미사일 형상 설계 효율화를 위한 변이형 오토인코더 기반 설계 공간 재설정

초록

Aerodynamic shape optimization is often hindered by high-dimensional design spaces containing large infeasible regions, which complicates the search for optimal solutions. This study proposes a methodology to redesign the design space for 3D missile geometry using a variational autoencoder (VAE) to overcome these challenges. A continuous, low-dimensional latent space that implicitly embeds the complex feasibility constraints is learned by training a VAE exclusively on a dataset of feasible shapes. The results demonstrate that this redesigned space significantly improves the generation rate of feasible designs compared to the original space. This study concludes that redesigning the design space with a VAE is a promising strategy to enhance the efficiency and stability of 3D shape optimization by transforming a complex, constrained problem into a more tractable, unconstrained alternative.

키워드