Using artificial intelligence in evaluating fire safety design of construction projects

No Thumbnail Available

URL

Journal Title

Journal ISSN

Volume Title

School of Engineering | Master's thesis

Date

2024-10-28

Department

Major/Subject

International Design Business Management

Mcode

Degree programme

Master's Programme in International Design Business Management

Language

en

Pages

98

Series

Abstract

Manual design evaluation in civil engineering has been found inefficient and error-prone, prompting the exploration of artificial intelligence (AI)-driven solutions , including large language models (LLM), to enhance accuracy and streamline the design evaluation process. Previous work showed that LLM, such as GPT, significantly enhances natural language processing tasks, enabling the translation of human language instructions into computable logic and performing reasoning tasks across various domains. However, there is a research gap in understanding how LLM can be leveraged to automate compliance analysis in evaluating fire safety design, particularly in processing complex, domain-specific documents that include multidimensional data formats such as tables and drawings. As such, this thesis evaluated the performance of GPT-4 in the design evaluation process for fire safety design in building projects, including rule extraction, design information extraction, and compliance analysis. The results show that GPT-4 proficiently extracts rules and design information from structured texts and simple tables but encounters difficulties with complex table formats and image analysis tasks. The customised GPT model integrated with a knowledge graph shows improved accuracy in compliance analysis because of the reduction in hallucination issues. This thesis recommends different prompting methods and data structure approaches in regulatory and design documents to enhance the accuracy and efficiency of the models in conducting design evaluation tasks. It also suggests future research into developing knowledge graph-based LLM models for a precise and reliable design evaluation in fire and other engineering fields.

Description

Supervisor

Seppänen, Olli

Thesis advisor

Nyqvist, Roope

Keywords

artificial intelligence, large language model, GPT-4, fire safety, civil engineering, construction, information extraction, compliance analysis, prompt engineering, knowledge graph, customised GPT model

Other note

Citation