A spectroscopic data-sparing modeling framework for crystallization composition prediction

Monitoring the dynamics of the crystallization process is critical in controlling the quality of the product, such as particle size distribution, crystal morphology and purity. Despite its popularity, current process analytical techniques for crystallization monitoring still suffer from many challenges, such as the need for large amount of dataset for calibration. Here we propose a Gaussian deconvolution-based quantification approach that leverages physics understanding of the system and only requires a minimal amount of experimental data to predict the composition of a multi-component solution. Using a model system representative of an antisolvent/cooling crystallization process, we demonstrate the proposed approach can lead to a simple linear model, with comparable prediction accuracy to conventional process analytical models, using a reduced amount of data Given the data-driven nature and the easiness in the setup, we anticipate the proposed approach to benefit a broad range of crystallization systems.