Sensor Fault Detection Of Flight Control Electromechanical Actuation System Based On Neutral Network

With the concept of More-Electric and All-Electric,the Electromechanical Actuator(EMA)is gradually used in new More-Electric aircraft.The electromechanical actuation system reduces the complicated circuits of the traditional hydraulic actuation system,and also reduces weight and simplifies maintenance procedures,so it has many advantages.The sensor fault has a great impact on the safety of the electromechanical actuation system,so it needs special attention.However,the current research on sensor fault detection has problems such as signal with noise,small detection range,long network detection time,and signal recovery only relying on fault strength.In order to solve the above problems,this paper first established a mathematical model of the electromechanical actuation system,and summarized the sensor types and corresponding fault types of the electromechanical actuation system.Considering the problem that the inherent noise of the signal affects the fault detection,a wavelet packet denoising method for the signal is designed,and the difference from the wavelet denoising is compared.Then based on neural network,and according to different fault types,a two-level neural network fault detection system was designed.The neural network takes the time series training method as an example,compares different learning methods.At the same time,the genetic algorithm was used for optimization for the neural network,and the result is compared with the traditional BP neural network.After the fault detection is completed,considering the uncertainty of the sensor signal,a dynamic neural network is used to reconstruct the sensor signal after the fault,thereby solving the problem of input randomness and signal recovery after the non-liquid fault.The difference between the results of the nonlinear dynamic neural network and the NARX network is shown.Finally,the experimental verification of the designed scheme is completed based on the electromechanical actuation experimental platform.Based on the research in this paper,a fault detection method for sensor of flight control electromechanical actuation system based on neural network is given.The introduction of wavelet packet denoising method solves the problem of noise interference of sensor signals.The two-level neural network architecture is adopted for fault detection,and the faults are divided into two types: liquid faults and non-liquid faults.This expands the scope of fault detection,reduces the amount of information required for network detection and shortens the time required for fault detection.The use of dynamic neural network solves the problem of signal recovery based on the fault strength.