A Fractional-Derivative Multi-Kernel Adaptive Learning Approach for Remaining Useful Life Prediction of Rotating...

Area of Science:

• Mechanical Engineering
• Prognostics and Health Management (PHM)
• Signal Processing

Background:

• Remaining Useful Life (RUL) forecasting is crucial for condition-based maintenance in rotating machinery.
• Current methods face challenges with nonlinear degradation, parameter calibration, or large datasets.
• Existing approaches often lack transparency ('black-box' models) or require extensive data.

Purpose of the Study:

• To develop a robust and transparent RUL prediction method for rotating machinery.
• To address limitations of physical models and deep learning approaches in small-sample scenarios.
• To enhance the accuracy and reliability of predictive maintenance.

Main Methods:

• A novel fractional-derivative multi-kernel adaptive learning approach is proposed.
• Integration of kernel adaptive learning with a multi-kernel mixture measure for a 'white-box' architecture.
• Incorporation of Hadamard fractional derivative to capture degradation memory and hereditary properties.
• An adaptive 3σ confidence interval with delayed triggering for First Prediction Time (FPT) identification.

Main Results:

• The proposed method demonstrates superior predictive accuracy compared to established baselines.
• Lower estimation errors and the lowest asymmetric penalty scores were achieved.
• Effective performance in practical small-sample scenarios was confirmed through evaluations.
• The method successfully captures complex long-range temporal dependencies in degradation processes.

Conclusions:

• The fractional-derivative multi-kernel adaptive learning approach provides robust and transparent RUL prediction.
• This method effectively overcomes the limitations of existing prognostic techniques.
• The findings contribute to advancing predictive maintenance strategies for rotating machinery.