The time-scale of these fluctuations is usually orders of magnitude smaller than that of the phenomena af practical interest. Therefore, if PHOENICS were required to simulate them in detail, the computer time would be orders of magnitude greater than would otherwise be needed.
Asociated with the variations with time are variations with position in space. Therefore, if PHOENICS is to simulate them, its computational grid must consist of cells which are exceedingly small in comparison with the dimensions of the whole geometric domain. This requirement further increases the computer time.
The variations in space and time can be thought of as caused by a population of eddies of a wide range of sizes, orientations and speeds of rotation, which are in ceaseless and random motion.
These eddies come into being spontaneously as a consequence of the inherent instability of even a steady flow when the velocity of the fluid is much greater than its kinematic viscosity divided by the distance to the nearest solid object.
For flow in a pipe for example, 'much greater' is found to signify 'more than one thousand times greater'; which happens when the so-called Reynolds Number (= velocity times diameter divided by kinematic viscosity) exceeds 2000.
For other geometries, the critical value of the Relnolds number will be different in value; but, from the practical point of view it can be said that it scarcely matters what the criterion is, because it is vastly exceeded: most flows which PHOENICS is called upon to simulate are turbulent.
These effects include:
Micro-mixing is handled in PHOENICS by the Multi-Fluid Model.
DNS is still practicable only for very simple flows and for modest Reynolds numbers; and its expense is far too great to be afforded in engineering practice. Nevertheless such results as have been published are beginning to serve the needs of turbulence-model developers who require empirical data for the calibration of their models.
Hitherto 'empirical' has been synonymous with 'experimental'; but now its meaning can be extended so as to include 'DNS-generated' as well.
DNS is not dicussed further in the present Encyclopaedia articles.
A third approach to the problem of predicting the behaviour of turbulence phenomena can be regarded as combining some aspects of both DNS and macroscopic modelling. It is called 'Large-Eddy Simulation', commonly abbreviated to LES. Its nature is this:
Although still much more expensive in respect of computer time than wholly macroscopic modelling, LES is beginning to be used in engineering practice.
PHOENICS is equipped with one version of LES, the nature of which is explained here.
A useful source of recent information concerning both DNS and LES (and its recent variants) is:
K Hanjalic, Y Nagano and S Jakirlic (Eds)
Turbulence, Heat and Mass Transfer 6
Begell House, 2009.
|Prescribed effective viscosity (CONSTANT_EFFECTIVE)||LVEL (LVEL)|
|Prandtl mixing-length (MIXLEN)||Van-Driest|
|Prandtl energy (KLMODL)||Two-layer low-Re k-ε model (KEMODL-2L)|
|k-ε model (KEMODL, KEMODL-YAP)||Chen-Kim k-ε model (KECHEN, KECHEN-LOWRE)|
|RNG k-ε model (KERNG)||MMK and KL k-ε model (KEMMK, KEKL)|
|Realisable k-ε model (KEREAL)||Lam-Bremhorst k-ε model (KEMODL-LOWRE, KEMODL-LOWRE-YAP)|
|Wilcox (1988) k-ω model(KWMODL, KWMODL-LOWRE)||Wilcox (2008) k-ω model (KWMODLR, KWMODLR-LOWRE)|
|Menter (1992) k-ω model(KWMENT, KWMENT-LOWRE)||k-ω SST model (KWSST, KWSST-LOWRE)|
|Saffman-Spalding k-Ω model (KOMODL)|
|Two-scale k-ε model(TSKEMO)||Reynolds-stress model (REYSTRS)|
Smagorinsky subgrid-scale (SGSMOD)
|Low-Reynolds no. (various)|
The keywords in () are the keywords used in the TURMOD command to activate the model, and also displayed in the VR-Editor, Main menu, Models, Turbulence models dialogs.