You are on page 1of 7

AND

6UILDII IG
ELSEVIER

Energy and Buildings 23 (1995) 33-39

CFD modelling of the air and contaminant distribution in rooms


Youchen Fan
VTT Building Technology, Indoor Environment and Systems, Box 1804, 02044 VT"T, Espoo, Finland
Received 12 August 1994; in revised form 18 December 1994; accepted 26 December 1994

Abstract

The k-E model is a wid!ely used model in engineering practice in handling indoor air quality problem. However, difficulties
may arise when using the high Reynolds number k-e model to simulate air flow patterns close to the boundaries of air and
the stagnant component as well as the low air flow fluctuation elsewhere in a room. When using the k-E model for low
Reynolds number cases, the correlations between turbulent coefficients and turbulent Reynolds number must also be defined.
By using the so-called Kolmogorov micro scale method, a new set of turbulent coefficient functions was deduced in this paper
for the k-e model in a case of low Reynolds number flow. Using the standard wall function leads to large differences between
the measured and calculated heat transfer coefficient. A special wall function valid for a viscous sublayer, a buffer zone and
a fully turbulent log-law zone is recommended in this paper. In addition, the modelling of air terminal devices in CFD
simulations is summarized by using a literature collection.
Keywords: Computational fluid dynamics modelling; Air contaminants; Indoor environment

1. Introduction

Computational fluid dynamics (CFD) modelling and


scale (or physical) modelling are, in principle, the two
alternative approaches for evaluating the global indoor
environments with the .aim to obtaining a proper design
of building service systems. Both approaches have their
advantages and limitations. CFD has, nevertheless, become a widely accepted alternative to the scale modelling
in recent twenty years for predicting the air motion in
buildings and indoor air quality. The main obstacles
in using CFD modelling', for HVAC research are assumed
to be on how to handle the mix-forced (momentum
and buoyancy) air flow and simulate the occupantbehaviour-related (OER) factors. Numerical problems
relating to the boundary layer flow are of secondary
importance in CFD application of HVAC engineering.
This is due to the fact that the CFD models were
originally developed by assuming the flow to be of a
high Reynolds number (HRN) type. This is not the
case particularly in modern buildings. Although a few
low Reynolds number (LRN) models already exist,
there is still a need to develop these models. For
simulating the OBR factors, wider use of statistical
mathematics will be required. At the moment, the
simulation research of OBR factors based on statistical
mathematics is totally missing.
0378-7788/95/$09.50 1995 Elsevier Science S.A. All rights reserved
SSDI 0378-7788(95)00916-L

The k-e model has been used worldwide since it was


created. However, for the H R N model, the so-called
wall function must be properly designed to correlate
the physical parameters between the LRN and H R N
flow zones. It has so far been believed that the wall
function has a large influence on the heat transfer near
to the solid boundaries. Thus, the problem is now to
find the correct wall function if it is needed. In addition,
for the LRN model, the relationship between turbulent
coefficients and turbulent Reynolds number was not
totally known until now. Another consideration with
respect to the LRN model for room air flow is that
these kinds of models, in fact, are trying to handle the
low turbulent Reynolds number effect both on the
boundary flow and on low fluctuation flow elsewhere
within the room. New findings for the above problems
will be addressed in the following sections. The research
method involved theoretical analysis and literature collections.

2. Theoretical analysis of the LRN model


The specific property of the k-e model refers to its
closure equations of Reynolds stress, i.e., the turbulence
kinetic energy and its dissipation rate transportation
equations. The general form of transportation equations
of k and e which are derived from Navier-Stokes (N-S)

34

Y. Fan I Energy and Buildings 23 (1995) 33-39

equations are as follows [1]:


k-equation
Dk
Dt

--

1_0
/x, +/~
r(
p axj I_\Cr~

Substituting Eqs. (6) and (7) into Eq. (5), gives

+ . , 0u,

/3 ~, OH
+

--

--gi--

o il
(1)

ar,

K2

C1 = C2

e-equation

r,C,f 2

(8)

in which K= Karman constant ( = 0.4), p, = 1.3 [3] and


2.0 > C2 > 1.71 [2].
By considering the influence of the turbulent Reynolds
number R,, the coefficient C2 has been interpreted as
a function of Rt [4] as follows
C2 = C20[1 - 0.3 exp( -Rt2)]

D--7 = p ~

~+/x

ax,j

(9)

Substituting Eq. (9) into Eq. (8), Ca can then be


expressed by

where

K2

Ca=C2o o.C~a
-c2

/3 e iz, OH

+c3 p--C k

-0.3C2o exp(-R, 2)

(10)

or

g'

(3)

C~ = C~o- 0.3(72o exp( - R , 2)

(11)

Each modified k-E model is concentrated on the selection of a set of turbulence parameters, i.e., C,, Ca,
Cz and C3. These modified models are called the LRN
model, which are developed mainly for phasing out the
wall functions. By analysing the calculation of these
parameters, some observations can be drawn.

If the turbulent constants in Eqs. (9) and (10) or (11)


are selected to be K=0.41, or,= 1.3, C~,= 0.09, Cm= 1.85,
then, we can have the following expressions of Ca and
C2 modified with respect to R,

2.1. Turbulentcoefficients

Some researchers [5] have introduced stronger nonlinear modification functions as follows

For the case of the boundary layer, the e transportation


equation of the k-e model has the form [2] as follows

Ca = 1.42 - 0.56 exp( - n t 2)

(12)

Ca = 1.85 - 0.56 exp( - R t 2)

(13)

CI=Clo

(Acl'~ 3]
1 + \ f~, ] 3

(14)

C2 = C2o[1 + exp(-Rt) z]

(15)

where

+Ca-~

-C2 ~

(4)

In the region close to a wall, the shear stress is uniform


and the length scale increases linearly with distance
from the wall (fully turbulent log-law region). The
convective terms can be omitted [3], which follows that
Eq. (4) reduces to

By considering equilibrium turbulence, the term


can be expressed as

m(au/ay)2/p

/L'
P (~)z= e

(6)

and
= C~, k2

by using the Boussinesq assumption.

(7)

At

f ~,= [1- exp( -A~,Rk)]2(l + -l~,)

(16)

It can be easily found that the derivatives of C~ with


respect to e and k are
aCl

a-~"(~(f/'L)--4;

aCl

"~

~(f~)--4

(17)

Thus, even a small variation in R t will produce


dramatic changes in the damping function value, which
increases the numerical stiffness of the model. Moreover,
the damping function for C2 will be effective only in
the region of y+ <5, which means that the distance
between the first grid node and the wall surface will
be only of the order of a millimeter. The foregoing
consideration could explain the reason why some researchers [6,7] have obtained diverging results by using
this model.
Through the model test, Fujisawa [8] found that for
the LRN model, the damping function of turbulent
coefficients which contain the wall parameter y+ are

35

Y. Fan / Energy and Buildings 23 (1995) 33-39

not suitable for reproducing the laminar behaviour or


laminarization.
In addition, the production term of the e-equation
(see Ref. [1]) should include three parts, which are
shear, gradient and turbulent production. By using the
Kolmogorov micro scale theory, the shear and gradient
production parts were dropped under the consideration
of high turbulent Reynolds number R,. Their relations
are as follows

Ps = C~R,- ~z2p,
P~=C~2R,-S/4P,

(18)
(19)

Eqs. (18) and (19) can be interpreted in a way that


if R, is not very high, then, P~ and Pg should not be
neglected. In this case, the turbulent coefficient C~ can
be expressed as
C1 ~--C1 + f l i R t -1/2 + C l 2 R t -5/4

(20)

Constants Cm~ and C~2 are difficult to determine. Nevertheless, the coefficient C~ should be a function of Rt,
when the turbulent Reynolds number is not sufficiently
high.
Based on the foregoiing analysis, it can be concluded
that the turbulent 'constants' have to be expressed as
a function of R, in the, air flow simulation of room(s)
where R, will not be very high. The difficulties caused
by near wall regions may be solved by the wall function
method even in the LRN model if the computation
time and computer memory are primitive issues.

