Turbulent flow in an intake-manifold

Peter Dittrich and Heinrich Reister*)

Three-dimensional, turbulent fluid flow analysis with computational methods has emerged as a
viable tool in the design process of engine components for passenger cars. i f applied in the early
stages of the component development such analysis can help reduce the product development time
drastically, and may support the design engineer to evaluate several competing design aspects
before product completion. However the grid generation for complex geometries still poses a
difficult and time consuming task which strongly influences the accuracy of the numerical solution.
The present paper studies the numerical sohaion of the flow in an inlet-manifoM of afire-cylinder
engine with two intake-ports for each cylinder. A fast and reliable grid generation technique is
discussed in detail. The finite volume based Navier-Stokes solver with schemes of first and second
order accuracy for the convective terms is used. Numerical results at several mass flow rates and
different boundary conditions at the int,:ke ports are compared with measurements. Finally, the
computational results are discussed with ,'espect to their applicability to support the design process
of an inlet-manifold.

1. Introduction the SIMPLE procedure by Patankar and Spalding [4]. The

The design of inlet-manifolds is supported by the computa- standard two equation k-e solution method is used for the
tion of the unsteady, compressible flow field, where the turbulence closure problem. The present approach allows the
crucial role of the geometry of the plenum and the runners use of a completely unstructured grid. The iterative proce-
are varied to study the effects of the filling process at various dure is continued until the residuals for momentum, pressure
speed ranges of the combustion engines. These computations and the turbulent values are all reduced three orders of
are based on one-dimensional, time varying gasdynamic magnitude.
models, which allow the study of pulsating flows (i.e. varia- The computational objectives of the three-dimensional
tion in speed, valve area, hatake-valve timing etc.) [ 1; 2]. For model should include the following aspects. A turbulence
intake systems pressure waves are usually timed at specific model should be able to predict flow separation. The verifica-
operating speeds to improve the induction process. tion of frictional losses for a wide range of mass flows should
However these gasdynamic models are based on the a s - support the one-dimensional, unsteady simulations (i.e. con-
sumption of one-dimensional gas flow. The multifaceted stant discharge coefficients for a large Reynolds-number do-
aspects of flow bifurcation combined with separation and main). In this light, the verification of the numerical method
subsequently high pressure losses can not be modelled realis- for steady, three-dimensional flows can be seen as a precur-
tically with the one-dimensional flow assumption. Also the sor for unsteady flow simulations.
cumulative frictional losses of a complete induction system
3. Grid generation
can be directly related to the power rating of a combustion
engine. Therefore it is interesting to study every geometric F o r the complex bifurcation problem of the inlet-manifold
detail of such a system. a partially block-structured approach was first attempted.
In the present investigation the three-dimensional steady The whole geometry is divided in several directly accessible
flow is investigated for several mass flow rates of the intake blocks, where each partial block can be automatically gener-
manifold. The verification of the numerical method is studied ated by a structured volume mesh based on hexahedral
with respect to the convective properties of the numerical elements. Then at the interface, all the structured mesh
discretisation schemes as well as with respect to the influence blocks were coupled with its neighbouring sides to form the
of the numerical grid. The azialysises are also compared with complete integration domain. Special attention was paid at
experimental results obtained on a flow bench. the approximation of the rounded entrance from the plenum
The verified numerical method together with a suitable and to the runners of the inlet-manifold as well as at the bifurca-
efficient grid generation technique can be used to study tion of a runner to each of the two intake ports, Fig. 1.
several geometric design aspects and thus may also support There, the curvature of the surface at the pipe entrance and
unsteady three-dimensional flow simulations. the radius of the leading edge o f the flow-dividing plane are
important design parameters to be studied by the computa-
2. Numerical method tional analysis.
In a first attempt to compute the flow field with the above
The numerical simulations are based on the Finite Volume grid generation strategy a convergent solution was achieved.
Method (FVM) discretisation of the fluid dynamics code However the velocity field over a cross section of an induc-
STAR-CD [3]. The upwind differencing (UD) and linear tion pipe showed f o u r distinct peaks at the most distorted
upwind differencing schemes (LUD) of first and second order grid elements in the vicinity of the wall, Fig. 2. The strong
accuracy are used. The numerical solution method is based on distortion of the volume elements is due to the fact that each
grid block spans the whole cross section of a pipe.
The use of a wall function approach for the approximation
*) Ph.D.P.Dittrich;Dr.-Ing. H. Reister, Mercedes-BcnzAG, Stuttgart. of the turbulent boundary layer is violated for such a numer-

