针对浮环密封装置在不同摩擦工况下振动信号特征微弱且难以识别的问题，传统深度学习网络在性能上取得了显著成果。然而，由于存在模型超参数多、训练时间长、迭代次数多、网络精度低以及计算成本过高等问题，其在实际使用时存在一定的局限性。因此，本研究提出了一种基于深度残差网络（ResNet）的方法，即基于快速注意力机制的残差网络（Quick Squeeze Excitation ResNet），以解决浮环密封摩擦振动信号分类中的问题。该方法通过引入注意力机制、调整网络残差块的连接方式并选择特定的优化器来提高模型性能，并与其他四种模型在同一个梅尔频谱图（Mel Spectrogram）数据集上进行对比测试。研究结果显示，QSE-ResNet的准确率达到了97%，比传统卷积神经网络（CNN）高出13%，同时缩短了55%的模型迭代次数，节约了30%的网络训练时间。此外，QSE-ResNet成功地解决了过拟合、梯度爆炸和梯度消失等问题，显著缩短了迭代次数、节省了网络训练时间并提高了测试精度，使得浮环密封摩擦振动的信号状态监控及设备部署更为便利。本研究提出的QSE-ResNet使得浮环密封摩擦振动模型具备更便捷的部署能力，为浮环密封摩擦振动信号的研究提供了新的思路。
Abstract
In order to solve the problem that the vibration signal characteristics of the floating ring sealing device are weak and difficult to identify under different friction conditions, the traditional deep learning network has achieved remarkable results in performance. However, due to the problems of many hyperparameters, long training time, many iterations, low network accuracy and high computational cost, the model has certain limitations in practical use. Therefore, this study proposes a new method based on the Deep Residual Network (ResNet), which is based on the Fast Attention Mechanism (Quick Squeeze Excitation ResNet), to solve the problem in the classification of friction and vibration signals of floating ring seals. This method improves the performance of the model by introducing an attention mechanism, adjusting the connection of the network residual blocks, and selecting a specific optimizer, and compares the test with the other four models on the same Mel spectrogram dataset. The results show that the accuracy of QSE-ResNet reaches 97%, which is 13% higher than that of traditional convolutional neural networks (CNN), while shortening the number of model iterations by 55% and saving 30% of network training time. In addition, QSE-ResNet successfully solves the problems of overfitting, gradient explosion and gradient vanishing, which significantly shortens the number of iterations, saves network training time and improves test accuracy, making the signal status monitoring and equipment deployment of floating ring seal friction vibration more convenient. The QSE-ResNet proposed in this study makes the floating ring seal friction and vibration model more convenient to deploy, and provides a new idea for the study of floating ring seal friction vibration signal.
关键词
浮环密封 /
摩擦振动 /
声发射 /
深度学习 /
特征分类
{{custom_keyword}} /
Key words
floating ring sealing /
friction and vibration /
acoustic emission /
deep learning /
feature classification
{{custom_keyword}} /
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 陈潇竹. 自适应分瓣式碳环密封间隙分析及性能研究[D]. 北京:北京化工大学,2022.
CHEN Xiaozhu. Gap analysis and performance of adaptive split-flap carbon ring seal[D]. Beijing:Beijing University of Chemical Technology, 2022.
[2] Duan W ,Chu F ,Kim C , et al. A bulk-flow analysis of static and dynamic characteristics of floating ring seals[J].Tribology International,2007,40(3):470-478.
[3] Zhang Erqing, Fu, et al.Mechanical Seal Opening Condition Monitoring Based on Acoustic Emission Technology[J]. Sensors Transducers, 2014, 172(6):139-146.
[4] 李双喜,黄柏淇,廖浩然等.航空发动机轴承腔气液两相润滑机械密封性能分析[J].润滑与密封,2023,48(03):30-39.
LI Shuangxi, HUANG Baiqi, LIAO Haoran, et al.Performance analysis of gas-liquid two-phase lubrication mechanical seal in aero engine bearing cavity[J].Lubrication and Sealing, 2023, 48(03): 30-39.
[5] 郑娆,陈潇竹,李双喜等.高速气膜镶装式浮环密封的开启特性[J].北京航空航天大学学报,2022,48(11):2111-2120.
ZHENG Rao, CHEN Xiaozhu, LI Shuangxi, et al. Opening characteristics of high-speed air film inlaid floating ring seal[J]. Journal of Beijing University of Aeronautics and Astronautics, 2022,48(11): 2111-2120.
[6] HUANG W, LIN Y, GAO Z, et al. An acoustic emission study on the starting and stopping processes of a dry gas seal for pumps[J]. Tribology Letters,2013,49(2): 379-384．
[7] 肖思文, 廖传军, 李学军. 小波尺度谱在AE信号特征提取中的应用[J]. 中国工程科学,2008(11):69-75.
XIAO Siwen, LIAO Chuanjun, LI Xuejun. Application of wavelet scale spectroscopy in AE signal feature extraction [J]. Engineering Science, 2008(11):69-75.)
[8] 林志斌, 傅攀, 张尔卿等. 基于声发射和小波神经网络的机械密封状态分类新方法[J].润滑与密封, 2014, 39(09): 40-45.
