This master’s thesis examines the potential for automating the preparation of the measurement book in construction using large language models (LLMs). Quantity surveying papers are essential for recording completed work and installed materials from the bill of quantities, enabling the monitoring of task execution and contractor billing. Traditional manual entry in Excel is time-consuming and involves repetitive tasks, leading to employee overload. The thesis first introduces LLMs, their features, their applicability in construction, and prompt engineering, which enables effective interaction with the models. The practical section is divided into three phases. The first phase tests the LLMs’ understanding of bills of quantities and evaluates different prompting techniques. In the second phase, prompts are designed to record items on the quantity surveying papers using two methods: one with an intermediate JSON file and one directly. The third phase focuses on optimising the prompts for the accurate preparation of the measurement book and on assessing the accuracy of the item and attribute records. Results indicate that LLMs can recognise items well, but due to variations in the structure of bills of quantities, the accuracy of item recording is limited, making the approach presented in this thesis not directly applicable in practice. Attribute records show high accuracy, suggesting that standardising the structure of bills of quantities and further LLM development could enable effective automation of construction documentation preparation. Automation using LLMs could bring significant time and cost savings in practice and simplify the preparation and use of documentation in the construction industry.
|
