2025 / Project
PINN vs MMC Fastener Optimization
A comparative research implementation for fastener-placement optimization across differentiable-physics and topology-based methods.
This codebase compares two routes to fastener-placement optimization: a Physics-Informed Neural Network workflow built around differentiable finite-element kernels, and Moving Morphable Components topology optimization.
The implementation separates datasets, experiments, analysis tools, figures, CAD definitions, and thesis-style writing assets. Multi-geometry evaluation covers L-brackets, tapered plates, and ribbed channels under multiple loads, with utilities for unified comparisons, generalization tests, and Ansys-ready exports.
The current multi-geometry surrogate reports close agreement with differentiable FEA for the documented high-energy cases and sub-millisecond inference, but the repository also records its remaining work: larger datasets, external Ansys validation, and propagation of the latest evidence into the research narrative.
I include it as an ongoing computational research implementation, not as a completed or published thesis result. That status is part of the project: useful research software should make its unfinished validation boundary visible.