基于SW-DBA-DCNN的滚动轴承故障诊断方法  

Rolling Bearing Fault Diagnosis Method Based on SW-DBA-DCNN

作　　者：张恒[1] 佘博[1] 王俊 王旋 ZHANG Heng;SHE Bo;WANG Jun;WANG Xuan(College of Weaponry Engineering,Naval University of Engineering,Wuhan 430033)

机构地区：[1]海军工程大学兵器工程学院,武汉430033

出　　处：《舰船电子工程》2023年第5期146-152,共7页Ship Electronic Engineering

摘　　要：针对滚动轴承故障诊断过程中出现的数据样本量不平衡问题,论文提出一种基于Sliding Window-Dynamic Time Warping Barycentric Averaging(SW-DBA)的数据扩增方法,并构建深度卷积神经网络模型用于故障诊断。首先,通过分析传统数据扩增方法,提出基于SW-DBA的数据扩增模型。其次,通过搭建深度卷积神经网络建立故障诊断模型,并将利用扩增的新数据序列作为非平衡样本的补充,实现非平衡样本下的故障诊断。最后,通过人为设置两组不平衡样本下数据扩增前后的对比实验,分析故障诊断准确率分别由90.32%和80.57%提升至93.33%和93.04%,验证论文提出方法能有效改善数据不平衡问题。Aiming at the unbalanced data sample size problem in the process of rolling bearing fault diagnosis,this paper proposes a data amplification method based on SW-DBA,and constructs a deep convolutional neural network model for fault diagnosis.Firstly,by analyzing the traditional data amplification method,a data amplification model based on SW-DBA is proposed.Secondly,the fault diagnosis model is established by building a deep convolutional neural network,and the amplified new data series is used as a supplement to the unbalanced sample to achieve fault diagnosis under the unbalanced sample.Finally,by artificially setting up the comparative test before and after data amplification under the two sets of unbalanced samples,the accuracy of analysis fault diagnosis is increased from 90.32%and 80.57%to 93.33%and 93.04%,respectively,which verifies that the proposed method can effectively improve the data imbalance problem.

关 键 词：故障诊断 数据扩增 DBA DCNN 

分 类 号：TH17[机械工程—机械制造及自动化]