Deep learning-based fault detection in railway wheelsets using time series analysis

Maintenance of Railway rolling stock is usually scheduled based. However, the mechanical parts, especially wheelset may wear down prematurely due to several factors such as excessive braking and traction forces environmental conditions. This makes maintenance less effective sometimes it results in derailments. paper presents a deep learning-based technique detect wheel conditions so that can be performed promptly efficiently. A time series dataset axle vibrations generated using simulation model wheelset. The then used train test learning model. Long short-term memory (LSTM) architecture selected for this application since designed perform better datasets. show good performance terms training testing accuracy. tested different defect scenarios mean square error prediction railway parameters around 15%.