You are on page 1of 30

Solution Techniques - 2

Solver Options
As discussed in the last lecture Fluent
has an unsteady solver and steady
solver options
The unsteady solver is used for dynamic
simulations where you are integrating in
time
The steady solver is designed to get to
get a steady flow or time averaged
steady flow solution
Solver Options
Both steady and unsteady can be used
with the segregated solver or the
coupled solver.
The segregated solver solves the
equation set sequentially
The coupled solver solves the equation
set in one go. This is the more
modern solver but consumes more
memory
My experiences
The segregated solver is the default
option. For most purposes this is good
enough. Ive never had to use the
coupled solver
I am of the view that you should think
more about the possible transient
aspects of the problem and worry about
steady vs unsteady rather than
segregated vs coupled
Convergence
Inevitably any iterative/numerical solution
procedure will only give a solution which is
converged relative to some criteria
The solution may be converged if:
All discretized transport equations are obeyed
to a specified tolerance defined by Fluents
residuals (but these may be too large)
The solution no longer changes with more
iterations (but the problem may be stuck)
Overall balances close (for a steady flow
simulation)
Convergence
Convergence is not the same as
accuracy. The solution is accurate if it
matches experimental data (which you
obviously need to make the judgement)
Thus there are qualitative and
quantitative aspects to convergence.
Convergence - Residual Plots
Fluent keeps track of the residuals throughout the
iteration. The default value of these is 1.0e-3.(except
for enthalpy - 1.0e-6 and species transport 1.0e-5)
You should plot these.
Solve-> Monitors -> Residuals -> Plot
The residual plots show when the residuals have
reached a specified tolerance
They can show which equations are having
convergence problems, though this is sometimes
difficult and there are other ways finding this out -
see multigrid verbose
The residuals measure the imbalance on conservation
equations
Convergence - Residual Plots
Residuals are not the full story
For some flow problems the default values are too
large
In my cyclone studies the solution appears to be
converged on some grids but the pressure profile is
unrealistic, This can only be resolved by a transient
solution and even then the change in the residuals
during the iteration is not much.
Some problems will display no apparent change in
residuals but the solution may be doing the right
thing from other criteria - my gravity sluice
Some flow problems display periodic convergence
behaviour - dont jump the gun!!!
Convergence
You can also monitor Lift, drag or
momentum or variables at points
surfaces or boundaries
Solve -> Monitors -> Force
Solve -> Monitors -> Surface
When these values stop changing the
solution may be converged
Check overall heat and mass balances if
the problem has a defined flow volume
Accelerating convergence
My opinion is that if the solution is converging then
leave it alone
Remember the equation set is non-linear
All numerical approaches to solving non-linear
equations estimate a new solution estimate using
some gradient of the solution at the current solution
estimate.
If the gradient is large then the new estimate may be
worse than the old estimate and the solution may
diverge. This will tend to happen at the beginning of
the iteration.
Accelerating convergence generally increases the
likelyhood of divergence happening
Under-relaxation
If you really must fiddle then you can:
Increase under relaxation factors

under-relaxation factors slow down the rate at
which the solution changes during the iteration
and in transient solutions increase the Courant
number:



Courant number is the ratio of a time step to a cell
residence time


( )
old calc old new
| | o | | + =
fluid cell
u x
t
C
/ A
A
=
Courant number
This specifies a maximum internal time step the
solver may take during the time integration and is
not the same as the time step of the simulation
The time step of the integration is used to
calculate a final time in a numerical integration.



