基于改进多尺度卷积网络的轴承故障诊断研究  

Research on Bearing Fault Diagnosis Based on Improved Multi-Scale Convolutional Networks

作　　者：贡莹莹 朱晓娟[1] GONG Yingying;ZHU Xiaojuan(School of Computer Science and Engineering,Anhui University of Science and Technology,Huainan 232001,China)

机构地区：[1]安徽理工大学计算机科学与工程学院,安徽淮南232001

出　　处：《现代信息科技》2025年第7期179-185,共7页Modern Information Technology

基　　金：安徽高校省级自然科学研究重点项目(KJ2020A0300)。

摘　　要：文章针对卷积神经网络在复杂环境中存在的问题,如易受干扰、固定感受野难以提取丰富故障特征以及诊断精度下降等,提出了一种改进的多尺度卷积网络轴承故障诊断方法。首先,对原始振动信号进行预处理;其次,利用不同感受野的卷积核提取多尺度特征,有效捕捉多样化的故障信息;接着,引入自注意力机制,使模型能够动态计算并调整特征图内各位置的权重,自适应地增强关键故障特征;最后,利用全连接层对提取的特征进行分类,实现精准诊断。实验结果表明,该方法在公开数据集上的诊断准确度达到约98%,并且在不同信噪比条件下展示出良好的抗噪性和泛化能力。In this paper,Improved Multi-Scale Convolutional Networks bearing fault diagnosis method is proposed to solve the problems of Convolutional Neural Network in complex environments,such as easy to be disturbed,difficult to extract rich fault features from fixed receptive field and low diagnosis accuracy.Firstly,the original vibration signal is preprocessed.Secondly,the convolution kernels of different receptive fields are used to extract multi-scale features to effectively capture diversified fault information.Thirdly,the Self-Attention Mechanism is introduced to enable the model to dynamically calculate and adjust the weight of each position in the feature map,and adaptively enhance the key fault features.Finally,the fully connected layer is used to classify the extracted features to achieve accurate diagnosis.The experimental results show that the diagnosis accuracy of the method on the public dataset reaches about 98%,and it shows good anti-noise and generalization ability under different signal-to-noise ratio conditions.

关 键 词：多尺度卷积网络 特征提取 自注意力机制 轴承故障诊断 

分 类 号：TH133.3[机械工程—机械制造及自动化]