【作者】 王红一；

【导师】 董峰；

【作者基本信息】 天津大学 ， 检测技术与自动化装置， 2011， 博士

【摘要】 气液两相流场的测量是目前多相流体力学研究中的一个难点。其中,水中气泡的运动机理和规律是气液两相流中的基本问题。研究气泡的特征参数及运动特性,对于指导两相流实验系统的设计、有效控制各种化学反应过程具有重要意义,并有助于进一步深入理解两相流的内部流动机理。本课题运用数字图像处理与分析技术,研究了对水中上升气泡的多种运动特性,主要研究工作包括以下几个方面:(1)为了获得不同生成条件下水中气泡的2D平面和3D空间的图像序列,建立了一套可视化两相流场图像采集实验系统;分别对图像的去噪、二值化、填充、边缘检测等预处理算法进行了讨论,并提出了一种基于图像背景像素点传染的气泡填充算法,实现了对边缘不闭合气泡的良好填充,进而实现了气泡特征参数的顺利提取;并通过对实验图像进行标定处理,计算了图像的缩放系数。在此基础上,对气泡的运动特性进行了研究和讨论。(2)利用图像技术对初始生成气泡的体积进行了计算,通过分析实验测量数据,提出了改进的气泡生成体积预测模型,解决了气体体积流量由小到大过渡时期的气泡体积难预测的问题,并提高了预测精度;通过对稳定上升时的单气泡进行受力分析,建立受力平衡方程,并结合数字图像测量到的气泡速度、加速度等参数,推导出了稳态上升气泡的体积计算公式;对于存在较弱相互作用的上升气泡,则介绍了一种基于Hough变换的3D气泡椭球体重构的方法,从而可以估测气泡体积。(3)提出了一种将贴标签法、互相关算法和小波多尺度分析法相结合的轨迹跟踪方法,对多个气泡同时上升的情况实现了良好的跟踪,在提高运算速度的同时保证了跟踪精度;观察了气泡轨迹的由线性上升到“之”字形上升,再到螺旋上升的转变过程,并讨论了气泡脱离频率对稳态上升时刻气泡轨迹的影响;通过观察气泡初始生成时期的上升速度曲线的震荡特性,将其划分为4个区域:气泡生长区、速度增长区、速度下降区和气泡稳态上升区,且同一工况下生成的气泡的初始速度在前三个区域具有较好的重复性;并进一步讨论气泡大小、气泡纵横比和气泡脱离频率对气泡末速度的影响,表明气泡的末速度与气泡的生成条件和气泡上升时的变形程度密切相关。(4)通过观察不同气体流量下生成气泡的高速图像,以气泡生成的周期性和气泡间相互作用的表现形式为依据,将气泡的生成流型分为5类:单气泡上升、气泡间存在相互作用但不发生聚并、气泡间存在聚并与破碎、间歇性喷射流、喷射流,对各种流型的运动特征进行了详细的描述;提取了基于GLCM和GLGCM的多种图像纹理特征,通过特征选择寻找最优的特征组合方式,并利用SVM多分类器对气泡的生成流型进行识别,实验结果表明StARMiner算法选择的纹理特征组合效果最佳,用于气泡生成流型时的识别率可达96%。更多还原

【Abstract】 The measurement of gas-liquid two-phase flow is one of the difficult issues in hydromechanics research field. The dynamic motion and mechanism of bubble rising in water is a basic subject for gas-liquid two-phase flow. The research of bubble parameter and behavior is significant for experimental system design and chemical reaction control and also helpful for the further understand of the gas-liquid two-phase flow mechanism. In the present work, the dynamic characteristics of bubble rising in water were studied based on digital image processing technique. The main conclusions can be draw as follows:(1) In order to get the image sequences with 2D plane and 3D space of bubbles generated at different situation, a suit of visible gas-liquid two-phase flow filed with video systems is established. Digital image processing methods of removing noises, binaryzation, filling bubbles and detecting edges were studied. Especially, the bubble filled method was improved to fill the bubbles with edge opening. And several bubble parameters can be extracted from the images. The scale coefficient, which is the ratio of the real distance and pixel distance, was acquired by images calibration method. Then, several bubble characters were studied based on the digital image processing method.(2)The formed volume of bubble were calculated by digital image processing. An improved model for the formed volume of bubble, with high accuracy and wide adaptive flow rate, was proposed based on the experimental data analysis. For the steady rising bubble, a volume computing formula was derived based on the force balance equation. The parameters (such as rising velocity and acceleration) used in the formula were measured from the bubble image by digital image processing method. And a 3D ellipsoid model was established based on Hough transition, which can be used to calculate the volume of bubbles with weak interaction.(3) In order to get the tracks of multi-bubble rising in the tank, an improved method based on the combination of labeling method, cross correlation method and wavelet multi-scale transform is proposed. This method not only saves the computing time, but also ensures the tracking accuracy. The process of bubble rising begin from linear to zigzag, and then to helical was investigated. And, the correlation between the detachment frequency and the bubble track was analyzed. The rising velocity of the formed bubble was divided into four regions based on the oscillating characteristics: bubble growing region, velocity increasing region, velocity decreasing region and the steady rising region. The first three regions have good repeatability when the bubbles generated at the same working situation. By discussing the impacts of the bubble size, aspect ratio and detachment frequency on the terminal velocity, we found that the terminal velocity is close related to the generating conditions and the distorting extent.(4) According to the periodicity of bubble formation and the appearance of bubble interaction at different gas flow rate, the bubble formed phenomena was classified into 5 regimes: single bubbling, bubbling in groups without coalescence, bubbling with coalescence and break up, chaining and jetting. And the main characteristics of each regime were detailed based on the image sequences. Several textural features based on GLCM and GLGCM were extracted and the best feature combination approach was selected by feature selection method. Then, bubble formed regimes were identified based on SVM multi-classifier. The Experimental results have shown that the feature combination selected by the StARMiner algorithm is most efficient for bubble formed regime recognition, the recognizing rate of which can reach up to 96%.更多还原