The simultaneous control of crystal size and shape is particularly important in fine chemical and pharmaceutical crystallization. These two quantities influence the dissolution rate and bioavailability of final drug products, and also contribute to the manufacturability and efficiency of downstream operations. Numerous practical issues are associated with the implementation of an industrially relevant crystal size and shape control algorithm, such as the limited number of commercially available measurement devices and tight productivity constraints. Model-based control algorithm is required to address these problems, which, however, brings an additional challenge: the high computational demand of the applicable process simulation. In this work we show that the model predictive control, relying on measurements coming from routinely applied, commercially available process analytical technology (PAT) tools may be feasible. The algorithm is based on GPU accelerated full 2D population balance model (PBM) solution, which does not require external computation units (i.e. cloud computing). The state estimation, which is a crucial part of any robust model-based control system, is carried out by fitting the model, through the re-adjustment of some model parameters, on the measured FBRM count, PVM based mean aspect ratio (AR) and solute concentration, in real time. Artificial neural network (ANN) based soft-sensor is employed to simulate the most likely mean AR of the bivariate crystal size distribution (2D CSD) calculated by the PBM, which is compared then to the mean AR measured by the in-situ imaging tool, as a part of real-time parameter re-adjustment.

• Purdue University, Department of Chemical Engineering

• Control

1. Control
2. crystallization
3. crystal shape
4. Model predictive control
5. Real time control
6. Review article