Design of a Spiral Coil for High-Frequency Wireless Power Transfer Systems Using Machine Learning

This article presents a novel design method based on machine learning (ML) for a spiral coil in a wireless power transfer (WPT) system operating at high frequencies. While a MHz frequency operation provides high power density for battery-powered applications, optimizing the spiral coil design becomes challenging due to the complex electromagnetic analysis, such as skin and proximity effects. Even though a 3-D electromagnetic simulator provides a practical analysis of different coil structures, it cannot quickly optimize the coil design due to its computing time. Therefore, an ML-based method is first proposed to estimate the
$Q$
factor of a spiral coil, a critical parameter to determine the efficiency of a WPT system. Toward this end, a feed-forward neural network is trained using around
$20 \times 10^{3}$
data samples collected by using a 3-D quasi-static electromagnetic field simulator. It is shown that this method is effective; that is, it ensures an accuracy of up to 96%. Then, a spiral coil design method leveraging the designed ML-based
$Q$
factor estimation is proposed. This method offers high performance (the intersection over union metric takes values up to 70%) and significant computation time savings (at least five orders of magnitude), compared to commonly adopted software simulators. Finally, the effectiveness of the proposed method is verified by the actual fabrication of several spiral coils.