基于SC-ResNeSt及频域格拉姆角场的滚动轴承故障诊断方法  

Fault diagnosis method of rolling bearings based on SC-ResNeSt and Gram angle field in frequency domain

作　　者：王海涛[1] 郭一帆 史丽晨[1] WANG Haitao;GUO Yifan;SHI Lichen(School of Mechanical and Electrical Engineering,Xi’an University of Architecture and Technology,Xi’an 710055,China)

机构地区：[1]西安建筑科技大学机电工程学院,陕西西安710055

出　　处：《计算机集成制造系统》2025年第4期1272-1286,共15页Computer Integrated Manufacturing Systems

摘　　要：在实际的工程环境中,滚动轴承通常在强噪声、变负载的条件下工作,针对传统深度学习模型在滚动轴承的故障诊断领域中面临着抗噪性、模型泛化性、鲁棒性差等问题,提出一种基于深度残差分散自校准卷积网络(SC-ResNeSt)和频域格拉姆角场(GAF)的滚动轴承故障诊断新方法。首先,利用GAF编码将振动信号转换为二维图像,并进行二维离散傅里叶变换(2D-DFT)将该图像从空间域转换到频域;其次,因为传统的卷积层缺少动态变化的感受野来提取更具代表性的特征,所以在分散注意力网络(ResNeSt)的基础上引入了自校准卷积模块(SC),提出了一种新的网络模型,即SC-ResNeSt;最后,以频域空间中的GAF作为SC-ResNeSt的输入,故障特征提取完成后,由Softmax分类器完成对故障特征的分类。为验证模型性能,采用美国凯斯西储大学(CWRU)轴承数据集和德国帕博德恩大学(PU)轴承数据集进行测试,实验结果表明所提方法在两种数据集中都取得了较高的故障识别准确率,从而证明了其良好的抗噪性、泛化性以及实用性。In the actual engineering environment,rolling bearings usually work under the condition of strong noise and variable load.Aiming at the problems of traditional deep learning model in rolling bearing fault diagnosis,such as noise resistance,model generalization and poor robustness,a new method of rolling bearing fault diagnosis based on Self-calibration Convolution network with Split-Attention Residual Networks(SC-ResNeSt)and Gram Angle Field(GAF)was presented in frequency domain.The vibration signal was converted into two-dimensional image by GAF coding,and the image was converted from spatial domain to frequency domain by 2D Discrete Fourier Transform(2D-DFT).Owing to lack dynamic receptive field to extract more representative features in the traditional convolution layer,the Self-calibrated Convolutions(SC)modules were introduced on the basis of Split-Attention Residua Networks(ResNeSt),and a new network model named SC-ResNeSt was proposed.By using GAF in frequency domain as the input of SC-ResNeSt,Softmax classifier was used to classify fault features after fault feature extraction.To verify the performance of the model,the bearing data set of Case Western Reserve University(CWRU)and Paerborn University(PU)in Germany were used to test.The experimental results showed that the proposed method on the two data sets had high fault identification accuracy,which proved good noise resistance,generalization and practicability of the method.

关 键 词：自校准卷积 分散注意力机制 格拉姆角场 故障诊断 

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]