Fault Diagnosis Method for Small Sample and Multi-Condition Based on Denoising Autoencoder and Convolutional Neural Network

To address the problem of fault diagnosis under multiple working conditions, a fault diagnosis method based on denoising autoencoder and convolutional neural network (CNN) is proposed. First, the multi-channel one-dimensional sensor data is processed into two-dimensional square matrix data, and the denoising autoencoder is trained using this data. The encoder part of the trained denoising autoencoder is used as a feature extractor to extract features from the two-dimensional square matrix data, which are then fed into the CNN for classification. Experimental results show that this method can achieve a diagnosis accuracy rate of 99.67% on the motor fault dataset from the subway train transmission systems simulation experiment. The effectiveness of incorporating the denoising autoencoder in the method is demonstrated through comparative analysis of the experimental results, as well as highlighting key considerations for data preprocessing.