Remaining Useful Life Prediction Based on Adaptive SHRINKAGE Processing and Temporal Convolutional Network.

Area of Science:

• Mechanical Engineering
• Machine Learning
• Predictive Maintenance

Background:

• Remaining Useful Life (RUL) prediction is crucial for industrial equipment reliability and maintenance.
• Traditional methods often require extensive feature engineering and struggle with long-term dependencies.
• Recurrent Neural Networks (RNNs) show limitations in capturing long historical data for accurate RUL prediction.

Purpose of the Study:

• To propose a novel RUL prediction method that overcomes the limitations of traditional approaches.
• To enhance RUL prediction accuracy by directly utilizing multi-channel sensor data.
• To improve the reliability and maintainability of industrial equipment through advanced predictive analytics.

Main Methods:

• A Temporal Convolutional Network (TCN) is employed as the core prediction model.
• An adaptive shrinkage processing sub-network is introduced to intelligently filter noise while preserving essential features.
• Multi-channel data is directly input into the network, eliminating the need for manual feature extraction.

Main Results:

• The proposed method significantly reduces Mean Absolute Error (MAE) by up to 52% and Root Mean Square Error (RMSE) by up to 64% on benchmark datasets.
• Experimental validation on PHM2012 and XJTU-SY rolling bearing datasets demonstrates superior performance.
• The adaptive shrinkage processing effectively balances noise reduction and feature retention.

Conclusions:

• The combined adaptive shrinkage processing and TCN model offers a highly accurate and efficient solution for RUL prediction.
• This method provides a valuable tool for enhancing the safety and supportability of industrial machinery.
• The approach shows significant potential for practical application in predictive maintenance strategies.