About this Abstract |
| Meeting |
2027 TMS Annual Meeting & Exhibition
|
| Symposium
|
Microstructure-Sensitive Design and Advanced Characterization: An MPMD/SMD Symposium Honoring David T. Fullwood
|
| Presentation Title |
A Journey in Quantifying and Mapping Microstructure: From Correlation Functions to Machine Learning Manifolds |
| Author(s) |
Stephen R. Niezgoda, Simon Mason, Dennis Dimiduk, Megna Shah, Jeff Simmons |
| On-Site Speaker (Planned) |
Stephen R. Niezgoda |
| Abstract Scope |
For many years, David Fullwood and I have worked on transforming microstructure images into quantitative information that can guide materials design. This work extends that long-running conversation by treating microstructure as a stochastic material state represented by an ensemble of realizations rather than a single image. Using phase-field simulations of spinodal decomposition, we evaluate a range of microstructure statistics for their ability to recover intrinsic process dimensionality, preserve neighborhood relationships, and support inversion to processing conditions. The resulting low-dimensional manifold provides both a quantitative geometry for comparing material states and a latent space for generative design. We use this representation to train a semantically aware diffusion model that generates realizable microstructures while enabling controlled navigation along physically meaningful attributes. A metric-space Kolmogorov–Smirnov test further quantifies whether two microstructure ensembles represent statistically distinguishable states, linking statistical representation, generative artificial intelligence, and closed-loop materials exploration for accelerated materials discovery and qualification. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Machine Learning, ICME, Other |