Anda di halaman 1dari 12

International Journal on Architectural Science, Volume 4, Number 1, p.

24-35, 2003

COMPARISON OF FOUR ALGORITHMS FOR SOLVING PRESSUREVELOCITY LINKED EQUATIONS IN SIMULATING ATRIUM FIRE
R. Yin and W.K. Chow
Department of Building Service Engineering, The Hong Kong Polytechnic University, Hong Kong, China
(Received 14 January 2003; Accepted 25 February 2003)

ABSTRACT
Four algorithms for solving the pressure-velocity linked equation were compared by simulating the pressure
distribution in an atrium fire. The four algorithms studied were the SIMPLE, the SIMPLER, the SIMPLEC and
the PISO. Four heat release rates of fire in an arbitrary atrium were considered. Different under-relaxation
factors for velocity components were assigned. It is found that the flow variables predicted by the four
algorithms are the same, though the pressure distributions are quite different. Two sets of predicted pressure
were found. The SIMPLE/SIMPLEC/PISO resulted in one set and the SIMPLER resulted in another.
Macroscopic parameters useful to the fire industry were also computed.

1.

INTRODUCTION

Studying pressure distribution induced by fire in an


atrium is very important in understanding the
smoke filling process [e.g. 1]. Accurate prediction
on that enables the design of good smoke control
systems as it helps in determining the dimensions
and positions of the vents for natural ventilation,
sizing the smoke extraction fans for mechanical
ventilation, and estimating the pressure level
required to pressurize the adjacent area. It is
difficult and expensive to carry out full-scale
burning tests. Fire zone models [e.g. 2,3], though a
popular design tool, are not supposed to predict the
pressure distribution in the compartment. In fact,
constant compartment pressure was assumed in
some fire zone models. Applying Computational
Fluid Dynamics (CFD) (known as the CFD/Field
model) is a possible solution to understand the
pressure distribution in an atrium fire [e.g. 1]. The
CFD/Field model [4] for fire studies (to some
extent, indoor aerodynamics of buildings as well
[5]) has three main characteristics:
y
y
y

The k- model is commonly used to calculate


the average flow variables induced by a fire.
The finite control volume method is used to
discretize the set of conservation equations.
As there is no explicit equation on pressure,
algorithms for solving the pressure-velocity
linked equation is required.

Questions are usually raised on the above three


characteristics in fire studies. Preliminary
comparison of turbulence models and algorithms
for solving pressure-velocity linked equation have
been made [6,7]. However, results are not yet
convincing to recommend a suitable turbulence

24

model nor a numerical scheme for fire simulations.


This paper compares different algorithms for
solving the pressure-velocity linked equation in
simulating fire-induced air flow. The findings of
which would help to recommend a suitable
algorithm for calculating the pressure distribution
induced by an atrium fire using the CFD/Field
model.
The SIMPLE (Semi-Implicit Method for Pressure
Linked Equations) type of algorithm [8] is
commonly employed in CFD studies. The method
was proposed years ago [8] but is still being used
for simulating practical problems nowadays. In
this paper, the SIMPLE, the SIMPLER (SIMPLE
Revised) [9], the SIMPLEC (SIMPLE Consistent)
[10] and the PISO (Pressure-Implicit with Splitting
of Operators) [11] algorithms were tested for
studying the fire-induced air flow in an atrium in
order to get a better understanding of the pressure
distribution. In the past two decades, there have
been reports on comparing them on simulating
different problems of fluid flow [e.g. 10,12], but
not on computing fire-induced air flow.

2.

