Mesh Generation with Gambit

Read the documentation!

Fluent has very good on-line
documentation
Gambit is not as good
Dont just read the documentation
but go to library and get hold of
some of the journal references
Mesh generation
Generating a good mesh is a large part
of the CFD problem
Generating a good quality mesh can be
hard work
What is a satisfactory mesh for a
problem will not automatically be so
when another model option is enabled
Grid Types
Fluent can solve flow problems in 2d or
3d
The cyclone simulation discussed in the
introductory lecture is a typical 3d
problem
The 2d channel is a typical 2d problem
2d Grids
consume much less
memory and take much
less time to solve
a 2d grid intrinsically
assumes that there are no
velocity gradients in the
direction normal to the grid
This assumption is really
only valid is the channel is
extremely wide in z
u/z =0
z
y
x
2d Grids
Can also be used
with an axisymmetric
flow such as swirling
flow in a centrifuge
again the
assumption of zero
velocity gradient
normal to grid
applies, but this is in
the direction

z
y
x
2d grids - types
mapped (preferred
if it possible to do
so)

paved with
triangles or quads
(unstructured
solver only)
2d quad mapped
2d tri paved
2d quad paved
2d grids
2d grids are drawn on faces in Gambit
2d face grids are used as a precursor to
creating 3d grids
3d grids
The grid you will generate for a full 3d
simulation of the flow
obviously consumes more memory and
takes longer to solve
generated for a volume in Gambit
mapped preferred.
complex 3d grids such as Cooper and t-
grid are unstructured grids
Structured and Unstructured
Solvers
Structured Solver (Fluent 4.5)
uses i,j,k indexing to locate neighboring cells
grid must be structured and based on a topologically
rectangular block in computational space
but this doesnt mean the grid cant be curved
all grid lines continue from one boundary to its opposite
Unstructured Solver (Fluent 5.0)
uses an internal data structure to link cells and faces
doesnt force an overall topology on the grid so mesh
volumes can be any shape
doesnt use indexing to locate neighboring cells
can read a structured grid
Structured and Unstructured
solvers
Structured solvers were written first because they
were easy to write
unstructured solves are more modern
however Fluent 4.5 (structured) and Fluent 5.0
(unstructured) had different sets of models
eg Fluent 5.0 doesnt have the Eulerian granular flow
model (4.5 does) but it is much easier to set up 5.0
for moving mesh problems. 5.0 has Large Eddy
Simulation
Fluent 6 is supposed to merge the two approaches
together
Structured Grids
Topologically rectangular
This means that the mesh volume is a
quadrilateral in 2d or a hexahedron in 3d
Each mesh volume is linked only to its
immediate neighbors
But the edges can be mapped around
curves and mesh volumes dont have to
be the same size
Reduces storage and CPU requirements
Solid regions like tank baffles can be
generated by blanking those mesh points
which overlap the solid region and making
them dead zones. Fiddly but still
generates a good grid
dead zones waste storage.
Unstructured Grids
Mesh volumes can be linked to
any other volume in the domain
And can be any shape
less computationally efficient than
a structured grid
but can still read a structured grid
topology (often still the best)
can use non-conformal grids
introduces flexbility but this
flexbility creates problems
Unstructured Grid Topologies
3d hexahedron
2d Prism
2d Triangle
Tetrahedron Pyramid Prism
Unstructured grids - Mapped
If you can generate a mapped mesh for
the domain then do so
Unstructured grids - Tgrid
Uses a 3d paving algorithm
uses the tetrahedron and pyramids
very easy to generate a grid for
complex geometry
but is inefficient and has problems of
numerical diffusion and skewness

Unstructured grids - Cooper
mesh
This grid projects a set of face meshes from one side
of the domain to another
the face meshes that are projected may be mapped
or paved
but the faces surrounding the ray of projection must
be mapped and must have the same number of grid
points along the ray
can lead to very small mesh volumes at flow
constrictions
but a better alternative to the tgrid if you can work
out how to generate it.
Pros and Cons
You will be using an unstructured solver (and grid) anyway
The unstructured tetrahedral mesh (tgrid) is the easiest mesh to
generate for a complex geometry
But it is inefficient in that more mesh points are often generated
compared to a hex mesh
it can also cause numerical problems because of skewness
and because the faces are not automatically aligned with the
flow you can get false diffusion. This can reduce the accuracy of
the problem
Even if it requires more work to generate a quadrilateral or
hexahedral (and cooper) mesh may give better results in the
final simulation

What is a good grid?
A high quality mesh
smooth variations in grid
spacing
minimal grid skewness (cell
interior angles equal
graded towards boundaries
and other areas where rapid
variations in parameters
occur
sometimes all this is hard to
achieve particularly with a
tetrahedral mesh
Avoid:
Skew
Large Volume changes
High aspect ratio
And try to align the grid with flow
streamlines
Assessing grid quality
Default measure of quality
is EquiAngle Skew
This should be minimised
Definition depends on cell
geometry
Hex and quad < 0.7
Tets < 0.8
Tri < 0.7
The tetrahedral mesh will
sometimes not meet this
criteria in all parts of the
domain

Hex or quads a
max
a
min
skew = max[ (a
max
90)/90, (90 a
min
)/90]
Tris or tets
Skew = Optimal face/cell size face/cell size
Optimal face/cell size
Optimal face
Actual face
Gambit - bottom up approach
create vertexs
link vertexs to make edges
create faces from edges
create volumes from faces (3d)
mesh edges
mesh faces
mesh volumes (3d)
Gambit - top down approach
create volumes or faces directly
generate complex 3d volumes by
uniting and splitting volumes
mesh volumes or faces directly
this can be a good shortcut way to
create the geometry and/ora grid
but Gambit may default to the t-grid
option when the grid is generated