基于因果路径图卷积神经网络的复杂机电系统故障检测方法  

Causal Paths-based Graph Convolutional Network for Fault Detection in Complex Mechatronic Systems

作　　者：郑舒文 王冲 刘杰 ZHENG Shu-wen;WANG Chong;LIU Jie(School of Reliability and Systems Engineering,Beihang University,Beijing 100191,China)

机构地区：[1]北京航空航天大学可靠性与系统工程学院,北京100191

出　　处：《系统工程》2024年第6期141-147,共7页Systems Engineering

基　　金：国家自然科学基金资助项目(52005027)。

摘　　要：为提高复杂机电系统故障检测的性能和可解释性,本文提出了一种基于因果路径的图卷积神经网络。首先结合经验约束和数据驱动混合的因果发现算法构建监测变量的因果图,用于描述变量间的因果关系。由于故障沿着因果关系传递影响,因此提取因果路径作为图卷积神经网络的输入感受野,并根据从因果路径提取的信息输出系统的故障检测结果。为验证提出方法的有效性,利用采集自高铁制动系统的故障检测真实数据集进行实验。结果表明:相较对比方法,本文提出的故障检测方法在两种性能指标的平均提升超过5%,并且有着较好的模型可解释性和分类泛化能力,在具有高维监测变量的复杂机电系统故障检测中有着良好的应用前景。To improve the performance and interpretability of fault detection in complex mechatronic systems,a graph convolutional network based on causal path(GCN-CP)is proposed in this paper.Firstly,hybrid causal discovery,combining empirical constraints with data-driven causal discovery,is employed to construct causal graph,which describes the causal relationships among monitoring variables.Since faults propagate along the cause-and-effect relationships,the causal paths are extracted as the receptive field for the proposed GCN-CP.The output of the proposed model is the system fault detection result based on information extracted from causal paths.A real case study concerning fault detection in high-speed train braking systems is carried out,to verify the effectiveness of the proposed method.Comparative experiments are carried out against several benchmarks,and the proposed method exhibits over 5%improvement with respect to two evaluation metrics.Additionally,the proposed method also demonstrates desirable interpretability and generalizability,indicating its potential for application in fault detection of complex mechatronic systems with high-dimensional monitoring variables.

关 键 词：故障检测 图卷积神经网络 机电系统 因果发现 高铁制动系统 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]