|About this Abstract
||2019 TMS Annual Meeting & Exhibition
||Computational Approaches for Big Data, Artificial Intelligence and Uncertainty Quantification in Computational Materials Science
||Bayesian CALPHAD: From Uncertainty Quantification to Model Fusion
||Pejman Honarmandi, Thien Duong, Seyede Fatemeh Ghoreishi, Douglas Allaire , Raymundo Arroyave
|On-Site Speaker (Planned)
Uncertainty quantification in CALPHAD is one of the most important steps in materials design under the ICME framework. As a manner of illustration, thorough parameter uncertainty analyses of four Hf-Si CALPHAD models are performed against the available data through a Bayesian Markov Chain Monte Carlo technique. The resulting parameter uncertainties are subsequently propagated to the phase diagrams. Since the resulting phase diagrams are biased towards the optimized parameter values obtained from ThermoCalc which may not be the global in high dimensional parameter spaces, finding a CALPHAD model with an appropriate number of parameters is very important that is usually costly. In this regard, A Bayesian hypothesis testing is proposed to fulfill the need of model selection. However, Bayesian model averaging and an error correlation-based model fusion are proposed here to fuse all of the given CALPHAD models and obtain a fused model instead of just considering the best selected model.
||Planned: Supplemental Proceedings volume