摘要
以卷积神经网络为代表的有监督深度神经网络凭借优异的提取能力,在轴承故障诊断领域得到广泛应用,不足是需要大量标签数据。在残差跳跃中嵌入注意力机制,构建注意力残差模块,并将注意力残差模块与卷积自编码器结合,得到注意力残差卷积自编码器。采用注意力残差卷积自编码器,在无监督学习下进行轴承振动信号的特征提取,显著提高轴承故障诊断能力。试验结果表明,注意力残差卷积自编码器具有优异的特征提取和选择能力,特征学习能力和故障诊断效果明显好于现有典型深度神经网络。
Supervised deep neural network represented by convolutional neural network has been widely used in the field of bearing fault diagnosis due to excellent extraction ability,but the disadvantage is that a large amount of label data is required.The attention residual convolutional autoencoder was obtained by embedding the attention mechanism in the residual jump,constructing the attention residual module,and combining the attention residual module with the convolutional autoencoder.The attention residual convolutional autoencoder was used to extract the feature of the bearing vibration signal under unsupervised learning,which significantly improved the bearing fault diagnosis ability.The experimental result shows that the attention residual convolutional autoencoder has excellent feature extraction and selection ability,and the feature learning ability and fault diagnosis effect are significantly better than those of the existing typical deep neural network.
出处
《机械制造》
2024年第5期85-90,96,共7页
Machinery
