Abstract: To address the challenge of integrating cross-scale time series data and document knowledge in intelligent operation and maintenance of bridge cranes,a Dual-System Collaborative Framework(DSCF) based on long and short term timescales was proposed,which employed large language models as reasoning engines for intelligent agents to construct based on the “dual-system”theory in cognitive science.On one hand,the lightweight“System-1”monitored sensor streams such as vibration,current,voltage and temperature in real-time at short timescales to detect anomalies.On the other hand,“System-2”leveraged domain-specific knowledge graphs and document knowledge to perform global reasoning and deep decision-making at longer timescales.To bridge the gap between short-term anomaly detection and long-term complex fault causal inference,a semantic event grid was introduced to map sensor signal segments into interpretable language descriptions,which were then integrated with industry standards,maintenance cases and other domain-specific prior knowledge to achieve multi-modal,multi-timescale information fusion.Experiments using bridge crane maintenance as a case study demonstrated that compared to single-mode or loosely collaborative multi-agent approaches,DSCF improved fault detection accuracy,the interpretability of anomaly causes and the rationality of operation and maintenance decisions.These results showcased the feasibility and potential of the proposed method in industrial operation and maintenance scenarios.

Key words: large language models, equipment operation and maintenance, knowledge graph, dual-system theory

摘要： 针对行车(桥式起重机)智能运维中跨尺度时间序列与文档知识的融合难题,提出了一种基于长短时间尺度的双系统智能体协作框架(DSCF)。该框架采用大语言模型作为智能体的推理引擎,基于认知科学中的“双系统”理论构建。一方面,轻量级的“System-1”在短时间尺度上对振动、电流、电压、温度等传感器流进行实时监测与异常捕捉;另一方面,“System-2”基于领域知识图谱与文档知识,在长时间尺度上构建全局推理与深度决策。为在短时突发的异常检测与长期复杂故障因果推断间实现衔接,提出了语义事件网格将传感器信号片段映射为可解释的语言描述,再结合行业标准及维护案例等领域先验知识实现多模态、多时间尺度的信息融合。实验表明,相比单一模式或松散协作的多智能体方案,DSCF在故障检测的准确度、异常原因的可解释性及运维决策的合理性方面均有提升,从而展示了所提方法在工业运维的可行性与潜力。

关键词: 大语言模型, 设备运维, 知识图谱, 双系统理论