基于PCA的重油分馏塔故障监测与诊断分析

PCA-based Fault Detection and Diagnosis with Application to Heavy Oil Fractionator

对系统过程数据进行主元分析,建立主元模型,可以在保留原有数据信息特征的基础上消除变量关联和部分系统噪声干扰,从而简化系统分析的复杂度。建立正确的主元模型,结合多变量统计过程控制图(Q统计图,HotellingT2图,主元得分图,贡献图),是对过程对象的进行检测和诊断的一项发展中的技术。通过对一个典型的重油分馏塔运行过程的故障监测与诊断分析,进一步说明了主元模型在确定故障特征方向和多变量统计控制图在监测和诊断故障源上的作用和有效性。同时采用了平均贡献图来直观明确地判别引起系统故障的主要原因。

Principle Component Analysis (PCA) is an effective way not only to eliminate correlation among process variables and reduce the influence of noise and disturbance on system, but also to reserve enough information of original data characteristics needed for modeling a industrial complex process. Based on principle component model, detection and diagnosis analysis is carried out on a typical Heavy Oil Fractionator with multivariate statistical techniques such as Q residuals plot, Retelling T2 plot, principle scores plot and contributions plot. In addition, mean contributions curve is used instead of common contribution histogram to diagnose the source cause of faults.