Generació de Decisions davant d'Incerteses

Àntoni Escobet Canal
Departament de Disseny i Programació de Sistemes Electrònics (DiPSE)
Escola Politècnica Superior d'Enginyeria de Manresa (EPSEM)
Universitat Politècnica de Catalunya
Avda. Bases de Manresa 61-73
08242 Manresa
Spain
Email: Toni@EPSEM.UPC.Edu

Abstract

This thesis deals with the Fuzzy Inductive Reasoning (FIR) methodology applied to fault detection and diagnosis systems. FIR, based on the General Systems Problem Solver (GSPS) proposed by Klir in 1989, is a methodological tool for data-driven construction of dynamical systems and for studying their conceptual modes of behavior. FIR is a qualitative modeling and simulation methodology that is based on observation of the input-output behavior of the system to be modeled, rather than on structural knowledge about its internal composition. This methodology has evolved over time with the aim of enlarging the class of problems that can be dealt with by FIR.

The work presented in this thesis aims to contribute to reducing modeling and simulation efforts of real industrial complex systems. Several methodological contributions have been made to increase FIR robustness as well as to develop a new methodology to create robust and efficient fault detection and diagnosis systems.

The main objective of this thesis is to reduce as much as possible the sensitivity of the FIR methodology, by maximizing its robustness, in such a way that it becomes a fundamental tool for developing efficient fault detection and diagnosis systems.

The main contributions of this thesis are:

To improve the robustness of FIR by creating a new tool, Visual-FIR, that identifies patterns and predicts future behavior of dynamical systems in a very efficient and simple to use environment.
To develop a new methodology for creating fault detection and diagnosis systems based on FIR. We have developed a detection technique, the enveloping, and a diagnosis measure, known as the acceptability measure, that allow improving and making more robust the fault detection and diagnosis processes of the FIRFDDS (fault detection and diagnosis system based on FIR).
To develop a tool that allows to easily create highly efficient FIRFDDS for specific applications. A platform, named VisualBlock-FIR, has been developed that allows the user to create, in a simple way, fault detection and diagnosis systems based on FIR.

In order to validate the methodological contributions and the developed tools a couple of case studies have been presented in this dissertation. The first corresponds to the benchmark problem of the Damadics automatic valve system, which proposes four failures of small and medium sizes that are detected and isolated / identified in a quick and highly efficient way. The second is a simulated fuel cell where five different faults are applied. The five faults are detected and identified correctly. Finally, we check the robustness of the FIRFDDS by adding white noise, at different magnitudes, to the outputs of the fuel cell.

Interested in reading the entire thesis? (220 pages, 3,683,388 bytes, pdf - only available in Catalan)
Interested in viewing the Powerpoint presentation? (only available in Catalan)

Homepage