Characterization of the Quality Factor in Spiral Coil Designs for High-Frequency Wireless Power Transfer Systems using Machine Learning

Abstract

This paper presents a machine learning-based char-acterization of the quality (Q) factor in spiral coil designs for wireless power transfer systems operating at MHz frequencies. Due to skin and proximity effects, at such frequencies, it is challenging to estimate the Q factor of the coupling coils, which is a critical parameter to determine the system's efficiency. A three-dimensional (3D) electromagnetic (EM) simulator allows us to precisely analyze the performance of different coil structures. However, the long processing time in the simulator is a bottleneck for quickly optimizing the coil design. To overcome this issue, we here propose a design method with a feed-forward neural network (FNN) to predict the parameters of the spiral coil. The FNN leverages the data set collected via the 3D quasi-static EM field simulator to train a predictor using the stochastic gradient descent algorithm. After optimization, the FNN model estimates the Q factor of the spiral coil without any delay. The proposed algorithm shows an accuracy larger than 96% under an arbitrary structure. Moreover, the proposed coil design method significantly reduces the computation time and hence, the analysis complexity.