A numerical integration will take steps in time At
i
.
which are less than At
s
.
The Courant number specifies a maximum value
of At
i
.
( )dt t
s
s
t t
t
t t
}
A +
A +
= ' | |
Supply Initial Conditions from a
previous solution
This is a good approach if the new solution is
close to a previous solution and is useful if
you need a family of solutions over a range of
flows
You can also start with a solution using a
simpler turbulence model or a solution which
is otherwise a subset of the problem
Get an approximation using lower order
discretization
Multigrid
The solvers use a methodology where the
grid is coarsened by merging several
neighbouring control volumes into one
volume and an approximate solution is
obtained for the coarser grid that is applied
as a starting point to your grid
The solver does this automatically but you
can adjust the parameters
Grid Independence
Finer grids are more accurate but are usually
less stable
Check your solution with more than one grid.
You can generate a finer grid using an
existing grid by grid adaptation. This
technique will be discussed in a later lecture
Boundary Conditions
Any solution of a set of PDEs requires a set of
boundary conditions for closure
From a physical perspective you need to specify
boundary conditions such an an inlet flow. However
boundary conditions are required at all boundaries
that surround the flow domain
For example a wall boundary condition specifies by
zero velocity at the wall by default. If you are solving
mass or heat transfer then you need to specify the
boundary conditions for these variables either as a
flux or boundary concentration
Fluent by default assumes a wall boundary condition
unless you specify otherwise
Boundary Condition Types
Flow inlet and outlet boundaries
Wall and repeating boundaries
Internal cell zones
Internal face boundaries
Define -> Boundary Conditions
Set boundary conditions
Change the type of boundary condition
Flow inlet and outlet boundaries
These specify boundaries conditions across
which there is a flow
Velocity inlet
Mass flow inlet
Pressure inlet
Pressure outlet
Outflow
Pressure far-field
Inlet vent, outlet vent, Intake fan, exhaust fan
These are typically used for flow problems
inside vessels or channels
Mass Flow inlet & Velocity inlet
b.c.
The simplest inlet bc is the Mass Flow inlet
However you can also specify a velocity inlet
bc with either a constant velocity or you can
specify a velocity profile such a fully
developed pipe flow profile
You also need to specify the turbulence if the
flow in a turbulent problem
You will need at least one outlet bc (either
Outflow or Pressure Outlet)
Fluent calculates the Pressure
Pressure inlet b.c.
Alternatively you can specify the Pressure at
the inlet and Fluent will calculate the flow
You still need to specify the turbulence
can be used for compressible and
incompressible flow problems
can also be used to define a free boundary in
an external or unconfined flow
Must be used in conjunction with a pressure
outlet b.c.
Pressure outlet b.c.
Here you specify the pressure at the outlet
from the flow domain
However this b.c has backflow parameters
allows for flow reversal. In this instance the
b.c behaves like a pressure inlet b.c.
Outflow b.c.
This is used to model flow exits where details
of flow velocity and pressure are not known
before hand.
Fluent extrapolates from the interior
You cant use this b.c. where
You have a pressure inlet b.c
The flow contains a density variation
the flow enters the domain (like a flow reversal)
gradients in the flow direction are significant
Outflow b.c. - multiple outlets
If you have multiple outflows you need to
specify the flow split otherwise Fluent
assumes that the flow is split equally
If you dont know the flow split, or your aim
is to predict the split, then you cant use this
b.c
Although you could use outlet b.c.s to get an
initial estimate and change them to a
Pressure b.c to accelerate convergence
Pressure far field b.c.
used to model a free stream condition at
infinity (or a long way away)
Also called a characteristic b.c
used typically in aerodynamic applications
such as flow around wings
Vent and fan b.c.
These are specialised b.c.s that model flows
at vents and fans - see the documentation
Wall boundary conditions
Obviously to model the boundary conditions at the
vessel or channel walls
They are also used to bound fluid & solid
regions(Fluent can solve heat transfer in solids)
Default is to assume zero velocity.
But you can specify the shear stress
This can be used to approximate a free surface
can also be used to model a momentum source
Also need to specify wall fluxes for heat and mass
transfer. This is either as a temp, concentration or
flux.
Walls can move. Moving boundaries, sliding meshes
etc will be treated in a later lecture
Repeating and periodic and
symmetry b.c.s
These are used when the flow geometry has some
symmetry that enables you simplify the flow problem
by solving a portion of the domain
Symmetry b.c.
Used when the physical
geometry and the flow
filed has a mirror
symmetry
At the symmetry plane
the following must be
satisfied
zero normal velocity
zero normal gradients of
all variables
If c|/cn = 0 for all variables at symmetry planes:
n
n
Symmetry planes reduce the computational effort in
this case only one quarter of the channel needs to be
modelled
Periodic b.c.
These must be in pairs
and they have to be set
up correctly in Gambit
They also have to be
physically identical
Here there is a
symmetry but there is a
flow normal to the b.c.
The flow field in at one
b.c equals the flow field
out at the other.
Mixing Tank (Rotationally or cyclically periodic)
u
Heat exchanger tubes (translationally periodic)
u
Other Boundary Conditions

Other Boundary conditions are axis, interrior
and interface boundary conditions
You can also specify the fluuid type and
whether the domain is fluid or solid via the bc
panel

You might also like