摘要
气膜密封装置是工业领域应用广泛的一种密封技术,其可靠的密封性能对于设备正常运行至关重要。气膜密封装置的动静密封环接触端面相对运动会产生摩擦,摩擦过程会产生复杂的声发射信号,这些信号往往隐含密封端面运行状况的重要信息。采用传统的方法往往难以准确识别和分类这些微弱的特征信号,因此需要开发更高精度的故障诊断方法。针对机械密封动、静环端面摩擦状态难以识别这一问题,以气膜密封装置为研究对象,提出了一种基于深度融合模型的气膜密封端面状态识别方法。首先,采用声发射传感器及采集设备,对密封端面的声发射信号进行了采集;其次,利用小波包变换方法对采集到的信号进行了滤波处理,并提取了时域和频域的微弱特征;然后,将深度随机森林(DRF)作为分类层融入卷积神经网络(CNN)形成了融合模型,对预先处理过的密封装置运行状态的特征信息进行了识别和分类;最后,根据实验的泄漏量,使用混淆矩阵和受试者工作曲线分析了两种模型的特征提取能力。研究结果表明:CNN-DRF融合模型对于密封端面声发射信号的两种特征识别精度分别为96%和98%,与传统的CNN模型相比,其可以充分提取信号特征信息,具有更出色的故障诊断能力。
The air film sealing device is one of the most widely used sealing technologies in the industrial field,its reliable sealing performance is essential for the normal operation of the equipment.The relative movement of the contact end faces of the dynamic and static sealing rings of the gas film sealing device will produce friction,and the friction process will produce complex acoustic emission signals,which often imply important information about the operation status of the sealing end faces.It is often difficult to accurately identify and classify these weak characteristic signals by traditional methods.Therefore,it is necessary to develop fault diagnosis methods with higher precision.In order to solve the problem that it was difficult to identify the friction state of the dynamic and static ring end faces of mechanical seals,taking the air film sealing device as the research object,a method for identifying the end face of air film seals based on the deep fusion model was proposed.Firstly,the acoustic emission signal of the sealed end face was collected by the acoustic emission sensor and acquisition equipment.Secondly,the wavelet packet transform method was used to filter the collected signal and extract the weak features in the time domain and frequency domain,and then the deep random forest(DRF)was integrated into the convolutional neural network(CNN)as a classification layer.Finally,according to the leakage amount of the experiment,the confusion matrix and the receiver operating curve were used to analyze the feature extraction ability of the two models.The research results show that the accuracy of the CNN-DRF fusion model for the two features of the sealed end-face acoustic emission signal is respectively 96%and 98%,which can fully extract the signal feature information and has better fault diagnosis ability than the traditional CNN model.
作者
刘伟
张书尧
李双喜
马亚宾
梁坤海
LIU Wei;ZHANG Shuyao;LI Shuangxi;MA Yabin;LIANG Kunhai(Collage of Mechanical and Electrical Engineering,Beijing University of Chemical Technology,Beijing 100029,China;Software R&D Department,Sage Energy Technology(Beijing)Co.,Ltd.,Beijing 100102,China)
出处
《机电工程》
CAS
北大核心
2024年第7期1198-1206,共9页
Journal of Mechanical & Electrical Engineering
基金
国家重点研发计划项目(2018YFB2000800)。
关键词
气膜密封技术
机械密封
声发射信号
小波包变换方法
融合模型
深度随机森林
卷积神经网络
特征提取
特征识别精度
air film sealing technology
mechanical seal
acoustic emission signal
wavelet packet transform method
fusion model
deep random forest(DRF)
convolutional neural network(CNN)
feature extraction
feature recognition accuracy
