【作者】 刘艳；

【导师】 童朝南；

【作者基本信息】 北京科技大学 ， 控制科学与工程， 2016， 博士

【摘要】 板带生产综合自动化控制系统是复杂工业过程控制最具代表性的系统。板带轧制生产中控制对象具有多变量、强耦合、非线性、大滞后和时变性等复杂性,对控制系统而言尤其以高精度、高速度为特征,是冶金企业过程控制系统中复杂程度最强、控制难度最高的综合自动化系统,其控制水平与规模在相当程度上代表和反映一个国家工业化、自动化、信息化的发展水平。相对于传统的板带轧机而言, “现代”板带轧机具有液压驱动的机械设备、交流调速的主辅电气传动、完备的三级计算机控制系统和齐全的检测仪表等特征,轧制过程必依赖于数学模型。为此,数学模型是现代板带轧机控制中的核心和关键技术。建立和应用高可用性、高精度的数学模型,是冶金企业能够持续开发和生产出满足需要的板带材品种,不断提高产品质量,保持稳定生产,节能降耗,降低成本,从而提升综合竞争力的技术基础。论文以板带轧机生产过程为对象,从理论建模和数据驱动的角度出发,研究板带轧机复杂的机理模型;研究以优化产品质量,提高生产效率为目标,充分利用工业过程的历史和实时数据,运用人工智能、统计及有限元分析等基于知识和数据驱动的建模方法。从而分析并建立现代板带轧机的关键数学模型,把数据资源的优势转化为生产效益和产品质量的优势。本文主要研究工作及创新点有以下4个方面：1)提出将遗传算法与有限元分析合理结合,建立轧辊温度场及热辊形模型,并分析轧制参数对温度场的影响,使热辊形对板形的影响最小化,为板形控制提供有效的技术支持。2)结合小波多分辨分析和多RBF神经网络的轧制力建模方法,将轧制力分解重构为对应于不同影响因素的子信号,建立一个多RBF神经网络模型。仿真结果表明该模型可以真实反映轧制力变化的内在机理,而且网络训练时间短、预测精度高,是轧制力建模的一种有效方法。3)对弯辊力设定中的关键参数——工作辊相对挠度进行详细推导,并采用显式动力学弹塑性有限元法,建立四辊轧机弯辊力模型,研究并分析液压弯辊力对轧制过程中板形参数以及工作辊参数的影响规律,为进一步建立和优化板形控制模型奠定基础。4)推导以勒让德多项式为基模式的板形识别模型,并运用真实过程数据仿真验证其优越性。该模型计算量小,多项式系数直接对应于板形控制的三种基本手段,物理意义明确,工程实施灵活快捷。研究成果已成功应用在某铝箔厂1850mm冷轧生产线中,取得良好的控制效果。论文的研究进一步丰富了基于知识与数据驱动的复杂工业过程的优化建模理论与方法,以信息化带动工业化,实现现代板带轧机的优化运行与控制。更多还原

【Abstract】 The integrated automation system for strip production is the most representative system of complex industrial process control.The control object in strip rolling production has the complex characteristics of multi-variable, strong coupling,nonlinear, large time-delay and time variation and so on.Furthermore, in which auto-control system is especially featured with high precision and high speed, then become the most complicated and difficult integrated automation system. Which level and scale are always highlighting development phase of a national industrialization, automation, informatization to a large extent.Compared with the traditional strip rolling mills, modern ones are characterized by hydraulic driven mechanical equipment. AC speed regulation of main and auxiliary electric drive, the complete three-level computer control system and all ready testing instruments etc.,their rolling process are supposed to rely on math models. Therefore, math model is becoming the key technology in modern strip mills’control. Establishing and applying high availability and high precision mathematical model are the fundamental technology to enhance an enterprise’s competitive strength with developing and producing satisfied steel products, improving the quality of products, maintaining the stability of making process, saving energy and reducing cost.Taking strip rolling process as the object, with the perspective of theory modeling and data-driven, this paper studied the complicated mechanism models and worked modeling method out which is based on knowledge and data-driven. It made full use of the historical and real time data made by industrial processes, adopted modeling method like artificial intelligence, statistics, finite element analysis with the goal of optimizing product quality and promoting product efficiency and established the key mathematical models of strip mill to fulfill the transformation from the advantages of data resource to productivity effect and product quality.The following 4 aspects are the main research work and innovation:1) A model including roll temperature field and thermal contour was put forward combining genetic algorithm and finite element analysis. The effective technical support for shape control was provided by analysizing the influence of rolling parameters on the temperature field as well as minimizing the impact of the heat roller on the flatness.2) A multiple RBF neural network model for rolling force was established by combing with methods of wavelet multi-resolution analysis and multiple RBF neural networks, where the rolling force is decomposed and reconstructed into sub signals which correspond to different influencing factors. The simulation results show that the model can truly reflect the intrinsic mechanism of the rolling force variation, and it is an effective method for rolling force modeling with the short network training time as well as the high prediction accuracy.3) The working roller flexibility, the key parameter on setting bending force, equation was deduced and then the roller bending force model for a four-roller mill was presented by adopting the method of explicit dynamic elastic FEM (Finite Element Method). The model can simulate the strip rolling process, study and analyze the influence law of the hydraulic bending roll on the flatness parameters during the rolling process as well as the working roll parameters. The work is fundamental for further model on flatness control and optimization.4) A flatness pattern recognition model was deduced based on Legendre polynomial. The model has less computation and clear physical meaning for its polynomial coefficients directly corresponding to the three basic means to shape control. It is easier and more flexible on project implementation. The results have been successfully applied in a 1850mm aluminum foil cold rolling production line, and got good control effect.The outcome further enriches the theory and method of optimization modeling based on knowledge and data-driven on complex industrial processes and promotes industrialization by informatization and optimization operation and control on modern strip mill.更多还原