TWS = |~bi(x~)LC(qN) d r = 0

(24)

i
fl

From the foregoing introduction of the TWS method,


it can be found that the weighted residual method was
executed to a resultant PDE by introducing an artificial
diffusive term to the governing PDEs (see Eq. (22)).
It was found by Baker et al. [9] that the stability of
the solution can be improved by this diffusive term.
This term causes a filtering effect to the high frequency
fluctuation of air flow, because the low frequency fluctuation is dropped first (Eq. (22)). After that, the
resultant governing equation which contains high order
fluctuation is minimized with respect to the approximate
solution (Eq. (24)).
In the TWS method, two problems arise. One is to
determine the/3 values and the other one is to choose
the 'known' functions of ~i and ~ in Eqs. (22)-(24).
It is obvious from the definition of Eq. (22) that the
13value should be tightly connected with the grid-point
distribution. Each problem must have its own/3 value,
which can handle the physical phenomenon properly.
In the case of selected or known functions, q~i and ~ ,
considerations must be made to avoid the physically
unrealistic. Linear interpolation would result in a similar
problem as in the central-difference scheme. When the
Peclet number is large for convection-diffusion problems, unrealistic results may be obtained [10]. Patankar
[10] has proposed the following exponential function
for a two-dimensional case:

pUX

+CY

3. Calculation of the partial differential equations

= A + B exp - - ~

For solving the governing equations of air and contaminant motion in room(s), Baker et al. [9] recently
presented a new method, called the Taylor weak statement (TWS). The basis of the method is similar to the
method of weighted residuals. The typical partial differential equation (PDE) has been expressed by:

The proper selection of 'known-functions' for threedimensional cases still needs further model tests and
demonstrations.
The bench mark analysis made by Baker and coworkers [9,11,12] using the TWS method as limited to
the LRN (in the order of 9 X 108) and large air change
rate (30 ach) flow which is far from that found in
practice. The reason for the good agreement of their
results was because the air flow can be identical in the
whole domain except for the region near the solid
boundaries. But this is not the case in reality. Most
difficulties in modelling room air flow reside in the
handling of multi-state (laminar, turbulent and transition flow) flow and transformation from one flow state
to another.

L(q) = 0

(21)

Using Taylor's expansion technique, a PDE companion


was defined in the form of

LC(q)-L(q)- [3 At -~j { A/I~

(22)

Further, an approximate solution qN which contains a


number of undetermined parameters Qj(t) was assumed,
the solution is then expressed by Baker et al. [9] as:
N

qfv(y, t) = ~ ~.(xOQj(t)

(23)

j--1

Substituting the approximate solution into Eq. (21)


leaves a residual. Then let the sum of the residual
(TWS) over a sub-domain (control volume or element)
of interest be minimized as follows

(25)

4. Modelling of air terminal devices (ATD) in rooms

In fact, both CFD simulation and scale modelling


are procedures of similarity. It is not difficult for a
CFD model to guarantee thermal and momentum similarity within the fluid domain. This is not the case
when considering the momentum similarity on bound-

36

Y. Fan / Energy and Buildings 23 (1995) 33-39

aries, particularly the ATD. The momentum similarity


on boundaries can mainly embody at the air supply
and exhaust devices in a ventilated space. This is because
the influence of air leakage (e.g. through windows) on
air flow similarity is limited typically. And on the other
hand, the momentum similarity is automatically satisfied
for impermeable walls due to the no slip law. Modelling
of air terminal devices has been becoming increasingly
important. In the following, the approaches in modelling
ATD will be summarized and discussed based on a
literature review.

