Simulation-Assisted Design of Experiments
Also known as: Simulation-based DoE, Virtual DoE, Computer-aided DoE, SA-DoE
Simulation-assisted design of experiments (SA-DoE) integrates computational simulation tools — such as finite element analysis (FEA), computational fluid dynamics (CFD), or discrete-event simulation — with classical DoE principles to systematically explore the factor space of a system. Rather than running costly or hazardous physical trials, researchers execute a structured set of virtual experiments across selected factor combinations, then fit a surrogate model to the simulation outputs to understand main effects, interactions, and optimal settings.
Key highlights
- Eliminates or drastically reduces the need for costly, time-consuming physical prototypes.
- Enables exploration of dangerous, impractical, or ethically constrained experimental conditions safely.
- The fitted surrogate model can be reused for subsequent optimization, sensitivity analysis, and uncertainty quantification with negligible additional cost.
- Scales well to high-dimensional factor spaces when combined with space-filling designs and advanced surrogates such as kriging.
- Produces a complete continuous response surface, not just point estimates at tested factor combinations.
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use SA-DoE when physical experimentation is too expensive, time-consuming, ethically constrained, or physically impossible, but a validated simulation model of the system exists. It is ideal for early-stage engineering design exploration, process optimization in manufacturing, and parametric studies in structural or thermal analysis. Do not use it when no validated simulation model is available — garbage in, garbage out applies: an unreliable simulator produces unreliable DoE results. Also avoid SA-DoE when physical variability (measurement noise, material scatter) is the primary concern, because deterministic simulators cannot reproduce random physical variation without explicit stochastic modeling.
Strengths & limitations
- Eliminates or drastically reduces the need for costly, time-consuming physical prototypes.
- Enables exploration of dangerous, impractical, or ethically constrained experimental conditions safely.
- The fitted surrogate model can be reused for subsequent optimization, sensitivity analysis, and uncertainty quantification with negligible additional cost.
- Scales well to high-dimensional factor spaces when combined with space-filling designs and advanced surrogates such as kriging.
- Produces a complete continuous response surface, not just point estimates at tested factor combinations.
- Results are only as trustworthy as the underlying simulation model; an unvalidated or poorly calibrated simulator propagates errors into all DoE conclusions.
- Surrogate accuracy degrades in highly nonlinear or discontinuous response surfaces unless a large number of simulation runs is used.
- High-fidelity simulations (e.g., large-scale FEA or CFD) can be computationally expensive, limiting the number of feasible design points.
- Purely deterministic simulation ignores physical variability; separate stochastic or robust design layers must be added if variability matters.
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
How is SA-DoE different from standard DoE?
Standard DoE is designed for physical experiments where random error is present and replication is needed to estimate it. SA-DoE targets deterministic computer simulators where running the same input twice gives the same output. Consequently, SA-DoE uses space-filling designs instead of factorial grids, and fits surrogate models (kriging, RBF) rather than relying solely on ANOVA.
What surrogate model should I choose?
Kriging (Gaussian process regression) is the most widely recommended choice for smooth, deterministic responses because it provides not only a prediction but also a prediction uncertainty, which is useful for sequential design. Polynomial response surfaces are simpler and adequate when the response is nearly linear or quadratic. Neural networks and radial basis functions suit highly nonlinear surfaces but require more training runs.
How many simulation runs do I need?
A common rule of thumb is 10 times the number of input factors for an initial space-filling design, followed by sequential augmentation until cross-validation error meets the accuracy target. There is no universal threshold; the required number depends on the nonlinearity of the response, the number of factors, and the accuracy needed.
Do I still need physical experiments after SA-DoE?
In most engineering workflows, yes — at least a small number of confirmation runs. The surrogate captures the simulation model's behavior, but if the simulation itself has model-form error, physical confirmation tests are essential to verify that the predicted optimum translates to real-world performance.
Can SA-DoE account for manufacturing variability?
Not automatically. Standard SA-DoE treats factor levels as exact. To incorporate variability, you must either use a stochastic simulator (e.g., Monte Carlo inside the simulation model) or combine SA-DoE with robust design analysis, propagating distributions through the surrogate rather than using fixed factor values.
Sources
- 1.Santner, T. J., Williams, B. J., & Notz, W. I. (2003). The Design and Analysis of Computer Experiments. Springer.ISBN 978-0387954202
- 2.Kleijnen, J. P. C. (2015). Design and Analysis of Simulation Experiments (2nd ed.). Springer.ISBN 978-3319185668
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Simulation-assisted design of experiments. ScholarGate. https://scholargate.app/experimental-design/simulation-assisted-design-of-experiments