| About this Abstract |
| Meeting |
2020 TMS Annual Meeting & Exhibition
|
| Symposium
|
Expanding the Boundaries of Materials Science: Unconventional Collaborations
|
| Presentation Title |
Regularization of Materials Failure Data for Damage Mechanism Categorization by Machine Learning |
| Author(s) |
John J. Hasier, Keo-Yuan Wu, Rachel Wittman |
| On-Site Speaker (Planned) |
John J. Hasier |
| Abstract Scope |
Material failures in power generation are messy, expensive events that generate large amounts of variable quality data. Machine learning classification of failure mechanisms can enable a move from material failure correction, currently requiring multidisciplinary teams of experts, to failure prevention. A machine learning driven expertise engine for operators and engineers can provide just-in-time knowledge to assist in critical decisions. Regularization of data is a critical and often trivialized step for the application of modern machine learning techniques to solve domain specific problems, as machine learning techniques fail to provide useful insights if the data fed to them is not well curated. This talk uses the creation of a training dataset based on decades of power generation failure analysis reporting as a vehicle to explore the challenges encountered by materials engineers and data scientists in sculpting real-world material science data into a useful input for modern data analytics techniques. |
| Proceedings Inclusion? |
Planned: Supplemental Proceedings volume |