Comparison of analytical, finite element and neural network methods to study magnetic shielding

Three types of methods are studied to calculate the shielding efficiency of a cylindrical ferromagnetic shield in Fe–Si steel, i.e. an analytical method, a finite element method and a neural network method. The methods take into account nonlinear hysteretic behaviour in the shield. All of the methods are compared with each other and with measurements: the shielding factor is computed and measured for several shield radii, thicknesses, field amplitudes and frequencies. The mean absolute error rates are calculated for AM, FEM and NN to be 9.16%, 6.45% and 5.54%, respectively. The analytical model is very fast and accurate for materials with mild nonlinearity. In case of highly nonlinear material and a strongly non-uniform field distribution in the shield, the modeling of the nonlinear behavior in the analytical model is not accurate. The FEM is very accurate for all considered shielding configurations if the mesh density is well chosen, but its evaluation time is high. The neural network has the disadvantage that it should be trained. It is, however, a fast method with the additional advantage that it can be used without any knowledge of the shield geometry and material properties if the training is based on measurements.