基于GAF-DarkNet的滚动轴承故障诊断方法  

Fault Diagnosis Method for Rolling Bearings Based on GAF-DarkNet

作　　者：虞浒 缪小冬[1] 顾寅骥 荀志文 隋天举 YU Hu;MIAO Xiaodong;GU Yinji;XUN Zhiwen;SUI Tianju(College of Mechanical and Power Engineering,Nanjing Tech University,Nanjing 211800,China;College of Computer Science and Technology,Nanjing Tech University,Nanjing 211800,China;Key Laboratory of Intelligent Control and Optimization for Industrial Equipment MOE,Dalian University of Technology,Dalian 116024,China)

机构地区：[1]南京工业大学机械与动力工程学院,南京211800 [2]南京工业大学计算机科学与技术学院,南京211800 [3]大连理工大学工业装备智能控制与优化教育部重点实验室,辽宁大连116024

出　　处：《轴承》2024年第2期66-73,81,共9页Bearing

基　　金：国家自然科学基金资助项目(52175465);工业装备智能控制与优化教育部重点实验室基金资助项目(LICO2020TB01)。

摘　　要：针对深度诊断模型较难处理信号紧邻特征点以及变工况导致模型诊断精度和泛化性能不足的问题,提出基于格拉姆角场(GAF)和DarkNet-53图像识别算法的滚动轴承故障诊断方法。通过GAF编码将原始振动信号转换为具有时序相关性的二维特征图像,将特征图像输入DarkNet-53进行特征自提取和故障诊断。基于凯斯西储大学(CWRU)滚动轴承数据集和南京工业大学转盘轴承数据集并通过变载荷工况分析对所提算法进行验证,同时与目前流行的智能诊断方法及二维重构诊断方法进行对比,结果表明变工况下GAF-DarkNet算法具有更好的泛化能力和故障识别效果。A fault diagnosis method is proposed for rolling bearings based on Gramian angular field(GAF)and DarkNet-53 image recognition algorithm to address the problem of difficulty in processing adjacent signal feature points and insufficient diagnostic accuracy and generalization performance caused by variable operating conditions in deep diagnostic models.The original vibration signal is converted into a two-dimensional feature image with temporal correlation through GAF encoding,and the feature image is input into DarkNet-53 for feature self-extraction and fault diagnosis.Based on rolling bearing dataset from Case Western Reserve University(CWRU)and slewing bearing dataset from Nanjing University of Technology,the proposed algorithm is validated through variable load condition analysis.At the same time,compared with current popular intelligent diagnostic methods and two-dimensional reconstruction diagnostic methods,the results show that the GAF-DarkNet algorithm has better generalization ability and fault recognition effect under variable operating conditions.

关 键 词：滚动轴承 故障诊断 深度学习 特征提取 图像识别 

分 类 号：TH133.331[机械工程—机械制造及自动化] TH165.3