A hybrid architecture for large-scale system design optimization of PDE-based models
Anugrah Jo Joshy, Jiayao Yan, John Tan Teng Hwang · AIAA SCITECH 2022 Forum · 2022
View Video Presentation: https://doi.org/10.2514/6.2022-1614.vid Large-scale system design optimization has recently started being used in many fields as a tool in the design of new concepts. This has exposed many of its limitations, especially in regard to its usability and computational efficiency. State-of-the-art optimization algorithms can solve problems with tens of thousands of design variables in just hundreds of model evaluations. However, this can be computationally very expensive for complex systems with costly simulations. A hybrid architecture known as SURF, which unifies the full-space and reduced-space architectures, can speed up optimization of computational models of engineering systems involving large number of implicit state variables. This paper will develop a new algorithm based on the SURF architecture to address some of the computational inefficienies experienced in the optimization of large-scale PDE-based models. The new algorithm will be able to automate adaptive selection of hybrid models generated using the SURF architecture in order to extract the best possible efficiency while also ensuring robustness of optimization. We also plan to validate the efficiency improvements on application problems such as nonlinear topology optimization, optimization of nonlinear elastic structures, and optimization of electric motors.