Mechanistic approach in powder blending PAT: Bi-layer mixing and asymptotic end point prediction

We derive a simple first order kinetics model for the powder mixing in a bin blender and adapt it to fit the NIR spectra obtained during the real-time mixing. The model is tested on a full drug product formulation mixing consisting of API and multiple excipients. It allows to extract a quantitative mixing rate for individual ingredients, which may be used in the drug development and production scale-up design. We further propose an algorithm for a calibration-free blending end-point determination from the NIR data, which allows for a robust and easily transferable blending process control. [Display omitted] •A new physical mixing model is derived for a bin blender.•The model fits well the acquired NIR real-time blending data.•The model defines a quantitative mixing coefficient for each ingredient.•The model allows to set a robust and universal blending end point criteria.•The methodology is perfectly transferable across production lines and products.