About this Abstract |
| Meeting |
MS&T26: Materials Science & Technology
|
| Symposium
|
Uncertainty Quantification in Ultra-High Temperature Materials Manufacturing
|
| Presentation Title |
Uncertainty Quantification via Deep Kernel Learning on Synchrotron Diffraction Patterns |
| Author(s) |
Ayorinde Emmanuel Olatunde, Ozan Dernek, Gabriel O. Ponon, Weiqi Yue, Qingzhe Guo, Oreofe Solarin, Amit Samanta, Donald W. Brown, Roger H. French, Anirban Mondal |
| On-Site Speaker (Planned) |
Ayorinde Emmanuel Olatunde |
| Abstract Scope |
The use of integrated methods for predictions made by Machine Learning models is increasing. Deep Kernel Learning (DKL) combines the strengths of principled Uncertainty Quantification (UQ) inherent in traditional UQ methods, such as Gaussian Process, with the merits of Deep Neural Networks for learning hierarchical representations from raw data, while incurring the demerit of increased computational complexity.In this work, we investigated the UQ capabilities of DKL models in the prediction of β-phase volume fraction from synchrotron X-ray diffraction patterns obtained from Ti–6Al–4V alloy during heat treatment. We considered two feature configurations: a full set with 4.1M raw pixel-intensity features and a reduced set with 262K downscaled pixel-intensity features. Using these two setups, we compared results across standard and study-defined metrics to determine whether the reduced space performs comparably to the full space and to draw relevant conclusions. |