A GNC group had hi-fidelity simulations of the flight behavior of an object that were too slow to run advanced analytics or perform control tasks. Like many real-world systems, the complex input output behavior of the simulations form black-box function predictions that can be modeled using emulators (aka, predictive models) when speed is necessary (i.e. embedded controls, virtual sensors etc.). Unfortunately, the level of variable interactions present in the system made building accurate emulators difficult.
Using a series of black box challenge problems from the Prime Defense Contractor, SmartUQ created and implemented a new emulator capable of handling highly interacting input systems. Combined with optimal DOEs, this created accurate emulators with very limited sampling sets.
Advanced design of experiments and emulation technologies reduced the number of samples well below the original targets and exceeded (<1% error) the prediction accuracy requirements across the entire design range for all the challenge problems presented.
