A Model for Prediction of Outer Race Defects of Rolling Contact Bearing based on Vibration Data Using Machine Learning Algorithms

The detection of bearing defects while the machinery is in use is essential for predicting the incipient failure and thereby providing an opportunity to take remedial measures for preventing the costly downtime and ensuring the safe and efficient operation of rotating machinery. With the increasing availability of vibration sensor data and the development of machine learning techniques, the ML methods have become a popular approach for automated fault diagnosis in bearings. In this paper, an attempt has been made to detect the faults in the outer race of bearing using different ML algorithms. An experimental setup has been designed and fabricated to conduct experiments on healthy and faulty bearings and the vibration signals were captured. The captured vibration signals were directly employed as images for training the ML algorithms without the need for conducting the spectral analysis. Six machine learning algorithms, namely, Linear Regression (LR), Decision Tree (DTR), KNN Regression (KNNR), Random Forest Regression (RFR), Convolution Neural Network (CNN), Naive Bayes (NB) were separately applied to classify the location of defects within the outer race of the ball bearing. The accuracy table are used to find the best suitable algorithm for the predictions. The methodology includes data preprocessing techniques, network architectures, training strategies, and evaluation metrics. It has been established that the use of ML technique is very effective in detecting the bearing defects and CNN is able to achieve 100% accuracy.