深度对抗域自适应模型提升轴承故障诊断准确性

126 浏览量更新于2024-08-27 1 收藏 2.85MB PDF 举报

本文主要探讨了"轴承故障诊断的深度对抗域自适应模型"这一主题，针对工业界普遍面临的挑战——在苛刻工作条件下确保滚动轴承故障诊断的准确性，提出了一种创新的方法。作者Zhao-Hua Liu（IEEE会员）与Bi-Liang Lu、Hua-Liang Wei、Lei Chen、Xiao-Hua Li以及Matthias Rätsch合作，深入研究了深度学习技术在解决实际应用中源领域和目标领域分布不一致问题上的潜力。该研究的核心是深度堆叠自编码器（Deep Stack Autoencoder，DSAE）与代表特征学习的结合。通过DSAE，作者构建了一个无监督学习方法，旨在进行维度降维，从而提取出轴承故障数据中的关键特征。在这个过程中，DSAE通过逐层抽象和解码，能够捕获数据的内在结构，减少冗余信息，提高特征的可解释性和区分度。然而，单纯依靠无监督学习可能不足以应对不同来源或环境下的数据分布差异，即域适应问题。为此，文章引入了深度对抗域适应网络（Deep Adversarial Domain Adaptation，DADA）。DADA利用对抗性训练策略，通过生成器和判别器之间的博弈，使得模型能够在源域和目标域之间找到一个共享的、泛化的特征表示，从而改善在新环境下（如不同的工作条件或设备型号）的故障诊断性能。在具体实施上，DADA模型会将源域数据和少量目标域数据输入到网络中，生成器负责将源域数据转换为目标域风格，而判别器则尝试区分这些变换后的数据与真正的目标域数据。通过反复迭代，模型逐渐学习到一个既能保留源域信息又能适应目标域分布的共享表示。这种方法有助于降低因数据分布变化导致的诊断精度下降，从而提高整体故障诊断系统的鲁棒性和可靠性。这篇文章的重要贡献在于提出了一种深度对抗域自适应模型，它将深度学习和领域适应技术巧妙地融合在一起，为在复杂工业环境中实现高精度的滚动轴承故障诊断提供了一种新的解决方案。未来，随着该模型在实际工业应用中的测试和优化，有望为提高旋转机械的可靠性和安全性带来实质性的进步。

This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination.

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICS: SYSTEMS 1

Deep Adversarial Domain Adaptation Model

for Bearing Fault Diagnosis

Zhao-Hua Liu , Member, IEEE, Bi-Liang Lu, Hua-Liang Wei , Lei Chen, Xiao-Hua Li, and Matthias Rätsch

Abstract—Fault diagnosis of rolling bearings is an essential

process for improving the reliability and safety of the rotating

machinery. It is always a major challenge to ensure fault diag-

nosis accuracy in particular under severe working conditions. In

this article, a deep adversarial domain adaptation (DADA) model

is proposed for rolling bearing fault diagnosis. This model con-

structs an adversarial adaptation network to solve the commonly

encountered problem in numerous real applications: the source

domain and the target domain are inconsistent in their distribu-

tion. First, a deep stack autoencoder (DSAE) is combined with

representative feature learning for dimensionality reduction, and

such a combination provides an unsupervised learning method to

effectively acquire fault features. Meanwhile, domain adaptation

and recognition classiﬁcation are implemented using a Softmax

classiﬁer to augment classiﬁcation accuracy. Second, the effects

of the number of hidden layers in the stack autoencoder network,

the number of neurons in each hidden layer, and the hyperpa-

rameters of the proposed fault diagnosis algorithm are analyzed.

Third, comprehensive analysis is performed on real data to vali-

date the performance of the proposed method; the experimental

results demonstrate that the new method outperforms the exist-

ing machine learning and deep learning methods, in terms of

classiﬁcation accuracy and generalization ability.

Index Terms—Adversarial network, bearing, deep learning,

deep neural networks, domain adaptation (DA), fault diagnosis,

feature extraction, machine learning, stack autoencoder (SAE),

unsupervised learning.

I. INTRODUCTION

OLLING bearings are widely used in industrial system,

such as wind turbine, aeroengines, and high-speed

railways, and it usually plays a pivotal role in their

functioning [1]–[4]. However, these devices often work with

Manuscript received May 15, 2019; accepted July 18, 2019. This work was

supported in part by the National Natural Science Foundation of China under

Grant 61503134 and Grant 61573299, in part by the Hunan Provincial Hu-

Xiang Young Talents Project of China under Grant 2018RS3095, and in part

by the Hunan Provincial Natural Science Foundation of China under Grant

2018JJ2134. This article was recommended by Associate Editor G. Nicosia.

