Multidisciplinary robust design optimization under combined parameter and surrogate model uncertainties: a new computational framework for sparse samples and extreme scenarios

Multidisciplinary robust design optimization (MRDO) demonstrates potential in engineering design under uncertainty. Existing MRDO methods primarily address parameter uncertainty, with limited research on the combined effects of parameter and model uncertainties. Additionally, effective robustness evaluation in sparse data is lacking. This work presents a tailored MRDO framework for sparse samples, incorporating the impact of extreme cases. The framework employs Kriging as a surrogate model and uses Monte Carlo simulation sampling to quantify the combined uncertainties of parameter and surrogate model. A linear moment method is presented to replace conventional moment estimation for robustness evaluation. The study also analyzes the effects of varying sample sizes and the presence of extreme scenarios on MRDO results. To validate the proposed method, three case studies are conducted with a mathematical example, an energy efficiency optimization problem for a gear hobbing machine, and the design optimization of a battery thermal management system for high-rate charge-discharge applications.