A combination of residual and long-short-term memory networks for bearing fault diagnosis based on time-series model analysis

Data-driven methods have been considered as an effective tool for detecting the nonlinear and complex changes of time-series data and extracting early fault features from bearing vibration measurements in industrial applications. Due to the lack of a feature extraction ability of the residual network, which is an existing typical intelligent fault diagnosis deep model of bearing vibration signal, it is difficult to capture the long-term dependence between the time-series data. To overcome this problem, we propose a combination of residual and long-short-term memory networks (Resnet-LSTM) and develop a fused time-series model. The two-dimensional signal of bearing vibration is input into the residual network and the local feature is extracted by embedding a residual layer. In addition, the bearing feature information is loaded into a long-term memory unit and the forgetting mechanism is introduced to extract the global features of the time-series data. The advantage of the proposed method is that it takes full advantage of all the local deep features and global time-series features from the bearing vibration signal. This approach enables us to learn sequential features in different interval lengths and capture the local sequence features of the data information flow, which can improve the fault diagnosis accuracy of existing methods. Experimental results demonstrate that the proposed method outperforms other common methods in single and compound fault diagnoses of bearings.

Automated summary

The proposed method in this paper combines residual and long short-term memory networks to improve fault diagnosis accuracy of bearing vibration signals. By extracting both local and global features, it can better capture the changes in time-series data.