About this Abstract |
Meeting |
2023 TMS Annual Meeting & Exhibition
|
Symposium
|
AI/Data Informatics: Computational Model Development, Validation, and Uncertainty Quantification
|
Presentation Title |
Large-scale Search of High-strength Aluminum Alloys at High Temperature Using Bayesian Learning for Neural Networks |
Author(s) |
Ayami Hattori, Shimpei Takemoto, Takeshi Kaneshita, Kenji Nagata, Yoshishige Okuno, Junya Inoue, Manabu Enoki |
On-Site Speaker (Planned) |
Shimpei Takemoto |
Abstract Scope |
We build a neural network to predict mechanical properties of 2000 series aluminum alloys at arbitrary temperatures to design high-strength ones at high-temperature. In this study, we perform Bayesian learning for neural networks, which allows us to evaluate the probability and uncertainty of the mechanical property of prediction. The Bayesian inference of neural networks is performed using the replica-exchange Monte Carlo method, one of the extended Markov chain Monte Carlo methods.
We also construct a Bayesian inverse design program of high-strength aluminum alloys at high temperatures. The program enables fast, large-scale exploration of design conditions for high-strength aluminum alloys. We also perform a thermodynamic analysis with the Thermo-Calc software to discuss the validity and strengthening mechanism of the alloy designs proposed by the neural network.
The program we have developed will be implemented in the MInt (Materials Integration by Network Technology) system as part of the Materials Integration Consortium. |
Proceedings Inclusion? |
Planned: |
Keywords |
Machine Learning, Aluminum, |