Research On Fault Diagnosis Of Underwater Vehicles Based On Data Mining

To make a comprehensive analysis on the running state and abnormal situation of the system, fault diagnosis technology uses various modern detection methods to diagnose the type and location of the fault, predict the fault trend and then make the right decision. The use of fault diagnosis technology can find out the happened system fault in time, avoid unnecessary losses, therefore the research on fault diagnosis technology has important value and significance.Fault diagnosis in the underwater vehicle system is a very complicated and difficult work. Different from general systems, the underwater vehicle system’s structure is complex, which has unique characteristics such as strong nonlinearity and uncertainty. Therefore, the single use of the common method based on signal processing, the method based on system model or diagnosis method based on knowledge is not enough for the underwater vehicle system. Its fault diagnosis should be timely and effective to achieve the comprehensive and reasonable conclusion. According to the characteristics of underwater vehicle system, this thesis presents a fault diagnosis method for underwater vehicle system based on data mining. So the complex system fault diagnosis process can be divided into the two-steps strategy of fault detection and fault identification in order to achieve the fault diagnosis of underwater vehicle system by itself.The main work is as follows:(1) On the basis of reading a lot of literatures, the background and significance of the research have been introduced. At the same time, this thesis has analyzed the underwater vehicle system, the present fault diagnosis research situation of equipments on related theory and data mining technology in domestic and foreign. The related knowledge of fault diagnosis technology and data mining technology has been expounded.(2) Research on unsupervised fault diagnosis of underwater vehicle. The traditional fault diagnosis system is described and the data mining technology, especially the outlier detection method has been studied in depth. Clustering algorithms such as K-mean algorithm, DBSCAN (Density-Based Spatial Clustering of Application with Noise) algorithm, ISODATA (Iterative Self-Organizing Data Analysis Technique) algorithm have been analyzed and compared. Then put forward an improved clustering algorithm IKD (Iterative K-mean DBSCAN) algorithm. This algorithm can deal with arbitrary shape distributed data, the clustering results can achieve self evaluation, IKD algorithm’s high accuracy could make it possible to accomplish outlier detection more suitable for practical application. The method based on IKD algorithm could implement a comprehensive unsupervised fault diagnosis of underwater vehicle.(3) Research on supervised fault diagnosis of underwater vehicle. The decision tree classification algorithm has been described in detail, and C4.5 algorithm has been applied to the fault diagnosis system. By using this decision tree algorithm to automatically find out the relationship between fault type and fault data, the fault diagnosis method can implement the system for autonomous learning and improve the diagnosis rules. On the basis of unsupervised fault diagnosis, it can realize fast fault detection and correct fault diagnosis under the condition of supervision.(4) The simulation experiment. First of all, simulation tests of outlier detection based on K-mean algorithm, DBSCAN algorithm, ISODATA algorithm and IKD algorithm have been performed separately by using the data from the propulsion system of underwater vehicle, the simulation results verify the feasibility of IKD algorithm. Secondly, through the experiments, it confirmed that the method of unsupervised fault diagnosis based on grey prediction and IKD algorithm can improve the speed of fault diagnosis. Finally, combined with the decision tree algorithm, the simulation experiments of supervised underwater vehicle fault diagnosis method have been performed. The test results showed that the obtained rules from fault diagnosis method based on data mining can provide decision support for fault diagnosis of underwater vehicle and enhance the practicability of fault diagnosis system.