School of Mechanical Aerospace and Civil Engineering

The equations we need to solve, are then

(U V ) P (U U ) + = + x y x x U x V x + y y U y V y + Su
(1a)

Pressure-Velocity Coupling
T. J. Craft George Begg Building, C41
Contents: Introduction to CFD Numerical solution of equations Finite difference methods Finite volume methods Pressure-velocity coupling Solving sets of linear equations Unsteady problems Turbulence and other physical modelling Body-tted coordinate systems Reading: J. Ferziger, M. Peric, Computational Methods for Fluid Dynamics H.K. Versteeg, W. Malalasekara, An Introduction to Computational Fluid Dynamics: The Finite Volume Method S.V. Patankar, Numerical Heat Transfer and Fluid Flow Notes: http://cfd.mace.manchester.ac.uk/tmcfd - People - T. Craft - Online Teaching Material

(U V ) (V V ) P + = + x y y x (V ) (U ) + =0 x y

+ Sv

(1b)

(1c)

where Su and Sv represent any other source terms that may be present (from buoyancy, rotation, turbulent stresses, etc). Before considering how to obtain the pressure eld, we rst examine alternative storage arrangements that can be used for the discretized ow variables, since these have a bearing on how source terms are represented, and thus on how the schemes to be described may be implemented.

- p. 3

Overview
Up to now, we have mainly considered discretization schemes, and how they are used for a single equation. In practical ow solvers, we have to solve a set of coupled pdes. In 3-D, this involves determining U , V , W and P , as well as any additional quantities (temperature, enthalpy, turbulence energy, etc) that may be required in the problem. We have already examined how to discretize the momentum equations and solve them to obtain U , V and W . As explained, this can be done in an iterative fashion, employing under-relaxation if necessary, to account for the coupling between the equations. However, we do not have a transport equation for the pressure P . Instead, the 4th equation we have is the continuity equation which does not even contain the pressure. In this lecture we are thus going to consider a strategy for handling the coupling between the velocity and pressure elds in a ow computation. For now, we restrict attention to 2-D ow, consider a steady-state ow, and work within a nite volume framework on a simple rectangular mesh. This is sufcient to demonstrate how schemes for coupling the pressure and velocity can be developed, and the same methods can readily be extended to 3-D and more complex problems.
- p. 2

Discretization
The U momentum equation can be integrated over the control volume shown below, resulting in a discretized equation of the form
au p Up = X au i Ui + Sup + Su
(2)

where Sup are the source terms arising from integrating the pressure gradient over the cell, and Su represents any other source terms. One convenient arrangement for storing the discretized variables is to use a single set of control volumes, so all variables are stored at the same locations (colocated). This has the advantage that all geometrical data only needs to be stored once. However, as outlined below, this arrangement does also have some disadvantages.
W w P N

If P is stored at the same position as U , we need to interpolate between PP and PE to get Pe , the pressure at the east face of the cell (and similarly for the west face). The source term Sup can then be approximated as
Sup = (P w P e ) (xy ) x
(3)
- p. 4

On a uniform grid, this interpolation gives Pw Pe = 0.5(PW PE ), so there is a relatively weak linkage between the velocity and local pressure eld (as PP is not used). This can lead to chequerboarding: a pressure eld that oscillates from node to node on the grid can still appear as uniform to the discretized momentum equation. (PE and PW can be the same value, for example, whilst PP could be signicantly higher or lower).

