Bearing Health Condition Monitoring using Time-Domain Acoustic Signal Features

Bearings are widely used in industries because of their low friction and high precision moments. As a result, they are used in almost every rotating machinery, making it essential to monitor them. Also, they are the most vulnerable part of the machine due to its often-working condition at high load and high speed. If such bearing damage goes unnoticed, it results in problems within the bearings and even affects other mechanical components. Usually, bearing damage occurs at the outer cage, inner cage, and ball mainly because of its worn-out condition due to metal-to-metal contact. Regular bearing health condition monitoring is a process to increase safety and reduce the machine's maintenance cost in time. This paper deals with the acoustic signals provided by the acoustic sensor at four different health conditions to monitor the bearings. These signals are segmented at one second to classify the data from time-domain features and compare the model performance with and without feature selection. Two prominent time-domain features, i.e., slope sign change and kurtosis of energy operator, are selected using feature selection to classify bearing health in the present work. Different machine learning algorithms such as Naive Bayes and Support vector machine was used for classification and obtained an accuracy of 99.70%, 99.69%, respectively.