Machine learning identification of multiphase flow regimes in a long pipeline-riser system

Multiphase flow regime identification is a promising technology for ensuring flow safety in marine gathering pipelines. An experimental study of air-water flow regime was conducted on a 1687 m long S-shaped pipeline-riser system. The flow regimes were quantitatively classified as severe slugging, oscillating flow and stable flow. By combining of two cheap and easily accessible differential pressure signals on the top of the riser as a sample, the flow regime classifier is established based on the support vector machine, and methods to improve the computational efficiency of the classifier are investigated. First, on the premise that the recognition rate is over 90%, a reasonable sample-size reduction strategy based on the K-means clustering method was designed. When the signal duration was 18.6 s, the number of samples was reduced from 7632 to 2658, and the hyperparameter iteration time of the classifier was shortened by 99.3% from 10681 s to 80 s. Second, with the combination of the single-feature recognition rate and the correlation between features, the number of features was reduced significantly. The recognition rate of the three preferred features was 96.3% with a sample duration of 18.6 s. Finally, by using the samples with a signal duration of 18.6 s as the training set, the flow regime classifier built with the three preferred features had a better generalization ability, and the average recognition rate of the testing set was higher than 90%. When the signal duration of the training set is in the range of 9.3 s–55.8 s, the maximum difference of the average recognition rate of the testing set is only 0.7%.     

Identification of flow patterns in long pipeline-riser system based on double-category labeling of samples near the transition boundaries

On-line identification of flow pattern provides an important guarantee for the flow assurance of long oil and gas pipeline. Two-phase flow patterns are experimentally investigated in a 1657 m pipeline-riser system with a wide range of air and water velocities. Samples near the transition boundaries are labeled as double-category flow patterns, and only the signals at the initial stage of severe slugging are applied to accurately identify the hazard before it occurs. The recognition rate of flow patterns with double-category labeling is significantly higher than that with single-category labeling, with the highest improvement of 8.6% for oscillating flow. The optimal signal combination is the signals DP₁ and DP₂ collected by the differential pressure sensor arranged above the water surface, with the highest overall recognition rate of 93.7%. The threshold range of double-category labeling and its corresponding minimum sample duration of 6.2 s are obtained under the premise of recognition rate higher than 90%.     

Intelligent identification of two-phase flow patterns in a long pipeline-riser system

Accurate and rapid identification of multiphase flow patterns in long-distance pipelines is an important means for flow assurance. In this paper, an experiment of air-water two-phase flow has been carried out on a pipeline-S-shaped riser system with a length of 1687 m. Based on the amplitude of riser differential pressure, the working conditions are categorised into three typical flow patterns, severe slugging flow, oscillatory flow, and stable flow. Only two difference pressure signals that are most practical near the offshore platforms are used. The data of liquid phase accumulation of severe slugging is adopted as samples, and the signal features are extracted by wavelet multiresolution analysis. The parameters of six common classifiers are optimized, and the effects of different classifiers with the optimal hyperparameters on flow pattern recognition are compared and analyzed. For neural networks, decision trees, support vector machines and k-Nearest Neighbors classifier with the optimal hyperparameters, the recognition rate of severe slugging is higher than 95.8% and the average recognition rate of the three flow patterns is higher than 94.2% when the sample duration is only 6.2 s. On the premise of achieving the highest recognition rate, the number of features is substantially reduced to 15.6% of the original number by principal component analysis.     

Identification of gas-liquid two-phase flow patterns in a horizontal pipe based on ultrasonic echoes and RBF neural network

This paper proposes a novel flow pattern identification method using ultrasonic echo signals within the pipe wall. A two-dimensional acoustic pressure numerical model is established to investigate the ultrasonic pulse transmission behavior between the wall-gas and wall-liquid interface. Experiments were also carried out at a horizontal air-water two-phase flow loop to measure the ultrasonic echo pulse signals of stratified flow, slug flow, and annular flow. It is interesting to find that the attenuation of the ultrasonic pulse at the wall-liquid interface is faster than the attenuation at the wall-gas interface. An RBF neural network is constructed for online flow pattern identification. The normalized envelop area and the area ratios of the echo spectrum are selected as the input parameters. The results show that the stratified flow, slug flow, and annular flow can be identified with an accuracy of 94.0%.     

