基于改进DenseNet模型的滚动轴承变工况故障诊断  

Fault Diagnosis of Rolling Bearing under Variable Working Condition Based on Improved DenseNet Model

作　　者：王焜 刘鑫[1] 杨嘉其 董增寿[1] WANG Kun;LIU Xin;YANG Jia-qi;DONG Zeng-shou(School of Electronics and Information Engineering,Taiyuan University of Science and Technology,Taiyuan 030024,China)

机构地区：[1]太原科技大学电子信息工程学院,太原030024

出　　处：《组合机床与自动化加工技术》2022年第6期78-81,共4页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：山西省回国留学人员科研资助项目(2020-126);山西省回国留学人员科研资助项目(2020-127);山西省重点研发计划(201903D321012)。

摘　　要：针对旋转机械传统故障诊断中浅层特征对振动信号故障信息表征能力不足以及在变工况条件下传统的网络模型诊断能力差的问题,提出了一种结合风格再校准模块(style-based recalibration module,SRM)和密集连接卷积神经网络(densely connected convolutional networks,DenseNet)智能故障诊断神经网络模型。将预处理得到的时频图输入到引入SRM的DenseNet网络模型中,通过对特征图进行重新加权以及特征复用,避免有效信息缺失,降低了无关信息的干扰,增强模型对故障特征的提取能力。分别进行单一工况和变工况实验验证,结果表明所提方法在变工况条件下的故障识别率均优于目前主流的SVM、WDCNN和ECACNN诊断方法。To solve the problems that the shallow features in traditional fault diagnosis of rotating machinery have insufficient ability to represent the fault information of vibration signals and the traditional network model has poor diagnostic ability under off-design conditions,an intelligent fault diagnosis neural network model combining style-based recalibration module(SRM)and dense connected convolutional networks(DenseNet)is proposed.The preprocessed time-frequency map is input into the DenseNet network model introduced into SRM.By re-weighting the feature map and reusing the features,the missing of effective information is avoided,the interference of irrelevant information is reduced,and the model's ability to extract fault features is enhanced.The results of single working condition and variable working condition experiments show that the fault recognition rate of the proposed method is better than the current mainstream SVM,WDCNN and ECACNN diagnostic methods.

关 键 词：旋转机械故障诊断 风格再校准模块 密集连接卷积神经网络 变工况 

分 类 号：TH133.3[机械工程—机械制造及自动化] TG66[金属学及工艺—金属切削加工及机床]