The Development of the Adaptive Diagnostic System for Navigation Sensors of Autonomous Underwater Vehicles * Vladimir Filaretov Alexey Zhirabok Alexander Zuev Robotics Laboratory Department of Automation and Control Laboratory of Problems of Optimal Control Institute of Automation and Control Processes Far-Eastern Branch of Russian Academy of Science Far Eastern Federal University Institute of Applied Mathematics Far-Eastern Branch of Russian Academy of Science 5, Radio str., 690041, Vladivostok, Russia 8, Sukhanova str., 690950, Vladivostok, Russia 7, Radio str., 690041, Vladivostok, Russia filaret@iacp.dvo.ru zhirabok@mail.ru zuev@iacp.dvo.ru Abstract - In this paper, the problem of the development of adaptive diagnostic system for navigation sensors of autonomous underwater vehicle (AUV) is studied. The diagnostic system is synthesized by using the observer-based methods. Herewith the big problem is uncertainty and variability of parameters of model of the underwater vehicles during performance of underwater mission. It does not allow to execute diagnostics of navigation systems of the device qualitatively. For solving this problem, the method of synthesis of special adaptive feedback for each diagnostic observer is offered. As a result the adaptive diagnostic system allows to detect and to isolate the faults of navigation sensors of the vehicles at performance of underwater missions in real time . The proposed adaptive diagnostic system was tested by simulation with parameters of real AUV. The results of simulation have completely confirmed the working capacity and high quality of the proposed robust observers with feedback in different regimes of work of underwater vehicle. Index Terms - of autonomous underwater vehicle, adaptive diagnostic system, feedback, observer. I. INTRODUCTION The autonomous underwater vehicles are complex technical systems, their safety and reliability are deserve particular attention. Navigation sensors are one of the most important components of the AUVs which are necessary to control a motion trajectory at performance of autonomous underwater missions. Since malfunctions and faults occurring in these sensors can lead to erroneous mission fulfillment or loss of the vehicle therefore it is necessary to detect and isolate a faulty sensor early. This task is very difficult problem because there is complex interdependence among the state vector elements of AUVs dynamic model. There are some different methods of fault diagnosis: signal-based, analytical model-based, knowledge based [2-3, 10-12, 14, 16]. For diagnosis of the AUV navigation sensors, the analytical model-based method can be applied that allows one to use the redundancy of the AUVs mathematical models which are well-known. In this paper, to provide the diagnosis process, observer- based method is used. In this case, the actual measurements of vehicle's sensors are compared with output signals of fault-free observers driven by the control signals and measurements of AUV's sensors. Difference between the actual sensor measurement and corresponding observer output signal is a residual signal that carries all possible information about the faults in the AUV's components. However, the parameters of AUVs model are known not precisely. For example it's not possible to calculate the added masses and know precisely hydrodynamic forces and moments etc. Therefore synthesized observers not always possess stability [9]. There are several approaches [4-5, 15] to robust diagnosing of AUVs subsystems, based on use of expanded filters Kalmana. Their advantage is relative simplicity of the realization, however thus synthesized observers are capable to detect faults effectively only at horizontal movement of the AUVs with low speed. In works [7, 13], the sliding observer is offered to use for the residual formation. Doubtless advantage of the given approaches is tolerance of the synthesized observers to unknown but slowly changing parameters of AUV. However, the big problem interfering practical introduction of such systems is the problem of «chattering». Also there are several methods [6], which use the neural networks for synthesis of the robust observers for navigation sensors of AUV. There the diagnostics is carried out with use of a combination of limiting values of variable parameters of AUV's model. The disadvantage of the given approaches is that observers are developed on the basis of the model which are not considering real nonlinearity at fulfillment of difficult movements by AUV. It this paper, to eliminate influence of external disturbance and inexact knowledge of model parameters of the real AUV on diagnosing process the special adaptive feedback are entered in observers. This provide robust features to diagnosis system.