Metal Foreign Object Detection in ICPT System via Multi-Parameter Information Fusion: Based on Transfer Dictionary Learning

Abstract

Aiming at the reliability challenge of metal foreign object detection (FOD) in inductive coupled power transfer (ICPT) systems, a fault classification strategy based on incremental transfer dual-dictionary learning is first proposed. The method first constructs a safety dictionary and a base foreign object dictionary, and introduces a reconstruction error ratio as a decision criterion to achieve fault diagnosis of metal foreign objects. Subsequently, to further enhance generalization capability, incremental transfer learning is integrated to improve the dictionary model. This involves dynamically updating the foreign object dictionary using small samples of unknown metals, with atom addition and fine-tuning to quickly adapt to new foreign object features. Finally, the trained model is evaluated. Experimental results demonstrate that the proposed method achieves a recall rate of over 95% for unfamiliar metal foreign objects after transfer, verifying the effectiveness of the detection approach. This method retains the advantages of unsupervised learning while requiring no detection coils, significantly enhancing its adaptability to complex application scenarios.