LIN Zhibin, FU Pan, ZHANG Erqing, et al. A new method for classification of mechanical seal state based on acoustic emission and wavelet neural network[J]. Lubrication and Sealing, 2014, 39(09): 40-45.)
[9] 陆俊杰,刘柱,丁雪兴等.局部特征尺度与小波阈值协同下密封摩擦面AE信号降噪研究[J].振动与冲击, 2023, 42(12):205-211+219.
LU Junjie, LIU Zhu, DING Xuexing, et al. Research on AE signal noise reduction of sealing friction surface under the synergy of local eigenscale and wavelet threshold[J]. Vibration and Shock, 2023, 42(12):205-211+219.
[10] Hossain ,Shahjahan M,Taheri , et al. In-situ process monitoring for metal additive manufacturing through acoustic techniques using wavelet and convolutional neural network (CNN)[J].The International Journal of Advanced Manufacturing Technology, 2021, 116(11-12): 1-16.
[11] 翁泽文, 力宁, 袁俊马, 等. 基于声发射时频分析与卷积神经网络的液膜密封摩擦状态识别[J]. 润滑与密封, 2023, 48 (1): 136-141.
WENG Zewen, LI Ning, YUAN Junma, et al. Friction State Recognition of Liquid Film Seal Based on Acoustic Emission time-frequency Analysis and Convolutional Neural Network [J]. Lubrication Engineering, 2023, 48 (1): 136-141.
[12] 刘怀顺. 基于声发射的液膜密封启动过程状态监测研究[D].中国石油大学(华东),2021.
LIU Huaishun. Research on state monitoring of liquid film seal start-up process based on acoustic emission[D].China University of Petroleum(East China),2021.
[13] 力宁,江平,翁泽文等.航空发动机浮环密封上浮性能试验研究[J].润滑与密封,2020,45(11):143-148.
LI Ning,JIANG Ping,WENG Wenhang,et al.Experimental study on floating performance of floating ring seals used in aero engine[J].Lubrication Engineering[13],2020,45(11):143-148.
[14] 雷杰，高鑫，宋杰，等. 深度网络模型压缩综述[J]. 软件 学报，2018，29(2)：251-266.
LEI Jie, GAO Xin, SONG Jie, et al. Survey of deep neural network model compression[J]. Journal of Software, 2018, 29(2): 251-266.
[15] 力宁,江平,翁泽文等.航空发动机浮环密封上浮性能试验研究[J].润滑与密封,2020,45(11):143-148.
LI Ning, JIANG Ping, WENG Zewen, et al.Experimental study on floating performance of aero engine floating ring seal[J]. Lubrication and Sealing, 2020, 45(11):143-148.
[16] 熊忠汾. 浮环气膜密封自振稳定性与不平衡响应分析[D].昆明理工大学.
XIONG Zhongfen. Analysis of natural stability and unbalanced response of floating ring gas film seal[D].Kunming University of Science and Technology.
[17] 李怀城,杨道武,温治芳,等. 基于Inception-CSA深度学习模型的鸟鸣分类[J]. 华中农业大学学报,2023,42(3):97-104.
LI Huaicheng, YANG Daowu, WEN Zhifang, et al. Inception-CSA deep learning model-based classification of bird sounds[J]. Journal of Huazhong Agricultural University, 2023, 42(3):97-104.
[18] Krizhevsky A, Sutskever I, Hinton E G. ImageNet classification with deep convolutional neural networks[J]. Communications of the ACM, 2017, 60(6): 84-90.
[19] Zhang P, Jiang W, Zheng Y, et al. Hydraulic-Pump Fault-Diagnosis Method Based on Mean Spectrogram Bar Graph of Voiceprint and ResNet-50 Model Transfer[J].Journal of Marine Science and Engineering,2023,11(9):
[20] R.G. S R , Saradhi G V ,Lakshmi R D .Resnet-based modified red deer optimization with DLCNN classifier for plant disease identification and classification[J].Computers and Electrical Engineering, 2023,1S05
[21] 刘乙萱, 苏鑫. 基于两阶段迁移学习的Multi-scale SE-ResNet50深度卷积神经网络的多标签航空图像分类问题研究[J]. 数学的实践与认识,2023,53(06):174-186.
LIU Yixuan, SU Xin. Research on multi-label aerial images classification based on two-stage transfer learning Multi-Scale SE-ResNet50 deep convolutional neural networks[J]. Practice and Understanding of Mathematics, 2023, 53(06): 174 - 186.
[22] Jie H ,Li S ,Samuel A , et al. Squeeze-and-Excitation Networks [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2019,42(8):1-1.
[23] 史红梅,郑畅畅,司瑾,等.基于动态加权的多尺度残差网络旋转机械故障诊断算法[J].振动与冲击,2022,41(23):67-74,93.
SHI Hongmei, ZHENG Changchang, SI Jin, et al. Rotating Machinery Fault Diagnosis Algorithm Based on Dynamic Weighted Multi-scale Residual Network [J]. Journal of Vibration and Shock, 2012,41(23):67-74,93.
{{custom_fnGroup.title_cn}}
脚注
{{custom_fn.content}}
PDF(3322 KB)
