Anti-noise Fault Diagnosis Model Based on Convolutional Neural Network

Abstract

Fault diagnosis methods based on deep learning have a strong ability to distinguish faults with unknown mechanisms in the field of mechanical fault diagnosis. However, when the noise interference is strong, the accuracy of the model will decrease to a certain extent. This paper proposes an anti-noise fault diagnosis model named APR-CNN. The model is designed based on a two-dimensional convolutional neural network, which uses the wavelet time-frequency images as input. According to the characteristic of the periodic transformation of the wavelet time-frequency image of the bearing signals, average pooling on rows method is used to compress the time domain information and extract the features effectively. Compared with classical methods on the open-source bearing fault dataset, experiments show that the APR-CNN model can still have an accuracy rate of 98% even in a noisy environment with SNR of −10, which is at least 30% higher than other methods.