A surrogate model assisted with a subdomain model for surface-mounted permanent-magnet machine

Linear subdomain analytical models allow general modeling of permanent-magnet (PM) motors, but fail to model the magnetic saturation effect well. On the other hand, surrogate models allow precise modeling of saturation. However, the bottleneck of the traditional surrogate model is the size of the training dataset because of the large design parameters space of PM motors. This paper provides a surrogate model assisted with a subdomain model (SMASM). It has the advantage of the precision of the surrogate model and generality of the subdomain model. The SMASM can reduce the size of the training dataset tremendously (from 1000k to 0.6k), by motor scaling and dividing the training dataset into linear and nonlinear parts. Compared to the subdomain model, the SMASM has a similar speed and higher precision when the motor works at a high saturation level. 2187 motors were tested. The average errors of the subdomain model and SMASM are about 24.3% and 0.4%, respectively. Their maximum errors are about 46.0% and 2.9%, respectively.