| description abstract | Civil engineering design problems are inherently complex, characterized by iterative processes, multiple criteria, and time-consuming manual design work. Traditional methods often struggle to rapidly reach optimal designs, lacking guarantees of achieving optimality. With the advent of recent advances in artificial intelligence (AI), this study attempts to answer the research question: How AI algorithms can expedite the civil engineering design process, enhancing efficiency and accuracy in reaching optimal solutions with fewer resources. The research employs a Markov decision process-based AI framework, integrating configuration design and refinement in a unified approach. The methodology begins with the Markov decision-making process to mathematically model the design process, followed by reinforcement learning for automatic design and refinement of solutions. Applied to a planar truss bridge design problem, the AI design agent produced feasible truss designs under various constraints efficiently, demonstrating superior capability and flexibility. The results indicate an average improvement of 12% in accuracy and 88% in computational efficiency over traditional methods. The meaning and significance of the results lie in the innovative integration of Markov decision-making and reinforcement learning into a unified two-stage design framework, significantly advancing the body of knowledge in civil engineering design automation. The speed and accuracy of the AI design agent validate the feasibility of the proposed model and highlight its potential in effectively solving complex civil engineering design problems. The directions for follow-up research are suggested to extend this framework to a wider array of design challenges and to refine the AI agent’s adaptability in more diverse design contexts. | |
