How can the integration of AI large language models and knowledge graph enhance fault diagnosis? A systematic literature review

The integration of Large Language Models (LLMs) and Knowledge Graphs (KGs) represents an emerging approach to improving fault diagnosis in industrial settings. Traditional fault diagnosis methods—including model-based, signal-based, and rule-based approaches—face persistent challenges in managing complex data, adapting to new failures, and ensuring reasoning accuracy. This systematic literature review evaluates 37 relevant studies to examine how the synergy between LLMs and KGs can mitigate these limitations. The review identifies two primary paradigms: LLM-augmented KGs, which automate the extraction of entities and relationships from unstructured industrial text, and KG-enhanced LLMs (GraphRAG), which ground model reasoning in structured causal paths to reduce hallucinations. While the integrated use of these technologies is a growing research trend, particularly with a peak of 11 papers published in 2025, the field remains in its early stages. Current research is heavily concentrated on rotating machinery, steel manufacturing, and computer numerical control (CNC) equipment, with significant gaps in industry coverage, multilingual support (beyond English and Chinese), and advanced evaluation metrics. Furthermore, many existing systems lack the explainability required for engineers to interpret reasoning routes in high-stakes environments. This study suggests that future research should focus on expanding diagnostic frameworks to specialized industries, improving domain-specific adaptations, and enhancing interpretability through interactive visualization and structured reasoning.