This talk presents the various methods based on AI, in a broad sense, and data-driven methods that make EM simulations and optimizations more effective.
As a first topic, the speaker will introduce the motor design based on topology optimization and its acceleration using deep learning. Then, the speaker will focus on the surrogate models such as neural networks and response surface methods that can evaluate the machine properties much faster than the finite element method. Some case examples of the surrogate methods will be presented showing that they actually make the optimal computations much faster. Finally, the speaker will talk about the data-driven methods in which the big data obtained through EM simulations are effectively used to reduce the size of the finite element equations. This approach also makes it possible to effectively evaluate various motor properties such as loss and torque ripples through the behavior model.

The conventional design method has a problem that it cannot cope with advanced design requirements of the rotor core for the motor.
We have tried to explore the best rotor core design applying the topology optimization analysis technology of JMAG.