Global and Local Sensitivity Analysis Applied to Electric Motor Design by Polynomial Chaos Expansion

The design of electric motors increasingly requires the integration of multi-physics simulations, multi-objective optimization, and uncertainty quantification. In this work, we present a framework for global and local sensitivity analysis of a V-shaped Interior Permanent Magnet (IPM) motor using surrogate modeling and Polynomial Chaos Expansion (PCE). A neural network surrogate is trained on a dataset of high-fidelity finite element simulations to predict torque and temperature as functions of eight design parameters. Global sensitivity analysis is performed via PCE, enabling the computation of Sobol' indices and statistical metrics with negligible computational cost. The surrogate-based PCE results are validated against finite element-based model evaluations, confirming their accuracy and convergence. A local analysis is also performed along the Pareto front, where input parameters are perturbed within a narrow range to estimate confidence intervals around optimal solutions. This enables the quantification of robustness and highlights trade-offs between performance and sensitivity. The proposed approach offers an efficient method for integrating uncertainty-aware decision-making into the early stages of electric motor design.