摘要
为了提高齿轮箱故障诊断能力,采用加权动态网络构建了一种振动传感器信号加权动态网络。通过加权动态处理方式来实现跨层连接特征图小波包系数矩阵的权重调节,完成各类频带信息的自适应加权计算,使密集连接网络具备更强的变转速工况表征性能,实现故障识别率的显著提升。研究结果表明:组成加权动态网络的密集块与过渡层结构。各其中一层可以实现卷积降维的效果,另一层则可以实现池化的效果。各密集块都包含了6个卷积模块、初始输入端卷积层和输出池化层以及全连接层。加权动态网络达到了最优性能,进行小波包系数加权动态计算时获得了最快收敛速率,并形成了稳定的最高值。采用本文加权动态网络可以对齿轮箱实现故障信号的准确诊断。
In order to improve the ability of gearbox fault diagnosis,a vibration sensor signal weighted dynamic network is constructed using the weighted dynamic network.The weighted dynamic processing method is used to realize the weight adjustment of the wavelet packet coefficient matrix of the cross-layer connection feature map,and the adaptive weighting calculation of various frequency band information is completed,so that the densely connected network has stronger performance of variable speed condition representation,and the fault recognition rate is significantly improved.The results show that the dense block and transition layer structure constitute the weighted dynamic network.One layer can achieve convolution dimension reduction,and the other layer can achieve pooling.Each dense block contains 6 convolution modules,an initial input convolution layer,an output pooling layer and a fully connected layer.The weighted dynamic network achieves the best performance,and the fastest convergence rate is obtained and the highest stable value is formed when the weighted dynamic calculation of the wavelet packet coefficients is carried out.The weighted dynamic network in this paper can be used to accurately diagnose the gearbox fault signal.
作者
张莉
赵江招
孟琨泰
高健
李峰
ZHANG Li;ZHAO Jiangzhao;MENG Kuntai;GAO Jian;LI Feng(Department of Information Engineering,Hebei Vocational College of Mechanical and Electrical Technology,Xingtai Hebei 054002,China;College of Information Science and Engineering,Hebei University of Science and Technology,Shijiazhuang 050018,China;Intelligent Sensor Network Application Technology R&D Center,Hebei Vocational College of Mechanical and Electrical Technology,Xingtai Hebei 054002,China;Cloud Technology Intelligent Product Innovation Application and R&D Center,Hebei Vocational College of Mechanical and Electrical Technology,Xingtai Hebei 054002,China;Department of Electrical Engineering,Hebei Polytechni of Mechanical and Electrical Technology,Xingtai Hebei 054002,China)
出处
《机械设计与研究》
CSCD
北大核心
2023年第3期97-100,108,共5页
Machine Design And Research
基金
河北省科学技术研究项目(20200909)。
关键词
密集连接网络
故障诊断
特征学习
小波包变换
齿轮箱
densely connected network
fault diagnosis
feature learning
wavelet packet transform
gear box
作者简介
张莉(1985-),女,硕士,讲师,主要从事软件技术应用,已发表论文4篇。E-mail:xiaozhangbing2022@163.com。
