A Study on the Effectiveness of Current Data in Motor Mechanical Fault Diagnosis Using XAI

Signals directly affected by the motor's mechanical fault are noise/vibration data. However, the phase current signal is also affected to some extent. A deep learning algorithm can classify minute changes in the current signal. By using XAI (eXplainable A.I) which outputs the data sensitivity of the deep learning model, it is possible to recognize the characteristic points of the current data from the model's point of view. In this study, the theoretical ideal current point and the XAI output are compared. In addition, when a model cannot be trained so that XAI is inapplicable, an analysis method using the Pearson correlation coefficient is proposed. Through this process, the effectiveness of current data in mechanical fault diagnosis is confirmed.