摘要
为满足深度学习与航天类故障诊断相结合的实验教学需求,应用倒立摆模型构建火箭姿态控制执行器故障诊断演示平台。利用半物理仿真模式,在Simulink中在线调参控制实物平台,演示火箭姿态的不同故障,无需编写代码;采用不同结构的卷积神经网络(Convolutional Neural Networks,CNN)进行故障诊断,比较其性能差异。平台的离线和实物仿真均能很好地演示火箭姿态执行器的故障问题,也显示了深度卷积神经网络相比浅层神经网络在故障诊断的优势。该演示平台应用于航天类故障诊断方法教学,既让故障问题从抽象到实际,也让深度学习故障诊断理论更加直观。
In order to meet the experimental teaching needs of the combination of deep learning and aerospace fault diagnosis,this paper construction of fault diagnosis demonstration platform for rocket attitude control actuator by inverted pendulum.By the semi-physic,online parameter adjustment is used to control the physical platform in Simulink,demonstrate different faults of rocket attitude,need not to write code;Different structure of CNN is used for fault diagnosis,performance differences were compared.Offline and physical simulation of the platform,all of them can demonstrate the fault of rocket attitude actuator very well,shows the advantage of CNN in fault diagnosis compared with shallow neural network.This demonstration platform is applied to the teaching of aerospace fault diagnosis methods,it not only makes the fault problem from abstract to practical,but also makes the deep learning fault diagnosis theory more intuitive.
作者
倪平
闻新
Ni Ping;Wen Xin(Shenyang Aerospace University,Shenyang 110136)
出处
《现代制造技术与装备》
2020年第12期8-11,共4页
Modern Manufacturing Technology and Equipment
