基于分布机器视觉的铝土矿选矿过程协调优化方法研究

Mineral processing-Bayer process for alumina production is a new technology developted only in our country to process high silica bauxite,which provides a new way to the sustainable development of aluminum industry in our country. But the mineral processing process is long . There are strong couplings between the flotation banks and between the process variables with uncertainties, which cause difficulty in modeling and optimal control of the process, so that the process is mainly controlled by operators experiencely according to the surface features of the froths.These cause fluctuation of the whole process and low recovery of valuable minerals. Therefore, multiple machine visions are fixed up distributedly in the process to collect distributed froth images in this project. Extraction method of the sensitive froth features-distribution from the distributed machine visions, and correlation model for the cascade banks based on the feature-reconstruction of distributed froth image series are studied to disclose the internal relationship between the banks. Prediction model for process indices based on the fusion of variation trends of the sensitive froth features-distribution is proposed. Based on these models, complex ore feed-oriented master-slave decomposition & coordination of the final object of the whole flotation process, and master-slave coordinative optimal setting of the multiple operating variables in the whole process is studied. The effectiveness and feasibility of the methods will be tested and validated by application of them to the industrial process. So that, systematic coordinative optimization theory and method is finally formed for optimal control of mineral processing process. The results are important to improve the quality of the concentrate and recovery of valuable minerals and to reduce the resources consumption in mineral processing, and they will also lay a theoretical foundation for high-efficient mineral processing, which is of great significance for both industrial application and scientific research.

我国自主创新的选矿-拜耳法氧化铝生产新工艺为我国铝工业的可持续发展开辟了新途径，但铝土矿选矿过程流程长、过程变量及作业间耦合关系复杂且存在不确定性使过程建模和优化控制困难，主要通过人工判断泡沫状态凭经验控制，工况波动大、资源利用率低。为此，本项目在铝土矿选矿流程中多点布置机器视觉，研究机器视觉敏感特征分布提取方法和基于分布机器视觉图像序列敏感特征重构的级联作业关联建模，揭示级联作业间的内在关联关系；建立基于机器视觉敏感特征分布变化趋势融合的过程指标预测模型；基于过程模型，研究面向复杂矿源的浮选过程总目标主从分解-协调方法和全流程多操作变量主从协调优化设定方法；将所提方法应用于铝土矿选矿过程验证其有效性，形成选矿过程主从协调优化理论和方法。本项目对提高铝土矿选矿产品质量和节能降耗具有重要意义，并为高效选矿奠定理论和方法基础，具有重要的工业应用价值和科学价值。

铝土矿浮选过程是我国自主开发的矿物加工工艺技术，对我国低品位铝土矿资源的有效利用和节约能耗具有重要意义。但因为浮选过程流程长、影响关系复杂且不确定性强，造成建模和优化控制困难，使实际生产过程中采用人工看泡沫状态进行操作，工作强度大且难以优化运行。本项目针对此问题研究基于分布机器视觉的铝土矿浮选过程协调优化问题。首先，针对物料流无法标记跟踪造成同一组物料参数不匹配的问题，提出一种基于时效关联分析的分布参数时间配准方法，实现分布机器视觉数据的时空配准，为建模和优化提供了数据基础；提出基于概率密度分布（PDF）的图像序列泡沫大小特征重构方法和基于趋势分割点的泡沫敏感特征序列变化趋势提取方法，建立了基于重构敏感特征趋势的精选与粗选泡沫表面特征关联模型，反映了不同浮选作业间的关联关系；针对指标影响因素过多且耦合严重，提出了基于敏感特征变化趋势和入矿参数等多信息融合的浮选精矿品位集成预测方法，和基于现象学与统计学习的入选矿浆粒度预测混合建模方法，充分利用了不同类型信息的特点分别建模再集成，实现了指标和关键参数的准确预测；在所建预测模型的基础上，提出了基于泡沫尺寸分布PDF最优的粗选加药量优化控制策略和基于泡沫敏感特征的pH值预测控制策略；综合利用分布机器视觉重构的敏感特征和入矿参数等信息，提出了基于置信规则和指标分解协调的粗选扫选和精选加药量主从协调优化方法，对不同作业加药量和不同药剂种类进行协调。工业数据仿真和工业试验验证了上述方法的有效性，并有部分方法在基于分布机器视觉的浮选过程监控系统中得到了集成应用，形成了较系统的选矿过程建模和协调优化理论和方法。所提方法也为其他矿物浮选过程的精细化优化控制提供了理论和方法依据。