pandas

NumPy

SciPy

OpenCV

scikit-image

Matplotlib

seaborn

PyTorch

scikit-learn

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

In this project, we fabricated mixed-dimensional aerogels using a curated set of nano-scale building blocks to tune the properties that fabricated samples exhibited. The first step of our workflow was to determine which recipes would successfully form an aerogel. Our fabrication procedure involved taking an aqueous mixture, freezing it across a temperature gradient, and then drying it at low pressure evcuate any solvent. This fascilated the formation of a monolithic gel matrix. One criteria for success was whether the self-assembled matrix would able to freely stand and maintain its shape under both these low-pressure and ambient conditions. We therefore proposed a volume retention cutoff that determined sample "feasibility". By training an SVM classifier on the normalized volumes of sample compositions and mass-loadings, we could then predict the feasibility of recipes throughout our entire design space. This enables us to cut down our design space as we perform further, more expensive characterization, and avoid experimental failures.

By using a set of calibration cubes with differing heights and the same cross-section, I compute the camera matrix and solve for the heights and top areas of samples. By computing the height and cross section of each sample, we can calculate an estimate for its volume. This process enables us to take images on different days, across different batches, with consistent results. Finally, taking into account the normalized volume of each sample, by defining a feasibility cutoff (e.g. 80%) we significantly reduced the "viable" region of our design space, decreasing the number of experiments required at later stages.