基于CNN+LSTM神经网络的电液伺服阀故障预测  

Fault Prediction of Electro-hydraulic Servo Valve Based on CNN+LSTM Neural Network

作　　者：贾春玉[1] 康凯旋 高伟 杨东 陈立娟 艾超[1] JIA Chun-yu;KANG Kai-xuan;GAO Wei;YANG Dong;CHEN Li-juan;AI Chao(School of Mechanical Engineering,Yanshan University,Qinhuangdao,Hebei 066004;School of Mechanical Engineering,Nanjing Institute of Technology,Nanjing,Jiangsu 211167)

机构地区：[1]燕山大学机械工程学院,河北秦皇岛066004 [2]南京工程学院机械工程学院,江苏南京211167

出　　处：《液压与气动》2020年第12期173-181,共9页Chinese Hydraulics & Pneumatics

摘　　要：针对电液伺服阀故障预测中故障类型复杂多变、早期故障较弱、时间序列难以处理等问题,构建了卷积神经网络(CNN)和长短期记忆神经网络(LSTM)相结合的电液伺服阀故障预测模型,取代人工特征选择和提取,解决故障预测的时序问题。以G761型电液伺服阀为例,利用AMESim软件对伺服阀阀芯磨损和孔板堵塞故障数据集进行了仿真,并用仿真故障数据验证了模型的预测精度。同时将LSTM,CNN,CNN+LSTM 3种模型针对电液伺服阀故障预测诊断的精度进行对比,CNN+LSTM故障预测模型训练时间更快,得到更高的预测精度,具有更好的适应性。Aiming at the problems of complicated and changeable fault types,weak early faults,and difficult time series in the fault prediction of electro-hydraulic servo valves,an electro-hydraulic servo comning convolutional neural network(CNN)and long-term and short-term memory neural network(LSTM)is constructed.The valve failure prediction model replaces manual feature selection and extraction to solve the timing problem of failure prediction.Taking the G761 electro-hydraulic servo valve as an example,the AMESim software is used to simulate the servo valve core wear and orifice clogging fault data set,and the simulation fault data is used to verify the prediction accuracy of the model.At the same time,the four models of LSTM,CNN,and CNN+LSTM are compared for the accuracy of the fault prediction and diagnosis of the electro-hydraulic servo valve.The results show that the CNN+LSTM fault prediction model has faster training time,higher prediction accuracy,and better Adaptality.

关 键 词：电液伺服阀 故障预测 卷积神经网络 长短期记忆神经网络 

分 类 号：TH137[机械工程—机械制造及自动化]