Modeling of a Three-Phase Industrial Batch Reactor Using a Hybrid First-Principles Neural-Network Model

We present an industrial case study of a three-phase reaction system in a batch reactor. For the successful modeling and prediction of the plant-scale performance, a hybrid model is used. Data from different scales were available for developing and testing the model. Laboratory data from a 1 kg vessel were used for the determination of the parameters related to reaction kinetics. Here, a first-principle approach is applied for modeling reactions and dissolution of one reactant that is introduced into the reactor in solid form. To model the large-scale production process, this model was extended with neural-network models to identify the missing parameters. In addition, a refitting of the kinetic parameters on the plant scale was required. With this hybrid model, a good prediction of the concentration courses in the industrial reactor was obtained and improved operating conditions were identified. This paper discusses the potentials and limitations of applying hybrid models in complex processes in the chemical industry.