改进残差结构的轻量级故障诊断方法  

Lightweight fault diagnosis method with improved residual structure

作　　者：刘芯志 彭成 满君丰[1,2] 刘翊 LIU Xin-zhi;PENG Cheng;MAN Jun-feng;LIU Yi(College of Computer Science,Hunan University of Technology,Zhuzhou 412007,China;College of Automation,Central South University,Changsha 410083,China;National Advanced Rail Transit Equipment Innovation Center,Ministry of Industry and Information Technology,Zhuzhou 412000,China)

机构地区：[1]湖南工业大学计算机学院,湖南株洲412007 [2]中南大学自动化学院,湖南长沙410083 [3]工业和信息化部国家先进轨道交通装备创新中心,湖南株洲412000

出　　处：《计算机工程与设计》2022年第8期2303-2310,共8页Computer Engineering and Design

基　　金：国家自然科学基金项目(61871432、61702177);湖南省自然科学基金项目(2019JJ60008、2019JJ60054、2020JJ4275、2020JJ6086);湖南省科技人才专项-湖湘青年英才基金项目(2019RS2062);湖南省研究生创新基金项目(CX20201050)。

摘　　要：针对大型机械装备环境噪声复杂,深度学习网络层数过深导致的巨大计算开销以及故障诊断人工特征提取的复杂性,提出改进残差结构的轻量级SCARN模型。SCARN模型使用蓝图可分离卷积代替常规卷积层,减少大量参数,设计轻量级空间通道注意力结构,加强特征表达能力,改进深度残差收缩模块,提高模型复杂噪声场景的鲁棒性。通过增加不同幅值的高斯白噪声模拟轴承信号复杂环境场景。实验结果表明,该模型4种评价指标均优于对比算法,具有良好的抗噪性能。Aiming at the complex environmental noise of large-scale mechanical equipment,the huge computational overhead due to the deep layer of deep learning network and the complexity of fault diagnosis artificial feature extraction,lightweight SCARN model with improved residual structure was proposed.The conventional convolution layer was replaced by blueprint separable convolution to reduce a large number of parameters.A lightweight spatial channel attention structure was designed to enhance the ability of feature expression.The depth residual shrinkage module was improved to improve the robustness of the model in complex noise scenes.The complex environment scene of bearing signal was simulated by adding Gaussian white noise of diffe-rent amplitude.Experimental results show that four evaluation indicators of the model are better than that of the comparison algorithm,and it has good anti-noise performance.

关 键 词：蓝图可分离卷积 空间通道注意力 深度残差收缩模块 轻量级 高斯白噪声 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]