Multi-fidelity data fusion for the design of multidisciplinary systems under uncertainty

Abstract

This thesis presents a multi-fidelity methodology to enable the incorporation of high-fidelity data into a conceptual design process. The methodology is based upon a fidelity weighted combination of Gaussian Process surrogate models that takes into account both the quality of the Gaussian Process approximation and the confidence of the designer in the disciplinary model being approximated. The methodology is demonstrated on the stability and control analysis of a Blended-Wing-Body aircraft's center of gravity limits. The results show that low-fidelity data is enhanced by the presence of high-fidelity data in key areas of the design space. At the same time, the presence of even sparse high-fidelity data is key to reducing the variance in the stability and control analysis, thereby improving the quality of the predictions of the center of gravity limits.