Fault Detection using a Reduced Data-Driven Methods

In this paper, the fault detection phase is presented using a reduced number of observations. Indeed, monitoring systems is a very important task, which helps to characterize normal variations and detect or determine any abnormal changes. In this area, Data driven approach is a very important context for system monitoring based on the principle of machine learning. Howver, the main objective of this paper is to validate the fault detection performance using two methods based on (Kernel Partial Least Square) KPLS method with optimal values determined by Tabu search algorithm. Then, a complete description is provided exclusively of the techniques used in static mode and in dynamic mode. In this work, a static mode is presented with the Reduced KPLS (RKPLS) method and the Moving Window RKPLS (MW-RKPLS) method in dynamic mode. The detection performance of these methods are compared and evaluated using a CSTR chemical process.