Model based fault detection and isolation in nonlinear dynamic systems

Any complex system or process is subject to the occurrence of faults. These systems may be mechanical, chemical, electrical or some combination of these and may or may not be controlled by a computer. Faults include the complete or partial failure of actuators, sensors or other components of the system, or the occurrence of events external to the system (disturbances) that prevent its normal functioning.; The ability to detect the occurrence of any fault, and identify its cause is critical for a number of reasons. In some cases faults can lead to great loss of life and property (e.g., aircraft, nuclear power plants, etc.). Further, early detection of faults can allow timely corrective action which in many cases will greatly reduce the incidence of expensive, unexpected breakdowns (e.g., machinery in factories). Environmental considerations are also of importance as in the development of systems for monitoring automobile engine emissions.; The increasing sophistication and decreasing cost of electronic systems has made viable the use of complex algorithms for the monitoring of dynamic systems. Over the past two decades much attention has been focused on the development of diagnostic algorithms for linear dynamic systems and a number of useful results have been developed. However very little attention has been paid to nonlinear systems and most hitherto developed techniques fail when applied to systems that exhibit significantly nolinear dynamic behavior.; This study focuses on the development of fault detection and isolation techniques for nonlinear dynamic systems. A general problem formulation that is applicable to both nonlinear and linear dynamic systems is postulated. Some of the hitherto qualitatively understood terms and concepts are formally defined for the general nonlinear case. Necessary and sufficient conditions are provided for the solvability of the fault isolation problem. It is shown that the isolation solution exists for those systems for which the fault variables satisfy certain invertibility conditions with respect to the available set of measurements. A systematic monitoring system design procedure that depends on the construction of forward and inverse models is provided for this class of systems. The fault detection algorithm does not depend on the particular type of modeling or inversion scheme used, requiring only that such models and inverses can be derived.; New results in sliding mode state estimation that are useful for the fault isolation problem are developed and validated through experimental work.; The derived results are experimentally validated through extensive experiments in the form of 3 case studies: (i) Internal Combustion engine fault detection using nonlinear input-output identification, (ii) Vehicle steering system monitoring using continuous time sliding mode observers, and (iii) Engine air-fuel-exhaust system monitoring using discrete sliding mode observers.; The results show that the scheme developed in this study is extremely successful, and applicable to all systems that satisfy certain invertibility conditions, and for such systems is capable of isolating multiple fault occurrences even under the influence of unknown disturbances.