The wheel of railway vehicles is a crucial component of railway operation safety. However, current railway vehicle wheel inspections are limited to post-maintenance procedures that only involve periodical visual inspections and maintenance by personnel in the event of physical abnormalities, such as vibration and noise, are detected during heavy overhaul and light maintenance of railway vehicles, rather than proactive maintenance.
Accordingly, predictive technology for wheel conditions became necessary to prevent railway vehicle damage, human casualties, and other accidents that are associated with defective wheels. However, the current lack of predictive technology for railway vehicle wheel conditions led to the background of this study.
This is a condition classification study examining the light rail vehicle wheel (rubber tires) conditions by employing machine learning methods. Experimentations were conducted by applying Support Vector Machine, Random Forest, and 1D-CNN algorithms by measuring five factors including three-axis acceleration, temperature, and pressure. The use of 1D-CNN algorithm yielded an exceptional accuracy rate of 98.9% in condition classification recognition.