Distributed Sensor Fault Detection and Isolation for Nonlinear Uncertain Systems ? Vasso Reppa * Marios M. Polycarpou * Christos G. Panayiotou * * KIOS Research Center for Intelligent Systems and Networks, Electrical & Computer Engineering Department, University of Cyprus, Nicosia 1678, Cyprus (e-mail: {vreppa, mpolycar, christosp}@ ucy.ac.cy). Abstract: This paper presents a design methodology and some analytical results for distributed sensor fault detection and isolation (SFDI) of a class of nonlinear uncertain systems. The proposed architecture is based on the design of local SFDI modules, which monitor the healthy operation of a set of sensors and aim to detect the faults in this set, using a dedicated nonlinear observer scheme. The multiple sensor fault isolation procedure is further enhanced by deriving a combinatorial decision logic that processes information from local SFDI modules. The performance of the proposed diagnostic scheme is analyzed in terms of its robustness with respect to the modeling uncertainties and conditions for ensuring fault detectability and isolability. Keywords: Fault detection, fault isolation, sensor faults, nonlinear systems, uncertain dynamic systems 1. INTRODUCTION During the last few years, we are witnessing a rapid in- crease in the use of sensing devices for monitoring and control applications, such as manufacturing, power sys- tems, environmental monitoring, etc. Technological ad- vances have facilitated the wide deployment of distributed sensors, which provide temporal and spatial information through wired and wireless links [Ding et al. (2006)]. The information provided by sensors is used for safety-critical tasks and decisions, therefore if one or more of the sensors gives erroneous information, then the efficiency of the system maybe degraded, or even worse, the system may become unstable, thus jeopardizing the safety of humans and/or expensive equipment. One approach to the problem of the sensor fault detection and isolation (SFDI) is the physical redundancy method. However, in most applications, the physical redundancy is not practical due to high cost of installation and mainte- nance, and space restrictions. Therefore, analytical redun- dancy approaches for SFDI are widely used [Chen and Patton (1999); Isermann (2006)]. Among them are the observer-based SFDI schemes, which have been extensively developed for linear dynamic systems. The general frame- work of an observer-based SFDI method consists of the generation of residuals, which are compared with fixed or adaptive thresholds. An initial classification of the existing observer-based ar- chitectures for SFDI in nonlinear systems is based on the different ways of modeling the inherent system non- linearity and the sensor faults. A usual approach is the linearization of the monitored nonlinear system either at ? This research has been funded by the European Commission 7th Framework Program, under grant INSFO-ICT- 270428 (i-Sense) a finite number of operating points [Montes de Oca et al. (2011)] or operating zones [Orjuela et al. (2010)]. Then, the well-established observer-based SFDI methods for linear systems can be applied. However, the linear approximation of the system introduces additional errors in the residuals affecting the fault detectability. Also, if the early detection of a fault is not achieved, it may cause the trajectory to move away from the operating point or zone, thus making the linearization even less effective. Recently, some research has been devoted to the detection and isolation of a single sensor fault in nonlinear uncertain systems [Zhang et al. (2005); Zhang (2011)]. The isolation of multiple sensor faults in nonlinear dy- namic systems is a challenging problem. Several re- searchers, using nonlinear models of the monitored sys- tems, have developed diagnostic schemes based on a single nonlinear observer, capable of isolating multiple sensor faults with specific time profile or magnitude, or assuming a maximum number of their multiplicity. Exciting research results has been obtained using: a diagonal residual feed- back nonlinear observer, for simultaneous sensor faults, whose number is less than the number of states minus the number of disturbances [Narasimhan et al. (2008)]; a sliding mode observer (SMO) [Yan and Edwards (2007)] and a neural-network based observer [Talebi et al. (2009)] for bounded sensor faults. Other approaches are based on a linear matrix inequalities-based observer for isolating at most two simultaneous sensor faults [Rajamani and Ganguli (2004)] and a diagnostic adaptive observer for sensor faults with constant magnitude [Vemuri (2001)]. On the other hand, there has also been some research activity in addressing the problem of detecting and iso- lating multiple sensor faults using a bank of observers. For consecutive but not simultaneous sensor faults, a Gen- 8th IFAC Symposium on Fault Detection, Supervision and Safety of Technical Processes (SAFEPROCESS) August 29-31, 2012. Mexico City, Mexico 978-3-902823-09-0/12/$20.00 © 2012 IFAC 1077 10.3182/20120829-3-MX-2028.00105