【作者】 许灿辉；

【导师】 阳春华；

【作者基本信息】 中南大学 ， 计算机应用技术， 2011， 博士

【摘要】 矿物浮选是涉及固体、液体、气体三相分选的复杂物理化学过程。影响分选效率的过程操作变量众多,且变量间相互耦合作用对浮选性能指标产生决定性影响,既有来自宏观层次的固体粒子传输等物理因素,同时也有来自微观层次的化学药剂因素和流体动力学因素等。浮选过程工序中可控操作变量直接影响泡沫层形成的质量,主要体现于气泡流动速度和气泡尺寸特征。气泡表观特征既能提供可控操作变量的指示符,同时亦可作为判断浮选分选效果的重要依据。浮选生产过程优化的瓶颈在于解决对泡沫表观特征如气泡速度和尺寸的准确描述、视觉特征量变化与操作参数间映射关系以及确定有效操作提高浮选指标性能等关键问题,这成为本文研究和探讨的内容和难点。在泡沫浮选机理分析基础上,依据气泡表观特征与工艺操作条件的关联性,提出了表层气泡流动速度和尺寸分布表观特征提取方法,开发了气泡表观特征在线提取系统,并成功应用于浮选工业现场泡沫监测和数据分析。论文主要研究工作及创新性成果如下：(1)针对浮选气泡体内部不透明、可视程度低等本质属性,对泡沫体内部气泡运动的轨迹进行动力学描述,建立了泡沫体内部气泡运动的拉普拉斯仿真模型。在讨论浮选泡沫体做不可压缩和无旋运动的基础上,给定流体流动区域边界上的运动参数即边界条件,建立的气泡运动拉普拉斯模型能反映泡沫体内部的动态特征,流场内流线趋势揭示相邻气泡的运动模式,气泡形状随着流线的密集而产生形变,给出泡沫体速度场分布的可视化输出。通过分析气泡运动模型的流场分布,确定表层气泡运动速度测量位置。(2)针对浮选表层气泡运动的连续性及工业泡沫图像的复杂性特点和速度特征提取实时性的需要,提出基于方向的快速六边形搜索宏块跟踪算法计算工业浮选表层泡沫流动速度位移。由于浮选泡沫连续帧图像在短时间内保持其形状结构的缓慢变化,适用于二维单气泡匹配的传统质心跟踪速度测量算法无法有效应用于跟踪环境错综复杂的三相工业泡沫图像的气泡质心。提出的速度算法适用于工业浮选泡沫速度特征的实时提取,对同水平位置的不同宏块进行跟踪匹配的实验结果表明该算法的准确性。在确定表层速度测量点和溢流口泡沫厚度测量点的基础上,在方形浮选实验槽进行大量实验,采用提出的速度特征计算空气回收率,发现了空气回收率的最优尖峰值。根据空气回收率和泡沫回收率的强依赖正比关系可知,将操作条件调整至空气回收率的最优尖峰值,可使浮选槽获得最优泡沫回收率。(3)针对采集的泡沫图像具有“无背景、全前景”的特点和传统分割算法的局限性,研究了适于矿物泡沫图像分割的改进形态学开闭重构分水岭分割算法和基于谷底边缘检测的浮选泡沫图像分割算法,克服白亮点现象带来的分割不稳定性。对泡沫图像进行分割后处理分析发现,泡沫尺寸概率密度为非高斯分布。针对现存气泡形状特征提取方法假设泡沫尺寸为正态分布,采用均值方差等不精准的单值特征,本文提出概率密度函数估计算法来准确描述气泡尺寸分布统计特性。但图像特征的统计分布数学模型是不可知的,也无法预先假设图像特征服从某种分布,采用非参数估计方法可描述不可知连续过程的气泡尺寸密度分布。根据泡沫尺寸分布曲线具有偏斜、长拖尾等特点,设计了非参数估计算子反映各种复杂关系。(4)针对泡沫尺寸分布特征与浮选药剂操作的关系,建立了基于非参数动态权系数的故障检测和诊断模型。对分割泡沫图像采用概率密度函数准确描述泡沫尺寸统计特性,设计非参数估计算子逼近概率分布曲线,从而将泡沫尺寸概率分布转化为动态权系数描述,建立了带时滞动态权系数模型。运用线性矩阵不等式得到合理的故障检测和诊断判据。将该模型应用于离线泡沫图像故障检测的仿真实验,验证了该模型的有效性。(5)以实际浮选过程为研究对象,开发了浮选泡沫图像特征分析系统,建立了基于对应气泡速度和尺寸表观特征的操作条件的品位回收率关系模型。通过实时观察操作变量在泡沫表观特征上的变化,调整浮选可控操作参数药剂用量和浮选槽充气量,在回收率品位曲线带上实现性能指标的优化,获得不同调整剂加入量和气泡运动速度条件下的浮选性能指标优化的结果,为实现浮选过程生产操作的实时优化奠定了基础。更多还原

