Details information of datasets.

With the in-depth development of industrial intelligence, as the core basic component of high-end equipment, the fault diagnosis and health management of rotating machinery has become a key link to ensure the reliability of complex systems. Although the intelligent diagnosis technology based on mechanical vibration signals has made remarkable progress, in complex mechanical systems, it is difficult to comprehensively cover the fault feature space using vibration signal data only.This paper proposes an intelligent diagnosis framework based on a large language model. By empowering the large language model through multimodal data feature fusion and constructing a ternary data system of “raw vibration signals - time-frequency spectrum features - fault knowledge text”, the framework realizes cross-modal joint representation of mechanical fault features and breaks through the bottlenecks of traditional methods, such as insufficient feature extraction capability under complex working conditions and limited cross-scenario generalization. The framework innovatively integrates the deep semantic understanding ability of pre-trained large language models with mechanical fault mechanisms. Through the method of plugging in principle knowledge bases, the model can not only output fault location results but also simultaneously generate interpretable reports including fault cause analysis and maintenance strategy suggestions.The model proposed in this paper has been strictly tested on bearing datasets. Experimental results demonstrate that the model exhibits excellent performance and adaptability in different industrial scenarios.

随着工业智能化的深度发展，旋转机械作为高端装备的核心基础部件，其故障诊断与健康管理已成为保障复杂系统可靠性的关键环节。尽管基于机械振动信号的智能诊断技术已取得显著进展，但在复杂机械系统中，仅依靠振动信号数据难以全面覆盖故障特征空间。本文提出一种基于大语言模型（Large Language Model, LLM）的智能诊断框架：通过多模态数据特征融合赋能大语言模型，并构建"原始振动信号-时频谱特征-故障知识文本"三元数据体系，该框架实现了机械故障特征的跨模态联合表征，突破了传统方法的瓶颈，例如复杂工况下特征提取能力不足、跨场景泛化能力受限等问题。该框架创新性地将预训练大语言模型的深度语义理解能力与机械故障机理相结合，通过接入原理知识库的方式，模型不仅可以输出故障定位结果，还可同步生成包含故障成因分析与维修策略建议的可解释性报告。本文提出的模型已在轴承数据集（bearing datasets）上完成严格测试，实验结果表明，该模型在不同工业场景中均表现出优异的性能与适应性。