基于KVAE-OCCA的质量相关故障检测方法及应用  

Quality-related Fault Detection Method and Application Based on KVAE-OCCA

作　　者：金雨婷 侍洪波[1,2] 吕晓龙 谭帅[1,2] 宋冰[1,2] JIN Yu-ting;SHI Hong-bo;Lü Xiao-long;TAN Shuai;SONG Bing(School of Information Science and Engineering,East China University of Science and Technology,Shanghai 200237,China;Key Laboratory of Smart Manufacturing in Energy Chemical Process,Ministry of Education,East China University of Science and Technology,Shanghai 200237,China;Shandong Wendeng Pumped Storage Co.Ltd.,Weihai 264200,China)

机构地区：[1]华东理工大学信息科学与工程学院,上海200237 [2]华东理工大学能源化工过程智能制造教育部重点实验室,上海200237 [3]山东文登抽水蓄能有限公司,山东威海264200

出　　处：《控制工程》2022年第2期348-355,共8页Control Engineering of China

基　　金：国家自然科学基金资助项目(62073140,62073141);上海市青年科技启明星计划项目(21QA1401800);上海市自然科学基金资助项目(19ZR1473200)。

摘　　要：为了解决非线性过程质量相关故障检测问题,提出了一种名为关键变量自编码器-正交典型相关分析(KVAE-OCCA)的方法。首先,为了挑选出与质量变量具有相关性的过程变量,计算过程变量和质量变量的互信息,选择具有较大互信息的过程变量。然后,利用自编码器对选择出的过程变量进行无监督学习,实现特征提取和降维。其次,利用正交典型相关分析方法建立质量相关故障检测模型,通过对系数矩阵奇异值分解得到质量相关和无关投影矩阵,构建统计量并估计控制限。最后,将提出的方法在典型测试案例上进行测试,以说明所提方法的有效性。In order to solve quality-related fault detection problem for the nonlinear process, a method called key variables auto encoder-orthogonal canonical correlation analysis(KVAE-OCCA) is proposed. Firstly, in order to select the process variables which have correlation with the quality variables, the mutual information between the process variables and the quality variables is calculated, and the process variables which have larger mutual information are selected. Then, unsupervised learning of selected process variables is realized by auto encoder(AE) to extract features and reduce dimensions. Secondly, the quality-related fault detection model is established by using the orthogonal canonical correlation analysis method, and the quality-related and quality-unrelated projection matrices are obtained by singular value decomposition of the coefficient matrix.Then, the statistics are constructed and the control limits are estimated. Finally, the proposed method is tested on a typical testing case to demonstrate the effectiveness.

关 键 词：故障检测 典型相关分析 自编码器 质量相关 关键变量挑选 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]