Status Report: CREATE Applied Surrogates Institute
HPC resources have traditionally been used to support high-fidelity analysis at discrete operating states.   In recent years, there has been growing interest in using HPC to support mission engineering, design optimization, and uncertainty quantification which all require parametric analyses over a large operating trade space. Surrogate models, which use AI/ML techniques to train models from multi-fidelity simulation data or experiment, provide a continuous representation of a system state that supports these applications in DoD systems engineering, design optimization, and digital twin efforts.  In 2023 the HPCMP began the CREATE Applied Surrogates Institute, or CASI, to build a software tool to enable HPC-based surrogate model building across the DoD.   The development team put together a Python-based toolkit called “Sage” with an easy-to-use interface for applications engineers that don’t have training in AI/ML or statistics.   Internally Sage supports advanced features like adaptive sampling, uncertainty quantification, multi-fidelity models, integration of test data, and time-dependent surrogates for different DoD applications.  Pre-releases of Sage are under evaluation by over 50 CREATE license holders across the DoD and an official production release is scheduled later this summer.  This talk will summarize the capabilities that have been added to Sage over the past year and discuss promising new capabilities that are under development.
IMPACT
Many DoD programs rely on home grown surrogate modeling software that run on desktop systems using low-order empirical models to generate training data.  CASI’s goal is to leverage HPC to perform larger scale AI/ML surrogate models based on much larger multi-fidelity datasets, using software supported by the HPCMP.   The enhanced quality of the surrogate models will ultimately lead to better decision making throughout the entire design cycle of an DoD weapons systems.
PRESENTER
Wissink, Andrew
andrew.m.wissink.civ@army.mil
650-604-5385
US Army AvMC
CO-AUTHOR(S)
Kaminsky, Andrew
andrew.kaminsky@cfd-research.com
CATEGORY
Surrogate Modeling for HPC
SECONDARY CATEGORY
AI/ML for HPC
SYSTEM(S) USED
Narwhal, Nautilus, Raider