Vol. 40, No. 3, Pp.

613~25, 1997
Copyright 1996 Elsevier Science Ltd
Printed in Great Britain. All rights reserved
0017-9310/97 $15.00+0.00

Int. J. Heat Mass Transfer.

~ ) Pergamon

P I I : S0017-9310(96)00117-2

A temperature wall function formulation for

variable-density turbulent flows with application
to engine convective heat transfer modeling
Z H I Y U H A N and R O L F D. REITZ~"
Engine Research Center, University of Wisconsin-Madison, Madison, WI 53705, U.S.A.
(Received 28 July 1995 and in final form 27 March 1996)

Abstract--A temperature wall function was derived for variable-density turbulent flows that are commonly
found in internal combustion engines. Thermodynamic variations of gas density and the increase of
the turbulent Prandtl number in the boundary layer are included in the formulation. Multidimensional
computations were made of a pancake-chamber gasoline engine and a heavy-duty diesel engine under firing
conditions. Satisfactory agreement between the predicted and measured heat fluxes was obtained. It was
found that gas compressibility affected engine heat transfer prediction significantly while the effects of
unsteadiness and heat release due to combustion were insignificant for the cases considered. Copyright
1996 Elsevier Science Ltd.

INTRODUCTION

An understanding of mechanisms of engine heat transfer is important because it influences engine efficiency,
exhaust emissions and component thermal stresses.
Accurate predictic,n of wall heat transfer is not only
needed for better understanding of heat loss mechanisms, but also necessary for improving the overall
accuracy of engine combustion simulations. Heat flux
through combustion chamber walls is mainly due to
gas-phase convection, fuel film conduction, and hightemperature gas and soot radiation. In many cases,
e.g. in premixed-charge engines and on surfaces of
diesel engines without spray impingement, gas-phase
convective heat transfer is the major concern. Soot
radiation is believed to become significant in large
bore engines [1].
Engine heat transfer phenomena have been studied
extensively for many decades. Numerous mathematical models have been proposed. The traditional
models (correlations) which are based on dimensional
analysis are useful from the viewpoint of global analysis [1]. However, (:hey cannot provide spatial resolution. AdditionaUy, these models lack a sound theoretical basis and the,ir predictions are often inaccurate
when applied beyond the conditions under which their
empirical constants are determined.
Approaches which solve the one-dimensional
energy equation have been reviewed recently by Yang
[2]. In order to solve the one-dimensional energy equation of a turbulent boundary layer, the equation is
linearized and no~xnalized so that an approximate
solution can be obtained for transient, compressible

t Author to whom correspondence should be addressed.

and low Mach number turbulent boundary layer

flows. Good heat flux predictions were achieved in a
motored engine by Yang and Martin [3].
Since the boundary layer of an engine in-cylinder
flow is thin relative to practical computational grid
size, velocity and temperature wall functions (or temperature profiles) are often used in multidimensional
computations to solve the near-wall shear stress and
heat transfer. The traditional temperature wall functions that are currently used are derived with the
assumptions of a steady and incompressible flow, no
source terms (terms that account for pressure work,
chemical heat release and sprays), and the validity of
the Reynolds analogy (e.g. the work of Amsden et aL
[4]). These assumptions are questionable when
applied to engine flows [2, 5, 6] because, in an engine,
the gas density varies significantly due to piston
motion and combustion; unsteadiness and chemical
heat release may invalidate the Reynolds analogy.
Another assumption used in the traditional wall
function formulations is that a constant turbulent
Prandtl number is used across the entire boundary
layer. Recent experimental results have revealed that
the turbulent Prandtl number increases in the buffer
and viscous sublayers [7]. In these regions, the
increased Prandtl number will affect heat transfer predictions.
Engine heat transfer predictions have not been satisfactory when the traditional temperature wall functions are needed. Great underprediction of wall heat
fluxes was found in previous studies [5]. Recently,
more sophisticated methods including two-equation
models [8] for heat transfer prediction have been
under investigation. However, wall-function-type
models are still of much practical interest due to their
relative simplicity and efficiency for engine corn613

614

Z. HAN and R. D. REITZ

NOMENCLATURE

Greek symbols
c~
reciprocal turbulent Prandtl number
~0
reciprocal molecular Prandtl number
e
dissipation rate of turbulent kinetic
energy
q~
dimensionless temperature
x
von K~irmfin constant
2
ratio of specific heat
#
dynamic viscosity
v
kinematic viscosity
v + = vt/v ratio of turbulent viscosity to
laminar viscosity
0
dimensionless time
p
density
z
shear stress tensor ; time scale ; the
transformed time.

Cp
f

specific heat
delay coefficient in the combustion
model
G
source term in energy equation
G + = Gv/qwU* dimensionless source
term
k
turbulent kinetic energy ; thermal
conductivity
I
turbulence integral length
scale
p
pressure
Pe
Peclet number
Pr
Prandtl number
q
heat flux
Qc
volumetrical heat release
r
combustion product fraction
SL
laminar flame speed
t
time
T
temperature
u*
friction velocity
U
magnitude of the gas velocity
y
distance to the wall
y+ = u*y/v dimensionless distance
Y
mass fraction.

