The aim of this work was to develop a PAT platform consisting of four complementary instruments for the characterization of electrospun amorphous solid dispersions with meloxicam. The investigated methods, namely NIR spectroscopy, Raman spectroscopy, Colorimetry and Image analysis were tested and compared considering the ability to quantify the active pharmaceutical ingredient and to detect production errors reflected in inhomogeneous deposition of fibers. Based on individual performance the calculated RMSEP values ranged between 0.654% and 2.292%. Mid-level data fusion consisting of data compression through latent variables and application of ANN for regression purposes proved efficient, yielding an RMSEP value of 0.153%. Under these conditions the model could be validated accordingly on the full calibration range. The complementarity of the PAT tools, demonstrated from the perspective of captured variability and outlier detection ability, contributed to model performance enhancement through data fusion. To the best of the author's knowledge, this is the first application of data fusion in the field of PAT for efficient handling of big-analytical-data provided by high-throughput instruments.
Keywords: Artificial neural networks; Data fusion; Electrospinning; Process Analytical Technology; Vibrational spectroscopy.
Copyright © 2019 Elsevier B.V. All rights reserved.