Study On Intelligent Identification Of Gas-liquid Two-phase Flow Patterns Using Sound Signals From Non-Return Valve

Gas-liquid two-phase flow is widely used in nature and industrial production process,which is much more common in oil transportation pipeline.Safety monitoring of crude oil transportation pipeline is of importance for the development of economy and industry which are highly depending on oil resources.Safety monitoring of oil pipeline is to detect gas-liquid flow parameters and recognize malfunctions in a timely warning manner.Gas-liquid two-phase flow pattern in oil pipeline is one of the most important parameters in oil transportation safety monitoring.Because of the complexity of crude oil flow and flammable risks,the identification of gas-liquid two-phase flow pattern has not been completely fulfilled so far.This paper proposed a new method for classifying the gas-liquid two-phase flow patterns by using the sound signals from non-return valve.The non-return valve is designed and manufactured under the actual working condition of oil transportation,which is suitable for experimental study.A visual detection system with air and water used as working fluids was built to study the four typical flow patterns including liquid flow,gas flow,slug flow and churn flow pattern.The sound signals from non-return valve under different conditions were captured.The reasonable sampling frequency and time was set by this paper for efficiency of signal processing.According to the nonlinear characteristics of gas-liquid two-phase flow,Hilbert-Huang Transform(HHT)is chosen to deal with sound signal in this paper.In this paper,the EMD denoising algorithm is used to process the sampled signals and the signals are reconstructed after denoising.After pretreatment,the collected sound signals will be pretreated with EMD and Hilbert spectrum analysis so that the original signals' energy of IMF component Hilbert spectrum and the Hilbert marginal spectrum can be obtained.The study found that the energy of the third IMF component and its Hilbert spectrum and the reconstructed signals'Hilbert marginal spectrum can represent the sound signal characteristics of different flow patterns.So 2D and 3D flow-pattern maps were constructed with the coordinates of the three characteristics.Verification tests demonstrated that the correct identification rates are above 98.1%with the developed flow pattern maps based on sound signal analysis of non-return valve.In order to optimize this method and achieve the goal of intelligent identification of gas-liquid two-phase flow pattern.The artificial neural network(ANN)was chosen as the classifier of flow pattern identification and an error back propagation(BP)neural network model is designed in this paper.The three kinds of energy characteristics which are used to draw the flow pattern are set as inputs.In this paper,four flow patterns are selected as outputs and neural network are trained with multiple sets of data.After training,the correct identification rates of the neural network are better than 97.5%,which realizes the intelligent identification of flow patterns.Therefore,the flow pattern intelligent classification method based on the sound signal from non-return valve has preferable spread and application prospects.