pandas

NumPy

SciPy

OpenCV

scikit-image

Matplotlib

seaborn

diffusers

scikit-learn

Predictive and Generative Modeling of Mixed-Dimensional Aerogels with Programmable Properties

This tool led to a major turning point in our project. Previously, our lab had published a paper in which from an SEM image of our sample, we extracted a single representative pore size to use in a tiled, brick-like pattern, to perform FE simulations. This was a crude representation of the intricate microstructure that our samples possessed. I proposed that, instead of these bricks, we instead extract a mesh of our sample's microstructures, and use this to perform the simulations. Not only does this method better preserve the individuality of our samples, but it also preserves the curvature and distribution of shapes and sizes that it contains. A crucial step towards implementing this was to install a periodic behavior to our patches. To this end I utilized an inpainting model and some clever augmentations to generate repeatable tiles. Additionally, I could build tiles from cropped regions of many different images, improving the diversity of any given patch.

Using these meshes, I have worked with collaboraters in the Materials Engineering and Mechanical Engineering Departments to build FE models to simulate Mechanical (deformation), Thermal, and Acoustic charcteristics of our samples. These models allow us to characterize virtual samples to a high degree of accuracy, while increasing our throughput by 1000+ fold.