AI-aided visual data analysis for granular gases: Complex particles, high density, and other challenges

Microgravity experiments with three-dimensional (3D) granular gases, i.e., ensembles of freemoving macroscopic particles which collide inelastically, produce large amounts of stereo video footage which require processing and analysis. The main steps of data treatment are particle detection, 3D matching and tracking in stereoscopic views, and quantification of ensemble statistical properties such as, e.g. mean kinetic energy or collision processes. Frequent overlapping or clustering of particles and their complicated movement patterns require smart software solutions. In recent years, Artificial Intelligence/Machine Learning (AI/ML) methods were successfully used for analysis of granular systems. We have applied such techniques to the granular gases of rod-like particles and developed a software tool which enables a full cycle of semi-automatic experimental data analysis. The approach is now tested on more complex, non-convex particles, shaped as 3D crosses (hexapods). Another challenge is optical analysis of dense granular gases, where individual particles cannot be tracked. We present a preliminary result of application of an ML method for number density profiles extraction in VIP-Gran experiment with dense ensemble of rod-like particles.