Flow regime identification of steam-water two-phase flow using optical probes,based on local parameters in vertical tube bundles

Flow regime identification based on local parameters of axial upward two-phase flow in vertical tube bundles, at high-temperature and high-pressure, was performed using optical probes. A staggered arrangement of the tube bundles was simulated inside a non-circular test channel, the tube size and pitch are same as that in a real steam generator of a PWR under design. Optical probes were utilized to acquire the void fraction, interface frequency, and fluctuation characteristics of the local void fraction at two typical locations (centroid of the three tubes, named op-1, and centre of the minimum gap between two tubes, named op-2). The system pressure ranged from 5 to 9 MPa, mass flux from 100 to 350 kg m⁻² s⁻¹, thermodynamic steam quality from 0 to 1, and inlet fluid temperature from 263.9 to 303.3 °C, depending on the saturation pressure. This study investigated local parameters and flow pattern characteristics of high-pressure steam-water two-phase flow in vertical tube bundles using optical probes, with the measurement error of less than 2%. Results showed that local void fraction at op-1 was much larger than that at op-2, and the local void fraction difference between op-1 and op-2 increased first and then gradually decreased, which was primarily affected by the transition in flow regimes. The flow pattern characteristics of steam-water two-phase flow were described based on three aspects, namely, variation in interface frequency with local void fraction, fluctuation characteristics of local void fraction, and statistical analysis of local void fraction deviating from the average. Additionally, the flow regime identification criteria, applicable to the steam-water two-phase flow in vertical tube bundles, were proposed based on local parameters.     

Experimental investigation of gas-liquid flow in a vertical venturi installed downstream of a horizontal blind tee flow conditioner and the flow regime transition

A venturi device is commonly used as an integral part of a multiphase flowmeter (MPFM) in real-time oil-gas production monitoring. Partial flow mixing is required by installing the venturi device vertically downstream of a blind tee pipework that conditions the incoming horizontal gas-liquid flow (for an accurate determination of individual phase fraction and flow rate). To study the flow-mixing effect of the blind tee, high-speed video flow visualization of gas-liquid flows has been performed at blind tee and venturi sections by using a purpose-built transparent test rig over a wide range of superficial liquid velocities (0.3–2.4 m/s) and gas volume fractions (10–95%). There is little ‘homogenization’ effect of the blind tee on the incoming intermittent horizontal flow regimes across the tested flow conditions, with the flow remaining intermittent but becoming more axis-symmetric and predictable in the venturi measurement section. A horizontal (blind tee) to vertical (venturi) flow-pattern transition map is proposed based on gas and liquid mass fluxes (weighted by the Baker parameters). Flow patterns can be identified from the mean and variance of a fast electrical capacitance holdup measured at the venturi throat.     

基于ERT技术的垂直管道两相流流型识别

在两相流测量问题的研究中 ,流型的准确识别是其它流动参数准确测量的基础。基于电学敏感原理的电阻层析成像技术 (ERT) ,由于可视化、无辐射、低成本等优点 ,在两相流动参数的检测中具有很广阔的发展前景。研究中以 ERT系统样机为测量工具 ,以垂直管道气 /液两相管流为研究对象。通过对模拟流动状态实验下几种典型流型的测量数据的采集、处理 ,采用模式识别中的特征提取方法降低数据维数 ,并利用数理统计和神经网络方法识别流型。研究结果表明采用距离判别法和径向基神经网络技术均得到较高的识别率     

基于EEMD-Hilbert谱的气液两相流钝体绕流流型识别

为了有效识别气液两相流的流型,以水和空气为实验介质,以涡街流量计为元件诱发钝体绕流,通过管壁差压法获取气液两相流钝体绕流的尾迹波动信号,采用集总经验模态分解对信号进行分解,通过Hilbert变换得到Hilbert边际谱,利用最大互相关系数法对固有模态函数进行筛选,选取特征固有模态函数能量比分别与体积含气率、两相雷诺数构建流型图。结果表明,构建的两类新流型图对单相水、泡状流、塞状流、弹状流等典型流型的识别率分别可达91.67%和88.89%,能较好地满足工程实际应用的需求。     

基于遗传算法/神经网络组合技术的气液两相流型辨识

通过采用遗传算法训练BP神经网络、优化网络权值的技术,对气液两相流的流型进行了辨识研究,在此基础上建立了基于遗传算法/神经网络组合技术的气液两相流流型的预测模型,从而为发动机轴承腔内润滑油气液两相流流型识别提供了技术支持,也为考虑轴承腔气液两相流的相关设计和实验工作提供了技术条件。     

