Condition Based Monitoring of Ball Bearings Using Machine Learning

Abstract

Ball bearings are used in rotary machines such as turbines, generators, automobiles, and electric motors just to name a few. They operate in harsh conditions, heavy loads and shocks which deteriorate their health and if not monitored timely, can lead to costly damage. With the advancements in technology, various sensors are integrated to monitor machines’ health. An important health indicator of such machines is vibration signal data which provides meaningful insight into a variety of mechanical faults. Traditionally, bearing faults are analyzed in time frequency domain which is incapable to classify the types of faults accurately. In this project, we used vibration signal data acquired by Korean Advanced Institute of Technology and ourselves and coupled it with machine learning algorithms to classify bearing faults. Additionally, we developed a Convolutional Neural Network that classifies these faults solely based on raw vibration signal data unlike machine learning algorithms that rely on an extensive number of features for greater accuracy.