Multivariate Process Monitoring Of Structural Changes in A CSTR System

Process monitoring traditionally using univariate process monitoring approach where each of individual variables is monitored separately. In this approach process variables interaction is difficult to be monitored and therefore multivariable statistical process monitoring (MSPM) was introduced to cater the drawback of univariate process monitoring. MSPM has a major advantage in detecting change in variables relationship or also known as structural changes. Despite of the advantage, most of studies are focusing on change in variables rather than the variables interaction. In this study, PCA based detection techniques performance including PCA, dynamic PCA and nonlinear PCA has been evaluated under change in reaction kinetic and change in heat transfer coefficient. Hotelling T2 and SPE chart are employed as the fault detection techniques. The project mainly focusing on fault detectability and fault detection time. All the PCA based approaches are able to detect the structural changes. Nonlinear PCA shows the fastest detectability followed by dynamic PCA and PCA. For highly nonlinear system, Nonlinear PCA are able to detects the fault the fast but the nonlinear PCA not performing the best when encounter with lesser degree of nonlinear data set.