Vous êtes sur la page 1sur 11

EUROTHERM 82

Numerical Heat Transfer 2005


September 13-16, 2005, Gliwice-Cracow, Poland
Eds.: A. Nowak, R.A. Biaecki
High Rayleigh Number Natural Convection in a Cubic Enclosure
Tomasz Michaek*
Institute of Fundamental Technological Research
Polish Academy of Sciences, witokrzyska 21, 00-049 Warsaw, Poland
e-mail: tmichal@ippt.gov.pl
Key words: natural convection, verification, validation, cavity heated from the side, high Rayleigh
numbers flows, Particle Image Velocimetry and Thermometry
Abstract
Experimental results concerning flows in water filled cavity heated from the side are presented.
Laboratory configuration was designed making use of sensitivity analysis for the sake of boundary and
initial conditions and material properties of water in order to determine sufficient accuracy of
measurements. This configuration allowed to analyze natural convective flows for Rayleigh number
values up to 10
9
. Two dimensional velocity and temperature fields in the central cross-section of the
cavity were obtained making use of Particle Image Velocimetry and Thermometry methods.
Subsequently, these results were used for validation of numerical simulations. Transition from laminar
to unsteady flow regime was investigated based on experimental and computational results.
1 Introduction
Credibility of computational simulations concerning fluid flow and heat transfer are evaluated through
two subsequent procedures: verification and validation. The main aim of verification is assessment of
numerical errors whereas in validation conceptual modeling errors are evaluated. The literature
referred to these methodologies [1,2,3] has several factors in common. First of all, these procedures are
defined as processes which concern accuracy. Secondly, verification should precede validation.
Finally, verification can bear no relation to experiments, whereas in validation experimental data are
essential.

Semantics and methods for verification are much more mature in comparison to validation
methodologies. Verification procedure is divided into two subsequent steps: verification of code and
verification of actual calculation. For each of these steps following methods were proposed: the
method of manufactured solutions [4] or analytical and numerical benchmarking for code verification,
and the generalized Richardson extrapolation [1], the Grid Convergence Index [2] and lately a least
square extrapolation [5] for the verification of an actual calculation. To date, there is no agreement for
Tomasz Michaek


any generic methods in validation. Some of the scientists, like Roache [2], claim that validation is
inherent and solely connected to experimental measurements and specific phenomena so generic
methodology is not possible but only general indications. On the other hand, a generic model for
validation was proposed by Sindir at all. [6], so called a building block approach. This approach
divides validation procedure into four phases: unit problems, benchmark cases, simple/partial flow
path and actual hardware. It was suggested that validation should start from simple cases and in turn
geometric and flow complexity should be increased. This approach allows meticulous validation
because it is possible to obtain reliable and accurate experimental data for simple problems. But with
the increase of flow complexity, decrease of data availability and accuracy can not be avoided.

Taking into account all mentioned issues, the aim of this paper is to present application of the
sensitivity analysis for designing validation experiments, which concern natural convection flows for
high Rayleigh numbers in enclosures. Particularly, the sensitivity analysis is used for identification of
fundamental parameters in experimental configurations and estimation of necessary precision in
measurements. We followed the suggestion by Aeschliman and Oberkampf [7] that >> beneficial
synergism arises when experiments are designed specifically for CFD Validation<<. In our opinion,
such approach enables us to reduce number of cycles usually performed in order to reach agreement
between experimental data and simulation results for non trivial physical phenomena. That is why that
can lead to succinct and fast validation.

Natural convection of water in the differentially heated cavity was chosen for validation study. This
configuration was investigated in previous research [8,9] for Rayleigh numbers up to 10
6
, so we
decided to investigate this configuration for higher values of Ra. Our experimental set-up permits to
obtain natural convective flows with Rayleigh numbers up to 10
9
. We studied natural convection of
water making use of Particle Image Velocimetry and Thermometry methods. That allows us to study
transition from laminar to unsteady flow regime. Assumption of stationary of flow belongs to
modeling assumptions in computational simulations and its correctness was checked through
validation procedure.

