Investigation of droplet dynamics in the hypermonotectic succinonitrile-water system in a temperature gradient and microgravity conditions supported by deep learning computer vision.

We have investigated the liquid-liquid phase separation from a homogeneous liquid phase and the subsequent liquid droplet phase evolution upon cooling, using in-situ observation. The transparent monotectic material succinonitrile-0.82 wt.-frac. water was used, in which succinonitrile-rich droplets with a slightly higher density than the liquid matrix form. They move to warmer regions in the sample due to thermocapillary forces in a temperature gradient. To exclude additional gravity effects on droplet movement and on thermosolutal convection, the experiment was carried out in six minutes of microgravity on the TEXUS-60 sounding rocket. A pre-trained deep-learning computer vision model based on mask R-CNN was used for droplet detection in microscopic images. The multi-object tracking model SORT was applied for statistical droplet evaluation from the detected objects. The moving droplets showed only slight deformation from spherical shape. Some droplets adhered to the boundary glass windows of the experimental container, and increased wetting behaviour and coagulation was observed at later stages of the experiment. Droplets approaching the container boundaries slow down their movement. Largely isolated and spherical droplets moving and dissolving in a slowly increasing temperature gradient follow the steady-state velocity model of Young et al. from 1959 very well, except for an offset from zero for vanishing droplet diameters. The model application to the experimental data reveals a temperature dependence of the surface tension of dσ/dT = - 0.99 ± 0.08 10-4 Jm-2K-1.