Experimental Evaluation of a Fault Detection and Identification Scheme for Chemical Processes

Abstract

This paper deals with the experimental implementation of an automatic scheme for fault detection and identification (FDI). It also provides redundant signals for performing an efficient control in case of a faulty situation, improving safety and reliability of plant operations. Two approaches for on-line parameter estimation are tested: a variant of Recursive Least Squares (RLS) algorithm for linear or linearized models and the GREG algorithm (Stewart et al., 1992), suitable for more general nonlinear models. The fault detection method is based on a statistical analysis of the deviation of the process parameters. Simultaneously, an Extended Kalman Filter (EKF) enables the fault identification to be performed, reducing false alarms and providing estimates of process variables for alternative control purposes. Applications take place in a industrial-scale pilot plant. The good performance obtained highlighted the strategy employed when sensor faults were introduced artificially on control loops.