GOVERNING EQUATIONS

Steady-state two-dimensional simulations were


performed to save CPU computing time. The
equation describing the conservation of flow
variables such as momentum and energy is:

(u) + (v) =
( ) + ( ) + S
x
y
x
x
y
y
(1)

International Journal on Architectural Science

The u-momentum equation for the control volume


centered at e as in Fig. 1 is:
a e u e = a nb u nb + S uU + (PP PE ) A e
(5)

The finite difference control volume method with


Centered Grid Nodes was used to discretize
equation (1) to give:
(2)
a P P = a i i + SU
i

Pressure distribution P* was estimated from the


velocities u* obtained by solving the u-momentum
equation (5):

The coefficient a P at node P (Fig. 1 refers) for is


expressed in terms of the coefficient at
neighbourhood a i (i = E, W, N, S) and SP as:

a P =

a i SP

a eu*e =

(3)

Detailed derivation of the finite difference equation,


application of boundary conditions and the
treatment of non-uniform mesh have been
discussed elsewhere [e.g. 4,13,14] and would not
be repeated here. The CFD software concerned is
also described in the literature [15].

P = P P*

1 0
a PP

(7)

and
u = u u*

An under-relaxation factor was assigned to


ensure convergence:

a i i + SU +

(6)

Velocity u given by equation (6) would not satisfy


conservation of mass unless a correct pressure field
is employed. Velocity and pressure fields are
corrected by adding u and P:

i=E, W ,N, S

a P
P =

a nbu*nb + SuU + (PP* PE* ) Ae

(8)

Substituting equation (6) into (5) would relate P to


u:

(4)

a e u 'e =

a nb u 'nb + (PP' PE' ) A e

(9)

where 0P is the value of P from the previous step.

The pressure P and velocity u satisfying both the


mass and momentum constraints are

3.

u = u* + u

PRESSURE-VELOCITY LINKED
EQUATION

(10)

P = P* + P

The main objective of this paper is to compare four


algorithms for solving the pressure-velocity linked
equation. The four algorithms are the SIMPLE, the
SIMPLER, the SIMPLEC and the PISO.

(11)

Attention is now paid to the method for solving P


using the four algorithms listed in Appendices A to
D, with a flow chart shown in Fig. 2.

Control
volume for u

v(i,j)

Grid point
at (i,j)

w P

P(i,j)
k(i,j)
(i,j)
T(i,j)

u(i,j)

v
Control
volume for v

25

International Journal on Architectural Science

Fig. 1: Control volume and staggered grid system


Start
Estimate P*, u*, v*, *
Calculate coefficient of equations
SIMPLER
Algorithm?

Yes
Calculate pseudo-velocities
using equation (B2)

No

Solve pressure P using


equation (B4)
Solve momentum equation
using equation (6)
Solve pressure correction
using equation (A7)

No

SIMPLEC
Algorithm?

Set u* = u
v* = v
P* = P
* =

Yes
Update velocity using equation (10)

No

Update velocity using equation (C2)

SIMPLER
Algorithm?

Update pressure using equation (11)

Yes
PISO
Algorithm?

Yes
Set P** = P, u** = u, v** = v

No

Solve second pressure correction using


equation (D3)
Update pressure using
equation (D2)
Solve other variables
equations

Convergence?

Yes
End

26

No

International Journal on Architectural Science

Fig. 2: Summary of the four algorithms

4.

The discretized equations for momentum and other


flow variables were solved by using the TDMA
method. The number of iterations required to solve
the equations was difficult to determine. There
were some recommendations [10] for judging
when to terminate the iteration process in solving
problems in which momentum equations are not
coupled to a scalar variable. However, those
methods are not suitable for fire simulations
because the air velocity depends on the local
density and temperature. In this paper, the number
of iterations was set to 5 for solving the corrected
pressure equation and the temperature equation;
and it was set to 1 for solving the other equations.