Before experimental data are shown, we succinctly describe the experimental configuration and
experimental techniques. Additionally, we present results of the sensitivity analysis performed for this
experimental configuration. Further, three kinds of experimental data are presented. First of all,
examples of a full two dimensional velocity and temperature fields are shown in the central cross-
section of cavity. Secondly, velocity fields are presented for a fragment of the central cross-section of
the cube located in the bottom right corner. These data characterize much higher precision, so
statistical moments and turbulence intensity were calculated and analyzed on their basis. Moreover,
experimental data in form of time series of velocity components were analyzed for several points
located in the bottom right corner. Finally experimental data were compared with computational
results for two cases Ra = 3x10
7
and Ra = 1.5x10
8
.
2 Experimental Validation
2.1 Sensitivity analysis for validation
Numerical solution, even very strictly verified, has only an academic value as long as it is not
validated, i.e. confronted with the real world physics. Before we can answer the prime question of
the validation procedure, is our code solving a proper set of equations, careful analysis of all
EUROTHERM 82, September 13-16, 2005, Gliwice-Cracow, Poland


physical details of analysed phenomena must be performed. To make it clear, it is not possible to
match physical phenomena to an idealized numerical benchmark . Nothing like isothermal or adiabatic
wall exists in reality, fluids are not ideal and their variable properties must be known. Hence, the
question arises, how exact description of the physical phenomena is necessary. In our opinion the
answer may be drawn from numerical sensitivity tests only [10].
Hence, before applying numerical model to simulate physical experiment a careful sensitivity analysis
of numerical results was performed to determine the most important parameters describing our
configuration. Moreover, we estimate the precision required for description of those parameters to
conduct a full validation procedure. Details of the applied sensitivity analysis will be presented in
forthcoming thesis (Michaek [11]).

The sensitivity analysis was conducted for the sake of boundary conditions, initial conditions and fluid
properties. Performed analysis revealed that considered system (see Fig.1) strongly depends on
thermal boundary conditions imposed on external walls. Heat fluxes not only through adiabatic but
also trough isothermal walls were analyzed. It appeared that small variations in heat flux Q
1

considerably changed the values of velocity components for monitored points inside the fluid square
domain. In opposite to configuration analyzed in previous paper [10] variations in fluxes Q
2
and Q
3

caused less noticeable changes, so the configuration turned out to be less sensitive to changes for those
parameters.

Sensitivity analysis for the sake of material properties was conducted by performing calculations with
altered values of dynamic viscosity , thermal conductivity and specific heat c
p
. Change in values of
material properties was determined through analyzing total variability of these properties with
temperature within applied temperature range. It was found that the most vulnerable for changes were
values of velocity components (10 %) due to change of specific heat and decrease of thermal
conductivity (about 10 %). Other alterations in values of material properties caused less noticeable
changes (about 4% observed for velocity components). It turned out that only with the 2D
measurements of velocity field it is possible to determine sensitivity of flow for the sake of material
properties. It is not possible on the basis of point temperature measurements.

Initial conditions appeared to have minor influence to final results of our calculations. Even strong
perturbation imposed in the initial temperature field did not cause any noticeable alterations in final
steady states. Taking into account results of the sensitivity analysis of the numerical model the
experimental setup was designed in order to meet all mentioned requirements of full validation
process.
2.2 Description of experiment and measurements
Experimental setup consists of cubic cavity with internal size 0.08 x 0.08 x 0.08 m, with two opposite
side-walls made of a 14 mm thick aluminium, and four remaining walls made of a 8 mm thick
Plexiglas. The left side aluminium wall was heated by coolant kept on the constant temperature T
H
,
whereas the right side aluminium wall was cooled by coolant kept on the constant temperature T
C
.
Natural convective flow of water was driven by temperature gradient T = T
H
T
C
. A set of
thermocouples was installed in the aluminium walls, the Plexiglas walls, and in the vicinity of the
cavity in order to monitor local air temperature T
ext
and precisely calculate heat fluxes. Position of
thermocouples in central cross-section of the cavity was depicted in Fig.1 (circles). Steady state
convection was assumed after running the experiment for several hours.
Tomasz Michaek


Figure 1: Sketch of the central cross-section of physical geometry used in the validation experiment.



Several configurations with different temperature boundary conditions (T
H
and T
C
) were investigated.
In this paper, we limit our consideration to four of them. Details of the boundary conditions with
dimensionless numbers describing considered configurations are presented in Table 1. Rayleigh
number is based on the obtained temperature difference between the internal aluminium walls T,
internal size of cavity L = 0.08 m and fluid properties at the reference temperature T
ref
= (T
H
+T
C
)/2.
Details of the experimental set-up and data for other configurations will be presented in [11].
Table 1: List of experiments.
No T
H
T
C
T
ext T Ra Pr
1 18.5 4.0 26.0 14.5 3x10
7
9.53
2 27.2 6.8 25.5 20.4 1.5x10
8
7.01
3 36.5 8.5 24.0 28.0 1.8x10
8
7.01
4 46.2 14.4 26.0 31.8 4.4x10
8
5.43

Thermochromic liquid crystals were used as tracers in order to measure simultaneously two-
dimensional velocity and temperature fields. Quantitative experimental data on velocity and
temperature fields in a central cross-section were obtained by making use of Particle Image
Velocimetry (PIV) and Particle Image Thermometry (PIT) techniques [12]. Figure 2a presents an
experimental image of liquid crystal tracers changing their colour with temperature between 19.5C
and 27.2C (red 19.5C-20.4C, yellow 20.4C 20.5C, green 20.5C22.9C, blue 22.9C-27.2C)
for the second configuration from Table 1. The pair of such images was used to obtained velocity field
by PIV technique. Resulting 2D velocity field is showed in Fig. 2b. Additionally, the temperature was
monitored during whole experiment in points depicted in Fig. 1 (circles) by set of thermocouples. That
allowed to estimate heat transfer coefficients with required accuracy and adopt proper thermal
boundary conditions (T
H
:= T
12
, T
C
:= T
16
, T
ext
:= T
E1
= T
E2
).
EUROTHERM 82, September 13-16, 2005, Gliwice-Cracow, Poland


Velocity fields were calculated for series of 40 images taken in time intervals t = 200 ms or 500 ms.
Averaged fields were obtained on the basis of those fields by calculating mean values. Precision of
such averaged velocity field was estimated by calculating dispersion of mean velocity. It amounted 0.9
mm/s for images taken with camera of working resolution 520 x 511 pixels. To enhance precision of
our measurements, we performed velocity measurements with high speed CMOS camera (maximal
resolution 1280x1024 pixels) for the fragment of investigated cross-section in the bottom right corner,
depicted on Figure 1 with dashed line. It was possible to perform time series of 150 images with time
intervals t = 100 ms or 200 ms. That allowed to increase precision of measurements, which amounted
0.48 mm/s.

On the basis of series of such images statistical characteristics were calculated for each of
configurations from Table 1. In particular, mean
N
, standard deviations
N
, skewness S
N
and kurtosis
K
N
for both velocity components were analyzed. We also calculated turbulence intensity I defined as a
ratio of standard deviation of velocity magnitude to its mean
I =
N
/
N
(1)
Additionally, time series of velocity components in several points located near the right wall were
analyzed (P1 = [0.078,0.024], P2=[0.078,0.015], P3 =[0.078,0.008], P4=[0.062,0.008], P5=
[0.050,0.008]). Changes in distribution of local velocity components were studied with the increase of
Rayleigh number.
3 Results and Discussion
Figure 2a show example of RGB Image, which is characteristic for considered configurations.
Application of thermochromic liquid crystals allowed to visualize a 2D temperature field. There are
visible temperature boundary layers near active vertical walls, whereas in core region temperature
stratification is observed with almost constant temperature gradient. In Fig. 2b velocity field is
presented for the same configuration. Kinetic boundary layers were observed near vertical walls and
the stagnant core region in central part of the cavity.

Figure 2: Experimental results for the second configuration from Table 1.


a)RGB Image(T
H
27.2C i T
C
6.8C).Thermochromic
Liquid Crystal sensitive between 19.5C and 27.2C
b) 2D Velocity field (T
H
27.2C i T
C
6.7C). Resolution
of camera 920x911 pixels.
Tomasz Michaek


With the increase of Rayleigh number kinetic and temperature boundary layers became thicker, what
made them difficult for precise measurements. Therefore, we decided to limit our measurements to a
fragment of cross-section depicted in Fig. 1 with dashed line. That allowed to enhance precision of
measurements through lengthening numbers of images taken in series (up to 150) and enlargement of
spatial resolution of images (1280x1024 pixels).

Two dimensional velocity fields for the fragment of cross-section in the bottom right corner are
presented in Fig.3. Width of the kinetic boundary layer was assessed for four considered
configurations. It varied from 7 mm for Ra = 3x10
7
, through 5 mm for Ra = 1.5x10
8
and 4 mm for Ra
= 1.8x10
8
and eventually about 3 mm for Ra = 4.4x10
8
. These values were greater in comparison with
theoretical values based on dimensional analysis (~Pr
1/2
LRa
-1/4
). Difference is due to strong
assumptions made in derivation of formula for thickness of kinetic boundary layer, namely isothermal
and adiabatic boundary conditions.

Figure 3: Velocity fields in the bottom right corner of the central cross-section of cavity

a) Ra = 3.1*10
7
(Exp No 1) b) Ra = 1.5*10
8
(Exp. No 2)

c) Ra = 1.8*10
8
(Exp. No 3) d) Ra = 4.4*10
8
(Exp. No 4)
EUROTHERM 82, September 13-16, 2005, Gliwice-Cracow, Poland


Turbulence intensity calculated on the basis of statistical moments is showed in Fig.4. Its value for
relatively low Rayleigh numbers (Ra < 10
8
) are lower than 0.3 in the most of the flow region (Fig.4a).
Turbulence intensity increases with the increase of Rayleigh number and it is in the range between 0.3
and 0.5 for Ra = 1.5x10
8
and grater then 0.5 for higher values of Ra. For the highest Rayleigh numbers
(Fig.4 c-d) it is visible that turbulence intensity along the horizontal bottom wall is equal about unity,
what means that value of the standard deviation is of the same magnitude what the average velocity.
Moreover, we noticed the change in value of calculated kurtosis estimator for these configurations. It
was grater than 5 for Ra = 3x10
7
and in the range between 1 and 4 for higher Rayleigh numbers. That
suggests that there is no longer steady flow observed. In order to verify this observation time series in
several points were analyzed on the basis of series of 150 images taken in time delays t = 100 ms.

Figure 4: Turbulence intensity in the bottom right corner of the central cross-section of the cavity
a) Ra = 3.1*10
7
(Exp. No 1)

b) Ra = 1.5*10
8
(Exp. No 2)

c) Ra = 1.8*10
8
(Exp. No 3) d) Ra = 4.4*10
8
(Exp. No 4)
Tomasz Michaek


In Fig.5 time series of vertical velocity component are presented in point P2 = [0.0078;0.0015]. For all
of tested configurations characteristic spikes were noticed in time series of both velocity components.
For lower Rayleigh numbers they appeared with frequency of 1, 3 or even 5 sec. (Fig.5a), whereas for
higher Rayleigh numbers the frequency was almost constant (for example 0.5 sec. for Ra = 1.5x10
8
,
Fig.5b). These oscillations caused increase in values of the standard deviation.

