A knowledge-graph enhanced large language model-based fault diagnostic reasoning and maintenance decision support pipeline towards industry 5.0

Industry 5.0 highlights the human-machine collaboration and the sustainability of intelligent manufacturing. Under this background, fault diagnosis, as a key technical component, imposes new requirements for efficient human-machine interaction. The ease of use and outstanding natural language processing capabilities of Large Language Models are believed to enhance the efficiency of human-machine interaction in fault diagnosis. But, LLMs usually exhibit limitations in their ability to incorporate new knowledge, the generation of hallucinations, and the transparency, rendering them unusable in the field of fault diagnosis. In this paper, we propose a novel fault diagnostic pipeline enhanced by knowledge graph, termed the Fault Diagnostic Reasoning Knowledge Graph LLM (FDRKG-LLM). This pipeline employs LLMs for complex fault diagnose tasks and construct knowledge graph to enhance the precise reasoning performance of the LLM. The effectiveness of the FDRKG-LLM is evaluated by a self-constructed product fault diagnose database. Experimental results demonstrate that the FDRKG-LLM outperforms existing retrieval-augmented generation models in assisting the analysis of mechanical equipment faults and providing reliable guidance. Hopefully, this research will pave the way for the widespread application of LLM-based solutions in the Industry 5.0.