关键词: 时频域信号比例优化器, 精准记忆TPA, 多重膨胀, 多核时间卷积网络, 轴承剩余使用寿命预测

Abstract: Rolling bearings were recognized as common key components in mechanical equipment. Accurate prediction of their remaining service life was considered crucial for safe and stable operation. A Multi inflated Multi kernel Time Convolutional Network (Mi-MkTCN) model was proposed to address current challenges in bearing life prediction. The model was based on a Time-Frequency domain signal Ratio Optimizer (TFRO). Three main problems were targeted:unclear bearing degradation characteristics,poor model generalization ability,and difficulty in capturing long-term data dependencies. The TFRO optimizer was designed to accurately retain important information. At each time node,past and current information were reassembled. Important time-frequency domain features from past information were proportionally allocated.Multiple dilation methods were employed in Mi-MkTCN to prevent loss of important features. A multi-kernel temporal convolutional network was then used to extract features at different scales. The effectiveness of the proposed model improvement method was demonstrated through ablation experiments. Algorithm comparison studies were conducted to verify the superiority of the TFRO-based Mi-MkTCN model. Performance metrics were recorded as follows: MAE(0.001 45),MSE(0.050 69),and RMSE(0.120 45). The experimental results showed that the proposed method significantly improved the prediction accuracy of the remaining service life of bearings,providing a high-precision and highly robust solution for predicting the remaining service life of bearings.

Key words: Time-Frequency domain signal Ratio Optimizer (TFRO), precision memory TPA, multi inflated, Multi inflated Multi kernel Time Convolutional Network(Mi-MkTCN), prediction of remaining useful life of bearings