, MLP-DT tree) 98%. high values indicate effectiveness classifying predicting outcomes. contributions work include development evaluation sensor layout effectiveness, application techniques improved flat detections.