Forschung im Ingenieurwesen--Engineering Research E]d, 61 (!995) Nr. 718 195

Turbulent flow in an intake-manifold

where y ~ is expressed in local non-dimensional wall-normal

coordinates
Outlet # I
y 9 = pC,, 1 / 4 k ' / 2 y h , .

The above expressions correspond to the VanDriest mix-

ing-length model for the overlap wall region, where the
Outlet/12 values E =9.0, C, =0.09 and ~c =0.42 were used in the
numerfcal simulation. The wall function for the overlap
region is applied to approximate the turbulent boundary
layer with a single grid cell whose central node has to be
Outlet #3
positioned at 10 to 100y + units away from the wall. The
non-dimensional y+ distribution for an initial numerical grid
can posteriorly checked for a given flow situation.
However, often the inspection of the numerical results
Outlet #4
indicate that y § values lie outside the range given above and
a severe departure of assumptions made for the turbulent
boundary layer approximation becomes apparent. The inspec-
tion of the initial computation for a specified mass flowrate of
Outlet r
0.12 kg/s shows four distinct peaks in the velocity distribution
near the wall, Fig. 2. The y * values are below the given range
Fig. !. Computational domain of intake-manifold for the validity of the boundary layer assumption and thus the
near wall velocities u + are overpredicted by the above relation.
Several attempts to modify the grid for the near wall cells
ical grid. For the approximation of a turbulent boundary
were unsuccessful. In order to fulfil the wall function criteria
layer the numerical procedure assumes a locally one-dimen-
one has to severely distort the numerical mesh such that a
sional flow behaviour, where the effects of pressure gradients
strong departure f r o m orthogonality completely inhibits a
and body forces are negligibly small.
numerical solution.
The assumption of a uniform shear stress in the wall
Instead a grid generation procedure was developed, such
elements leads to the well known algebraic relation for the
that from any initially crude approximation of a complex
local tangential fluid velocity
geometry the numerical grid could be. modified in the vicinity
u + = I/K In(E),+), of the wall to control the boundary layer approximation. In

MAGNITUDE VELOCITY
M/S
LOCAL MX= 26.93
LOCAL MN= 13.91

25.00
24.07
23.14
22.21
21.29
20'.3~
" 19.43
18.50
17.57
~ ", 16.64
<: 15.71

r~ lZO~

Fig. 2. W a l l velocities for block structured grid

196 F o r s c h u n g im I n g e n i e u r w e s e n - - E n g i n e e r i n g R e s e a r c h Bd, 61 (!995) Nr. 718

 r
MAGNITUDE VELOCITY
M/S
LOCAL MX= 24,85
LOCAL MN= 11.96
24.65
23.93
23.01
22.09
21.17
20.25
19.33
16.41
9 : 17.48
~;~ 16.56
~C~'L~ 15.64
14.72
13.80
12.88
11.96

