The paper presents a multi-disciplinary framework for design under uncertainty that leverages surrogate models trained using risk-adaptive statistical learning. We consider multiple competing quantities of interest coming from two different disciplines and describing the hy-drodynamic as well as structural features of a hydrofoil. The surrogates predict superquantile risk (s-risk) of quantities of interest, which adapt to the risk averseness level of a particular design stage. In order to construct accurate surrogate models we leverage two computational frameworks for both hydrodynamic and solid mechanic analysis. Each framework provides data at low- and high-fidelity levels through a multi-resolution approach for the hydrodynamic (RANS at high and low resolution) and a multi-fidelity approach for the solid mechanic predictions (full 3D and a simple beam). The framework is demonstrated through the design of a complex super-cavitating hydrofoil, but the method is generally applicable to the design complex physical system under uncertainty.
展开▼
