卷积深度森林滚动轴承故障诊断方法  

ROLLING BEARING FAULT DIAGNOSIS METHOD BASED ON CONVOLUTIONAL DEEP FOREST

作　　者：顾辉 邵星[1] 王翠香[1] 皋军[1] GU Hui;SHAO Xing;WANG CuiXiang;GAO Jun(School of Information Engineering,Yancheng Institute of Technology,Yancheng 224051,China;School of Mechanical Engineering,Yancheng Institute of Technology,Yancheng 224051,China)

机构地区：[1]盐城工学院信息工程学院,盐城224051 [2]盐城工学院机械工程学院,盐城224051

出　　处：《机械强度》2024年第6期1279-1286,共8页Journal of Mechanical Strength

基　　金：国家自然科学基金项目(61502411,62076215);教育部新一代信息技术创新项目(2020ITA02057);江苏省研究生科研与实践创新计划项目(SJCX22_XZ035,SJCX22_XY061)资助。

摘　　要：针对滚动轴承振动信号非线性、样本量少、传统机器学习诊断算法需要专家经验等问题,提出了一种卷积深度森林(Convolutional Deep Forest,CDF)的故障诊断方法。首先对一维振动信号进行归一化和转图片预处理,接着利用卷积神经网络对图片训练,完成端到端的特征提取,然后使用级联森林对特征进行分析并分类,最后在轴承数据集上验证了CDF的有效性。试验结果表明,CDF针对4种负载下的大小样本数据均能取得较高的准确率,基于二维信号转图片的卷积神经网络和CDF的准确率均高于一维,证明了基于信号转图片数据预处理操作的有效性。Aiming at the vibration signal of rolling bearing with problems of nonlinear,small sample size and traditional machine learning based diagnosis algorithm required expert experience,a convolutional deep forest(CDF)based rolling bearing fault diagnosis algorithm was proposed.Firstly,the one-dimensional vibration signal was preprocessed through normalization and transformation into image.Then the convolution neural network was exploited to train the image to complete the end-to-end feature extraction,and the cascade forest was used to analyze and classify the features.Finally,the effectiveness of CDF was verified on the bearing data set.The experimental results show that CDF can achieve high accuracy for small or big sample data under four loads.In addition,the accuracy of convolution neural network and CDF based on two-dimensional image are higher than one-dimensional,which proves the effectiveness of data preprocessing operation based on signal to image.

关 键 词：故障诊断 滚动轴承 深度森林 卷积神经网络 

分 类 号：TH133.33[机械工程—机械制造及自动化] TP212[自动化与计算机技术—检测技术与自动化装置] TH212.3[自动化与计算机技术—控制科学与工程]