Gaussian process computer model validation method
In this paper, the problem of computer model validation, based on a set of experimental results, is addressed. A global statistical modelling of the validation framework is proposed. This modelling is based on a Gaussian process modelling of the bias between the computer model and the physical system. The application of classical statistical inference to this statistical modelling yields a rigorous validation method. This Gaussian process validation method simultaneously calibrates the computer model, and adds to its predictions a statistical correction based on the set of experimental results. This statistical correction can substantially improve the calibrated computer model, for predicting the physical system on new experimental conditions. Furthermore, a quantification of the uncertainty of this prediction is provided. Physical expertise on the calibration parameters can also be taken into account in a Bayesian framework. Finally, a real case study on the thermal-hydraulic code FLICA 4 is presented, in a single phase friction model framework. The Gaussian process validation method allows to consistently assimilate a set of experimental results and to significantly improve the prediction capability of the thermal-hydraulic code FLICA 4.