基于MDAM-GhostCNN的滚动轴承故障诊断方法  

Fault diagnosis method of rolling bearing based on MDAM-GhostCNN

作　　者：郭俊锋[1] 谭宝宏 王智明[1] GUO Junfeng;TAN Baohong;WANG Zhiming(School of Mechanical and Electronical Engineering,Lanzhou University of Technology,Lanzhou 730050,China)

机构地区：[1]兰州理工大学机电工程学院,兰州730050

出　　处：《北京航空航天大学学报》2025年第4期1172-1184,共13页Journal of Beijing University of Aeronautics and Astronautics

基　　金：国家自然科学基金(51465034)。

摘　　要：针对传统故障诊断方法特征提取不充分、计算复杂及在变工况下识别准确率低的问题,提出一种基于混合域注意力机制(MDAM)-GhostCNN的滚动轴承故障诊断方法。采用马尔可夫转移场(MTF)将轴承振动信号转化为具有时间相关性的二维特征图;利用Ghost卷积计算精简的优点,构造出GhostCNN;设计一种MDAM,使网络从通道和空间2个维度充分捕获特征信息,实现特征通道间相互依赖的同时让网络有效关注特征空间信息。由此,构建出MDAM-GhostCNN模型。将MTF二维特征图输入到MDAM-GhostCNN模型中进行训练并输出诊断结果。采用凯斯西储大学和江南大学(JNU)轴承数据集进行实验验证,并对其数据集进行加噪处理。结果表明:在变工况下,所建模型有着更高的识别准确率、抗噪性能和泛化性能。A rolling bearing fault diagnostic approach based on mixture domain attention mechanism(MDAM)-GhostCNN is developed to address the issues of inadequate feature extraction,complicated computation,and low recognition accuracy under varied working conditions in conventional fault detection methods.First of all,the Markov transfer field(MTF)is used to transform the bearing vibration signal into a two-dimensional feature graph with time correlation.Secondly,taking advantage of the simplification of Ghost convolution calculation,GhostCNN is constructed.Then,a MDAM is designed,which makes the network fully capture the feature information from the two dimensions of channel and space,and makes the network pay attention to the feature space information effectively while realizing the interdependence between the feature channels,and construct the MDAM-GhostCNN model.Finally,the MTF two-dimensional feature map is input into the MDAM-GhostCNN model for training and output diagnosis results.Experimental verification and noise processing were performed on the bearing data sets from Jiang Nan University(JNU)and Case Western Reserve University.The results show that under variable working conditions,the constructed model has higher recognition accuracy,noise immunity and generalization performance.

关 键 词：滚动轴承 故障诊断 马尔可夫转移场 Ghost卷积 注意力机制 

分 类 号：TH133.33[机械工程—机械制造及自动化]