Download as pdf or txt
Download as pdf or txt
You are on page 1of 29

Lecture 22:

SIMPLEC Algorithm
Application to Body-Fitted Meshes
Introduction to Co-Located Schemes
Last Time …

z Did example problems using SIMPLE


z Started looking at improvements to SIMPLE
» SIMPLER
This Time …

z Look at another variant of the SIMPLE family of


pressure-based schemes
» SIMPLEC algorithm
z Consider how to apply staggered mesh ideas to body-
fitted meshes
z Discuss difficulties in applying staggering ideas to
unstructured meshes
z Start looking at co-located schemes and how to avoid
checkerboarding
SIMPLEC Algorithm

z SIMPLE-Corrected (SIMPLEC) algorithm seeks to


mitigate the effects of dropping velocity neighbor
correction terms:
SIMPLEC Velocity Correction Equation

z SIMPLEC retains neighbor velocity correction terms,


but makes an approximation:

z Thus
Velocity Correction (Cont’d)

z Therefore, re-define d coefficients:

Need under-relaxation in
the momentum equation
to avoid division by zero
for Sp =0, steady flow
Discrete Pressure Correction Equation

aP p = ∑ anb p + b
'
P
'
nb
z Form of equation is
nb the same as for
SIMPLE
z b term is still the
amount by which the
starred velocities do
not satisfy continuity
z Only d coefficients
are different from
SIMPLE
SIMPLEC Algorithm

z The overall algorithm is identical to that for SIMPLE


z Only the d coefficients used to drive the p’ equation
are different.
z With SIMPLEC, because the ∑ a u '
nb nb nb and ∑ a v '
nb nb nb

are not dropped, there is no need to under-relax the p’


correction. Thus:
SIMPLEC Performance

z SIMPLEC is found to converge faster that SIMPLE


» Gains of 20-30% can be found for many problems
» Does not solve extra equations like SIMPLER
» Cost per iteration approximately the same as
SIMPLE
z However, like SIMPLE, the SIMPLEC algorithm will
also destroy a good velocity field unless there is also a
good pressure guess.
Curvilinear/Body-Fitted Meshes

z For the last few lectures, we were looking at path-to-


solution issues
z Let’s now go back to the issue of staggered storage
and see whether it can be applied to body-fitted
meshes
z What coordinate system to use ?
» Cartesian velocity components?
» Other?
z Let’s see what happens if we use Cartesian
components of velocity on a curvilinear mesh
Curvilinear Meshes (Cont’d)

Grid-Following
Velocity
One Cartesian
Component on
Face
Covariant Velocity System

z Look at curvilinear coordinate system


z Covariant velocity components are components in the
(ξ,η) directions
z Store velocity parallel to ξ direction on ξ=constant face
z Similarly for η
Contravariant Velocity System

z Contravariant velocity components are components


perpendicular (ξ,η) directions
Advantages

z Can use either co- or contravariant velocity


components in a staggered system
z Only one velocity component per face
z Non-overlapping staggered control volumes can be
defined just as for Cartesian meshes
z Can devise SIMPLE family algorithms for solution
z Widely used in the 80’s and 90’s.
Disadvantages

z Major disadvantage is complexity


z Because these velocity components are curvilinear,
recasting momentum equations into these
components generates complicated curvature terms
» Result of the fact that linear momentum is
conserved; hence Cartesian velocity momentum
equations can be written in conservation form
» Curvilinear velocity momentum equations always
have extra terms because “curvilinear” momentum
not conserved!
Disadvantages (Cont’d)

z Need Cartesian velocity gradients for turbulence


production terms, viscosity for complex rheologies,
etc.
» Difficult to cast these terms in curvilinear
coordinates
» Typically need to carry both curvilinear and
Cartesian components
» Find Cartesian components from curvilinear
components each iteration and then find production
terms etc.
Unstructured Meshes

z Generally difficult to apply staggering to unstructured


meshes
» Geometric complexity too great
z Many finite element methods and node-based finite
volume methods use unequal-order interpolation
» Some node-based finite volume methods store
pressure on a coarser mesh than velocity
» Some FEM use lower-order interpolation functions
for pressure compared to velocity (p linear; v
quadratic)
Unequal-Order CVFEM

z Baliga and Patankar (1983)

• One-fourth fewer
pressures than
velocities in 2D –
accuracy?
• How to chop up
tetrahedra in 3D?
Co-Located Storage

z Store (u,v,P,φ) at cell


centroid
z Same control volume
for all
z Also sometimes called
equal-order
interpolation or a non-
staggered formulation
Discrete Momentum Equation

z Let’s do this in 1D to avoid algebra; assume uniform


mesh
z Discrete momentum equation for velocity up:

We are writing an
equation for uP but
we need ue
Momentum Equation (Cont’d)

z Interpolating pressure linearly, we get:


Pressure Checker-Boarding

z Divide momentum equations by center coefficients:

z Here:
Pressure Checker-Boarding (Cont’d)

z Interpolating from momentum equations:

Any pressure field in which pW = pP and pWW = pP =


pEE is seen by these equations as a uniform field
Discrete Continuity Equation

z Discretizing continuity equation over control volume:

z Interpolate linearly:

z Obtain: Supports velocity


checkerboarding
Co-Located Formulation

z Idea is to write face velocities ue and uw in terms of


adjacent pressures rather than alternate pressures
z If we interpolated cell velocities to the face linearly, we
would get:

z Instead, we use
Co-Located Formulation (Cont’d)

z Velocity interpolation to the face:

z We are subtracting the ‘alternate’ pressure term and


adding in an ‘adjacent’ pressure term
Discussion

z Face velocities are designed to be driven by adjacent


pressure differences, not alternative ones
z Thus, these velocities are sensitive to checker-
boarding and will not allow a checker-boarded
pressure field
z The momentum equation still supports a checker-
boarded pressure field, but since the continuity
equation does not, the final solution is not checker-
boarded.
Discussion (Cont’d)

z Another way to think about this is to think of ue as a


sort of staggered velocity
z Instead of creating a staggered control volume and
associated geometry, uˆe , uˆw are interpolated from
cell values, and an ”adjacent” pressure term is added
on
z This type of interpolation is called “momentum
interpolation”; sometimes also called an “added-
dissipation” scheme
z Similar ideas used in the compressible flow literature
Closure

In this lecture, we
z Looked at improvements to the SIMPLE algorithm
» SIMPLEC
z Considered covariant/contravariant velocity forms for
use with staggered body-fitted meshes
» Too cumbersome
z Saw that there was no easy way to apply staggering to
unstructured meshes
z Started looking at co-located schemes

You might also like