Autoregressive data generation method based on wavelet packet transform and cascaded stochastic quantization for bearing fault diagnosis under unbalanced samples

Abstract

The capacity to diagnose faults in rolling bearings is of significant practical importance to ensure the normal operation of the equipment. However, because it is challenging to obtain a sufficient number of labeled fault samples in practical engineering applications, the collected datasets will be unbalanced, which will greatly affect the performance of diagnosis. To tackle the above mentioned challenges, an autoregressive data generation method based on wavelet packet transform and cascaded stochastic quantization for fault diagnosis under unbalanced samples is proposed in this paper. Firstly, we propose a cascaded autoregressive data generation model. The hierarchical structure can better learn the multi-scale deep features of the input signal. Secondly, to increase the diversity of pseudo samples, we use wavelet packet decomposition to obtain high and low frequency information, which is used as the training data for the generative model, then reconstruct the generated high and low frequency information to obtain pseudo samples. Finally, a parallel attention-guided quantization mechanism and a multi-scale quantization feature fusion module are proposed to better integrate latent variable features of different scales. The experimental results validate the effectiveness of the proposed method, and demonstrate its significant application potential in fault diagnosis under unbalanced samples.