A Comparative Study of Computational Algorithms used in the Automatic Generation of Reduced-Order Models from CFD Simulations

In indoor thermal environment modelling applications
where dynamic local effects of fluid flows are critical,
classic zonal models are not always suitable. On the other
hand, CFD simulations can give accurate solutions at very
high computational cost. Reduced-order models (ROMs),
extracted from CFD simulations, can preserve CFD model
accuracy while being characteristically of low
computational cost. The authors propose a method,
known as CFD-ROM, capable of rapidly and automatically
generating, from CFD simulations, zones, mass, and heat
flows, and boundary conditions (BCs) for ROMs. This
paper presents a comparative study of automatic zone
generation algorithms as a necessary initial step to
developing the CFD-ROM method. Zone generation
algorithms compared in this paper are: (1) Mean Values
Segmentation; (2) Classic Watershed; and (3) Coarse Grid
Interpolation. The methods were compared on the bases of
their accuracy against the original validated CFD
simulation results, and their time to zone generation. The
Mean Values Segmentation method yields promising
results, providing a mean error below 0.2K for 15-zone
models generated in under 28 seconds. The next
immediate steps for the development of CFD-ROM are (i)
construction of a ROM solver, and (ii) testing its ability to
predict thermal conditions when CFD BCs and ROM BCs
differ