Uncertainty Propagation With Model Form Error Estimation Using Surrogate Models for Airfoil Aerodynamics

Computational simulations inherently contain errors and uncertainties which must be addressed through verification and validation processes to justify their use in engineering applications. Model validation and predictive capability assessment of an aerodynamic tool requires quantification and propagation of all relevant components of uncertainty ranging from parametric input over numerical to experimental measurement uncertainties. In particular, the quantification of a computational model’s inability to represent the real world system is critical and becomes even more difficult at possible operating conditions for which no experimental data for model validation is available. Three different metrics for model validation are analyzed herein. Gaussian Process Regression is used to interpolate and extrapolate the estimated model form uncertainty to achieve also predictive capabilities. An efficient surrogate-based Monte Carlo sampling framework for propagation of mixed uncertainties is successfully applied to provide total uncertainty estimates. However, especially the generalization of model form uncertainty remains challenging. All capabilities are demonstrated for the 2nd AIAA UQ challenge problem using the DLR Surrogate Modeling for Aero-Data Toolbox in python.