Fault Sign Detection Method for Rotators based on One-class Support Vector Machine with Successive Multi-level Class Construction

Abstract

A new fault sign detection method for rotators based on one-class support vector machine is proposed. The vibration of rotators occurs not only during faulty operation, but also during normal operation. Furthermore, faults are rare events, and it is difficult to obtain an indication of fault occurrence beforehand. To overcome this issue, a one-class support vector machine with successive multi-level class construction is proposed. The advantage of this method is that the time varied feature can be classified according to the class-set. The experiments are conducted by using actual rotator data. The results show that the class-set feature is capable of indicating the fault sign.