Research On Fault Diagnosis Of AUV Thruster Based On Improved SVDD Method

With the expansion of human marine development strategy, as the most important tool to explore the marine, underwater vehicle technology has attracted people's attention and developed rapidly. AUV (Autonomous Underwater Vehicle) can work autonomously in unstructured ocean environment without person or cable. The security and reliability of AUV are the valuable premises towards a practical process. As the heaviest working load parts of AUV, thrusters have become the most common and serious fault source. Combining with characteristics of AUV's complicated working environment and the strongly nonlinear system of AUV itself, researching on the thruster fault diagnosis technology has important significance to improve the intelligent security level for AUV.To solve the problems of unbalance, incompleteness of data set and fault identification,this paper establishes the thruster fault classification model and reteaches on the parameter optimization and fault identification technology for classification model, based on the design of AUV experiment platform system as well as the study of thrusters system.AUV carrier technology and propulsion system have been researched. In order to achieve the condition monitoring of AUV thruster system and intelligent diagnosis together with testing the feasibility of proposed method, AUV master control system and propulsion system have been designed, AUV experimental platform have been developed, and AUV thruster dynamics models have been established based on the pool experiment.AUV thruster fault classification method has been researched. During to the unpredictable disturbance such as ocean current and AUV's characteristics of strong nonlinear and large time delay, the traditional methods, which are based on the analytical model, are extremely limited, and it is difficult to establish an AUV precise mathematical model. To solve this problem, an AUV thruster fault mode classification method based on SVDD(Support Vector Domain Description) algorithm is proposed. This method establishes the least enclosed super-sphere model for data sets, thus establishes AUV thruster fault classification model based on historical data set in fault mode. Via the classification results of AUV's normal and various degrees of fault data obtained in pool experiment, the effectiveness of the proposed SVDD method has been tested.Parameters optimization method of classification model has been researched. The way SVDD describes data has been more flexible with the help of kernel tricks,which in turn makes SVDD classification performance depends too much on its kernel parameters. To solve the optimization problem for kernel parameters in SVDD classification model, a method to describing the distribution form and law of the mapping data sets in high-dimensional space has been proposed, analyses kernel parameters affect its classification performance according to experimental data. On basis of the method, this paper has optimized the measurement of kernel parameter according to the principle of maximum entropy to improve the classification performance of SVDD model. The effectiveness under varies fault types of the proposed parameter optimization method for the SVDD classification model has been verified.Fault degree Identification method has been researched. Due to the traditional SVDD method is a single classification method and does not have the ability to identify unknown fault degree. Meanwhile its completeness to obtain fault data via experiments is limited. To solve the problems above, the paper has proposed an identification method for unknown fault degree by analyzing the super-sphere space structure of known degree fault. The effectiveness of proposed fault identification method has been verified via the results of pool experiments.