• CN:11-2187/TH
  • ISSN:0577-6686

机械工程学报 ›› 2019, Vol. 55 ›› Issue (7): 65-72.doi: 10.3901/JME.2019.07.065

• 基于深度学习的机械装备故障预测与健康管理 • 上一篇    下一篇

利用FCKT以及深度自编码神经网络的滚动轴承故障智能诊断

杨蕊, 李宏坤, 王朝阁, 郝佰田   

  1. 大连理工大学机械工程学院 大连 116024
  • 收稿日期:2018-05-28 修回日期:2018-11-08 出版日期:2019-04-05 发布日期:2019-04-05
  • 通讯作者: 李宏坤(通信作者),男,1974年出生,教授,博士研究生导师。主要研究方向为机械设备动态分析与故障诊断,信号处理。E-mail:lihk@dlut.edu.cn
  • 作者简介:杨蕊,女,1990年出生,博士研究生。主要研究方向为机械设备微弱故障特征提取、智能故障诊断。E-mail:yr90223@mail.dlut.edu.cn
  • 基金资助:
    国家自然科学基金资助项目(51175057)。

Intelligent Fault Detection for Rolling Element Bearing Based on FCKT and Deep Auto-coding Neural Network

YANG Rui, LI Hongkun, WANG Chaoge, HAO Baitian   

  1. School of Mechanical Engineering, Dalian University of Technology, Dalian 116024
  • Received:2018-05-28 Revised:2018-11-08 Online:2019-04-05 Published:2019-04-05

摘要: 实时、快速、批量地对振动信号进行处理成为故障诊断领域的未来发展趋势,但是可能会带来数据维数灾难问题。针对在样本较大情况下深度学习运行时间较长,以及层与层之间节点数的减少使得故障识别准确率降低问题,提出首先计算原始时域信号的频谱在不同偏移点数下的相关峭度值(FCKT)作为新的样本数据,并结合深度自编码神经网络实现轴承的智能故障分类。新样本相对于原始样本,实现了数据的维数约减,缩短了样本集的分析时间。同时,在保持各样本数据原有信息的基础上,使得样本之间差异性更突出。另外,该方法在避免深度学习算法层与层之间的权值根据经验设定的同时,解决了通过逐层减少隐含层节点数来提高计算效率时带来的分类识别准确率降低的问题。最后,通过试验数据对比分析验证了算法的有效性。

关键词: 不同偏移点数相关峭度值(FCKT), 深度自编码器, 智能故障分类

Abstract: Real-time, fast, and batch processing of vibration signals have become a future development trend in the field of fault diagnosis. However, it may lead to data dimensional disasters. In view of the fact that the long running time and the low fault identification accuracy of deep learning algorithm under the condition of large sample. The frequency spectrum correlation Kurtosis of original time domain signal is calculated under different iteration periods (FCKT) as new samples data and use the deep auto-coding neural network is used to realize the intelligent fault classification of planetary gearbox. Compared with the original samples, the new samples reduce the data dimension and shorten the analysis time. At the same time, the differences between the samples are more prominent based on the original information of each sample data. In addition, the method solves the problems that the weight parameters between layers of deep learning algorithm are set according to experience and the classification accuracy is reduced by reducing the number of hidden nodes layer by layer to improve the calculation efficiency. Finally, the validity of the algorithm is verified by comparison of experimental data.

Key words: correlation Kurtosis for different iteration periods (FCKT), deep auto-coding, intelligent fault classification

中图分类号: