基于磨音/振动信号和给料粒度图像特征的半自磨流程运行优化方法

In semi-autogenous mill, by using both big ore-rocks and balls as grinding media, the ore is ground through attrition and impact between the media and the ore caused by tumbling of the mill shell. The semi-autogenous grinding process combines the functions of both crushing and grinding. It has the advantages of large-scale, higher capacity, higher running efficiency and breakage efficiency, but lower energy consumption, so that it is a trend to use semi-autogenous process to replace the process of secondary crushing, fine crushing and primary grinding. However, semi-autogenous process is easy to be influenced by the variation of the ore properties and size. Meanwhile, the internal status of the mill cannot be measured. And particularly, the ore properties fluctuate frequently in our country. These make it especially difficult to control the process optimally. Therefore, the following issues will be studied in this project: deep extraction method for the features of the acoustic and vibration signals of the semi-autogneous mill and separation of different features, based on the fusion of the online measured parameters (acoustic and vibration signals, the motor current etc.) and the knowledge of mechanistic and experiences, Naive Bayesian network-based model for collision type between the balls and the materials and for mill internal parameters online monitoring will be built; online monitoring of the feed size distribution and ore type classification based on statistics of the 3D features of feed image series; ore-type oriented and parameters correlation degree-based multi-objective optimization method for the semi-autogneous grinding process using multi-objective decomposition-coordinate strategy; industrial test to these methods, and therefore to form a set of monitoring and optimization methods for semi-autogneous grinding process. The results will lay a theoretical and technical foundation for the optimal control of the mineral processing process, and are important to save energy and improve production efficiency for green manufacturing, which is of great significance for both industrial application and scientific research.

半自磨机以大块儿矿石和钢球做介质，通过磨机筒体带动介质和矿石摩擦、碰撞实现矿石磨碎，兼具破碎和磨碎双重功能，易大型化，处理量、运转率和破碎效率高，而能耗更低，用其代替中、细碎和第一段磨矿是发展趋势。但它受给料性质的影响严重、且磨机内部状态无法检测，特别是我国矿石性质波动频繁，优化运行困难。因此，本项目研究半自磨机磨音/振动信号特征深度提取方法，融合磨音/振动信号特征、磨机电流等实时信息和机理/经验知识，基于朴素贝叶斯网络，建立磨机内部球与物料的碰撞类型识别模型和磨机内部关键参数在线监测模型；研究基于矿石图像序列三维特征统计分析的给料粒度分布在线监测和矿石类型识别；针对不同类型矿石研究基于变量相关度的半自磨流程多目标分解协调优化方法; 并进行工程验证，形成一套半自磨流程监测与优化运行方法。项目成果为矿物加工过程节能和提效提供理论和技术基础，对其绿色生产意义重大，具有重要的科学价值和实用价值。

半自磨流程兼具破碎和磨碎双重功能，使其应用范围越来越广。矿石性质不断下降、性质频繁波动，半自磨设备本身超重型和黑箱、机理复杂，矿石性质和磨机内部参数以及粒度指标等关键参数检测困难，这些问题使得半自磨流程的优化控制极其困难。.本项目首先提出基于趋势信息的自适应多变量过程稳态检测方法，自定义过程稳态指标，有效检测过程稳态和非稳态；针对稳态生产过程，提出基于参数时间配准的多层数据协调方法，实现不同层次数据的协调，为过程建模和优化提供准确数据。提出了基于AR谱减法降噪和多信息趋势融合的磨机负荷监测方法，能有效区分磨机内部负荷的5种状态（空磨、偏空磨、正常、偏饱磨、饱磨），为更精细化的控制提供依据；提出自适应分水岭分割入磨矿石粒度分布检测方法和基于Gan-Unet图像深度学习的矿石图像分割方法，可自动识别矿石是否含泥并采用不同的参数进行分割，能有效检测矿石粒度分布的变化。.基于矿石粒度分布等特征和矿石成分建立了矿石类型的LSSVM识别模型，设计了面向不同矿石类型的基于目标分解协调的磨矿分级过程多目标优化方案，基于不同方法实现不同的优化目标，包括：面向能耗和钢耗优化，提出了基于DEM仿真和响应曲面建模的磨机转速优化和控制方法；提出了基于自抗扰控制的半自磨过程多目标优化控制，实现能耗、钢耗最小化和处理量最大化；融合专家、机理和数据知识，提出了基于智能协调和模糊逻辑的磨矿分级过程智能优化控制方法，在保证指标的同时，优化能耗和处理量；针对工况变化问题提出基于聚类分析的扩展置信规则知识自动提取方法等。研发了磨机负荷检测和入磨矿石粒度检测嵌入式系统，工程测试表明能为生产提供有效的检测信息；开发了磨矿分级过程智能优化控制仿真系统，采用大量实际生产数据仿真验证了方法的有效性，表明可有效降低能耗、提高处理量并保证指标满足要求。为半自磨机等磨矿分级过程智能化检测和运行优化制提供理论和方法基础。