Subscripts
1
laminar
m
species index
ss
steady state
t
turbulent
us
unsteady
w
wall quantity.

putations. Because of gas density variation in engine

flows, an attempt is made to formulate a model that
includes gas compressibility in this study. A model
which assumes quasi-steady conditions is derived first
and then is applied to engine predictions. The effects
of unsteadiness and heat release are then discussed.
TEMPERATURE WALL FUNCTION
FORMULATIONS

In the near wall region of wall-confined in-cylinder

engine flows, it is assumed that :
(1) gradients normal to the wall are much greater
than those parallel to the wall ;
(2) the fluid velocity is directed parallel to a flat
wall ;
(3) pressure gradients are neglected, i.e. p = p(t) ;
(4) viscous dissipation, and the Dufour and
enthalpy diffusion effects on energy flux are
neglected ;
(5) radiation heat transfer is neglected ;
(6) the gas is ideal.
Hence, the general energy conservation equation
can be written as
Oq
or
dp
cOy = - p C p - ~ + - ~ - Q ,

where

(1)

OT
q = -- (k + kt) ~ y .

(2)

The above assumptions are commonly included in

many heat transfer modeling studies of engine incylinder flows, and consequently, equation (1)
becomes the starting point of analysis under these
assumptions. The first term of the right-hand-side of
equation (1) is the transient term and accounts for the
energy change with time in the control volume. The
second one is the pressure work term and the third one
is the heat generation term due to chemical reactions.
It is known that engine wall heat transfer is
unsteady in nature as described by equation (1). However, we approximate the process by invoking the
quasi-steady assumption. The effect of unsteadiness
will be discussed later.
Integration of equation (1) from the wall (y = 0)
and using the relation between conductivity and viscosity gives
# . #t'~dT
-cp ~r+ ff~rt)-~y=q,+Gy,

(3)

where q, is the heat flux through the wall, and G = Q

(the average chemical heat release). Let
v + = --;
Yt
v

y+ = -u*y
-;
v

G + = Gv
qwU*"

After rearrangement, equation (3) becomes

(4)

Temperature wall function formulation for turbulent flows

pcpu* d T -

q*

(1

v-~

dY+

G +y

+ ~ dy + .
v )

+(1

(5)
Equation (5) is integrated from 0 to y+. The left-handside of equation (:5) then becomes
T+ = I r p CpU* d T = p%u*Tln(T[T,,,)
Jr.

qw

(6)

q~

However, correlations describing the changes of Prt

and vt are needed in order to integrate the right-handside of equation (5). Reynolds proposed [9] that
v + = 1-~-xy+[1-exp(-y+~/A2)]

or (11), a relation which describes the variation

of the turbulent Prandtl number with y+ can be
obtained.
Figure 1 illustrates the variations of v+/Prt with y+
using different combinations of the mentioned correlations. Very similar trends can be seen between
these combinations. However, analytical integration
of equation (5) is difficult if the above correlations
are used directly. A curve-fit technique was used to
construct a simplified functional form to make analytical integration of equation (5) possible. The simplified expression for v +/Prt is given as
V+

Prt

(7)

(xY+)'

(8)

(xy+)3 + 328.5"
Experiments have revealed that the turbulent
Prandtl number increases in low-Reynolds-number
regions (near the wall) and from its equilibrium value
(far from the wall). Based on experimental data, Kays
formulated the coxTelation as [7]
0.7

Prt = ~

+0.85

(9)

where the turbule~t Peclet number is defined as

(10)

Pet = v+ Pr.

On the other hand, the R N G theory of Yakhot

and Orszag [11] also provides a relation between the
turbulent Prandtl number and Reynolds number as
1
i c- 1.3929 10"6321 (~+ 1.3929 0.3679
--~
v = , ~o -- 1.39~9 [
~ 1.~
.

(11)

By combining equation (7) or (8) and equation (9)

80

70

Pr--~t= my+

50

30

y+ > y+"

.....-.':
...~:':~"

...e.."~""

20

40

(12)

The constants are set to be a = 0.1, b = 0.025,

c = 0 . 0 1 2 and m = 0 . 4 7 6 7 when x = 0 . 4 1 and
Pr = 0.7 are used. These constants were determined
such that the experiments-based equation (9) was favored. However, it was found that different sets of
constants with curve-fits covering the range of the
curves given in Fig. 1 only caused a very little change
in the heat-flux predictions. The transition value of
y~- was chosen as 40.0 in this study. It is worthwhile
to point out that the use of this value ofyJ- is mainly
due to mathematical consideration and is less physically oriented. In transitional wall-function analysis,
a critical y~ is used to describe the transition from the
viscous sublayer to turbulent layer, which is determined from benchmark experiments. For the engine
flows considered here, there is no such experimental
data available that can be referred to. The newly formulated expression with the given constants is also
shown in Fig. 1. for comparison. It matches the curves
with the use of equation (9) well.
Substituting equation (12) into equation (5) and

. . . . . . . Reynolds + Yakhot & Orszag

. . . . . Mellor + Yakhot & Orszag
- - - Reynolds + Kays
. . . . . Mellor + Kays
,
Equation (12)

60

_ a + b y + + c y +~ y+ <~y~

y+

where x = 0.41, A = 26; and Mellor suggested [10]

that
v+ =

615

y*

60

80

100

Fig. 1. Variation of the ratio of dimensionless viscosity to turbulent Prandtl number.

