Application of Deep Learning in Motor Vibration and Noise Suppression Based on Negative Magnetostrictive Effect

The suppression of vibration and noise is one of the critical problems in the optimal design of the motor. The vibration and noise caused by the magnetostrictive effect can not be ignored. In this paper, the vibration and noise caused by the magnetostrictive effect are suppressed by drilling holes in the stator and filling the holes with negative magnetostrictive material. Firstly, the electromagnetic–mechanical coupling model is established. By traversing all the drilling positions by finite element analysis (FEA), the stress value is calculated, and the drilling position corresponding to the minimum local stress is determined. It is found that the resource consumed by FEA is expensive, and the calculation time is prolonged. Deep learning is introduced as an alternative model to calculate the hole’s position and size. The Pix2Pix, Pix2PixHD, CycleGAN, and StarGAN are experimented with based on the PyTorch environment. Through the experiment, it is found that deep learning has an absolute advantage over the FEA in time which ensures a high accuracy rate. Finally, an experimental platform for measuring the strain of the motor stator is designed. According to the best positions and apertures of holes are obtained by the simulation and the actual situation, we drill holes in the motor stator and fill the holes with Ni, a negative magnetostrictive material. The result shows that the stator’s local stress is reduced by about 16%. The vibration and noise caused by the magnetostrictive effect are suppressed effectively.