A remaining useful life prediction method based on PSR-former.

Area of Science:

• Mechanical Engineering
• Data Science
• Machine Learning

Background:

• Rotating machinery generates complex, non-linear vibration data crucial for fault analysis and Remaining Useful Life (RUL) prediction.
• Environmental complexities often corrupt vibration data, hindering accurate Health Index (HI) formation for Prognostics and Health Management (PHM).

Purpose of the Study:

• To propose a novel PSR-former model for enhanced RUL prediction from vibration data.
• To improve the accuracy of HI estimation in challenging mechanical environments.
• To provide a more stable and efficient deep learning architecture for PHM.

Main Methods:

• A Phase Space Reconstruction (PSR) layer is employed for feature fusion and in-depth vibration data understanding.
• A Transformer layer with an attention mechanism and layer-hopping connections is utilized for stable and convergent RUL prediction.
• The proposed model was validated on the Intelligent Maintenance Systems (IMS) bearing dataset.

Main Results:

• The PSR-former model achieved a Root Mean Square Error (RMSE) of 1.0311, demonstrating high prediction accuracy.
• Comparative analysis against LSTM, GRU, and CNN models showed the superiority of the proposed method.
• The model effectively establishes a precise RUL prediction model.

Conclusions:

• The PSR-former model offers a significant advancement in RUL prediction accuracy for rotating machinery.
• The proposed architecture effectively handles complex vibration data for improved PHM.
• This method provides a robust solution for predicting machine health and RUL.