Pressure Correction Schemes

The problem in obtaining the pressure eld arises because, although we have 3 equations for U , V and P , the continuity equation does not explicitly contain P . The momentum equations provide us with a set of discretized equations which can be solved for U and V , if we know the pressure eld. However, we cannot use the continuity equation directly to obtain P . Instead, we consider how an iterative procedure can be used to adjust the pressure eld in order to ensure that the resulting velocity eld does satisfy continuity. Such schemes are generally referred to as Pressure Correction schemes. At each iteration a correction to the pressure distribution is calculated, designed to drive the local velocity eld towards one that satises both momentum and continuity equations.

i-2

i-1

i+1

i+2

Methods are available to overcome the above problem. However, for now we consider an N alternative arrangement that results in a stronger linkage between adjacent pressure and velocity values.
w P e E

- p. 5

- p. 7

A much stronger coupling between the velocity and pressure eld nodal values is obtained by using a staggered grid. In this arrangement the velocity components are stored at the centres of the faces of the pressure control volume. The disadvantage of this is that there are separate control volumes for U , V and P , so more geometrical information has to be stored.
PW Uw

The SIMPLE Scheme

PN Vn PP Vs PS Ue PE

A widely-used scheme for coupling the pressure and velocity is the SIMPLE (Semi Implicit Method for Pressure Linked Equations) scheme of Patankar (1980), which is outlined in the following. The discretized momentum equations for Ue and Vn are written as:
au e Ue = X au i Ui + Sup + Su av n Vn = X av i Vi + Svp + Sv
(5)

Such overheads become particularly cumbersome in non-orthogonal and 3-D grid arrangements. The pressure gradient term in the equation for Ue is now simply
Sup = (P P P E ) (xy ) x
(4)

Note that, because of the staggered grid, Ue denotes a nodal value of the U velocity, and Vn a nodal value of V . The summation on the right hand sides represents the contributions from the surrounding nodal values, Sup and Svp are the pressure-related source terms, and Su and Sv any other source terms. After dividing through by the diagonal coefcient, the discretized momentum equations can be written in the form
Ue = X au Ui i + Du (PP PE ) + Su au e X av Vi i + Dv (PP PN ) + Sv av n
(6a)

so there is a direct coupling between the velocity and the pressure values at adjacent nodes. For now, we thus work within a staggered grid arrangement. In the CFD II course we do consider how to handle the case when all quantities are stored at the same location.
- p. 6

Vn =

(6b)

v where Du = y/au e , Dv = x/an .

- p. 8

Starting off with some initial values for the pressure eld, the above equations can be solved to obtain U and V . However, these will not, in general, satisfy the continuity equation. Now suppose we add corrections U , V , P to the velocities and pressure so that the corrected variables
U = U + U V =V +V P = P + P
(7)

In the SIMPLE scheme, the two parts of the corrections are dened such that
Ue 1 = Du (PP 1 PE 1 ) Ue 2 =

X au U i i + Du (PP 2 PE 2 ) au e

(11a)

Vn 1 = Dv (PP 1 PN 1 )

Vn 2 =

X av V i i + Dv (PP 2 PN 2 ) av i

(11b)

satisfy both the momentum and continuity equations. By substituting these expressions into the discretized momentum and continuity equations, we now attempt to solve for the corrections U , V , and P . Substituting these new values into the discretized momentum equations gives:
X au i )= (Ui + Ui ) + Du (PP PE ) + Du (PP PE ) + Su (Ue + Ue au e X av i (Vn + Vn )= (Vi + Vi ) + Dv (PP PN ) + Dv (PP PN ) + Sv av n
(8a)

, V and P Note that the choice of this split ensures a very simple linkage between U1 1 1 values. The more complicated term involving the summation of velocity corrections over neighbouring nodes is put into the second part of the correction.

