Intelligent data-driven approach for enhancing preliminary resource planning in industrial construction

Abstract Industrial construction is often fast-tracked, with engineering and construction phases progressing simultaneously. In contrast to projects where detailed engineering information is available prior to construction, fast-tracked projects rely on the subjective experience of practitioners to derive preliminary resource plans—often resulting in plans that fail to reflect true project design. Here, a data-driven approach for predicting module classes and preliminary resource requirements from historical resource data of similar modules is proposed. The approach deploys semi-supervised machine learning to categorize construction modules at a micro-level based on key design elements extracted from low level-of-detail 3D models. Using historical data, the approach summarizes resource requirements for each module class, thereby connecting preliminary 3D models to historical resource requirements and enhancing preliminary resource planning in the absence of detailed design information. Notably, the method can incorporate updated design information as it becomes available, ensuring that outputs of the approach remain up-to-date.