基于多尺度频谱与模糊规则的湿式球磨机负荷集成模型及在线更新

Ball mills are bottleneck mechanical devices for achieving the global optimization control of the mineral grinding process. Therefore, mill load is one of the most important factors for improving production quality and efficiency, saving energy consumption of grinding process. In industry processes, mill load is always measured indirectly based on shell vibration and acoustical signals, and current signal of the ball mill motor or estimated by domain experts using their rich practical experiences. However, the shell vibration and acoustical signals have characteristics of non-linearity, nonstationary and multi-component. Moreover, there is complementary and redundant among these multi-source signals. And mill load estimation results based on the knowledge experts are not stable due to their limit vigor. Another problems is that the grinding process has the characteristic of time variant. Aim at these problems, this project will make a research on multi-scale frequency spectra and fuzzy rule based wet ball mill load ensemble model and its online updating algorithm. At first, numerical simulation model of the ball mill grinding mechanism and shell vibration mechanism model are built, which integrated with multi-component signal adaptive decomposition algorithm, such as empirical mode decomposition to analyze shell vibration. The results can support theoretical basis for multi-scale frequency spectral features selection and mechanical knowledge ruler extraction. Then, ensemble soft measuring model based on multi-scale frequency spectra optimization selection and experts mechanical fuzzy rule is constructed. Thus, the data-driven and the fuzzy inference model are fused complementally. Base on above results, new samples that can represent concept drift of the grinding process are identified intelligently to on-line construct multi-layer ensemble model. Then, stability of the on-line updating algorithm are analyzed. Finally, we will carry out the experimental validation and industrial application research. The successful application of this project makes a foundation for real-time online measuring mill load, optimal control and energy saving of grinding process.

球磨机是磨矿过程实现优化控制的瓶颈设备。磨机负荷对产品质量、生产效率及节能降耗至关重要。工业界常基于磨机振动/振声和电流等信号间接测量或领域专家凭经验估计。针对振动/振声信号的非线性、非平稳和多组分特性，多源信号的互补冗余性，专家知识估计负荷的不稳定性，以及磨矿过程的时变特性，拟开展基于多尺度频谱和模糊规则的湿式球磨机负荷集成模型及在线更新研究。首先建立研磨机理数值仿真和筒体振动机理模型，并结合经验模态分解等技术进行多组分振动信号分析，为频谱特征选择和机理知识提取奠定理论依据；接着建立基于优化选择频谱特征和专家机理知识规则的集成模型，实现数据驱动和模糊推理模型的互补融合；在此基础上，智能识别表征磨矿过程特性漂移的更新样本，用于在线构建多层集成软测量模型，并进行更新算法的稳定性分析；最后开展所提方法的实验验证和工业应用研究。该项目对实现磨机负荷实时检测、磨矿过程优化控制和节能降耗意义重大。

磨矿过程是利用矿物在物理或者物理化学性质上的差异性，通过磨矿设备获取有用矿物的过程。球磨机是该过程所必需的重型旋转机械设备，保证其运行在最佳负荷是实现磨矿过程运行优化与反馈控制的关键因素之一。磨机封闭旋转、连续运行的工作特性导致难以对磨机负荷进行直接检测与监视，同时也难以构建有效的机理模型。工业界通常基于磨机研磨产生的振动、振声、电流等信号构建数据驱动软测量模型。针对研磨机理不清、建模数据难以获得、检测模型难以合理阐释等问题，提出由实验设计、数值仿真、信号处理三部分组成的筒体振动分析策略，研究基于EDEM和ABAQUS的磨机筒体振动生成系统,结合真实实验磨机进行数值仿真和分析。针对具有小样本特性的高维共线性数据建模难的问题，提出基于双层遗传算法的选择性集成核偏最小二乘建模方法。针对磨矿过程的连续不间断运行特性和机械设备旋转封闭的特殊性导致获取完备训练样本的经济性差和周期性长等问题，提出了基于虚拟样本生成技术的多组分机械信号建模方法。针对磨机筒体振动/振声多尺度频谱与磨机负荷参数间的模糊特性、多源多尺度频谱间的冗余性与互补性等问题，提出了基于多尺度频谱核潜在特征的选择性集成模糊推理软测量方法。针对潜结构映射模型和模糊推理模型在建模机理上存在差异性、难以模拟领域专家认知磨机负荷等问题，提出了基于数据和模糊推理的混合集成磨机负荷软测量方法。针对近似线性依靠程度和预测误差等指标只能片面反映建模对象的漂移程、领域专家结合工业过程需要依据上述指标和自身经验进行更新样本有效识别等问题，提出了基于更新样本智能识别算法的自适应集成建模策略；进一步，提出了基于样本分布统计检验的双窗口概念漂移检测方法。结合太原理工大学搭建了多通道实验磨机机械信号采集系统并进行大量实验，接着基于针对实验磨机机的分析结果和工业磨机的实际运行情况，搭建了多通道工业磨机机械信号采集系统并开展工业实验研究。本项目为磨机负荷实时检测和磨矿过程的运行优化提供有效支撑。