A Practical Evaluation of Qualitative and Quantitative Chemometric Models for Real-Time Monitoring of Moisture Content in a Fluidised Bed Dryer Using near Infrared Technology

Abstract

Qualitative and quantitative chemometric models were evaluated to monitor moisture content of a wet granulation in a fluidised bed dryer using near infrared (NIR) technology. A principal component analysis (PCA) model was evaluated to obtain qualitative information. Multiple linear regression (MLR), principal component regression (PCR), partial least squares regression (PLS-R) and support vector machine regression (SVM-R) were evaluated using The Unscrambler® X. The PLS-R method was selected to demonstrate real-time monitoring of the moisture content. An ABB FT-NIR spectrometer with a Galileo direct-contact fibre-optic diffuse reflectance probe was used for NIR measurements. The Unscrambler® X Process Pulse was employed to upload the PLS-R model in measuring real-time moisture content. The PCA model successfully projected test batch data for the process signature and the PLS-R model successfully predicted (root mean square error of calibration: 0.5799 and R²: 0.9898; root mean square error of cross validation: 0.6595 and R²: 0.9868) the moisture content of the granulation during fluidised bed drying. Future work includes implementing the developed models for routine manufacture of drug product at commercial scales. Real-time determination of the end point for loss on drying (LOD) will eliminate the conventional, off-line LOD measurements currently in practice.