Removing planning constraints is crucial for ensuring a reliable construction plan and workflow. However, essential information related to these constraints is often scattered and fragmented, leading to the creation of information silos that hinder efficient decision-making. This research aims to develop a framework that integrates disjointed planning constraint attributes, thereby facilitating accurate and reliable access to constraint-related information.
This study develops a constraint knowledge graph to organize and integrate constraint attributes with building information modeling (BIM) objects. A knowledge graph-enhanced retrieval-augmented generation (graph RAG) framework was developed to enrich large language models’ (LLMs) knowledge in responding to project constraint-related queries. The framework’s effectiveness was validated using a real-world building project with 1,122 BIM objects and 37 planning constraints.
Evaluation results demonstrate that the constraint knowledge graph effectively organizes and represents planning constraints information. The Graph RAG framework correctly answered all 16 testing queries, significantly outperforming traditional RAG in delivering accurate and reliable responses. A 40.6% improvement in correctness score and a 28.3% improvement in ROUGE score compared with baseline RAG were found during end-to-end performance evaluation.
This study is the first to apply a knowledge graph for organizing construction planning constraint attributes alongside BIM objects. It innovatively utilizes a knowledge graph to enhance LLMs performance in answering constraint domain-specific queries. This research offers a practical tool for improving construction managers’ awareness of planning constraints through a conversational interface. The developed framework is generalizable, allowing for expansion and representing extensive construction knowledge that facilitates knowledge-informed decision-making.