616

Z. HAN and R. D. REITZ

splitting the integration into two parts gives

T + = fig

1
dy + + fY+ l dy +
Pr-I +a+by+ +cy +2
Jy+ my +

+,+

pr - l + a + b y + + c y 2

dy + +

f/o+

--dy +

~ m

dYm
(13)

where Pr-J is neglected in the second part of the

integration, thus assuming the turbulent effect is
dominant in this region of boundary layer. Finally,
the temperature profile equation (wall function) is
given as
r + = 2.1 l n ( y ) + 2 . 1 G y +33.4G +2.5

here. The model details can be found in the given

references.
With the characteristic-time combustion model, the
time rate of change of the partial density of species
m, due to conversion from one chemical species to
another, is given by
dt

2.1 In(y+) + 2.5

(15)

If the source term G can be neglected, equation (15)

then becomes
qw =

pCpu*Tln(T/T~)
2.1 ln(y) +2.5

(16)

TURBULENCE AND COMBUSTION MODELS

Turbulence in the high-Reynolds-number regions

was modeled using a modified R N G Ice model [12].
It was shown that the original R N G model [11] gave
better predictions for separated flows [13] and the
modified version which includes gas compressibility
[12] improved engine combustion simulation [14].
The velocity wall function of Launder and Spalding
[15] was used to calculate the wall momentum flux
~w = p U O ,

(17)

where Tw is the wall shear stress, U is the magnitude

of the gas velocity in the wall cell and
U*

C=-y+

y+ <~10.18

0 = - ln(Ey +)

y+ > 10.18.

(18)

In equation (18), E is equal to 9.8, and u* was calculated from turbulent kinetic energy k as
u* = k x / / ~1/2.

(19)

To simulate engine combustion process, the characteristic-time combustion model of Abraham et al. [16]
and the one of Kong et al. [17] were used for gasoline
and diesel engine, respectively. The basic idea of these
two models is the same. For the sake of brevity, only
the essence of the combustion models is discussed

(20)

where Ym is the mass fraction of species m, Y* is the

local and instantaneous thermodynamic equilibrium
value of the mass fraction, and % is the characteristic
time to achieve such equilibrium. The characteristic
time is assumed to be the sum of a laminar (kinetic)
timescale Zl and a turbulent timescale % that is
Zc =

pcpu*Tln(T/Tw) - (2. ly + + 33.4)Gv/u*

(14)

and the corresponding formulation for wall heat flux

is given as
qw --

Ym-Y*

Z l +f'c t

(21)

where f is a delay coefficient which gradually introduces the controlling role of turbulent effects in a
developing flame kernel. The laminar timescale is
modeled based on Arrehenius kinetics and the turbulent timescale is assumed to be proportional to the
eddy turnover time, k/e, which is calculated from the
turbulence model.
Different formulations of % zt a n d f w e r e used for
modeling of gasoline engine combustion and for diesel
engine combustion in this study due to lack of a generalized model currently. One of the differences among
the formulations is in the delay coefficient f In gasoline engines, the time of ignition is known and this
is followed by a well-defined flame growth process.
Hence, in the model of Abraham et aL [16], when the
flame kernel grows to be comparable to the turbulence
eddy size, it is assumed to become influenced by the
turbulence, f i s given as
f = 1 -- e - ('- '~)/~d

(22)

where % = Cmtl/SL, and ( t - ts) is the time after spark,

SL is the laminar flame speed, l is turbulence integral
length scale and Cml is a model constant with a typical
value of 7.4.
However, the combustion process is more complicated in diesel engines than that in gasoline engines.
The wide range of equivalence ratio and non-homogeneous spray droplet combustion makes the monitoring of flame kernel growth difficult. In the model
of Kong et al. [17], the idea that laminar chemistry
initiates combustion, and then turbulence influences
combustion gradually, is also used. To account for
the separate effects of laminar chemistry and turbulence, the appearance of products is used as an indicator of mixing following the initiation of combustion
events a n d f i s formulated as
1 -e

f=

-r

0.632'

(23)

where r is the local ratio of the amount of products

to that of total reactive species at each point in the
combustion chamber, i.e.

