Advanced image analytics to study powder mixing in a novel laboratory scale agitated filter dryer
Powder mixing unit operations are essential to the manufacturing of pharmaceutical drug substance and drug products. To model powder mixing, computational approaches have been used to evaluate mixing indices for different types of mixers, while also necessitating experimental methods for tracking of particle dynamics for validation. In this work, the experimental mixing performance of two particle species of different particle size and densities in a novel laboratory agitator filter dryer (AFD) was evaluated by performing 3D X-ray microCT (μCT) analysis. A new workflow, in which the μCT images were pre-processed with image analytics of increasing complexity and fidelity from image thresholding to advanced AI-based image segmentation and 3D reconstruction, was demonstrated to derive powder mixing indices. The AFD device which enables a customizable laboratory equipment for scale-down experimentation of agitated drying was shown to be capable of achieving a uniform powder mixture with micromixed cluster domain size.
Powder mixing unit operations are essential to the manufacturing of pharmaceutical drug substance and drug products. To achieve drug product quality, a well-mixed formulation containing uniform distribution of the active pharmaceutical ingredient (API) and excipients is key to ensure blend uniformity, as a prerequisite for further processing into the final dosage form with the desired dose and weight. The importance of a uniform powder mixture is further exemplified with low drug loading (low dose) formulations, wherein the active ingredient could account for <0.5% of the blend. In continuous direct compression manufacturing process, blend uniformity at the blender outlet is critical to the content uniformity of the tablets .
Furthermore, the drug substance manufacturing steps may also involve powder mixing unit operations, such as agitated or fluidized drying , acoustic mixing , or dry coating . In agitated drying, a semi-dry powder bed of the active ingredient is mixed intermittently to re-distribute localized solvent and to enhance heat and mass transfer for drying effectiveness, such that localized unbound solvents can be removed to reach target solvent acceptance criterion [, , ]. Pharmaceutical dry coating applications could involve a concomitant process of micronization and surface coating of the API to create an interactive powder mixture, or utilization of a screening mill to disperse low amount of guest (glidant) particles onto API surfaces as part of the delumping process to create a predominantly drug substance blend with superior flow and dissolution properties [8,9]. The performance of the intermediate blend generated from such processes are heavily dependent on the distribution of host and guest particles in the mixture.
Mixing can be classified into ordered, random or interactive mixing [10,11]. The type of powder mixtures is associated with intrinsic material properties such particle size, density, shape, surface and mechanical properties of the particles, and also extrinsic constraints such as compositions, process volume, mixing time, intensity and mode. To define the degree of mixing, numerous mixing indices have been proposed, and can be classified into sample variance-based, contact-based and distance-based methods . Sample variance mixing indices include Lacey Index , Kramer index , entropy of mixing , variance reduction ratio, amongst many others. Contact-based variance methods such segregation index and coordinate number  requires determination of particle contacts within the system, whereas distance-based methods such as Siiria method index take into account of particle location and distance. A review of mixing indices is detailed elsewhere [12,16].
The expanded use of computational approaches such as discrete element method (DEM) or continuum method (CM) to model mixing behavior and mixing indices of granular matters for different types of mixers necessitates experimental methods for real-time tracking of particle motion alongside positional data for validation [, , ]. Specifically, experimental mixing studies using velocimetric techniques such as positron emission particle tracking (PEPT) and X-ray stereography; tomographic methods such as nuclear magnetic resonance imaging (MRI) and x-ray computed tomography (X-ray CT); spectroscopic methods such as NIR and Raman, have been reviewed . X-ray imaging is an attractive, non-intrusive technique capable of achieving a broad range of spatial resolutions, sample size requirements and field of view (FOV). X-ray CT allows for multiple 2D projections of the object to be acquired in a manner such that individual cross-sectional slices and the entire 3D structure can be reconstructed. MicroCT (μCT) systems are capable of focal spot size within a few μm with achievable spatial resolution in range of 5–50 μm. Higher resolution images can be achieved with nanofocus systems (nanoCT, submicronCT)  or using synchrotron-based X-ray imaging (SR-μCT) to reach submicrometric spot size [22,23]. X-ray CT has been commonly utilized to characterize internal features including microstructure, particle distribution and local porosity based on its sensitivity to density differences within the samples.
Mixing index can be calculated by leveraging suitable image analytics to elucidate the spatial distribution of particles within the powder system. Image analytics can identify the spatial distribution of particles based on the contrast in the colors of different particle types and this information can be used to estimate the degree of mixing between the particles. This approach can be followed for images at time points to get a temporal evolution of the mixing index. With more complex system containing multiple phases, AI-based image processing can allow textural pattern signature for the phases to be learnt from a human analyst through iterative training on multiple images, thereby offering robust and quantitative analytics of complex multiphase systems .
In this work, the mixing performance of two particle species in a novel laboratory device was evaluated by 1) 3D X-ray μCT analysis of the powder system as a function of impeller revolutions, and 2) pre-processing of the 3D X-ray μCT images with methods of increasing fidelity from basic binary image segmentation, to automatic image thresholding, to advanced AI-based image segmentation and 3D reconstruction, and 3) applying different mathematical approaches for the experimental determination of powder mixing indices. This work aims to demonstrate that particle mixing can be successfully investigated through X-ray μCT with approaches of varying degree of complexity, while yielding important insights in the mixing dynamics.
The mixing device is a novel laboratory-scale agitated filter dryer (AFD), which was developed in collaboration with the University of Leeds and Freeman Technology, as an auxiliary accessory for use with the FT4 Powder Rheometer® . The AFD module provides essential features of an AFD, i.e. interchangeable impeller blade design, ability to precisely regulate rotational velocity, accurate measurement of torque, as well as filtration, heating, cooling and monitoring capability. It enables a customizable laboratory equipment for scale-down experimentation that could mimic experimental conditions for scale-up of agitated drying. Previously, the novel device was demonstrated in a particle breakage study . Particle mixing study is important for the design of the AFD protocol because it can directly impact particle attrition or agglomeration [7,25,26]. Herein, the experimental characterization of the mixing characteristics is also reported.
Anhydrous dibasic calcium phosphate (DCP) (DI-CAFOS A60, Chemische Fabrik Budenheim KG, Budenheim, Germany) and d-mannitol (Pearlitol 400 DC, Roquette, Keokuk, USA) were used as the binary mixing components.
Raimundo Ho, Yujin Shin, Shawn Zhang, Aiden Zhu, Prashant Kumar, Himanshu Goyal, Advanced image analytics to study powder mixing in a novel laboratory scale agitated filter dryer, Powder Technology, 2023, 118273, ISSN 0032-5910,