基于ICA-PCA的化工流程仪表故障诊断

Instrument fault monitoring in process industry with independent component analysis and principal component analysis

传统的多元统计过程控制(MSPC)的故障诊断方法要求观测变量数据服从高斯分布,然而实际化工流程中的仪表数据中难以满足这一要求。针对这一问题,提出在仪表数据中提取分离出非高斯信息和高斯信息,并分别利用独立元分析法和主元分析法建立不同的故障诊断模型。在检测到发生故障后,通过改进的贡献度算法定位出发生故障的仪表。通过对Tennessee Eastman(TE)过程数据进行仿真研究,验证了ICA-PCA故障诊断法在化工流程仪表不同故障诊断中的有效性。

Multivariate statistical process control（MPSC） method assumes that the monitored variables have a Gaussian distribution. In fact, most of the Instrument data don＇t have a Gaussian distribution. A instrument fault monitoring method based on independent component analysis（ICA） and principal component analysis（PCA） is proposed to extract the Gaussian and non-Gaussian information for fault detection and diagnosis. The fault source of instrument can be determined by contribution algorithm. The proposed fault diagnosis method is proved to bc effective by simulation with the data from the Tennessee Eastman process.