基于LSTM神经网络模型的液压管路故障诊断方法  

Fault diagnosis method for pipe bending of hydraulic line based on LSTM neural network model

作　　者：孟秋静 杨钢[2] MENG Qiu-jing;YANG Gang(Sino-German Institute of Engineering,Shanghai Technical Institute of Electronics&Information,Shanghai 201411,China;School of Mechatronics and Vehicle Engineering,Chongqing Jiao-tong University,Chongqing 400074,China)

机构地区：[1]上海电子信息职业技术学院中德工程学院,上海201411 [2]重庆交通大学机电与车辆工程学院,重庆400074

出　　处：《机电工程》2022年第10期1374-1381,共8页Journal of Mechanical & Electrical Engineering

基　　金：国家自然科学基金资助项目(52178273)。

摘　　要：在航空发动机液压管路故障信号中,因含有噪声的干扰,导致针对液压管路故障的识别准确率较低,为此,提出了一种基于长短期记忆(LSTM)神经网络的航空液压管路故障诊断方法。首先,采集了航空发动机液压管路故障的振动信号,根据管路信号的特点设计并确定了LSTM模型;然后,开展了实例分析,将采集的液压管路原始振动信号加入了高斯噪声,并创建成液压管路数据集,利用所建长短期记忆神经网络模型对液压管路数据集进行了时序信息融合;最后,针对液压管路不同的故障情况,采用LSTM神经网络模型与循环神经网络(RNN)、卷积神经网络(CNN)、支持向量机(SVM)和反向传播神经网络(BPNN)等模型,进行了对比分析,验证了LSTM模型对航空液压管路故障分类的可行性和有效性。研究结果表明:在识别故障管路精度上,LSTM神经网络模型明显优于SVM和BPNN等传统的浅层神经网络模型;在抗噪性能方面,LSTM明显优于近年来所用的CNN和RNN诊断方法;这说明LSTM神经网络故障诊断方法对航空发动机外部液压管路故障诊断具有适应性和实用性。Aiming at the problem that low recognition accuracy caused by noise interference in the fault signals of aircraft engine hydraulic pipelines,a fault diagnosis method based on long short-term memory(LSTM)neural network was proposed.Firstly,vibration signals were collected for the hydraulic fault pipeline of aeroengine,and LSTM model was designed and determined according to the characteristics of pipeline signals.Then,the original vibration signal of hydraulic pipeline was added to Gaussian noise through case analysis,and the hydraulic pipeline data set was created,and the time sequence information of hydraulic pipeline data set was fused by the established long and short-term memory neural network model.Finally,the long and short-term memory neural network model was compared with the recurrent neural network(RNN),convolutional neural network(CNN),support vector machine(SVM)and back propagation neural network(BPNN)models to analyze the fault diagnosis results of hydraulic pipelines.The results show that both short-term and long-term memory neural network model on the identification accuracy of the fault line is better than SVM and BPNN traditional shallow neural network model,the anti-noise performance is superior to the CNN and RNN diagnosis methods in recent years,explain LSTM neural network fault diagnosis method for aeroengine hydraulic circuit fault diagnosis has the external adaptability and practicability.

关 键 词：长短期记忆神经网络模型 流固耦合振动特性 振动信号全局特征 高斯噪声 健康状态识别 时间信息融合 

分 类 号：TH137.86[机械工程—机械制造及自动化] 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]