Acoustic emission (AE) technology can capture early weak faults in rolling bearings. However, due to the nonstationary nature introduced by high sampling rates, traditional methods often struggle to extract useful features effectively. Existing models mostly use only one type of domain information and lack multidomain cooperation. To solve this problem, this article proposes a deep learning framework called wide-kernel dual-domain bidirectional residual network (WDBRN). It combines wide kernel convolution with dual-domain adaptive fusion. The framework builds two parallel branches in the time domain and the frequency domain. It uses wide convolutional kernels to get a large receptive field. It also uses a bidirectional gated recurrent unit module to capture temporal feature evolution and spectral correlations. An adaptive gated fusion mechanism is designed to adjust the weights of time and frequency features based on signal quality. This allows the model to use information from both domains together. Experimental results show that the proposed method gets an average diagnosis accuracy of 98.72% under different speed conditions. Notably, WDBRN outperforms convolutional neural network (CNN)-long short-term memory (LSTM) and multi-scale convolution neural network (MSCNN) in both classification accuracy and feature clustering quality. This study provides a practical engineering reference for AE signal-based bearing fault diagnosis in key petrochemical rotating machinery operating under high noise environments.
Wu et al. (Tue,) studied this question.