电子学报 ›› 2022, Vol. 50 ›› Issue (1): 106-115.DOI: 10.12263/DZXB.20200826

• 学术论文 • 上一篇    下一篇

特征融合与灰色回归的滚动轴承性能退化评估

杨创艳1,2, 马军1,2, 王晓东1,2, 罗亭1,2, 李卓睿1,2   

  1. 1.昆明理工大学信息工程与自动化学院,云南 昆明 650500
    2.昆明理工大学云南省人工智能重点实验室,云南 昆明 650500
  • 收稿日期:2020-08-03 修回日期:2021-02-05 出版日期:2022-01-25 发布日期:2022-01-25
  • 作者简介:杨创艳 女,1994年9月出生于云南省大理市.现为昆明理工大学信息工程与自动化学院硕士研究生.主要研究方向为机械故障诊断及性能退化评估. E-mail:yangchuangyan@stu.kust.edu.cn
    马 军(通信作者) 男,1989年9月出生于云南省镇雄县.2016年毕业于昆明理工大学并获得博士学位,现为昆明理工大学信息工程与自动化学院副教授、硕士研究生导师.主要研究方向为大型机械设备结构健康监测. E-mail:mjun@kust.edu.cn
  • 基金资助:
    国家自然科学基金(51765022);云南省科技计划(2019FD042)

Feature Fusion and Grey Regression for Performance Degradation Assessment of Rolling Bearings

YANG Chuang-yan1,2, MA Jun1,2, WANG Xiao-dong1,2, LUO Ting1,2, LI Zhuo-rui1,2   

  1. 1.Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming,Yunnan 650500,China
    2.Yunnan Key Laboratory of Artificial Intelligence,Kunming University of Science and Technology,Kunming,Yunnan 650500,China
  • Received:2020-08-03 Revised:2021-02-05 Online:2022-01-25 Published:2022-01-25

摘要:

针对传统退化指标无法准确反映滚动轴承全寿命周期内退化状态的问题,提出一种特征融合与灰色回归的滚动轴承性能退化评估方法.该方法提取滚动轴承振动信号的高维退化特征,构建基于单调性、相关性和鲁棒性的综合评价准则,选择有效退化特征并构建敏感指标集;提出核独立成分分析(Kernel Independent Component Analysis,KICA)和马氏距离(Mahalanobis Distance,MD)相结合的方法,计算敏感退化指标KICAMD;融合灰色回归模型和3δ原则,判定敏感退化指标KICAMD是否存在虚假波动并修复,获得轴承健康指标(Health Index,HI);最后,基于HI时间序列的转折突变点,自适应确定初始故障时间和定量评估轴承退化状态.两组滚动轴承全寿命周期振动实验数据及对比分析表明,所提方法构建的性能退化指标能有效表征轴承全生命周期的运行状态.

长摘要
针对传统退化指标无法准确反映滚动轴承全寿命周期内退化状态的问题,提出一种特征融合与灰色回归的滚动轴承性能退化评估方法。首先,提取滚动轴承振动信号的时域统计、能量和熵特征,构造了高维退化特征集D;并建立了含单调性、相关性和鲁棒性的综合评价准则Z,基于综合评价准则Z选择有效退化特征构建敏感指标集F。其次,提出了核独立成分分析(Kernel Independent Component Analysis,KICA)和马氏距离(Mahalanobis Distance,MD)相结合的方法,对敏感指标集F进行特征融合,得到敏感退化指标KICAMD;KICAMD相比PCAMD和K-medoids信噪比分别提高了75.27%和62.83%,相关系数分别增加了57.75%和43.58%。然后,融合灰色回归模型和3δ原则,判定敏感退化指标KICAMD是否存在虚假波动并加以修复,获得轴承退化指标HI;对比线性回归和指数回归方法,灰色回归模型(Grey-regression Method, GM)虚假波动修复方法的RMSE最小,与原始信号相关度最大。最后,基于HI时间序列的转折突变点将退化指标HI划分为呈梯度上升的状态区间,自适应确定初始故障时间(Start Failure Time, SFT)和定量评估轴承退化状态。通过NASA和XJTU-SY两组滚动轴承全寿命周期振动数据实验及对比分析表明,所提方法构建的性能退化指标能有效表征轴承全生命周期的运行状态。

关键词: 滚动轴承, 特征融合, 灰色回归模型, 初始故障时间, 退化状态定量评估

Abstract:

Aiming at the problem that the traditional degradation indicator cannot accurately reflect the degradation state of the rolling bearing in the whole life cycle, a method of performance degradation evaluation based on feature fusion and grey regression is proposed. The high-dimensional degradation features of vibration signal are extracted, and the comprehensive evaluation criteria based on monotonicity, correlation and robustness is constructed. A method combining kernel independent component analysis(KICA) and mahalanobis distance(MD) is proposed to calculate the sensitive degradation indicator KICAMD. Then, a novel based-the gray regression model and 3δ principle method is introduced to determine in advance whether the sensitivity degradation indicator KICAMD is false fluctuation ahead of time repair, and then the bearing degradation health indicator HI is obtained. Based on the abrupt transition point of HI time series, the start failure time is determined adaptively and the rolling bearing degradation state is quantitatively evaluated. The experiment and comparative analysis of two groups of rolling bearing life cycle vibration show that the constructed performance degradation index can effectively characterize the running state of the rolling bearing.

Extended Abstract
Aiming at the problem that the traditional degradation indicator cannot accurately reflect the degradation state of the rolling bearing in the whole life cycle, a method of performance degradation evaluation based on feature fusion and grey regression is proposed. Firstly, the time-domain statistics feature, energy feature and entropy feature of rolling bearing vibration signals are extracted, and the high-dimensional degradation feature set D is constructed. And a comprehensive evaluation criterion Z including monotonicity, correlation and robustness is established. The sensitive indicator set F is constructed by selecting effective degradation features based on criterion Z. Secondly, a method combining Kernel Independent Component Analysis (KICA) and Mahalanobis Distance (MD) is proposed to perform feature fusion on the sensitive indicator set F, and the sensitive degradation indicator KICAMD is obtained. Compared with PCAMD and K-medoids, the SNR values of KICAMD increased by 75.27% and 62.83%, and the correlation coefficients increased by 57.75% and 43.58%, respectively. Compared with linear regression and exponential regression methods, the RMSE value of Grey-regression Method(GM)is the smallest, and the greatest correlation compared to the original signal. Thirdly, combined with the GM and 3δ principle, the sensitivity degradation indicator KICAMD is determined to determine whether there is false fluctuation and repair, and the bearing degradation indicator HI is obtained. Finally, the degradation indicator HI is divided into the state interval with gradient rising based on the abrupt transition point, the Start Failure Time (SFT) is determined adaptively and the rolling bearing degradation state is quantitatively evaluated. The experiment and comparative analysis of two groups of NASA and XJTU-SY rolling bearing life cycle vibration data, show that the running state of life-cycle is effectively represented by the performance degradation indicator constructed by the proposed method.

Key words: rolling bearing, feature fusion, grey regression model, start failure time, quantitative assessment of degenerate states

中图分类号: