改进深度卷积神经网络及其在变工况滚动轴承故障诊断中的应用

张西宁; 刘书语; 余迪; 雷建庚; 李霖

doi:10.7652/xjtuxb202106001

您当前的位置：

首页 >

文章列表页 >

改进深度卷积神经网络及其在变工况滚动轴承故障诊断中的应用

更新时间：2025-07-09

- 改进深度卷积神经网络及其在变工况滚动轴承故障诊断中的应用
- Improved Deep Convolutional Neural Network with Applications to Bearing Fault Diagnosis Under Variable Conditions
- 西安交通大学学报 2021年55卷第6期页码：1-8
- 作者机构：
  
  西安交通大学机械制造系统工程国家重点实验室,西安,710049
- 作者简介：
- 基金信息：
  
  : 2020-12-02。作者简介: 张西宁(1965—),男,教授,博士生导师。基金项目: 国家自然科学基金资助项目(51275379);国家自然科学基金创新研究群体资助项目(51421004)
- DOI：10.7652/xjtuxb202106001
  中图分类号： TH17
- 网络首发：2021-06-10，
  
  纸质出版：2021
- 稿件说明：
移动端阅览
张西宁, 刘书语, 余迪, 等. 改进深度卷积神经网络及其在变工况滚动轴承故障诊断中的应用[J]. 西安交通大学学报, 2021,55(6):1-8.

Improved Deep Convolutional Neural Network with Applications to Bearing Fault Diagnosis Under Variable Conditions[J]. 2021, 55(6): 1-8.
张西宁, 刘书语, 余迪, 等. 改进深度卷积神经网络及其在变工况滚动轴承故障诊断中的应用[J]. 西安交通大学学报, 2021,55(6):1-8. DOI： 10.7652/xjtuxb202106001.

Improved Deep Convolutional Neural Network with Applications to Bearing Fault Diagnosis Under Variable Conditions[J]. 2021, 55(6): 1-8. DOI： 10.7652/xjtuxb202106001.

摘要

为了解决目前常用的最大池化丢失大量信息和平均池化模糊重要特征的问题

提出了一种小尺度卷积核以跳动的方式进行降采样的方法。该方法用步长为2、激活函数为Rectified Linear Unit(ReLU)的小尺度卷积层代替传统的池化层

既可以使输出图像的尺寸变成输入的一半

实现降采样的功能

又能让小尺度卷积核在训练中自动调整权重挑选有效的特征。与最大池化相比

该方法可有效地提高神经元激活比例并增加神经元活性值的多样性。综合采用提出的池化方法、深度可分离卷积核和全局平均池化层3个策略改进的深度卷积神经网络

在实验室变转速多滚动轴承数据集上进行测试

结果表明

改进后的网络识别正确率达到98.4%

高于作为对比的其他网络

同时还大幅提高了网络的稳定性

减少了40%以上的训练时间。提出的方法可以为科研技术人员在搭建深度卷积神经网络和变工况滚动轴承故障诊断时提供参考。

Abstract

The commonly used max pooling only takes the maximum value of neuron activity in the region to bring about loss of a large amount of information

and the average pooling assigns the same weight to the neuron activity values to weaken important features. A novel strategy for down sampling with small-scale convolution kernels is proposed. This strategy replaces the traditional pooling layers with small-scale convolution layers

where the stride of convolution kernels is 2 and the activation function is chosen as Rectified Linear Unit(ReLU)

thus the size of the output image only gets half of the input to achieve down sampling and the small-scale kernels can automatically adjust the weight to select effective features during training. Compared with max pooling

the ratio of active neuron and the diversity of neuron activity values are more effectively heightened. The improved deep convolution neural network(DCNN)with the proposed strategy

depth wise separable convolution and global average pooling is tested and compared on a variable-speed and multi-rolling bearings dataset. The results show that the improved DCNN recognition accuracy rate reaches 98.4%

higher than the other competing networks. Meanwhile

the stability of network is greatly enhanced and training time is shortened by 40%. The proposed strategy for down sampling with small-scale convolution kernels may provide a reference to construction of DCNN and fault diagnosis of rolling bearings under variable conditions.

关键词

Keywords

references

李国杰, 程学旗. 大数据研究: 未来科技及经济社会发展的重大战略领域: 大数据的研究现状与科学思考 [J]. 中国科学院院刊, 2012, 27(6): 647-657.

LI Guojie, CHENG Xueqi. Research of big data: an important strategic field of future technology and economic and social development: research status and scientific thinking of big data [J]. Bulletin of the Chinese Academy of Sciences, 2012, 27(6): 647-657.

雷亚国, 贾峰, 孔德同, 等. 大数据下机械智能故障诊断的机遇与挑战 [J]. 机械工程学报, 2018, 54(5): 94-104.

