(c) 1995, W. Nuninger

. N.

R

.

E

I

.

N

Institut National Polytechnique de Lorraine

LO

Centre de Recherche en Automatique de Nancy

P

FMO and diagnosis W. Nuninger HCM Meeting, Duisburg, 1995

R AI

Finite Memory Observers (FMO) and Diagnosis

In collaboration with

W. Nuninger, F. Kratz, J. Ragot Based on some results developed by

A. Medvedev HCM Meeting Duisburg, october 27, 1995

CONTENTS

n

Introduction

n

Why Finite Memory Observers

n

Finite Memory Observers Formulation ›deterministic systems

n

Residual generation and Detection Window

n

Link with Parity Space approach

n

Other results and application

n

Conclusion

1

INTRODUCTION

n

Control of Industrial Processes ›Safety of men ›Security of the process ›Production Performances

‡ Fault Detection and Isolation Problem ›Sensor and actuator failures to be detected

‡ Diagnostic Procedure ½Step 1: Residual Generation ½Step 2: Decision Function Generation

‡Further System Accommodation n

Robustness with respect to ›Systems uncertainties ›Measurement noises

Why Finite Memory Observers ?

n

Residual Generation based on ›Parity

space method

›Observer ›Kalman

filter identification

›Parameter

n

Drawbacks ›Infinite

memory ý Insensitivity to recent measurements ›Error in the model + Infinite memory ý Divergence phenomenon ›Time delays ý Performance degradation n

New Approaches ›Fading

Memory Filter Sorenson, Sack 1985 Observer Medvedev 91; Muske 93 ›Finite Memory Obs. Medvedev, Toivonen 91 ý Intrinsically finite memory ý No state vector integration ý Detection using sliding window ›Deadbeat

2

Formulation for Deterministic Systems

n

Systems

xÝ(t) =Ax(t)+Bu(t)  yi ≡y(t −τi ) =Cx(t − τi )

(1) 

i from 0 to k

n

Hypothesis ›(k+1) time delays, τo< τ1