Bearing Damage Detection using Support Vector Machine

Abstract

Vibration can be used for preventive condition monitoring in industrial equipment. Vibration measurement may avoid unexpected damages. The measurement data can be analyzed either by using signal processing or machine learning. This work assessed bearing damage detection as bearing is intensively used in industrial mechanical equipment. The assessment was performed experimentally by developing system prototype and measurement tools, including the use of laser vibrometer and a labjack as well as support vector machine to predict some types of damages. The work has been able to predict the BPFI, BPFO, BSF, and FTF damages by 93.125% accuracy. However, accuracy plunged to 61.25% when bearing damages were of three types: BPFO, BSF and FTF.