Implementation of a Radial Basis Function control strategy for the crystallization of Ibuprofen under uncertainty

Batch crystallization is still one of the most commonly used processes in the pharmaceutical industry. Indeed, due to the flexibility of the process and the possibility of changing the product specification, the batch process is still the main choice. However, crystallization is a highly non-linear process and therefore difficult to model and simulate. Moreover, uncertainty in physical parameters and uncertainty propagation from previous process outputs has an important and critical influence on the process risk and therefore the control strategy needed to achieve robust process performance.

To this end, a two dimensional population balance model of ibuprofen batch crystallization is developed and used to analyse and study the operability space, where uncertainty is included. The uncertainty comes from previous synthesis steps and from reported uncertainty in the process parameters, described in the literature. A nonlinear control strategy is then applied, comparing different radial basis functions (quadratic, cubic, Gaussian…), in order to achieve the desired mean size by manipulating and updating the cooling profile of the same process.

The resulting three operation strategies are benchmarked with respect to key performance metrics (reference trajectory, deviation and input variation): open-loop, open-loop strategy under process uncertainties, and closed loop operation subject to process uncertainties. The final mean crystal size (MCS) is reported as a confidence interval, in order to be used for further downstream processing.