Fault Diagnosis of Rolling Bearing Under Different Conditions Using Modified Inception and LSTM with Attention Network

Fault diagnosis of rolling bearing is very important for health prediction and maintenance of rotating machinery. However, the bearing operating conditions are complicated in the practical industrial environment, which is challenging to diagnose fault efficiently and accurately. In this paper, an end-to-end intelligent fault diagnosis method of modified inception structure cascades to Long Short-Term Memory (LSTM) with Attention Mechanism (AM) is proposed. Firstly, the modified inception as the feature extraction module can effectively extract the multi-scale feature information from different fault vibration signals by using its multi-branch local sparsity, which can also improve the generalization of this model. Then, the feature information is input into the classification module composed of LSTM based on AM. It can further learn the temporal relevance of feature information. Meanwhile, the AM can adaptively map weights, which significantly enhances the memory ability of LSTM for important information and improves the robustness of this model. In addition, the internal working principle of the overall network is explained by the visual analysis of the key steps and results. Finally, the proposed method is verified by experimental results including single and compound fault bearing under different conditions. It is shown from the results that the proposed end-to-end intelligent fault diagnosis method can accurately identify different bearing fault types under different speeds or background noise.