Fault Prediction for DC-DC Power Converters Based on Sample Entropy and LSTM

As an important component of the power supply system in industrial production, the power converters often operate continuously under high thermal and electrical stress for a long time. Once severe degradation occurs, it will affect the production efficiency and product quality of the entire assembly line. Therefore, in order to improve the reliability and availability of power converters, a fault prediction method based on a combination of time series signal decomposition and deep learning is proposed in this article, which can address the problem of low prediction accuracy caused by the nonlinear trend of circuit performance degradation. Firstly, the performance degradation test platform was built for power converters under thermal stress and energized working modes, while the output voltage is collected at regular intervals. Afterwards, the output voltage signal should be preprocessed and the sample entropy is used as its fault feature parameter. Then, the sample entropy time series are decomposed by CEEMDAN to extract the monotonic trend component. Finally, the fault prediction has been completed by each decomposed component using LSTM algorithm. The experimental results show that the proposed method obtains high prediction accuracy which verity its feasibility and effectiveness.