16.08.2013 Views

The computation of turbulent natural convection flows - Turbulence ...

The computation of turbulent natural convection flows - Turbulence ...

The computation of turbulent natural convection flows - Turbulence ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

421<br />

results <strong>of</strong> the 3D numerical simulation were mostly two-dimensional. <strong>The</strong> 3D<br />

simulation using the k-ε-AWF produced similar but weaker movement near<br />

the top <strong>of</strong> the cavity, compared with the experimental data, but it was not<br />

strong enough to penetrate all the way down to the bottom <strong>of</strong> the cavity, as<br />

was seen in the experimental data. <strong>The</strong> numerical simulation also produces<br />

similar circulations at the bottom with the opposite direction. Comparisons<br />

<strong>of</strong> normal stresses showed that while near the end walls the normal stresses<br />

were severely under-estimated, over the rest <strong>of</strong> the cavity thek-ε-AWF returns<br />

the same fluctuation levels as those found in the measurements. <strong>The</strong> inconsis-<br />

tency is due to the existence <strong>of</strong> 3-dimensional flow conditions reported in the<br />

experimental study <strong>of</strong> the 15 ◦ stable test case.<br />

<strong>The</strong> version <strong>of</strong> the k-ε-AWF which included a limiter for the <strong>turbulent</strong> vis-<br />

cosity based on a realisability criteria was then employed to simulate the flow.<br />

<strong>The</strong> calculation without the <strong>turbulent</strong> viscosity limit produced steady-state re-<br />

sults even in a time dependent simulation. <strong>The</strong> calculation with the <strong>turbulent</strong><br />

viscosity limit, on the other hand, resulted in a time-dependent solution. <strong>The</strong><br />

reason or this behaviour is that in the original model the level <strong>of</strong> <strong>turbulent</strong><br />

viscosity was sufficient to damp out any instabilities which might have led to<br />

large-scale unsteady structure. <strong>The</strong> limiter became active in certain regions<br />

<strong>of</strong> the flow, acting to reduce the <strong>turbulent</strong> viscosity, and hence allowing such<br />

structure to develop. <strong>The</strong> <strong>turbulent</strong> viscosity limit is based on a physical con-<br />

straint (Section 3.3.2). In general the flow pattern was similar to the results <strong>of</strong><br />

the k-ε-AWF without the limiter. <strong>The</strong>se comparisons revealed that the inclu-<br />

sion <strong>of</strong> the <strong>turbulent</strong> viscosity limit does not make significant changes to the<br />

temperature pr<strong>of</strong>iles. <strong>The</strong> comparisons showed that introduction <strong>of</strong> the viscos-<br />

ity limiter, in addition to enabling the k-ε-AWF to return unstable flow condi-<br />

tions, led to some improvements in the predicted time-averaged velocity field,<br />

but there were large deviations between the predicted and measured flow<br />

fields. Improvements in the predicted rms velocities were evident, most no-<br />

tably near the end walls, where the predicted levels were significantly higher<br />

than those <strong>of</strong> k-ε-AWF predictions without the viscosity limiter, and close to<br />

the measured values. <strong>The</strong>se predictive improvements are most likely to re-<br />

sult from the fact that the predicted rms velocities included both the modelled

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!