Active Learning - CFD Integrated Surrogate-based Framework for Shape Optimization of LTA Systems

Lighter-than-Air systems including airships are characterized by heightened design parameters compared to conventional lifting bodies. The aerodynamic shape optimization of these systems is complicated from time as well as computational requirements. Unlike, traditional optimization schemes involving the random selection of diverse data points within the pre-determined design space and performing multiple numerical or experimental evaluations, current framework makes use of an automated active learning - CFD integrated surrogate model, coupled with Particle Swarm Optimizer (PSO) to achieve optimal shape of an airship design for drag mitigation. The proposed active learning technique utilizes sampling techniques that consider both the design and output space to ask meaningful unlabeled queries for improved understanding of the design space. These unlabeled queries are then labeled using automated numerical simulations. The model actively trains on these strategically selected samples during each iteration such that higher accuracy is obtained with a small number of training instances. The surrogate model developed within this framework demonstrates a mean absolute percentage error of less than 3%, showcasing superior performance compared to existing sampling techniques. This paper provides the methodology and results of fully automated shape optimization of an airship hull profile, which significantly reduces the computational time and cost. This framework is shown to produce optimal design solutions for a typical airship hull profile, while displaying potential application for more complex airship configurations to carry out design optimisation.