Comparative study based on KPCA and KPLS monitoring techniques using the Chemical Stirred Tank Reactor (CSTR) nonlinear process

In this paper, a comparison of performances between the Kernel Partial Least Square (KPLS) monitoring method, as well as the Kernel Principal Component Analysis (KPCA) monitoring technical is presented. Among the Fault Detection (FD) methods, the techniques based on Kernel principal have attracted much attention in industrial field. Furthermore, the KPCA and KPLS methods are the most popular for the diagnosis of non-linear systems. Our main contribution is shown the performances and the properties differences of a method over the other. However, the FD performances, from this comparison, are essentially based of the False Alarm Rate (FDR), the Good Detection Rate (GDR) and finally the Computation Time (CT). These methods are compared and presented by handling a Continuous Strirred Tank Reactor (CSTR) system.