AI-Based Condition Monitoring and RUL Estimation for DC/DC Converters Deployed on Embedded Systems

Power electronics often fail due to capacitor or semiconductor degradation. Lifetime and reliability analyses based on physics-of-failure models and mission profiles help ensure durability, but deviations in real-world conditions can lead to premature failures or unnecessary replacements. The solution developed in this study detects and identifies specific component degradations in a DC/DC boost converter using spectral analysis of voltage and current signals. A machine learning classifier, trained on real waveforms from a demonstrator, determines which component has reached its end-of-life, achieving over 92.5% accuracy. Digital signal processing, recursive feature elimination, and hyperparameter tuning based on randomized grid search are performed using only input and output waveforms. Beyond impending component failure detection, this method enables remaining useful life estimation for predictive maintenance and promotes sustainability by reducing premature replacements and electronic waste.