Serial artificial neural networks characterized by Gaussian mixture for the modelling of the Consigma25 continuous manufacturing line

Abstract

In this research, the Consigma25 Continuous Manufacturing (CM) Line is statistically analysed and modelled. First, the main effects plot is employed to examine the effects of different process parameters on the granules size and the tablet strength. Second, a modelling framework based on serial interconnected artificial neural networks is proposed to model the CM line by mapping these parameters to the granules size and the tablet strength. Then, Gaussian mixture models (GMMs) are adopted to characterize the error resulting from these networks in a way that helps in extracting more information and, as a result, improves the performance of the modelling framework. Validated on an experimental data set, the proposed interconnected framework can anticipate the characteristics of the granules and tablets produced using a specific blend of excipients with an absolute error percentage value of less than 12.3%. In addition, the GMMs have improved the predictive performance by 9.7%.