Fault Diagnosis Algorithm,Experiment And Application Of UAV Actuator Systems Based On Deep Neural Networks

With the development and popularization of multi-rotor aircraft,the fault diagnosis of the multi-rotor actuator system has not stopped.At present,there are still three problems to be solved:First,the fault diagnosis of the actuator system is mostly based on model analysis.It takes a lot of time and effort to build an accurate model.Second,most of the current methods for actuator systems cannot be diagnosed or the accuracy is not high.Third,the general method can only detect whether the UAV actuator system is faulty.It is not possible to pinpoint the fault of the specific axis of the drone.In order to solve these problems,this study proposes a method for intelligent fault diagnosis using deep neural networks,which can accurately locate faults of a single axis when diagnosing minor faults in the actuator system.The main contents of the paper include:(1)In order to realize the deep neural network algorithm of the UAV actuator system,this paper firstly built the experimental six-rotor UAV,and based on this,designed the actuator system failure experiment and constructed the initial data set.(2)This paper proposes the application of intelligent diagnosis based on deep neural network to the fault diagnosis of UAV actuators.The initial BP neural network algorithm has a 5-fold cross-validation accuracy of 94.01%.In the later research,the deep neural network model was further improved,and the 1D-CNN model and the Bi-GRU(Gated Recurrent Unit)neural network model were designed.Finally,the average accuracy of the two algorithms after the 5-fold cross-validation diagnosis was 96.90%and 96.73%.In addition,this paper proposes a circular convolutional neural network algorithm,which combines the advantages of cyclic neural network and convolutional neural network.The average fault diagnosis accuracy of the final GRU+CNN hybrid model is 97.33%.This result should be slightly better than the neural network model constructed above.(3)This paper designs and implements the UAV health management platform to monitor the status of the UAV in real time,and embeds the method obtained in the previous research into the designed software platform,and applies the deep neural network fault diagnosis model to the UAV fault diagnosis.In summary,this thesis deeply studies the deep neural network fault diagnosis of the UAV actuator system,designs the algorithm,completes the experimental verification,and develops the health management application software platform.For the fault diagnosis research on the UAV actuator system,the proposed way has a certain theoretical significance and application value.