(714e) Development of Process Models Suitable for Controller Design - Use Case: Flow Synthesis of Mesalazine

The implementation of continuous manufacturing routes – both in primary and secondary manufacturing – requires the design of suitable process control concepts. Besides traditional approaches based on standard controllers like proportional-integral-derivative (PID), also model based control concepts, e.g., model predictive control (MPC), are increasingly realized in industry. Such concepts allow the systematic consideration of process constraints and the intuitive formulation of the control objective. However, these model-based approaches require the availability of a suitable process model. The process model needs to describe the real system dynamics at a sufficient accuracy, while keeping the model complexity - in terms of model structure and computational effort needed for performing the predictions - low enough.

In the talk, modeling approaches that are suitable for controller design are presented. Next to mechanistic methods based on mass- & energy-balances and kinetic modeling, also data driven concepts based on the local linear model tree (LOLIMOT) algorithm are outlined. The former one offers more process insight, but might be infeasible for complex processing steps where the process understanding is insufficient to apply the mechanistic approaches. The latter one allows the modelling of reactions where the underlying mechanisms are not known, but a sufficient amount of experimental data is available. The concepts are demonstrated via the use case of continuous Mesalazine synthesis, which consists of three main steps: nitration, substitution and hydrogenation. After each of the steps, PAT devices are installed to measure the respective reaction outcome. The captured data is used to parameterize the models and to test the model accuracy.