NUMERICAL EXPERIMENTS

Numerical experiments were performed in a


section of an atrium of length 13.5 m and height 9
m as shown in Fig. 3. A heat source of length 1 m
and height 0.5 m was located at the center of the
atrium floor with fire of 4 heat release rates: 0.5
MWm-1, 1 MWm-1, 1.5 MWm-1 and 2 MWm-1.
There was a 2 m high soffit at the atrium ceiling
boundary. The atrium was divided into 32 parts
along the x-direction and 32 parts along the ydirection with a non-uniform grid system.
Under-relaxation factors used in solving the
equations are listed in Table 1.
Table 1: The under-relaxation factors for flow variables

Flow
variables

Turbulence
kinetic energy k

Turbulence
energy
dissipation

Temperature
T

Density

Turbulence
viscosity

Pressure
P

Underrelaxation
factor

0.7

0.7

0.5

1.0

0.6

0.7

2m

9m

2m

9 parts

4 parts
Free
boundary

12 parts

Heat source
5 parts

9 parts

2 parts

9 parts

10 parts

4m
9m
13.5 m
Fig. 3: Geometry for numerical experiments

27

International Journal on Architectural Science

under-relaxation factors of the velocity component


u and v were set to 0.5; the number of iterations
in the TDMA method in calculating flow variables
was set to 50; and the number of sweeps was set to
10000. The predicted results were roughly the
same as those predicted from the above numerical
conditions, with a maximum deviation less than
0.1%.

Even an Intel Pentium 166 MHz personal computer


is capable of carrying out two-dimensional
simulations. The associated CPU time required for
simulation with different algorithms were recorded.
The convergence criterion was 7 10-4 and the
maximum sweep was 8000. That means computing
would stop at 8000 sweeps even when the
maximum residential errors do not satisfy the
convergence criterion. Most cases satisfied the
convergence criterion within 4000 sweeps. The
additional 4000 sweeps were demonstrated to be
useless. A summary of the number of steps and
CPU time required for the four algorithms with
different relaxation factors for velocities u and v
is shown in Table 2.

Further, changing the number of iterations in the


TDMA method would not decrease the total
number of sweeps, but it would require longer
CPU time and would even give divergent results.
A possible reason is because it is a buoyancydriven air flow problem.
Setting the initial
conditions to static situation by assigning all flow
variables equal to the ambient conditions would not
give converged results by increasing only the
number of iterations in the TDMA without
correcting the temperature field.

Results of velocity vectors, pressure and


temperature contours are shown in Figs. 4 to 6.
Predicted results on the flow variables using the
four different schemes were basically the same
except for the pressure field. The maximum
deviation for velocity, temperature, turbulent
kinetic energy k and turbulent energy dissipation
computed from the four algorithms was less than
0.3%. However, a large deviation was found for
the predicted pressure field. It is observed that
using the schemes SIMPLE, SIMPLEC and PISO
gave a set of similar results; whereas using the
SIMPLER gave another set as shown in Figs. 5a
and 5b.

5.

PRESSURE DISTRIBUTION

Although the pressure Pij predicted at the node (i,j)


from these four algorithms were quite different, the
pressure differences P between adjacent cells
were quite the same. For the heat release rate of 1
MWm-1, values of the pressure Pi,j at nodes (10, 20),
(10, 21) and (9, 20) computed by using the
SIMPLE, SIMPLEC and PISO are:

Predictions were repeated for heat release rate of


500 kWm-1 to check the convergence criteria and
the number of steps in the TDMA methods. The

P10,20 = 11.94 Pa; P10,21 = 13.83 Pa; and P9,20 =


12.71 Pa

5 ms-1-1
5 ms

y/m

/m
y
3

10

x/m

x/m
Fig. 4 Velocity vectors -1
(with heat release rate 1M Wm )
Fig. 4: Velocity vectors (with heat release rate 1 MWm-1)

28

15

International Journal on Architectural Science

29

International Journal on Architectural Science

150oC

y/m

100oC
3

50oC

200oC

300oC

0
0

12

x/m
Fig. 6: Temperature contours (with heat release rate 1 MWm-1)

