改进图神经网络的自动扶梯故障诊断方法研  

Study on Fault Diagnosis of Escalator by Improved Graph Neural Network

作　　者：连志祥 LIAN Zhixiang(China Railway SIYUAN Survey And Design Group Co.,LTD.,Wuhan 430063)

机构地区：[1]中铁第四勘察设计院集团有限公司数智化部,武汉430063

出　　处：《铁道勘测与设计》2025年第1期105-110,共6页Railway Survey and Design

摘　　要：针对自动扶梯分布广泛、运行时间长且故障诊断耗时久的特点,提出了一种改进图神经网络的自动扶梯故障诊断方法。整个方法包含数据预处理、本体图构建、特征提取、图分类四个部分。并在特征提取阶段提出残差图神经网络,以增强特征的丰富程度,同时使用图神经网络生成节点向量,代替以往的均值向量,以最大程度保留特征图中每个节点的信息。将上述方法在郑州地铁二号线中的247条自动扶梯故障数据上进行训练和测试。实验结果表明,所提出的方法在故障检测任务上的Top1准确率为74.5%,Top3准确率为82.4%,可以准确诊断出故障位置。最后,在故障的维修策略推荐任务上再次验证,实验表明,此方法的Top1准确率为66%,Top3准确率为78%,可以推荐合适的维修策略。An improved graph neural network-based fault diagnosis method for escalators is proposed,addressing the escalators'extensive distribution,prolonged operation,and lengthy fault diagnosis.The method comprises four key stages:data preprocessing,ontology graph construction,feature extraction,and graph classification.During the feature extraction phase,a residual graph neural network is introduced to enrich the features,and node vectors are generated using graph neural networks,replacing traditional mean vectors,ensuring maximum retention of node-specific information within the feature map.The method was trained and tested on 247 escalator fault data from Zhengzhou Metro Line 2.Experimental results revealed a Top1 accuracy of 74.5%and Top3 accuracy of 82.4%for fault detection,enabling precise fault location diagnosis.Subsequently,the method demonstrated a Top1 accuracy of 66%and Top3 accuracy of 78%in verifying recommended maintenance strategies for fault conditions.

关 键 词：自动扶梯 图神经网络 故障诊断 维修策略推荐 

分 类 号：TH2[机械工程—机械制造及自动化]