The design of aerospace systems requires the use of computationally intensive numerical solvers. Most often, the designer has access to simulation codes with different levels of fidelity, characterized by different accuracy and computational cost. In addition, certain phenomena manifest a stochastic nature that needs to be accounted for in the design process. Incorporating multiple sources of uncertainty through reliability-based design optimization (RBDO) then becomes a challenging task. Among the potential strategies, decoupled approaches such as SORA iterate between deterministic optimization and reliability analysis. However, when employing high-fidelity solvers, such methods still induce significant computational burdens. A way to further alleviate the computational cost issues is to involve models of fidelity throughout the RBDO problem. In this paper, a Bayesian approach with multifidelity surrogate models is proposed to tackle the optimization problems within SORA. This approach facilitates the incorporation of additional sources of information provided by lower fidelity models. Furthermore, the surrogate models are built in an augmented space allowing to reuse information along the RBDO iterations. The efficiency of the proposed framework is compared to reference approaches on four test cases with increasing complexity, and two aerospace realistic cases concerning the optimization of a solid-propellant rocket booster and a sounding rocket.
Espoeys et al. (2025) studied this question.