Inferential modeling in pharmaceutical crystallization

The main limitation to improving the online operation of crystallization processes for the past twenty years has been the lack of reliable and informative real-time sensors. Inferential models are developed to fill this need. The inferential modeling problem for these processes is especially challenging due to the low quantity of data relative to the dimensionality of the measurement vector that serves as the input to the inferential model. Quantifying prediction intervals is critical in these applications, which involves more sophisticated chemometric analysis techniques than those popular in the process control literature. The principles and techniques are illustrated through application to the batch crystallization of a pharmaceutical chemical