摘要
针对卷积神经网络难以处理时间序列数据和循环神经网络难以提取数据深层特征的问题,提出了一种基于深度卷积网络和循环神经网络相结合的滚动轴承故障诊断方法。首先,使用格拉姆角场(GAF)编码将一维轴承振动信号构造为时序图像并划分为训练集、验证集和测试集;然后,将训练集和验证集输入VGG16模型进行特征提取,将提取到的特征输入RNN进行训练;最后,用测试集验证CNN-RNN模型的有效性。XJTU-SY和CWRU轴承数据集的试验结果表明:相对于HHT和GASF编码方法,GADF编码方法对原始信号故障特征的表达能力更强;相对于独立的CNN模型或RNN模型,CNN-RNN模型的识别效果更好;GADF编码方法与CNN-RNN模型相结合时具有更高的识别率。
Aimed at the problems that the convolutional neural network is difficult to process the time series data and the recurrent neural network is difficult to extract the deep features of data,a fault diagnosis method for rolling bearings is proposed based on combination of deep convolutional network and recurrent neural network.Firstly,Gramian Angle Fields(GAF)coding is used to construct the one-dimensional bearing vibration signal as a time series image and divide it into training set,validation set and test set;then,training set and validation set are input to VGG16 model for feature extraction,and the extracted features are input into RNN for training;finally,the test set is used to verify the effectiveness of CNN-RNN model.The test results of XJTU-SY and CWRU bearing data sets show that:compared with HHT and GASF coding methods,GADF coding method has a stronger ability to express the fault features of original signal;compared with independent CNN model or RNN model,CNN-RNN model has better recognition effect;GADF coding method has higher recognition rate when combined with CNN-RNN model.
作者
姚立
孙见君
马晨波
YAO Li;SUN Jianjun;MA Chenbo(College of Mechanical and Electrical Engineering,Nanjing Forestry University,Nanjing 210037,China)
出处
《轴承》
北大核心
2022年第2期61-67,共7页
Bearing
基金
国家重点研发计划资助项目(2018YFB2000800)。
关键词
滚动轴承
卷积神经网络
门控循环单元
格拉姆角场
故障诊断
rolling bearing
convolution neural network
gated recurrent unit
Gramian angular fields
fault diagnosis