4.1. Momentum similarity at A T D

(b)
1.00

0.80

In this method, a complicated diffuser is modelled


by a simple opening which lets the supply air have the
same momentum and same direction as reality.
Two versions to the basic approach exist. In the first,
a simple opening has the same effective area as the
real diffuser, then the same velocity as each nozzle (if
the diffuser has many nozzles). This version has been
used by Heikkinen [13], and Skovgaard and Nielsen
[14]. From Nielsen's results [15], it can be found that
a simple opening can fairly simulate a complicated
diffuser, especially the jet decay far from the ATD.
The measurements made by Heikkinen [13] also confirmed Nielsen's conclusion. However, the jet spread
in horizontal and thickness in vertical obtained by this
simplified ATD model can be very different in comparison to the measured air jet generated by a complicated diffuser (see Ref. [13]).
The second version lets the momentum force defined
by Eq. (26) and the supply air direction be the same
between the simple opening and the complicated diffuser. The opening area can, somewhat freely, be chosen.

F = J p U z dA

(a)

0.40
0.20
0. I(

0.0(
1

0.0,

* Full s c a l e . Diffuser type D


Model. Nozzle

0.02
0.00'1

~ ,; ; ;/o 20 do~o6'o8o

(c)
Fig. 1. Modelling of ATD [15]. (a) End wall mounted diffuser; (b)
nozzle directed against the ceiling in 45; (c) velocity decay in wall
jet along the ceiling in a room and in a model.
X0
r

.e-

(26)
-~y'0

This version has been used earlier by Chen et al. [16].

4.2. Momentum similarity in front of the A T D


This approach is based on well-developed wall-jet
techniques. The air velocity field within a small space
in front of a diffuser is self-similar. Thus, the momentum
similarity of the ATD can be defined away from the
ATD by using the well-developed wall-jet technique.
The approaches developed in the above way were
referred by Nielsen [17] as the Box method and the
prescribed velocity method. These are sketched in Figs.
1 and 2. At surface b there is no flux. In addition,
universal or dimensionless physical parameters' (~b)
profiles for a given diffuser at surface a must be predefined. This method is more difficult than the method
described above. However, it was believed that better
results could be achieved by using it [17].

Fig. 2. Modelling of ATD [17].

All the above methods have been used in CFD


simulation. The first one seems to be more user-friendly,
particularly in assessing a system design. Whereas, for
the second one, measurements are normally required.
5. Wall function

The wall function was originally a method to deal


with the boundary conditions for the HRN model. The
HRN model was based on the assumption that the g t
value is so high that the turbulent parameters C1, C2,
C~, and C3 could be treated as constant. This assumption
does not hold in the region near the solid boundaries
at any time. Hence, a 'known' function (so-called wall
function), is needed to connect the boundary conditions
with the physical parameters in the domain where the

37

Y. Fan / Energy and BuiMings 23 (1995) 33-39

The air flow over the walls is seldom in parallel to


the wall surface. Therefore, each wall function, in fact,
is based on the assumption that the flow is similar.
Hence, any model plus the wall function has more or
less the same difficulties as the mixing-length model.
The LRN model has phased out the usage of the
wall function. Though it still has some problems that
have not yet been solved, it offers an opportunity to
handle the room air flow.

influence of Rt on turbulent parameters is negligible.


The velocity and temperature distributions (VTD) of
the fiat plate boundary layer flow or the Couette flow
have been selected as the wall function. In a model
test, Chen [1] found that if the VTD within the boundary
layer were expressed by linear and logarithmic equations,
the computed convective heat transfer coefficient near
a wall of a ventilated room could be about 40-50%
lower than the measured one. Later he improved the
wall function by dividing the boundary layer into three
regions, the viscous sublayer, buffer zone and fully
turbulent log-law zone then evaluated the VTD by
using different formulas for each region. It was concluded [1] that modified wall function could reduce
the differences between computed and measured convective heat transfer coefficient to about 20%, but it
is still high. The velocity distribution in the whole
boundary layer over a flat plate had been deduced by
Spalding [18] more than 30 years ago, which is expressed
by Eq. (27). Unfortunately, Spalding's equation seems
to have not been noticed by researchers of CFD modelling. A comparison of different wall functions is
shown in Fig. 3.

