Many industrial processes involve non-spherical particles, yet numerical investigations often oversimplify by assuming a spherical shape. This leads to errors in drag force calculations and hampers engineering design solutions. To address this, the goal of our project is to develop an AI-driven platform to automatically classify particles into different shape and size classes and extract the corresponding orientation tendencies. Both information can enhance drag coefficient models for non- spherical particles. Synthetic images will be used to overcome the need for a large labeled dataset and to investigate automatic labeling strategies, especially in the presence of a domain shift.