Physics-Informed Neural Networks

Under the mentorship of Postdoctoral Researcher Dr. Pankaj Rohilla, I worked on optimizing microneedle geometry using physics-informed machine learning. Microneedles are used for minimally-invasive drug delivery and diagnostics, and their performance depends heavily on precise geometric parameters that are expensive to explore with traditional finite-element simulation alone.

Physics-informed neural networks (PINNs) fold the governing physical equations directly into the network’s loss function, so the model is constrained to learn solutions that are physically consistent rather than purely fitting simulation or experimental data. This lets the network generalize across geometries with far fewer simulation runs than a standard finite-element sweep would require, making it practical to search a much larger design space for microneedle shape.