Reduction of model dimension in nonlinear finite element approximations of electromagnetic systems

Abstract

A method to reduce the order of finite element (FE) based models of electromagnetic components is presented. In the technique proposed, a full order FE model is transformed into a user specified (or error determined) reduced order system using empirical eigenvectors (EE). The EE method uses an observation of the response of a full model to construct a reduced basis that replicates a full system. Nonlinear attributes resulting from saturation are naturally preserved in the reduction, while the numerical effort required to model the component is greatly reduced. The EE method has been applied to model an iron-core toroidal inductor. A four-fold increase in the speed of computation has been obtained for a linear, fixed time step analysis and a seventy-one fold increase for a nonlinear, variable time step analysis, without observable loss in accuracy.