Multi-agent coordination of data-driven and physics-based models for automated design of ultra-high-performance concrete beams

Efficient and effective design of structures plays essential roles in the construction industry. Advancements in information technologies and data science have improved efficiency, while the reliability of data-driven models remains limited. To overcome the challenges, this paper presents a data-physics hybrid framework integrating data-driven and physics-based models for automated structural design. The data component comprises a large language model and a genetic optimizer, and the physics component is a finite element model. The large language model coordinates the framework, supports human–computer interactions, and integrates finite element modelling that provides reliable predictions of structural behavior. The optimizer then explores the optimal design solutions via an evolutionary process. The framework has been implemented into the design of ultra-high-performance concrete beams, reducing human effort by 88 % versus manual design and 75 % versus traditional optimization methods. Results demonstrate effective coordination between data-driven and physics-based models for combined efficiency and reliability.