A multivariate approach to predict the volumetric and gravimetric feeding behavior of a low feed rate feeder based on raw material properties

Abstract

In this study, the volumetric and gravimetric feeding behavior of 15 pharmaceutical powders on a low feed rate feeder was correlated with their material properties through a multivariate approach. The powders under investigation differ substantially in terms of material properties, making the selected powders representative for powders typically used in pharmaceutical manufacturing. The material properties were described by 25 material property descriptors, obtained from a rational selection of critical characterization techniques that provided maximal information with minimal characterization effort. From volumetric feeding experiments (i.e., powder feed rate not controlled), the maximum feeding capacity (maximum feed factor (FF_max)) and optimal hopper fill level at which the feeder should be refilled during gravimetric feeding (feed factor decay (FF_decay)) were obtained. During gravimetric feeding experiments (i.e., powder feed rate controlled), the variability on the feed rate (relative standard deviation (RSD)) and the difference between the setpoint and mean feed rate (relative error (RE)) were determined. Partial least squares (PLS) regression was applied to correlate the volumetric and gravimetric feeding responses (Y) with the material property descriptors (X). The predictive ability of the developed PLS models was assessed by predicting the feeding responses of two new powders (i.e., validation set). Overall, the volumetric feeding responses (FF_max and FF_decay) were predicted better than the gravimetric feeding responses (RSD and RE), since in gravimetric mode the impact of material properties on the feeding behavior is reduced due to the control system of the feeder. Especially RE was weakly correlated with material properties as RE of most powders varied around zero with only a small numerical variation. Interestingly, this confirms that the control system is working properly and that the feeder is capable of feeding different powders accurately at low feed rates. The developed models allowed to predict the feeding behavior of new powders based on their material properties. Consequently the number of feeding experiments during process development can be greatly reduced, thereby leading to a more efficient and faster development of new drug products.

Keywords: Continuous manufacturing; Material characterization; Material properties; Multivariate data analysis; Twin screw feeding.