A Fault Diagnosis Method Based on Wavelet Denoising and 2DCNN under Background Noise

This paper develops a novel method named wavelet denoising convolutional neural network (WDECNN) for fault diagnosis with background noise. The continuous wavelet transform (CWT) is first applied to transform the measured raw vibration data into time-frequency images which serve as the inputs of WDECNN. Then, a light-weight two-dimensional CNN (2DCNN) model is incorporated in WDECNN to simplify the network architecture, while a wavelet denoising module is also applied in it to achieve high accuracy of fault identification in the noisy environment. Particularly, the wavelet denoising module which consists of wavelet decomposition and denoising is parallel to the 2DCNN model, and the denoising results are integrated into pooling layers in the 2DCNN model. Thus, the denoised information is added to the 2DCNN model to improve its feature learning ability. Finally, the effectiveness of the developed method is validated on Paderborn bearing dataset, which illustrates its fault diagnosis capability under background noise.