FAULT DIAGNOSIS OF A BALL BEARING USING VIBRATION ANALYSIS

Abstract

Failure of ball bearings is a major cause of rotatory machine failure resulting in large economic losses and possible injury to human lives. Correct diagnosis helps to identify bearing faults and use the bearings effectively, preventing catastrophic failures of rotating machines. Detecting potential problems early, and condition monitoring allows for proactive maintenance and reduces the downtime of machines. Improved equipment reliability and efficiency lead to lower maintenance costs and increased productivity. The study aimed to classify three types of bearing faults: Inner Raceway, Outer Raceway and Ball fault. Inner Raceway and Outer Raceway faults were introduced in the 608-deep groove ball bearing via an electric grinder making 1.5mm line cuts through the axis of the bearing. For ball fault, one ball was removed out of the 7 present. An accelerometer with sampling frequency of 1000Hz was fixed on the drive end of the AC induction motor to acquire the vibration signals. Models for Support Vector Machine (SVM), Convolutional Neural Network (CNN) (both 1D CNN and 2D CNN) and Long Short Term Memory Network (LSTM) were developed. Raw bearing fault data from an open-source database, CWRU was fed into the models to check their accuracy. A minimum accuracy of 92.47% was acquired from the raw CWRU data, thus validating the models. The acquired fault data from the accelerometer was processed through a (20Hz, 500Hz) band pass filter before feeding into the machine learning and deep learning models. 70% of the vibration data was used for training the models while the remaining 30% was used for testing. Out of the 4 models compared, 1D CNN gave a maximum test accuracy of 98.35%.