LEI Yaguo, JIA Feng, KONG Detong, et al. Opportunities and challenges of machinery intelligent fault diagnosis in big data era [J]. Journal of Mechanical Engineering, 2018, 54(5): 94-104.

ZHANG Qingchen, YANG L T, CHEN Zhikui. Deep computation model for unsupervised feature learning in big data [J]. IEEE Transactions on Services Computing, 2016, 9(1): 161-171.

任浩, 屈剑锋, 柴毅, 等. 深度学习在故障诊断领域中的研究现状与挑战 [J]. 控制与决策, 2017, 32(8): 1345-1358.

REN Hao, QU Jianfeng, CHAI Yi, et al. Deep learning for fault diagnosis: the state of the art and challenge [J]. Control and Decision, 2017, 32(8): 1345-1358.

邱锡鹏. 神经网络与深度学习 [M]. 北京: 机械工业出版社, 2020: 11-12.

贾京龙, 余涛, 吴子杰, 等. 基于卷积神经网络的变压器故障诊断方法 [J]. 电测与仪表, 2017, 54(13): 62-67.

JIA Jinglong, YU Tao, WU Zijie, et al. Fault diagnosis method of transformer based on convolutional neural network [J]. Electrical Measurement Instrumentation, 2017, 54(13): 62-67.

ZHANG Wei, LI Chuanhao, PENG Gaoliang, et al. A deep convolutional neural network with new training methods for bearing fault diagnosis under noisy environment and different working load [J]. Mechanical Systems and Signal Processing, 2018, 100: 439-453.

张西宁, 向宙, 唐春华. 一种深度卷积自编码网络及其在滚动轴承故障诊断中的应用 [J]. 西安交通大学学报, 2018, 52(7): 1-8, 59.

ZHANG Xining, XIANG Zhou, TANG Chunhua. A deep convolutional auto-encoding neural network and its application in bearing fault diagnosis [J]. Journal of Xi'an Jiaotong University, 2018, 52(7): 1-8, 59.

李彦冬, 郝宗波, 雷航. 卷积神经网络研究综述 [J]. 计算机应用, 2016, 36(9): 2508-2515, 2565.

LI Yandong, HAO Zongbo, LEI Hang. Survey of convolutional neural network [J]. Journal of Computer Applications, 2016, 36(9): 2508-2515, 2565.

张西宁, 向宙, 夏心锐, 等. 堆叠自编码网络性能优化及其在滚动轴承故障诊断中的应用 [J]. 西安交通大学学报, 2018, 52(10): 49-56, 87.

ZHANG Xining, XIANG Zhou, XIA Xinrui, et al. Optimization of stacking auto-encoder with applications in bearing fault diagnosis [J]. Journal of Xi'an Jiaotong University, 2018, 52(10): 49-56, 87.

HUBEL D H, WIESEL T N. Receptive fields, binocular interaction and functional architecture in the cat's visual cortex [J]. The Journal of Physiology, 1962, 160: 106-154.

XIANG Zhou, ZHANG Xining, ZHANG Wenwen, et al. Fault diagnosis of rolling bearing under fluctuating speed and variable load based on TCO spectrum and stacking auto-encoder [J]. Measurement, 2019, 138: 162-174.

曾雪琼, 黎杰. 基于卷积神经网络的时频图像识别研究 [J]. 机械与电子, 2016, 34(5): 25-29.

ZENG Xueqiong, LI Jie. Time-frequency image recognition based on convolutional neural network [J]. Machinery Electronics, 2016, 34(5): 25-29.

袁建虎, 韩涛, 唐建, 等. 基于小波时频图和CNN的滚动轴承智能故障诊断方法 [J]. 机械设计与研究, 2017, 33(2): 93-97.

YUAN Jianhu, HAN Tao, TANG Jian, et al. An approach to intelligent fault diagnosis of rolling bearing using wavelet time-frequency representations and CNN [J]. Machine Design Research, 2017, 33(2): 93-97.

HINTON G E, SRIVASTAVA N, KRIZHEVSKY A, et al. Improving neural networks by preventing coadaptation of feature detectors [J]. Computer Science, 2012, 3(4): 212-223.

KINGMA D P, BA J L. Adam: a method for stochastic optimization [EB/OL].[2020-11-07].http:∥arxiv.org/abs/1412.6980.pdf.

浏览量

下载量

CSCD

文章被引用时，请邮件提醒。

提交

工具集

关联资源

采用自适应基因粒子群算法优化隐马尔科夫模型的方法及应用

基于多维缩放和随机森林的轴承故障诊断方法

融合多注意力机制的卷积神经网络轴承故障诊断方法

采用多尺度自适应选择卷积神经网络的轴承故障诊断研究

融合多小波分解的深度卷积神经网络轴承故障诊断方法