Fault diagnosis for rapid transit using pattern recognition and classification techniques

For many electromechanical systems, early fault detection is invaluable as pre-emptive maintenance can result in tremendous savings for the operator, instead of dealing with the faults when they occur. In this work, we investigate the use of pattern recognition and classification techniques for fault diagnosis in rapid transit vehicles. Operational data are processed using the principal components analysis method to reduce their dimensionality, and are then clustered and classified into identifiable behaviors or classes. Faulty data are examined and compared to normal behaviors. This proof-of-concept demonstration shows promising results to warrant further investigation in the use of pattern recognition techniques in fault diagnosis for rapid transit vehicles.