Generation, comparison and contrast of polymer chemistry and neural network models of a solution styrene-butadiene rubber production process

Full scale solution anionic styrene-butadiene rubber plant data fed the development of the Polymer Chemistry Model, which predicts Mooney viscosity by relating initiator and monomer flows to Mooney viscosity via polymer chemistry principles and rheological relationships, as well as the Neural Networks Model, which predicts Mooney viscosity by training a Neural Network system with historical input and output process data. The Polymer Chemistry Model was able to fit test data to an R2 of 0.24, while the best iteration of the Neural Networks Model was able to fit test data to an R2 of 0.65.;If applied to the same process in future control, the Polymer Chemistry Model and Neural Networks Model could improve the process performance for within specification Mooney from 15,600 defects per million opportunities to 400 defects per million opportunities and less than 1 defect per million opportunities respectively.