The pressure differences P1 and P2 among the


nodes (10, 20) and (10, 21); (9, 20) and (10, 20)
are:
P1 = P10,20 P20,21 = 1.89 Pa
P2 = P9,20 P10,20 = 0.77 Pa
Using the SIMPLER, values of the pressure at
nodes (10, 20), (10, 21) and (9, 20) are:
P10,20 = 13.81 Pa; P10,21 = 16.53 Pa; and P9,20 =
14.56 Pa
The corresponding pressure differences are:
P1 = P10,20 P10,21 = 1.88 Pa
P2 = P9,20 P10,20 = 0.75 Pa
It is observed that the pressure differences P1 and
P2 predicted from the four algorithms are similar.
A possible reason is the method used in calculating
the pressure field in the SIMPLER algorithm is
different from the other three algorithms. For the
SIMPLE/ SIMPLEC/PISO, the pressure field is
estimated first, then corrected by the pressurecorrection equations in the next iteration. Pressurecorrection is not needed in SIMPLER since
pressure is calculated directly from velocity field.

30

To solve the equations at the free boundary, instead


of setting the pressure on the free boundary to
certain values (such as setting directly to 0 Pa),
coefficients Spp and Sup in the discretized pressurecorrection equations given by (2) and (3) are
assigned as:
Spp = 1030
Snp = 0.0
Pressure will not be corrected under free boundary
conditions. If the velocity field is calculated from
the mass conservation equation, results might not
satisfy the momentum conservation equation. The
pressure calculated at those points using the
momentum conservation equation (for SIMPLER)
would be different from the pressure estimated
from the mass conservation equation (for
SIMPLE/SIMPLEC/PISO).
Inside the computing domain, velocities calculated
by different methods are the same, giving similar
pressure differences.

6.

ERROR ANALYSIS

Three kinds of errors should be considered [e.g.


16]: error in original data, truncation error and
round-off error.

International Journal on Architectural Science

Error in original data arose because the initial


conditions, particular values of k and have to be
estimated.
The solution would be changed
depending on the sensitivity of the equation
towards those parameters, and whether the errors
can be reduced by iterative methods. If the
equations are too sensitive to those parameters,
divergent solution would be resulted. Underrelaxation method might improve the situation.

Using the SIMPLEC algorithm would not


reduce the CPU time. This observation is
different from the conclusions reported by
Van Doormaal and Raithby [10]. Results
predicted by this scheme were slightly more
stable than those by using the SIMPLE.

Very stable results were predicted by using


the SIMPLER algorithm, but simulations with
this scheme took longer CPU time.

Truncation error is due to truncating some terms in


discretizing the equations.
First order discretization methods are applied in this study and so
truncation errors will be expected. However,
results are useful in estimating macroscopic
parameters.

Results were also stable by using the PISO


algorithm. In fact, this scheme is found to be
much better than the other three and it took
less CPU time.

Round-off error comes from floating-point


arithmetic because of the fixed word length in the
computer. A real number cannot be expressed
exactly because of the hardware limitation and the
associated errors are grouped under round-off error.
The round-off error is very small initially, but it
becomes bigger after several iterations. Finally,
the result may be deviated from the expected
values. This type of error is especially important
when subtracting two quantities of similar
magnitude.
Decreasing the under-relaxation
factors would improve the situation and so
selecting suitable under-relaxation factors is very
important.
Overflow or not converged results (labelled as
NC) were encountered in cases listed in Table 2.
By trial and error, 8000 sweeps was sufficient to
reduce error to a value less than the defined
residual errors, provided that the round-off errors
were small enough.

7.

COMPARISON
ALGORITHMS

OF

THE

FOUR

Different values of the under-relaxation factor for


the momentum equation u and v were tested
using the four algorithms for the four heat release
rates, and are shown in Table 2. Longer CPU time
was required for cases with smaller underrelaxation factors (up to 506 s for SIMPLER with
v and v of 0.2).

The under-relaxation factors u and v cannot be


too big nor too small. However, it is possible to
use larger values of u and v for the schemes
SIMPLER and PISO as shown in Table 2.

