Development of feed factor prediction models for loss-in-weight powder feeders

Abstract

In this paper, data-driven predictive models are proposed for pharmaceutical powders fed through a loss-in-weight (LiW) feeder. First, material-specific partial least squares (PLS) regression models are developed for each excipient, including Lactose Anhydrous, Magnesium Stearate, Croscarmellose Sodium, and Microcrystalline Cellulose. It is demonstrated that, using only feeder configuration data, these material-specific models can accurately predict the feed factor profile. Furthermore, principal component analysis (PCA) is performed to group the excipients and Acetaminophen (APAP) grades into clusters with similar material properties. Then, for each cluster, a generic PLS model is developed which uses feeder configuration data and powder properties to predict the feed factor profile for a broad range of materials. The accuracy of these generic models is demonstrated through validation on new/unseen grades of powder materials. Finally, a workflow using the developed models is proposed to speed up characterisation of drug manufacturing process whilst minimising powder consumption and experimental effort.