Meta-models in computer experiments: kriging vs artificial neural networks

Non-stochastic simulation models, such as finite element or computational fluid dynamics, often support real experiments in industrial research. It has become a common practice to provide a meta-model as computer experiments can be highly complex and time-consuming, and the design space is often broad. The meta-model is an approximation of the computer experiments response adapted both globally and locally on the design space, in order to capture local minima/maxima. The Kriging model, first proposed in Geostatistics, is doubtlessly the most popular meta-model because of its recognized ability to provide high- quality predictions. The underlying correlation structure can be evaluated either by estimating the parameters of correlation or by means of a variogram. In this paper, the performance of the Kriging model is compared with an Artificial Neural Network meta-model in order to determine which model guarantees higher accuracy in predicting the result of four-dimensional com- putational fluid dynamics experiments for low pressure turbines where energy loss values are provided.