Rolling Element Bearing Fault Diagnosis Using Hybrid Machine Learning Models

Rolling element bearings are widely employed in a variety of mechanical equipment. An effective fault diagnosis can limit the occurrence of accidents, maximize the operating profile of the bearing, and save maintenance costs. Faults in a rolling element bearing commonly include the faults in inner race, outer race, rolling elements, and cage. In this work, an approach based on machine learning algorithms was used. For testing, a public data set from the Case Western Reserve University bearing center (CWRU) was used. After testing different combinations for feature extraction and selection, the best method of processing the original data according to the machine learning model was chosen to be used for the multi-class classification problem. The developed diagnostic methods demonstrated the ability to classify and detect damage to roller bearings. The logistic regression, support vector machine, and random forest algorithms were tested, and the best classification result was obtained using the random forest algorithm on selected variables using the permutation importance of features from the time, frequency, and time-frequency domains. The algorithm based on the Random Forest method achieved the highest average value of accuracy in predicting faults in the roller bearing.