RUL prediction method based on sequential health index evaluation with multidimensional coupled degradation data.

Area of Science:

• Engineering
• Data Science
• Machine Learning

Background:

• Remaining Useful Life (RUL) prediction is vital for predictive maintenance.
• Challenges include limited labeled life-cycle data and complex degradation patterns.
• Accurate health index (HI) construction is difficult due to multidimensional data coupling.

Purpose of the Study:

• To develop an advanced RUL prediction method addressing data scarcity and complex degradation.
• To propose a novel approach integrating a CNN-Transformer model with sequential health index evaluation.
• To reduce model complexity and computational load while enhancing prediction accuracy.

Main Methods:

• A CNN-Transformer hybrid model with a chunk-interaction mechanism for reduced complexity.
• A sequential health index evaluation scheme using Mahalanobis distance and Sequential Evaluation Ratio (SER).
• Dynamic HI construction that eliminates the need for high-quality labeled life-cycle data.

Main Results:

• The proposed method demonstrates superior performance compared to LSTM, Transformer, and Att-BiGRU models.
• Achieved higher prediction accuracy and robustness across multiple datasets.
• Effective in label-scarce scenarios, highlighting its practical applicability.

Conclusions:

• The integrated CNN-Transformer and sequential HI evaluation method offers a robust solution for RUL prediction.
• This approach effectively handles data scarcity and complex degradation patterns.
• It provides a significant advancement for predictive maintenance strategies.