基于泡沫特征的浮选预测研究进展

Research Progress on Flotation Prediction Based on Froth Characteristics

  • 摘要: 浮选是根据矿物颗粒表面物理化学性质差异,从矿石中分离有用矿物的技术方法,其工艺指标的实时检测是提高资源利用率的有效前提。而传统的离线化验方法由于采样、送样及分析耗时长,表现出严重的滞后性,难以实现对矿物加工过程的实时调控。近些年来,随着机器视觉与人工智能技术不断发展,浮选泡沫图像因其蕴含着丰富的过程状态信息,已成为浮选工艺指标软测量研究中重要的数据来源。借助建模对浮选泡沫图像进行特征提取并量化,能够为浮选过程的动态实时监测、指标预测提供可靠的支持,从而推动矿物加工领域在资源有效利用、效率提高、智能化发展等方面取得进步。文章梳理了从数据采集与处理到特征提取,再到预测模型建立与优化的全流程技术研究现状:在前期数据准备方面,总结了浮选泡沫图像采集装置、采样方法、图像预处理方法,在特征提取环节,回顾了泡沫静态特征与动态特征提取的典型技术,如气泡尺寸、形态、纹理、泡沫运动行为等多个维度,基于此,又进一步综述了深度学习算法在灰分、品位等指标预测方面的应用,同时,归纳了遗传算法、粒子群优化算法等不同智能优化算法对模型调优中所起到的关键作用,期望能够为浮选技术的智能化转型提供一定参考。

     

    Abstract: Flotation is a technical method for separating valuable minerals from ores based on the differences in physical and chemical properties of mineral particles. The real−time detection of its process indicators is an effective prerequisite for improving resource utilization. However, due to the long sampling period, long sample delivery and analysis time, the traditional off−line test method shows a serious lag, and it is difficult to achieve real−time control of mineral processing. In recent years, with the continuous development of machine vision and artificial intelligence technology, flotation froth image has become an important data source in the soft measurement research of flotation process indicators because of its rich process state information. With the help of modeling, the feature extraction and quantification of flotation froth images can provide reliable support for dynamic real−time monitoring and indicators prediction of flotation process, so as to promote the effective utilization of resources, efficiency improvement and intelligent development in the field of mineral processing. This paper reviews the research status of the whole process technology from data acquisition and processing to feature extraction, and then to prediction model establishment and optimization. In the aspect of preliminary data preparation, the flotation froth image acquisition device, sampling method and image preprocessing method are summarized. In the feature extraction process, the typical techniques of froth static feature and dynamic feature extraction are reviewed, such as bubble size, shape, texture, froth motion behavior and other dimensions. Based on this, the application of deep learning algorithm in the prediction of ash content, grade and other indicators is further reviewed. The key roles of different intelligent optimization algorithms such as genetic algorithm and particle swarm optimization algorithm in model optimization are summarized. It is expected to provide some reference for the intelligent transformation of flotation technology.

     

/

返回文章
返回