基于频域信息裁剪和多尺度补丁嵌入的设备剩余使用寿命预测OA

Remaining useful life(RUL)prediction for mechanical equipment has great significance in ensuring production safety and improving production efficiency.To address the sensitivity of rotating support equipment to sudden events during long-term operation,a novel RUL prediction method is pro-posed.Firstly,leveraging the low-rank property of traditional feature frequency-domain information,Fourier components are randomly selected,and a re-selection mechanism is employed to highlight the influence of high-frequency components,enhancing the model's sensitivity to sudden events represented by high-frequency features while preserving information density.Secondly,a multi-scale patch embed-ding technique breaks through the limitation of traditional fixed patch scales,adopting a multi-window patch fusion approach to further improve the method's capability of capturing information at different frequencies.In terms of computational efficiency,an improved efficient computational attention mecha-nism is combined with frequency-domain information pruning and multi-scale patch embedding,reducing the time complexity to linear time,which makes it conducive to deployment and use in production envi-ronments.Experimental results on 2 public datasets,XJTU-SY and PHM2012,demonstrate that the proposed method achieves improvements in prediction performance.Compared with baseline methods,the root mean square error(RMSE)and mean absolute error(MAE)are reduced by 19.16%and 19.60%on average,respectively,under the same prediction horizon,while effectively improving com-putational efficiency.