| About this Abstract |
| Meeting |
2023 TMS Annual Meeting & Exhibition
|
| Symposium
|
Advanced Characterization Techniques for Quantifying and Modeling Deformation
|
| Presentation Title |
Alloy Rupture Strength Prediction Using Machine Learning and Microstructure Analysis |
| Author(s) |
Ram Devanathan, Osman Mamun, Mohammad Taufique, William E. Frazier, Arun Sathanur, Keerti Kappagantula, Jing Wang, Marissa Masden, Madison Wenzlick, Kelly Rose |
| On-Site Speaker (Planned) |
Ram Devanathan |
| Abstract Scope |
We present a machine learning approach to predict the creep rupture strength of 9-12% Cr ferritic martensitic steels and austenitic stainless steels using curated experimental datasets. We evaluated three algorithms, namely Gaussian Process Regression, Neural Network, and Gradient Boosted Decision Tree (GBDT). We identified the most important features that govern the rupture strength for these two classes of alloys. The GBDT algorithm showed excellent predictive performance for unseen test data as shown by correlation coefficient better than 0.95 for both alloy datasets. To further improve the predictive power by including microstructural features, we have evaluated the potential of several deep learning models to identify grain boundaries in realistic steel microstructures We also used a topology-based loss function to improve recognition of grain boundaries, which is a challenging task. Our approach, when used with high quality training data, can reduce the need for a time consuming and expensive mechanical test campaign. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Characterization, Machine Learning, Mechanical Properties |