Is it possible to quantify the flow without adding any polymeric beads? I'm interested to know whether there's dominant capillary or Marangoni flow. Aby suggestions would really be helpful.
If specific patterns are anticipated based on the prevalence of either Marangoni or capillary flows, for example:
Then, it's plausible that an AI system could prove instrumental in identifying the existence of such patterns, consequently determining the dominant flow, ideally in an unsupervised manner. Sufficient examples would be needed to train the model, though.
Also, capturing the drying process from multiple perspectives (similar to a time-lapse video) - including calculations on contact angles - could offer insights at the surface level of the droplet. This data could potentially establish correlations with the internal dynamics of the droplet (I'm not an expert in microfluidics, I am just imagining possibilities!). Developing 3D vision algorithms could be valuable here, for example, for reconstructing surface and surface normals of the drying droplet and obtaining the gradient of reconstructions along time could be of help in characterizing the drying process at the surface level, like telling the love story between the interfaces of water and air.
Thoughts?
Maria Olga Kokornaczyk
Yojana Jautzi P. Carreón
Jorge González-Gutierrez
This can be done using numerical simulation. In an experiment, you can measure the velocity field using tracers. But from these data, it will be difficult to understand what contribution the Marangoni flow makes and which relates to the capillary flow. This will require additional theoretical assessments. If you do not use tracers, then it is unclear how to measure the flow velocity.
Thanks, @Mario Castelán, for your thoughts. This can be done, but we might not have enough data in the literature just yet to train these. And also, with machine learning, we can predict the outcome but exactly can we obtain the quantitative data?
Hello, Anusuya Pal. Apologies for the late response! Indeed, data availability is a significant obstacle. However, the development of highly realistic synthetic data generation algorithms for microscopic images is advancing rapidly (I'm currently writing a paper on this topic for cooper coating micrographs). This progress is revolutionizing data augmentation techniques for deep learning applications.
Regarding quantitative prediction, achieving exact quantitative data may not always be feasible. Nonetheless, approximations are possible. For example, unsupervised deep learning can be used to cluster regions of interest (ROIs) in micrographs, which can then be characterized into low-dimensional vectors. Our recent papers with Maria Olga Kokornaczyk address these challenges.
Thanks, @Mario Castelán, for your response. I will go through the paper you mentioned.
- Badges
- Science method
- Similar topics
- Methodology
- Crystallography
- X-ray Diffraction