Parameter Shift Prediction of Planar Transformer Based on Bi-LSTM Algorithm

Abstract

The reliability issues of magnetic elements become more and more prominent with the wide-range application of high-power-density power electronics. Normally, high-frequency planar transformers take up above 30% of the weight and volume of the converter. They suffer various reliability stresses, such as high operating temperature and high-frequency voltages, which can lead to parameter degradations and even failure during operation. To achieve a reliability-oriented design and minimal lifetime maintenance cost of the high-power-density converter, the lifetime prediction of planar magnetics is essential. This paper proposes a method to predict the parameter of planar transformers under thermal reliability stress using a deep learning algorithm. A deep learning neural network is established using the existing parameter shift data in the accelerated aging test. Then, based on the Bi-LSTM model, the future parameter shift of the planar transformer is predicted. Finally, multiple methods are compared to show the advantage of the proposed method. Compared with the conventional curve fitting method, the deep learning algorithm is more suitable for lifetime prediction in terms of filtering data, considering the weight factor, and predicting future change trends.