A novel nonlinear adaptive Mooney-viscosity model based on DRPLS-GP algorithm for rubber mixing process

摘要

Rubber-mixing process is a typical non-linear batch process with very short operation time (commonly, 2-5 min). The large measurement delay of Mooney-viscosity, one of the key quality indexes of mixed rubber, strongly restricts further improvement of the quality of final rubber products and the development of rubber-mixing process control. A novel nonlinear adaptive Mooney-viscosity prediction model based on Discounted-measurement Recursive Partial Least Squares-Gaussian Process (DRPLS-GP) algorithm is developed. Using rheological parameters as the input variables, which could be measured online, the measurement delay of Mooney-viscosity is markedly reduced from about 240 min to 2 min. In DRPLS-GP model, to overcome the noise and the multi-collinearity of original data, orthogonal latent variables (LVs) are extracted by Discounted-measurement Recursive Partial Least Squares (DRPLS) firstly, and then the LVs are inputted to Gaussian Process (GP) as predictors for further regression. Thus relying on the nonlinear regression power of GP and the multivariate regression power of DRPLS, the nonlinear relationship between rheological parameters and Mooney-viscosity could be regressed successfully by DRPLS-GP. In particular, this method could update Mooney-viscosity prediction model without increasing the computation and sampling burden, so it is very practical for industrial application. Moreover, the flexibility of discounted-measurement factor of the novel method ensures the high precise prediction of Mooney-viscosity of different mixed rubber formulas. The results which are obtained by using of 1006 industrial data sampled in a large-scale tire factory located in east China confirm that the predictive performance of DRPLS-GP is superior to other approaches.