Fault diagnosis of bearings

Abstract

A rolling element bearing is an essential component of rotating machinery system to reduce friction and support loads while allowing a relative motion between two components and transmitting power. Since rolling element bearings are so widely used in machines, their failure can cause the breakdown of machine. As a result, they are most important components in industrial applications. Thus, it is important to detect defects like cracks, spalls, pits, wear etc. at the initial stages because these damages can lead to permanent failure of the machine. So, the fault detection and diagnosis of bearings are done by using vibration measurement techniques. In this project, condition monitoring of the bearing is carried out. Vibration signals of bearings at different conditions are recorded. Statistical features are calculated from the signal as they are the good indicator of bearing health and train the machine learning model SVM and DT for classification of fault. Vibration responses for various bearing faults are presented in a frequency spectrum, and then a traditional envelope analysis is performed using a band-pass filter to reveal location of fault with the help of spectrum analysis. Another approach is also performed in which vibration signal is decomposed into sub bands using Tunable Q wavelet transform (TQWT) and feature extraction of sub-bands signal is done. Two distinct machine learning algorithms, support vector machine (SVM) and decision tree (DT), are employed for automated defect identification in the rolling element bearing. SVM was found to be capable of detecting 96% of the samples, but DT was only able to classify 95% of them.