Sensitivity Analysis in Design
Sensitivity analysis ranks which inputs actually move the outcome, so effort and measurement go where they change the answer.
The question it answers
A model may have fifty inputs, but usually a handful control the result. Sensitivity analysis identifies which inputs matter, by how much, and whether they interact. That ranking tells engineers where to spend measurement and modeling effort and which uncertainties to reduce first.
Local versus global
- Local sensitivity measures the response to a small change around one point, useful near a chosen design.
- Global sensitivity explores the whole input range, capturing interactions and nonlinearity.
- Variance-based methods attribute the spread in the output to each input and to combinations of inputs.
Why global matters here
Fusion models are strongly coupled and nonlinear, so local gradients can mislead. A global view across the configuration space reveals interactions, such as field and density jointly setting confinement, that a one-at-a-time study would miss.
Making it affordable
Global analysis needs many model evaluations, which is why it runs on surrogates trained on high-fidelity runs. The surrogates make thousands of samples cheap, and any surprising region is re-checked at full fidelity.
Applied to Kronos
For the Hyperion breeder, sensitivity analysis shows which inputs drive the tritium breeding ratio and the peak heat flux, focusing materials and neutronics work. For the burner, it highlights the plug parameters that dominate confinement.
Feeding back
The ranking directly shapes uncertainty-driven decisions and the design of experiments, closing the loop between what is uncertain and what is worth measuring.