机理和数据融合的304不锈钢极薄带轧制力模型  

Rolling force model for 304 stainless steel ultra-thin strip based on mechanism and data fusion

作　　者：任鹏帆 王振华 贾燚[1,2,3] 刘元铭 王涛 REN Pengfan;WANG Zhenhua;JIA Yi;LIU Yuanming;WANG Tao(College of Mechanical and Vehicle Engineering,Taiyuan University of Technology,Taiyuan 030024,Shanxi,China;Engineering Research Center of Advanced Metal Composites Forming Technology and Equipment,Ministry of Education,Taiyuan University of Technology,Taiyuan 030024,Shanxi,China;National Key Laboratory of Metal Forming Technology and Heavy Equipment,China National Heavy Machinery Research Institute Co.,Ltd.,Xi’an 710018,Shaanxi,China)

机构地区：[1]太原理工大学机械与运载工程学院,山西太原030024 [2]太原理工大学先进金属复合材料成形技术与装备教育部工程研究中心,山西太原030024 [3]中国重型机械研究院股份公司金属成形技术与重型装备全国重点实验室,陕西西安710018

出　　处：《钢铁》2024年第10期64-76,共13页Iron and Steel

基　　金：国家自然科学基金资助项目(52105390);中国博士后科学基金面上资助项目(2023M742576);海安太原理工大学先进制造与智能装备产业研究院开放研发资助项目(2023HA-TYUTKFYF031);金属成形技术与重型装备全国重点实验室开放课题资助项目(S2208100.W02,S2308100.W18)。

摘　　要：304不锈钢精密极薄带广泛应用于数码电子、航空航天、医疗设备、汽车制造等高端领域。在冷轧生产过程中,轧制力是一项至关重要的参数,它的设定精度直接影响轧制过程的稳定性以及最终产品的厚度精度。为了提高冷轧304不锈钢极薄带轧制力的预测精度,在森基米尔轧机上针对304不锈钢极薄带进行轧制试验,并据此建立了适用于超薄规格304不锈钢极薄带的动态变形抗力模型,修正了传统Bland-Ford-Hill轧制力机理模型。将修正的轧制力机理模型与遗传算法(GA)优化XGBoost参数相融合构建了高精度GA-XGBoost轧制力预测模型,该模型不仅考虑了轧制力的非线性问题,还将机理与数据相融合来实现轧制力的精准预测。与相同数据集下其他典型机器学习算法所建立的轧制力预测模型进行对比分析,采用相关系数(R~2)、平均绝对误差(E_(MA))、均方根误差(E_(RMS))和平均绝对百分误差(E_(MAP))多个指标全方位、多角度对模型的泛化性能进行评价,结果表明,所提出的GA-XGBoost轧制力预测模型的预测结果最为精确,其测试集各指标分别为R~2值为0.984,E_(MA)值为3.040,E_(RMS)值为6.410以及E_(MAP)值为0.041%。研究结果证明了所提出的机理和数据双驱动的轧制力预测模型的准确性和优越性。该研究为304不锈钢精密极薄带材轧制力的设定提供了一种新思路。304 stainless steel precision ultra-thin strip is widely used in high-end fields such as digital electronics,aerospace,medical equipment,and automotive manufacturing.In the cold rolling production process,rolling force is a crucial parameter,and its setting accuracy directly affects the stability of the rolling process and the thickness accuracy of the final product.In order to improve the prediction accuracy of rolling force for ultra-thin strips of coldrolled 304 stainless steel,the rolling experiment of 304 stainless steel was carried out on the Senkimir rolling mill,and the dynamic deformation resistance model suitable for ultra-thin 304 stainless steel was established,and the traditional Bland-Ford-Hill rolling force mechanism model was modified.A high-precision GA-XGBoost rolling force prediction model was constructed by integrating the modified rolling force mechanism model with genetic algorithm(GA)optimized XGBoost parameters.This model not only considers the nonlinear problem of rolling force,but also integrates the mechanism with data to achieve accurate prediction of rolling force.Compare and analyze the rolling force prediction models established by other typical machine learning algorithms under the same data,and the generalization performance of the model is evaluated comprehensively and from multiple perspectives using multiple indicators such as correlation coefficient(R~2),mean absolute error(E_(MA)),root mean square error(E_(RMS)),and mean absolute percentage error(E_(MAP)).Research has shown that GA-XGBoost has the most accurate prediction performance,with various indicators in its test set being R~2 value of 0.984,E_(MA)value of 3.040,E_(RMS)value of 6.410,and E_(MAP)value of 0.041%.This proves the accuracy and superiority of the mechanism and data driven rolling force prediction model proposed.It provides a new approach for setting the rolling force of precision ultra-thin strips of 304stainless steel.

关 键 词：冷轧 304不锈钢极薄带 轧制力预测 集成学习 遗传算法 

分 类 号：TG335.56[金属学及工艺—金属压力加工]