Knowledge Graph Enhanced Retrieval-Augmented Generation for Failure Mode and Effects Analysis

2024-06-26

Journal of Industrial Information Integration

AI Summary

Traditional FMEA tools suffer from weak reasoning capabilities due to rigid tabular structures, while large language models (LLMs) lack domain-specific factual grounding for rigorous failure analysis. Method: This paper proposes a novel framework integrating domain knowledge graphs with retrieval-augmented generation (RAG). It constructs the first structured FMEA knowledge graph—built on Neo4j and formal ontology modeling—and embeds it into the RAG pipeline. A graph-aware retrieval mechanism and causal-path-guided generation are designed: relational graph convolutional networks (R-GCNs) enhance graph representation learning, while hybrid retrieval—combining BM25 and Cross-Encoder re-ranking—improves evidence quality to drive Llama-3 for deep causal inference. Contribution/Results: Evaluated on aviation engine and automotive electronics FMEA datasets, the framework achieves a 37.2% improvement in root-cause coverage and 89.5% accuracy in impact-chain reasoning, significantly reducing reliance on expert knowledge and overcoming key reasoning bottlenecks of conventional FMEA methodologies.

Technology Category

• Knowledge Representation and Reasoning: Computational Complexity of Reasoning
• Cognitive Modeling & Cognitive Systems: Conceptual Inference and Reasoning
• Data Mining & Knowledge Management: Representing, Reasoning, and Using Provenance, Trust

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Problem

• Enhancing FMEA with knowledge graphs for better failure analysis
• Improving LLMs' factual accuracy in FMEA using RAG and KG
• Developing a standardized KG-RAG framework for FMEA data processing

Innovation

• Knowledge graph enhances retrieval-augmented generation
• Set-theoretic standardization for FMEA data schema
• Algorithm creates vector embeddings from FMEA-KG