摘要
针对基于修正线性单元(ReLU)的滚动轴承故障诊断方法导致分类不准确的问题,本文提出了一种基于bReLU激活函数改进的卷积神经网络(CNN)和长短时记忆神经网络(LSTM)的滚动轴承故障模型。该网络模型首先在CNN网络使用bReLU激活函数完成自适应特征提取,并在卷积层和激活函数之间加入批量标准化层(BN),最后叠加LSTM网络对时序特征进行学习。该模型在凯斯西储大学的故障数据集的准确率可以达到99%以上,较于其它类型的CNN-LSTM故障模型准确率更高,训练时长减少66.7%。实验表明所提出的模型更加适用于工业设备轴承的故障诊断。
Aiming at the problem that the bearing fault diagnosis method based on Rectified Linear Unit(ReLU)leads to inaccurate classification,a bearing fault diagnosis model based on Convolutional Neural Network(CNN)and Long Short-Term Memory(LSTM)of bReLU activation function was proposed.The CNN network based on bReLU was used to adaptively extract features,and the batch standardization layer(BN)was added between the convolution layer and the activation function.Finally,LSTM layer was used to learn timing characteristics.The accuracy of improved method can reach more than 99%on the fault dataset of Case Western Reserve University,which is higher than other CNN-LSTM models,and the training time was reduced by 66.7%than other CNN-LSTM fault methods.The experiments show that the proposed improved method is more suitable for bearing fault diagnosis of industrial equipment.
作者
项璇
曹少中
杨彦红
XIANG Xuan;CAO Shaozhong;YANG Yanhong(School of Information Engineering,Beijing Institute of Graphic Communication,Beijing 102600,China)
出处
《北京印刷学院学报》
2023年第3期22-26,共5页
Journal of Beijing Institute of Graphic Communication
基金
北京市自然基金和北京市教委联合项目(KZ202010015021)
大兴区科技发展计划项目(KT201901162)
北京印刷学院科研项目(Eb202103,Ec202002)。
关键词
滚动轴承
故障诊断
卷积神经网络
长短时记忆网络
rolling bearing
fault diagnosis
Convolutional Neural Network(CNN)
Long Short-Term Memory(LSTM)
