融合物理冶金学与机器学习的组织性能预测及热轧工艺优化  

Microstructure and properties prediction and optimization of hot rolling process based on physical metallurgy and machine learning

作　　者：李鑫[1] 周晓光[1] 曹光明[1] 崔春圆 吴思炜 刘振宇[1] LI Xin;ZHOU Xiaoguang;CAO Guangming;CUI Chunyuan;WU Siwei;LIU Zhenyu(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：《冶金自动化》2023年第2期16-26,共11页Metallurgical Industry Automation

基　　金：国家重点研发计划项目(2021YFB3702404);中国博士后科学基金(2022T150205)。

摘　　要：传统的钢材组织性能预测和工艺优化技术(物理冶金模型和神经网络模型)无法兼顾钢材性能预测的高精度和物理冶金学基本规律,因此,对钢材生产的实际指导作用有限。基于对工业大数据的关联、清洗等处理,将数字孪生技术应用于钢铁材料组织性能预测及热轧工艺优化中。通过对有效工业数据的机器学习,在保证物理冶金学数学模型规律的条件下,实现了力学性能的高精度预测。以轧制力预测为例,介绍了融合物理冶金学与机器学习的预测方法。利用人工智能算法,对生产工艺进行针对性设计,实现了智能化热轧,主要体现在热轧钢材的定制化生产、减量化合金设计、钢种归并、性能稳定性控制等。最后,对组织性能预测及热轧工艺优化领域的成果进行了介绍。The traditional techniques of microstructure and properties prediction and process optimization(physical metallurgy model and neural network model)can not give attention to the high precision of steel material properties prediction and the basic laws of physical metallurgy,so the practical guidance for steel production is limited.Digital twin technology is applied to predict microstructure and mechanical properties as well as optimization of hot rolling process of steels after industrial big data are related and cleaned.Under condition of physical metallurgy regular correctly,high accuracy prediction of mechanical properties is realized through machine learning of effective industrial data.Taking rolling force prediction as an example,the prediction method based on physical metallurgy and machine learning was introduced.Smart hot rolling is realized through targeted design of production process using artificial intelligence algorithm.Smart hot rolling is mainly reflected in the customized production of hot rolled steel,the design of reducing alloy,the merging of steel grades,and the stability control of mechanical properties,etc.Finally,the achievements in the field of microstructure and mechanical properties prediction and optimization of hot rolling process of steels were introduced.

关 键 词：数字孪生 机器学习 数学模型 组织性能预测 热轧工艺优化 

分 类 号：TG335.11[金属学及工艺—金属压力加工] TF01[冶金工程—冶金物理化学] TP181[自动化与计算机技术—控制理论与控制工程]