Research On Fault Diagnosis Method Of Aviation Auto-Transformer Rectifier Unit

With the increasing use of high-power devices and power electronics in avionic systems,the failure rate and repair rate of avionic equipment have rapidly increased.In order to ensure the maintainability and continued airworthiness of avionic equipments throughout a flight cycle,an efficient and reliable fault diagnosis process used to timely and automaticly detect and handle the faults or potential failures in the airborne electronic equipment is of great significance.The airborne autotransformer rectifier unit is a crucial part in the avionic secondary power supply system.Therefore,this thesis carries out research on the fault diagnosis method of ATRUThis thesis firstly introduces the working principle and some typical structures of ATRU.Based on the study of the failure modes and effects of ATRU,a fault tree model is established to clarify the fault mechanism of the whole system.Then,the fault simulation model of a P-type 12-pulse ATRU is constructed to verify each failure mode,which is globally used as a study object in this thesis.A C-language-based program has also been written in PLECS to realize automaticly faults injection and simulations under all failure conditions.With the populization of artificial intelligence recently,artificial neural network algorithms are widely used in fault diagnosis of complex nonlinear systems.In chapter 3,a Radial Basis Function Neural Network(RBFNN)based on Fast Fouria Transformation technique is ultilized to realize fault diagnosis of ATRU.The RBFNN model is constructed and trained by Matlab ANN toolbox and the hidden layer nodes are optimized according to the training parameters MSE.The experiment shows that the FFT-bases RBFNN can classify the complex fault modes of ATRU.In order to further improve the adaptive ability and classification accuracy of diagnostic model,chapter 4 proposes a two-level diagnosis based on fusion features.The ATRU multi-source signal feature system is constructed by mathematical modeling and simulation analysis of several electrical signals.Then,the two-level RBF network groups are constructed considering the ATRU topology characteristics.Several testing experiments are designed to compare the classification performance of the methods in this thesis and other existed method.The results show that the two-level model based on multi-source features has better comprehensive classification performance.