Wheel flat detection and severity classification using deep learning techniques

Wheel flats are one of the most common types defect found in railway systems. can result decreasing passenger comfort and noise if they slight, or serious incidents such as derailment severe. With increasing demand for transport, speed weight rolling stock tend to increase, which results relatively rapid deterioration. The occurrence wheel is also affected by this demand. To perform preventative maintenance flats, keep wheelsets a proper condition minimise costs, ability detect classify required. This study aims apply deep learning techniques flat severity. used neural network (DNN), convolutional (CNN) recurrent (RNN). 1608 samples, simulated using D-Track, dynamic behaviour simulation software package, develop machine models. Three different aspects models evaluated, namely overall accuracy, from show DNN has highest accuracy 96%. In addition, be with nearly 100% accuracy. CNN performs better than RNN terms detection. However, severity classification. Overall, offers best approach detecting classifying their