基于深度特征与随机过程的轴承剩余使用寿命预测  

Remaining Useful Life Prediction of Bearings Based on Deep Features and Wiener Process

作　　者：庞世杰 韩晓明 PANG Shijie;HAN Xiaoming(College of Electrical and Power Engineering,Taiyuan University of Technology,Taiyuan 030024,China)

机构地区：[1]太原理工大学电气与动力工程学院,太原030024

出　　处：《轴承》2025年第5期102-110,共9页Bearing

基　　金：山西省回国留学人员科研资助项目(2020-041)。

摘　　要：为提高轴承剩余使用寿命(RUL)的预测精度和对预测结果进行不确定性量化,提出一种深度特征提取与随机退化过程交互联动的轴承RUL预测方法。首先,从振动信号中提取均方根等时域特征,并引入新的自上而下时间序列分割算法,将退化过程划分为多个阶段;其次,采用累积变换增强时域特征的趋势性,并将时域累积特征与振动信号傅里叶变换的频域特征作为深度特征提取网络的输入;然后,筛选在不同个体中有相似趋势的深度特征与表征退化阶段的模式特征进行融合构建健康指标;最后,通过目标函数建立特征提取模块与随机模型的联系,实现数模联动,并在PHM 2012轴承数据集上验证了该方法的优越性。In order to improve the prediction accuracy of Remaining Useful Life(RUL)of bearings and quantify the uncertainty of prediction results,a RUL prediction method for bearings is proposed based on interaction of deep feature extraction and Wiener degradation process.Firstly,the time-domain features such as root mean square are extracted from vibration signals,and a new top-down time series segmentation algorithm is introduced to divide the degradation process into multiple stages.Secondly,the cumulative transformation is used to enhance the tendency of time-domain features,the cumulative time-domain features and frequency-domain features of vibration signals after Fourier transform are used as inputs for deep feature extraction network.Then,the deep features exhibiting similar trends among different individuals are screened and fused with pattern features representing degradation stages to construct health indicators.Finally,the collaboration between feature extraction module and Wiener model is established by objective function to achieve data-model interaction,and the superiority of the method is verified on PHM 2012 bearing dataset.

关 键 词：滚动轴承 剩余寿命预测 不确定性量化 特征提取 健康指标 

分 类 号：TH133.33[机械工程—机械制造及自动化] TP183[自动化与计算机技术—控制理论与控制工程]