Temperature wall function formulation for turbulent flows

r-

Yco~+ YH~o+ Yco + YH2

1-YN2

(24)

and the parameter r indicates the completeness of

combustion in a specific region.
Both the models of Abraham etal. [16] and Kong
etal. [17] have been shown to work well [17, 18]. In
the gasoline engine calculations, the spark ignition
process was modeled by adding energy to the charge
in the computational spark cell at a specified rate [4].
Other submodels used in the diesel engine simulations
include the wave spray atomization model [19], the
Shell autoignition model [20], etc., which are not discussed here.
COMPUTATION OF A PREMIXED-CHARGE
ENGINE

To test the proposed heat transfer model, predictions were carried out in a premixed-charge sparkignition engine and the results were compared with
experimental data. This engine has a pancake-shaped
combustion chamber geometry and the spark is
located at the center of the cylinder head. The engine
specifications and operating conditions are listed in
Table 1 [21].
The heat transfi~r measurements were described by
Alkidas [22] and data is available for the case listed in
Table 1 and a motored-engine case at 1500 rpm. The
wall heat transfer measurements were made at four
radial locations situated on the engine head at 18.7,
27.5, 37.3 and 46.3 mm from the cylinder axis, and at
one axial location on the cylinder liner, 6.3 mm from
the head. These head-flux probe locations are referred
to as HT-1 to HT-5, respectively, in the study (see
Fig. 3). Measured heat transfer data is available for
the motored-engine case at the radial locations HT-1
to HT-4. The averaged wall temperature is 420K for
the fired-engine case and 380K for the motored-engine
case [22].
The computations were made using the KIVA-II
code [4]. Two-dimensional computations were carried
out for computational efficiency due to the symmetry
of the chamber geometry. In the baseline case, the
typical mesh size i~ about 2.5 mm in radial direction

and 1.26 mm in axial direction. The effect of grid size

on the heat transfer predictions was studied by using
finer grids as well, as will be addressed later. The
numerical time step used was 5 ~ts which was found
sufficient to give time-step-independent results.
It should be pointed out that different treatments
of heat transfer influenced the combustion predictions
somewhat which, in turn, affected the heat transfer
predictions. Since heat transfer was the major concern
in this study, the approach adopted was to adjust a
constant of the combustion model (Cml see equation
(22)) such that the computed cylinder pressure agreed
with the measured one in each case as necessary. In
this way the computed heat fluxes using different
models could be compared under similar thermal conditions.
The computed cylinder pressure of the fired-engine
case is shown in Fig. 2. It agrees satisfactorily with
the measured data. Additional details about the flame
structure and the flame development are summarized
in Fig. 3 which shows the predicted temperature contours in the combustion chamber at 10, 20 and 30
after the spark (spark occurs at 27 BTDC). The flame
kernel forms in the vicinity of the spark cell and then
propagates radially into the combustion chamber
space.
The predicted wall heat fluxes for this case by using
the present model are shown in Fig. 4. As can be seen,
satisfactory prediction is obtained in terms of both
the phase and magnitude of the heat flux. It is also
seen that the heat flux peak values occur at later crank
angle from HT-1 to HT-5 as the flame propagates
across the chamber. Another feature seen in Fig. 4 is
that both the magnitude and variation of the heat
flux are similar at each location. This indicates that
homogeneity in wall heat transfer exists in the considered SI engine.
Figure 5 shows the computed wall heat fluxes for the
motored engine case. Again, good level of agreement
between computation and measurement is obtained.
It was of interest to examine the effect of grid size.
This was not only due to the consideration of numerical accuracy, but also due to the fact that as grid size
decreases, the resolved boundary layer flow will be
within different wall flow regimes. Three different grid

Table 1. Specificationsand operating conditions of the engines

Bore x stroke (mm x mm)

Cylinder displacement (1)
Compression ratio
Connecting rod length (ram)
Intake-valve closing (deg. BTDC)
Equivalence ratio
Engine speed (rpm)
Injection timing (deg. BTDC)
Spark timing (deg. BTDC)
Swirl r~ttio (nominal)
Fuel

617

Gasoline engine

Diesel engine

105.0 x 95.25
0.82
8.56
158.0
117.0
0.87
1500

137.2 x 165.1
2.44
15.0
261.6
147.0
0.46
1600
11

27
0
C3H8

1
Amoco Premier no. 2

618

Z. HAN and R. D. REITZ

2500

2000

"

"

measured

--

computed

10

20

1500

13_

1000

500
-30

-20

-10

Crank angle

30

(degree)

Fig. 2. Comparison between computed and measured cylinder pressure--the fired-engine case.

Spark

111'-1 HT-2 HT-3 liT-4

. . 1 ..

