A 3D preform design method based on a generative artificial intelligence algorithm

This study introduces a novel 3D preform design method that leverages generative artificial intelligence to optimise complex forging geometries. The 3D preform shape is simplified by expressing it in a two-dimensional manner based on the geometric features of the forged shape. A beta-variational autoencoder ($\beta$-VAE) serves as the generative model, paired with a surrogate model using deep neural networks (DNN) to explore and optimise preform geometries efficiently. The initial training dataset for the generative model is created using isosurfaces derived from Laplace’s equation. A multi-objective optimisation framework is developed to minimise forging load and flash volume and mitigate underfill and lap defects. Preforms were designed using this method and validated through forging experiments on two target geometries: a brake calliper and an electric vehicle (EV) manifold. The forging results of the optimised preforms were compared with traditional isosurface methods, demonstrating that the proposed method reduces forging load and flash volume while preventing underfill and lap defects. This integrated approach combines advanced computational techniques with practical validation, offering a systematic and adaptable solution for the challenges associated with 3D preform design in forging processes.