Abstract:

In order to overcome the shortage of fault isolation in multivariable statistical process control, a new fault isolation algorithm was proposed that included Fisher discriminant analysis (FDA) based feature extraction, Fisher linear classification and support vector data description (SVDD) based pattern classification in nonlinear kernel space.FDA-SVDD based cascade and series parallel combination forms were constructed, and a novel SVDD based weighted normalization radius discriminance was designed.The simulation of a non-isothermal continuous stirred tank reactor (CSTR) process showed that the series parallel combination form had better fault isolation performance than the simple FDA method and the cascade combination form.

摘要：

为了克服多变量统计过程控制在故障分离上的缺陷， 提出了一种新的故障分离方法。 新方法由基于Fisher判别分析（FDA）的特征提取、Fisher线性分类和基于支持向量数据描述（SVDD）的非线性核空间模式分类等算法组成。构造了基于FDA-SVDD的串级和混联融合方式， 并设计了基于SVDD的加权归一化半径模式判别准则。非等温连续搅拌槽 （CSTR）过程仿真验证了混联融合比单纯的FDA分类算法和串级融合具有更优良的故障分离效果。