(149q) Model-Based Control for Liquid-Liquid Extraction

Liquid-liquid extraction is a common task in downstream processing of diverse compounds of chemical and pharmaceutical interest, either to remove impurities from a product stream or to carry a substance of interest into another liquid phase. To support the selection of the most suitable process route, process models are of high interest. These models predict critical quality attributes (CQA) of the pharmaceutical product based on critical process parameters (CPP) of the process investigated.

Increasing attention is given to process models available in early development stages of continuous manufacturing setups.Virtual testing of different process routes, operation scenarios and control concepts can reduce the amount of costly experimental test runs. By now, final process designs are frequently based on results from such virtual experiments. Suitable control concepts help to identify optimization potentials (e.g., reducing the amount of waste generated during processing). In the past, however, different models for different process routes were required, resulting in high modeling effort in early development stages.

Here, we present a systematic approach for the development of models and control concepts applied to liquid-liquid extraction. The core of our highly efficient approach is to reuse models for different process routes: The unit operations of the different routes are split into sub-processes and similarities of these sub-processes across the different process routes are identified. Consequently, the effort in developing process models for different process routes is reduced significantly.

Based on that, the process models obtained are used to compare different control concepts on each process route and in between different process routes, resulting in a cost-benefit analysis of using simple control strategies against more sophisticated ones. This comparison uses defined criteria, including tolerable ranges for individual values of CQAs and the required efforts in including process analytical tools.

We demonstrate this approach by successfully applying it to two validated models of liquid-liquid extraction process routes. The results obtained form the basis for decisions on the process route for an industrial-scale real-world process. The models can be used as a digital twin and be combined to capture more complex process lines. Recommendations for process control and required real-time measurements are provided by the demonstrated approach.