~h~,~.

hhhhl,

h h i , l, h h h h h

rrr.s

.~?F-]

Fig. 3. Predicted temperature contours showing the flame structure. Top: crank = - 1 7 ATDC; H =
2250K; L = 902K, middle: crank = - 7 ATDC; H = 2290K; L = 951K, bottom: crank = 3 ATDC;
H = 2370K ; L = 1030K. The heat flux probe locations of the experiment [22] are indicated schematically
by HT-1 to HT-5.

sizes, 1.26, 0.63 and 0.315 m m (at TDC) in the axial

direction were used, and the grid size in the radial
direction was kept the same (1.25 mm).
Figure 6 shows the values o f y at the first grid cell
when different grid sizes were used in the fired-engine
case at location HT-1. The small steps on the curve
of the coarsest mesh were due to the grid-chopping
technique used in the K I V A code as the mesh was
compressed [4]. In all cases, there exists a very sharp
drop of y that indicates the arrival of the flame at
the monitoring location HT-1. y decreases because
the gas viscosity increases as the gas temperature
increases. Figure 6 clearly shows that as the grid size
decreases to 0.315 mm, the values of y are smaller
than 10 and the viscous shear stress is calculated
according to equation (18) in this case.
As can be seen in Fig. 7, in the fired-engine case,
the present heat transfer model gives results that are
insensitive to the grid sizes considered. Importantly,

the computed heat flux was not altered when the finest
grid resolution was used in which the shear stress was
calculated from the viscous sublayer correlation. In
the present model, the increased turbulent Prandtl
n u m b e r that occurs w h e n y is less than 10 is included,
and it is beneficial to the heat transfer prediction in
the viscous sublayer. The forgoing results indicate that
the present model can be used in both turbulent and
laminar regimes. For the motored-engine case, also
shown in Fig. 7, the computed heat flux is seen to be
satisfactorily insensitive to grid resolution as well.
In traditional engine wall treatments, gas compressibility is not accounted for. Typically the righthand-side of equation (5) can be integrated by
assuming the gas density does not vary with the distance from the wall, hence
T+ = pcpu*(T-- Tw)
q.

(25)

Temperature wall function formulation for turbulent flows

2.5

2,S

measured

. . . . . computed

t<F" 2

x 1.5

~--

619

HT-2

-'"\

1
"I" 0.5

"1- 0.5

~..~

0-30

-20

-10

10

20

30

0
-30

C r a n k a n g l e (degree)
2.5

- .o
I

-20

-10

10

20

angle

Crank

2.5

HT-3

30

(degree)
,

HT-4

1.5

"1" 0.5

~'0.5
0

-30

-20

-10

.---.
c.e

angle

Crank
2.5

10

20

30

(degree)

-10

10

20

,.~...~oJ~~

-30-20-10
Crank

0
angle

10

20

30

(degree)

HT-5

:1" 0.5

-30

~"-mr

-20

-'-"-'r

30

C r a n k angle (degree)

Fig. 4. Comparison between predicted and measured wall heat flux--the fired-engine case.

This dimensionless temperature is the one usually

defined for incompressible flow [23]. Substituting
equation (25) into equation (14) can give the model
that excludes gas compressibility. This model was then
used to test the effect of gas compressibility. In
addition, a similar model for incompressible flows
proposed by Launder and Spalding [15] was also used
for comparison. It is expressed as
T

ertu*
-

(~.lpf)

n/4--('4"~SlPr l'~(Prt'~25
+Prtsin(n/4)~,x} \Pr,- J\--~rJ

(26)

where T is defined by equation (25) and the Van

Driest constant A is 26.

The effect of compressibility was found to influence

wall heat transfer predictions significantly as indicated
in Fig. 8 in which the model using equation (25) is
referred to as the incompressible model. The results
clearly show that the models excluding gas compressibility significantly underpredict heat transfer
when the engine is fired, while the present model that
includes gas compressibility reproduces the measurement well. (As noted earlier, since heat transfer affects
combustion predictions, the constant Cm~in the combustion model was adjusted slightly when the incompressible models were used such that the computed cylinder pressures also match the measured
data.)
In the motored-engine case, the incompressible
models also underpredict the measured heat flux

620

Z. HAN and R. D. REITZ

0.5

.....

0.~

0.5

measured

computed

HT-2

-0.4

HT-1

v
X
0.7=

0.3
0.2

"1-

0.1

-40

0.1
i

-30

-20

- 0

10

-40 -30 -20 - 0

20

10

20

Crank angle (degree)

Crank angle (degree)

0.5

0.5

~ 0.4
E

HT-3

HT-4

0"41
~0.3

~0.3
~- 0.2

0.2

/"

"i-

"1" 0.1

-40

-30

-20

0,1 ,soo.

- 0

10

20

-40

-30

-20

- 0

10

20

Crank angle (degree)

Crank angle (degree)

Fig. 5, Comparison between predicted and measured wall heat flux--the motored-engine case.

120

100

/'-,J,J"
80 _V_._..

~
I
/

dy=1.26mm
dy=0.63 mm
. . . . . dy=0.315 mm

60
40
20

i ......
,--- _-7 -'---~- . . . . . . . . . . . . . . .

