Motor diagnosis based on vibration imaging and CNN

In modern industry, motor failures are difficult to avoid. Traditional diagnosis methods are limited by feature extraction, and manual feature extraction can cause great variability. This paper proposes a diagnosis method based on vibration image and Convolutional Neural Network (CNN). By normalizing and binarizing the vibration data in one dimension, it forms the two-dimensional vibration images. Then, CNN completes the feature extraction and learning of image data, avoiding the limitations of traditional methods. A variety of common fault data can be generated by the motor fault state simulation of the experimental platform. After the test of experimental data, the proposed method can achieve 100% accuracy and realize error-free discrimination.