Enhancing engineering risk analysis with knowledge graph-driven retrieval-augmented generation

Tools

Ouyang, Linghan and Li, Haijiang and Liu, Jiucai; Moreno-Rangel, Alejandro and Kumar, Bimal, eds. (2025) Enhancing engineering risk analysis with knowledge graph-driven retrieval-augmented generation. In: EG-ICE 2025. University of Strathclyde Publishing, GBR, pp. 475-484. ISBN 9781914241826 (https://doi.org/10.17868/strath.00093296)

[thumbnail of Ouyang-etal-EG-ICE-2025-Enhancing-engineering-risk-analysis-with-knowledge]
Preview
 Text. Filename: Ouyang-etal-EG-ICE-2025-Enhancing-engineering-risk-analysis-with-knowledge.pdf
Final Published Version
License: Creative Commons Attribution 4.0 logo
Download (2MB)| Preview

Abstract

The increasing complexity and volume of unstructured risk-related data in engineering projects pose significant challenges for timely and accurate risk analysis. While Retrieval-Augmented Generation (RAG) enhances large language models (LLMs) with external knowledge, traditional RAG systems struggle with context fragmentation and cross-chunk reasoning. This paper proposes a Knowledge Graph-enhanced RAG (KG-RAG) framework that integrates structured semantic relationships into the RAG pipeline to improve information retrieval and response generation. By extracting entities and their interrelations from textual risk assessment reports, the system builds a graph-based knowledge base that enables hierarchical summarization and precise risk identification. It supports both global risk summarization and causal chain tracing through a dual-mode retrieval strategy. A case study on the Jiaozhou Bay Second Subsea Tunnel project illustrates the efficacy of KG-RAG in analysing complex engineering risks, outperforming naïve RAG methods in accuracy, traceability, and decision support. The results suggest KG-RAG offers a scalable and intelligent solution for automating engineering risk assessment.

CORE (COnnecting REpositories)