OPTIMAL IDENTIFICATION OF TAKAGI-SUGENO FUZZY MODELS FOR NONLINEAR FDI

Abstract

The design of Fault Detection and Isolation (FDI) systems is generally based on models identified by minimizing adjustment error. For non-linear processes, the use of observers based on Takagi-Sugeno fuzzy models have been proposed. However, from a detection and isolation point of view, models obtained using known identification algorithms are not necessarily optimal. This is showed through evaluations made using a benchmark hydraulic system simulator, with a confusion matrix, mean detection time and mean time between false alarms as performance indicators. Based on these results, we propose a new design procedure employing multiobjective optimization that improves on the above-mentioned time indicators by a factor varying between 1.5 and 3.7.