This paper presents a tunable terahertz metasurface based on vanadium dioxide (VO 2 ), in which broadband linear-to-linear tunable polarization conversion is achieved by regulating the conductivity of VO 2 . The unit cell consists of five layers arranged sequentially, including a gold backplane, a polyimide dielectric spacer, a thin VO 2 film, a second polyimide layer, and a hybrid metal–VO 2 patterned top layer. When VO 2 is in the insulating state, simulation results demonstrate that the proposed metasurface realize polarization conversion ratio (PCR) exceeds 90% from 3.50 to 8.00 THz under linear-to-linear polarized wave. As the conductivity of VO 2 changes, the polarization conversion response is flexibly tuned from broadband behavior to dual-band and triple-band high-efficiency conversion. The Jones-vector analysis, electric- and magnetic-field distributions, and surface-current distributions show that the broadband response arises from the overlap and coupling of multiple electric and magnetic resonant modes. Owing to its broadband conversion performance and conductivity-controlled tunability, the proposed metasurface shows promising potential for applications in terahertz communications, imaging and wavefront manipulation.
We have investigated the range of validity of an electromagnetically thin model for calculating eddy current losses in a non-magnetic cylindrical conducting shell rotating in an external transverse low-frequency magnetic field. In fact, “thin” means that the shell can be of any thickness but must be thinner or at most comparable to the diffusion skin depth. The reduction in thickness (exclusion of the interior of the shell from the computational domain) is accounted for in the continuity conditions applied to the interface between air regions. Using this model, we derived a very simple closed-form engineering expression for the Joule losses in a shell rotating in a low-frequency magnetic field. This expression can also be used for a stationary shell in a time-harmonic magnetic field and, with proper interpretation of the symbols, also for a shell rotating in a static magnetic field. A comparison with the results obtained using the finite element method and the separation of variables method for real thicknesses of the shell is presented.
Electric motors are applied to many industrial machines. Among them, the demand for interior permanent magnet synchronous motors (IPMSMs) is drastically increasing because IPMSMs are used in EVs, which are becoming popular as a means of mitigating global warming. Therefore, designing a high-power IPMSM is required. In the design of IPMSM, high-stiffness rotor core is also required because it must be protected from damage and deformation of rotor core due to centrifugal force when it rotates at a high speed. To realize a new rotor core design with high-power and -stiffness, the topology optimization is effective from the viewpoint of its high degree of freedom regarding the structure of magnetic core. However, unmanufacturable region (gray scale) tends to appear in a coupled topology optimization problem based on magnetomechanical analysis. In this paper, the magnetomechanical coupled topology optimization of IPMSM for EVs using the coupled strategy of the density method (DM) and level-set method (LSM) is carried out. Furthermore, a sensitivity filter was implemented to enhance the manufacturability of the optimized structure. As a result, high torque, and low von Mises stress structure of IPMSM was obtained.
This study proposes the finite element physics-informed neural networks (FE-PINNs) method , which integrates physics-informed neural networks (PINNs) with the finite element (FE) to estimate static magnetic fields, aiming to accelerate the shape optimization of electromagnetic devices. Conventional PINNs, which directly learn static magnetic field equations involving partial derivatives, often exhibit low solution accuracy even for simple structures and face challenges in regions where the physical characteristics change sharply. To address this issue, the proposed method learns FE equations obtained by discretizing static magnetic field equations. By incorporating FE equations into the loss function during training, the model can predict vector potential distributions consistent with static magnetic field equations across various material distributions. This study employs the perceiver IO-graph message passing (PIO-GMP) network for predicting static magnetic fields. When comparing PIO-GMP with a conventional graph convolutional network (GCN) as the prediction network in FE-PINNs, estimation accuracy improves by up to 99%. Furthermore, the constructed FE-PINNs serve as a surrogate model for finite element method (FEM) in topology optimization of magnetic shields.
