基于卷积注意力机制的阀门内漏声发射识别方法  

Acoustic emission recognition method for valve internal leakage based on convolutional attention mechanism

作　　者：黄鑫[1] 屈文忠[1] 肖黎[1] HUANG Xin;QU Wenzhong;XIAO Li(Department of Engineering Mechanics,Wuhan University,Wuhan 430072,China)

机构地区：[1]武汉大学工程力学系,武汉430072

出　　处：《振动与冲击》2024年第9期105-114,共10页Journal of Vibration and Shock

基　　金：国家自然科学基金(51975581)。

摘　　要：阀门结构作为核电厂的关键设备部件之一,因长期处于高温高压环境下,其闸板或阀瓣易发生热变形或磨损导致密封不良,进而引发内漏事故。实时在线识别阀门的内漏状态,对提升核电机组热效率、提高阀门可靠性具有重要意义。因实际工业现场的基底噪声极易掩盖阀门内漏的声发射信号,进而造成阀门内漏状态的误判。为实现阀门内漏状态的快速准确识别,搭建了阀门内漏检测试验台架,开发了基于声发射方法的阀门内漏检测分析系统,将卷积注意力机制引入卷积神经网络中,实现高效快速地识别阀门内漏状态。结果表明,基于阀门内漏的声发射信号频域数据,利用卷积注意力机制神经网络能有效准确地识别阀门内漏状态,在内漏率为26 L/h时,识别准确率高达98%,并且具有较好的可靠性和鲁棒性。As one of key components of nuclear power plant,valve structure is prone to thermal deformation or wear of its gate or valve disc due to long-term exposure to high-temperature and high-pressure environment to cause poor sealing and ultimately lead to internal leakage accident.Real time online identifying valve internal leakage status is of great significance for improving thermal efficiency of nuclear power units and enhancing valve reliability.Due to substrate noise in actual industrial site being easy to cover acoustic emission signals of valve internal leakage,valve leakage status was easy to misjudge.To realize fast and correct recognition of valve leakage status,a valve internal leakage detection test bench was built,and a valve leakage monitoring and analysis system based on acoustic emission was developed.Convolutional attention mechanism was introduced into convolutional neural network to realize efficient and fast identification of valve leakage status.The results showed that based on frequency domain data of acoustic emission signals of valve internal leakage,the convolutional attention mechanism neural network can effectively and accurately identify valve internal leakage status;when the internal leakage rate is 26 L/h,the proposed method’s recognition accuracy is up to 98%with better reliability and robustness.

关 键 词：阀门结构 内漏 声发射 卷积注意力模块 卷积神经网络 

分 类 号：TH134[机械工程—机械制造及自动化] TG115.28[金属学及工艺—物理冶金]