6. C o n c l u d i n g r e m a r k s

The k-e model in simulating the room air flow was


reviewed. It was concluded that the turbulent parameters
should be functions of the turbulent Reynolds number
even in the case where the wall function was to be
used. It is evident that the air flow in modem buildings
is very slow. Thus, the basic assumption of fully turbulent
flow in the region far from the walls for the H R N
model will no longer hold. The recommended correlation, Eq. (20), is based on the exact time-averaged
N-S equations and Kolmogorov micro scale theory. If
the constants in this equation could be defined by CFD
analysis or by experiments, better results could be
expected. It was found that the damping function of
turbulent coefficients which contain the wall parameter
y+ fails to handle the room air flow. The damping
function of turbulent coefficients should affect not only

y+ = u + +0.1108
X [ exp '4" + - 1 - 0 . 4 u
u + - u(p/rw)lr2;

(0.4u+) 2
+

(0.4~+)3]

2!

(27)

y + -y(,r,,, p//z) ' a


2,0
2:8
26

./

.~

24
22
20

u + = 0.25 In(y ) + 5.5 j ~

',8
U

16
14

~2

//

10
8

y+=

U +-

1;y~

2
0

In (y*)
Fig. 3. Comparison of different wall functions.

10

38

Y. Fan / Energy and Buildings 23 (1995) 33-39

the near wall region air flow but also the air flow in
the central zones of the room.
To predict the momentum and heat energy distribution over a finite cell will form the basis of a calculation
method of the partial differential equations. These kinds
of distributions are typically considered to be linear.
Patankar has proposed an exponential distribution in
his two-dimensional convection--diffusion simulation
[10]. If this function works for three-dimensional cases,
further model tests and demonstration will be needed.
Each method mentioned in Section 4 shows the benefit
from evaluating the jet decay, but the prediction on
jet spread and thickness may be far from reality. This
could be indicated by unbelievably uneven computed
temperature and contaminant distributions. Further
research is required.
The k-e model (LRN and HRN models) together
with wall function has more or less the same problem
as the mixing length model. This point can be proved
by using the von Karman's similarity hypothesis [19].
In order to improve the ability of the k-6 model in
handling the circulated flow, the wall function must be
phased out. It is true that the use of the LRN model
without wall function will increase computational cost.
Each turbulent model, which includes the k-e model,
Reynolds stress model and vortex model may be analogized by the Taylor's expansion. The differences
between them are in how many terms have been kept
in the expansion. The k-e model is the simplest one.
If the turbulent coefficients can be correlated correctly
to the turbulent Reynolds number, then use of the
model in engineering will be more promising.

7. Nomenclature

A
Aj
Acl. A.,

constant (-)
kinematic flux vector jacobian (J/m2)
A t constants (-)
constant (-)
B
constant (-)
C
C1, C2, C3 turbulence parameters (-)
turbulence parameters (-)
C~o
constant (-)
turbulence parameter (-)
C.
momentum force (N)
F
gravitative acceleration vector in direction
gi
i (m/s2)
time-averaged enthalpy (J/kg)
H
kinetic energy (J/kg)
k
pressure (Pa)
P
shear production (J/kg s2)
es
turbulent production (J/kg s2)
Pt
gradient production (J/kg s2)
P,
q(xj, t)
a vector (={u 1, 19, ~b, p}T, j = 1, 2, 3) (--)
turbulent Reynolds number (=k2p/lze) (-)
R,

gk

U
u

X
Y
Y

turbulent Reynolds number ( = kl/2yp/lz) (-)


time-averaged velocity (m/s)
velocity near wall (m/s)
coordinate (-)
coordinate normal to X (-)
distance from the wall (m)

Greek
p6
IZt
P
Ok, O"H

,(x)

coefficient of thermal expansion (l/K)


dynamic viscosity (N s/m2)
turbulent viscosity (=C~pkZ/E) (N s/m2)
density (kg/m3)
equivalent turbulent Prandtl number (-)
shear stress at the wall surface (Pa)
resultant variable which connects velocity
field and pressure through continuity equation (-)
temperature (K)
known function (-)
known function (-)

Acknowledgement
The author thanks Professor Dr Markku Virtanen
of our laboratory for his intelligent comments and
editorial help with this paper.

