AI-Assisted PDF Extraction and Rule-Based Production Time Estimation in Structural Steel Assembly Using Historical Production Data

Abstract

Offsite construction improves quality control, accelerates project completion, and minimizes disruptions at the construction site compared with traditional onsite construction methods. Within this paradigm, steel fabrication plays an essential role by enabling structural elements to be produced under controlled conditions, thereby increasing precision and supporting a more efficient assembly process. Despite these advantages, inaccurate production time estimation in steel fabrication plants remains a persistent challenge, contributing to cost overruns, project delays, suboptimal resource allocation, and difficulties in short-term production planning. A primary driver of this inaccuracy is an interoperability gap: the detailed parametric data required for reliable time estimation is typically locked within unstructured design files, such as 2D shop drawing PDFs, making it inaccessible for automated prediction models. To bridge this gap, the present study proposes a structured, design-driven workflow that extracts and leverages latent information from shop drawings to estimate structural steel assembly times. The methodology operates in two sequential stages. First, an AI-assisted extraction framework parses unstructured PDF data into structured inputs, isolating key assembly variables including member sectional dimensions, weight, crane handling requirements, and the quantity of attached components. Second, these inputs are integrated into a rule-based computational model that calculates production time using historical productivity rates and logic-driven algorithms. By creating a direct pipeline between static design documentation and dynamic production planning, this approach establishes a practical foundation for early-stage estimation and look-ahead scheduling. A proof of concept using a representative shop drawing confirms the workflow's feasibility. The findings demonstrate that this method provides a scalable, consistent, and practically implementable framework for estimating production time, laying the groundwork for advanced scheduling and decision-support systems in steel fabrication.