Efficient adjoint methods for steady and unsteady flow optimization
Date
2025-05
Authors
Fang, Lean
Major Professor
Advisor
He, Ping
Hu, Hui
Sharma, Anupam
Abdelkhalik, Ossama
Passalacqua, Alberto
Committee Member
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Many emerging aerospace systems, such as urban air mobility aircraft and hypersonic re-entry vehicles, present a major challenge to the traditional design process due to their complexity. Design optimization with high-fidelity models, such as computational fluid dynamics (CFD), can significantly speed up the design process. The adjoint method efficiently computes gradients for systems with many inputs and has been widely used for large-scale gradient-based optimization in CFD-based design problems. However, significant gaps still exist, such as efficient adjoint methods for unsteady simulations, and my Ph.D. research aims to address these existing gaps. We first develop a duality-preserving (DP) adjoint formulation for segregated steady-state Navier-Stokes (NS) solvers. The DP-adjoint solver leads to an almost identical convergence rate as the primal solver and is advantageous for ensuring the optimization’s robustness. We then develop a PIMPLE-Krylov adjoint solver that enables field inversion machine learning (FIML) augmented turbulence modeling of time-resolved unsteady flow. The FIML-trained model accurately predicts the spatial-temporal variations of unsteady flow fields and exhibits reasonably good prediction accuracy for flows and geometries outside of the training parameter space. Finally, we develop a fully implicit Runge-Kutta (FIRK) method to speed up segregated unsteady flow simulations and adjoint computations. The proposed FIRK-PIMPLE method significantly reduces the flow simulation runtime while maintaining accuracy and has the potential to make unsteady large-scale high-fidelity design optimization computationally more feasible.
Series Number
Journal Issue
Is Version Of
Versions
Series
Academic or Administrative Unit
Aerospace Engineering
Type
article