FAULT DETECTION AND DIAGNOSIS IN PROCESS PLANTS USING FINITE-STATE AUTOMATION

Abstract

This thesis addresses the problem of Fault detection and diagnosis (FDD) in process plants using Finite-State Automaton (FSA) Models. In the first part of the work a review of various FDD methods for process plants is presented and their respective merits and demerits are discussed. This is followed by a discussion on Finite-State Automaton theory and its use to mathematically model discrete-event systems (DES) for e.g. process plants. Some parallels are also drawn between FSA and hybrid system models existing in the literature with a conclusion that FSA model is another member in the class of hybrid system models. FSA model partitions the state-space into finite regions and contains information on system trajectory across these regions. Finite-state Automaton Tables (FATs) offer a convenient tool to record these state-transition details. A framework for modeling faults as discrete inputs using FSA models is then discussed. The on-line generation of FATs is them explained. This is to prevent the explosion of table size for "real-life" applications with large number of state variables. A fault-detection shell which dynamically generates automaton tables is developed which helps to examine the areas of interest in the state-space selectively. The algorithm is illustrated with a case study on a bend-top heat-exchanger pilot plant. Snap-shots from the case-study are presented and some conclusions are drawn. Recommendations for future work includes discussion on the diagnosibility issues related to selection of state-boundaries and the development of a set of diagnostic tools to diagnose faults which are not modeled.