One-Shot Optimization for the Inverse Design of a Quasi 1-D De Laval Nozzle
Please login to view abstract download link
Modern component design often takes place in virtual environments, which are essential for both modeling and optimizing performance. Optimization involves solving complex Partial Differential Equations (PDEs), and the computational cost can be prohibitive, especially with many design variables. Gradient-based methods are preferred for their convergence efficiency, and the adjoint method is commonly used to reduce gradient evaluation costs. However, adjoint-based optimization can still be expensive. To address this, we propose a one-shot acceleration technique that solves PDEs and the optimization problem simultaneously in a coupled iteration. Using a Newton method for the optimality conditions, the inverse design of a nozzle has been performed and the solution has been achieved in just 6 Newton iterations.
