Variable fidelity modelling in modern aircraft design

Abstract

The key to successful application of surrogate modelling is the ability to quickly generate high quality data over a large parameter space. This can be achieved through application of Variable Fidelity Modelling (VFM), which is a data fusion technique that allows data from different sources to be combined. VFM can be used to correct a dataset obtained by a simple and fast method with a small number of high quality samples in order to generate a high-fidelity surrogate at a marginal cost. This paper presents the results of a feasibility study of a VFM technique based on Hierarchical Kriging in the context of aerospace applications. A generic process for generation of a high quality surrogate using VFM and assessment of its accuracy is presented and validated for a number of realistic use case scenarios. Investigation of the impact of various parameters of the method is conducted, highlighting the importance of selection of the appropriate data correlation functions. The potential of VFM to significantly reduce the computational cost required for generation of a high-fidelity surrogate is demonstrated for the case of a parametric study of wing cruise performance used at the early stages of the aircraft design process. In addition, an application where the VFM method is used to predict the effects of wing section optimization is shown. The VFM approach presented in this paper can be implemented in a number of different scenarios and can offer significant improvements for the surrogate generation process by quickly generating high-fidelity response surfaces.