Extended Relevance Vector Machine-Based Remaining Useful Life Prediction for DC-Link Capacitor in High-Speed Train

Area of Science:

• Engineering
• Reliability Engineering
• Machine Learning

Background:

• Remaining Useful Life (RUL) prediction is crucial for component health management.
• Accurate RUL prediction under uncertainty remains a significant challenge.
• Existing methods may suffer from limitations in handling evidence approximation.

Purpose of the Study:

• To enhance the accuracy of RUL prediction in uncertain conditions.
• To develop a robust method for estimating degradation tendency.
• To improve the reliability of component health management.

Main Methods:

• Extending the Relevance Vector Machine (RVM) to a probability manifold.
• Selecting tendency features based on batch samples.
• Building a dynamic multistep regression model to describe uncertainties.
• Estimating degradation tendency for continuous monitoring.
• Utilizing the First Hitting Time (FHT) method for RUL prognostication.

Main Results:

• The proposed RVM extension compensates for evidence approximation weaknesses.
• Accurate estimation of degradation tendency is achieved.
• The dynamic multistep regression model effectively describes uncertainty influences.
• The FHT method provides reliable RUL prognostication based on estimated tendency.

Conclusions:

• The developed approach significantly improves RUL prediction accuracy under uncertainty.
• The method offers a reliable way to monitor component degradation status.
• The study demonstrates the effectiveness of the proposed scheme in a real-world case study involving train traction system capacitors.