About this Abstract |
Meeting |
2024 TMS Annual Meeting & Exhibition
|
Symposium
|
Algorithm Development in Materials Science and Engineering
|
Presentation Title |
Inverse Problem Analysis of Phase Fraction Prediction in Aluminum Alloys Using Differentiable Deep Learning Models |
Author(s) |
Yu Okano, Takeshi Kaneshita, Shimpei Takemoto, Yoshishige Okuno |
On-Site Speaker (Planned) |
Yu Okano |
Abstract Scope |
Nowadays, efficient alloy development is being pursued, and there is a need for faster calculations to optimize alloy properties. Thermo-Calc based on the CALPHAD method is widely used to calculate thermodynamic states affecting alloy properties.
For this reason, we have developed a deep learning model that provides fast and accurate prediction of temperature variation of equilibrium phase fractions calculated by Thermo-Calc. The architecture of this deep learning model is based on Transformer, which is being used in Natural Language Processing tasks.
This deep learning model can calculate the 6000 series of industrial aluminum alloys (Al-Mg-Si based alloy) more than 100 times faster than the calculation speed of Thermo-Calc. Furthermore, by utilizing the back propagation of the model, we developed a method to search the material composition for the desired phase fraction in an inverse problem, and the results of the material search are presented in this presentation. |
Proceedings Inclusion? |
Planned: |
Keywords |
Aluminum, Machine Learning, Computational Materials Science & Engineering |