Fault detection and diagnosis of nonlinear processes using improved kernel independent component analysis (KICA) and Support Vector Machine (SVM) (Retracted article. See vol. 58, pg. 20858, 2019)
Fault detection and diagnosis of nonlinear processes using improved kernel independent component analysis (KICA) and Support Vector Machine (SVM) (Retracted article. See vol. 58, pg. 20858, 2019)
复制标题
DOI：
10.1021/ie071496x
复制
发表时间：
2008-09-17
影响因子：
4.2
通讯作者：
Zhang, Yingwei
中科院分区：
文献类型：
--
作者：
Zhang, Yingwei
本文提出了一种基于核独立分量分析的非线性动态过程监测方法。与支持向量机（SVM）方法相比，KICA方法是无监督的，可用于故障检测。因此，在本文中，KICA用于检测故障。由于特征空间的维数远小于核矩阵的秩，因此选择特征空间中的一个基。具体而言，本文首先基于一组数据的相似性因子构建特征空间中的基。贡献图是不可能的，因为从输入空间到特征空间的非线性映射函数是未知的。因此，KICA很难用于非线性故障诊断。在本文中，一旦检测到故障，将直接引入改进KICA的核变换分数作为支持向量机的输入进行故障诊断。在选择相同独立分量（nic）的情况下，SVM +改进KICA的分类率高于SVM + KICA的分类率。原因是改进后的KICA + SVM的负熵比原来的KICA + SVM更能考虑到原始输入的有用信息。支持向量机加改进KICA的训练时间比支持向量机加KICA的训练时间短，因为前者减少了昂贵的计算量。将该方法应用于田纳西伊士曼过程和污水处理过程的故障检测与诊断。应用表明，与传统方法相比，该方法有效地捕获了过程变量的非线性动态，并具有更好的故障可检出性。
In this article, the nonlinear dynamic process monitoring method based on kernel independent component analysis (KICA) is developed. Compared to the Support Vector Machine (SVM) method, KICA is unsupervised and available for fault detection. Hence, in this article, KICA is used to detect faults. Because the dimension of the feature space is far less than the rank of kernel matrix, a basis in feature space is selected. Specifically, the basis in feature space is first constructed based on the similarity factor of data in one group in this article. A contribution plot is impossible, because the nonlinear mapping function from input space into feature space is unknown. Therefore, KICA is difficult for nonlinear fault diagnosis. In this article, once a fault is detected, the kernel-transformed scores from improved KICA will be directly introduced as the inputs of SVM to diagnose the fault. The classification rate of SVM plus improved KICA is higher than the classification rate of SVM plus KICA when the same number of independent components (nICs) is selected. The reason is that the negentropy in improved KICA plus SVM could take into account the more-useful information of original inputs than that of original KICA plus SVM. The training time of SVM plus improved KICA is shorter than that of SVM plus KICA, because the former attenuates the expensive computation load. The proposed approach is applied to the fault detection and diagnosis in the Tennessee Eastman process and a wastewater treatment process (WWTP). Applications indicate that the proposed approach effectively captures the nonlinear dynamic in the process variables and shows superior fault delectability, compared to conventional methods.