Prompt-based automation of building code information transformation for compliance checking
Fan Yang, Jiansong Zhang
Abstract
Transforming building code information into a machine-processable format is essential for automated compliance checking, yet it presents significant challenges. A prompt-based framework was developed to automate the conversion into a logic programming language. Its effectiveness was assessed by testing the framework on 51 requirements from the International Building Code (IBC) 2015, achieving 97.37 % precision and 95.88 % recall at the logic clause level, with only 2 % of the data used for training. Further testing on crash report transformation enhanced efficiency, reducing the average code generation time to approximately 60.8 s, thereby achieving a 27.8 % time savings compared to existing rule-based methods. This paper contributes to the body of knowledge by introducing an effective, versatile, and user-friendly approach to automated building code information transformation, markedly decreasing the reliance on training data , time, and manual efforts. • Developed a prompt-based framework to automate building code transformation for simplifying compliance checks. • Achieved 97.37 % precision and 95.88 % recall in code transformation with only 2 % training data. • Extended the framework to processing crash reports, demonstrating versatility and reducing code generation time by 27.8 %. • Produced a valuable dataset that converts building regulations into a logic programming language. • Combined rule-based precision with machine learning adaptability and user friendliness, reducing training data needs.