| About this Abstract |
| Meeting |
2025 TMS Annual Meeting & Exhibition
|
| Symposium
|
High Performance Steels
|
| Presentation Title |
Nanoindentation and Machine Learning, it’s all About the Features! |
| Author(s) |
Claus O.W. Trost, Stanislav Žák, Megan J. Cordill |
| On-Site Speaker (Planned) |
Claus O.W. Trost |
| Abstract Scope |
Nanoindentation has evolved into a rapid mechanical property characterisation method, allowing for the collection of vast amounts of data. To condense the data, elastic modulus and hardness maps are plotted. If the constituents making up the material are different enough, their properties can easily be extracted. Otherwise, the data need to be deconvoluted. Therefore, unsupervised machine learning methods such as the K-means algorithm are often used. Usually, the indentation curves are condensed into elastic modulus and hardness, thereby losing information that the full P-h curve provides. In the reverse-nanoindentation problem, other features, such as the loading curvature, are extracted. These features are well established but have not yet been used in clustering data. The advantages of such features will be shown on a High-speed Steel dataset consisting of 3300 human-labelled indents. By using unsupervised and supervised (explainable)machine learning, it will be demonstrated that maps can be deconvoluted using these features. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Mechanical Properties, Iron and Steel, |