Robust Model-based Fault Diagnosis for Unmanned Underwater Vehicles Using Sliding Mode-observers

The early detection of the malfunctions and faults as well as their compensation is crucial both for maintenance and for mission reliability of unmanned underwater vehicles (UUVs). Among the di erent fault detection methods using analytical redundancy, the rst distinction rises between model-free and model-based approaches [1, 2]. Model-free methods are well-suited for large-scale systems, where the development of a model is too expensive. The lumped parameter model of an underwater vehicle can be easily described by a small set of well-known equations with highly uncertain parameters [3]. This uncertainty suggests the introduction of robustness requirements in the model-based residual generation for UUVs. Robustness can be addressed in many di erent ways: for instance, the reader can refer to [4] for the parity equations and to [5] for the observer-based methods. According to the same view, the so-called unknown input observer (UIO) has been proposed (see [6]). These approaches share the common idea of decoupling residuals and noises by eliminating (or at least reducing in some optimal sense) the in uence of the disturbances on the residuals. Another way of facing the problem consists in applying the H1 lter design to the residual generator [7, 8]. All the methods described up to this point refer to linear models, while the general framework for underwater vehicles presents nonlinearities due to the vehicle dynamics and to the features of the drag occurring in the water. The extension of the residual generation methods to nonlinear systems has been the subject of many studies and an account of all the results can be found in [9]. Fault detection based on parameter estimation may be applied to nonlinear systems when the dynamic equations are linear in parameters, but the uncertainty of common UUV models introduces di culties. Unknown input observers for nonlinear systems require a complex state-space coordinate transformation, often di cult to be computed (see the references in [9]). Research about the extension of parity equations to nonlinear systems is only just beginning (the only paper known to the authors is [10]). In this paper, sliding-mode observers [11, 12] are proposed for the purpose of residual generation for UUV applications. In Section 2 the problem of robust observation for dynamical systems with \Lipschitz" nonlinearities is addressed by means of sliding-mode observers. A robust observation scheme for the detection of actuator faults in the Roby 2 UUV is presented in