0
-30

-20

-10

10

20

Crank angle (degree)

30

Fig. 6. y+ at the first grid cell with different grid sizes.

greatly as seen in Fig. 8. However, by comparing the

results of the fired-engine and the motored-engine case
it is seen that the effect of gas compressibility is relatively less important in a motored engine than in a
fired engine. This is simply because the gas density
undergoes smaller variation under non-reacting
conditions.

In the derivation of the present heat transfer model,

the quasi-steady assumption is invoked. As a result,
the transient temperature term and the pressure work
in the energy equation are absent in the model. In the
work of Reitz [5], the transient part of the wall heat
flux was calculated by using the expression of Yang
and Martin [3]

Temperature wall function formulation for turbulent flows

2.5

=
-

0.5

measured
- . d y = 1 . 2 6 mm

......... d~/----0.63mm
~ .... dy=0.315 mm

~-.

..."f'~.
~<
:_3

0.4

621

measured
- d y = 1 . 2 6 mm

.......... dy--0,63 mm
. . . . . dy=0,315 mm

0.3
x

~- 0.2

<D
1; O.5

"1- 0.1

(9

-30

-20-10

10

20

(~

0-40-30-20-10

30

10

Crank angle (degree)

Crank angle (degree)

20

Fig. 7. Comparison of predicted wall heat flux showing the effects of grid size. Left : fired-engine case.
Right: motored-engine case. Location HT-I.

2,5

0.5

a measured
. . . . . compressible model
- - - incompressible model
~'-- 2 -......... L-S m~del

I-"

.//

~<

measured

. . . . . compressible model
- - - incompr.e.ssible model
O.4 _ .........
L-S mooel

.-.

~1.5

~,~ 0.3

N~

.."
0.2

$
:r- 0 5

"1"0.1

. ~ .'F''~''" . . . . ":'~'"

-'3(~-20-10

10

20

Crank angle (degree)

30

-30

-20

"

-10

10

Crank angle (degree)

20

Fig. 8. Comparison of predicted wall heat flux showing the effect of compressibility. Left : fired-engine
case. Right: motored-engine case. Location HT-1.

q.s = - k T o ~ o ) "~-]/'~tCpoU*
l~.

the energy equation would not cause large errors

in wall heat flux predictions for this engine
considered.

COMPUTATION

where subscript 0 represents initial conditions (at

intake valve closure). Equation (27) was also used in
the present study to examine the effect of unsteadiness.
The results are illustrated in Fig. 9 in which q, was
computed by using equation (15) and q , + q , , is computed by using equations (15) and (27). It is seen
from Fig. 9 (left) that the effect of unsteadiness is
insignificant.
The effect of chemical heat release is included in the
present model. However, Fig. 9 (fight), in which the
results of equation (15) (including heat release) and
equation (16) (excluding heat release) are compared,
indicates that the neglect of the heat release source in

OF

DIESEL

ENGINE

The present heat transfer model was further applied

to a Caterpillar heavy-duty truck engine which is also
described in Table 1. A 60 sector mesh was adopted
in the computations (i.e. sector symmetry was
assumed to save computing time since the spray nozzle
had six injector holes) with 20 cells in the radial direction, 30 cells in the azimuthal direction and 5 cells in
the squish region at TDC. Typical grid size is about 2.2
mm in axial direction and 3.6 mm in radial direction in
the piston bowl. The azimuthal grid spacing is 2 .
Three locations on the cylinder head were chosen
to monitor wall heat flux computations. They are
located at 15, 45 and 60 mm from the cylinder axis
and are referred to as HT-1 to HT-3, respectively, as
indicated in Fig. 10. HT-1 is located near the injector,
HT-2 is located over the piston bowl close to the edge

622

Z. HAN and R. D. REITZ

2.5

&----

.....
2 .....

2.5

measured
computed, q u
computed, qn+qus

,<--

"1-

-'30

-20

-10

10

.o

1.5

,.q

Crank angle (degree)

.-

20

'

./

0.5

~'~l'~

G=O
G=Q c

-r-

0.5

measured

.....

gl

1.s

.....

30

-20-10

10

20

Crank angle (degree)

30

Fig. 9. Comparison of predicted wall heat flux showing the effects of unsteadiness (left) and chemical heat
release (right). Location HT-1.

Injector

HT-1

HT-2

HT-3

10

8
a.
~E

"

measured

6
4

Piston

13_

Liner

-50

50

I O0

Crank angle (degree)

350

of the bowl and HT-3 is located over the squish area

near the cylinder liner. In the Caterpillar engine, spray
droplets impinge on the piston-bowl surface shortly
after injection, and there is minimal spray wall interaction on the cylinder head. Hence, the neglect of
spray effects in the present heat transfer model is still
valid for the analysis of heat transfer process on the
cylinder head.
Comparison between computed and measured cylinder pressures and heat release rates [24, 25] are
shown in Fig. 11. Good agreements between the computations and measurements are seen. The computed
wall heat fluxes are shown in Fig. 12 in which the heat
release rate is also given for reference. It is seen that,
unlike the homogeneous charge engine discussed in
Fig. 4, the wall heat-flux distribution in this diesel
engine is highly nonuniform. Two heat-flux peaks are
seen at location HT-I. The first peak corresponds
to the initial premixed combustion after ignition, as
indicated by the heat release rate curve. The second
peak is caused by the high-temperature gas and the
flame reentering the piston-bowl due to the reverse
squish flow. Heat flux at HT-2 increases rapidly at
about 10 ATDC. At this time, both computations
and experimental flame images [26] indicate that the
high-temperature flame reaches the edge of the piston-

0
-100

Fig. 10. Schematic of chamber geometry of a Caterpillar

engine showing the heat flux monitoring locations.

