Optimization of Stacking Auto-Encoder with Applications in Bearing Fault Diagnosis

doi:10.7652/xjtuxb201810007

您当前的位置：

首页 >

文章列表页 >

Optimization of Stacking Auto-Encoder with Applications in Bearing Fault Diagnosis

更新时间：2025-07-09

- Optimization of Stacking Auto-Encoder with Applications in Bearing Fault Diagnosis
- Vol. 52, Issue 10, Pages: 49-56+87(2018)
- 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：
- DOI：10.7652/xjtuxb201810007
  CLC： TH17
- Online First：10 October 2018，
  
  Published：2018
- 稿件说明：
移动端阅览
Optimization of Stacking Auto-Encoder with Applications in Bearing Fault Diagnosis[J]. 2018, 52(10): 49-56+87.
DOI：

Optimization of Stacking Auto-Encoder with Applications in Bearing Fault Diagnosis[J]. 2018, 52(10): 49-56+87. DOI： 10.7652/xjtuxb201810007.

摘要

为了解决堆叠自编码网络在参数较多时的梯度弥散问题

对网络每层的编码值进行了统计分析

发现大部分分布于激活函数的饱和区

这直接导致了神经元权值梯度的消失。为此

引入了一种标准化策略

将神经元按照样本进行归一化

然后引入两个待学习参数进行缩放和平移

最后通过激活函数输出到下一级神经元。运用带标准化的堆叠自编码网络进行滚动轴承故障诊断

将振动信号的频谱输入到网络中。与普通堆叠自编码网络相比

该标准化策略可有效地使网络编码值均匀分布

如将第一层编码值的熵从0.88 bit提高到了16.29 bit。带标准化的堆叠自编码网络可有效提高网络的抗噪能力和训练速度:在凯斯西储大学滚动轴承数据集上

当人为添加噪声信号的信噪比为0 dB时

识别正确率从16.18%提高到了100%; 在实验室实测数据集上

不仅训练时间下降了37.22%

而且识别正确率从97.93%提高到了99.95%。对网络的编码值进行分析以及引入的标准化策略

可为科研技术人员构建堆叠自编码网络时提供参考

也为滚动轴承故障诊断提供了一种策略。

Abstract

To solve the problem of gradient dispersion in stacking auto-encoder(SAE)with large number of parameters

we analyze the distribution of encoding values in each hidden-layer of network. It is found that most of them are distributed in the saturation area of the activation function

which directly leads to a weight gradient loss

thus a normalizing strategy is introduced. The node is normalized according to the sample

then two parameters are introduced to scale and move the encoding values. Then the modified values are passed to the activation function to next layer. The fault diagnosis of rolling bearing is carried out by the normalized SAE

and the spectrum of vibration signal is input into the network. Compared with ordinary SAE

the encoding values of normalized SAE are more well-distributed. For example

the entropy of encoding values in the first level is increased from 0.88 bit to 16.29 bit. The normalized SAE has higher anti-noise ability and faster training rate. When the signal to noise ratio(SNR)is 0 dB

the recognition accuracy is increased from 16.18% to 100% on the rolling bearing data sets of Case Western Reserve University. On laboratory data sets

the training time is decreased by 37.22%

and the recognition accuracy is increased from 97.93% to 99.95%. The introduced normalizing strategy provides a reference for subsequent research on the construction of SAE

and also provides a strategy for fault diagnosis of rolling bearings.

关键词

Keywords

references

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]. Chinese Journal of Mechanical Engineering, 2018, 54(5): 94-104.

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(8): 1345-1358.

雷亚国, 贾峰, 周昕, 等. 基于深度学习理论的机械装备大数据健康监测方法 [J]. 机械工程学报, 2015, 51(21): 49-56.

LEI Yaguo, JIA Feng, ZHOU Xin, et al. A deep learning-based method for machinery health monitoring with big data [J]. Chinese Journal of Mechanical Engineering, 2015, 51(21): 49-56.

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

RUMELHART D E, HINTON G E, WILLIAMS R J. Learning representations by back-propagating errors [J]. Nature, 1988, 323(6088): 399-421.

