Enhancing accuracy and security in BIM-based construction cost management leveraging AI and blockchain technologies

Construction projects have been suffering from cost overrun problems that lead to stakeholder disputes and potential failures in project deliveries. This highlights the critical role of effective construction cost management in ensuring the project adheres to a pre-agreed feasible cost framework. Nevertheless, traditional construction cost management methods based on 2D drawings have been considered inefficient and error-prone. In response, Building Information Modeling (BIM) offers a more automated and reliable approach. However, challenges persist in crucial cost-related activities in BIM-based construction cost management, including Quantity Take-off (QTO) results incompliant with measurement standards, element classes inconsistent with Bill of Quantities (BQ) item classification systems, inaccurate and inefficient construction cost estimation, and insecure construction cost information management in the digital era. To tackle these problems, this research develops an integratedframework to improve the accuracy and security of cost-related deliverables from design to construction stages comprehensively.Firstly, a data-driven BIM model auditing approach based on Knowledge Graph (KG) is proposed to ensure BIM models have consistent modeling styles and sufficient semantic information for code-compliant BIM-based QTO. A BIM-KG data model incorporating QTO information requirements on modeling styles and semantic information is established, based on which a transformation mechanism is developed to convert BIM data into BIM-KG representations. A KG embedding model is improved to train computable embeddings for the BIM-KG representations. Automatic mechanisms are then designed to leverage the embeddings for identifying BIM model mistakes against the QTO requirements.Secondly, an end-to-end BIM element classification approach based on Multi-Modal Deep Learning (MMDL) is proposed to classify BIM elements in consistency with the BQ item classification system for accurate BQ generation. Multi-modal (i.e., graphical and non graphical) element features are transformed from BIM models. Given numerous BIM element features, a feature selection algorithm, Hierarchical Sequential Forward Selection (HSFS), is designed to determine relevant ones for classification automatically. Afterward, an MMDL model for multi-modal feature fusion is developed to classify BIM elements into BQ item categories. Thirdly, an end-to-end construction cost prediction method based on hypergraph deep learning is developed to predict the actual costs of construction projects accurately and efficiently. A hypergraph formulation is proposed to represent construction cost factors and the interrelationships among them. Based on the formulated hypergraph, a hypergraph deep learning model is developed and deployed to predict construction costs. Then, the model training outcomes are leveraged to interpret the importance of cost factors for better understanding of cost patterns.