300
"~

measured

250

15o
~oo
so
o
"7"

-50
-20

I
-10

I
0

I
I
I
I
10
20
30
40
Crank angle (degree)

I
50

60

Fig. 11. Computed and measured [24, 25] cylinder pressure

and heat release rate in a Caterpillar engine.
bowl where HT-2 is located. Hence, the value of heat
flux at HT-2 is as high as 10 M W m -2. Later, at about
20 ATDC, flame extends into the squish region and
the wall heat flux at HT-3 increases. Due to lack
of experimental data the computed heat flux is not
compared with measurements. However, the predicted heat flux changes are consistent with the flame
development process shown by the endoscope combustion images [26], and the magnitudes are also in
the range of previous measured values in heavy-duty
diesel engines [1, 27].

Temperature wall function formulation for turbulent flows

12

C,.,

10

<

623

8
4

3oo

4~

200,_,
100~
0
70

"1-

-2
-4

-10

10

20

30

40

50

Crank angle (degree)

60

Fig. 12. Computed wall heat flux. Heat release rate is given for reference.

12

.~
: "-.
l
\

---,10

od

HT-I
- - .... H T - 2
..... HT-3

4
~,~,.

10

20

30

40

50

60

Crank angle (degree)

Fig. 13. Comparison between the present heat transfer model and the KIVA model.

Predictions of l:he present model were compared

with those of the standard heat transfer model in
KIVA-II [4] as shown in Fig. 13. The standard KIVA
model uses an incompressible wall function approach.
Details of the model can be found in ref. [4]. It is seen
in Fig. 13 that the KIVA model predicts much lower
heat fluxes at all the monitoring locations compared
with those predicted by using present model. The peak
value at HT-2 is only about 2 M W m -2 which is too
low compared wit]h data found in the literature.
Effects of heat release and unsteadiness on the predicted heat flux are summarized in Fig. 14. The same
approach as disc~Lssed previously was used. As can
be seen the most noticeable difference between these
treatments is that the heat flux becomes somewhat
lower after it reaches its peak value when the model
that excludes all the sources and unsteadiness is used.
Therefore, it is cortcluded that the transient effect and

chemical heat release terms could be neglected for the

studied engine without significant errors.
CONCLUSIONS
A temperature wall function (temperature profile)
formulation was derived from the one-dimensional
energy conservation equation. This function is suitable for density-variable turbulent flows which are
commonly found in internal combustion engines. Due
to the gas density variation, the wall heat flux is found
to be proportional to the logarithm of the ratio of the
flow temperature to the wall temperature instead of
to the arithmetic difference of the two temperatures,
which is the case for incompressible flows.
The temperature wall function was implemented in
the KIVA-II computer code and applied to gas/wall
convective heat transfer predictions in a premixed-

624

Z. HAN and R. D. REITZ

,
HT-1

ness and chemical heat release permit the use of a

model which neglects unsteadiness and heat release
for engineering calculations without severe errors.

- - . . . . G=O

E 5

G=Qc

i'l"

Acknowledgements--This work was supported by Ford

Motor Co., Caterpillar Inc., and DOE/NASA-Lewis. Support for the computations was provided by Tacom, San
Diego Supercomputer Center and Cray Research, Inc.

v
tml . .

REFERENCES

'11
0

-10

12

10

20

30

40

50

60

Crank angle (degree)

t

10
HT-2

O4

/ ~

~X ' ~ 6

==_
"1"

2
I

-10

10
20
30
40
Crank angle (degree)

50

60

Fig. 14. Comparison showing the effects of unsteadiness and

chemical heat release on heat flux prediction in a Caterpillar
engine.

charge spark-ignition engine and a heavy-duty diesel

engine. Satisfactory agreement between the predicted
wall heat fluxes and the measured ones at several
monitoring locations was obtained for the premixedcharge engine under both firing and motoring
conditions. The heat transfer predicted by using the
present model was found to be independent of grid
size. Although no comparison with measurement was
made for the diesel engine predictions due to a lack
of measured data for that engine, the heat transfer
predictions indicated a significant nonuniform heat
flux distribution on the cylinder head and peak heat
flux values of 5-10 M W m -2 depending on the
location. Higher values were found over the edge of
the piston bowl. These characteristics and the magnitudes are supported by previous measurements in
other heavy-duty diesel engines found in literature.
It was found that gas compressibility affected heat
flux prediction significantly. Wall heat flux can be
greatly underpredicted with the use of a wall function
formulated for incompressible flows. Although the
present compressible flow model is derived under
quasi-steady condition, it was found that the effect of
unsteadiness was insignificant for the cases
considered. The effect of chemical heat release was
also found to be small. The weak effects of unsteadi-

