Definition
A surrogate model is a fast, approximate model trained to predict the output of a slower, more detailed simulation or calculation, without running that full simulation each time.1 It is used when the original simulation is too computationally expensive to run repeatedly, such as during an interactive design or estimating process, trading some accuracy for a large gain in speed.
Why a surrogate model's training range limits where it's reliable
A surrogate model's usefulness depends on how well it was trained to approximate the specific range of cases it will actually be used for; a surrogate trained on one range of parts or conditions can give unreliable predictions outside that range.
Where surrogate models apply
Not on the drawing; a surrogate model is a computational technique used within engineering or manufacturing software, not something specified by a part's requirements.
Common mistakes
Using a surrogate model outside the range of cases it was actually trained or validated on, where its approximation can become unreliable without any obvious warning sign.
A surrogate model's usefulness depends on how well it was trained to approximate the specific range of cases it will actually be used for; a surrogate trained on one range of parts or conditions can give unreliable predictions outside that range.
Read the researchFrequently asked questions
Is a surrogate model the same as a simulation?
No, a surrogate model approximates a simulation's results quickly, without running the full simulation; it trades some accuracy for speed.
Why not just always run the full simulation?
Because a full simulation can be too computationally expensive to run repeatedly, especially in an interactive process needing fast feedback.
Can a surrogate model be trusted for any input?
Only within the range of cases it was trained and validated on; predictions outside that range can be unreliable.
Notes & references
- Queipo, N. V., Haftka, R. T., Shyy, W., Goel, T., Vaidyanathan, R. & Tucker, P. K. (2005). Surrogate-based analysis and optimization. Progress in Aerospace Sciences, 41(1), 1–28. https://doi.org/10.1016/j.paerosci.2005.02.001 The standard foundational review of surrogate-based analysis and optimization in engineering design, used here as the grounding for this definition.
