Surrogate models

Surrogate models are high-speed mathematical approximations used to replace computationally expensive simulations in engineering and design.

When a single high-fidelity simulation (like a Navier-Stokes CFD run for an aircraft wing) takes hours or days, surrogate models step in to bridge the gap. These 'models of models' use data-driven techniques like Kriging, Radial Basis Functions, or Neural Networks to map inputs to outputs based on a limited set of intelligently sampled data points. In practice, this allows engineers at firms like NASA or SLB to perform thousands of design iterations in seconds, often achieving 100x faster results with minimal loss in accuracy. By replacing brute-force computation with efficient statistical emulators, teams can optimize complex variables (such as aerodynamic drag or structural mass) that would otherwise be too costly to explore.

https://en.wikipedia.org/wiki/Surrogate_model