基于路径重积分特征提取的轴承故障诊断
Feature Extraction of Time Series Signal Path Signature and Its Application in Fault Diagnosis
张浩然 1马萍 1张宏立 1王聪 1李新凯1
作者信息
- 1. 新疆大学电气工程学院,乌鲁木齐 830017
- 折叠
摘要
滚动轴承振动信号呈现非线性和非平稳特征,为充分挖掘滚动轴承振动信号的有效信息,提高故障诊断准确率,提出基于振动信号路径重积分(path signature,PS)的滚动轴承故障信号特征提取方法.首先,对一维故障振动信号进行时延重构,构成有限维路径空间;其次,对路径进行多重迭代积分得到高阶路径积分特征作为故障振动信号初始特征,利用主成分分析(principal component a-nalysis,PCA)对其进行降维得到能充分表征故障信号本征信息的特征;最后,将不同故障信号基于路径重积分的故障特征构成故障特征集,输入到支持向量机(support vector machine,SVM)完成故障的识别和分类.实验结果表明,该方法在公开数据集上的10 种故障类型诊断准确率达99.33%,对比其他几种方法,所提方法能快速准确地识别滚动轴承不同故障类型.
Abstract
The vibration signal of rolling bearing presents nonlinear and non-stationary characteristics.In order to fully exploit the effective information of the vibration signal of rolling bearings and improve the ac-curacy of fault diagnosis,a feature extraction method of rolling bearing fault signal based on path signature(PS)of vibration signal is proposed.Firstly,the one-dimensional fault vibration signal is delayed recon-structed to form a finite-dimensional path space.Secondly,the high-order path integral feature is obtained by multiple iterative integration of the path as the initial feature of the fault vibration signal,and the princi-pal component analysis(PCA)is used to reduce the dimension to obtain the feature that can fully charac-terize the intrinsic information of the fault signal.Finally,the fault features of different fault signals based on path signature constitute the fault feature set,which is input into the support vector machine(SVM)to complete the fault identification and classification.The analysis results show that the accuracy of this meth-od in diagnosing 10 types of faults on public datasets in 99.33%.Compared with other methods,the pro-posed method can quickly and accurately identify different fault types of rolling bearings.
关键词
路径空间/路径重积分/截断阶数/特征提取/故障诊断Key words
path space/path signature/truncation orders/feature extraction/fault diagnosis引用本文复制引用
基金项目
国家自然科学基金资助项目(52065064)
国家自然科学基金资助项目(52267010)
新疆维吾尔自治区自然科学基金资助项目(2022D01E33)
新疆维吾尔自治区自然科学基金资助项目(2022D01C367)
出版年
2024
NETL
NSTL
万方数据