VII.5 Moment (and related) methods
In a previous lecture we discussed turbulence modelling, in which the aim was to remove the
‘small scale’ detail of turbulence and replace it with a (cheaper) turbulence model. This is
done by an averaging process. A similar approach is used in combustion modelling to eliminate
the small scale turbulent effects and to derive models for the combustion processes. Since the
density of the fluid is variable we introduce a density-weighted average called a Favre
average. As usual we split the flow up into mean and fluctuating '' components
| (VII.5) |
For a normal averaging procedure we would assume that ux' = 0. Favre averaging is
constructed by requiring that the density-weighted fluctuating component
| (VII.6) |
This is equivalent to
| (VII.7) |
Of course, Favre averaging can be applied to any flow variable, not just the velocity component
ux.
This density-weighted average is more difficult to evaluate, requiring simultaneous
measurement of and ux. However when it comes to manipulating the fluid flow equations
(NSE, plus the transport equations for i), Favre averaging is much easier to use. Frequently
we find ourselves manipulating terms such as
Using a conventional average (non-density weighted) we end up with 4 terms
none of which can be removed. Using Favre averages produces a much simpler expression
Favre averaging can be used in a similar manner to that described in the previous
lecture, to derive mean flow equations and equations for k and for compressible
flow. Thus a version of the k - model for turbulent compressible flows can be
derived.
We can also apply Favre averaging to equation (VII.4). After some manipulation we arrive
at the equation
| (VII.8) |
All the terms on the l.h.s. contain variables already known (from the solution of the mean flow
equations). Those on the r.h.s. however are unknown as yet, and typically require modelling.
In high Re flows, the molecular diffusivity D is small, so the term .Dii can safely be
ignored. This leaves two terms
- | . | | representing turbulent transport, and | | | |
|
| | | the mean chemical source term | | | | |
Both terms cause problems for modelling.
Various approaches have been applied to modelling these various terms, leading to a wide
range of combustion models ranging from simple, cheap and inaccurate to sophisticated,
accurate and computationally very expensive. Approaches include
-
Eddy Break Up model.
-
Devised by Spalding (1971), this is based on the assumption that the turbulent
cascade mixing process controls the chemical reactions - i.e. that mixing of the
reactants is the rate-determining step in the process. This can be used to derive a
simple model for the chemical source term. However this model requires detailed
tuning of coefficients, an unsatisfactory arrangement, and is not physically very
accurate. It has been formulated primarily for premixed combustion, although it
is possible to develop formulations for non-premixed combustion.
-
PDF transport model
-
Turbulence can be discussed statistically, as a quasi-random flow with particular
statistical properties. These properties can be described in terms of correlations
between flows at different points, or the same point at different times. The basic
quantity involved is the PDF, or Probability Density Function. Simple engineering
models such as the k - model ignore this, providing information only about
the mean levels of turbulence. However it is possible to model turbulence in a
statistical manner, providing information about these turbulent correlations as a
PDF transport equation. Such models are computationally costly but can be very
accurate.
Extending this to cover combustion involves providing a transport equation for the
so-called joint pdf of velocity and reactive scalars. This is denoted
and has the interpretation that
is the probability that the flow at point at time t has velocity and is in the reactive
state described by . This can provide the information necessary to specify the
additional terms in (VII.8). However, although this method is potentially more accurate
than other approaches, and can be applied to premixed, non-premixed and
partially premixed cases, it is also significantly more expensive, and so is more
used for theoretical investigation of combustion than for solving engineering
problems.