(149c) Optimizing an Industrial Batch Process Using the Design of Dynamic Experiments Methodology

The recently proposed Design of Dynamic Experiments (DoDE) methodology (Georgakis, 2013) which generalized the classical Design of Experiments (DoE) approach, is applied to an industrial batch polymerization process in silico. The detailed knowledge-driven model of this process, unknown to the first author, is used in simulation mode by the industrial side to perform the simulated experiments. The fist author is provided with no details about the inner characteristics of the process, besides that it is a polymerization reaction which should have no more than 1% unreacted monomer and no more than 1% of impurities in the product an the end of the batch. He is also provided the operational conditions of three batch runs, from which only the first produces acceptable product quality. The optimization objective is to decrease the batch duration while producing acceptable quality product. An initial set of 12 experiments is designed around the single successful recipe by considering time-dependent variations of the initial monomer feeding profile and the reactor temperature. None of these runs violated the quality constraints by more than 10%, the normal variability of the process. The simulated “experimental” results leed to an initial data-driven model, which calculates the operating conditions of a run with 6.5% reduced batch duration and a product of acceptable quality. This predication is confirmed through a follow-up in silico experiment by the industrial side. I second set of experiments is then designed to complement the first set and aiming to reduce the batch time by 20% while keeping the expected product quality within the desired bounds. The presentation will present details on the overall methodology and an assessment of the overall results of the two sets of experiments. It will also commend on the use of the DoDE approach in real rather than simulated processes.

GEORGAKIS, C. 2013. Design of Dynamic Experiments: A Data-Driven Methodology for the Optimization of Time-Varying Processes. Industrial & Engineering Chemistry Research, 52, 12369-12382.