基于累积和等距映射的缓变故障检测方法

摘要

针对化工过程存在的缓变微小故障，提出一种基于累积和等距离映射（CUSUM-ISOMAP）的多变量过程故障检测方法。该方法首先运用累积和控制图的思想，分别对每一个变量计算均值偏差累积和及方差偏差累积和，之后建立扩展增广矩阵，对增广数据运用基于ISOMAP的降维特征提取算法建立统计量进行故障检测。传统的ISOMAP算法无法获取输入输出数据之间的映射关系，不能处理新的采样数据。引入核岭回归算法获得新采样点的降维输出。CSTR过程的仿真结果表明了算法对过程微小故障实施故障检测的有效性。%In order to detect the ramp minor fault in chemical processes,we proposed a new multivariable process fault detection method which is based on CUSUM-ISOMAP.It first calculates the mean deviation cumulative sum and variance deviation cumulative sum for each variable using the idea of CUSUM control chart,and then creates extended augmented matrix and employs the ISOMAP-based dimensionality reduction feature extraction algorithm on the augmented data to set up statistics and carry out fault detection.Traditional ISOMAP algorithm is unable to get the mapping relationship between inputs and outputs,so it cannot handle the new sampling data.In the paper we introduce Kernel ridge regression algorithm to obtain the dimensionality-reduced outputs of the new sampling data.Simulation result of CSTR process shows the effectiveness of the method proposed in the paper on exerting fault detection for minor process fault.