摘要
针对滚动轴承故障样本稀疏、卷积神经网络(convolutional neural network, CNN)池化层效率低的问题,提出一种基于胶囊网络的小样本学习方法模型。基于孪生神经网络,通过相同或者不同类别的样本对进行特征学习,根据特征之间的差异进行故障分类。在标准的凯斯西储大学(Case Western Reserve University, CWRU)轴承故障数据集进行的实验结果表明,该模型在有限数据样本下对故障诊断更为有效。通过添加不同幅值能量的高斯白噪声开展实验,其结果表明,所提方法在抗噪性方面具有优势。
Aiming at the problems of the sparse rolling bearing faults samples and the low efficiency of convolutional neural network pooling layer,the few-shot learning approach model based on capsule network was proposed.Based on the Siamese neural network,feature learning was performed through same or different sample classes,and faults were classified,according to differences between features.Experimental results of the standard Case Western Reserve University(CWRU)bearing fault dataset demonstrat that the proposed model is more effective for fault diagnosis under limited data samples.Simultaneously,experiments were carried out by adding Gaussian white noise with varying amplitude energy.The advantage in noise immunity is revealed using the proposed method.
作者
汪祖民
聂晓宇
王颖洁
季长清
秦静
WANG Zu-min;NIE Xiao-yu;WANG Ying-jie;JI Chang-qing;QIN Jing(College of Information Engineering,Dalian University,Dalian 116622,China;College of Physical Science and Technology,Dalian University,Dalian 116622,China;College of Software Engineering,Dalian University,Dalian 116622,China)
出处
《计算机工程与设计》
北大核心
2023年第4期1259-1266,共8页
Computer Engineering and Design
基金
国家自然科学基金项目(62002038)
大连市科技创新基金项目(2020JJ26SN058)。
关键词
滚动轴承
故障诊断
小样本学习
胶囊网络
孪生神经网络
抗噪性
卷积神经网络
rolling bearing
fault diagnosis
small sample learning
capsule network
twin neural network
noise resistance
convolutional neural network