Since we want the corrected velocities to satisfy continuity, we now examine the discretized continuity equation. Integrating the continuity equation over the pressure control volume leads to
w Uw )y + (n Vn s Vs )x = 0 (e Ue

(12)

or
w Uw )y + (n Vn s Vs )x = Sm (e Ue

(8b)

(13)

where Sm = (e Ue w Uw )y + (n Vn s Vs )x is simply the mass imbalance arising from the original U , V eld.
- p. 9 - p. 11

Subtracting equations (6) from these, gives

= Ue

X au U i i + Du (PP PE ) au e X av V i i + Dv (PP PN ) av n

(9a)

, V , P ) As a rst approximation, we now assume that the second part of the corrections (U2 2 2 may be neglected. Substituting equations (11) into the discretized continuity equation then results in

Vn =

(9b)

y [Du ]e (PP 1 PE 1 ) [Du ]w (PW 1 PP 1 )

which provides relations between the corrections to the pressure and velocity that ensure the momentum equations are still satised. To simplify the analysis, we now split the velocity and pressure corrections into two parts: + U etc. Then we can write U = U1 2
Ue 1 + Ue2

+ x [Dv ]n (PP 1 PN 1 ) [Dv ]s (PS 1 PP 1 ) = Sm

(14)

The above equation can then simply be written in the form

a p PP 1 = a e P E 1 + a w P W 1 + a n P N 1 + a s P S 1 + Su

(15)

where
ae = y [Du ]e aw = y [Du ]w an = x[Dv ]n S u = Sm as = x[Dv ]s

X au U i i = + Du (PP 1 PE 1 ) + Du (PP 2 PE 2 ) au e X av V i i + Dv (PP 1 PN 1 ) + Dv (PP 2 PN 2 ) av n

(10a)

ap = ae + aw + an + as
(10b)

Vn 1 + Vn2 =

can be solved, using methods This set of linear equations for the pressure correction P1 considered in earlier lectures. and V , are then obtained from The corresponding corrections to the velocities, U1 1 equations (11), and the pressure and velocities are thus all updated.
- p. 10 - p. 12

Since the SIMPLE scheme is typically embedded within an iterative ow solver, the whole algorithm is: 1. Start with initial values 2. Solve momentum equations to get U , V 3. Calculate coefcients and source terms for the pressure correction equation (14)
4. Solve for the pressure corrections P1 , V 5. Calculate velocity corrections U1 1

These expressions are then substituted into the discretized continuity equation, which now becomes (17) (e Ue 2 w Uw2 )y + (n Vn2 s Vs2 )x = 0 We thus get
y [Du ]e (PP 2 PE 2 ) [Du ]w (PW 2 PP 2 ) + x [Dv ]n (PP 2 PN 2 ) [Dv ]s (PS 2 PP 2 ) = " # # " X au U X au U i i1 i i1 y y au au p p w e " # " # X av V X av V i i1 i i1 + x x av av p p
n

(18)

6. Update P , U , V 7. Repeat from step 2 until solution has converged

This can again be written in the generic form

a p PP 2 = a e P E 2 + a w P W 2 + a n P N 2 + a s P S 2 + Su

(19)

correction, and the where the ai coefcients are the same as those in the equation for the P1 source term Su is now given by the right hand side of equation (18) above.

In the PISO scheme, therefore, between steps 6 and 7 of the SIMPLE algorithm, the source is solved and the result used as terms Su of equation (19) are calculated, the equation for P2 a second correction to the pressure.
- p. 13 - p. 15

The PISO Scheme

One problem with the SIMPLE scheme as described is that it can be rather slow to converge.
, V and P . One reason for this is the neglect of the corrections U2 2 2

There are a number of variants on the SIMPLE scheme, designed to improve convergence speeds. One such variant, which is very similar in structure, and may sometimes lead to better convergence, is the PISO (Pressure Implicit solution by Split Operator method) scheme proposed by Issa (1982). In the PISO scheme, the same decomposition of corrections to the velocity and pressure is , V and P are computed as described earlier. made as in the SIMPLE scheme, and U1 1 1 However, a second corrector stage is now added, in an attempt to account for the neglected , V and P in the rst stage. U2 2 2 In this second corrector stage the second parts of equations (11) are approximated by
Ue 2 =

X au U i i1 + Du (PP 2 PE 2 ) au e X av V i i1 + Dv (PP 2 PN 2 ) av i

(16a)

Vn 2 =

(16b)
- p. 14