HINTON G E, SALAKHUTDINOV R R. Reducing the dimensionality of data with neural networks [J]. Science, 2006, 313(5786): 504-507.

SCHÖLKOPF B, PLATT J, HOFMANN T. Greedy layer-wise training of deep networks [C]∥International Conference on Neural Information Processing Systems. Cambridge, MA, USA: MIT Press, 2006: 153-160.

VINCENT P, LAROCHELLE H, BENGIO Y, et al. Extracting and composing robust features with denoising autoencoders [Z]. New York, USA: ACM, 2008: 1096-1103.

AMARAL T, KANDASWAMY C, SANTOS J M. Using different cost functions to train stacked auto-encoders [C]∥Mexican International Conference on Artificial Intelligence. Piscataway, NJ, USA: IEEE, 2014: 114-120.

JIA Feng, LEI Yaguo, LIN Jing, et al. Deep neural networks: a promising tool for fault characteristic mining and intelligent diagnosis of rotating machinery with massive data [J]. Mechanical Systems and Signal Processing, 2016, 72/73(1): 303-315.

SHAO H, JIANG H, ZHAO H, et al. A novel deep autoencoder feature learning method for rotating machinery fault diagnosis [J]. Mechanical Systems and Signal Processing, 2017, 95(3): 187-204.

崔江, 唐军祥, 龚春英, 等. 一种基于改进堆栈自动编码器的航空发电机旋转整流器故障特征提取方法 [J]. 中国电机工程学报, 2017, 37(19): 5696-5706.

CUI Jiang, TANG Junxiang, GONG Chunying, et al. A fault feature extraction method of aerospace generator rotating rectifier based on improved stacked auto-encoder [J]. Proceedings of the CSEE, 2017, 37(19): 5696-5706.

WANG L, ZHAO X, PEI J, et al. Transformer fault diagnosis using continuous sparse autoencoder [J]. SpringerPlus, 2016, 5(1): 448-463.

SHAHERYAR A, YIN X, HAO H, et al. A denoising based autoassociative model for robust sensor monitoring in nuclear power plants [J]. Science and Technology of Nuclear Installations, 2016, 3(1): 1-17.

YANG Zhixin, WANG Xianbo, ZHONG Jianhua. Representational learning for fault diagnosis of wind turbine equipment: a multi-layered extreme learning machines approach [J]. Energies, 2016, 9: 379.

IOFFE S, SZEGEDY C. Batch normalization: accelerating deep network training by reducing internal covariate shift [C]∥32nd International Conference on Machine Learning. Cambridge, MA, USA: IMLS, 2015: 448-456.

HINTON G E, SALAKHUTDINOV R R. Reducing the dimensionality of data with neural networks [J]. Science, 2006, 313(5786): 504-507.

KINGMA D P, BA J L. Adam: a method for stochastic optimization [EB/OL].[2018-04-01]. http: ∥arxiv.org/pdf/1412.6980v8.pdf.

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

Case Western Reserve University Bearing Data Center. Bearing data file [EB/OL].[218-01-11]. http: ∥csegroups.case.edu/bearingdatacenter/home.

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Feature Extraction for Incipient Fault Diagnosis of Rolling Bearings Based on Nonlinear Manifold Learning

Spectral Kurtosis of Multiwavelet for Fault Diagnosis of Rolling Bearing

Rolling Bearing Fault Diagnosis Approach Based on Case-Based Reasoning

Adaptive Morphological Filtering Algorithm with Applications in Bearing Fault Diagnosis

Fault Diagnosis Method for Rolling Bearing under Variable Working Conditions Using Improved Residual Neural Network

Related Author

栗茂林 1

王孙安 1

梁霖 1

王晓冬

何正嘉

訾艳阳

付新哲

Related Institution

School of Mechanical Engineering, Xi'an Jiaotong University

College of Electrical Engineering and Information Engineering, Lanzhou University of Technology

Key Laboratory of Gansu Advanced Control for Industrial Processes

National Experimental Teaching Center of Electrical and Control Engineering, Lanzhou University of Technology

Key Laboratory of Gansu Advanced Control for Industrial Processes, Lanzhou University of Technology

⁰