Comparison of optimization algorithms applied to aerodynamic design.

摘要

This thesis investigates the performance and robustness of three gradient-based optimization algorithms. The algorithms are the BFGS quasi-Newton algorithm, which uses the quadratic penalty approach for constraints, the KSOPT algorithm, which uses the Kreisselmeier-Steinhauser function to combine the objective function and constraints, and the SNOPT algorithm, which implements a sequential quadratic programming method for solving constrained optimization problems. These approaches are applied to the two-dimensional Navier-Stokes aerodynamic shape optimization problem. Accurate gradients are computed using the discrete adjoint method. The performance of the optimizers is demonstrated for several design examples, including inverse design, maximization of lift-to-drag ratio, maximization of endurance factor, lift-constrained drag-minimization, and multi-point optimization. The SNOPT optimizer is more robust and efficient than the BFGS and KSOPT optimizers for the majority of test cases presented.