Fault Diagnosis Based On Data Driven For Autonomous Underwater Vehicle Sensor

Autonomous Underwater Vehicle (AUV) is a kind of intelligent platform between unmanned submarine and surface ship. It is very significant to ocean resource exploration, hydrologic data collection and military strategy penetration. AUV obtains all the information of external environment and self-states mainly by varied sensors. Thus, the reliability of sensors is utmost important to AUV. However, fault diagnosis technique can avoid the catastrophic events to some extent.Though there is a long time and rapid development for AUV sensor fault detection and diagnosis technology, there are still many problems unsolved. Fault diagnosis technology based on data driven will also be constantly improved. As to fault diagnosis method based on data driven still need further research. The major research work is as follows:(1) In order to overcome the limitations of the model based method, the sensor fault diagnosis method based on dynamic prediction of Taylor series is proposed. First of all, applying accumulated generating algorithm to sensor output data sequence, a regular sequence with exponential growth can be constructed. Then, Taylor series predictive model is established based on the sequence constructed above and future output data of the sensor can be predicted and estimated based on the predictive model. Simulation results for the four main types of AUV sensor faults using this algorithm show that this algorithm can quickly, accurately diagnose the sensor faults. And this method can be used to reconstruct the fault signals.(2) To overcome the problems of the lack of data, experience and so on, the sensor fault diagnosis method based on grey correlation analysis is presented. Grey correlation analysis can reflect the similarity degree of sequence geometric curve and the proximity of sequence variation trend. Before applying this fault diagnosis method, it is needed to get the sensor output data under fault condition to establish a fault pattern sequences set. Then, calculate the comprehensive grey incidence degree between measurement data sequences and the fault pattern sequences. So, the fault diagnosis of the measurement data vectors will be achieved. Through the analysis of simulation results, it can be drawn that the sensor fault diagnosis algorithm based on grey correlation analysis is effective and feasible.