This conference paper was submitted for presentation at the NAFEMS World Congress 2025, held in Salzburg, Austria from May 19–22, 2025.

Abstract

œHow can we develop better products faster?' is a critical question confronting manufacturing industries worldwide. Traditional development processes, reliant on repetitive and human'driven design and analysis, often increase costs and prolong time'to'market. To address these challenges, this research presents Deep Generative Design, a paradigm shift from conventional simulation'based design to an AI'driven methodology. The proposed framework is built upon three core AI technologies. First, Generative AI automatically produces large sets of novel design concepts through 3D deep learning. By leveraging historical data while ensuring engineering feasibility, it provides designers and engineers with a broad range of innovative ideas, thereby inspiring fresh solutions. Second, Predictive AI employs 3D deep learning'“based CAE/CAM techniques to evaluate each design'™s engineering performance, manufacturability, production cost, and novelty in real time. This significantly reduces the cost and duration of analysis, helping to minimize the sequential and iterative cycles commonly required for design, simulation, and test. Third, Optimization AI applies deep learning to identify optimal design solutions under specified performance targets and constraints. By simply inputting desired performance metrics, engineers can generate the best'fitting solutions in real time, as well as compare trade'offs across multiple performance indicators. By securing product'specific big data and AI solutions, companies can accelerate digital transformation and efficiently establish data standards for future use. Replacing repetitive tasks in CAE and optimization with AI dramatically shortens development cycles, lowers production costs, and enables engineering teams to focus on more creative, value'adding activities. In addition, AI autonomously generates, evaluates, and optimizes designs, offering significant market opportunities for innovation. Finally, the framework reduces dependence on domain experts by allowing multiple users to access and share the latest AI models via cloud'based platforms. Consequently, anyone can perform end'to'end engineering tasks'”including design, analysis, and optimization'”without specialized assistance, ultimately paving the way for faster, more cost'effective, and higher'quality product development. This integrated Deep Generative Design framework has been validated across a broad range of industries'”including mobility, electronics, heavy industries, and robotics'”demonstrating its robust applicability. In this presentation, we focus on several mobility'related examples, specifically in automotive wheel and brake design generation and vehicle crash performance prediction. First, we introduce an AI system that, given a user'specified reference image and style via text input, can generate a multitude of 3D wheel designs for vehicles. It then predicts each design'™s stiffness and strength, recommending the most promising solutions to designers. Second, for the brake caliper, we describe an AI approach that learns from a limited set of design samples to produce large'scale synthetic data, which in turn is used to accurately predict stiffness. Finally, we address vehicle crash scenarios: by training on video test results, the AI can predict outcomes of small'overlap and side'pole impact tests using only the vehicle'™s geometric information, as well as propose optimized design solutions. These examples illustrate how the Deep Generative Design framework can be seamlessly applied to real'world engineering challenges, underscoring its potential to transform the entire product development process.