Multimodal Degradation Prognostics Based on Switching Kalman Filter Ensemble.

Area of Science:

• Prognostics and Health Management (PHM)
• System Degradation Modeling
• Machine Learning for Predictive Maintenance

Background:

• Accurate system prognostics require determining current health and predicting future degradation.
• Switching models are used to represent system degradation phases, offering advantages in phase identification and handling nonlinearities via piecewise linear approximation.
• Existing methods have limitations, including discretized remaining useful life predictions and insufficient robustness due to single-model reliance.

Purpose of the Study:

• To develop a hybrid ensemble and switching method to overcome limitations in existing prognostic approaches.
• To improve the accuracy and robustness of remaining useful life (RUL) predictions.
• To enable both continuous and discrete predictions for system RUL and degradation phase.

Main Methods:

• A hybrid approach combining ensemble methods with switching methods.
• Implementation of a Switching Kalman Filter (SKF) for classifying linear degradation phases.
• Utilizing appropriate Kalman filters for each phase to predict fault dimension propagation.

Main Results:

• The proposed method achieves both continuous and discrete prediction values for RUL and degradation phase.
• Demonstrated superior accuracy and robustness against noise compared to existing methods on benchmark aeroengine data.
• The SKF effectively detected switching points between different degradation modes.

Conclusions:

• The hybrid ensemble and switching method provides a more accurate and robust solution for system prognostics.
• The SKF is effective in identifying transitions between degradation phases, enhancing prognostic capabilities.
• This framework offers significant improvements for predictive maintenance and system health management.