基于CNN-VIT模型的非接触式机械密封多源故障状态识别方法

扫码查看

原文链接

国家科技期刊平台
NETL
NSTL
万方数据

中文摘要：针对非接触式机械密封多源故障信号频率高、识别难和易受干扰的难题,搭建密封试验台,模拟4种典型故障工况和正常运行工况,布置声发射(AE)测试系统测取10 000个特征样本,将卷积神经网络(CNN)结构与Vision Transformer模型相结合,并在CNN输出数据中引入Patch概念构建了一种改进模型,随后应用于密封故障状态识别,提出了一种新型的非接触式机械密封状态智能识别方法,并分析了改进模型各层网络决策对密封AE数据的影响.结果表明,AE技术能够高效监测非接触式机械密封多源故障状态信号,该方法对密封状态识别的准确率最高可达99.42%;改进模型更加关注稳定运行状态信号中频率为0～50 kHz和100～150 kHz的噪声频带数据,动环表面剥落状态信号中频率为100～150 kHz的噪声频带数据和270±40 kHz的摩擦AE频带数据.研究结果可为机械密封的智能化、高可靠、长寿命发展提供理论基础.

外文标题：Multi-source fault state recognition of non-contact mechanical seal based on CNN-VIT model

外文摘要：In response to the problems of high-frequency,recognition challenge,and susceptibility to interference of multi-source fault signals of non-contact mechanical seal,a seal test bench was built to simulate four characteristic fault conditions and standard operational conditions.An acoustic emission(AE)system was employed to collect 10 000 feature samples.By combining the Convolutional Neural Network(CNN)architecture with the Vision Transformer model and introducing the Patch concept into the CNN output,a modified model was developed and applied to seal fault state recognition.A novel method for the intelligent recognition of non-contact mechanical seal conditions was proposed,and the influences of the modified model's network layers on the seal AE data.The results show that the AE technique can effectively monitor the multi-source fault state signals of non-contact mechanical seals.This technique can achieve a maximum accuracy rate of 99.42%in recognizing seal conditions.The modified model more notably focuses on the noise frequency band data between 0～50 kHz and 100～150 kHz in the stable operation state signals,as well as the 100～150 kHz noise band data and the 270±40 kHz friction AE band data in the signals indicative of rotating ring surface peeling.The research results can provide a theoretical basis for future development of mechanical seals toward enhanced intelligence,high reliability and long service life.

外文关键词：

acoustic emissionnon-contact mechanical sealfault diagnosisdeep learningVision Transformer

作者：

陈金鑫、丁雪兴、陆俊杰、徐洁、张帅

展开 >

作者单位：

兰州理工大学石油化工学院,兰州 730050

浙大宁波理工学院宁波市极端密封重点实验室,浙江宁波 315000

关键词：

声发射非接触式机械密封故障诊断深度学习 Vision Transformer

基金：

国家自然科学基金项目宁波市自然科学基金-青年博士创新研究项目浙江省自然科学基金项目"科技创新2025"重大专项"科技创新2025"重大专项"科技创新2025"重大专项宁波市自然科学基金项目

项目编号：

519054802022J152LY22E0500102023Z0052022Z0542022Z0072023J271

出版年：

2024

DOI：

10.3969/j.issn.1005-0329.2024.09.012

流体机械

中国机械工程学会

流体机械

CSTPCD北大核心

影响因子：1.418

ISSN：1005-0329

年,卷(期)：2024.52(9)