A cost-effective machine learning-based sensorless method for metallic object detection in wireless power transfer systems

Reliable detection of metallic foreign objects is essential to ensure the safety and operational efficiency of wireless power transfer (WPT) systems, particularly in electric mobility charging environments where sensor integration may be constrained. Traditional detection techniques, such as thermal or magnetic sensors, often increase hardware complexity, cost, and calibration burden. This paper introduces a sensorless foreign object detection approach based on real-time efficiency variation analysis, eliminating the need for additional sensors or auxiliary coils. A decision tree classifier is trained using experimental data and enhanced through Synthetic Minority Over-sampling Technique-based balancing and k-fold cross-validation to ensure robustness under various interference scenarios. The proposed method was achieving an F1-score of 0.81 across multiple metallic object positions and sizes. This cost-effective, measurement-driven technique provides a practical and interpretable solution for anomaly detection in WPT systems operating under stable and controllable conditions, offering a reliable foundation for further advancements toward robust performance in more complex and dynamic real-world applications. • A sensorless foreign object detection method is proposed, using efficiency variation without additional hardware to reduce cost and complexity. • The decision tree model combined with SMOTE improves classification accuracy under imbalanced object position and size data. • A practical resonant wireless charging prototype is implemented to validate robust detection across multiple object sizes and load conditions.