Customized Regression Model for Improving Low Fidelity Analysis Tool

Abstract

Multidisciplinary design exercises generally restrict the number of evaluations of a high fidelity, computationally time intensive analysis because of requirements on design cycle time. Instead, analysis tools of lower level of fidelity that are less computationally time intensive are often complemented with an uncertainty model and used within a robust design framework. In this paper, a regression technique is customized to discover a functional form between the Low Fidelity Analysis response and High Fidelity Analysis response and develop an uncertainty model that estimates the high fidelity function with greater confidence in the regions of interest to the designer. The method is demonstrated for a synthetic function example and for a typical disciplinary metric in hypersonic vehicle design. Results show that the customized regression has enabled to reduce the fidelity uncertainty in the regions of interest. The high fidelity responses are within the three sigma prediction bounds.