Dassault Systèmes Expert and speaker
Abstract
Physical simulations play a crucial role in science-based, data-driven design processes by providing high-accuracy simulation results. However, the high execution times often limit the design exploration capabilities. To retain comprehensive 3D simulation results while reducing executing time, we present an approach utilizing neural networks for 3D interactive design exploration.
Fast executing neural networks are trained as 3D surrogates using parametric design data generated from physical simulations, and then deployed in a collaborative design environment for design exploration, allowing designers to evaluate a high number of parameters within the design space. By delivering high-fidelity results that covers both transient physical responses and 3D fields, this proposed method can accelerate the design iteration process, and enhance the efficiency and effectiveness of science-based product design approaches.
Presentation:
