International Conference on Mechanical And Industrial Engineering (ICMAIE2015) Feb.

8-9, 2015 Kuala Lumpur (Malaysia)

Validations Of Openfoam Steady State

Compressible Solver Rhosimplefoam
Umran Abdul Rahman, and Faizal Mustapha

Milovan PeriC [7]. The concept of solution is based on

AbstractOpenFOAM steady state solver rhoSimpleFoam discretization of the flow equations (Navier Stokes equations
was tested. Reynolds Average Navier Stokes (RANS) coupled with in integral form integral form as FVM is the base) into
kOmegaSST turbulent model simulation was performed. The algebraic equations in the form of,
simulation was based on 0.8395 Mach with Reynolds Number of
11.72 x 106 of air flow parameters over ONERA M6 wing section at Ax b (1)
3.06 degrees angle of attack; an overall close fitting to the Where A is the coefficient matrix, x is the variable vector to
experimental values of Coefficient of Pressure (Cp) were observed. be solved and b is the boundary conditions. Looking at the
Some minor differences in a few pressure tapping ports, especially at simplistic nature of (1), direct deduction can be made i.e. the
20% span cross section and at 99% span cross section were also
number of discretized equations and unknowns must be
noticed. The typically called Lambda shaped wave pattern was
also observed on the wing top surface.
equivalent to the number of finite control volumes in the
calculation domain. Since the involved equations are
Index TermsRhoSimpleFoam, Compressible, Transonic, non-linear to start with, direct solution will be too expensive
Onera M6. therefore the sequential solution is adopted. The technique
involved solving each variable one at a time while holding the
I. INTRODUCTION rest of the dependent variables as constants. This is where
there are two iteration loops occurred, the first loop is the
OpenFOAM is an open source CFD package which uses inner iteration for obtaining the value of the individual
primarily Finite Volume Method (FVM) in solving CFD variable, while the next is the outer iteration where the saved
cases. Built in with its own meshing tools, multiple type of values from inner loop being used to solve the next variables.
solvers and a third party data visualizer ParaVIEW, these Between each outer loop, if the solved variable values
complete packages are there for the pickings of CFD users being used directly in solving the next variable, chances are it
and researchers alike. will cause the next variable values to be in an invalid physical
In this paper, the concentration will be to compare the domain. For example, the density variable may become a
results of wind tunnel testing against steady state negative entity at certain region of the grid and this
compressible solver of OpenFOAM i.e. rhoSimpleFoam. non-physical situation will definitely blow up the iteration
The test subject is the ONERA M6 wing panel where it was process. To curb this, relaxation factors for each variable will
built and tested in a wind tunnel back in 1979 by Schmitt, V & be dialed in especially during the initial phase of the iteration.
F. Charpin [1]. This particular reference being selected as it Convergence of the solution can be observed thru the
has been the benchmark in validating many other CFD minimum residuals (the difference between previous variable
software [2][5] which provides solvers of similar capability. values to current values) of each variable post inner loop
It has been known that the unpredictable nature of transonic and/or the consistency of the value of the variables itself thru
flow poses a great challenge for any solver to closely predict the iterations. Later is, at times, is a better indication of
the flow conditions and this is more so on a relatively complex convergence as there are occasions where individual variable
boundary; therefore this validation test is an important step in residuals will not fall down to machine zero yet the value of
establishing the solver for practical application in solving the variables itself doesnt change any longer even though the
CFD cases of similar flow parameters. iterations were being pro-longed.
II. BACKGROUND III. MODELLING
rhoSimpleFoam (from OpenFOAM v2.3.0 package [6]) Apart from the given ONERA D top half airfoil coordinates
is the extension of incompressible steady state solver (73 points), given also the plan form drawing which can be
simpleFoam which uses the Semi-Implicit Method for used to produce 3D drawings of the symmetrical wing.
Pressure Link Equations (SIMPLE) algorithm. This algorithm However, during production of the 3D model of the wing,
was introduced by Patankar & Spalding (1972) and later more there are a few discrepancies found on the given dimensions.
detail utilization concept was presented by Joel H. Ferziger & The data presented in TABLE I shows the actual plan form
dimensions as per the AGARD 138 report [1].
Manuscript sent December 6, 2014. This work was supported in part by
the Ministry of Education, Malaysia under MyBrain15 scholarship for TABLE I
Industrial Doctorate Student (GS34199) University Putra Malaysia (UPM). PLAN FORM DETAILS AS PER AGARD 138
Umran is with the Malaysia Airlines. He is now the engineer in charge for Aspect Ratio 3.8
A380 fleet while pursuing his doctorate studies with UPM (e-mail:
[email protected]). L.E sweep 30 degrees
Prof Dr. Faizal Musapha is with Department of Aerospace Engineering,
T.E sweep 15.8 degrees
UPM, Serdang, Selangor, Malaysia (e-mail: [email protected]).

http://dx.doi.org/10.15242/IAE.IAE0215214 61
 International Conference on Mechanical And Industrial Engineering (ICMAIE2015) Feb. 8-9, 2015 Kuala Lumpur (Malaysia)

Taper Ratio 0.562

Mean Aerodynamic Chord (MAC) 0.64607 m
Semi span Length 1.1963 m
Semi span Area 0.7532 m2
Root chord 0.8059 m
Tip chord (based on the above Taper Ratio) 0.452916 m
Rectangular length (dimension bound) 1.5787 m
Fig. 2 Tip Design
Rectangular width (dimension bound) 2.7345 m
Tip shape - half round rotation of airfoil N/A

By using the given taper ratio of 0.562, sweep angles and

root chord, the semi-span area worked out to be 0.75296 m2
and MAC to be 0.64590 m where these values are lower than
the data presented in the report. The significant effect of these
differences is the Reynolds Number calculation later for flow
similarities and of course the net effect of the pressure
gradient. In order to maintain the effective area and MAC as
per the document, the taper ratio needed to be increased to
0.562516 instead of 0.562. Consequently the tip chord grows
slightly to 0.453332 m instead of 0.452916 m resulting in
MAC to be 0.646068 m and the effective area becomes
0.753209 m2. By percentage of rounding off errors, this
reduces the error down to only 0.0003%.
There are also issues in shaping the wing tip as there are no
actual coordinates given, however the report do mention Fig. 3 Plan Form
about the tip being formed by rotational extrusion of the tip
airfoil half way. If this statement being applied directly, the IV. MESHING
sweep Leading Edge (LE) will meet with a forward bulge of From the given Onera D airfoil coordinates; by extrusion,
the tip due to the rotation axis is in parallel with stream flow swept translation, tip scaling and lastly rotation, the basic 3D
while the LE is slanted rearwards. From the report photo (Fig. volume coordinates were created based on Fig. 3 plan form
1), this doesnt seem to be the case. To alleviate this issue, the dimensions. These initial volume coordinates together with
rotation axis maintained, however, as the tip airfoil being the extrusion path coordinates up to the domain boundary
rotated; at each of the rotated tip airfoil coordinate, rearwards faces being fed into the block meshing tool (blockMesh) of
transformation being applied according to each coordinate OpenFOAM in creating a fully structured mesh. Special
swept angle from the root airfoil coordinates, resulting in a care was given in ensuring:
smooth LE of the tip against the swept angle, this is however 1. All cells grow in volume towards external boundaries
purely an estimation. Fig. 2 shows the tip design while Fig. 3 with controlled aspect ratio.
shows the plan form dimension of the generated 3D model. 2. Proper ratio of cell height and width between adjacent
cells.
3. Flatness of cell walls.
4. Skew angle of cell walls being kept minimal as the
cell grows towards external boundaries.
5. Only hexahedral cells used.

Mesh Statistic
Points: 690279; Faces: 2032484; Internal Faces: 1994476;
Cells: 671160.
Overall Domain Dimension: 30 m x 30 m x 14 m (length x
Fig. 1 Actual Onera M6 Wing, Note The Tip Shape height x width).
Wing Location: Wing root planted on symmetry boundary
where the root LE is at 10m from inlet and 15m equidistant
from top and bottom boundaries, refers Fig. 4.

http://dx.doi.org/10.15242/IAE.IAE0215214 62
 International Conference on Mechanical And Industrial Engineering (ICMAIE2015) Feb. 8-9, 2015 Kuala Lumpur (Malaysia)

V. SIMULATION PARAMETERS
Mach number: 0.8395; Angle of Attack: 3.06 degrees;
Reynolds Number: 11.72 x 106. These parameters were given
in the report where the stagnation temperature ranges between
292 and 315 Kelvin (K). In order to match the flow
parameters, Reynolds Number equation,
VL
Re (2)

Where V is velocity, L is the effective length (MAC in this
case, which is 0.646068 m), is density and is dynamic
viscosity, coupled with Mach number equations,
V
M (3)
Fig. 4 Overall Domain a
Equation of State,
p
(4)
RT
Sonic equation,
a RT (5)
Where is the local sonic velocity, R is the gas constant
(287 J/Kg.K), is the ratio of thermal constants (1.4) and T is
the static temperature, and also the Sunderlands equation [8]
which linked the temperature to dynamic viscosity,
Fig. 5 Surface Mesh 3
CT 2
(6)
T S
Where C is the constant (1.458 x 10-6 kg/msK), S is the
Sunderlands Temperature (110.4 K) and by combining the
equations from (3) to (6) into (2), fixing a pressure value
(taken as standard sea level air pressure of 101325 N/m2) and
inserting the rest of the known variable values, the Reynolds
Number equation reduced to the function of temperature only.
By applying iterative procedure, a temperature of
305.7335559 K was found to satisfy the effective flow
Reynolds Number. By substituting this temperature value into
Fig. 6 Nose Tip
(5) and later into (3) and by using the given Mach number, the
relative air velocity was determined to be 294.236957 m/s.
These 3 values (temperature, pressure and velocity) were used
as the boundary conditions of the calculation domain where
the velocity being vectored accordingly to achieve 3.06
degrees angle of attack.
A minor argument can be made here; there is no mentioned
of actual static air pressure of the flow at Mach 0.8395 but it
does mention about dynamic air pressure between 1600 to
64000 N/m2 for the whole tested range of Mach numbers
between 0.27 to 1.33. It does also mention the stagnation
pressure values fall within 30,000 to 210,000 N/m2 for Mach
Fig. 7 Tail Tip
0.8. With this info alone, it is relatively vague in determining
the actual static air pressure at the inlet boundary, therefore
the question will be, how exactly we know that by taking the
standard sea level ambient air pressure is sufficed to mimic
the inlet static air pressure during the wind tunnel testing?
First of all, we dont, however, instead of simulating air
being blown over the wing such as what the tunnel testing did;
this simulation can be seen as moving the wing itself into a
static air where the actual velocity of the wing and the
resultant airflow temperature which passes the wing can be
controlled at the domain boundary level. This is where the
value hold and the effective Mach and Reynolds Number
Fig. 8 Horizontal Cut (Cell Growth)
matched.

http://dx.doi.org/10.15242/IAE.IAE0215214 63
 International Conference on Mechanical And Industrial Engineering (ICMAIE2015) Feb. 8-9, 2015 Kuala Lumpur (Malaysia)

VI. TURBULENCE MODELLING

Noting that the flow is in a relatively high Reynolds
Number region, turbulent flow is expected. As this simulation
is a RANS type simulation, turbulent eddies cant be fully
resolved as in Direct Numerical Simulation (DNS), therefore
it requires modeling. The trade off here is, due to some level
of approximations are needed in turbulent modeling,
expectedly, some level of precision will have to be sacrificed.
This sacrificial however, resulting in the grid count that
doesnt need to be at a level where a typical modern desktop
will not be able to cope with in order to obtain a reasonable
flow prediction. In this case study, the turbulence was
Fig. 11 Results At 50,000 Iterations
modeled using k-omega SST (Shear Stress Transport)
developed by Menter, F.R [9] where the inlet boundary values As for numerical comparisons, the selected span sectional
of k and omega were generated using empirical estimations locations were based on the available data from wind tunnel
based on [10]. test results. Post simulation, the surface pressure data at those
wingspan locations were sampled and the inversed Cp values
VII. RESULTS & DISCUSSIONS were generated by passing the data through (7).
The simulation was allowed to run up to 50,000 iterations 2 p
Cp 1 (7)
as this will permit the observation on residual trending M 2 p
together with the consistency of the solved variables.
These data sampling and producing the output Cp values
were done in ParaView. It was achieved by the usage of
Calculator and Plot On Intersection Curves filters. As the
available wind tunnel data at each selected location on the
span were normalized in terms of its chord, these numerically
obtained data were normalized as well in order to have a
consistent plot axis. Both data (at the same location) were
plotted against each other using Microsoft Excel and can be
inspected in Fig. 12 through Fig. 18.

Fig. 9 Residual Trending

Based on Fig. 9, it can be observed that the simulation
heading into a converged solution where the rippling
frequency of each variable residual continually trending
towards steady state as the iterations progressed. With
reference to Fig. 10 and 11, it can be readily seen that the Cp
(inversed) at the selected cross sections of the wing marked
very subtle differences between 30,000 and 50,000 iterations.
This is another indication depicting a converged solution.
From the same figures as well, the sharply defined
(lambda) shaped Cp gradient on the top wing can be observed,
Fig. 12
indicating the expected occurrence of shock waves on the top
surface were fully resolved.

Fig. 13

Fig. 10 - Results at 30,000 Iterations

http://dx.doi.org/10.15242/IAE.IAE0215214 64
 International Conference on Mechanical And Industrial Engineering (ICMAIE2015) Feb. 8-9, 2015 Kuala Lumpur (Malaysia)

k-omegaSST turbulent model has the ability to closely predict

the resultant surface pressure from a transonic flow condition.
For a more quantitative comparison; by taking the nett
different of the area formed under each curve between
experiment Cp and numerical Cp values, in terms of
percentages, TABLE II below was produced.

TABLE II
NETT DIFFERENCE OF AREA BETWEEN THE CP CURVES
Span Location Top Diff (%) Bottom Diff (%) Average Diff (%)
20% 6.717 9.054 7.886
44% 2.570 1.462 2.016
Fig. 14 65% 0.775 5.314 3.044
80% 0.570 5.799 3.185
90% 3.168 2.066 2.617
95% 0.715 1.949 1.332
99% 11.814 3.862 7.838

The average differences can be found to be less than 10%

indicating a close proximity of CFD solutions as compared to
wind tunnel data where at locations away from the extremities
i.e. between stations 44 to 95, the average differences are even
lesser.
By going back to Fig. 12 and Fig. 18, minor discrepancies
can be seen and this can be confirmed by making cross
Fig. 15
reference to TABLE II. At these extremities locations, those
differences may be attributed to the differences in modeling.
In Fig. 1, the actual tunnel testing of the semi span wing, one
can see a divider plate was installed at the root of the wing.
The dimension and thickness of this plate is not available and
so does its distance from the tunnel wall. On this plate, there
will be a buildup of boundary layer which grows in the span
wise direction. This boundary layer may not only affecting the
pressure gradient on the wing surface, but it may also (to a
certain extend) affecting the location at which the formation
of the shock waves on the wing surface. Where else in the
CFD simulation, an ideal symmetrical boundary condition
Fig. 16 was applied on the wall at which the wing was attached. As for
the station 99 (Fig. 18), the tip shaping during 3D modeling
can differ slightly from the actual wing therefore a slightly
different result expected.
It is worth to mention as well, at 80% span (Fig. 15), the
second shock on the wing surface is not clearly visible when
compared to wind tunnel test data. The contributing factor to
this phenomena is the junction of wave (the point where all
the three arms met) on the wing top surface (refer Fig. 10 &
11) was shifted more towards the wing root in CFD simulation
as compared to wind tunnel testing. This shift can be the
product from inaccuracy of tip shaping as well as the
Fig. 17 approximations involving turbulence modeling.

