Considering inelasticity in the real-time monitoring of particle size for twin-screw granulation via acoustic emissions

Abstract

A recently developed process analytical technology (PAT) using artificial intelligence to form the framework of its model, combining frequency-domain acoustic emissions (AE) and elastic impact mechanics to accurately predict complex particle size distributions (PSD) in real-time. This model was modified in this study to give more accurate predictions for the more highly cohesive granules typical of pharmaceutical solid oral dosage formulations. AE spectra were collected from the granulated impacts of various formulations with ranging characteristics from largely elastic to highly inelastic collision responses. A viscoelastic (Hertzian spring-dashpot) and elastoplastic (Walton-Braun) contact force model were compared to understand how these different micro-mechanical approaches would affect the prediction accuracy of particle sizes relevant to granulation. Retraining the artificial intelligence model with the Walton-Braun transformation and a more comprehensive dataset of AE spectra spanning a broad range of granulated formulations showed the prediction error drop to as low as 2% compared to the original elastic version showing errors as large as 18.6% with representative formulations of the industry. The improved PAT shows good applicability to monitoring bimodal PSD that are typical of continuous twin-screw granulation.