基于子空间混合相似度的过程监测与故障诊断  

Process monitoring and fault diagnosis based on subspace mixed similarity

作　　者：杨英华[1] 魏玉龙[1] 李召 秦树凯[1] 

机构地区：[1]东北大学信息科学与工程学院,沈阳110004 [2]许继电源有限公司,许昌461000

出　　处：《仪器仪表学报》2013年第4期935-941,共7页Chinese Journal of Scientific Instrument

基　　金：中央高校基本科研业务费专项资金(N100404018)资助项目

摘　　要：针对现代工业过程多变量、过程数据通常同时包含高斯性和非高斯性分布的特点,提出了一种基于混合子空间的系统性能监控与故障诊断方法。首先使用小波去噪、PCA和ICA方法来进行过程检测,然后将基于PCA特征子空间距离相似度和基于ICA子空间余弦相似度的方法结合,建立故障诊断库,计算混合相似度,确定各类故障的诊断阈值。最后对在线的数据进行监控,判断过程是否正常。当有故障发生时,利用混合子空间相似度确定故障类型。该方法可以充分利用过程数据中的高斯和非高斯信息。通过对Tennessee Eastman(TE)过程的仿真研究,验证了该方法的可行性与有效性,与变量贡献图方法相比可以更加有效地监测出故障原因。Aiming at the features that modern industrial processes always have a large number of process variables and the process data usually contain both Gaussion and non-Gaussion information at the same time, a new system per- formance monitoring and fault diagnosis method based on mixed subspace is proposed in this paper. First,wavelet de- noising, PCA and ICA methods are used to carry out process monitoring. Then, the distance similarity method based on PCA characteristic subspace and the cosine similarity method based on ICA subspace are combined to set up fault diagnosis library, calculate the mixed similarity and determine the diagnosis thresholds of various faults. Finally, the online process measurement data are monitored to judge whether the process is normal. We could judge the source and type of the fault in a short time based on mixed similarity when abnormal condition appears. This method could make full use of Guassion and non-Guassion information in the process data. A simulation case study of Tennessee Eastman （TE） process indicates that the proposed method is feasible and efficient. Compared with the variable con- tribution plot method, the proposed method can locate the fault more quickly and efficiently.

关 键 词：混合相似度 田纳西-伊斯曼过程 过程监测 故障诊断 

分 类 号：TP277[自动化与计算机技术—检测技术与自动化装置]