1. G. L. Borman and K. Nishiwaki, International combustion engine heat transfer, Prog. Energy Combust. Sci.
13, 1 (1987).
2. J. Yang, IC engine gas-wall convective heat transfer-history, problem, and solutions, in Transport Phenomena
in Combustion (Edited by S. H. Chan). Taylor and
Francis, London (in press).
3. J. Yang and J. K. Martin, Approximate solution--onedimensional energy equation for transient, compressible,
low roach number turbulent boundary layer flows,
A S M E J. Heat Transfer 111,619 (1989).
4. A. A. Amsden, P. J. O'Rourke and T. D. Butler, KIVAII--a computer program for chemically reactive flows
with sprays, LA-11560-MS, Los Alamos National Labs
(1989).
5. R. D. Reitz, Assessment of wall heat transfer models for
premixed-charge engine combustion computations, SAE
Paper 910267 (1991).
6. G. L. Borman, In-cylinder heat transfer research at the
U. W. Engine Research Center, COMODIA 90 (1990).
7. W. M. Kays, Turbulent Prandtl number--where are we?,
A S M E J. Heat Transfer 116, 284 (1994).
8. H. Hattori, Y. Nagano and M. Tagawa, Analysis of
turbulent heat transfer under various thermal conditions
with two-equation models. In Engineering turbulence
modelling and experiments 2 (Edited by W. Rodi and F.
Martelli). Elsevier Science, Oxford (1993).
9. W. C. Reynolds, Computation of turbulent flows, A.
Rev. FluidMech. 8, 183 (1976).
10. G. L. Mellor, Proc. Symp. Fluidics Internal Flow,
Pennsylvania State University (1968).
11. V. Yakhot and S. A. Orszag, Renormalization group
analysis of turbulence. I. Basic theory, J. Sci. Comput.
1, 3 (1986).
12. Z. HanandR. D. Reitz, Turbulence modeling ofinternal
combustion engines using RNG k-e models, Comb. Sci.
Tech. 106, 267 (1995).
13. D. Choudhury, S. E. Kim and W. S. Flannery, Calculation of Turbulence Separated Flows Using a Renorrealization Group Based k-~ Turbulence Model, Vol. FED
149, p. 177. ASME, New York (1993).
14. Z. Han and R. D. Reitz, A RNG k-e model with application to diesel combustion modeling, In Transport
Phenomena in Combustion (Edited by S. H. Chan),
Taylor and Francis, London (in press).
15. B. E. Launder and D. B. Spalding, The numerical computation of turbulent flows, Comput. Meth. Appl. Mech.
Engng 3, 269 (1974).
16. J. Abraham, F. V. Braceo and R. D. Reitz, Comparison
of computed and measured premixed charge engine combustion, Combust. Flame 60, 309 (1985).
17. S. C. Kong, Z. Han and R. D. Reitz, The development
and application of a diesel ignition and combustion
model for multidimensional engine simulations, SAE
Paper 950278 (1995).
18. R. D. Reitz and T. W. Kuo, Modeling of HC emissions
due to crevice flows in premixed-charge engines, SAE
Paper 892085 (1989).
19. R. D. Reitz, Modeling atomization processes in highpressure vaporizing sprays, Atomization Spray Technol.
3, 309 (1987).

Temperature wall function formulation for turbulent flows

20. S.C. Kong and R. D. Reitz, Multidimensional modeling
of diesel ignition and combustion using a multistep kinetics model. 3". Engng Gas Turbines Power 115, 781
(1993).
21. E.G. Groff, A. C. Alkidas and J. P. Meyers, Combustion
data for an axisymmetric homogeneous-charge sparkignition engine, GM Research Publication GMR-3577
(1981).
22. A. C. Alkidas, Heat transfer characteristics of a sparkignition engine, Z Heat Transfer 102, 189 (1980).
23. W. M. Kays anti M. E. Crawfold, Convective Heat and
Mass Transfer. McGraw-Hill, New York (1980).

625

24. D. A. Nehmer and R. D. Reitz, Measurement of the

effect of injection rate and split injections on diesel engine
soot and NOx emissions, SAE Paper 940668 (1994).
25. C. T. Tow, A. Pierpont and R. D. Reitz, Reducing particulates and NOx emissions by using multiple injections
in a heavy duty D.I. diesel engine, SAE Paper 940890
(1994).
26. S. C. Kong, L. M. Ricart and R. D. Reitz, In-cylinder
diesel flame imaging compared with numerical computations, SAE Paper 950455 (1995).
27. J. B. Heywood, Internal Combustion Engine Fundamentals. McGraw-Hill, New York (1988).