Research On Fault Detection Methods Of The Electric Servo Actuator Based On Neural Networks

The electric servo actuator is the power source of the unmanned autonomous vehicle’s rudder surface control system.Since the precise output and reliable control of the actuator are essential to the vehicle’s safe and stable flight,it has great practical significance to study the rapid and accurate fault detection of the actuator.The traditional electric actuator fault detection mainly analyzes the output value of its sensor to establish the mapping relationship between the fault and the detected output value.This method relies on the mathematical model of the actuator.In the actual system,it is difficult to obtain an accurate model of the actuator due to the complexity of its structure and the uncertainty caused by the sensor interference.To solve this problem,this paper researches a set of neural networks fault detection model algorithm,which does not need to obtain the precise model of the servo,and can automatically perform feature extraction and fault detection.First,this paper establishes an actuator fault detection model based on the LSTM neural network.For the servo fault data,the actual fault test data of the aircraft servo is used as the neural network training data set,and then preprocessed.On 15 types of faults for three different types of servos,including pitch,left roll and right roll respectively,this model introduces different types of optimization algorithms for training and parameter optimization,and sets the number of neurons in different hidden layers to compare with time step,thus selecting the best performance LSTM neural network configuration.The results show that the detection accuracies of all three servos reach high level,and the LSTM model can effectively carry out fault detection.Secondly,by introducing the convolutional layer and the pooling layer for feature pre-extraction and optimizing neural network parameters,a CNN-LSTM neural network actuator fault detection model is established.The convolutional layer and pooling layer of the convolutional neural networks can amply extract the data features,reduce the data dimension,and accelerate the training speed.Therefore,to select the optimal hyper-parameters,a convolution layer and a pooling layer are added for feature pre-extraction before the LSTM layer of the original model,and the model is trained and tested under different hyper-parameters,such as the number of convolution kernels,the number of iterations,the batch size,the size of the convolutional layer and the pooling layer.The results show that,in the case of the same data set,the CNN-LSTM model not only improves the accuracy of fault detection,but also improves the training speed for the three types of servos.Finally,through simulation verification and comparative analysis,the designed neural networks fault detection model is tested for its anti-noise capability and robustness under different SNR noise conditions.Gaussian white noise is added to the LSTM model and the CNN-LSTM model to verify the change of the fault detection accuracy of the two models under different signal-to-noise ratio noise conditions.The BN layer is added to the input of the LSTM network layer,the Dropout layer is added to the output of the LSTM network layer,and a comparative analysis is carried out.The results show that the LSTM and CNN-LSTM models have better anti-noise capability,and the BN layer and Dropout layer(especially the BN layer)both improve the antinoise capability and robustness of the three servos under different signal-to-noise ratio noise conditions.