A model was developed for predicting the feed factor profile of a powder, processed through a gravimetric feeder, as a function of material properties and process parameters. Predictive models proposed in existing literature have often used excipients and active pharmaceutical ingredients (APIs) with good powder flow characteristics in their development. In this work, a material properties library containing a large proportion of APIs, as well as excipients and co-processed blends, was used to build the model and enhance the prediction of feed factor profile for cohesive powders. Gravimetric feeder trials were performed at varying mass flow rates and screw geometries to determine the feed factor profiles. A semi-empirical exponential model, with parameters fmax, fmin, and β, was then used to fit the experimental feed factor profiles. Bayesian optimisation and Support Vector Regression (SVR) modelling techniques were utilised to optimise and predict the exponential model parameters as a function of material properties. The parameters found to strongly influence the model were particle size, bulk density, FFC and FT4 rheometer parameters. Results showed low prediction errors between the estimated and experimental data. The final model produces good estimations of the feed factor profile and requires minimal powder consumption.
Keywords: Continuous manufacturing; Data driven model; Flow prediction; Gravimetric feeding; Materials property library.
Copyright © 2022 Elsevier B.V. All rights reserved.