A model-based technique for fault identification of sensors in autonomous systems using adaptive neural networks and extended Kalman filtering

In this study, an adaptive scheme for autonomous underwater vehicle systems is developed that utilizes a model of the complex nonlinear dynamics and control of the vehicle to enable detection of sensor faults and failures. Our framework for design of fault identification and risk management, incorporates a neural network-based nonlinear observer to monitor the input and output of the control system for detection of a variety of faults in the sensors. The training occurs online and parameters of the recurrent neural network are updated by an extended Kalman filter. The fault detection and identification system was developed and integrated for a nonlinear model of a Remus-100 underwater vehicle. The results obtained from the numerical simulation shows the system's ability for prompt detection and isolation of a variety of sensor faults. Further study is needed for development of experimental validation and verification and computational efficiency of the proposed algorithm.