8.

MACROSCOPIC PARAMETERS

Macroscopic parameters useful to the fire industry


were calculated from the predicted flow variables.
These included the intake air flow rate Fin, hot air
out flowing rate Fout, average air temperature of the
atrium Tav and maximum height hmax of the neutral
plane (0 Pa pressure with respect to the ambient).
The air intake flow rate Fin was calculated from the
horizontal incoming air speed uiin and density i of
the ith control volume at the opening (i.e. x = 9 m):
Fin =

i u i in

(12)

The outflowing rate Fout of hot air was calculated in


a similar manner using the outflowing air speed
uiout:
Fout =

i u i out

(13)

The average temperature Tav was found from the


temperature Ti and the area Ai of the ith control
volume:

Ti A i
Tav = i
Ai

(14)

The followings are observed:


y

Results predicted by using the SIMPLE


algorithm were unstable, especially for large
value under-relaxation factors but it took
roughly the same CPU time as the other three
algorithms.

The maximum neutral plane height hmax was judged


from the predicted pressure distribution and
measured from the atrium floor.
The macroscopic parameters for the four heat
release rates were calculated and shown in Table 3.

31

International Journal on Architectural Science

In general, the inflow rate and average temperature


of the atrium increased as the heat release rate
increased. However, the neutral plane height
decreased to the floor level when the heat release
Table 3: Macroscopic parameters

rate increased. This point is very important in


understanding the smoke filling process in the
atrium.

Heat release
rate
/MWm-1

Inflow rate at
opening
Fin/kgs-1m-1

Outflow rate at
opening
Fout/kgs-1m-1

Average
temperature
of atrium
Tav/C

0.5
1.0
1.5
2.0

-6.88
-8.10
-8.73
-9.13

6.89
8.10
8.72
9.12

96
145
205
248

Maximum Height of Neutral Plane


hmax/m
SIMPLE/
SIMPLEC and
SIMPLER
PISO
1.58
2.50
1.21
1.92
1.06
1.60
0.96
1.43

3.

G.P. Forney and W.F. Moss, Analyzing and


exploiting numerical characteristics of zone fire
models, NISTIR 4763, National Institute of
Standards
and
Technology,
Gaithersburg,
Maryland, U.S.A. (1992).

4.

H.K. Versteeg and W. Malalasekera, An


introduction to computational fluid dynamics - The
finite volume method, Longman, Essex (1995).

5.

D. Etheridge and M. Sandberg, Building


ventilation: Theory and measurement, Wiley &
Sons, New York (1996).

6.

W.K. Chow and W.K. Mok, On the simulation of


forced-ventilation fires, Numerical Heat Transfer,
Part A: Applications, Vol. 28, pp. 321-338 (1995).

7.

Therefore, it can be concluded that both the PISO


and the SIMPLER are suitable numerical schemes
for solving the pressure-velocity linked equation in
atrium fire simulation.

W.K. Chow and Y.L. Cheung, Comparison of the


algorithms PISO and SIMPLER for solving
pressure-velocity linked equations in simulating
compartmental fire, Numerical Heat Transfer,
Part A: Applications, Vol. 31, No. 1, pp. 87-112
(1997).

8.

S.V. Patankar and D.B. Spalding, A calculation


procedure for heat, mass and momentum transfer
in
three-dimensional
parabolic
flows,
International Journal of Heat and Mass Transfer,
Vol. 15, pp. 1787-1806 (1971).

ACKNOWLEDGMENT

9.

S.V. Patankar, A calculation procedure for two


dimensional elliptic situations, Numerical Heat
Transfer, Vol. 14, pp. 409-425 (1984).

10.

J.P. Van Doormaal and G.D. Raithby,


Enhancements of the SIMPLE method for
predicting incompressible fluid flows, Numerical
Heat Transfer, Vol. 7, pp. 147-163 (1984).