基于神经网络的航空发动机轴承腔气液两相流流型辨识方法研究

基于航空发动机轴承腔润滑中所存在的气液两相流问题，采用基于神经网络的理论方法建立预测模型，以便进行轴承腔内气液两相流流型的识别。研究以管道气液两相流为原型，采用3种典型的神经网络对流型进行模式识别，通过考察3种网络的辨识率，发现BP网络的识别方法具有较高的准确性。     

Identification of gas/liquid two-phase flow regime through ERT-based measurement and feature extraction

Gas/liquid two-phase flow is of great importance in various industrial processes. As the most important characteristic of a two-phase flow, the flow regime not only characterizes the flow condition in an explicit way, but also determines the measurement model in each measuring method. Based on the application of Electrical Resistance Tomography (ERT) to a gas/liquid two-phase flow on a vertical pipe, features reflecting the characteristics of gas/liquid two-phase flow are extracted directly from the measured data without reconstruction of the cross-sectional images. The statistical features are derived through time series statistical analysis. Meanwhile features in the wavelet-scale domain are derived through both one-dimensional and two-dimensional wavelet transform. All extracted features are considered as the input of a Support Vector Machine (SVM) algorithm to recognize the flow regime. The preliminary results show that the feature extraction methods of multi-feature fusion and high-dimensional wavelet transform are suitable for the identification of gas/liquid two-phase flow regimes.     

基于多尺度信息熵特征和RBF神经网络的气液两相流流型识别方法

根据小波包变换能将信号按任意时频分辨率分解到不同频段的特性，提出一种基于小波包多尺度信息熵的流型识别方法。该方法首先对采集到的压差波动信号进行4层小波包分解，在通频范围内得到分布在不同频段内的分解信号，进而建立流型的多尺度信息熵特征向量。并以此特征向量作流型样本对RBF神经网络进行训练，实现流型的智能化识别。试验结果表明，训练成功的RBF网络能很好地识别水平管内的4种流型，为流型识别开辟了一条新的途径。     

Experimental and numerical investigation on the performance of small-sized cavitating venturis

Cavitating venturis (CVs) are simple devices which can be used in different industrial applications to passively control the flow rate of fluids. In this research the operation of small-sized CVs is characterized and their capabilities in regulating the mass flow rate were experimentally and numerically investigated. The effect of upstream and downstream pressures, as well as geometrical parameters such as the throat diameter, throat length, and diffuser angle on the mass flow rate and critical pressure ratio were studied. For experimental data acquisition, three CVs with throat diameters of 0.7, 1 and 1.5 mm were manufactured and tested. The fabricated CVs were tested at different upstream and downstream pressures in order to measure their output mass flow rate and to obtain their characteristic curves. The flow inside the CVs was also simulated by computational fluid dynamics. The numerical results showed agreement with the experimental data by a maximum deviation of 5–10% and confirmed that the numerical approach can be used to predict the critical pressure ratio and mass flow rate at cavitaing condition. It is found that despite the small size of venturis, they are capable of controlling the mass flow rate and exhibit the normal characteristics. By decreasing the throat diameter, their cavitating mode became more limited. Results also show that increasing the diffuser angle and throat length leads to a decrease in critical pressure ratio.     

Three-way PLS regression and dual energy gamma densitometry for prediction of total volume fractions and enhanced flow regime identification in multiphase flow

Dual energy gamma densitometry and 3-way partial least squares regression were applied to quantify the total volume fractions and improve flow regime identification in multiphase flow. Multiphase flow experiments were carried out with formation water, crude oil and gas from different North Sea gas fields in Statoil׳s High Pressure Multiphase Flow Loop in Porsgrunn, Norway. Four different flow regimes were investigated (stratified wavy, slug, dispersed and annular). A traversable dual energy gamma densitometer was used to measure the fluid densities in the pipe. Partial least squares regression was previously applied to identify multiphase flow regimes and quantify volume fractions of gas, oil and water. That study showed promising results for flow regime identification but the predictions of the total volume fractions were not acceptable. In this study a new method combining gamma densitometry and 3-way partial least squares regression was applied in order to improve the quantitative estimation of the total volume fractions gained in the previous study. The proposed 3-way regression approach allows prediction of the total volume fractions directly using one model instead of multiple models which was reported earlier. The improved quantification of the volume fractions of gas, oil and water was used to improve the flow regime identification plots and increase the interpretability.The new 3-way prediction results for the volume fractions were significantly better than what was found earlier based on individual PLS models. The root mean square error of prediction for gas, oil and water from the 3-way PLS models were 4.1 %, 4.3 % and 4.6% respectively. All models reported were validated based on independent data (test set validation).     

