This paper presents an intelligent system based on a large-scale language model (LLM) for preventive maintenance of industrial machinery. This system goes beyond conventional anomaly detection to provide actionable maintenance recommendations. Developed based on the LAMP framework for numerical data analysis, this system converts bearing vibration frequency analysis data (BPFO, BPFI, BSF, and FTF frequencies) into natural language, enabling high-accuracy anomaly detection with just a few attempts using LLM. Fault types (inner race, outer race, ball/roller, and cage failures) are classified and severity levels assessed. A multi-agent component uses vector embedding and semantic search to process maintenance manuals and access comprehensive procedural knowledge and up-to-date maintenance practices through web searches, providing more accurate and in-depth recommendations. The Gemini model generates structured maintenance recommendations that include immediate actions, inspection checklists, corrective actions, parts requirements, and schedule specifications. Experimental validation on a bearing vibration dataset demonstrates its effectiveness in detecting anomalies and providing context-sensitive maintenance guidance. This system successfully bridges the gap between condition monitoring and actionable maintenance planning, providing intelligent decision support for industrial practitioners. This study advances the application of LLM in industrial maintenance, providing a scalable predictive maintenance framework across a variety of machine components and industries.
