基于GADF融合RDSAN的跨工况轴承故障诊断
Bearing Fault Diagnosis Across Working Conditions Based on GADF and RDSAN
瞿红春 1韩松钰 1贾柏谊 1马文博 1詹亦宏 1台合泽1
作者信息
- 1. 中国民航大学航空工程学院,天津 300300
- 折叠
摘要
针对在不同工况下获取到的滚动轴承振动数据特征分布存在差异,以及由于存在噪声而导致故障特征不明显的问题,提出了一种融合格拉姆角差场(GADF)与残差深度子领域自适应(RD-SAN)模型的跨工况轴承故障诊断方法.首先,为充分利用GADF在故障特征差异化显示上的优势,利用GADF来生成滚动轴承一维振动时域信号对应的图像数据集;其次,将数据集输入RDSAN模型,其中使用由改进图像集预训练的ResNet-18 网络结构进行源域与目标域通用特征的进一步提取,并引入局部最大均值差异(LMMD)计算匹配条件分布距离进行子领域自适应;最后,在添加0.5 dB高斯白噪声的CWRU滚动轴承数据集上进行跨工况试验验证,结果表明所提方法的平均诊断精度达到96.8%;将所提出的方法与不同的诊断方法进行比较分析,结果验证了该方法的有效性和优越性.
Abstract
In order to solve the problem that the characteristic distribution of vibration data obtained under different working conditions is different and the fault characteristics are not obvious due to the existence of noise,a cross-working bearing fault diagnosis method based on qualified ram angle difference field(GADF)and residual depth sub-domain adaptive(RDSAN)model is proposed.Firstly,in order to make full use of the advantages of GADF in the differential display of fault characteristics,the image data set cor-responding to the time domain signal of one-dimensional vibration of rolling bearings is generated by GADF.Then,the data set is fed into the RDSAN model,where the common features of source domain and target domain are further extracted using the ResNet-18 network structure pre-trained by the improved im-age set,and the local maximum mean difference(LMMD)is introduced to calculate the matching condi-tion distribution distance for subdomain adaptation.Finally,cross-working test is carried out on CWRU roll-ing bearing data set with 0.5 dB Gaussian white noise.The results show that the average diagnostic accura-cy of the proposed method reaches 96.8%.The proposed method is compared with other diagnostic meth-ods,and the results prove the effectiveness and superiority of the proposed method.
关键词
滚动轴承/故障诊断/跨工况/格拉姆角差场/子领域自适应Key words
rolling bearing/fault diagnosis/variable operating conditions/gram angle difference field/sub domain adaptation引用本文复制引用
基金项目
中国民航大学科研基金项目(05yk08m)
中央高校基本科研项目(ZXH2010D019)
出版年
2024
NETL
NSTL
万方数据