摘要
随着非平稳信号的时频故障特征表示技术的优势以及卷积神经网络(CNN)的出现,转子碰摩故障诊断已经达到了高精度。然而,基于CNN的故障诊断方法有许多缺点,提出了一种基于轻量级卷积神经网络(GHSQNet)与时频图相结合的转子碰摩故障诊断方法。该方法能够在保证故障诊断精度的同时占用更小的内存。实验结果表明,与其他轻量级卷积神经网络相比,提出的模型性能表现更加优越。识别准确率达到了99.39%,而模型大小只有1.07 MB,在移动端有很高的实用价值。
With the advantages of time-frequency fault feature representation technology of non-stationary signals and the appearance of Convolutional Neural Network(CNN),the rotor rub-impact fault diagnosis has reached high precision.However,the fault diagnosis method based on CNN has many shortcomings.A rotor rub-impact fault diagnosis method based on lightweight convolutional neural network(GHSQNet)and time-frequency graph is proposed,which can maintain the fault diagnosis accuracy and occupy less memory at the same time.Experimental results show that compared with other lightweight convolutional neural networks,the proposed model performs better.The recognition accuracy rate is 99.39%,and the size of the model is only 1.07 MB,which is of great practical value on mobile terminal.
作者
闻章
于洋
WEN Zhang;YU Yang(School of Information Science and Engineering,Shenyang University of Technology,Shenyang 110870,China)
出处
《电子设计工程》
2024年第15期161-165,171,共6页
Electronic Design Engineering
关键词
故障诊断
声发射
碰摩故障
卷积神经网络
fault diagnosis
acoustic emission
rub-impact fault
Convolutional Neural Network
