Fire risk analysis can be greatly improved using high fidelity CFD models. Unfortunately these models can take hours to days to run, limiting the number of scenarios that can be explored. This cost also makes CFD uncommon in quantitative risk analysis where the additional accuracy would be quite valuable.
Emulation allows quantitative uncertainty analyses where computationally intensive modeling codes have been applied. Once built, a sufficiently accurate emulator can rapidly generate response surfaces and prediction uncertainty estimates. Advanced emulators and statistical tools can even handle functional inputs and outputs like heat release rate profiles and temperature response fields.
Using a typical fire simulation CFD, a series of functional emulators were fit to predict the mean and variance of the temperature at different locations over the course of a simulated fire event with respect to several input variables. These emulators were then used to evaluate the likelihood of exceeding temperature limits for a number of scenarios all without requiring further CFD simulations.
