首页|不平衡数据下基于改进门控卷积网络的轴承故障诊断

不平衡数据下基于改进门控卷积网络的轴承故障诊断

扫码查看
原文链接
  • 万方数据
  • 维普
深度学习在滚动轴承故障诊断中具有广泛的应用,然而,现实中的监测数据往往具有不平衡性,这就会对模型的诊断性能产生很大影响.因此,提出一种基于改进门控卷积神经网络(Improved Gated Convolutional Neural Network,IGCNN)的故障诊断方法,用于数据不平衡条件下的故障诊断.首先,提出改进门控卷积层以增强特征提取能力,通过批量归一化技术提高模型的泛化能力.然后,使用标签分布感知边界(Label-distribution-aware Margin,LDAM)损失函数提高模型对少数类的敏感度,减小数据不平衡对模型的影响.将所提算法应用在两组故障轴承数据上,在数据不平衡率为20:1的情况下,所提算法仍然可达到92.71%和94.47%的故障识别率,而对比的其他主流深度学习模型在该情况下只有60%～72%的准确率,表明所提方法在数据集严重不平衡情况下具有很强的诊断能力和鲁棒性.
Bearing Fault Diagnosis Based on Improved Gated Convolutional Network with Imbalanced Data
Intelligent methods based on deep learning have been widely used in the fault diagnosis of rolling bearings.However,the actual data from industrial plant is severely imbalanced,which will greatly affect the diagnostic performance of the model.To solve this problem,a fault diagnosis method based on improved gated convolutional neural network(IGCNN)is proposed.Firstly,an improved gated convolution layer is proposed for feature extraction,with the batch normalization layer applied in this layer to enhance the generalization performance of the model.Then,the label-distribution-aware margin(LDAM)loss function is employed to raise the sensitivity of the model to the minority class and mitigate the influence of imbalanced data on the model.The proposed method is applied to two sets of faulty bearing data.It is found that when the imbalance ratio of the data sets is 20:1,the recognition accuracy of the proposed method can still achieve 92.71%and 94.47%.While the other compared mainstream deep learning models only achieve 60%-72%accuracy.It shows that the proposed method has strong diagnostic capability and robustness for the datasets with severe imbalance.

fault diagnosisimbalanced dataimproved gated convolutional neural networklabel-distribution-aware margin lossrolling bearings

郗昌盛、梁小夏、田少宁、杨杰、冯国金、甄冬

展开 >

河北工业大学 机械工程学院,天津 300401

河北工业大学 先进装备研究院有限公司,天津 300401

故障诊断 数据不平衡 改进门控卷积神经网络 标签分布感知边界损失函数 滚动轴承

国家自然科学基金资助项目天津市自然科学基金资助项目2022年河北省春晖计划资助项目

5227510121JCZDJC00720E2022202101

2024

10.3969/j.issn.1006-1355.2024.04.023

噪声与振动控制
中国声学学会

噪声与振动控制

CSTPCD北大核心
影响因子：0.622
ISSN：1006-1355
年,卷(期)：2024.44(4)
  • 5