基于模糊信息融合的气固流化床流型及其转换的识别研究

本文在模糊集理论的基础上介绍了多传感器、多参数识别气固流化床流型的信息融合模型。将压力脉动信号的算法复杂性Cn、涨落复杂性Cf和香农熵En作为融合的特征参数，进行特征层的多参数融合；根据特征参数建立了过渡流型的隶属度函数；对多个传感器的特征层识别结果进行决策层融合，得到了多传感器对不同流化状态的最终识别结果。实验结果表明，采用香农熵特征参数能较好地解决鼓泡与湍动2种流化状态转换的识别；应用多传感器、多参数数据融合对流态化不同流型及其转换的识别能得到较好的效果。     

故障事件严酷度类别的模糊模式识别法

运用模式识别的主要思路,即针对一个事物进行特征提取并采取最大隶属原则归类。并基于模糊数学理论,建立了严酷度的隶属函数,从而使不易归类的故障事件的严酷度得到相对精确的归类。这里探索了用模糊模式识别法识别故障事件严酷度类别的方法,为今后产品可靠性分析过程中的故障模式影响分析的自动化、机械化、程序化提供一定的理论参考。     

基于监控视频的异常事件识别

提出了一种基于监控视频的异常事件识别模型,该模型可以实时监测视频中的前景目标,并通过分析目标的运动信息判断是否有异常事件的发生。首先,采用背景建模的混合高斯算法提取前景目标;然后,用金字塔迭代的L-K特征点跟踪算法得到前景的光流运动信息,并通过分析前景的面积比例、速度方差、整体熵判断视频中是否有异常事件的发生;最后,利用爆炸、人群短时聚集和分散两种异常事件做仿真实验。结果表明,该模型可以准确提取前景目标区域,并可以快速、精准地判断监控视频中的异常事件,可以为管理部门及时发现和控制异常事件提供有效的帮助。     

静电分离器中电场和电流体特性试验研究

利用静电分离技术从空气中分离平均直径为2μm的油滴,对小尺度的线—板式静电分离器的电晕放电特性和流型特征进行试验研究,放电电极接高压电源正极。测试在不同的气体流速、油滴浓度条件下分离器的伏安特性曲线。利用高速摄像机对其中的电流体流型特征进行可视化研究,试验测定进口流速为0.2 m/s、0.3 m/s、0.4 m/s,施加电压从0～16 kV条件下电流体流型的变化,得到在不同的施加电压和气体流速下的流型分布图。试验表明,进口流速对分离器的电场影响并不明显,而油滴浓度对电场有很大的影响,在相同的施加电压下电流密度随着油滴浓度的增大而降低。电流体流型的变化取决于进口流速和施加电压的相互作用。得到在不同的描述惯性力和电场力关系的量纲一参数雷诺数和电流体数下5种代表的电流体流型。     

基于网络的人脸识别系统的设计

随着人们对社会安全的重视,以及网络广泛应用,对网络视频进行了研究,设计了一个基于以太网的人脸识别系统。系统基于ARM＋Linux实现了一个嵌入式的网络摄像头设备,该设备通过网络传输视频至服务器,服务器接收视频帧,调用OpenCV图像处理库对获取到的视频帧进行处理,从而实现了远程的人脸检测、识别功能。     

Identification of two-phase flow pattern in porous media based on signal feature extraction

The statistical analysis methods based on differential pressure signals of two-phase flow are employed in the present study to identify the flow patterns in packed porous bed. The typical flow pattern images of two-phase flow in the packed porous beds are recognized and the corresponding differential pressure signals are recorded based on the visualization experiments. Then the statistical analysis methods, including probability density function (PDF), power spectral density (PSD), and wavelet energy spectrum (WES), are employed to extract the features of differential pressure signals in the time domain, frequency domain, and time-frequency domain respectively. The dimensionless parameters are proposed as the evaluation index to quantify the differences among flow patterns. The results show that the PDF, PSD, and WES methods can effectively characterize different flow patterns in the time, frequency, and time-frequency domain, respectively. The comprehensive recognition efficiency is about 88.5% using the introduced dimensionless parameters.