On aerodynamic design with a POD surrogate model


Wednesday, 30 November, 2016 - 11:30

Speaker: Valentina Dolci (Politecnico di Torino)

Venue: Room 005, ground floor, SISSA main campus, Via Bonomea 265. Trieste​


Since the last few years, surrogate models are becoming a promising research field for engineering applications. A surrogate, or reduced-order model (ROM), is a mathematical tool able to extract the main features of a more computational demanding high-order model, starting from a reduced set of information. Once the surrogate model is built, it can be used to perform faster analyses of the problem. The fields of optimization and database generation in aerodynamics can be best candidates for the application of surrogate models. In the case of aerodynamic optimization, a cost function should be evaluated several times requiring many CFD simulations in order to achieve sufficient information to identify an optimum target. This operation requires a great amount of computational time and effort and the adoption of a surrogate model replacing the CFD high-order model can be attractive. However the construction of an accurate and robust surrogate model is a delicate process. Particular attention should be paid to the type of surrogate model chosen for a specific problem and to the number and position of the initial set of high-order simulations. In this construction phase there is a constant trade-off between required accuracy and computational effort reduction: the building of the surrogate model must be fast and not computational demanding preserving in the meantime the primary characteristics of the problem. In the present work surrogate models are constructed using the proper orthogonal decomposition technique applied in the parameter space. A reduced snapshot set is used, adopting full and fractional factorial planes together with quadtree distribution for the initial positioning of the snapshots. Response surface methodology is employed to compute the POD coefficients. Surrogate models are applied in the analysis of transonic flows, three-dimensional aircraft configurations and shape optimization of vehicles. A posteriori error estimates have been performed and the models showed good agreement with the CFD reference solution.