【作者】 李志明；

【导师】 彭艳； 于丙强；

【作者基本信息】 燕山大学 ， 机械设计及理论， 2012， 博士

【摘要】 本文以整辊镶块式板形辊的信号处理和板形闭环控制方法为主要内容，展开了理论、工业验证及仿真研究，所取得的成果对我国掌握具有自主知识产权的板形检测与控制技术具有积极的理论和现实意义。以整辊镶块式板形辊为对象，在分析其信号产生及提取原理的基础上，以提高板形信号检测精度及系统稳定性为目标，对板形信号处理的有关内容展开全面研究，从激磁信号产生方法、板形信号处理系统架构、信号补偿模型及抗干扰方法等方面对板形信号处理理论及方法进行了补充和完善。基于单片机产生激磁信号方法，采用D/A转换方式产生的激磁信号具有稳频、稳幅、稳相特性，可有效避免板形信号在处理过程中的精度损失。在分析传统系统架构模型的不足及其成因的基础上，提出“信号处理板卡+上位机软件”板形信号处理系统架构模型，将信号提取、数据采集与处理等实时性要求很强的功能下移至信号处理板卡并由DSP芯片控制；该模型抗干扰能力强，上位机工作负载小，可有效增强整个系统的实时性、稳定性及可靠性。利用自主研制的变包角实验装置，通过物理实验方法，寻求径向压力与带钢包角、旋转频率之间的关系，建立带钢包角补偿和幅频特性补偿模型；通过理论分析方法，建立适应带钢跑偏情况的带钢边部覆盖补偿模型。同时，研究板形信号处理过程中不同阶段的抗干扰方法，并用于指导板形信号处理系统的研制，从而为提高板形信号处理系统的抗干扰能力及稳定性奠定基础。以板形信号处理理论及方法为指导，自主研制整辊镶块式板形仪信号处理系统。经标定及通讯测试后，将整辊镶块式板形仪与AFC系统共同投入到鞍钢1250HC冷轧机的工业应用中，并从抗干扰能力、板形检测精度、板形闭环控制效果等方面对板形信号处理系统的性能进行分析与验证。工业应用表明，该系统具有抗干扰能力及实时性强，信号提取及转换精度高，稳定可靠等优点。针对现有人工神经网络方法求解板形控制影响矩阵的不足，提出基于影响矩阵自学习的板形闭环控制方法。该方法利用事先确定的关键影响因素，建立影响矩阵先验值表；根据实际工况点在影响矩阵先验值表中的位置，利用质心插值原理在线计算影响矩阵，实现板形调控机构调节向量的计算；在各种板形影响因素共同作用下，影响矩阵自学习模型利用轧制过程数据不断修正典型工况点的影响矩阵先验值，使其与板形调控机构的实际调控性能不断接近，进而为板形闭环控制快速地进入稳定轧制期、达到精度要求奠定基础。采用质心插值原理实现影响矩阵的在线求解，通过影响权重因子及自学习速度因子的动态调节实现影响矩阵的自学习过程；与人工神经网络方法相比，该方法具有计算速度快、稳定性好、可靠性高、适合在线应用、便于实施等特点。利用1250HC冷轧机的有关实测数据，对基于影响矩阵自学习的板形闭环控制方法进行了仿真研究。结果表明，该方法板形控制精度较高，控制过程平稳，收敛速度较快。更多还原

【Abstract】 Focusing on the signal processing on shape meter of entire roller embedded withelatic bolocks and shape closed-loop control method, theoretical research, industrialapplication and simulation research were carried out in this paper, and the newachievements gained has positive theoretical and practical significance to master the shapemeasuring and controlling technology with proprietary intellectual property rights.Taking the shape measuring roller embedded with elatic bolocks as research objects，on the basis of analysis on shape signal generation and extraction principle, and aiming toimprove the measuring precision of shape signal and system stability, a body of theory andmethod which is used for the research and development of shape siganl processing system,is supplemented and completed by the related research on shape signal processing fromthe standpoints of excited signal generating method，system architecture of shape signalprocessing, signal compensation model and anti-disturbance methods. This theory mainlyincludes the following content. A method generating high quality excited signal based onsinglechip, which generates the excited signal with the characters of frequency, amplitudeand phase position stabilization by D/A convertion, and this can effectively avoid lossingthe shape signal precision. According to disadvantage and its reason of traditional singleprocessing system architecture model, the system architecture model “signal processingbaord+upper machine software” is proposed. In this model, the functions with highrequirement of real-time, such as signal extraction, data acquisition and Data Processing,are lowered into the signal processing board and controlled by DSP chip. This model hasstrong anti-disturbance ability, lesser work load on the upper machine, and is efficient toenhance the real-time, stability and reliability of this system. The relations among theradial pressure, wrap angle and rotational frequency are gained by physical experimentwith the self-made alterable wrap angle experimental installation, further, thecompensation models for strip wrap angle and amplitude-frequency characteristic arecreated. The compensation model for marginal measuring unit coated by clod strip,suitable for the situation of strip wandering, is created by theoretical analysis.Simultaneously, the anti-distance methods of the different stage in the shape signal processing procedure are researched, and are applied into the development of shape signalprocessing system in order to improve the system anti-distance ability and stability.The signal processing system of shape meter of entire roller embedded with elaticbolocks is developed under the direction of the above theory. After calibrating anddocking testing in laboratory, the shape meter of entire roller embedded with elaticbolocks and the AFC system are applied together into the industrial application, and fromthe angle of anti-disturbance capacity, shape measuring precision and effect of shapeclosed-loop control, the performances of signal processing system are analyzed andverified. The industrial application presented that the shape signal processing system hasstrong anti-disturbance ability, powerful real-time performance, high precision ofextraction and convertion, more stability and reliability.Directing at the defects of artificial neural networks method to obtain shape controleffective matrix, a shape closed-loop control method based on self-learning of effectivematrix is proposed. The effective matrix empirical value table is created by keyinfluencing factors determined beforehand. According to the position of actual workingcondition point in the effective matrix empirical value table, the value of effective matrixis calculated on-line by centroid interpolation principle, further, the regulating variables ofshape control actuator are gained. Under the combined action of the various shapeinfluencing factors, self-learning model of effective matrix makes use of the rollingprocess data to correct the effective matrix empirical value of the relevant typical workingcondition points. This makes the effective matrix empirical values approximate to theactual regulation performances, and provides the foundation for the shape closed-loopcontrol which enters the stability rolling period and reachs to the required precision asquickly as possible.The on-line calculating of effective matrix is realized by centroid interpolationprinciple, and the realization of self-learning process of effective matrix is made by weightfactors and dynamic regulating of self-learning speed factor. Compared with the artificialneural networks method, this method has rapid computation speed, good stability andreliability, and is suitable to on-line application and easy to implement. Using the relevantreal data of1250HC cold mill, the simulation of the shape closed-loop control method based on self-learning of effective matrix is carried out. The simulation result shows thismethod has better shape control precision, smooth control process and higher convergingspeed.更多还原