A Physics-Informed Gaussian Process Regression-Based Meta Model for Rapid Characterization of Permanent Magnet Synchronous Machines

Optimizing an Interior Permanent Magnet Synchronous Machine (IPM) requires evaluating multiple working points for each candidate design. In case the design domain has many dimensions, multiple working points evaluations would require an impractical number of finite element method (FEM)-based simulations. This study proposes a novel strategy to build a meta-model to reduce the number of FEM-based simulations for a given optimization process. The study proposes a novel, physics-informed meta-model based on Gaussian Process Regression (GPR) aiming for rapid characterization of any given machine design. The meta-model uses an adapted version of Posterior Standard Deviation (PSD) to allow for an exact and detailed adaptive sampling strategy. The results show that the proposed meta-model presents a data-efficient approach capable of computing performance parameters with low error. Additionally, the characterization from the proposed meta-model agrees with the experimental data.