Fault Diagnosis of Ball Bearings Using Machine Learning of Vibration Signals

One of the major reasons for the failure of rotating machines are rolling element bearing defects. Failure of these bearings leads to unplanned maintenance shutdowns and unsafe working conditions in a plant. For these reasons, it is very important to detect and identify the defects in rolling element bearings in their early stage. Vibration signals are known to monitor the condition of rotating machinery. The performance of conventional intelligent fault diagnosis methods depends on the feature extraction of vibration signals, which requires signal processing techniques, good proficiency, and human expertise. In recent times, deep learning algorithms have been extensively used in the health monitoring of machinery. Here in this study, a machine learning-based model for the detection of bearing defects is analyzed. The bearings used for this analysis is 6305 deep groove ball bearing. Defects like ball defect, outer race defect, and inner race defect were considered. A motor-driven variable speed test rig rotor supported on ball bearings are used and for the different defects, vibration responses were saved and analyzed. Vibration data for a healthy bearing is also collected. TensorFlow is used to implement machine learning models. The results obtained shows that machine learning-based fault diagnosis is very efficient in identifying the bearing defects. This method also reduces the post-processing of the vibration data, which is the most time consuming and human expertise required phase. These Machine learning-based fault diagnosis systems can be developed for online defect detection also.