Comparing one and two class classification methods for multiple fault detection on an induction motor

This paper shows that one class classification methods combined with wavelets are capable of detecting the majority of faults on a 3 phase induction motor learning only from healthy data. It has important applications for condition monitoring of electro-mechanical machines in industry as it means that rare and expensive-to-obtain faulty data is not required. The experiments were carried out under laboratory conditions on a small, well worn, 3 phase induction motor, which had bearing faults, imbalance faults, broken rotor bar faults and winding faults imposed on it. A two class support vector machine (SVM) was trained on equal amounts of healthy and faulty data to demonstrate that it has high accuracy when faulty data is readily available. The combination of the one-class SVM and wavelets to the best of the author's knowledge has not been previously attempted but shows acceptable results.