References
[1] Q. Chen, Indoor airflow, air quality and energy consumption
of buildings, Ph.D. Thesis, Delft Technical University, Netherlands, 1988.
[2] V.S. Arpaci and P.S. Larsen, Convective Heat Transfer, PrenticeHall, Englewood Cliffs, NJ, 1984.
[3] B.E. Launder and D.B. Spalding, The numerical computation
of turbulent flow, Comput. Methods Mech. Eng., 3 (1974) 269-289.
[4] W.P. Jones and B.E. Launder, The calculation of low-Reynoldsnumber phenomena with a two-equation model of turbulence,
Int. Z Heat Mass Transfer, 16 (1973) 1119-1130.
[5] C.K.G. Lam and K. Bremhorst, A modified form of the k-E
model for predicting wall turbulence, J. Fluids Eng., 103 (1981)
456--460.
[6] P.L. Betts and A.A. Dafa'Alla, Turbulent buoyant air flow in
a tall rectangular cavity. Significant questions in buoyancy
affected enclosure or cavity flows, in J.A.C. Humphrey, C.T.
Avedisian, B.W. Le Tourneau and M.M. Chen (eds.), ASME,
New York, 1986, pp. 83--91.
[7] H. Nguyen, A Petrov-Galerkin least-squares finite element
algorithm for predicting of room air motion, Proc. Indoor Air
"93, HelsinM, Finland, 1993, Vol. 5, pp. 325-330.
[8] N. Fujisawa, Calculations of transitional boundary-layers with
a refined low-Reynolds number version of a k-e model of
turbulence, in W. Rodi and E.N. Gani6 (eds.), Engineering
Turbulence Modelling and Experiments, Elsevier Science, 1990,
pp. 23-32.
[9] A.J. Baker, P.T. Williams and R.M. Kelso, Development and
validation of a robust CFD procedure for predicting indoor

Y. Fan / Energy and BuiMings 23 (1995) 33-39

[10]
[11]

[12]

[13]

room air motion, Proc. Indoor Air '93, HelMnki, Finland, 1993,
Vol. 5, pp. 183-188.
S.V. Patankar, Numerical Heat Transfer and Fluid Flow, Hemisphere, Washington, I)C, 1980.
R.M. Kelso, S. Roy and A.J. Baker, A CFD prediction of
thermal comfort distribution in a 3-D space with high ventilation
rates, Proc. Indoor Air "93, Helsinki, Finland, 1993, Vol. 5, pp.
313-318.
S. Roy, A.J. Baker and R.M. Kelso, Airborne contaminant
CFD modeling studies for two practical 3-D room air flow
fields, Proc. Indoor Air '93, Helsinki, Finland, 1993, Vol. 5, pp.
349-354.
J. Heikkinen, Modelling of a supply air terminal for room air
flow simulation, 12th AIVC Conf., Ottawa, Canada, 1991, Vol.
3, pp. 213-230.

39

[14] M. Skovgaard and P.V. Nielsen, Modelling complex inlet geometries in CFD -- Applied to air flow in ventilated rooms,
12th A1VC Conf., Ottawa, Canada, 1991, Vol. 3, pp. 183-200.
[15] P.V. Nielsen, Model experiments for the determination of airflow
in large spaces, Proc. Indoor Air '93, Helsinki, Finland, 1993,
Vol. 5, pp. 253-258.
[16] Q. Chen, A. Moser and P. Suter, A numerical study of indoor
air quality and thermal comfort under six kinds of air diffusion,
ASHRAE Trans., 98 (1) (1992) 203-217.
[17] P.V. Nielsen, Description of supply openings in numerical
models for room air distribution, ASHRAE Trans., 98 (1) (1992)
963-971.
[18] D.B. Spalding, A single formula for the 'law of the wall', J.
AppL Mech., (1961) 455-458.
[19] L.C. Burmeister, Convective Heat Transfer, Wiley, New York,
1983, Ch. 9.

You might also like