About this Abstract |
| Meeting |
2027 TMS Annual Meeting & Exhibition
|
| Symposium
|
Verification, Calibration, and Validation Approaches for Mechanical Modeling of Metals
|
| Presentation Title |
Corrected high speed nanoindentation mapping properties using mechanical-crystallograpic latent space |
| Author(s) |
Damien Texier, Pierre Belamri, Henry PROUDHON, David RYCKELYNCK, Quentin SIRVIN |
| On-Site Speaker (Planned) |
Damien Texier |
| Abstract Scope |
High-speed nanoindentation mapping enables precise characterization of elastic anisotropy and hardness at microstructural scales, but tip erosion during repeated testing distorts contact area measurements—compromising mechanical property accuracy.
To address this, we propose a self-supervised multimodal framework leveraging a quaternion Vision Transformer (pretrained on EBSD data) integrated into a Joint-Embedding Predictive Architecture (JEPA). This model extracts latent representations of nanoindentation curves while accounting for local crystallographic context—without requiring labeled hardness or modulus data. Trained on 90,000 Inconel 718 measurements, it reveals that tip blunting is encoded as interpretable structure, not noise.
The learned latent space decodes acquisition order via a sub-linear power-law wear signal, enabling continuous correction of contact-area inflation. Validation across >800,000 indents confirms the model’s scalability—extrapolating erosion trends beyond training data—while independent calibration validates progressive tip degradation. This approach transforms latent-space insights into actionable corrections for hardness and modulus mapping. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Characterization, Machine Learning, Mechanical Properties |