An automatic machine fault identification method using the knowledge graph–embedded large language model
Pengcheng Wu, Xun Mou, Leihao Gong, Haobei Tu, Linqiong Qiu, Bo Yang
Abstract
Abstract Computer numerical control (CNC) machines are prone to various faults during long-term operation, which can compromise production safety. However, accurately identifying fault sources and obtaining rapid troubleshooting solutions remains a significant challenge. To tackle this challenge, this study proposes an automated machining process decision-making system that integrates a knowledge graph and a large language model (LLM). The system first constructs a graph-based fault knowledge representation using BERT-Transformer-CRF for machining knowledge extraction. The developed machining process knowledge graph is then enhanced and endowed with knowledge reasoning capabilities via the large language model. By combining the graph-structured machining process representing form and knowledge inference ability, the proposed system significantly improves fault diagnosis and troubleshooting efficiency. To evaluate its performance, a functional test was conducted, comparing the system with conventional approaches in terms of accuracy, knowledge inference ability, and user-friendliness. Experimental results in practical industrial scenarios demonstrate that the proposed model achieves 97.50% accuracy in fault diagnosis and troubleshooting. Additionally, subjective evaluations indicate high usability, with scores of 9.4 for user-friendliness and 9.1 for knowledge inference ability. These findings highlight the system’s potential to enhance fault troubleshooting in industrial machining processes.