基于CNN的飞机升降舵液压系统故障诊断  

Research on fault diagnosis of aircraft elevator hydraulic system based on CNN

作　　者：张鹏 李广道 ZHANG Peng;LI Guangdao(Engineering Techniques Training Center,CAUC,Tianjin 300300,China;College of Electronic Information Engineering,CAUC,Tianjin 300300,China)

机构地区：[1]中国民航大学工程技术训练中心,天津300300 [2]中国民航大学电子信息与自动化学院,天津300300

出　　处：《中国民航大学学报》2023年第1期35-40,52,共7页Journal of Civil Aviation University of China

摘　　要：针对民机液压系统故障诊断对专家经验的依赖和深层网络诊断模型退化的问题,提出改进的一维卷积神经网络算法。首先,将仿真故障数据直接输入一维卷积神经网络,再对卷积层使用残差块机制来提高信息的利用率,引入挤压与激励网络对卷积层特征向量进行加权表示,从而减少无效信息,达到抗干扰的效果;其次,使用一维全局均值池化层处理末层信息,降低神经网络参数的数量和诊断时间;最后,为了验证所提方法的有效性和实用性,通过实验室仿真平台得到的飞机升降舵液压系统故障数据对该方法进行测试,同时与主流算法进行对比。实验结果表明:本文所提方法测试集准确率高达99.3%,相比其他网络在液压系统故障诊断方面准确率和泛化性有明显的提升,在加入20%噪声环境下本文网络相比传统卷积网络诊断准确率提升4.4%,且具有较强的实用性。Aiming at the dependence of expert experience on fault diagnosis of civil aircraft hydraulic system and the degradation of deep network diagnosis model,an improved one-dimensional convolutional neural network algorithm is proposed.Firstly,the simulated fault data is directly inputted into the one-dimensional convolutional neural network,and then the residual block mechanism is used to improve the utilization of information in the convolutional layer.The squeeze and excitation network is introduced to weight the feature vectors of the convolutional layer,so as to reduce the invalid information and achieve anti-interference effect.Then one-dimensional global average pooling layer is used to process the information of last layer to reduce the neural network parameters and diagnosis time.Finally,in order to verify the effectiveness and practicability of the proposed method,this method is tested by the fault data of the aircraft elevator hydraulic system obtained by the laboratory simulation platform,and compared with the mainstream algorithm.The experimental results show that the test set accuracy of the proposed method is as high as 99.3%.Compared with other networks,the accuracy and generalization of fault diagnosis of hydraulic system are significantly improved.With 20%noise added environment,the accuracy of the proposed network is 4.4%higher than that of the traditional convolutional network,and it has strong practicability.

关 键 词：故障诊断 民机液压系统 卷积神经网络 残差结构 全局均值池化 挤压与激励网络 

分 类 号：V267[航空宇航科学与技术—航空宇航制造工程] TP183[自动化与计算机技术—控制理论与控制工程]