Abstract: In order to resolve the problem of model validation with limited test data in the untested domain, this paper presented an extrapolation method together with qualitative knowledge and quantitative Bayesian inference. Qualitative information such as the subject matter experts’ opinions is transformed to prior probability in the proposed quantification method and applied to Bayesian inference. The Bayesian network  with Monte Carlo method which is limited in sampling range is explored for extrapolating quantitatively the inference from the validated domain at the component level to the applied domain at the system level. And Bayesian interval hypothesis testing is performed on the evaluated quantity to assess the model validity. A simplified version of a static frame challenge problem developed by Sandia National Laboratories demonstrates that the method provides a valid approach to facilitate rational decisions in confidence extrapolation.

Key words: model extrapolation, qualitative knowledge, quantitative Bayesian inference, Bayesian network, Monte Carlo method