Design and Application of the Sage Surrogate Modeling Software
Andrew L. Kaminsky, Alec M. House, L Jensen, Matthew Liu, William Chapman, Alessandro P. Brown, Andrew M. Wissink, Nathan S. Hariharan, David McDaniel
Abstract
In the Department of Defense (DoD), the ability to rapidly and accurately characterize system performance across trade spaces is critical for timely decision-making and effective design optimization. Engineers often rely on computational physics models of varying fidelities to estimate performance, accelerate design processes, and improve outcomes. However, these models frequently face significant limitations due to high computational costs and long runtimes, which hinder their ability to deliver broader trade space exploration, high accuracy, and quantified uncertainty—key factors in achieving mission-critical objectives. Surrogate modeling provides a practical solution by approximating system behavior with reduced computational cost while maintaining accuracy. Data-driven surrogate models map input-output relationships from observed data to estimate system behavior across the entire trade space, enabling faster and more efficient exploration. This paper introduces Sage, a modular and user-focused software designed to address these challenges. Sage enables automated creation, evaluation, and refinement of high-quality surrogate models, supporting rapid and accurate system performance estimates while providing robust tools for uncertainty quantification. By offering these capabilities, Sage serves as a critical resource for the DoD community, empowering engineers to meet complex design and analysis needs with efficiency and precision.