Image-based simultaneous particle size distribution and concentration measurement of powder blend components with deep learning and machine vision

This work presents a system, where deep learning was used on images captured with a digital camera to simultaneously determine the API concentration and the particle size distribution (PSD) of two components of a powder blend. The blend consisted of acetylsalicylic acid (ASA) and calcium hydrogen phosphate (CHP), and the predicted API concentration was found corresponding with the HPLC measurements. The PSDs determined with the method corresponded with those measured with laser diffraction particle size analysis. This novel method provides fast and simple measurements and could be suitable for detecting segregation in the powder. By examining the powders discharged from a batch blender, the API concentrations at the top and bottom of the container could be measured, yielding information about the adequacy of the blending and improving the quality control of the manufacturing process.

Development of a Near-Infrared Spectroscopy (NIRS)-Based Characterization Approach for Inherent Powder Blend Heterogeneity in Direct Compression Formulations

With the advent of continuous direct compression (CDC) process, it becomes increasingly desirable to characterize inherent powder blend heterogeneity at a small batch scale for a robust and CDC-amenable formulation. To accomplish this goal, a near infrared spectroscopy (NIRS)-based characterization approach was developed and implemented on multiple direct compression (DC) blends in this study, with the intended purpose of complementing existing formulation development tools and enabling to build an early CMC data package for late-phased process analytical technology (PAT) method development. Three fumaric acid DC blends, designed to harbor varied degrees of inherent blend heterogeneity, were employed. Near infrared spectral data were collected on a kg-scale batch blender via both time- and angle-based triggering modes. The time-triggered data were used to investigate the blending heterogeneity with respect to rotation angles, while the angle-triggered data were used to provide blending variability characterization and compare against off-line HPLC-based results. The time-triggered data revealed that the greatest blend variability was observed between revolutions, while the blending variability within a single revolution stayed relatively low with respect to rotation angles. This confirmed earlier literature findings that the bottom layer of powder blends tends to move with the blender within each revolution, and the most intense powder mixing takes place across revolutions. This also indicates the use of blending speed and the number of co-adds are not able to increase sampling volume to improve signal-to-noise ratio under a tumble-bin blender as what were typically done in a feedframe application. The angle-triggered data showed that there is a consistent trend between NIRS and HPLC-based methods on characterizing blend heterogeneity across the blends at a given sample size. This study contributes to establishing NIRS as a potential characterization approach for inherent powder blend heterogeneity for early R&D. It also highlights the promise of continuous characterization of inherent powder blend heterogeneity from gram scale to mini-batch CDC scale.