Towards an adaptive POD/SVD surrogate model for aeronautic design
Résumé
A computational methodology is presented to obtain a model reduction of the steady compressible Reynolds-Averaged Navier–Stokes equations with a high-dimensional parameter space. It combines both a reduced-basis method and an adaptive sequential sampling technique. The reduced basis is made of the leading eigenvectors computed by a Singular Value Decomposition of the snapshot span basis. The sampling method uses an a posteriori error estimator combined with a leave-one-out algorithm. This methodology allows a dynamic building of surrogate models for aerodynamic flight data generation and for multidisciplinary design optimization with a small number of full numerical flow simulations. The efficiency of the method is assessed on an analytic test case and a classical aerodynamic transonic flow around a 2D profile.