自适应递推核学习及在橡胶混炼过程在线质量预报的工业应用

实时测取混炼胶门尼粘度是橡胶和轮胎厂十分关心和亟待解决的问题. 采用两阶段递推核学习建模方法, 按配方快速建立橡胶混炼过程门尼粘度的预报模型, 并对模型进行递推更新以适应过程的快速变化. 结合混炼过程的特点, 提出一种适合门尼粘度的性能指标, 并推导了采用快速留一交叉验证法对核学习模型参数进行自适应选择, 避免人为选取参数的片面性. 所研发的先进密炼信息集成与控制系统已在国内多家大型橡胶和轮胎厂上线应用. 门尼粘度实时预报的工业应用结果表明了其实用和有效性, 对橡胶混炼过程具有重要的现实和经济意义.

Adaptive recursive kernel learning with application to online quality prediction of industrial rubber mixing process

Mooney viscosity having significant impact on the properties of the polymer is very difficult to be measured online. A new modeling method using two-stage recursive kernel-learning is proposed for online modeling and prediction of Mooney viscosity in the rubber mixing processes. The model can be established online for each recipe and recursively updated to adapt fast changes of the process. In the present method, a novel error evaluation index is formulated based on the mixing properties. The model parameters are online selected adaptively, using the fast leave-one-out cross validation criterion, to overcome the embarrassment of parameter selection. An industrial system named as Smart Mixing Information Integrated & Control System has been developed and successfully applied to several large-scale rubber and tire manufacturers in China. The results of Mooney viscosity online prediction show that the developed method is very efficient and thus has real economic importance for rubber mixing processes.