Fault Tree And Neural Network-based Launch Vehicle Failure Diagnosis Of Key Technologies

As strictly testing the spacecraft is an important step for succeed in launching the spacecraft, it is the key stage of the testing and launching that finding fault, reappearing fault and eliminating fault timely during the testing procedure. Our country's CZ-serial model Launch Vehicle is an advanced international Launch vehicle, and the domestic space ranges has accumulated abundant experiences in testing and faulting diagnosis of this series of Launch Vehicles. Artificial intelligent technology has maturely applied in fault diagnostic fields, hence researching on the techniques and methods of fault diagnostic for the Launch Vehicle based on artificial intelligence is of great importance and practical values for enhancing the technical skills of experts and optimizing for the efficiency and precision of the fault diagnosis of the Launch Vehicle by combined with accumulated testing data from the space range, fault diagnostic experiences and Artificial intelligent technology.According to fault diagnosis of the Launch Vehicle, this thesis proposes two fault diagnostic methods based on the fault tree with weight calculated rule and SVM algorithms Neural Networks with improved kernel function. The diagnostic steps of fault tree with weight diagnostic algorithms which based on weighted calculated rule of the systemic fault tree is established by testing measurement data link and different node weights are calculated by node weight calculated rule that is established by criticality importance for directly shown the measurement of impact that different conclusion brought by different causation and by transformed fault tree with weight into 3 databases–condition table, ruled table, conclude table. At same times, failure probability and probability of importance are derived from quantitative analysis and quality analysis of testing data. The diagnostic steps of SVM Propagation algorithms based on improved kernel function methods is geometrical trait of the kernel function and distributing trait of the testing data are analyzed and improving kernel function by multiplied conformal function to reduce redundancy vector and established SVM Neural Network. Finally, the thesis provided the theoretical foundation for further systemic development by compared with specific applied conditions of the two diagnostic algorithms. Performed the fault diagnose analyze based on fault tree and SVM Propagation algorithms by the detail applied condition and finally confirm the location and cause of the failures and put forward the diagnostic conclusion and maintenance recommendations so that ultimately bring about the Launch Vehicle fault entirely ruled out。In this way, can not only provide the reliable helpless and quickly diagnostic way but also make technician quickly study and master technologies of the fault diagnoses while shooting range always has no actual equipments.Some help is provided for the fault diagnosis of the Launch Vehicle based on the fault tree with weight calculated rule algorithms and SVM algorithms Neural Networks with improved kernel function. The experiment which input the testing data and fault data of the spaceflight shooting range into fault tree with weight calculated rule fault diagnostic algorithms and SVM algorithms Neural Networks with improved kernel function shown that the two algorithms finally identify the location and cause of the failures of the Launch Vehicle, arrived to the design request and enhanced the precision and efficiency of the fault diagnosis of Launch Vehicle.