Design Method for Reducing Risk of Electromagnetic Noise Generation in Induction Motors via Machine Learning Inverse Problem

The noise of electric motors can be classified as mechanical noise and electromagnetic noise caused by electromagnetic forces. Mechanical noise depends on the rotation speed, the cooling type and size of the motor, and it is possible to predict the noise level with a certain degree of accuracy. In contrast, electromagnetic noise depends on the electromagnetic design parameters and mechanical properties of the construction, making its prediction difficult due to the complexity of the correlation. Therefore, we examined a method to judge the risk of excessively harsh and auditory electromagnetic noise in induction motors via machine learning (ML) using actual data of electric motors as the training dataset. However, the design of an electric motor requires that design parameters that do not cause excessive electromagnetic noise are determined efficiently. In this study, we examined a method for determining the design parameters to reduce the risk of electromagnetic noise generation, as a machine learning inverse problem by constructing a ML model that uses the target noise value as input and the design parameters as output using about 2,000 actual data of induction motors as training data.