板带热轧过程耦合机器学习及生产过程多目标优化  

Coupling machine learning for plate and strip hot rolling and its application to the process multi-objective optimization

作　　者：刘振宇[1] 周晓光[1] 曹光明[1] 吴思炜 王国栋[1] LIU Zhenyu;ZHOU Xiaoguang;CAO Guangming;WU Siwei;WANG Guodong(State Key Laboratory of Rolling and Automation,Northeastern University,Shenyang 110819,China)

机构地区：[1]东北大学轧制技术及连轧自动化国家重点实验室,辽宁沈阳110819

出　　处：《轧钢》2022年第6期29-36,45,共9页Steel Rolling

基　　金：国家十四五重点研发计划项目(2021YFB3702404);国家自然科学基金项目(52104370);中国博士后科学基金项目(2022T150205)。

摘　　要：针对板带热轧过程集成化程度极高、工序较多且材料与力学现象强耦合的特点,按常规自动化控制方法只能关注局部优化,时常以降低其他性能为代价,成为阻碍热轧板带综合质量提升的主要瓶颈。为解决这一难题,采用人工智能理论的机器学习方法,将物理冶金学理论模型与数据(热轧生产大数据+冶金与材料学实验数据)有机结合,开发了综合反映热轧板带全流程组织演变、表面氧化及力能参数变化的数字孪生系统。以此为基础,采用多目标智能优化理论和算法,开发出热轧工艺参数全局寻优的逆向优化系统并在实际生产中得到应用,实现了板带热轧过程中氧化铁皮厚度与结构演变、显微组织演变及道次轧制力的在线软测量。结合多目标优化,在提升热轧板带表面质量及其性能稳定性方面在工业生产中效果明显,为综合、全面解决热轧板带组织性能优化、表面质量提升及板形与尺寸精确控制方面存在的关键难题提供了新的技术方案。In view of the characteristics of a high integration degree, complex processes, and strong coupling between materials properties and mechanical behaviors during plate and strip hot rolling, conventional automatic control methods only focus on local optimization, which is often at the expense of reducing other properties and qualities, hindering the improvement of the comprehensive qualities of hot rolled plate and strip. To solve this problem, by adopting the coupling machine learning methods and combining the physical metallurgical models with data(hot rolling production big data+metallurgical and material science experimental data), a digital twin system was developed to comprehensively describe the microstructure evolution, oxidation and mechanical property in the whole process of hot rolling. Based on this, a reverse optimization system for global optimization of hot rolling process parameters was developed by using multi-objective intelligent optimization theory and algorithm, which was applied in practical production. The online soft measurement of the thickness and microstructure evolution of oxide scale, microstructure evolution and rolling force in the hot rolling process of plate and strip was realized. Combined with multi-objective optimization, it has obvious effect in industrial production in terms of improving the surface quality and property stability of hot rolled plate and strip, and provides a new technical solution for comprehensively solving the key problems in the optimization of microstructure and property of hot rolled plate and strip, improvement of surface qualities, and precise control of plate shape and size.

关 键 词：板带 热轧过程 耦合机器学习 工艺逆向优化 综合质量 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TG335.11[自动化与计算机技术—控制科学与工程]