Fault Detection of Bearing using Signal Processing Technique and Machine Learning Approach

Abstract

In large- or small-scale industries, machines have rotary element supported by bearings for accurate drive and fixed support. Fault diagnosis has gained much importance in recent times due to increased bearing failures. This demands an efficient diagnosis methodology to detect faults in bearings. In this work, fault diagnosis for the acquired vibration signals of healthy and fault seeded in rolling element bearings were investigated using signal processing technique and online machine learning approach. The research work is carried out in two phases. The first phase of research work investigates fault detection of bearing using conventional signal processing techniques such as time domain analysis and spectrum analysis. The results show that signal processing techniques may be useful for revealing post fault detection information. It was also concluded that the use of different signal processing techniques is often necessary to achieve meaningful diagnostic information from the signals. The second phase of research work describes fault diagnosis of bearing using machine learning approach. Using MATLAB, Discrete Wavelet Features (DWT), were extracted from acquired signals for different rolling element bearing conditions. J48 algorithm was implemented to extract most significant features. Extracted features were used as input to different classifiers to obtain maximum classification accuracy of rolling element bearings. The results showed that machine learning technique could be used to detect and classify the different fault sizes effectively with vibration signals.