A Fast Calculation Method of Electromagnetic Parameters of Superconducting Magnet Based on Generating Convolution Network

摘要

In the process of designing magnets stacked by the second-generation superconducting ReBCO annular plates, the magnetic field distribution under the assumed current distribution can often determine the relevant performance of the magnet. However, when the number of annular superconducting plates stacked along the axis of magnet becomes large, the time cost of electromagnetic analysis of magnet by finite element method (FEM) increased at a geometric level, even can up to many days. In order to improve the calculation speed in this process, this paper proposes a fast calculation method of electromagnetic parameters of superconducting magnet based on generating convolution network. According to the size of the magnet, the generalized graph of the current distribution is established and input to the convolution neural network to obtain the magnetic field distribution of the magnet. When the number of conductor pieces is inconsistent with the number of training pieces, the method of input zeroing and dislocation superposition are used to realize the compatibility of the model. Then, the calculation time of magnetic field distribution of superconducting magnet with different number of rings is measured by this method, and the calculation accuracy of this method is compared with that of finite element method. The results show that the method can complete the calculation faster than the FEM, and the calculation accuracy can meet the engineering requirements.