Defect Diagnosis of Rolling Element Bearing using Deep Learning

Abstract

The bearing element is an important part of many mechanical facilities. This element is very important for the maintenance of the machine and for the detection of machine faults. The presence of a rolling fault in the machines can affect the overall performance of the machine. For this reason, the bearing faults must be diagnosed in order to be monitored for the machine in a healthy manner. The generated vibration signals as a result of the machine's rolling action should be monitored in order to diagnosis the machine defects. It is assumed that the vibration signals are composed of large-scale features and noise components. The performance of traditional diagnostic methods is based on characterizing faulty vibration signals. The utility of traditional techniques requires signal processing, expert knowledge, and human effort. In the most basic sense, the diagnosis of defective vibration signals is based on a comparison of defective vibration signals to healthy signals. In this study, deep learning technology which has been applied to many fields in recent years has been used to detect defective vibration signals. The ability to learn the complex features of the deep learning architecture provides superiority over traditional diagnostics to fault diagnosis. In this study, four classes of vibration signals have been used as input data. Training and test data for each class have been created. These images have been then trained on the CNN model that has been developed in Keras deep learning library. This model has been developed to automatically diagnose vibration signals. The developed model produced results with high accuracy.