A Novel Bidirectional Sparse Filtering Method for Bearing Fault Diagnosis Under Noise Interference

Traditional bearing fault diagnosis methods often suffer from the interference of signal noise, which is inevitable in real-world industrial settings and obscures valuable fault features embedded within the bearing vibration signals. Consequently, accurately extracting effective bearing fault features becomes exceedingly difficult. To address this issue, this article introduces a new bidirectional sparse filtering (BiSF) method to process and extract features from bearing vibration signals. Specifically, the BiSF method designs a novel bidirectional normalization strategy to improve the feature extract ability under noise interference conditions. Meanwhile, squeeze and excitation (SE) attention is embedded in BiSF to conduct adaptive weighting of input data and autonomously learn optimal weights. Then, the learned features are input to the softmax classifier for accurate diagnosis of bearing faults. A simulation study and a bearing experimental case are adopted to testify to the effect of the BiSF model. The experimental results illustrate that the proposed BiSF can effectively eliminate noise components than the traditional SF method, and achieve a high diagnosis accuracy of 97.45% +/- 0.32 % for bearing fault classification.