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

Abstract:

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.