Fig. 3. Wall velocities for grid with additional surface layers

this approach several new layers of hexahedral cells were 4. Open Flow M a n i f o l d
created on the surface of the model, which were then subse-
quently shrunk back on the original surface of the model. With For this test case all ten outlets were opened wiih a
the information Of the initial computation based on the results constant pressure imposed. Then the mass flux distribution
obtained with the preliminary grid, the thickness of the outer can freely adjust within the manifold. Fig. 4 shows the mass
layer was determined so that the wall function approximation flux distribution for the open flow situation. The outer
is satisfied under all flow situations. runners receive more mass flow than the inner runners. Both
With this grid generation strategy, which was applied for two the UD and the LUD results agree very well with measure-
additional surface layers, the initial computation was repeated. ments. The higher order results (LUD) match better with the
The velocity profiles then showed no peaks in the wall cells and experiment.
the y § values could be controlled to lie within the range 20 A section cut through the plenum of the manifold explains,
to I00, Fig. 3. Fig. 5, why the filling rate of the inner tubes is less than for
The proposed grid generation strategy was applied twice to the outer tubes. First the flow bifurcates in two main streams
the initially block-structured grid of the intake manifold. The at the stagnation region below the inlet tube of the manifold.
final body fitted layer was determined after initial computa- The main flow is attached to the lower region of the plenum
tions allowed the estimation of the wall function assumptions. and two vortical structures can be identified, whose center is
The final numerical grid consists of approximately 180 000 close to the inlet region of the inner runners. While the outer
hexahedral cells and was used for the evaluation of the steady runners are being fed in a direct manner, the filling of the
flow results. inner tubes is mainly provided by a reverse flow in the upper
In order to evaluate the spatial resolution of the complex half of the plenum.
manifold domain, the discretisation of the nonlinear, convec- A comparison of the local velocities for UD and LUD
tive terms in the Navier-Stokes equation were varied. The differencing, Fig. 6 and 7, for a vertical section cut between
variation of the spatial step sizes for three-dimensional com- the runners 3 and 4 of the plenum shows larger velocity
putations is often inhibited because of the excessively large gradients for the higher order scheme due to higher numeri-
number of hexahedral elements (i.e. computer memory stor- cal diffusion of the first order scheme. The maximum veloc-
age). Therefore, it is important that the numerical grid allows ities of the u-velocity component as well as the v- and
accurate solutions for higher discretisation schemes. Computa- w-velocity component differ considerably between the two
tions were made using upwind differencing (UD) of first order numerical simulations.
accuracy, and with linear upwind differencing (LUD) tbr The need for higher order discretization schemes and the
second order accuracy on the convective terms. influence of the grid spacing truncation errors for in-cylinder

Forschung im Ingenieurwesen--Engineering Research Bd. 61 (1995) Nr. 7J8 197

 Turbulent flow in an intake-manifold

14

12

1 lO
8

0 Fig. 4. Mass flow distribution for open manifold

011 01r 021 02r 031 03r 041 04r 051 05r

VELOCITY MAGNITUDE
M/S
LOCAL MX= 44.58
LOCAL MN= .3022
"PRESENTATION GRID"
46.OO
42.71
39.43
36.14
32-84
29.57
26.29
23.O0
19.71
16.43
13.14
B.857
6.571
3.286
-.1431E-O5

Fig. 5. Velocities in section cut

Y Z through plenum

flows has been determined by Watkins et al. [5]. The coarse First a variation of four different mass flows were investi-
grid solution for the flow in the plenum of .the intake gated for the runner #4. A comparison for the UD and
manifold needs similar experimental verification as in the LUD differencing scheme with measurement is shown in Fig.
in-cylinder investigation of Watkins et. al. before the local 8. The quadratic dependence on mass flow on pressure drop
flow features of unsteady simulations can be interpreted. is apparent. Both differencing schemes agree within 5% with
measurement. As in the open flow case, the LUD results
5. Single Cylinder Flow match closer with experiment than the UD results.
For this~ type of calculation four of the five runners were A comparison was made of the pressure drop for all five
closed and only the two outlets leading to the intake-ports runners with the same mass flow. Each pressure drop was in
were open. At both outlets, again a constant pressure was the same order of magnitude for all five intake pipes with a
imposed. Then the mass flow distribution of the flow split in maximum difference of 4% between the highest and the
each individual runner can be studied. lowest levels. This fact reinforces the good design of the

198 Forschung im Ingenieurwesen--Engineering Research Bd, 6 t (1995) Nr_ 7/8

 40 | | | | | i r | | , |

V_UD - -
35 L U D ---" 4 V _ L U D ....
W _ U D ......
30 W _ L U D --"'7'
,,
25

2O

0 ................. "~'~" . . . . . . . . . . . . . . . . . . ,r"

m lO

:
5

o
-5
-I0
-40
I
-30
I
-20
I
-I0
distance (mm)
I
0
I
I0 20
I
-4

-40 -30
I I
-20 -i0
I I
0
I
i0
j
20
distance (mm)

