深度概率优化的VAE轴承状态评估

摘要
图/表
参考文献(25)
相关文章 (0)

全文: PDF (1855 KB) HTML (1 KB)
输出: BibTeX | EndNote (RIS)

摘要基于振动信号的VAE（variational auto-encoder，VAE）轴承状态评估方法，由于VAE近似后验分布简化高斯假设，其隐变量低维空间表示过于简单，往往无法捕捉到振动信号真实的潜在故障特征，且利用变分证据下界评估运行状态，存在估计不准确以及受样本数目影响较大等问题。研究分布变换优化VAE近似后验分布，利用优化采样算法优化计算VAE边缘概率密度，建立一种基于深度概率优化的VAE轴承状态评估模型。通过标准化流（normalizing flows）实现VAE中的分布优化，构造复杂灵活的近似后验分布，自适应学习健康状态下轴承振动信号频谱概率分布；采用AIS（annealed importance sampling，AIS）算法，通过一系列中间分布，采样完成边缘概率密度的优化计算，建立评价指标。滚动轴承对比实验表明，所提方法对滚动轴承退化过程更为敏感，证明了该方法在轴承状态评估中的有效性。

	服务

	把本文推荐给朋友
	加入我的书架
	加入引用管理器
	E-mail Alert
	RSS
	作者相关文章
	尹爱军
	陈小敏
	谭建
	王昱

关键词 ：深度概率优化, 变分自编码器, 标准化流, 退火重要性采样, 轴承状态评估

Abstract：Variational auto-encoder (VAE) can be utilized to assess bearing operation condition based on vibration monitoring.A limitation of traditional VAE based evaluation method is the simplified Gaussian posterior distribution.The spatial representation of low-dimensional hidden variable is too simple to capture the real potential fault characteristics of vibration signal.Moreover, the evidence lower bound in traditional VAE is subject to inaccurate estimation.In this paper, distribution transformation was utilized to optimize VAE approximate posterior distribution.With the edge probability density calculated by applying optimized sampling algorithm, a bearing condition evaluation model was established based on deep probability optimization.Firstly, normalizing flows (NF) was employed to construct a complex and flexible approximate posterior distribution to realize distribution optimization.Then the Annealed Importance Sampling (AIS) algorithm with a series of intermediate distributions was introduced to complete optimal calculation of the edge probability density and establish the evaluation indicator.Comparative experiments on rolling bearing indicate that the proposed method is more sensitive to the degradation process, which verifies the effectiveness of proposed bearing condition assessment method.

Key words： deep probabilistic optimization variational auto-encoder normalizing flows annealed importance sampling bearing condition assessment

收稿日期: 2020-06-08 出版日期: 2021-10-28

引用本文:

尹爱军，陈小敏，谭建，王昱. 深度概率优化的VAE轴承状态评估[J]. 振动与冲击, 2021, 40(20): 186-192.
YIN Aijun,CHEN Xiaomin,TAN Jian,WANG Yu. Bearing condition assessment of VAE based on deep probability optimization. JOURNAL OF VIBRATION AND SHOCK, 2021, 40(20): 186-192.

链接本文:

http://jvs.sjtu.edu.cn/CN/ 或 http://jvs.sjtu.edu.cn/CN/Y2021/V40/I20/186

［1］GUSTAVO N P L, ALEX M A, PEDRO A C R,et al.Entropy measures for early detection of bearing faults［J］.Physica A: Statistical Mechanics and its Applications,2019,514:458-472.

［2］佘道明,贾民平,张菀.一种新型深度自编码网络的滚动轴承健康评估方法［J］.东南大学学报(自然科学版),2018,48(05):801-806.
SHE Daoming, JIA Minping ,ZHANG Wan.Deep autoencoder network method for health assessment of rolling bearings［J］.Journal of Southeast University (Natural Science Edition),2018,48(5):801-806.

［3］康守强,王玉静,崔历历,等.基于CFOA-MKHSVM的滚动轴承健康状态评估方法［J］.仪器仪表学报,2016,37(9):2029-2035.
KANG Shouqiang,WANG Yujing,CUI Lili, et al.Health state assessment of a rolling bearing based on CFOA-MKHSVM method［J］.Chinese Journal of Scientific Instrument,2016,37(9):2029-2035.

［4］WANG S S,CHEN J,WANG H, et al.Degradation evaluation of slewing bearing using HMM and improved GRU［J］.Measure-ment, 2019,146:385-395.

［5］JIANG W, ZHOU J Z,LIU H, et al.A multi-step progressive fault diagnosis method for rolling element bearing based on energy entropy theory and hybrid ensemble auto-encoder［J］.ISA Transactions, 2019,87: 235-250.

