Conveniently Identify Coils in Inductive Power Transfer System Using Machine Learning

TL;DR

This work tackles the practical problem of identifying coil parameters $L$ and $Q$ for very-high-frequency IPT without bulky measurement devices or disassembly. It introduces a CNN-based approach that takes coil images and the operating frequency as input to predict $L$ and $Q$, trained on a diverse set of coil configurations. The dataset includes variations such as ferromagnetic cores, differing wire gauges, and multiple shapes, with measurements across several frequencies. Experimental results show the method achieving low MSE and a measurable 21.6% identification error at 85 kHz, indicating feasibility and potential for rapid, portable coil parameter identification in manufacturing and field applications.

Abstract

High-frequency inductive power transfer (IPT) has garnered significant attention in recent years due to its long transmission distance and high efficiency. The inductance values L and quality factors Q of the transmitting and receiving coils greatly influence the system's operation. Traditional methods involved impedance analyzers or network analyzers for measurement, which required bulky and costly equipment. Moreover, disassembling it for re-measurement is impractical once the product is packaged. Alternatively, simulation software such as HYSS can serve for the identification. Nevertheless, in the case of very high frequencies, the simulation process consumes a significant amount of time due to the skin and proximity effects. More importantly, obtaining parameters through simulation software becomes impractical when the coil design is more complex. This paper firstly employs a machine learning approach for the identification task. We simply input images of the coils and operating frequency into a well-trained model. This method enables rapid identification of the coil's L and Q values anytime and anywhere, without the need for expensive machinery or coil disassembly.

Figures (9)

• Figure 1: Basic IPT system
• Figure 2: Experimental setup
• Figure 3: Experimental efficiency
• Figure 4: CNN Network
• Figure 5: Proposed architecture based ML