Data-driven Leakage Inductance Modeling of Common Mode Chokes

The accurate leakage inductance modeling of common mode chokes (CMCs) can reduce trial-and-error filter debugging efforts and help avoid the saturation issue caused by the leakage flux. However, the analytical leakage inductance model is difficult to derive due to the complex leakage flux distribution. This paper presents a data-driven approach to model the CMC leakage inductance. A large amount of the leakage inductance data is collected by sweeping the selected input variables for the 3D finite element method (FEM) simulation. Then the data is trained by Artificial Neutral Network (ANN) to regress the nonlinear relationship between the leakage inductance and selected input variables. To verify the proposed method, core ZW4310TC is selected to model the relationship between the leakage inductance and winding parameters. Three CMCs with core ZW4310TC were built with different winding parameters and their leakage inductances were measured to verify the model. Compared with a previous analytical model, the error was reduced from 42.9% to 1.3% at the case where the previous model has the worst accuracy.