［6］宁永杰. 基于机器学习的滚动轴承状态评估与剩余寿命预测［D］.徐州：中国矿业大学,2019.

［7］刘国增,赵建民,张鑫,等.基于小波包AR能量熵和平滑样条的轴承退化状态评估［J］.轴承,2019(8):58-63.
LIU Guozeng, ZHAO Jianmin, ZHANG Xin,et al.Assessment on degradation state for bearings based on wavelet packet ar energy entropy and smoothing spline［J］.Bearing, 2019(8):58-63.

［8］张小强,朱文辉,康铁宇,等.基于人工免疫算法的离散隐马尔科夫故障诊断模型优化［J］.装备环境工程,2019,16(1):63-67.
ZHANG Xiaoqiang, ZHU Wenhui, KANG Tieyu, et al.Optimization of discrete hidden markov fault diagnosis model based on artificial immune algorithm［J］.Equipment Environmental Engineering,2019,16(1): 63-67.

［9］刘美芳,余建波,尹纪庭.基于贝叶斯推论和自组织映射的轴承性能退化评估方法［J］.计算机集成制造系统,2012,18(10): 2237-2244.
LIU Meifang, YU Jianbo, YIN Jiting.Bearing performance degradation assessment based on bayesian inference and self-organizing map［J］.Computer Integrated Manufacturing Systems, 2012,18(10):2237-2244.

［10］PAVLE B, MATEJ G, DEJAN P, et al.Bearing fault prognostics using Renyi entropy based features and Gaussian process models［J］.Mechanical Systems and Signal Processing, 2015,52:327-337.

［11］DOERSCH C.Tutorial on variational autoencoders［J］.Stat, 2016, 1050: 13-37.

［12］HOU X X, SUN K,SHEN L L,et al.Improving variational autoencoder with deep feature consistent and generative adversarial training［J］.Neurocomputing, 2019,341:83-194.

［13］KING D P, WELLING M.Auto-encoding variational bayes［C］.Internationl Conference on Learning Representations,2014:1-14.

［14］HSU C C,HWANG H T,WU Y C，et al.Voice conversion from non-parallel corpora using variational auto-encoder［C］∥Global Con-ference on Signal and Information Processing.IEEE ,2017：1-6.

［15］马波,赵祎,齐良才.变分自编码器在机械故障预警中的应用［J］.计算机工程与应用,2019,55(12):245-249.
MA Bo, ZHAO Yi,QI Liangcai.Application of variational auto-encoder in mechanical fault early warning［J］.Computer Engineering and Applications,2019,55(12): 245-249.

［16］LAROCHELLE H, MURRAY I.The neural autoregressive distribution estimator［J］.Journal of Machine Learning Research, 2011, 15:29-37.

［17］SALMAN H, YADOLLAHPOUR P, FLETCHER T, et al.Deep diffeomor-phic normalizing flows［J］.ArXiv preprint, 2018:1-13.

［18］NEAL R M.Annealed importance sampling［J］.Statistics and computing, 2001, 11(2): 125-139.

［19］WU Y, BURDA Y, SALAKHUTDINOV R, et al.On the quantitative analysis of decoder-based generative models［J］.ArXiv preprint, 2016:1-17.

［20］REZENDE D J, MOREZENDE D, MOHAMED S.Variational inference with normalizing flows［C］.International conference on machine learning.PMLR, 2015: 1530-1538.

［21］张锐戈. 滚动轴承振动信号非平稳、非高斯分析及故障诊断研究［D］.西安：西安电子科技大学,2014.

［22］夏均忠,郑建波,白云川,吕麒鹏,杨刚刚.基于NAP和RMI的滚动轴承性能退化状态识别与评估［J］.振动与冲击,2019,38(23):33-37.
XIA Junzhong, ZHENG Jianbo, BAI Yunchuan, et al.Perfor-mance degradation status identification and assessment for rolling bearing based on NAP and RMI［J］.Journal of Vibration and Shock,2019,38(23): 33-37.

［23］LEE J, QIU H, YU G, et al.IMS, University of Cincinnati.Bearing data set, NASA ames prognostics data repository ［R/OL］.Moffett Field,CA,USA:NASA Ames Research Center,2007.［2018-12-12］.http://ti.arc.nasa.gov/project/ prognostic-data-repository.

［24］李鹏.基于高斯混合模型的变分自动编码器［D］.哈尔滨：哈尔滨工业大学,2017.

［25］ZHANG B, ZHANG S, LI W.Bearing performance degradation assessment using long short-term memory recurrent network［J］.Computers in Industry, 2019, 106:14-29.