Research On Oil Bubble Flow Pattern Identification And Coalescence Detection In Vertical Upward Pipe

There exists much oil-water mixed flow in the petroleum industry.As a key parameter of two-phase flow,the flow pattern is important in modeling and controlling the oil-water two-phase flow system.Precisely identifying the oil-water flow pattern and understanding the flow pattern evolution character is of great importance for the flow measurement instrument design.Also,the flow pattern evolution information is important guidance for the two-phase flow control system optimization.Due to the phase interaction between oil and water,the oil-water two-phase flow patterns exhibit quite complex character.How to accurately identify the flow pattern and detect the flow state transition is still a problem to be solved.Especially,when the oil-water mixing flow rate is low,the interaction between the dispersed oil phases leads to be obvious coalescence phenomenon.There still lacks effective method to accurately detect the coalescence phenomenon of oil bubbles.We first collected flow pattern images and conductance fluctuation signals with an oil-water two-phase flow experiment.Based on the typical deep convolutional neural network structure,three models of oil-water two-phase flow pattern identifier are proposed.The three flow pattern identifiers are models based on LeNet-5,AlexNet and VGGnet-16,respectively.We compared the flow pattern identification results of the three proposed model,and the results showed that all the three models had a relatively high accuracy for the oil slug pattern identification.With the increase of network depth,flow pattern identification model based on VGGnet-16 has much higher identification accuracy for bubble flow and VFD than that of LeNet-5 and AlexNet.Based on recurrent neural network structure,we designed a model to predict the oil-water two-phase flow coalescence.A multivariate characteristic series representation of the flow pattern conductance signals is designed as the input of the recurrent neural network.The number of layers of the recurrent neural network based model is determined by model parameter optimization and experiments.With the proposed model,we successfully identify the oil bubble flow coalescence phenomenon.The oil bubble coalescence dynamics also have been studied with the coalescence detecting results and the flow condition.We find that the phase volume fraction is a key factor that affecting the coalescence phenomenon.With the increase in water phase volume,the number of coalescence is significantly decreased and the coalescence tends to be more inhomogeneous,which shows that the asymmetry of the flow structure is gradually increased.We also in this research employ the Visual Studio platform to establish an oil-water two-phase flow pattern identification system.The system identifies the flow patterns images through graphical interfaces,and it has the advantages of simplicity,fastness and intuition.In addition,the system can predict the bubble coalescence phenomenon in real time by analyzing the collected oil-water two-phase flow fluctuation signals.