基于深度学习与SAE网络的火箭推力下降故障诊断

A Fault Diagnosis Method for Launch Vehicle Thrust Descent Based on Deep Learning and Stacked Auto Encoder

针对运载火箭动力系统在发动机推力下降故障诊断中存在的推力下降程度及故障时间测算不精确的问题,提出了一种基于深度学习的故障诊断方法。不同时刻及程度的推力故障下,利用运载火箭六自由度运动学模型生成的过载信息作为故障训练样本,采用堆栈自动编码器方法训练网络,利用训练好的网络辨识发动机推力下降程度,带入六自由度仿真模型中可以实现在线故障诊断。数字仿真证实:该方法可以对火箭发动机的不同时刻与不同推力下降程度的推力损失进行故障诊断,与普通神经网络方法相比,精确性更高。

To improve thecalculation accuracy of the thrust drop and the fault time during the engine thrust descent fault of launch vehicle power system during flight,a fault diagnosis method based on deep learning was proposed.By the six-degree-of-freedom kinematics model of the launch vehicle,the overload information was generated as the fault training sampleunder thrust faults at different times and with different degrees of drop.The stack autoencoder method was used to train the network,and the degree of engine thrust decline was identified by the trained network.Theonline fault diagnosis could be realized by applyingit in the 6-DOF simulation model.The digital simulation showed that the method in this paper could diagnose the thrust loss of launch vehicle engine and the accuracy was better than that of the traditional neural network method.