基于建筑信息模型数据驱动的铁路设备运维多模态知识图谱构建

Construction of a Multi-Modal Knowledge Graph for Railway Equipment Operation and Maintenance Based on Building Information Model Data-Driven Approach

铁路信号设备是保障行车安全、提高运输效率的核心装备,加强信号设备智能运维是降低铁路运行风险的必要基础保障。目前,针对我国基于建筑信息模型(BIM)的智能运维平台存在不能精准映射各设备的行为规律和相互之间互馈作用的机理,须同时依靠经验知识进行推断等问题。首先构建了铁路设备运维文本知识图谱;其次构建卷积神经网络(CNN)-团组图卷积神经网络(cgGCN)模型对BIM图像模态数据进行处理,完成对20种铁路信号设备零件图信息的标注,实验结果表明模型准确率达到95.38%,精确率和召回率的调和平均值F1达到95.58%;最后将BIM图像信息以视觉模态嵌入运维文本知识图谱,利用Neo4j图数据库实现多模态知识图谱可视化展示,从而精准映射各信号设备相互之间互馈作用的机理,为后续现场铁路运维人员实施安全管理和运维决策提供在线服务和指导。

Railway signal equipment is essential for ensuring traffic safety and improving transportation efficiency.Strengthening the intelligence operation and maintenance of signal equipment is essential to mitigate the risks associated with railway operations.Currently,the intelligence operation and maintenance platform based on building information model(BIM)in China is unable to accurately depict the behavior and mutual feedback mechanism of each device,thus relying on experiential knowledge for inference.To address this issue,initially,the knowledge graph was constructed using the text related to the operation and maintenance of railway equipment;Subsequently,a convolutional neural networks-clique group graph convolutional neural networks(CNN-cgGCN)model was developed to process BIM image modal data and annotate the information of 20 specific railway signal equipment part drawings.The experimental results show that the accuracy of the model reaches 95.38%,and the harmonic mean F1 of precision and recall reaches 95.58%;Finally,BIM image information is integrated into the visual knowledge graph of operation and maintenance text.This multi-modal knowledge graph is then visualized using the Neo4j graph database,so as to accurately map the mechanism of mutual feedback between signal equipment,and offer online services and guidance to on-site railway operation and maintenance personnel,facilitating safety management and operational decision-making.