Bayesian Surrogate-Based Calibration of Finite Element Models for Motor Performance Analysis

Abstract

In the realm of electric motor design and performance evaluation, the integration of computational simulations with experimental observations is vital to ensure accurate predictions and efficient designs. Finite Element (FE) models have long been utilized to simulate the physical behavior of motors under various operating conditions. However, the fidelity of these models is often limited by uncertainties in model parameters, measurement noise, and simplifications inherent in numerical formulations. This paper introduces a Bayesian surrogate-based calibration framework aimed at enhancing the accuracy of FE models for motor performance analysis. By leveraging Gaussian Process (GP) surrogates and Bayesian inference, the proposed methodology efficiently aligns simulation predictions with experimental data while quantifying the uncertainties involved. We perform a detailed case study on a brushless DC motor, showcasing the effectiveness of the approach through parameter sensitivity analysis, calibration accuracy, and predictive reliability. Experimental validation confirms the framework’s ability to reduce model discrepancies significantly, thus promoting more robust and interpretable motor performance predictions.