Inverse Design of Brake Seal Groove Geometry Using Conditional Variational Autoencoders

초록

Inverse design using deep learning enables rapid generation of design candidates that satisfy target performance without expensive iterative optimization. In automotive brake systems, seal groove geometry strongly affects Apparent Piston Travel (APT), which determines pedal feel and responsiveness. This paper proposes a conditional variational autoencoder (CVAE)-based inverse design framework that generates diverse brake seal groove geometries for specified APT targets. Seal shapes are represented as 208× 208 binary images, and the CVAE learns a low-dimensional latent space conditioned on APT performance. Experiments using topology-optimized seal datasets demonstrate that the proposed framework generates physically plausible geometries satisfying target APT values (R² = 0.8, RMSE = 0.08 ) while providing rich design diversity.

키워드