Fault Diagnosis and Fault Tolerance for Mechatronic Systems: Recent Advances

This book is a product of a workshop on Fault Diagnosis and Fault Tolerance for Dynamic Systems, held in Vancouver in October 2002. Each lecturer has expanded his contribution to form a chapter of the current volume. This book therefore represents a fairly up-to-date snapshot of the control engineer’s approach to fault diagnosis via state-space models. The book comprises six substantial articles by experts in the field. Although the discussion is fairly mathematical in places, there are many application examples. These are not restricted to robotic systems, but cover thermofluid systems, underwater vehicles, and aircraft. The first chapter, by Saif and Xiong, adopts an observer-based approach to the fault detection, isolation, and accommodation (FDIA) problem. In this approach, the observer is used to generate estimation errors or residuals, so that the residuals can bemonitored for large excursions,which signal deviations from the base model and hence faults. The article deals with both sensor and actuator faults. To be specific, the authors use a sliding mode observer (SMO). The SMO used is a generalization of that devised by Walcott and Zak (WZ). The WZ observer is exact for a certain class of non-linear uncertain systems, but implies strong structural constraints, and therefore, is of limited application. The new designs proposed in the article reduce the structural constraints. The first extension proposed is the sliding mode output observer (SMOO), which proves to be good for detecting large faults, but less effective at detecting incipient faults. This leads the authors to propose further extensions: the sliding mode functional observer (SMFO) and the sliding mode output functional observer (SMOFO). The methods are illustrated on an application to engine diagnostics. The second chapter by Wang, considers the problem of fault diagnosis and fault tolerant control for non-Gaussian stochastic systems with random parameters. The approach is to detect changes in the probability density function (PDF) of the parameters. Previously, this was accomplished by monitoring the mean and variance, an approach that is inadequate for non-Gaussian systems. A recursive estimator for the random parameters is derived using the measured PDF, and faults are signalled by unexpected changes in the parameter vector. An approach to fault diagnosis via the minimization of residual entropy is proposed and discussed. The application example is a refining system in a paper making process. The editors Caccavale and Villani present the third chapter on fault diagnosis for industrial robots. This is concerned with observer-based FDI again and concentrates on applications to robot manipulators. The diagnostics are designed in discrete-time to avoid problems with discretization. A simple model-based FDI scheme is first discussed and found to have drawbacks. This leads to observerbased FDI schemes. The approach is to achieve a sharp diagnosis by minimizing uncertainties. The uncertainties are dealt with by using three methods. The first is a robust time-delayed compensation scheme, the second – which proves better – is a recursive estimator, and the last is an adaptive approach that assumes a parametric model of the uncertainties. The three approaches are illustrated and compared on a six degree-of-freedom robot with both abrupt and incipient simulated faults. The next article, by Antonelli, is concerned with autonomous underwater vehicles (AUVs) and remotely operated vehicles (ROVs). A specific problem for the diagnosis and control of such systems is that they operate in an unstructured hostile environment. Undetected failures can lead to the loss of vehicles. Fault tolerance is discussed in the specific context of thruster redundancy. The article gives a summary of experienced failures. Both fault detection and fault tolerance are discussed and past experiments are reviewed. Chapter 5, by Boskovic and Mehra, is concerned with failure detection, identification, and reconfiguration (FDIR) in flight control. The particular challenges of FDIR design for aircraft, where a single 1015