| About this Abstract |
| Meeting |
2024 TMS Annual Meeting & Exhibition
|
| Symposium
|
Frontiers of Materials Award Symposium: Physics-Informed Machine Learning for Modeling and Design of Materials and Manufacturing Processes
|
| Presentation Title |
Interpretability and Generalizability of Constitutive Models using Symbolic Regression |
| Author(s) |
Jacob D. Hochhalter, Karl Garbrecht, Donovan Birky, Nolan Strauss, Geoffrey Bomarito, Laurent Capolungo, John Emery |
| On-Site Speaker (Planned) |
Jacob D. Hochhalter |
| Abstract Scope |
Data-driven (i.e., empirical) methods for constitutive model development have recently been a focus of the materials research community due to the many promising developments within the machine learning community. However, it is also known that empirical constitutive models often lack (among several things) the ability to generalize across materials or processing methods, due in part to lacking rationale for the model form. Physics-informed machine learning methods can help regularize developed models to improve generalizability and promote known or expected characteristics of the model, e.g., monotonicity. However, such methods do not yet guarantee satisfaction of a particular constitutive model form that would be expected, based on past theoretical developments. We present results of symbolic regression, an inherently interpretable machine learning method, to learn microstructure-dependent material models for which rational model forms can be imposed with simultaneous stochastic parameter estimation. |
| Proceedings Inclusion? |
Planned: None Selected |