Convolutional Neural Network (CNN) based Planar Inductor Evaluation and Optimization
Xiaoyan Liu, Mengxuan Wei, Maohang Qiu, Shuai Yang, Dong Cao, Xiaofeng Lyu, Yanchao Li
Abstract
Magnetic component as one of the most lossy and bulky components in power electronic converters has been researched on optimization through calculation, experimental and FEM simulation. However, the traditional methods are normally time-consuming or inaccurate. A novel method that combined FEM simulation and convolutional neural network (CNN) is discussed in this paper, which can predict the inductance and core loss efficiently and accurately. Experimental result shows the accuracy of CNN prediction. Based on the CNN inductor inductance and loss prediction, a novel optimization method is presented which can comprehensively and quickly provide the optimization result considering power loss and power density.