Rolling Bearing Diagnosis Based on Deep Learning Enhanced by Various Dataset Training

DOI Web Site

Bibliographic Information

Other Title
  • 多様データセットを用いた深層学習による転がり軸受の損傷診断

Description

<p>In recent years, there has been an increasing interest in deep learning technique for bearing flaking diagnosis, because it is possible to select vibration features and set diagnostic thresholds without domain knowledge of bearing diagnosis. The authors has proposed previously the CNN-LSTM model trained by using various dataset which would be better generalization performance than that in studies ever reported , i.e., the model might be available for actual rotating machinery, in which vibration feature is affected by type of bearings, various operating conditions and unknown disturbance. In this study, the model was analyzed by Grad-CAM, which was known as a visualization tool for deep learning model for image data, to know how the model detects the flaking. The analysis of Grad-CAM has shown that the periodic impulsive waveforms were detected when the test signal derived were of fault bearings as well expertized engineer will do. Furthermore, it is proved that the extracted feature is still available even though the waveforms were contaminated with white noise. In addition, the analysis also revealed the over-fitting situation of trained model. Therefore, it was concluded that Grad-CAM analysis was able to evaluate the trained deep learning models of bearing vibration diagnosis.</p>

Journal

Keywords

Details 詳細情報について

Report a problem

Back to top