Research And Application Of Fault Diagnosis Based On Simulink/Arduino

With the development of modern industry,the fault in aerospace field have been become higher complicated、nonlinear and strongly coupled,the fault diagnosis method of shallow neural network is difficult to achieve good results,deep learning has more powerful feature extraction ability than shallow neural network.When the theory is applied to practical engineering,algorithm needs to be constantly modified to meet the demand,Hardware In Loop simulation can greatly reduce the time of this process.In order to solve application of deep learning in fault diagnosis of modern spacecraft.The models of rocket swing servo control system and satellite attitude control system were established,carried out failure mode analysis and collected data;By changed the parameters of weight matrix、activation function、training method and network layers,studied the optimal fault diagnosis structure of four deep learning algorithms;The accuracy of the four algorithms is 94.2%-98.8%,the method of adjusting learning rate by loss function improves the accuracy by 17.9%,the activation function adopts the improved linear rectification unit,and the accuracy is increased by 4%.The results show that,deep learning algorithm can show strong identification ability for modern spacecraft faults through improvement.In order to solve application of fault diagnosis system to practical problems,set up a physical model,took Simulink as host computer,PD control algorithm and two kinds of fault diagnosis algorithm were established.Through hardware in loop simulation mode,the parameters could be adjusted online,reduced the model construction time,observed the performance of fault diagnosis algorithm.In the hardware in loop simulation experiment,one dimensional convolution neural network fault diagnosis accuracy is 52.5%,the accuracy of the deep neural network initialized by contrast divergence algorithm is 87.5%.The results show that for the spacecraft data with discontinuous,nonlinear,strong coupling and few parameters characteristics,multi parameter network has good engineering performance.