Fig. 6. Comparison of u-velocity component in plenum section cut Fig. 7. Comparison of v- and w-velocity components in plenum section
between runner 3 and 4 cut between runner 3 and 4

geometrical transition from the plenum to the runners of the that the severe mesh distortions are removed for the wall
intake manifold. cell-layers. With this procedure, the wall function implemen-
tation of the turbulent flow model can be controlled so that
6. Conclusion its underlying assumptions are never violated.
Three-dimensional fluid flow computation for the steady With this mesh generation strategy, steady flow computa-
filling process of an intake manifold are presented. The tions were performed for the intake-manifold. A comparison
computations were made with the commercially available with measurement was made for the case with all ten outlet
code STAR-CD which is based on the finite volume method. ports open (open flow manifold) and for the case with only
A two-equation k-e turbulence model together with an al- two outlet ports open (single cylinder flow). These two
gebraical model for the turbulent boundary layer was steady flow cases can be understood as extremes between
applied. which the real unsteady process will be situated and the
A first, numerical solution showed a strong grid depen- validation with experiments can be understood as a necessary
dency in the vicinity of the wall. This was due to strong mesh assumption for unsteady computations.
distortions which resulted from the geometrical complexity of For the comparison of mass flux distribution and pressure
the solution domain. A method was proposed which allows loss between inlet and outlet of the solution domain the
to modify a distorted mesh to be reformed in such a way, computations show good agreement with the experimental

35
mbar ]
30

25

20

15

10

5
Fig. 8. Comparison of pressure drop with mea-
0 surement for several mass flows
0.0183 0.0370 0.0561 kg/s 0.0765

Forschung im Ingenieurwesen--Engineering Research Bd. 61 (1995) Nr, 7/8 t09

Turbulent flow in an intake-manifold

results. The second-order L U D solution shows better agree- References

ment than the first-order UD differencing scheme, however [1] Seifert, H.: Die Berechnung instation~irer Strrmungsvor-
both solutions deviate by less than 5%, so that one can g~inge in den Rohrleitungs-Systemen yon Mehrzylinder-
presume sufficiently accurate numerical resolution with the
motoren. MTZ Motortechnische Zeitung 74 (1972) 11.
mesh spacing of the present grid.
[21 Seifert, H.: Erfahrungen mit einem mathematischen Mod-
A comparison, of the local flow field in the plenum of the ell zur Simulation yon Arbeitsverfahren in Verbrennungs-
manifold reveals a 9 dependence on the discretisation 9motoren. MTZ Motortechnische Zeitung 39 (1978) 7/8.
scheme. The L U D scheme shows steeper velocity gradients [3] STAR-CD Users Manual, 1993. Computational Dynam-
than the UD scheme. Thus care must be taken when inter- ics, London.
preting the local flow field which may be especially important [4] Patankar, S.V. and Spalding, D.B.: A Calculation Proce-
for unsteady computations. dure for Heat, Mass and Momentum Transfer in Three-
After the successful validation of the steady flow field, Dimensional Parabolic Flows. Int. J. Heat, a. Mass
important design features may then be studied. The effect of Transfer 15 (1972) pp 1787-1806.
geometry variations as a remedy to prevent flow separation
[51 Watkins, A.P.: Kanellakopoulos, P. and Lea, C.J.: An
can be analysed and the computed frictional losses can be Assessment of Discretisation Schemes and Turbulence
used for computations based on one-dimensional gasdynamic Models for In-Cylinder Flows. International Symposium
models. With proper 9unsteady boundary conditions for the C O M O D I A 90:499-504 (1990).
intake-ports the model can also be used to study unsteady,
three-dimensional gasdynamic effects. Received December 16, 1994 F 4208

200 Forschung im Ingenieurwesen---Engineering Research Bd. 61 (1995) Nr. 7/8