Premium
Bayesian hierarchical uncertainty quantification by structural equation modeling
Author(s) -
Jiang Xiaomo,
Mahadevan Sankaran
Publication year - 2009
Publication title -
international journal for numerical methods in engineering
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.421
H-Index - 168
eISSN - 1097-0207
pISSN - 0029-5981
DOI - 10.1002/nme.2550
Subject(s) - structural equation modeling , bayesian hierarchical modeling , bayesian probability , bayesian inference , uncertainty quantification , computer science , bayesian network , hierarchical database model , latent variable , inference , bayes factor , bayes' theorem , multilevel model , bayesian average , reliability (semiconductor) , data mining , algorithm , machine learning , artificial intelligence , power (physics) , physics , quantum mechanics
Uncertainty quantification is playing an increasingly important role in assessing the performance, safety, and reliability of complex physical systems in the absence of adequate amount of experimental data. Simulation of a complex system involves multiple levels of modeling, such as material (lowest level) to component to subsystem to system (highest level). This paper presents a Bayesian structural equation modeling approach to quantify both epistemic and aleatoric uncertainties in hierarchical model development. A generalized structural equation modeling with latent variables is presented to model three sets of relationships in hierarchical model development, namely, model predictions vs experimental observations at each individual level, model predictions at lower vs higher levels, and experimental data at lower vs higher levels. The three sets of relationships are represented by a hierarchical Bayes network, and the influencing factors between them are estimated by a Bayesian regression approach. Both measurement and prediction errors at various levels are quantified through the Bayesian method. The variability of input variables in the computational model is updated and quantified using various levels of measurement data via Bayesian inference and the structural equation modeling parameters. The proposed methodology is illustrated with a transient heat conduction example problem. Copyright © 2009 John Wiley & Sons, Ltd.