Navigation Sensors Fault Diagnosis Based On Information Entropy

Under the demand of development of aviation and space technology, it is becoming a trend for the vehicle such as aeroplane and naval vessel etc that they are equipped with many different navigation equipments, which makes the Integrated Navigation System (INS) with multifunction and high precision to be crucial navigation equipment. INS is responsible for ship navigating, orientating and posture controlling, so its reliability concern directly the security of the vehicle and stuff. Therefore the technology of Fault Detection and Diagnosis (FDD) for navigation equipment is drawing more and more people's attention, which makes the research of FDD ensuring INS works normally significant.With the navigation sensors of INS as research objects and the technology of feature extraction, the thesis apply information entropy to FDD of navigation sensors. Base on the analysis of different information entropy measures, this paper priority research the method extracting the feature of wavelet packet-energy entropy and then put it into the feature exertion of gyro fault. That will be better for recognizing the fault state from the extraction of sign and the FDD method based on information entropy is realized.Firstly, statuses of domestic and foreign research on FDD and information entropy theory are analyzed based on opinions from lots of reference materials. As the base of the technology of FDD, the theory of the Ship INS and correlative navigation equipments are presented simply.Secondly, theories and application on information entropy are presented and information entropy feature exertions in different space are defined. For FDD realized more easily, the wavelet analysis and its application on FDD are defined based on information entropy. Then this thesis combines the wavelet analysis with information entropy and researches a wavelet entropy measure with mutual characters of the wavelet analysis and information entropy. The superiority of the wavelet entropy measure on feature exertion of FDD is proving by example. Lastly, this thesis prior researches the eigenvector exertion of wavelet packet-energy entropy (WP-EE) and then summarizes the exertion process. Based on the eigenvector exertion, the fault model sorter on BP neural network is constructed. The research is artificially analyzed by using inertia equipment gyro. The three layers BP neural network is selected and the eight dimensional eigenvector of wavelet packet-energy entropy (WP-EE) is served as the network input. The BP neural network is trained and trained BP neural network is test. The simulation results indicate that this method mentioned in this thesis can diagnose faults effectively.