About this Abstract |
Meeting |
MS&T25: Materials Science & Technology
|
Symposium
|
Applications of Uncertainty Quantification (UQ) in Science and Engineering
|
Presentation Title |
Uncertainty Quantification via Deep Kernel Learning for Predicting Multimodal β-phase Volume Fraction from SXRD Patterns |
Author(s) |
Ayorinde Emmanuel Olatunde, Ozan Dernek, Gabriel Ponon, Weiqi Yue, Qingzhe Guo, Amit Samantha, Donald W. Brown, Roger H. French, Pawan K. Tripathi, Anirban Mondal |
On-Site Speaker (Planned) |
Ayorinde Emmanuel Olatunde |
Abstract Scope |
The processes of obtaining scientific experimental data and statistical modelling are prone to significant errors, which are reducible but not avoidable. There are various approaches for quantifying these uncertainties, but the methods used in these quantifications also have their limitations. One of such limitations is finding the estimates used for prediction with standard kernels and not the exact kernel of the data set.
In this work, we extend our ongoing efforts by incorporating deep kernel learning (DKL) before modelling with the Gaussian Process (GP) used in quantifying the uncertainties in the prediction of the multimodal β-phase volume fraction from synchrotron X-ray diffraction patterns obtained from Ti–6Al–4V alloy during heat treatment.
DKL allows us to learn the kernel of our data, thereby modelling the GP with the learned kernel with for potential better performance. We will compare our results with those obtained with standard base kernels for benchmarking. |