About this Abstract |
Meeting |
2024 TMS Annual Meeting & Exhibition
|
Symposium
|
Algorithm Development in Materials Science and Engineering
|
Presentation Title |
Bayesian Interpretable Machine Learning of Yield Surface Models with Noisy Data |
Author(s) |
Donovan Birky, Nolan Strauss, Jacob Hochhalter |
On-Site Speaker (Planned) |
Donovan Birky |
Abstract Scope |
While component-scale material model parameters are often idealized as deterministic, materials exhibit a range of intrinsic variability, e.g., characteristic microscale defects, which implies a need for non-deterministic model parameters. Additionally, machine learning methods continually demonstrate improvements in model form accuracy through incorporation of increasingly complex mechanisms. We develop an interpretable machine learning method, genetic programming based symbolic regression (GPSR), to learn implicit models, e.g., yield surfaces, while quantifying intrinsic and extrinsic uncertainty in the training data. Preliminary results demonstrate that the Bayesian implementation mitigates overfitting to noisy data, improves interpretability, and accurately quantifies uncertainty in the modeled dataset. |
Proceedings Inclusion? |
Planned: |
Keywords |
Machine Learning, Computational Materials Science & Engineering, Modeling and Simulation |