基于最大信息系数和深度残差图卷积的工业过程故障诊断方法  

Fault diagnosis method in industrial processes based on maximal information coefficient and depth residual graph convolution

作　　者：任佳[1,3] 孙思宇 鲍克 REN Jia;SUN Si-yu;BAO Ke(School of Information Science and Engineering,Zhejiang Sci-Tech University,Hangzhou 310018,China;Zhejiang Electronic Information Products Inspection and Research Institute,Hangzhou 310007,China;Key Laboratory of Intelligent Manufacturing Quality Big Data Tracing and Analysis of Zhejiang Province,China Jiliang University,Hangzhou 310018,China)

机构地区：[1]浙江理工大学信息科学与工程学院,浙江杭州310018 [2]浙江省电子信息产品检验研究院,浙江杭州310007 [3]中国计量大学浙江省智能制造质量大数据溯源与应用重点实验室,浙江杭州310018

出　　处：《高校化学工程学报》2023年第1期111-119,共9页Journal of Chemical Engineering of Chinese Universities

基　　金：国家自然科学基金(51876194);浙江省公益技术研究项目(LGG20F030007);浙江省电子信息产品检验研究院(浙江省信息安全重点实验室)开放课题;浙江省智能制造质量大数据溯源与应用重点实验室开放课题(ZNZZSZ-CJLU2022-01)。

摘　　要：针对工业数据非线性、时变性、时空特征耦合的特点,提出一种基于最大信息系数和残差图卷积网络的工业过程故障诊断算法(MIC-RGCN)。引入最大信息系数(MIC)挖掘变量之间的相关关系,将高维变量之间的相关信息转换为空间距离信息,构建相关性矩阵作为图卷积层的邻接矩阵输入;构建改进的深度残差图卷积网络(GCN)模型对数据的时空特征进行深度融合提取并精准分类。在田纳西-伊斯曼过程和三相流过程数据集上将该算法与4种典型机器学习和深度学习算法进行对比测试。实验结果表明,该算法有效地提升了故障诊断的准确率。Aiming at the nonlinear, time-varying, and spatiotemporal characteristics of industrial data, a fault diagnosis algorithm(MIC-RGCN) for industrial processes was proposed based on maximal information coefficient and residual graph convolutional network. The maximal information coefficient(MIC) was introduced to mine the correlation between variables, the correlation information between high-dimensional variables was converted into spatial distance information, so as to construct a correlation matrix as the adjacency matrix of graph convolutional layers. An improved deep residual graph convolutional network(GCN)was designed to perform deep fusion extraction and accurate classification of the spatiotemporal features in data.The algorithm was compared with four typical machine learning and deep learning algorithms on Tennessee Eastman process and three-phase flow process datasets. The results showed that the algorithm improves the accuracy of fault diagnosis effectively.

关 键 词：故障诊断 最大信息系数 图卷积网络 田纳西-伊斯曼过程 三相流过程 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]