# Knowledge Management for Automobile Failure Analysis Using Graph RAG

**Authors:** Yuta Ojima, Hiroki Sakaji, Tadashi Nakamura, Hiroaki Sakata, Kazuya Seki, Yuu Teshigawara, Masami Yamashita, Kazuhiro Aoyama  
**arXiv:** [2411.19539](https://arxiv.org/abs/2411.19539)  
**Date:** 2024-11-29  
**Summary page:** https://www.summarizepaper.com/en/paper/2411.19539v1/

## Key points

- Knowledge management system tailored for automobile failure analysis using RAG, LLMs, and KGs
- Importance of transferring failure analysis knowledge from experienced to new engineers in the automotive industry
- Challenges faced by novices in analyzing complex failure events occurring in a cascading manner
- Value of knowledge graphs in representing failure events due to their ability to describe semantic relationships and structure information
- Difficulty inexperienced engineers face in extracting and comprehending sub-graphs from KGs due to vast amounts of information
- Introduction of Graph RAG as a variant merging LLMs and KGs for enhanced knowledge management
- Obstacles in employing the Graph RAG framework with existing knowledge graphs specific to automobile failures
- Focus on refining the Graph RAG pipeline for pre-existing knowledge graphs to address challenges
- Utilization of an original Q&A dataset showcasing significant improvements with an average 157.6% enhancement in ROUGE F1 scores compared to conventional methods
- Efficacy of proposed approach in facilitating automobile failure analysis processes within the automotive industry

## Summary

This paper by Yuta Ojima, Hiroki Sakaji, Tadashi Nakamura, Hiroaki Sakata, Kazuya Seki, Yuu Teshigawara, Masami Yamashita, and Kazuhiro Aoyama introduces a knowledge management system tailored for automobile failure analysis utilizing retrieval-augmented generation (RAG) with large language models (LLMs) and knowledge graphs (KGs). The automotive industry faces a pressing need to transfer failure analysis knowledge from seasoned engineers to younger counterparts. However, the complex nature of failure events occurring in a cascading manner poses challenges for novices in analyzing them effectively. Knowledge graphs prove valuable in representing failure events due to their ability to describe semantic relationships and structure information by illustrating component relationships. Despite their effectiveness, extracting and comprehending sub-graphs from KGs remains arduous for inexperienced engineers due to the vast amount of information contained within them. The paper delves into the emerging interest in Graph RAG—a variant of RAG that merges LLMs and KGs for enhanced knowledge management. Nonetheless, employing the current Graph RAG framework with an existing knowledge graph specific to automobile failures presents several obstacles as generating executable queries for a non-LLM constructed knowledge graph database proves challenging. To tackle this issue head-on, the authors concentrate on refining the Graph RAG pipeline for pre-existing knowledge graphs. Through the utilization of an original Q&A dataset, the proposed method showcases significant improvements with an average 157.6% enhancement in ROUGE F1 scores compared to conventional methods. The findings underscore the efficacy of the proposed approach in facilitating automobile failure analysis processes. This research is slated for publication at the 2024 IEEE International Conference on Big Data (BigData) and spans 7 pages with 6 figures included. The comprehensive exploration conducted by these authors sheds light on innovative strategies leveraging Graph RAG technology to streamline knowledge transfer and enhance analytical capabilities within the automotive industry's realm of failure analysis.