Spatio-Temporal Feature Fusion for Residual Life Prediction of Bearings Under Cloud-Edge Collaboration
In order to address the issues of feature dependencies,degradation trend representation,and real-time concerns in the prediction of remaining useful life(RUL)for rolling bearings,this paper introduces a method for predicting the remaining life of bearings that integrates spatial and temporal features in a cloud-edge collaborative framework.Firstly,in the offline phase,noise reduction is performed on historical degra-dation data of bearings based on expert prior knowledge.Secondly,temporal and frequency domain degra-dation features are extracted from the denoised signals,and the extracted degradation features are analyzed and screened.Finally,in order to better capture the interaction between features,the method uses the Pear-son correlation coefficient to analyze the similarity correlation of the selected degradation features,and con-structs a feature space map based on the similarity correlation parameter as an input to the graph convolu-tional network(GCN)-Transformer model for training and testing in the cloud-side collaboration.The real-time prediction phase is tested to reduce the burden on the cloud to improve the real-time prediction.In the experiments on the XJTU-SY dataset,the proposed method reduces the MAE and RMSE metrics by 10.5%and 11.3%,respectively,and reduces the average response time(real-time metrics)to 0.363 using the cloud computing strategy compared with other literature prediction methods.The experimental results vali-date the effectiveness of the proposed method.
rolling bearingsremaining useful lifespatio-temporal feature integrationcloud-edge collabo-rationgraph convolutional networks