About this Abstract |
| Meeting |
2027 TMS Annual Meeting & Exhibition
|
| Symposium
|
Verification, Calibration, and Validation Approaches for Mechanical Modeling of Metals
|
| Presentation Title |
Integrated Bias Calibration Leveraging Orthogonality and Shrinkage for Decoupled Uncertainty Quantification |
| Author(s) |
Liam A. Myhill, Enrique Martinez, Sez Russcher |
| On-Site Speaker (Planned) |
Liam A. Myhill |
| Abstract Scope |
Calibration of complex, inexact computer models has followed the framework posed by Kennedy and O’Hagan; whereby model-form error is considered as an additive discrepancy field acting in conjunction with a physics simulator and calibration parameters are provided as weakly informative prior distributions. We propose a philosophically dissimilar framework, which encodes both a model’s structural bias and parameter bias while preserving best estimate values for the model parameters. A Gaussian Process (GP) model is constructed per calibration parameter to capture parameter corrections as a function of the application domain, and an additive discrepancy GP captures the structural model bias. Our method evaluates simulator predictions at best-estimate inputs, and leverages shrinkage priors and orthogonality constraints to ensure the calibration parameter deviation from best estimates is supported by data. The governing philosophy is that best estimate values serve as an informative anchor, and any deviations that can be explained via parameter corrections should. |
| Proceedings Inclusion? |
Planned: |
| Keywords |
Modeling and Simulation, Computational Materials Science & Engineering, Other |