首页|基于同步挤压小波变换和Transformer的轴承故障诊断模型

基于同步挤压小波变换和Transformer的轴承故障诊断模型

扫码查看
原文链接
  • NETL
  • NSTL
  • 万方数据
针对采用神经网络对滚动轴承进行故障诊断时,故障信息利用不充分,特征提取困难的问题,提出了一种基于同步挤压小波变换(SST)-Transformer的滚动轴承智能故障诊断方法.首先,以同步挤压小波变换作为信号处理模块,将一维振动信号转为时频图;接着,设计了一种最大程度保留故障信息的时频图分割方式,将时频图分割为一系列图像块序列;然后,将序列输入到具有强大的处理序列数据能力的Transformer模型中,进行了特征提取;最后,将特征数据输入分类器进行了分类,对比了不同的时频图分割方式的诊断效果,并将SST-Transformer模型与基准算法相比较.研究结果表明:相较于其他分割方式,基于SST-Transformer的滚动轴承智能故障诊断方法的诊断准确率提升了3.45%,并大幅提升了模型训练的收敛速度;相比于其他基准算法,该方法的平均准确率至少提升了1.05%.该方法有较高的诊断准确率和较好的稳定性.
Fault diagnosis model of rolling bearing based on synchro squeezing wavelet transform and Transformer
A rolling bearing intelligent fault diagnosis method based on synchronous squeezing wavelet transform(SST)-Transformer was proposed to address the problem of insufficient utilization of fault information and difficulty in extracting features,when using neural networks for fault diagnosis of rolling bearings.Firstly,synchronous squeezing wavelet transform was used as the signal processing module,the one-dimensional vibration signal was transformed into a time-frequency map.Next,a time-frequency map segmentation method was designed to preserve fault information to the greatest extent possible,dividing the time-frequency map into a series of image block sequences.Then,the sequence was input into a Transformer model with strong processing capabilities for sequence data,and feature extraction was performed.Finally,the feature data was input into the classifier for classification,and the diagnostic performance of different time-frequency map segmentation methods was compared.The SST Transformer model was also compared with the benchmark algorithm.The research results show that compared to other segmentation methods,the intelligent fault diagnosis method for rolling bearings based on SST Transformer has improved the diagnostic accuracy by 3.45%and significantly improved the convergence speed of model training.Comparing to other benchmark algorithm,the average accuracy has improved by at least 1.05%.The method has high diagnostic accuracy and good stability.

intelligent fault diagnosisneural network(NN)fault feature extractionattention mechanismdeep learningsynchro squeezing wavelet transform(SST)Transformer model

张向宇、王衍学

展开 >

北京建筑大学 机电与车辆工程学院,北京 100044

故障智能诊断 神经网络 故障特征提取 注意力机制 深度学习 同步挤压小波变换 Transformer模型

国家自然科学基金国家自然科学基金

5227507951875032

2024

10.3969/j.issn.1001-4551.2024.06.007

机电工程
浙江大学　浙江省机电集团有限公司

机电工程

CSTPCD北大核心
影响因子：0.785
ISSN：1001-4551
年,卷(期)：2024.41(6)