Artificial intelligence and deep learning have been widely used in recent years to explore the possibility of accelerating computation. However, it has very few applications in magnetic field calculation. This article employs a conditional generative adversarial network (cGAN) to approximate the magnetic field of a coaxial magnetic gear and calculate the magnetic torque through postprocessing. The working principle of magnetic field approximation using cGAN and the training process is introduced in this study. We adopted conditional image-to-image translation technology in cGAN and compared different loss functions and residual structures combinations. Then, we found the best combination, which can accelerate convergence, reduce errors, and improve the generator's performance. Numerical experiments have verified the effectiveness of the proposed cGAN, and the average numerical error can be as small as 1%. At the same time, its speedup ratio is as high as 200 compared to the finite element method.

The demand for fast magnetic field approximation for the optimal design of electromagnetic devices is urgent nowadays. However, due to the lack of a publicly available dataset and the unclear definition of each parameter in the magnetic field dataset, the expansion of data-driven magnetic field approximation is severely limited. This study presents a physics-informed generative adversarial network (PIGAN), as well as a permanent magnet linear synchronous motor (PMLSM)-based magnetic field dataset, for fast magnetic field approximation. It includes the current density, material distribution, electromagnetic material properties, and other parameters of the electric machine. Physics-informed loss functions are utilized in the training process, making the output governed by Maxwell’s equation. Different slot-pole combinations of the PMLSM are involved in the dataset to extend the generalization of PIGAN. Some indicators for the further evaluation of magnetic approximation performance, including image-based metrics and calculation methods for the performance of electric motors, are presented in this study. Some challenges of magnetic field approximation using PIGAN are also discussed. The effectiveness of the physics-informed method is verified by comparing the magnetic field approximation results and the performance analysis results of the PMLSM with FEM, and the speed of PIGAN is approximately 40 times faster than that of FEM, while the accuracy is similar.

In this work, a novel general pattern of assisted flux barriers in an asymmetric V-shape interior permanent magnet (AVIPM) machine is presented. The AVIPM machine has a symmetric permanent magnet (PM) structure and an asymmetric rotor core structure to realize the magnetic-field-shifting (MFS) effect. The general pattern can represent four possible types of assisted flux barriers at different positions on the rotor core and the final structure can be automatically determined by using optimization method. The advantage of the proposed optimization pattern is that the optimal design of assisted flux barriers in V-shape interior permanent magnet (VIPM) machines with high torque and low torque ripple can be generated within a short computing time. The proposed optimization method is applied to improve the structure of a conventional 8-pole 48-slot VIPM machine, which is commonly used for driving electric vehicles (EVs). A non-dominated sorting genetic algorithm II (NSGA-II) is used for the global optimization of both a VIPM machine and an AVIPM machine. The electromagnetic performance is computed using a finite element analysis (FEA). Moreover, a frozen permeability (FP) method is applied to perform an accurate separation of PM torque and reluctance torque. The results exhibit that the peak output torque of the AVIPM machine is increased by 8.7% compared to a conventional VIPM machine with the same PM volume due to the notable MFS effect. In the meantime, the AVIPM machine has a better overload capability due to the larger contribution of reluctance torque. Moreover, the proposed machine has high efficiency of over 97%.

As asymmetric interior permanent magnet synchronous motor (AIPMSM) has excellent performance but complicated topological structure, a novel high-resolution encoding and edge smoothing method is proposed for topology optimization of the asymmetric rotor of interior permanent magnet synchronous motor (IPMSM) in this study. This method aims to solve complex electromagnetic design problems with time-dependent performance through a multi-objective genetic algorithm (MOGA) integrated with a high-resolution encoding and edge smoothing method. The complex structure is represented by a high-resolution image-like matrix and then vectorized by the edge smoothing method. Therefore, the commonly used discrete binary encoded variables related to the finite element (FE) model are replaced with a vectorized topological structure and other control variables. In this sense, high-resolution matrix and edge smoothing methods are used for the first time to represent the rotor topology of AIPMSMs. Compared with the traditional topology optimization method, the proposed method has the advantage of expressing more complex and vectorized topological structures; meanwhile, the obtained performance is accurate and trustworthy using conventional FE simulation. Numerical results show that a stable convergence is achieved with the avoidance of checkerboards and material overlapping. It is shown that the proposed method can find solutions with better performances, in comparison with the reference model.

In this article, a novel pre-processing method is proposed for magnetic field approximation based on neural networks. The process of calculating magnetic fields based on neural networks is presented first. In order to save the training time and reduce the estimation error, an excitation distance function, which is inspired by an analytical formula of magnetic potential, is proposed to integrate the input data, such as the geometry, excitations, and boundaries. This pre-processing or roughly reduces the dimension of the input layer by three quarters, while the training process is accelerated by a more integrated input layer. Preliminary experiments show that the predicted results by the proposed pre-processing method are close to the ground truth, with a significant reduction of computation time compared with the traditional finite element method (FEM).