ScholarWorks@성균관대학교

초록

Electric vehicle propulsion motors operate under diverse conditions, with input currents controlled via both maximum torque per ampere (MTPA) and flux-weakening strategies. These variations result in nonlinear behavior in air-gap electromagnetic force harmonics (AEFHs), which can induce undesirable vibrations. Effective mitigation of electromagnetically induced vibrations requires incorporating both the electromagnetic excitation sources and geometry-dependent mechanical responses. Traditionally, this necessitates AEFH reduction techniques in conjunction with coupled electromagnetic-structural analyses, an approach that entails significant computational cost. To address this challenge, we propose a deep neural network-based framework capable of predicting d-q axis flux linkages, AEFHs, and the stator transfer function (TF) from geometric parameters. Magnetic nonlinearity is handled by training separate models for each current combination. For TF, the models learn the maximum magnitude and the corresponding frequency as functions of geometry. These predictions are subsequently integrated to assess both electromagnetic performance and structural response. The proposed approach facilitates efficient and accurate vibration-oriented optimization across various operating conditions while significantly reducing computational demands.

키워드