Accelerated Kernel Canonical Correlation Analysis with Fault Relevance for Nonlinear Process Fault Isolation

In the field of multivariate statistical process monitoring (MSPM), fault isolation has attracted increasing attention, due to its importance in ensuring process reliability and product quality. However, the existing fault isolation methods are mostly limited to linear settings with single variable isolation. For nonlinear modeling, the kernel method is commonly used, but the time for solving a kernel matrix and its storage required in the traditional method increase sharply with large sample size. To solve these issues, a multivariate fault isolation method based on accelerated kernel canonical correlation analysis (AKCCA) is proposed. In the new method, kernel canonical correlation analysis is utilized to associate variables with process anomaly and extracting nonlinear structures. Furthermore, full rank factorization is embedded in kernel matrix approximation while performing eigenvalue decomposition (EVD), which substantially reduces the storage and computational expense. In addition, faulty relevance of each variable is newly calculated, which improves the accuracy of fault isolation for nonlinear processes. The feasibility of AKCCA and its computational advantage are illustrated by a numerical case and the Tennessee Eastman benchmark.