Generative Al for structural design of buildings

Tsinghua University has developed generative artificial intelligence (generative AI)-based design solutions for building structures. Generative AI conducts structural design after learning existing data and knowledge, and the AI-based design creates a novel design paradigm.
They proposed data-mechanics co-driven generative networks, which are composed of the generative network, discriminative network, knowledge evaluator, and mechanics evaluator. The discriminator guides the generator to learn from data features, while the knowledge and mechanics evaluators guide the generator to simultaneously learn mechanical principles and knowledge. Moreover, advanced AI technologies, such as generative adversarial network (GAN), graph neural network (GNN), and Attention, have been successfully combined with the demand for structural design.
They invited engineers to conduct a large number of blind tests to identify whether the structural design was created by humans or AI. The results indicated that about half of the AI designs were judged to be engineers’ designs.
The corresponding design platform has also been successfully integrated into the daily design workflow. They have developed the necessary pre- and post-process tools. These tools are compatible with mainstream design software.