Automating adjoint sensitivity analysis for multidisciplinary models involving partial differential equations
Structural and Multidisciplinary Optimization · 2024-08 · 67 (8)
DOI: 10.1007/s00158-024-03847-2 ↗Kurz gefasst
Fachartikel in der Zeitschrift Structural and Multidisciplinary Optimization, online erschienen am 14.08.2024. Verfasst von Ru Xiang und 5 weiteren Autor:innen. Die Arbeit wurde bisher 18-mal zitiert (Crossref) und nennt 46 Literaturangaben. Gefördert von: National Aeronautics and Space Administration.
Abstract (Originalsprache)
We present a general framework for incorporating partial differential equation (PDE)-based models into gradient-based optimization of multidisciplinary systems by integrating FEniCSx with the recently developed Computational System Design Language (CSDL). CSDL is a domain-specific language designed to facilitate adjoint sensitivity analysis multidisciplinary design optimization (MDO). We use CSDL’s abstractions to link together sub-models representing different disciplines, not all of which are necessarily modeled by PDEs. For the subsystems which are modeled by PDEs, we use FEniCSx to compute partial derivatives of problem residuals, which CSDL can combine with derivatives from other disciplines using the chain rule and the adjoint method. The development of this framework is motivated by the problem of optimizing designs of electric vertical takeoff and landing (eVTOL) aircraft where, due to the relative novelty of this class of vehicle, there is currently a large, unexplored design space. Predicting the performance of eVTOL aircraft requires PDE-based modeling of various sub-problems, including electric motors, composite shell structures, and thermoelasticity and electrochemistry of battery packs. For system-level analysis and optimization, these must be coupled to non-PDE components, such as low-fidelity aerodynamic models, geometric design tools, and lumped-parameter battery and circuit models. In this work, we demonstrate the modeling flexibility and efficiency of this framework through classic optimal control and topology optimization problems with known solutions, and also challenging eVTOL-related applications including shape optimization of an electric motor, and aeroelastic coupling for gust response of an eVTOL wing. Given the generality of this framework, we expect it to facilitate research on a wide range of PDE-constrained MDO problems beyond eVTOL applications.
Förderung
- National Aeronautics and Space Administration
Zitieren
APA
Xiang, R., Schie, S. P. C. V., Scotzniovsky, L., Yan, J., Kamensky, D., & Hwang, J. T. (2024). Automating adjoint sensitivity analysis for multidisciplinary models involving partial differential equations. Structural and Multidisciplinary Optimization, 67(8). https://doi.org/10.1007/s00158-024-03847-2
Vancouver
Xiang R, Schie SPCV, Scotzniovsky L, Yan J, Kamensky D, Hwang JT. Automating adjoint sensitivity analysis for multidisciplinary models involving partial differential equations. Structural and Multidisciplinary Optimization. 2024;67(8). doi:10.1007/s00158-024-03847-2
BibTeX
@article{Xiang2024,
author = {Xiang, Ru and Schie, Sebastiaan P. C. van and Scotzniovsky, Luca and Yan, Jiayao and Kamensky, David and Hwang, John T.},
title = {Automating adjoint sensitivity analysis for multidisciplinary models involving partial differential equations},
journal = {Structural and Multidisciplinary Optimization},
year = {2024},
volume = {67},
number = {8},
doi = {10.1007/s00158-024-03847-2}
}
EUVIMED