基于XGBoost的气液两相流流型超声识别方法  

Ultrasonic identification of gas-liquid two-phase flow patterns based on XGBoost

作　　者：苏茜 白凡 刘振兴[1,2] 刘彰 SU Qian;BAI Fan;LIU Zhenxing;LIU Zhang(Engineering Research Center of Metallurgical Automation and Measurement Technology,Wuhan University of Science and Technology,Wuhan 430081,Hubei,China;School of Artificial Intelligence and Automation,Wuhan University of Science and Technology,Wuhan 430081,Hubei,China;Hunan Valin Xiangtan Iron and Steel Co.,Ltd.,Hunan Iron and Steel Group,Xiangtan 411100,Hunan,China)

机构地区：[1]武汉科技大学冶金自动化与检测技术教育部工程研究中心,湖北武汉430081 [2]武汉科技大学人工智能与自动化学院,湖北武汉430081 [3]湖南钢铁集团湘潭钢铁集团有限公司,湖南湘潭411100

出　　处：《化工进展》2025年第4期1786-1793,共8页Chemical Industry and Engineering Progress

基　　金：国家自然科学基金(61903281,61901423,51907144);湖北省自然科学基金(2019CFB145);中国博士后科学基金(2018M642932);武汉市知识创新专项曙光计划(2022010801020311)。

摘　　要：气液两相流现象广泛存在于石油开采与运输、能源化工、航空航天等诸多领域。针对气液两相流流型识别问题,基于有限元多物理耦合仿真技术,建立典型气液两相全稳态流型的二维几何剖分仿真模型。设计双发四收的超声换能器收发方式以及三时段组合采样的采样模式对气液两相流全流型进行测试,结合超声波在气液流体中的传播机理对声压信号进行特征映射,并作为极限梯度提升树(XGBoost)分类算法的输入参数,实现对气液两相流分层流、泡状流、环状流和塞流4种流型分类。在此基础上,通过挖掘超声机理对分层流和塞流两类流型进行细分,即区分平滑分层流,波状分层流和塞状流、段塞流流型,从而实现对气液两相流全流型分类。超声传播机理特征与时域特征分类效果对比结果表明:搭建的基于超声的多接收分布式超声测试系统能提取更具流型识别性的超声机理特征参数,相较于时频特征具有较高的识别率,气液两相流分层流、泡状流、环状流和塞流总体识别率为98.5%,其中平滑分层流和波状分层流最高识别率为96.15%,气液塞状流和气液段塞流最高识别率为96.85%。The phenomenon of gas-liquid two-phase flow widely exists in many fields such as petroleum extraction and transportation,energy and chemical industry,aerospace and so on.Based on the finite element multiphysics coupling simulation technology,a two-dimensional geometric section simulation model of typical gas-liquid two-phase full-steady-state flow was established to recognize the gas-liquid two-phase flow pattern.The design of the ultrasonic transducer transmitting and receiving modes with two transmitters and four receivers and the sampling mode of three-time combined sampling was used to test the gas-liquid two-phase flow patterns,and the feature mapping of the sound pressure signals combined with the ultrasonic propagation mechanism in gas-liquid fluids was used as the input parameter of the extreme gradient boosting(XGBoost)classification algorithm to realize the classification of the gas-liquid two-phase flow patterns into four types:laminar flow,vesicular flow,annular flow,and plug flow.On this basis,the two types of laminar flow and plug flow were subdivided by exploiting the ultrasonic mechanism,i.e.,smooth laminar flow,undulating laminar flow and plug flow,segmental plug flow,so as to realize the classification of all flow types of gas-liquid two-phase flow.Comparison of ultrasonic propagation mechanism features and time-frequency domain features showed that the ultrasonic-based multi reception distributed ultrasonic testing system constructed in this study was able to extract ultrasonic mechanism feature parameters with better flow pattern recognition,and had a higher recognition rate compared with the time-frequency features.The recognition rate of gas-liquid two-phase flow,laminar flow,vesicular flow,annular flow,and plug flow,was 98.5%.Among these,the highest recognition rate of the smooth laminar flow and undulating laminar flow was 96.15%,while that of plug flow and segmental plug flow reached 96.85%.

关 键 词：超声测试 有限元仿真 气液两相流 极限梯度提升树 流型识别 

分 类 号：TE81[石油与天然气工程—油气储运工程] TB551[理学—物理]