基于扩散模型的滚动轴承小样本故障诊断研究  

Research on Small Sample Sault Diagnosis of Rolling Bearing Based on DDPM Diffusion Model

作　　者：吴静远 舒启林[1] 王耿 李明昊 魏永合[1] WU Jingyuan;SHU Qilin;WANG Geng;LI Minghao;WEI Yonghe(School of Mechanical Engineering,Shenyang Ligong University,Shenyang 110159,China)

机构地区：[1]沈阳理工大学机械工程学院,沈阳110159

出　　处：《组合机床与自动化加工技术》2024年第12期117-122,共6页Modular Machine Tool & Automatic Manufacturing Technique

基　　金：国家自然科学基金资助项目(51875368);辽宁省科技厅应用基础研究资助项目(2022JH2/101300230);辽宁省教育厅基本科研资助项目(LJKQZ20222448)。

摘　　要：足够的训练样本数量是智能故障诊断达到较高准确率的基础,一方面传统小样本问题的解决方法,其扩充过程不稳定、生成样本质量不高;另一方面去噪扩散概率模型(denoising diffusion probability model,DDPM)在高质量图像生成等领域被广泛应用。基于此,提出一种基于DDPM的小样本故障诊断方法。首先,将轴承原始振动信号通过连续小波变换得到二维时频图;然后,利用DDPM对小样本进行扩充,将扩充样本集用于训练基于卷积神经网络的故障诊断模型。实验结果表明,该方法能有效提高故障诊断的准确率,具备有效性和优越性。A sufficient number of training samples is the basis for intelligent fault diagnosis to achieve higher accuracy.On the one hand,the traditional solution to the small sample problem is unstable in the expansion process and the quality of the generated samples is low.On the other hand,the denoising diffusion probability model(DDPM)is widely used in fields such as high-quality image generation.Based on this,this paper proposes a small sample fault diagnosis method based on DDPM.First,the original vibration signal of the bearing is transformed through continuous wavelet transformation to obtain a two-dimensional time-frequency diagram.Then,DDPM is used to expand the small sample,and the expanded sample set is used to train a fault diagnosis model based on convolutional neural network.Experimental results show that this method can effectively improve the accuracy of fault diagnosis and is effective and superior.

关 键 词：故障诊断 小样本 DDPM 滚动轴承 

分 类 号：TH165[机械工程—机械制造及自动化] TG659[金属学及工艺—金属切削加工及机床]