Multi-object aerodynamic design optimization using deep reinforcement learning

Aerodynamic design optimization is a key aspect in aircraft design. The further evolution of advanced derivatives requires powerful toolbox. Reinforcement learning (RL) tool but has rarely been utilized the aerodynamic It can potentially obtain results similar to those human designer, by accumulating experience from training. In this work, popular RL method called proximal policy (PPO) proposed investigate multi-object optimization. By observing performances different airfoils, PPO updates reasonable generate optimal airfoils single step. Pareto problem with constraints, only 15% computational time non-dominated sorted genetic algorithm (II) achieve same accuracy. testing show that agent learns ∼4.3%–10.1% improvements performance compared baseline.