基于UMAP改进的多域特征提取方法及轴承故障诊断
Improved Multi-Domain Feature Extraction Method and Bearing Fault Diagnosis Based on UMAP
尹泽明 1王彩年 2王智 1毛范海1
作者信息
- 1. 大连理工大学机械工程学院,大连 116024
- 2. 通用技术集团大连机床有限责任公司,大连 116620
- 折叠
摘要
针对传统多域特征提取方法占用计算资源过大、分类精度不足等问题,提出了一种基于统一流行逼近与投影算法(UMAP)改进的多域特征提取方法.通过对原始信号进行多域特征采集结合UMAP的全局信息提取能力进行信息融合与低维映射重构特征集;在此基础上将特征集输入到支持向量机中进行模型训练,实现轴承的故障识别与诊断.基于某大学公开的滚动轴承实验数据集对比分析了几种典型的优化算法与传统多域特征提取方法,证明所提方法识别滚动轴承故障状态的成功率为100%,验证了该方法的优越性.
Abstract
In order to solve the problem that the traditional multi-domain feature extraction method occu-pies too much computing resources and has insufficient classification accuracy,a multi domain feature ex-traction method is proposed based on unified manifold approximation and projection algorithm(UMAP).By combining the multi domain feature collection of the original signal with the global information extrac-tion capability of UMAP,information fusion and low dimensional mapping are performed to reconstruct the feature set;On this basis,the feature set is input into the support vector machine for model training to a-chieve bearing fault recognition and diagnosis.Based on the publicly available experimental dataset of roll-ing bearings at western reserve university in the united states,several typical optimization algorithms and traditional multi domain feature extraction methods were compared and analyzed.It was proved that the success rate of the proposed method in identifying rolling bearing fault states was 100%,and the superiority of this method was verified.
关键词
故障诊断/多域特征提取/统一流形逼近与投影/支持向量机Key words
fault diagnosis/multi-domain feature extraction/uniform manifold approximation and projec-tion/support vector machine引用本文复制引用
基金项目
辽宁揭榜挂帅项目(2021JH1/10400099)
大连市科技创新基金项目(2022JJ12GX032)
大连理工大学引进人才科研专项项目(DUT21RC-3-112)
出版年
2024
NETL
NSTL
维普
万方数据