基于多尺度特征交叉融合注意力的滚动轴承故障诊断方法

Fault Diagnosis Method for Rolling Bearings Based on Multi-Scale Feature Cross-Fusion Attention

针对现有滚动轴承故障诊断方法信息利用不充分导致诊断精度不高的问题,提出一种基于多尺度特征交叉融合注意力的滚动轴承故障诊断方法。首先,利用卷积神经网络Vgg16和Swin Transformer网络并行提取原始轴承振动信号的全局和局部特征;然后,在全局和局部特征图上进行自注意力特征捕获,并将其送入交叉注意力特征提取模块,实现特征交互,构造多尺度交叉注意力特征图;最后,利用全连接层和Softmax函数实现轴承故障诊断。所提方法在凯斯西储大学轴承数据集、某工业传输技术中心自采集数据集和德国Paderborn数据集上分别实现了97.74%、99.01%和98.31%的诊断精准率,优于其他主流对比方法,验证了所提模型在滚动轴承故障诊断中的可靠性。

A fault diagnosis method for rolling bearings based on multi-scale feature cross-fusion attention is proposed to address the problem of low diagnostic accuracy caused by insufficient information utilization in existing methods.Firstly,the global and local features of original bearing vibration signals are extracted parallelly by using convolutional neural network Vgg16 and Swin Transformer network.Then,the capture of self attention features is carried out on global and local feature maps,and the self attention features are fed into cross attention feature extraction module to realize the feature interaction and construct the multi-scale cross attention feature maps.Finally,the fully connected layers and Softmax function are used to realize the fault diagnosis for the bearings.The proposed method achieves diagnostic accuracies of 97.74%,99.01%,and 98.31%on bearing data set from Case Western Reserve University,self collected data set from an industrial transmission technology center and Paderborn data set in Germany,respectively,outperforming other mainstream comparative methods and verifying the reliability of proposed model in fault diagnosis for the bearings.