Search

In this study, we propose a fast topology optimization (TO) method based on a deep neural network (DNN) that predicts the current-dependent motor torque characteristics using its cross-sectional image. The trained DNN is shown to provide the current condition that provides the maximum torque under the assumed motor control method. The proposed method helps perform TO with a reduced number of field computations while maintaining a high search capability. [ABSTRACT FROM AUTHOR]

This study presents a novel two-step optimization method that incorporates explainable neural networks into topology optimization. The deep neural network (DNN) is trained to infer the torque performance from the input image of the motor cross section. The sensitive region that has a significant influence on the average torque is extracted using gradient-weighted class activation mapping (Grad-CAM) constructed from the DNN. Then, the optimization with respect to the torque ripple is performed only in the incentive region with little influence on the average torque. The proposed method is shown to increase the average torque of an interior permanent magnet (IPM) motor by 14% and reduce the torque ripple by 79% compared with the original model. [ABSTRACT FROM AUTHOR]

The computational cost of topology optimization based on the stochastic algorithm is shown to be greatly reduced by deep learning. In the learning phase, the cross-sectional image of an interior permanent magnet motor, represented in RGB, is used to train a convolutional neural network (CNN) to infer the torque properties. In the optimization phase, all the individuals are approximately evaluated by the trained CNN, while finite element analysis for accurate evaluation is performed only for a limited number of individuals. It is numerically shown that the computational cost for the topology optimization can be reduced without the loss of optimization quality. [ABSTRACT FROM AUTHOR]

This paper presents the fast topology optimization methods for rotating machines based on deep learning. The cross-sectional image of electric motors and their performances obtained during a multi-objective topology optimization based on the finite-element method and genetic algorithm (GA) is used for training of the convolutional neural network (CNN). Two different approaches are proposed: 1) CNN trained by preliminary optimization with a small population for GA is used for the main optimization with a large population and 2) CNN is used for screening of torque performances in the optimization with respect to the motor efficiency. [ABSTRACT FROM AUTHOR]

This paper presents a new topology optimization of interior permanent magnet (IPM) motors using the genetic algorithm with aid of the deep leaning. The data composed of the rotor shape of an IPM motor and its performance, obtained by a prior topology optimization process, is input to a convolutional neural network (CNN). After the learning process, CNN is shown to provide fairly accurate estimate of the motor performance. During the posterior topology optimization, the finite element analysis (FEA) is carried out only for the limited number of individuals; probability that FEA is performed increases with the motor performance evaluated by CNN. It is shown that the computing time is reduced to about 1/10 without deterioration of the optimization performance with aid of the deep learning. [ABSTRACT FROM AUTHOR]

This paper presents a new topology optimization method based on basis functions for design of rotating machines. In this method, the core shape of a given rotating machine is represented by the linear combination of the basis functions. The shape is then freely deformed by changing the weighting coefficients to the basis functions to find the optimal shape. The proposed method is compared with the conventional shape optimization based on polygon morphing. The former is shown to outperform the latter in both interior permanent magnet and synchronous motor models. [ABSTRACT FROM AUTHOR]