An Autonomous Bearing Fault Diagnosis System

Abstract

The rolling element bearing plays a significant role in any mechanical manufacturing industry. The failures in bearing leads to sudden shut down of the machinery causing heavy losses to the industry. Therefore, bearing fault diagnosis using vibration signal processing has been a vital subject of study in past years. A sensitive and reliable fault diagnosis system is required to identify bearing faults in the early stages to prevent sudden failures in machines. This research work proposes autonomous bearing fault diagnosis techniques which are capable of diagnosing three kind of bearing faults from vibration signals. Two different fault diagnosis techniques are proposed and analyzed in this work for efficient classification of vibration signals. Firstly, Tunable Q-factor Wavelet Transform based feature extraction technique is proposed. In proposed technique, vibration signals are decomposed into time-frequency coefficients using Tunable Q-factor Wavelet Transform and entropy based features are computed from decomposed time-frequency coefficients. Classification of vibration signals is performed using three soft computing methods viz. Support Vector Machine, Artificial Neural Network and Random Forest Tree classifier. Another technique of feature extraction based on double decomposition of vibration signals using Empirical Mode Decomposition and Tunable Q-factor Wavelet Transform is proposed and evaluated for bearing fault diagnosis. In this work, vibration signals are first decomposed using Empirical Mode Decomposition and then using Tunable Q- factor Wavelet Transform. Further, Higuchi‟s Fractal Dimension is evaluated from the decomposed sub-bands as features and feature vector is prepared. This feature vector is fed to Support Vector Machine classifier for recognition of bearing faults. Multiple experiments are carried out with varying values of user defined parameters and classification is performed to attain the best set of parameters to attain the highest classification performance in bearing fault diagnosis task. Classification results revealed that proposed techniques have a potential to design and develop a real time bearing fault diagnosis system.