Adaptive Fault Estimation for Unmanned Surface Vessels with a Neural Network Observer Approach

This paper is concerned with the fault reconstruction observer design problem with unknown nonlinearities, external disturbances and faults. First, a neural-network-based fault estimation approach is developed to generate the estimations of the actuator failures. In this design, the neural network strategy is utilized to approximate the totally unknown nonlinear functions. Then, an iterative adaptive observer is designed to offer the accurate estimations of the sensor faults, where the estimations in the previous iteration are applied in the current iteration to guarantee the convergence of sensor fault estimation errors. The developed neural network observer design approach can reconstruct the states, actuator and sensor faults for the unmanned surface vessel simultaneously. Finally, a practical example of the unmanned surface vessel is presented to illustrate the effectiveness and the potential of the proposed observer technique.