Adaptive least contribution elimination kernel learning approach for rubber mixing soft-sensing modeling

Rubber mixing process is a typical non-linear fed-batch process without well developed mechanism. Soft-sensing modeling of the mixture´s Mooney viscosity is crucial and challenging since this index is an important process criterion to judge the quality of rubber compounds while the measurement of Mooney viscosity is time-consuming and laborious to assay. Furthermore, the mixing process is drifting and volatile even noisy; only few data samples could be used to modeling. In present paper, an adaptive least contribution elimination kernel learning (ALCEKL) approach is proposed to predict the Mooney viscosity. It adopts a sparsity strategy of least contribution elimination and presents a buffer based learning algorithm associated with improved space angle index (SAI) strategy. Experiments on field data indicate that proposed approach is more robust and accurate than other kernelized modeling methods with feasible computational complexity under field circumstances.