121st NIA CFD Seminar: Tangent and Adjoint Problems in Partially Converged Flows
Date: Tuesday, August 20, 2019
Time: 11:00 a.m. – noon (EDT)
Room: NIA, Rm137
Speaker: Emmett Padway
Link: http://nia-mediasite.nianet.org/NIAMediasite100/Catalog/Full/fe54023273ef446084620d8a1a25ea5821
Speakers Bio: Emmett did his undergraduate work in aeronautical engineering at McGill University, where he worked as a research assistant for Prof. Siva Nadarajah in the McGill Computational Aerodynamics Group. He began his Ph.D. at the University of Wyoming supervised by Prof. Dimitri Mavriplis in the fall of 2015. In 2016, he was awarded a NASA Aeronautical Graduate Scholarship/Fellowship through the NASA ASTAR program. His research in mechanical engineering is focused on adjoint method development with an application towards MDO and AMR.
Abstract: As computers and algorithms have developed, the field of CFD has expanded to include a larger role for Aerodynamic/Multidisciplinary Shape Optimization and Adaptive Mesh Refinement. Both these applications solve first the primal problem (or the governing equations), and then the adjoint problem, derived off a zero residual condition, to obtain the objective sensitivities or output-based error estimate respectively. As the field of CFD has moved to more difficult problems: higher-order formulations, blunt geometries, or time-accurate simulations, convergence of the governing equations to a zero residual condition has become difficult or impractical to achieve. Despite these limitations, the use of the adjoint technique has continued unabated even in cases where the zero residual constraint has not been fulfilled. These systems are highly sensitive to the state at which they are linearized and can provide inaccurate sensitivity calculations or error estimations, negatively impacting design optimization and/or mesh refinement results. This work presents the derivation and application of the tangent and adjoint problems based off the linearization of the primal solver, which allows for sensitivity calculations and error estimates in partially converged flows.