Study On Parameter Measurement For Gas-liquid Two-phase Flow In Small Channel Based On Image Processing

Gas-liquid two-phase flow widely exists in nature and industrial production process.In recent years,due to the development and dissemination of small-scale industrial devices,gas-liquid two-phase flow in small channels has attracted more and more attention.However,research on parameter measurement of gas-liquid two-phase flow in small channels is still limited,lacking perfect theory and accurate mechanism model,and related measurement technique is not advanced.Therefore,it has important scientific research value and engineering application prospect to study the parameter measurement of gas-liquid two-phase flow in small channels.High speed photography is a commonly used method for parameter measurement of gas-liquid two-phase flow,which has the advantages of visualization,noninvasion and instantaneity.In this thesis,flow pattern identification and void fraction measurement.in 'small channels 'are studied based on image processing.The main work and innovation are listed as follows:1.A gas-liquid two-phase flow parameter measurement system in small channels is established based on high speed photography.The system consists of three components:a fluid control unit,a virtual binocular image acquisition unit and a void fraction calibration unit.Experiments are carried out in horizontal and vertical channels with the inner diameter of 2.1mm,3.0mm and 4.0mm,respectively.In horizontal channels,bubble flow,slug flow,stratified flow and annular flow are observed,while in vertical channels,bubble flow,slug flow and annular flow are observed.2.A flow pattern identification method based on convolutional neural network is presented.Firstly,a convolutional neural network model for flow pattern identification is designed,including 12 network layers.Secondly,typical flow pattern images are captured,and then the data set is expanded.Finally,the model is trained and validated by using Keras deep learning framework.Experimental results show that the identification accuracies of typical flow patterns are all above 96%in horizontal and vertical channels.Convolutional neural network model is used to avoid the explicit feature extraction process and to achieve end-to-end learning.3.A void fraction measurement method based on quasi-three-dimensional model is presented.Firstly,gas-liquid two-phase flow images are captured from two orthogonal directions by the virtual binocular image acquisition unit.Secondly,binary image pairs are obtained by pre-processing operation.Then,the binary image pairs are reconstructed to obtain quasi-three-dimensional model images.Finally,void fraction is calculated by quasi-three-dimensional model images.Experimental results show that the absolute error between measurement values and reference values are all less than 7%.Compared with the use of the images from single direction,the results have certainly improvement.In this thesis,the convolutional neural network is introduced to identify the flow pattern,and the method of quasi-three-dimensional reconstruction is introduced to measure the void fraction.Experimental results show that this method is feasible and effective,and can provide a good reference for parameter measurement of gas-liquid two-phase flow in small channels.