A Data-Driven Inverse Flux Map-Based on Finite Element Analysis for High-Fidelity Induction Motor Simulation Model

Abstract

This article proposes a data-driven inverse flux map for induction motors (IM) using a deep neural network (DNN). The proposed method enables the direct mapping of stator and rotor currents from flux linkages, which is essential for implementing flux map-based IM model, derived from finite element analysis (FEA). Thanks to this flux map-based IM model, fast and accurate IM simulation combined with inverter circuit models and control algorithms is possible. In the proposed method, the DNN is employed as a universal function approximator to model the inverse of the nonlinear multivariable vector function. The training data is obtained from static FEA. To enhance learning performance and modeling accuracy, the proposed method applies a whitening transformation and incorporates a tangent activation function. The proposed inverse flux map is validated based on the characteristics of the inverse function. The overall flux map-based IM model, integrated with the proposed inverse map, is further validated against time-stepping FEA (TS-FEA) and experimental results using two commercial IMs under various operating conditions.