Automated system for the on-line monitoring of powder blending processes using near-infrared spectroscopy: Part II. Qualitative approaches to blend evaluation

Abstract

Near Infrared (NIR) spectroscopy is seen as a very powerful tool in a variety of applications involving powder characterisation. Here we deal with a typical pharmaceutical application of powder blend monitoring. A D-optimal experimental design is used to cover the 85–115% range of the target formulation which is comprised of the active component at 3.5% w/w, Microcrystalline Cellulose (Avicel PH102) at 62%, Dibasic Calcium Phosphate Anhydrous at 31.5%, Sodium Starch Glycolate (Explotab) at 2%, and 1% Magnesium Stearate. A miniature Flobin blender has been modified to enable the use of a fibre optic probe for on-line NIR spectral data collection. The experiments were successful in detecting spectral changes which eventually converged to constant variance. While the NIR spectrum of a powdered sample is rich in information which is representative of both the physical and chemical characteristics of the sample, it is at times difficult to select the appropriate mathematical treatments in order to extract the desired information. This article investigates several possible pre-treatments (including detrending (DT), standard normal variates (SNV), second derivatives, and the combination of SNV and DT) together with several ways in establishing blend homogeneity, which includes the running block standard deviation, dissimilarity calculations and principal components analysis (PCA). The focus of this work is to investigate qualitative tools of analysis for blend homogeneity determinations, while future work will focus on quantitative data interpretation.