Hybrid non-intrusive local/global approaches for robust optimization leveraging surrogate models

This work extends non-intrusive local/global coupling methods to multiscale robust design optimization, building upon and further developing the framework introduced in Karaouni et al. (2025). The approach enables efficient global optimization while explicitly accounting for local variabilities at different scales. A hybrid surrogate model is constructed, combining kriging for the global design variables and Polynomial Chaos Expansion to represent local uncertainties. To enhance efficiency, a two-level multifidelity surrogate model is also employed to accelerate robust optimization. Overall, the proposed approach fully exploits the non-intrusive local/global coupling algorithm to generate low-cost multifidelity datasets accounting for local uncertainties. This surrogate framework allows for cost-effective uncertainty quantification and robust optimization, which is implemented using the NSGA-II algorithm to efficiently explore the design space. The methodology is illustrated on 2D examples sensitive to local perturbations, highlighting its potential for non-intrusive hybrid robust design applications.