11.

R.I. Issa and A.D. Gosman and A.P. Watkins,


The
computation
of
compressible
and
incompressible recirculating flows by a noniterative
implicit
scheme,
Journal
of
Computational Physics, Vol. 62, pp. 66-82 (1986).

12.

D.S. Jang, R. Jetli and S. Acharya, Comparison of


the PISO, SIMPLER, and SIMPLEC algorithms
for the treatment of the pressure-velocity coupling

9.

CONCLUSIONS

Numerical experiments in an atrium with four heat


release rates were performed to test four algorithms
for solving the pressure-velocity linked equation.
The four algorithms are the SIMPLE, the
SIMPLER, the SIMPLEC and the PISO. The
under-relaxation factors were tested by applying
each scheme. A suitable algorithm is recommended
for solving the pressure-velocity linked equation in
atrium fire simulations.
It is found that all the four schemes gave the same
results on flow variables except for pressure. The
SIMPLE, the SIMPLEC and the PISO gave similar
pressure predictions but the SIMPLER gave
another set of results. Further, the PISO gave more
stable results and took less CPU time.

The paper is jointly funded by a PolyU-ASD


account A-038 and B-Q063 from Research Grants
Council of Hong Kong.

REFERENCES
1.

W.K. Chow, Simulation of fire environment for


linear atria in Hong Kong, ASCE Journal of
Architectural Engineering, Vol. 3, No. 2, pp. 80-88
(1997).

2.

J. Quintiere, Fundamentals of enclosure fire


zone models, Journal of Fire Protection
Engineering, Vol. 1, No. 2, pp. 99-119 (1989).

32

International Journal on Architectural Science

13.

14.

in steady flow problems, Numerical Heat


Transfer, Vol. 10, pp. 209-228 (1986).

where

W.K. Chow and N.K. Fong, Application of field


modelling technique to simulate interaction of
sprinkler and fire-induced smoke layer,
Combustion Science and Technology, Vol. 89, pp.
101-151 (1993).

aE = (Ad)E ; aW = (Ad)W ; aS = (Ad)S ;


aN = (Ad)N

(A8)
(A8)

aP = aE + aW+ aN + aS

(A9)

W.K. Chow and W.M. Leung, A short note on


achieving convergent results in simulating building
fire using the k- turbulent model, Numerical
Heat Transfer, Part A, Vol. 17, pp. 495-501 (1990).

S pU = (u*A)w (u*A)e + (v*A)s (u*A)n (A10)

15.

R. Yin and W.K. Chow, Studies on thermal


responses of sprinkler heads in atrium buildings
with fire field models, Fire and Materials, Vol. 25,
No. 1, pp. 13-19 (2001).

16.

C.F. Gerald and P.O. Wheatley, Applied numerical


analysis, 5th edition, Addison Wiley, New York
(1994).

APPENDIX A: REVIEW
SIMPLE ALGORITHM

OF

THE

The exact equation for P derived from equation (9)


is not suitable for quick calculation. In the
SIMPLE procedure, u e is given by:

Values of P might be too large in this


approximation, giving slow convergence rate or
even divergence would be resulted during
computation. An under-relaxation factor is
employed in the momentum and pressure equations
to avoid that.
Summary of the SIMPLE procedure:
Estimate a pressure field P*.
Evaluate the coefficients of the momentum
equations given by equation (6) and solve for
u* and v*.
Evaluate the source term given by equation
(A10) and solve for P by equation (A7).
Correct the velocity field given by equation
(10) and pressure field given by equation (11).
Solve for other equations and update the
coefficients.
Take the corrected P as new P*, repeat step (2)
to step (6) until convergence.

