Mass and energy balance-assisted data-driven modeling and optimization of batch processes: The case of a batch polymerization process

Abstract

In this paper, we use two data-driven modeling methodologies that we have been recently developed: the design of dynamic experiments (DoDE) (Georgakis 2013) and the dynamic response surface methodology (DRSM) (Klebanov and Georgakis 2016, Wang and Georgakis 2017). DoDE allows time-varying inputs, and DRSM models time-varying process outputs. We combine the above two data-driven tools and partial process knowledge to develop an integrated data and knowledge-driven model. We use this hybrid model to improve process productivity. The material and energy balances we use here lack a quantitative description of all rate phenomena. The optimization of the process is evolutionary and thus done in cycles. After three cycles, we have increased the productivity of the process, defined as the ratio of polymer produced over the batch duration, by 39.6% compared to the base case productivity. We also show how this productivity increase could have been as high as 81%.