基于CEEMD-PCA-XGBoost的滚动轴承故障诊断方法  

Rolling bearing fault diagnosis method based on CEEMD-PCA-XGBoost

作　　者：马东 何毅斌[1] 李铭 唐权 胡明涛 MA Dong;HE Yi-bin;LI Ming;TANG Quan;HU Ming-tao(College of Mechanical and Electrical Engineering,Wuhan Institute of Technology,Wuhan 430205,China)

机构地区：[1]武汉工程大学机电工程学院,湖北武汉430205

出　　处：《机电工程》2023年第2期186-194,共9页Journal of Mechanical & Electrical Engineering

基　　金：国家自然科学基金资助项目(51775390);化工装备强化与本质安全湖北省重点实验室开放基金资助项目(2018KA01);武汉工程大学研究生教育创新基金资助项目(cx2021052)。

摘　　要：在轴承的故障诊断过程中,往往会存在因数据缺失或分布不均,从而导致其运算速度慢和分类准确率低的问题,为此,提出了一种互补集成经验模态分解结合主成分分析和极限梯度提升(CEEMD-PCA-XGBoost)的故障诊断方法。首先,基于互补集成经验模态分解(CEEMD)对第三方轴承故障数据集进行时域和频域的特征提取,实现了数据初步筛选的目的;然后,采用主成分分析法(PCA),降低了分解后的本征模态函数分量(IMF)的特征值维度;将提取的特征量作为输入量,输入到极限梯度提升(XGBoost)模型中,并采用栅格法优化了模型的参数;最后,通过2种不同轴承数据集对该方法进行了验证,并从分类精度、准确度等角度出发,将该方法所得结果与采用其他算法所得到的结果进行了对比分析。实验结果表明:经美国凯斯西储大学轴承数据集检验,采用优化后的算法模型得到的分类准确率为100%,运算时间为11.264 s;经IEEE PHM 2012数据集验证,采用该算法得到的轴承寿命预测曲线拟合效果优于其他算法。研究结果表明:在运算速度和分类准确率方面,该轴承故障诊断方法具有较好的综合性能。Aiming at the problems of slow operation speed and low classification accuracy due to data loss or uneven distribution in bearing fault diagnosis,a fault diagnosis method combined with complementary ensemble empirical mode decomposition and principal component analysis and extreme gradient boosting(CEEMD-PCA-XGBoost)was proposed.Firstly,based on complementary ensemble empirical mode decomposition(CEEMD),the third-party bearing fault data sets were extracted in the time domain and frequency domain to achieve preliminary data screening.Then,the principal component analysis(PCA)method was used to reduce the eigenvalue dimension for the decomposed intrinsic mode functions(IMF).The extracted eigenvalues were input into the extreme gradient boosting(XGBoost)model as input quantities,and the model parameters were optimized by the raster method.Finally,the method was verified by two different bearing data sets,and the results obtained by this method were compared with those obtained by other algorithms from the perspective of classification precision and accuracy.The experiment results show that the classification precision of the optimized model is 100%,and the calculation time is 11.264 s,verified by the bearing dataset of Case Western Reserve University;verified by the IEEE PHM 2012 dataset,the fitting effect of the life prediction curve is better than other algorithms.The research results show that,the method has good comprehensive performance in terms of operation speed and classification accuracy.

关 键 词：特征提取 互补集成经验模态分解 主成分分析 极限梯度提升 分类准确率 特征值维度 

分 类 号：TH133.33[机械工程—机械制造及自动化]