| With the continuous expansion of the aviation industry,the reliability and flight safety of the aircraft have become more and more important.At this time,the FCS(flight control system),which is the core of the aircraft,is still quite complex.Sex puts forward higher demands.At the same time,with the rapid development of sensor technology and storage technology today,a large amount of historical data during aircraft operation can be accurately measured and completely stored.Therefore,how to use data-driven methods to diagnose faults in flight control systems to ensure The safe and stable operation of aircraft is of great research value.Combined with the FCS project,this paper mainly studies the sensor fault diagnosis of the flight control electromechanical actuation system cross-linked by the flight control system.Four nonlinear models of flight control system sensor faults in different states are established,and the signal sources of sensor faults are simulated.The signal samples of different fault types are denoised,feature extraction and neural network training in turn.This paper mainly completes the following tasks:(1)Based on the flight control signal data of the external simulation excitation library in the project,the simulation engineering software is used to establish the sensor fault simulation model of the flight control electromechanical actuation system.Each output signal data is denoised,and the characteristic parameters of different fault types are extracted to provide relevant fault sample data for subsequent neural network training.(2)Firstly,the classical BP neural network is selected for preliminary analysis of fault detection,the basic neural network fault detection model is constructed,and various performances of the network model are analyzed.Then,based on the genetic algorithm(GA)optimization neural network and radial basis function(RBF)neural network model,K-means clustering optimization algorithm was introduced to construct the K-RBF network model.Finally,the ability of the optimized neural network to diagnose and classify the corresponding faults is compared and analyzed,which improves the fault diagnosis accuracy of the flight control system to a certain extent.(3)Using the memory and feedback characteristics of the dynamic time series neural network itself,a fault recovery model for the loss of the original signal is established.Further,according to the characteristics of flight control fault signals,a nonlinear active self-regression model(NARX)dynamic neural network model is introduced to restore the sensor signal.At the same time,it compares with the typical nonlinear dynamic neural network and analyzes the simulation experiment,and optimizes the constructed fault recovery network model.Finally,through the neural network algorithm model,the signal after the fault recovery can be output to replace the fault simulation sensor signal to ensure the normal operation of the system. |
