When utilizing large models containing numerous uncertain parameters, model calibration becomes a critical step in the analysis. Traditional methods of calibration involve adjusting uncertain parameters based on expert opinion or best estimates. While this traditional calibration may lead to better model predictions, it usually only yields better estimates for certain specific conditions. This drastically reduces the functionality of the model in question. Bayesian calibration is an alternative to traditional calibration methods which utilizes available information (simulation results and/or real world measured values) to iteratively refine uncertain parameters (either assumed or measured uncertainty) while considering not only parametric uncertainty, but also model, observational, and residual uncertainties at every step of the calibration process. Various methods have been previously developed for executing the calibration process, several of which can drastically reduce the computational expense of Bayesian calibration. A numerical example is included to illustrate the functionality of Bayesian calibration.
Bayesian Large Model Calibration Using Simulation and Measured Data for Improved Predictions
Sae International Journal of Passenger Cars. Mechanical Systems
Sae Int. J. Passeng. Cars - Mech. Syst
SAE 2015 World Congress & Exhibition ; 2015
Sae International Journal of Passenger Cars - Mechanical Systems ; 8 , 2 ; 415-420
2015-04-14
6 pages
Conference paper
English
Improved Predictions of Flight Delays Using LMINET2 System-Wide Simulation Model
British Library Conference Proceedings | 2009
|Bayesian Calibration of the QASPR Simulation
AIAA | 2007
|Convergent Multifidelity Optimization Using Bayesian Model Calibration
British Library Conference Proceedings | 2010
|