Shinji Wakao

This paper proposes a warm-start design optimization method based on potential-shaped latent representations learned by a variational autoencoder (VAE). The latent space is structured so that objective function values correspond to a quadratic potential, concentrating high-performance designs near the origin without explicit latent-space optimization. Candidate designs are generated by sampling the latent space around the origin and subsequently refined using level-set-based optimization. The effectiveness of the proposed method is demonstrated through rotor flux-barrier shape optimization of a synchronous reluctance motor.

The Darwin model for electromagnetic quasistatic (EMQS) field analysis is of particular interest for the analysis of power electronic devices, such as power transformers and wireless power transfer systems. This model is an approach that neglects the second derivative term in time of the displacement current and considers the electrostatic field. This approximation allows conventional techniques for magneto-quasistatic (MQS) field analysis, such as considering nonlinear magnetic material properties and coupled circuit analysis, to be appropriated, and capacitance effects to be included. Several methods have been proposed for frequency-domain analysis. This is because it is difficult to properly derive a formulation of time domain analysis. We have already proposed a potential formulation in frequency domain analysis and succeeded in obtaining appropriate calculation results. In this study, we propose a time-domain analysis of the low-frequency stabilization formulation in the finite element method. The proposed formulation is based on the Coulomb-type gauge condition, which defines no additional variables and redundant variables to improve the convergence characteristics. In addition, the system matrix can be made symmetric for solving by general preconditioners with iterative solvers.

The density method is one of the powerful topology optimization methods of magnetic devices. The density method has the advantage that it has a high degree of freedom of shape expression which results in a high-performance design. On the other hand, it has also the drawback that unsuitable shapes for actually manufacturing are likely to be generated, e.g. checkerboards or grayscale. The purpose of this paper is to develop a method that enables topology optimization suitable for fabrication while taking advantage of the density method. This study proposes a novel topology optimization method that combines convolutional neural network (CNN) as an effective smoothing filter with the density method and apply the method to the shield design with magnetic nonlinearity. This study demonstrated some numerical examples verifying that the proposed method enables to efficiently obtain a smooth and easy-to-manufacture shield shape with high shielding ability. A network architecture suitable as smoothing filter was also exemplified. In the field of magnetic field analysis, very few studies have verified the usefulness of smoothing by using CNN in the topology optimization of magnetic devices. This paper develops a novel topology optimization method that skillfully combines CNN with the nonlinear magnetic field analysis and also clarifies a suitable network architecture that makes it possible to obtain a target device shape that is easy to manufacture while minimizing the objective function value.

This paper aims to further improve the efficiency of multi-objective optimization design of synchronous reluctance motors (SynRMs) using the level set (LS) method, which has the advantage of obtaining a practical shape. The solutions obtained by gradient methods tend to be local ones due to the multi-modality of the objective function, especially when multiple objective functions. A huge number of trial calculations are required to obtain a high-quality and broadly distributed Pareto front. Therefore, it is indispensable to effectively get out of the local solutions in the optimization process with the LS method. The authors propose a novel method appropriately switching multiple objective functions with high independence of sensitivity information. The authors adopt highly independent mathematical expressions for the objective functions of the average torque and torque ripple. In addition, the authors repeatedly perform the optimization while appropriately selecting the sensitivity information of one objective function from multiple ones, which enables the authors to effectively break out of local solutions in the optimization process. The proposed method was applied to the shape optimization of SynRM flux barriers and successfully searched a more extensive and advanced Pareto front in comparison with the conventional method. The proposed method adopts search spaces with mathematical high independence for average torque and torque ripple. In the optimization process, when the solution search is judged to get stuck by several criteria, the search space is alternately switched to effectively get out of local solutions.

We proposed a new multi-objective optimization approach combining both gradient calculation in actual high-dimensional design space and that in a low-dimensional latent space that is an appropriately compressed design space with Auto-Encoder. The proposed approach enables us to quickly and precisely search for more global pareto solution group in multi-objective optimization by using the optimization results in the low-dimensional space as the initial value of the gradient calculation in the actual design space.

In our previous study, we carried out confidence interval estimation of next‐day photovoltaic (PV) output with just‐in‐time modeling hour by hour. When the target reliability is high in particular, the width of the confidence interval becomes large certainly. Therefore, in this paper, based on behavior of the measured PV output in the morning of the targeted day, we predict of PV output transition in the estimated confidence interval in the afternoon. The proposed method, which is consistently based on black box modeling, enables us to easily and effectively narrow the confidence interval width, i.e., sharpness, of PV output. Some numerical examples, which demonstrate the effectiveness of the proposed method, are also reported. © 2020 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.