Prediction of Parameter Drifts in Planar Transformers Based on Deep Learning

As power electronic converters evolve towards higher frequencies, smaller volumes, and higher power densities, the operating electromagnetic and thermal environments of magnetic components gradually deteriorate, leading to increasingly prominent reliability issues. To achieve reliability-first design and minimize maintenance costs for high-power-density converter systems, a more precise reliability analysis of magnetic components is imperative. This study employs accelerated aging experiments to acquire time-series data of magnetic components in planar transformers. The empirical mode decomposition (EMD) method is utilized to process the time-series data, and kernel principal component analysis (KPCA) is adopted to reduce the dimensionality of the data. Subsequently, the reduced-dimensional data is input into a bidirectional long short-term memory (Bi-LSTM) network to predict subsequent changes in the magnetic components. Predicted failure probabilities are then deduced through failure equations. Comparative analysis with alternative methods demonstrates the feasibility of reliability prediction for magnetic components using a Bi-LSTM network grounded in empirical mode decomposition and kernel principal component analysis.