1.
2.
3.
4.
5.

u e = u e1 + u e 2

(A1)

P = P1 + P2

(A2)

where
u e1 = (PP 1 PE 1 )

u e 2 =

APPENDIX B: REVIEW
SIMPLER ALGORITHM

Ae
ae

P2

PE 2 )

Ae
ae

(A4)

THE

u e = u *e + d e (PP PE )

Using the SIMPLER algorithm, the pressure field


is calculated from the velocity field which is also
corrected.
The momentum equation is expressed as:

First order approximation would give zero value of


u e 2 :
(A5)

ue

a nb u nb + S uU A e

=
+
(P

Ae
ae

(A6)

ae

ae

PE )

(B1)

u e is defined as:
The pseudo-velocity ~

where
de =

OF

(A3)

a nb u nb + (P
ae

6.

~
ue =

a nb u nb + S uU
ae

(B2)

Putting equation (B2) into equation (B1) gives:


Putting equation (A5) into the continuity equation
gives:
+ a N PN + a S PS + S pU
a P PP = a E PE + a W PW

ue = ~
u e + d e (PW PE )

(B3)

(A7)

33

International Journal on Architectural Science

Substituting the above equation into the continuity


equation, the pressure equation can be derived
from equation (A7):
a P PP = a E PE + a W PW + a N PN + a S PS + S PU

(B4)

de =

Ae

(C3)

a e a nb

However, P should not be under-relaxed.

where
aP = (Ad)E ; aW = (Ad)W ; aS = (Ad)S ;
aN = (Ad)N

(B5)
(B5)

aP = aE + aW + aN + aS
S PU = (~
uA) w (~
uA) e + (~
vA) s (~
vA ) n

(B6)
(B7)

APPENDIX D: REVIEW OF THE PISO


ALGORITHM

The coefficients of equation (B5) are the same as


those in the pressure correction equation (A7).
However, the source term is evaluated using the
pseudo-velocities.

The major difference between the SIMPLE and the


PISO is on the pressure field correction. In the
SIMPLE, only one correction is used to update the
pressure field. However, two corrections are
introduced in the PISO.

The SIMPLER procedure is summarized as:


1.
2.
3.
4.
5.
6.
7.
8.

Evaluate the coefficients of the momentum


equations given by equation (6).
Calculate pseudo-velocities using equation
(B2).
Solve for pressure field P* given by equation
(B4).
Solve for u* and v*.
Evaluate the source term in equation (B7) and
solve for P given by equation (A7).
Correct the velocity field given by equation
(10).
Solve for other equations; update the
coefficients.
Repeat step (1) to step (8) until convergence.

APPENDIX C: REVIEW
SIMPLEC ALGORITHM
The term

a nb u nb

OF

THE

in equation (9) is neglected in

the SIMPLE algorithm. To introduce a consistent


approximation, the term
a nb u e is subtracted

from both sides of the equation to give:


(a e

a nb )u 'e = a nb (u 'nb u 'e ) + (PP' PE' ) A e


(C1)

In

the SIMPLEC approximation, the term


a nb (u nb u e ) in equation (C1) is neglected.

The velocity correction equations is changed to:

u e = u*e + de (PP PE )
where

34

(C2)

After correcting the velocity and pressure fields,


equations (10) and (11) become:
u = u* + u = u* + u1 + u2 = u** + u2

(D1)

P = P * + P = P * + P1 + P2 = P ** + P2

(D2)

Substituting the above equations into the


discretized momentum equations and the mass
equation. The equation for the second pressure
correction is:
+ a N PN + a S PS + b
a P PP = a E PE + a W PW

(D3)

where
aE = (Ad)E ; aW = (Ad)W ; aS = (Ad)S ;
aN = (Ad)N

(D4)
(D4)

aP = aE + aW + aN + aS

(D5)

A
A
) W a nb unb ( ) E
a
a
A
A
( )S a nb vnb ( ) N
a
a

b=(

a nbunb +

(D6)

a nb vnb

Applying the above method, the second correction


pressure field can be calculated from equation (D2).

International Journal on Architectural Science

35

Anda mungkin juga menyukai