Sensorless metal object detection for wireless power transfer using machine learning

Purpose This study aims to realize a sensorless metal object detection (MOD) using machine learning, to prevent the wireless power transfer (WPT) system from the risks of electric discharge and fire accidents caused by foreign metal objects. Design/methodology/approach The data constructed by analyzing the input impedance using the finite element method are used in machine learning. From the loci of the input impedance of systems, the trained neural network (NN), support vector machine and naive Bayes classifier judge if a metal object exists. Then the proposed method is tested by experiments too. Findings In the test using simulated data, all of the three machine learning methods show high accuracy of over 80% for detecting an aluminum cylinder. And in the experimental verifications, the existence of an aluminum cylinder and empty can are successfully identified by a NN. Originality/value This work provides a new sensorless MOD method for WPT using three machine learning methods. And it shows that NNs obtain high accuracy than the others in both simulated and experimental verifications