矿物浮选过程泡沫图像多敏感特征分布控制

The characteristic of bubble image is the key criterion for dosage addition in the mineral floatation process. But the characteristics are multiple and correlated. The sensitivity of different characteristics to the dosage addition is also various with non-Gasussian stochastic distribution and uncertainty. Consequently the distribution control of the bubble image characteristics is difficult. The manual operation of chemical dosage addition based on visual observation for bubbles is employed in practice at present, which badly influences the production quality, quantity and mineral recovery with resource waste and dosage over consumption. In the view of sensitive characteristic of bubbles, the project will investigate the abstraction method of sensitive characteristic set of bubble image based on analysis of sensitivity to dosage variation and the effective representation of correlated stochastic distribution, design the acquisition method of principal component weight vectors projecting from tensors to vectors based on local linear imbedding, explore the coupled modeling method based on multiple sensitive characteristic distribution of bubble image, build the model of the relationship between chemical dosage addition and principal component weight vectors and the data-driven predictive model of principal component weight vectors, propose dosage addition control strategy with minimum information entropy of joint deviation of principal component weight vectors, so as to track the expected multiple characteristic distribution of bubble image. A systematic distribution control method based on multiple sensitive characteristic distribution of bubble image will be constructed, which will build the foundation for the stabilization of the floatation production and significantly improve the production quality and resource recovery.

矿物浮选泡沫图像特征是浮选过程药剂添加量调控的重要依据，但图像特征多且相互关联，各特征对药剂添加量的敏感程度不一，并具有非高斯随机分布特性和不确定性，泡沫图像特征分布控制困难。现有人工观察泡沫状态手动调节的方法严重影响精矿质量和矿物回收率，造成资源浪费严重、药剂消耗大。为此，本项目从泡沫图像敏感特征的视角，研究基于药剂变化量敏感度分析的泡沫图像敏感特征集提取及其关联随机分布的有效表征方法，构建基于局部线性嵌入的张量到向量投影的主元权向量获取方法，提出基于泡沫图像多敏感特征分布信息的浮选过程耦合建模方法，建立药剂添加量与主元权向量间关系模型、数据驱动的主元权向量变化预测模型，提出基于主元权向量联合偏差信息熵最小的加药量调整策略，实现泡沫图像多敏感特征随机分布的跟踪控制，形成泡沫图像多敏感特征分布控制理论与方法，为矿物浮选过程稳定生产提供理论基础，对提高产品质量和资源回收率具有重要意义。

矿物浮选泡沫图像特征是浮选过程药剂添加量调控的重要依据，人工观察泡沫状态手动调节的方法严重影响精矿质量和矿物回收率，造成资源浪费严重、药剂消耗大，亟需突破浮选过程泡沫图像特征分布控制难题。本项目从泡沫图像敏感特征的视角，提出了基于药剂变化量敏感性指数和相关性分析的泡沫图像敏感特征提取方法，确定了表征浮选过程工况的泡沫图像敏感特征集；给出了关联泡沫图像敏感特征随机分布的表征方法，提出了基于气泡尺寸与形状的联合特征构建方法，采用双三次B样条表征联合特征分布曲面；提出了基于泡沫图像多敏感特征分布信息的浮选过程耦合建模方法，分别建立了基于KPCA与FCM_PSVAR的入矿品位预估、最优泡沫图像特征优化设定、主元权向量变化预测等模型，研究了自适应模糊神经网络药剂量控制策略，实现泡沫图像多敏感特征随机分布的跟踪控制，研发了基于分布机器视觉的浮选过程监控系统，应用于湖南新龙矿业有限股份责任公司的金锑矿浮选，使金锑精矿品位完全达标率提高20%，成效明显。.形成泡沫图像多敏感特征分布控制理论与方法，为矿物浮选过程稳定生产提供理论基础。获中国有色金属工业科学技术一等奖1项，发表学术论文17篇，其中SCI收录4篇、EI收录9篇，合作出版专著1部，授权国家发明专利5项，申请国家发明专利1项，软件著作权3项；课题负责人获国家自然科学基金委杰出青年基金资助，培养副教授1名，博士3名、硕士8名。