Abstract

Machine parameters such as flux linkages and inductances play a key role in achieving optimal torque control of synchronous machines (SMs). However, it is challenging to identify these parameters online based on the SM model, which has complex and nonlinear characteristics. A fully connected feedforward neural network (NN) is a promising candidate owing to its capability to approximate complex nonlinear functions. Therefore, this study proposes an online physics-informed learning framework for identifying the parameters of SMs using an NN. The proposed method enables the NN-modeled flux linkages and inductances to be learned online in compliance with the governing physical laws of SMs. Consequently, the NN can more effectively capture the nonlinear characteristics of SMs within the constraints imposed by these governing physical laws. The learned parameters can be employed as online estimators and applied to online MTPA control. As a result, the effectiveness of the proposed method is validated through simulations conducted on a 35-kW interior permanent magnet synchronous motor (IPMSM) drive.

Additional Comments