SIMULINK BLOCKSET FOR FAULT DETECTION USING INTERVAL MODELS S. Tornil, T. Escobet and V. Puig Dept. ESAII, Universitat Politecnica de Catalunya Rambla Sant Nebridi 10, 08222 Terrassa (SPAIN) Tel. +34937398621, Fax. +34937398628 E-mail: {stomil.vpuigj@esaii.upc.es. teresa@bages.eupm.upc.es Abstract: This paper describes a tool to aid in the analysis and design of fault detection systems for processes which present bounded structured uncertainty (only bounds for the values of their physical parameters are known). This uncertainty is considered in the fault detection methodology by using interval models to represent the processes. The tool has been implemented using the well-known Matlab/Simulink CACSD framework. This framework is very useful in the analysis and design of control systems and can also be used in the analysis and design of supervisory systems (which must include fault detection capabilities). Copyright © 2000 IFAC Keywords: Fault detection, uncertain dynamic systems, intervals, optimisation, software tools. 1. INTRODUCTION The detection and isolation of faults (FDI) in complex industrial systems is one of the most important tasks assigned to the computers supervising such systems. The early indication of incipient faults is critical in avoiding performance degradation, product deterioration, major system damage and loss of safety in safety-critical systems. The quick and correct isolation of the faulty component allows that appropriate actions can be applied as soon as possible. In model-based FDI methods, the actual behaviour of the system is checked for consistency with a mathematical model that describes its non-faulty operation (analytical redundancy) . Consistency checking is normally achieved through a comparison between the measured value of a system variable and the estimated value obtained using the system model. 165 In most practical applications, it is not possible to obtain an accurate and complete model of the process (system), due to some kind of uncertainty in the available knowledge about it. Sometimes it is not possible to know the whole structure of the process (unstructured uncertainty) and sometimes the structure is known but the value of the model parameters is not exactly known (structured uncertainty). When bounded structured uncertainty is present, interval models appear as a natural framework to deal with it. A Simulink blockset has been developed to manage interval models. The manipulation of such models involves the solution of some optimisation problems. The 'Optimization Toolbox' included in Matlab implements the classical optimisation algorithms, but these algorithms only present local convergence. In the actual version of the 'blockset', the GIA InC++ library is integrated into the Matlab/Simulink