基于FDVI和CDDPM的小样本岸桥齿轮箱多类故障诊断  

Multi-class and small-sample fault diagnosis of shore bridge gearboxes based on FDVI and CDDPM

作　　者：袁九海 张氢[1] 张建群 冯文宗 孙远韬[1] YUAN Jiuhai;ZHANG Qing;ZHANG Jianqun;FENG Wenzong;SUN Yuantao(School of Mechanical Engineering,Tongji University,Shanghai 201804,China)

机构地区：[1]同济大学机械与能源工程学院,上海201804

出　　处：《振动与冲击》2025年第6期306-317,共12页Journal of Vibration and Shock

基　　金：国家自然科学基金面上项目(52075389)。

摘　　要：岸桥齿轮箱零部件数量多、故障类型丰富,且故障数据难以获取,其诊断面临小样本、多分类的问题。针对上述问题,提出了一种基于频域振动图(frequency domain vibration image,FDVI)和条件去噪扩散概率模型(conditional denoising diffusion probabilistic model,CDDPM)的故障诊断方法。首先,将获取的振动信号转为FDVI图像,充分表征各故障的振动信号的特征信息;然后,使用CDDPM对小样本数据进行扩充,将标签信息输入到模型以控制生成故障样本类别,同时采用跳层采样加快样本生成速度;最后,将扩充后的样本集输入卷积神经网络分类器中进行训练,提升分类器对小样本多类故障诊断的效果。在对CWRU数据集的17种故障类型和岸桥缩尺试验台数据集的29种故障类型的小样本诊断试验表明:样本扩充后CWRU数据集故障识别率由89.86%提高到99.30%;岸桥数据集故障识别率由68.63%提高到95.75%。上述分析表明所提方法能完成小样本条件下岸桥齿轮箱多类故障诊断任务。The number of components and fault types of the shore bridge gearbox is large,while the fault data is difficult to obtain,and its diagnosis faces the problem of small sample and multiple classification.To address the above problems,a fault diagnosis method based on the frequency domain vibration image(FDVI)and the conditional denoising diffusion probabilistic model(CDDPM)was proposed.Firstly,the obtained vibration signals were transformed into images using the FDVI method,fully characterizing the feature information of vibration signals for each fault.Then,the CDDPM was used to expand the small sample data,and the labeling information was input to the model to control the generation of fault sample categories,while skip-layer sampling was used to accelerate the sample generation speed.The expanded sample were input into a convolutional neural network classifier for training to improve the classifier’s performance in diagnosing multi class faults with small samples.The small-sample diagnostic experiments on 17 faults in the CWRU dataset and 29 faults in the shore bridge scaling experimental platform dataset show that:after the sample expansion,the fault recognition rate of the CWRU dataset is increased from 89.86%to 99.30%;the fault recognition rate of the shore bridge dataset is increased from 68.63%to 99.30%.The above analysis shows that the method can accomplish the task of multi-class fault diagnosis for shore bridge gearboxes under small sample conditions.

关 键 词：频域振动图 条件去噪扩散概率模型 小样本 岸桥齿轮箱 故障诊断 

分 类 号：TH212[机械工程—机械制造及自动化] TH213.3