Bearing Fault Location Based on Feature Visualization of Convolutional Neural Network

In industrial production, bearing is one of the most common and indispensable key components. It is of great significance to prevent safety accidents caused by bearing failure, timely fault type identification, and accurate fault location positioning. In this paper, we propose a method for bearing fault location based on feature visualization of convolutional neural network (CNN). In this method, the bearing vibration signals are transformed into images and the Grad-CAM feature visualization method is used to find the feature regions of the images. Thus, the points of the feature regions are traced back to the one-dimensional signals to roughly locate the faults. It has been proved that the experimental results of bearing fault location by this method are in good agreement with the theoretical values.