Research On Wire-Mesh Sensor Super Resolution Of Two-Phase Flow Based On Complex Network And Deep Learning

Two-phase flow widely exists in industrial production.Understanding the dynamics underlying flow patterns is of great significance to develop the national economy and national defense.Wire-Mesh sensor is a kind of intrusive sensors,which can directly measure the local conductivity and then convert it to the local gas fraction.Therefore,it has been widely used in the field of two-phase flow detection.In recent years,deep learning has been widely applied in the field of image processing.Based on its successful experience in the tasks of image classification and image segmentation,lots of deep learning super resolution methods are proposed for image reconstruction.Compared with the traditional method,this method can restore more detailed features of the image and obtain better imaging effect.In the past decade,complex network theory has been developed rapidly,showing great potential for characterizing the nonlinear dynamics of complex systems.Particularly,it has been successfully applied in the field of non-linear time series analysis.In order to characterize the complicated gas-liquid flow behavior governing the flow transitions in a 50mm-inner-diameter vertical pipe,a series of studies have been carried out in this dissertation.The main work are as follow:(1)We design the two-phase flow measurement system based on Wire-Mesh sensor to acquire the local flow signals.According to the working principle of the sensor,the system mainly consists of three parts: the excitation module,the acquisition module and the communication module.The excitation module is designed to excite the transmitting electrodes cyclically by the square wave;The acquisition module is designed to acquire the signals synchronously at the receiver electrodes;The communication module is designed to send out the acquisition data to the host computer.On this basis,the gas-liquid two-phase flow experiment in a50mm-inner-diameter vertical pipe is carried out.Different flow conditions are generated by keeping the gas velocity constant and changing the liquid velocity.Then the local flow signals under different conditions are acquired by the Wire-Mesh sensor measurement system.(2)We reconstruct the gas-liquid distribution map of the flow section based on the method of deep learning super resolution,in order to obtain the better imaging effect than traditional methods.Three super resolution models,SRCNN,FSRCNN and SRGAN,are applied to raise the blurred images resolution and generate the clear images,which are compared with the images generated by bicubic interpolation method.We use the evaluation index called ‘No Reference Structure Similarity'(NRSS),which is proposed to quantify the high frequency components in the image,to evaluate the image quality.The results show that compared with the traditional interpolation method,the method based on the deep learning super resolution can restore more abundant information such as the edge details and texture details,leading to the better imaging effect.(3)We propose a Wire-Mesh sensor measurements analysis framework based on complex network,aiming to characterize the flow behavior in the transition from bubble flow to slug flow in a 50mm-inner-diameter vertical pipe.We infer a complex network from Wire-Mesh sensor measurements in terms of the Mutual Information between time series,and then calculate network measures to quantitatively characterize the network topological features associated with flow behavior.The networks corresponding to different flow patterns show different topological features,which allows us to characterize the transition of flow behavior.The analysis results show that the complex network analysis framework proposed in this dissertation can effectively reveal the complex flow behavior of gas-liquid two-phase flow.