Data-driven deep learning prediction of full molecular weight distribution in polymerization processes

Abstract

The mathematical modelling of the full molecular weight distribution (MWD) results in a large set of ordinary differential equations (ODEs), which usually requires considerable computation time because of stiffness behaviour. This study applies state-of-the-art deep learning (DL) methods to model three academically and industrially relevant polymerization processes: free radical polymerization (FRP), reversible addition–fragmentation (RAFT), and coordination catalyst polymerization (CCP). The DL models were trained with datasets generated from the numerical solution of the first principles kinetic model of each polymerization process. Then, the applied DL models were used to predict the conversion rate, average molar weights, and molecular weight distributions with minimum deviations and reduced computational load. Therefore, by reducing the large computational load, this type of DL models can make feasible the application of on-line optimal control strategies to complex and economically important polymerization processes.