【作者】 张材；

【导师】 谭建平；

【作者基本信息】 中南大学 ， 机械电子工程， 2004， 博士

【摘要】 薄带坯铸轧是新一代铝加工发展的重要方向和铝加工技术的制高点,而板形控制是当今铸轧技术领域的一个前沿课题,是保证铸轧工艺顺利进行,控制产品质量的关键技术。它包括板形模式识别技术与板形控制技术。 本文以国家计委产业化前期关键技术及成套装备研制与开发项目“铝及铝合金铸轧新技术与设备研制开发”(计高技[1998]1985)为背景,以铸轧板形控制技术与装置为研究对象,分析总结了铸轧板形形成机理与控制特性,提出了铸轧板形控制的思路,并设计了控制方案,研究了BP网络和遗传算法人工智能技术用于板形模式识别的参数设计原则,应用神经网络初始权重多维空间优化方法,建立了GA-BP网络板形模式识别模型;根据铸轧板形控制的思路,引入人工智能预测控制的思想,建立了GA-BP网络板形在线预测模型和预测控制模型;开发了铸轧板形控制系统的主要子系统(板形检测系统、金属流场控制系统、数据采集系统和软件系统),对板形控制系统进行了技术集成:在此基础上,将铸轧板形控制系统及模型应用到了快速减薄铸轧工业试验,并进行了工业试验验证。这对于我国薄带坯铸轧技术的发展具有重要的理论意义和工程实用价值。 论文的主要研究内容和成果如下: 1.通过分析总结铸轧过程板形形成机理与控制特性,指出由于来料不存在板形问题,铸嘴内金属温度场和速度场是影响板形的关键因素,提出了以铸嘴内金属流场调节为主、外冷为辅来调节热辊型,压下缸为辅控制板形的铸轧板形控制思路,并设计了铸轧板形控制系统总体方案。 2.研究了BP网络用于板形模式识别时快速BP算法的适用性,以及主要参数的变化对板形模式识别效果的影响,得出了BP网络板形模式识别参数设计的基本原则。指出BP网络模式识别的重要问题是提高网络对未学习样本的识别能力。 3.研究了把高级语言对象封装的思想引入到遗传算法的优化过程,建立神经网络初始权重多维空间优化模型的方法,应用该方法建立了GA-BP网络板形模式识别模型。研究了遗传算法主要参数对遗传优化性能的影响,设计了以训更多还原

【Abstract】 Thin-gauge roll casting is an important and new development in the technology of aluminum industry. And flatness control is the latest research field in the roll casting, which includes flatness recognition and control technology, and it is the key technology to the roll casting technics and product quality.Funded by the industrialization key technology and whole set equipment research of China Special Plan Project—"New Technology and Equipment Manufacture of the Aluminum and Aluminum Alloy Roll Casting"(No.[1998] 1985), this paper research on the roll casting flatness control technology and equipment. After analyzing flatness forming mechanism and flatness control particularity, the flatness control rules are concluded, and the control project are designed; study the rules of parameter densign with artificial intelligence technology of BP Network and Genetic Algorithms in the flatness pattern recognition, the GA-BP flatness pattern recognition models are innovated and improved by the method of optimizing the neural network initial right in multi-dimension; based on the flatness control rules, with the idea of artificial intelligence in predictive control, the GA-BP flatness on-line predictive and control models are formed; design and empolder the main subsystem(flatness measure system, flow control system, data sampling system and software system), integrate the flatness control technology, and apply the flatness control system to the high-speed thin-gauge roll casting line, which is validated by the industry test. All those are very important to the development of the high-speed thin-gauge roll casting both in theory and application. In this paper, the main research work and achievement as follows:1. After analyzing and summarize flatness forming mechanism and flatness control particularity, point out that the flatness problem dose not exist in raw material, while the tempreture field and velocity field are main factors to the flatness, control method has been concluded which take adjusting hot roller flexibility with flow as primary means, and take controlling the outer cooling and balance jar as subsidiary means, on this point, the control project are designed.2. After researching the applicability of the fast BP learning algorithm in the flatness pattern recognition and the effect of main parameter change to therecognition, the basic rules of the parameter designing are put forward, and point out that improving the recognition ability of the un-training samples is the key to BP network flatness pattern recognition.3. Reseach the method that brings the object encapsulation idea of advanced program into the GA optimization process, and establishs the models of optimizing the neural network initial right in multi-dimension, by this method, the GA-BP flatness pattern recognition models are established. Analyze the effect of main parameter change to the optimization’s capability, design the GA fit function which involve the training sample errors and test sample errors, find the synthetically training strategy that the fit value evolution can’t be too more in process of optimizing the neural network initial right, and the network goal after optimizing must be smaller than before optimizing. Improve the BP network’s recognition ability effectively to the un-training samples.4. According to the flatness control project, the process control models are established, introduce the idea of artificial intelligence in predictive control, use the time-lag of flatness control and real-time data of roll casting mearsurement, establishing GA-BP flatness on-line predictive models and control models in roll casting process, which is validated by the off-line simulation.5. To measure the flatness, design a laser scanner that based on a pair of laser bean difference measure theory, it realize that measure flatness and thickness at the same time in low cost and high precision; develop the flow control system, it’s core is PLC; make a distributed data sample system which based on management site, master control site and slave sample site; adopt the modularization idea to design the software system of roll casting flatness control; based on above systems, the flatness control technology is integrated.6. With the industrial trail data of High-speed thin-gauge roll casting, to validate the GA-BP flatness models and the online predictive models which have been founded, the industrial trail results indicate that the models are correct, and it established a basic in theory and trails for roll casting flatness control.更多还原