The utility of simulation-based design is dependent on adequately characterizing the impact of model-form uncertainties on performance metrics. This is especially true in aerodynamic design, where uncertainty in turbulence models is often a limiting factor in the credibility of computational design solutions. In this work, we use data-driven techniques based on field inversion and machine learning to extract a representation of model-form uncertainties. This representation is embedded in a predictive solver and applied in robust aerodynamic design optimization of aircraft engine nozzles. The data is obtained from a few simulations of a higher fidelity model, and uncertainty representations are embedded within a lower fidelity model (eddy viscosity-based Reynolds Averaged Navier—Stokes). The worst-case model-form uncertainty is treated as an interval-based estimate and the robust design optimization approach seeks to minimize this interval. Results from the robust design process are compared to corresponding deterministic design solutions with and without mass constraints.
展开▼
