铁道科学与工程学报2025，Vol.22Issue(9)：4260-4270,11.DOI:10.19713/j.cnki.43-1423/u.T20241859
基于GRCMFDE与CNN的轮对轴承故障诊断方法研究
Fault diagnosis of wheelset bearings based on GRCMFDE and CNN
摘要
Abstract
To solve the difficulty in extracting fault signal features under actual working conditions of wheelset bearings,a new entropy-based nonlinear dynamic feature extraction algorithm—generalized refined composite multiscale fluctuation dispersion entropy(GRCMFDE)was proposed.First,a single coarse-grained time series was generated into multiple sequences through composite coarse-graining to make up for the defect of single coarse-graining of multiscale fluctuation dispersion entropy,so as to mine more characteristic information.Then,the traditional composite process was optimized through fine composite to prevent the undefined entropy from appearing due to the shortening of the length of the composite coarse-grained sequence,which could make it impossible to calculate the mean entropy at this scale.Finally,through generalized coarse-graining processing,the mean calculation was changed to the calculation of the variance,so as to avoid the dynamic mutation phenomenon of the original fault signal that was neutralized by the mean calculation of the multiscale fluctuation dispersion entropy.Extracting fault feature information of NJ2323WB wheelset bearings using GRCMFDE,and then inputting the fault features into convolutional neural network(CNN)for pattern recognition.Collecting wheelset bearing signals of different types of actual damage on DF4 locomotives on the JL-501 locomotive bearing detection platform and extracting features.The results show that the feature extraction performance of GRCMFDE is better than that of RCMFDE and MFDE,and the fault recognition performance of CNN is better than that of support vector machine.The fault identification rate of the GRCMFDE-CNN diagnostic model is 100%.This method can quickly and accurately identify different faults that occur in bearings during actual working conditions and has certain application value.关键词
广义精细复合多尺度波动离散熵/特征提取/卷积神经网络/模式识别/轮对轴承/故障诊断Key words
GRCMFDE/feature extraction/CNN/pattern recognition/wheelset bearing/fault diagnosis分类
机械制造引用本文复制引用
彭刘禹,胡俊锋,张龙..基于GRCMFDE与CNN的轮对轴承故障诊断方法研究[J].铁道科学与工程学报,2025,22(9):4260-4270,11.基金项目
国铁集团科技研发重点课题(N2023J042) （N2023J042）
江西省自然科学基金重点项目(20224ACB204017) （20224ACB204017）
江西省教育厅项目(GJJ2405311) （GJJ2405311）
