Structural modelling and Bayesian inference for model validation and confidence extrapolation

The hierarchical validation is recommended for assessing the confidence of the large-scale simulation model, through comparing the model predictions with experimental observations at each level. However, the system-level experimental observations may be scarce in the full-scale testing. A new structural modelling approach is proposed to implement the higher-level model validation using lower-level data. The proposed approach relates the computation model to corresponding system responses and lower-level outputs to higher-level responses. Bayesian network with Markov chain Monte Carlo (MCMC) sampling is used to represent the relationship sets and estimate the structural modelling parameters and input variables. Highest posterior density (HPD) confidence range is employed to quantify the model credibility, the confidence intervals of model errors, and the effect of lower-level data on the higher-level model assessment. The proposed methodology is implemented for hierarchical model validation and confidence extrapolation of a guidance model on a flight vehicle using time series data.