RUL Prediction Based on xLSTM-Transformer Neural Network for Rolling Element Bearings Under Different Working...

Area of Science:

• Engineering
• Computer Science
• Data Science

Background:

• Predictive maintenance relies on accurate Remaining Useful Life (RUL) prediction for rolling bearings.
• Traditional deep learning models face challenges in capturing both short-term dynamics and long-term degradation trends in health indicators (HI).

Purpose of the Study:

• To propose a novel hybrid deep learning model for enhanced RUL prediction of rolling bearings.
• To address limitations in capturing local temporal features and global degradation trends simultaneously.

Main Methods:

• Developed a hybrid model integrating extended Long Short-Term Memory (xLSTM) with Transformer architecture.
• Employed an encoder-decoder structure incorporating Multi-Head Attention with xLSTM modules.
• Validated the model on the XJTU-SY and PHM2012 rolling bearing datasets.

Main Results:

• The proposed model demonstrated superior performance over comparative methods on key metrics like RMSE and R².
• Achieved significant improvements in prediction accuracy and multi-dataset generalization capability.
• Effectively captured both short-term dynamic features and long-term degradation patterns.

Conclusions:

• The hybrid xLSTM-Transformer model offers a more accurate and generalizable solution for bearing health assessment and RUL prediction.
• The model shows significant potential for advancing intelligent health management in industrial equipment.
• This approach enhances the reliability and efficiency of predictive maintenance strategies.