摘要
针对传统行星齿轮箱故障诊断网络的人工特征提取困难的问题,提出了将小波包函数约束的自适应小波包分解(AWPD)卷积核融入CNN网络,构建行星齿轮箱端到端故障诊断网络MSAWPD-CNN。首先将2种频率尺度的AWPD卷积核融入到诊断网络的第1层,使诊断网络聚焦于可以有效表达行星齿轮箱故障的时频特征;其次将2种频率尺度的时频特征分别通过2层1D卷积进行特征压缩;最后将压缩后的特征堆叠送入骨干CNN进行深层特征提取与故障模式分类。实验结果表明,所提出的网络可以有效聚焦时频特征,加速训练结果的收敛,并且在测试集中诊断准确率最高。
Aiming at the difficulty of artificial feature extraction in traditional planetary gearbox fault diagnosis network,an end-to-end planetary gearbox fault diagnosis network MSAWPD-CNN is constructed by integrating the convolution kernel of adaptive wavelet packet decomposition(AWPD)constrained by wavelet packet function into CNN network.Firstly,the AWPD convolution kernels of two frequency scales are incorporated into the first layer of the diagnostic network,enabling the diagnostic network to concentrate on the time-frequency features that can effectively represent the faults of planetary gearboxes;secondly,the time-frequency features of the two frequency scales are respectively compressed via two layers of 1D convolution;finally,the compressed features are stacked and delivered to the backbone CNN for deep feature extraction and fault pattern classification.The experimental results indicate that the network proposed in this paper can effectively focus on the time-frequency features,expedite the convergence of the training results,and achieve the highest diagnostic accuracy in the test set.
作者
李斌
张砦
李泓锟
LI Bin;ZHANG Zhai;LI Hongkun(College of Automation Engineering,Nanjing University of Aeronautics and Astronautics,Nanjing 211106,China)
出处
《机械与电子》
2025年第7期74-80,共7页
Machinery & Electronics
关键词
行星齿轮箱
故障诊断
端到端
自适应小波包分解
卷积神经网络
planetary gearbox
fault diagnosis
end-to-end
adaptive wavelet packet decomposition
convolutional neural network
作者简介
李斌(1996-),男,江苏宿迁人,硕士研究生,研究方向为行星齿轮箱故障诊断;通信作者:张砦(1980-),男,安徽歙县人,博士,副教授,研究方向为航天器测试、智能故障诊断和数字系统容错,E-mail:wolnyzhang@nuaa.edu.cn;李泓锟(2000-),男,四川绵阳人,硕士研究生,研究方向为行星齿轮箱故障诊断。
