基于多级BP神经网络的无线电罗盘多故障诊断研究

Multiple Fault Diagnosis of the Radio Compass Based on Multistage BP Neural Network

为了快速准确诊断出无线电罗盘多故障模块,针对诊断过程中的过拟合现象提出了基于提前停止法的学习率可变BP算法,并运用多级BP神经网络诊断思想,得出基于多级BP神经网络的多故障诊断方法;文中根据多级BP神经网络的多故障诊断方法,对具体的机载无线电罗盘测向电路建模仿真,将复杂的无线电罗盘电路分解为3个子网络,并对每个子网络建立合适的故障集,按顺序依次诊断得出无线电罗盘电路中的故障模块;此方法可快速准确定位电路中的多处故障模块,准确率较高且缩短了诊断时间。

In order to diagnose failure modules of radio compass accurately, the thesis proposes the Vector variable BP algorithm based on the early stop method for the over fitting phenomenon in the process of simulation, and uses the ideas of multistage BP neural network in the process of simulation. It concludes the fault diagnosis method of multistage BP neural network. The method of BP neural network theory and network tearing principle is applied to the WL -- 7 radio compass circuit in this thesis which presents a multiple fault diagnosis method based on multilevel BP neural network. Complex avionics are decomposed into several sub networks that is built up the fault sets. Finally, it gets the verification results through the Matlab simulation software in the paper. This method can quickly pinpoint Failure modules in the circuit, and shorten the diagnostic time.