(Corresponding author: Zhao-Hua Liu.)

Z.-H. Liu, B.-L. Lu, L. Chen, and X.-H. Li are with the School of

Information and Electrical Engineering, Hunan University of Science and

Technology, Xiangtan 411201, China (e-mail: zhaohualiu2009@hotmail.com;

1197393632@qq.com; chenlei@hnust.edu.cn; lixiaohua_0227@163.com).

H.-L. Wei is with the Department of Automatic Control and Systems

Engineering, University of Shefﬁeld, Shefﬁeld S1 3JD, U.K. (e-mail:

w.hualiang@shefﬁeld.ac.uk).

M. Rätsch is with the Image Understanding and Interactive Robotics

Group, Reutlingen University, 72762 Reutlingen, Germany (e-mail:

matthias.raetsch@reutlingen-university.de).

Color versions of one or more of the ﬁgures in this article are available

online at http://ieeexplore.ieee.org.

Digital Object Identiﬁer 10.1109/TSMC.2019.2932000

heavy loads or under some severe environments (e.g., high

speed, high humidity, high temperatures, variable speed, etc.),

which makes rolling bearings prone to fault attacks. The

high failure rate of rolling bearings also increases the oper-

ation and maintenance costs. Moreover, the cases in which

the potential faults of rolling bearings are not detected, there

would be a high risk exists of the breakdown of the entire

equipment [5]–[10]. Therefore, it is always desirable and

necessary to diagnose potential rolling bearing faults in time.

In the Internet of Things (IoT) and Industry 4.0 era,

large amounts of real-time data have been collected from

the device-monitoring systems. The data, together with mod-

ern data mining techniques, makes it possible to effectively

mine features and diagnose faults using artiﬁcial intelli-

gence methods, such as support vector machine (SVM) [11],

artiﬁcial neural network (ANN) [12], [13], stack autoen-

coder (SAE) network [14], and deep belief network (DBN)

[15], [16]. For example, Jiang et al. [13] proposed an approach

for rolling bearing fault identiﬁcation using multilayer deep

convolutional neural network. Sun et al. [14] designed an

intelligent bearing fault diagnosis method combining com-

pressed data acquisition and deep neural network architecture.

Chen and Li [15] presented a novel method to implement bear-

ing fault diagnosis utilizing the integration method of sparse

autoencoder (AE) and deep belief network. However, although

these intelligent fault diagnosis methods achieve good classiﬁ-

cation performance in experimental testing, they do not exhibit

satisfactory performance when applied in practical applica-

tions, in which the classiﬁcation accuracy is usually much

lower than that for test data. This can be explained from two

aspects as follows. First, these artiﬁcial intelligence methods

require a large amount of labeled data to train the model.

However, in many real applications, it is very expensive or

difﬁcult, even not possible; to collect labeled training data that

has the same distribution as the test set. In conclusion, it is

difﬁcult to collect sufﬁcient labeled data and then train a reli-

able diagnosis model in engineering scenarios. Second, it is

assumed that the training data set and the test set of the

model are generated under the same working conditions in

the experimental testing. In other words, it is assumed that

all data obey the same distribution and possess the same fea-

ture space. In reality, however, during the operation of the

rotating machinery differ, the mechanical working conditions

vary, the signal acquisition methods are different, and the

mechanical workloads are varying. As a consequence, these

intelligent diagnosis methods have poor generalization abil-

ity in real application and, therefore, bring poor diagnostic

2168-2216

 2019 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.

See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

下载后可阅读完整内容，剩余9页未读，立即下载

weixin_38591223

粉丝: 7
资源: 911

深度对抗域自适应模型提升轴承故障诊断准确性

deep-transfer-learning:深度域自适应算法的实现集合

通过深度域自适应对图像和形状进行联合分析

基于域自适应迁移学习的轴承故障诊断

应用于轴承故障诊断的LMS自适应滤波器及其FPGA实现.pdf

ZSL-ADA:随附论文“具有对抗域自适应的零镜头学习生成框架”的代码

混合系统故障诊断的模态转换自适应阈值研究 (2014年)

航空发动机故障诊断：一种自适应模型方法

部分对抗域自适应：解决标签空间不完全共享的问题

掌握CyCADA：ICML 2018提出的周期一致对抗域自适应技术

混合系统故障诊断：模态转换自适应阈值方法

最新资源