Research On Positioning/Foreign Obeject Detection For Wireless Charging System Based On Data-Driven Technology

The positioning and foreign object detection technology in electric vehicle wireless charging system is an important guarantee for the efficient,safe and stable operation of the wireless charging system.The distributions of magnetic field on the surface of the coil have different characteristics under three different states of coil alignment,coil offset and introduction of foreign object.In this paper,a method of positioning/foreign object detection based on feature learning of magnetic field data is proposed.The data acquisition of magnetic field on the surface of the coil is realized by building a highly reusable hardware detection system The training datasets are constructed based on the acquired magnetic field data,and the feature extraction of magnetic field data is realized by model leaming.Based on the complete-learning model,accurate coil positioning and foreign object detection are realizedFirstly,the characteristics of magnetic field distributions under different states are theoretically deduced and simulated.It is proved that the characteristics of magnetic field distributions are different under different states,which provides the theoretical basis for the positioning/foreign object detection method based on feature learning of magnetic field data,and proves the feasibility of the methodSecondly,this paper designs the architecture of the positioning/foreign object detection system.The system is divided into two parts,the data acquisition part and the model recognition part.For the data acquisition part,the TMR sensor matrix is used to acquire the data of plane magnetic field.By means of scanning,the TMR sensors of each row/column are gated and switched The multiplexing of signal excitation and processing circuits in data acquisition part is realized,and the cost,volume,complexity and manufacturing difficulty of the system are reduced.For the model recognition part,this paper constructs the training datasets and test datasets on the experimental data acquired from different states.For the tasks of positioning and foreign object detection,a multi-label positioning model based on gradient boosting regression tree(GBRT),a foreign object detection model based on convolutional neural network(CNN)and a multi-task joint model based on feature-extraction layer and share-bottom layer are proposed respectively to learn the features of magnetic field data on the training datasets,predict and evaluate on the test datasets,and verify the effect of the modelsFinally,the hardware and software development of the data acquisition part of the positioning/foreign object detection system are completed,and the wireless charging system prototype is built,based on which data acquisition of the magnetic field is completed.The three models proposed in this paper are used to learn the characteristics of magnetic field data,and the complete-learning models are applied to the positioning/foreign object detection system for the experiments of positioning and foreign object detection.The experimental results show that the system achieves accurate positioning and foreign object detection,which verify the effectiveness and accuracy of the method proposed in this paper.