基于连续曲面的轧制模型自学习方法

发布时间：2019-07-12 10:29

【摘要】：针对基于层别数据的传统轧制模型自学习方法导致相邻层别上自学习系数跳跃大、不连续等问题,提出"机理模型+特征点+拟插值+自学习"的轧制模型构建新机制,用多维空间连续曲面代替原来的层别,实现轧制模型结构的升级。构造用特征点表征的连续曲面,采用连续函数对空间中各个特征点上的自学习系数进行拟插值的方法,获得光滑曲面方程。从相邻层别不连续变为多维空间上连续可导,使模型自学习系数精确到空间任意点,对提升模型设定精度有质的突破。该模型自学习方法目前已成功应用于国内某大型热连轧机组变形抗力在线计算,实际生产应用表明,新方法上线后变形抗力与轧制力的预报精度显著提高,带钢因厚度超差导致的预封锁量减少了44%,满足热连轧带钢稳定轧制的生产要求。
[Abstract]:In order to solve the problems of large jump and discontinuity of self-learning coefficient on adjacent layers caused by the traditional self-learning method of rolling model based on layer data, a new rolling model of "mechanism model feature point quasi-interpolation self-learning" is proposed, and the multi-dimensional spatial continuous surface is used instead of the original layer to upgrade the rolling model structure. The continuous surface represented by feature points is constructed, and the self-learning coefficients on each feature point in space are quasi-interpolation by continuous function, and the smooth surface equation is obtained. From the discontinuity of adjacent layers to the continuous derivation of multidimensional space, the self-learning coefficient of the model is accurate to any point in the space, and there is a qualitative breakthrough in the setting accuracy of the lifting model. At present, the model self-learning method has been successfully applied to the on-line calculation of deformation resistance of a large hot strip mill in China. The practical production application shows that the prediction accuracy of deformation resistance and rolling force after the new method is significantly improved, and the pre-sealing locking amount caused by the thickness difference of the strip is reduced by 44%, which meets the production requirements of stable hot strip rolling.
【作者单位】：武汉科技大学信息科学与工程学院;武汉科技大学冶金自动化与检测技术教育部工程研究中心;东北大学研究院;
【基金】：国家自然科学基金资助项目(51774219) 湖北省教育厅科学技术研究资助项目(D20161103) 武汉市青年科技晨光计划资助项目(2016070204010099) 东北大学轧制技术及连轧自动化国家重点实验室开放课题基金资助项目(2017RALKFKT004)
【分类号】：TG335

【相似文献】