Generative Design of Structural Forms Using Deep Generative Models: A New Paradigm in Structural Engineering

This study explores the application of deep generative models, specifically Conditional Generative Adversarial Networks (cGANs) and Diffusion Models, for automated generation of optimized structural forms in engineering design. A comprehensive dataset of parametric and simulation-based designs, containing geometric configurations and performance metrics, was created to train the models. Both cGAN and Diffusion Models were conditioned on key design parameters such as span, material type, and maximum allowable displacement. The diffusion model demonstrated superior performance compared to cGAN in terms of training stability, output diversity, and accuracy in adhering to design constraints. Structural performance evaluation revealed that diffusion-based designs had ۲۶% lower displacement and ۱۱% higher material efficiency than those generated by cGANs. Additionally, diffusion models showed improved stress distribution and better overall structural utilization. The study also introduced a feedback loop between the generative model and structural analysis software, allowing for iterative refinement of the generated forms. The results highlighted that diffusion models produced more structurally efficient and material-conscious designs while maintaining high levels of diversity and novelty. The diffusion model’s ability to generalize and maintain design coherence in unfamiliar scenarios further supports its potential in real-world applications. This research establishes that deep generative models, particularly diffusion models, are a promising tool for AI-assisted structural design, with potential applications in parametric design, early-stage conceptualization, and performance-driven design tasks.

Keywords:

Generative Design , Diffusion Models , Structural Optimization , Deep Learning in Structural Engineering , Performance-based AI Design