FDS 6 Options

[Kevin]

I am starting this thread because some of you at our workshop (April
2009) here at NIST expressed an interest in some of the features that
are currently being tested in FDS. These features are not described in
the User's Guide because they are still in the research stage.
Nevertheless, we'll post some basic information in this thread to let
those who are interested understand better what we are doing.

[rmcdermo]

Dear All:

At our meeting last week I had several requests for a better
description of the options that are currently in beta testing for
FDS6. Here I will provide a more thorough list than the one I gave in
my talk and also give some discussion on each parameter.

I must emphasize that these options are in beta testing (at best), so
use them cautiously!

Note that WERNER_WENGLE_WALL_MODEL has been removed as a MISC
parameter because this wall model is now the default in the latest SVN
of FDS (will be default for the FDS 5.4.0 release but has not gone
through the full V&V suite yet). SLIP_FACTOR and default boundary
layer thickness have been removed. The user may still specify a free
slip wall by setting FREE_SLIP=.TRUE. on the SURF line.

First, I will give a list of the new parameters (or new default
settings for old parameters) with their default FDS6 values and show
where they belong in the input file. A description of each of the
parameters then follows below.

Cheers,
Randy

Parameters and FDS6 defaults:

&MISC FDS6 = .TRUE. (sets all the FDS6 parameters, overwrites any of
these parameters if they are also specified in the input file)

&MISC CFL_VELOCITY_NORM=1, FLUX_LIMITER=2, DYNSMAG=.TRUE.,
BAROCLINIC=.TRUE., CHECK_KINETIC_ENERGY=.TRUE.

&REAC HRRPUV_AVERAGE=1.E10

&SURF ROUGHNESS=0.

Description:

FDS6 -- (LOGICAL) If .TRUE., set and overwrite all FDS6 options

CFL_VELOCITY_NORM -- (INTEGER) Possible values = [0,1,2]. For a
uniform grid spacing, dx, the stability restriction for the advective
time step is calculated by dt <= CFL*dx/||U||. The default CFL in FDS
is 1.0 (this can be changed by setting CFL_MAX on MISC). The time
step then depends on how you specify the velocity norm ||U||. The
current practice in FDS is to take the max over the velocity
components. This corresponds to CFL_VELOCITY_NORM=0, which is the
least restrictive option. For CFL_VELOCITY_NORM=1 we take ||U|| as
the max over all cells of the L1 norm of the velocity vector, which is
the most restrictive (and hence the most stable). For
CFL_VELOCITY_NORM=2, we use the L2 norm, which is more restrictive
than the current default but not as restrictive as the L1. If you are
unfamiliar with the L norm jargon, let's look at a simple example
which I think will clarify. Suppose the cell with the maximum
velocity has a velocity vector given by [u,v,w]=[1,2,3]. For
simplicity, suppose dx = 1 and CFL = 1. For CFL_VELOCITY_NORM=0, we
have ||U|| = max([1,2,3]) = 3, and hence dt = 1/3 = 0.33. For
CFL_VELOCITY_NORM = 1, we have ||U|| = |1| + |2| + |3| = 6, and dt =
1/6 = 0.17. For CFL_VELOCITY_NORM = 2, we have ||U|| = sqrt(1^2 + 2^2
+ 3^2) = 3.7, and dt = 0.27. We choose L1 for the velocity based on
both accuracy and stability considerations (which tend to be linked in
explicit methods). There is no proof of stability for the flux
limiter schemes discussed below for 3D problems. However, Godunov's
scheme can be shown to be stable for L1 for multi-dimensional problems
(see e.g. E.F. Toro, "Riemann Solvers and Numerical Methods for Fluid
Dynamics: A Practical Introduction", 2nd Ed. Springer, 1999, p 556).
Our experience so far has been that this bound also works well for the
second-order flux limiting schemes we have now implemented in FDS. It
should be noted that the tighter time step restriction does not
necessarily translate into smaller time steps (and hence slower
simulations) than the current FDS default because the flow fields are
different. In fact, because the dynamic model tends to produce more
dynamics, more mixing, and better "puffing" of the fire, we see slower
local velocities in many cases. The net result is that the time steps
for FDS6 seem to be roughly the same as for FDS5.

FLUX_LIMITER -- (INTEGER) Possible values (-1,0,1,2,3,4). FL = -1 is
the current FDS default which is essentially central differencing with
a boundedness correction applied if the scalar goes out of the range
[0,1]. Stanford uses a similar "Bounded QUICK" scheme. For FL>=0 we
discretize the conservative form of the scalar transport equations (a
staggered-grid finite-volume scheme). FL=0 simply sets central
differencing and this option should look very similar to FL = -1. FL
= 1 sets first order upwinding (Godunov's scheme). FL = 2 sets
Superbee. FL = 3 sets MINMOD. And FL = 4 sets CHARM. FL = 2 should
be used for LES and FL = 4 for DNS because CHARM is convergent for
smooth solutions. Currently, the scalar transport scheme reverts to
first-order upwinding at a mesh boundary. This is necessary because
the data structure in FDS only allows for one layer of ghost cells and
two ghost cells are required to maintain second-order limiters. This
enhancement will be made for v6. See the E.F. Toro reference above
for a more detailed description of these limiters.

DYNSMAG -- (LOGICAL) If .TRUE., implements the variable density
formulation of the dynamic Smagorinsky model. The model coefficient
is averaged over the test filter volume and any values which are
negative after this averaging procedure are clipped to zero.

BAROCLINIC -- (LOGICAL) If .TRUE. (current FDS default is .FALSE.),
apply time-lagged baroclinic correction to the right hand side of the
momentum equation. There is already ample discussion of this
parameter in the Tech Guide. In previous versions of FDS, especially
for multiple mesh calculations, the baroclinic correction was thought
to be the source of instability. However, as we analyzed the
formulation we found that the fault was in the numerics: asymmetry of
the stress tensor caused at a mesh boundary caused spurious generation
of angular momentum (vorticity). In the spirit of including the best
physics that we have available, we now see no reason to exclude this
correction. The computational cost is insignificant and by including
the correction by default we eliminate the question of whether this
term is "applicable" or "necessary". It will turn itself off if it is
not needed.

CHECK_KINETIC_ENERGY -- (LOGICAL) If .TRUE., estimate the subgrid
kinetic energy after the corrector step and calculate MTR(x,t) -- the
"measure of turbulence resolution". This quantity may be output using
QUANTITY = 'TURBULENCE RESOLUTION'.

HRRPUV_AVERAGE -- (REAL) This parameter is an ad hoc bound on the heat
release rate that was introduced into FDS5 in order to help match
flame height profiles. It lacks a sound physical basis and we should
therefore get away from depending on it as soon as possible. The
bound set in FDS6 is simply a way to eliminate this parameter from the
chemistry model. I expect this parameter will no longer exist in the
release of v6.

ROUGHNESS -- (REAL) This is a roughness height (in meters) for the
given surface. The current implementation of the WERNER WENGLE model
applies a roughness correction that is qualitatively correct (increase
the roughness and it will increase the stress) but we are still in the
process of developing this model for rough walls. Currently the user
may need to tune this value if roughness is of critical importance to
the calculation.

[Mohamed]

Dear Sir
Could i participate in the testing process for FDS 6.

Regards,
Mohamed Aly

[rmcdermo]

Mohammed:

By all means. Please simply provide feedback through the discussion
group. No one needs special permission. FDS 6 options are available
with the latest SVN of the code.

Thanks,
Randy