Learning Observer Based Fault Estimation for a Class of Unmanned Marine Vehicles: The Switched System Approach

This paper concerns with the fault estimation problem for a class of unmanned marine vehicles with sensor faults and non-differentiable actuator faults. Firstly, the presented unmanned marine vehicles are modeled as a class of continuous-time switched systems. In order to resolve the non-differentiable actuator faults, two kinds of learning observer methods are proposed to allow the augmented systems independent of the time-derivative of actuator faults. The first one is a kind of timeline-based learning observer, which produces the satisfactory estimation results by using the estimated information from the previous moment and the measured output estimation error information. The second one is an iterative learning observer, where the actuator faults are accurately estimated by the iterative process. The monotonic convergence of the iterative process is guaranteed based on the derived linear matrix inequality conditions. Under arbitrary switching signals, the simultaneous estimations of states, faults and disturbances can be achieved via the proposed observer design approaches. Finally, simulation results are provided to illustrate the effectiveness of the developed two kinds of learning observers.
Note to Practitioners
—The mass of unmanned marine vehicles (MVs) is switched back and forth as the unmanned MV performs the tasks, such as loading, dispatching and retrieving detectors. Hence, in order to better describe the dynamic behaviour, the unmanned MVs are described as the mass-switched systems. In general, any actual systems with switching properties can be characterised as the switched systems. At present, the fault estimation problem of the mass-switched unmanned MVs has not yet been investigated to date. This paper presents a theoretical study. Existing approaches to fault estimation of switched systems generally require differentiable actuator faults, which may be difficult to guarantee. In order to relieve this restriction, this paper proposes the new learning observer-based estimation method for switched linear systems. Based on the physical data of the unmanned MV, simulation results show that the proposed method is feasible.