Reinforcement Learning for Accelerated Aerodynamic Shape Optimisation
Researchers have developed a novel reinforcement learning (RL) algorithm designed to accelerate aerodynamic shape optimization. This method employs an actor-critic policy evaluation approach, allowing for the temporal freezing of certain optimization parameters to reduce computational effort. The algorithm aims to improve global optimization speed by using local parameter changes informed by intermediate computational fluid dynamics simulations, provided that local neighborhood estimates are accurate. AI
IMPACT This research could lead to more efficient design processes in fields requiring complex aerodynamic simulations.