Online prediction of Mooney viscosity in industrial rubber mixing process via adaptive kernel learning method

Abstract

Mooney viscosity is an important while difficult-to-measure quality index with a long-term laboratory assay in nowadays internal rubber mixing processes. In this study, an adaptive kernel learning (AKL) algorithm suitable for nonlinear multi-input-multi-output process modeling is applied to online prediction of Mooney viscosity. The developed AKL algorithm utilizes a sequentially sparse strategy to control the model complexity and adopts a two-stage recursive learning mechanism to update the network topology effectively. Consequently, the AKL model can trace different characteristics of the internal mixing process. The developed AKL modeling method has been successfully applied to online prediction of Mooney viscosity in several rubber and tire manufactories in China. The industrial applications show that the AKL model exhibits good modeling ability and predicts Mooney viscosity successfully. Furthermore, the comparison results indicate that AKL is superior to conventional recursive partial least squares method.