Parameter estimation and dynamic optimization of an industrial fed-batch reactor

Modeling and optimization of fed-batch reactors with several multi-step reaction pathways is challenging due to the nonlinear dynamic system behavior and large number of kinetic parameters. We showcase the model-based optimization of an industrial (20 m3) fed-batch reactor by using our open-source dynamic optimization software DyOS. First, we build a detailed mechanistic model of the fed-batch reactor. Second, we conduct parameter estimation of the mechanistic model with 25 states and 44 fitting parameters using historic time-series industrial production data. Third, we perform dynamic multi-stage optimization and identify optimal feeding profiles for the operation, targeting improvements in economic profit over the established experience-based production routine. We demonstrate substantial economic improvement: The optimized production recipe can save up to 10% of raw material at the same yield of main product. Our findings underline the strong capabilities of model-based process optimization and its application to industrial challenges in process design and operation.