An image-to-image translation model for ultra-fast performance prediction of permanent magnet synchronous machines

Recent advancements in artificial intelligence (AI) and deep learning (DL) have shown great potential in accelerating computational tasks. Notably, image-to-image translation (I2I) has gained significant attention due to its broad applications in computer vision and image processing problems. This paper presents a data-driven I2I model based on conditional generative adversarial networks (cGAN) to predict the performance of an interior permanent magnet synchronous machine (IPMSM). The generator of the cGAN can plot tangential and radial magnetic flux density maps from the input cross-sectional images of an IPMSM with varying geometric features, dimensions, pole-slot configurations and excitations. Additionally, a comparison is made between traditional cGAN networks and a modified multidomain version that incorporates a physics-informed loss function. Model test shows that the proposed method significantly reduces computation time. For a single design, the improvement of evaluation speed is approximately 90.23% compared to the FEA simulation. Besides, the multidomain evaluative model can reduce training time consumption whilst improving the prediction accuracy. Experimental validation indicates that the predicted magnetic flux density maps have an average error of less than 0.13% compared to FEA results. Furthermore, post-processing the generated images enables the calculation of the torque, flux linkage and induced voltage of the IPMSM. The accuracy of performance calculation using the I2I model is also verified against data obtained from FEA software.