Comparison Between Statistical and Observer-Based Approaches for Fault Detection and Isolation in a Chemical Process

Abstract

Abstract This paper summarizes the results of a comparison including three different fault detection and isolation (FDI) methods for dynamic systems. The techniques studied have recently been proposed in the literature and are Dynamic Principal Component Analysis (DPCA), Canonical Variate Analysis (CVA) and Subspace Model Identification (SMI). The aim of this study is to contrast the performance of each method in detecting and isolating incipient fault conditions. Utilizing real data from a debutaniser distillation tower, this study yields that the observer approach based on an identified SMI model is most sensitive for fault detection but performs poorly in isolating the fault condition. This method failed to correctly diagnose the fault condition using fault isolation approach. In contrast, DPCA offered a correct picture of this event using variable reconstruction and contribution charts, whilst CVA only yielded satisfactory results using variable reconstruction. For this study, it is therefore concluded that both approaches have complementary strengths and weaknesses.