基于MTF-CNN的滚动轴承故障诊断方法

Rolling bearing fault diagnosis method based on MTF-CNN

针对传统故障诊断方法在滚动轴承实际工况复杂多变、数据集较小时对轴承故障诊断识别准确率较低的问题,提出了MTF-CNN滚动轴承故障诊断模型。首先采用马尔科夫转移场(MTF)编码方式将原始一维振动信号转化为具有时间相关性的二维特征图像,然后将特征图作为卷积神经网络(CNN)的输入进行自动特征提取和故障诊断,最后实现对不同故障类型的分类。为了验证所提方法的有效性和优越性,选用凯斯西储大学滚动轴承数据进行试验验证,并在负载改变时和不同数据集规模下对所提出方法的泛化性能进行测试,同时与传统智能算法进行对比分析。结果表明,相较于其他常用的故障诊断方法,所提出模型在数据集较小、负载改变的环境下对滚动轴承故障诊断具有更好的泛化性能和识别效果。

Aiming at the problem of traditional fault diagnosis method having lower accuracy in bearing fault diagnosis during actual working conditions of rolling bearing being complex and changeable and measurement data set being smaller,a rolling bearing fault diagnosis model was proposed based on MTF-CNN.Firstly,the original 1 D vibration signal was transformed into a 2 D feature image with time correlation using Markov transfer field(MTF)coding,and then the feature image was taken as the input of convolutional neural network(CNN)to do automatic feature extraction and fault diagnosis,and finally realize classification of different fault types.In order to verify the effectiveness and superiority of the proposed method,the rolling bearing data of Case Western Reserve University were selected to do experimental verification,and the generalization performance of the proposed method was tested under changing load and different data set sizes.Meanwhile,the proposed method and traditional intelligent algorithms were analyzed contrastively.The results showed that compared with other common fault diagnosis methods,the proposed model has better generalization performance and recognition effect for rolling bearing fault diagnosis in environment of smaller data set and changing load.