MoDiBrace: multimodal diffusion model for generating internal bracing systems in deep excavations

Internal bracing in deep excavations (IB) is widely deployed in large-scale infrastructure, residential construction, and public building projects, constituting a significant subdomain of structural design. Accelerating the IB design process remains a pressing challenge. With the rapid progress of artificial intelligence, generative approaches to structural design have achieved notable breakthroughs; however, existing models predominantly target shear walls and relatively simple frame systems and thus struggle to handle IB, which is characterized by large scale, numerous constraints, and complex configurations. To address this gap, this study establishes the first multimodal dataset for IB and, within a diffusion-model framework, proposes and trains a generative model for internal bracing (IB). The model employs multimodal conditional inputs, enabling controllable generation with respect to structural scale and bracing types. The outputs are further vectorized to interface directly with professional structural analysis software, thereby improving overall design efficiency.