VIII. CONCLUSIONS
OpenFOAM usage may require a steep learning as the
users need to be accustomed to text input files rather than the
usual graphic interface which normally found on commercial
CFD packages; this however, a very small price to pay when
compared to its ability to expand without restrictions and its
zero cost factor.
In engineering, the results obtained from this validation
work are more than enough in meeting the industry standards,
however, from scientific point of view, perhaps more can be
Fig. 18
pursued such as running a DNS simulation on the 3D mesh
Based on the comparative presented data, the numerically
generated from laser scanning of the actual ONERA M6 wing
obtained solutions by rhoSimpleFoam solver coupled with

http://dx.doi.org/10.15242/IAE.IAE0215214 65
 International Conference on Mechanical And Industrial Engineering (ICMAIE2015) Feb. 8-9, 2015 Kuala Lumpur (Malaysia)

panel. The Author was born in 1967, Apr12. Received

Aircraft Maintenance Engineers License in 1992
For all practical purposes, this presentation has established
from Civil Aviation Authority, United Kingdom.
rhoSimpleFoam, the steady state transonic solver from Obtained a Bachelor in Mechanical Engineering
OpenFOAM is on par with (if not better than) many other from University Kebangsaan Malaysia (1999) and
commercial codes of similar capability. later completed a Masters of Science course from
Institute of Technology Bandung, Indonesia in
year 2006. Currently pursuing his Doctorate in
ACKNOWLEDGMENT Industrial Engineering from University Putra
The Author would like to thank the Board Members of Malaysia.
Work as an Engineer for Malaysia Airlines since
School of Graduate Studies, University Putra Malaysia 1992 and currently being tasked to manage the A380 fleet. Have produced
(UPM). Thanks also to D.Eng program advisors Dr. unpublished reports on the fleet performance and defect analysis for aircraft
Mohamed Thariq and Dr. Azmin Shakrine for the guidances manufacturer review, which include identifying root causes of certain
given in preparing this paper. To Prof. Dr. Abdul Aziz and critical defects on the fleet.
Mr. Umran Abdul Rahman being engaged in consulting work on UAVs
Prof. Dr. Nik Abdullah from University Malaysia Pahang by UPM, open source CFD system implementation in UMP and also being
(UMP), thanks for the interest in Authors work. Special engaged by an independent contractor to perform aerodynamic analysis on
thanks to The Ministry of Education, Malaysia for the granted UAV platform suggested for Ministry of Defense, Malaysia.
scholarship under MyBrain15 program. Prof. Dr. Faizal Mustapha is a distinguish Professor in UPM on Structural
Analysis. Obtained his PhD in Structure Health Monitoring from University
Of Sheffield, U.K. Currently engganging in various sectors which needed
REFERENCES consultation on structure condition monitoring.
[1] Schmitt, V. & F. Charpin, "Pressure Distributions on the
ONERA-M6-Wing at Transonic Mach Numbers," Experimental Data
Base for Computer Program Assessment. Report of the Fluid
Dynamics Panel Working Group 04, AGARD AR 138, May 1979.
[2] Christopher L. Rumsey and Thomas B. Gatski, Summary of EASM
Turbulence Models in CFL3D With Validation Test Cases,
NASA/TM-2003-212431, June, 2003.
[3] Tatum, K.E. & J.W. Slater, "The Validation Archive of the NPARC
Alliance", AIAA Paper 99-0747, January, 1999.
[4] Taku Nagata et al, Validation of new CFD Tool Using Non-orthogonal
Octree with Boundary-fitted Layer Unstructured Grid AIAA
2012-1259, 2012.
[5] Francisco Palacios et al, Stanford University Unstructured (SU2):
Open-source Analysis and Design Technology for Turbulent Flows,
AIAA paper 2014-0243, January 2014.
[6] http://www.OpenFOAM.org/download/ubuntu.php (25/07/2014)
[7] Joel H. Ferziger and Milovan PeriC, Computational Methods for Fluid
Dynamics, ISBN 3-540-42074-6, 3rd Edition, 2002.
[8] The viscosity of gases and molecular force. Philosophical Magazine,
S.5, volume 36 (1893), 507-531. (Obtained the ref summary from
http://www.cfd-online.com/Wiki/Sutherlands_law - 22/04/2014)
[9] Menter, F. R., "Two-Equation Eddy-Viscosity Turbulence Models for
Engineering Applications," AIAA Journal, Vol. 32, No. 8, August
1994, pp. 1598-1605.
[10] http://www.cfd-online.com/Wiki/Turbulence_free-stream_boundary_
conditions - 18/07/2013.

http://dx.doi.org/10.15242/IAE.IAE0215214 66