TY - GEN
T1 - Parallelization Strategies for Efficiently Computing CAD-based Sensitivities for Design Optimization
AU - Dannenhoffer, John F.
N1 - Publisher Copyright:
© 2022, American Institute of Aeronautics and Astronautics Inc.. All rights reserved.
PY - 2022
Y1 - 2022
N2 - Multi-disciplinary analysis and optimization (MDAO) has been a long-standing goal in the aerospace community. In order to employ MDAO effectively, one needs to be able to compute the sensitivity of the objective function with respect to the driving parameters in a robust and efficient manner. As models get very large, there is a need to compute these sensitivities in parallel, especially since most optimization methods already employ parallel solvers. Contained herein is a study of two different parallelization strategies for efficiently computing sensitivities. They are compared on a model problem, where the objective is to find the CAD-like design parameters that most-closely match a set of given mass properties. Although quite simple compared with CFD-based optimizations, this model problem allows one to really examine the efficiency and robustness of the parallelization strategies. The results are that for small design changes, a linearized approach to the geometry can be very effective. But for large changes, a non-linear approach, involving rebuilds and computing the sensitivities in parallel with the flow solver, has been found to be the best approach.
AB - Multi-disciplinary analysis and optimization (MDAO) has been a long-standing goal in the aerospace community. In order to employ MDAO effectively, one needs to be able to compute the sensitivity of the objective function with respect to the driving parameters in a robust and efficient manner. As models get very large, there is a need to compute these sensitivities in parallel, especially since most optimization methods already employ parallel solvers. Contained herein is a study of two different parallelization strategies for efficiently computing sensitivities. They are compared on a model problem, where the objective is to find the CAD-like design parameters that most-closely match a set of given mass properties. Although quite simple compared with CFD-based optimizations, this model problem allows one to really examine the efficiency and robustness of the parallelization strategies. The results are that for small design changes, a linearized approach to the geometry can be very effective. But for large changes, a non-linear approach, involving rebuilds and computing the sensitivities in parallel with the flow solver, has been found to be the best approach.
UR - http://www.scopus.com/inward/record.url?scp=85123356766&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85123356766&partnerID=8YFLogxK
U2 - 10.2514/6.2022-0971
DO - 10.2514/6.2022-0971
M3 - Conference contribution
AN - SCOPUS:85123356766
SN - 9781624106316
T3 - AIAA Science and Technology Forum and Exposition, AIAA SciTech Forum 2022
BT - AIAA SciTech Forum 2022
PB - American Institute of Aeronautics and Astronautics Inc, AIAA
T2 - AIAA Science and Technology Forum and Exposition, AIAA SciTech Forum 2022
Y2 - 3 January 2022 through 7 January 2022
ER -