Multivariate analysis for product design and control

Abstract

Abstract Fluidised bed coating of pellets, granules or particles plays a very important role in the development and production of oral controlled release formulations. Based on design of experiments for robustness testing of the manufacturing process, a 2 4−1 fractional factorial was planned. During the experiments in-line NIR spectroscopy was combined with a theoretical film growth model in order to monitor and control the coating process. The increase in film thickness during the process evolvement was calculated and compared with the film thickness obtained by a reference method off-line, image analysis. NIR spectra were aquired during coating by means of a fiber optic probe positioned in the fluid bed. Time series of NIR spectra were calibrated with the corresponding theoretical film thickness by a multivariate analysis method, PLS (partial least square) regression. Film thickness predicted by the PLS model was then compared with the actual median thickness, as measured by image analysis off-line. The calibration of in-line NIR-spectra to film thickness calculated with a growth model is successful with PLS. The PLS model contains three principal components, which describe 99% of the variance in NIR-spectra and the variation in film thickness calculated by the growth model. The root mean square error (RMSE) between the film thickness predicted by NIR-spectra and the film thickness calculated by the theoretical growth model is 1.2 μm.