【Abstract】 Mineral flotation is a complex physical and chemical process involving three phases such as solid, liquid and air. It is a multivariate process influenced by many factors including physical variables like solid particles in macro level and chemical variables like reagent or dynamic fluid kinetic factors in micro level, and additional coupling effect among the variables can complicate the decisive reasons for flotation performance change. From the perspective of flotation mechanism, froth visual features can characterize the combining effect of multiple operating conditions on flotation, which is the indicator of flotation separation performance. The bottleneck of flotation optimization lies in guaranteeing accurate flotation froth features description, relation between image features and operational variables, flotation performances optimization based operation condition strategies. All these unsolved problems will be discussed and explored in this thesis.Based on froth flotation mechanism analysis and the relationship between froth visual features and operation conditions, surface bubble motion velocity measurement and bubble size distribution estimation algorithms are proposed. With the online feature data and operation conditions from the developed froth image feature monitoring system, relation on froth visual features, operational variables and flotation performances is established and applied on froth flotation industry. Main research work and innovative achievements are as follows:(1) To describe foam flow within a flotation cell which has inherent properties like opacity and invisibility, a Laplace base bubble motion model is established from kinetic analysis. With the assumption of incompressible and irrotational flow, boundary conditions are given for the established model. It can provide visual output of the bubble velocity distribution which is one of the important dynamic features in froth layer, and the trend of streamline within reveals flow pattern of adjacent bubbles which proves the bubble shape distorts with the density of streamline. The Laplace model based bubble motion velocity distribution can theoretically point out rational position in a flotation cell to measure the flow velocity of bubble in surface layer.(2) Considering the characteristics of complex surface bubble flow and its demand of real-time velocity feature extraction in flotation industry, a direction-oriented fast hexagon search based block matching velocity measurement algorithm is proposed. Since the traditional centroid tracking based algorithm fitting for 2D foam velocity measurement is no longer applicable in complex 3D industry froth images, the proposed algorithm suits the real-time extraction of velocity measurement with the approval of its accuracy from experimental results of two target blocks at same horizontal position but different height. Given the right position to measure the velocity and the froth depth over the lip, various experiments carried out shown that air recovery calculated from velocity feature appeared a peak with the increase of inlet gas velocity. When operating in condition of peak air recovery, the opportunity of optimal flotation performances increases consequently.(3) For froth images has an unpleasant property of no void background for conventional segmenting algorithms, two effective froth segmentation algorithms are explored based on the features of froth images to overcome the limitation of traditional methods such as segmenting instability of white spot effect, including improved morphological reconstruction watershed segmentation method and valley edge detection based segmentation method. The post-segmentation analysis has shown the fact that probability density distribution (PDF) of bubble size is non-Gaussian. Unlike traditional method applying singular feature such as mean or variance with the assumption that the distribution is normal, probability density distribution is suggested to accurately describe statistical feature of froth structure. The fact that the mathematical model of distribution is unknown makes nonparametric estimation method fitting to depict the unknown continuous process of froth flotation. Considering that the bubble size distribution is highly skewed with long tails, nonparametric descriptor is designed to reflect the variety of the curve.(4) In order to establish the relationship between bubble size distribution and reagents addition, a fault detection and diagnosis scheme using nonparametric estimation based dynamic weights model is proposed. The designed nonparametric estimation is applied so that the output PDF is formulated in terms of dynamic weights, by which a non-linear model with time delay is established. Effective fault detection and diagnosis is achieved by using linear matrix inequation method. The reagent fault is successfully detected and diagnosed on the industrial data of off-line froth images.(5) Taking certain alumina industry froth flotation as a research case, a froth visual feature extraction system is developed. Operational conditions with corresponding bubble velocity and size features based grade recovery model is proposed. By manipulating the cell operating condition such as chemical reagent addition or inlet air flowrate, the on-line monitoring of froth visual features can reflect the changes, and consequently the circuit performance can shift among the coordinates on the grade recovery curve to achieve optimal trade-off between grade and mineral recovery. The metallurgical results clearly indicate that changes in air rate or reagent addition result in the variation of flotation performance, which laid the foundation for implementation of real-time process optimization.更多还原