Figure 5: Time series of vertical velocity component in point P2 = [0.0078;0.0015]


a) Ra =3x107(Exp. No 1) b) Ra =1.5x108(Exp. No 2)

Accurate experimental measurements allowed for application of appropriate boundary conditions into
the computational model. Two kind of numerical solutions were validated. First, in which we
calculated natural convective flow driven by temperature gradient in a square domain with vertical
isothermal walls and horizontal adiabatic walls. Second, in which not only flow in the square domain
was calculated but also heat transfer in horizontal Plexiglas walls and vertical aluminium walls (see
Fig.1). First kind of calculations were done with the use of SOLVSTR code described in detail in
Michalek [11], and second kind of calculations were done with commercial code Fluent. Correctness
of SOLVSTR code was verified by solving 2D problem of air flow in the differentially heated cavity
for high Rayleigh numbers and direct comparison with benchmark solution [13]. Additionally, the
performance and the accuracy of the code were evaluated making use of previously defined numerical
benchmark [14]. Calculations for four configurations listed in Table 4 were performed on uniform,
Cartesian grids (65x65, 129x129, 257x257 nodes) with the use of SOLVSTR code. Similarly,
calculations performed with the use of Fluent code were done on uniform grids with 40x40, 80x80,
160x160 nodes. That allows to estimate simulation uncertainty U
SN
by calculation numerical error as a
difference between extrapolated solution (Richardson extrapolation) and the one for which uncertainty
is evaluated. Estimated values of simulation uncertainty U
SN
are presented in Table 2.

We adopted validation procedure proposed by Stern at all. [1], which involves not only comparison
between experimental data and computational results but also takes into account its relations to so
called key validation metric. This metric involves estimation of experimental data uncertainty U
D
,
simulation uncertainty U
SN
and simulation uncertainty due to use of previous data U
SPD
. In case of
validation of velocity fields, experimental data uncertainty was estimated through calculation the
estimator of dispersion of the mean velocity calculated on the basis of series of images. Simulation
uncertainty U
SN
was calculated making use of Richardson extrapolation as mentioned above.
EUROTHERM 82, September 13-16, 2005, Gliwice-Cracow, Poland


Simulation uncertainty due to use of previous data was neglected because date at the input of our
calculation consist of material properties of water, which are well known.

Table 2: Validation procedure (experiment versus computational results).

Experiment Numerical simulation (Fluent) Numerical simulation (SOLVSTR)
Variable D U
D
S U
SN
E U
V
S U
SN
E U
V
T
12
18,67 0,38 18,92 0,02 0,25 0,38
T
16
4,05 0,38 3,83 0,02 0,22 0,38
Not available for this calculations because heat
transfer was not solved within the walls
T
7
18,22 0,48 18,39 0,02 0,17 0,48 17,99 0,07 0,23 0,49
T
10
17,76 0,63 17,64 0,02 0,12 0,63 17,17 0,07 0,59 0,63
T
15
20,17 0,47 19,45 0,02 0,72 0,47 Not available
U
min
-0,66 0,24 -0,73 0,01 0,07 0,24 -0,65 0,01 0,01 0,24
U
max
0,69 0,24 0,68 0,01 0,01 0,24 0,65 0,01 0,04 0,24
V
min
-2,6 0,24 -2,22 0,05 0,38 0,25 -2,40 0,09 0,20 0,26
V
max
2,42 0,24 2,22 0,05 0,20 0,25 2,40 0,09 0,02 0,26
V
P1
-2,48 0,58 -1,99 0,01 0,49 0,58 -1,99 0,04 0,49 0,58
V
P2
-1,85 0,42 -1,77 0,02 0,08 0,42 -1,71 0,04 0,14 0,42
U
P3
-0,24 0,09 -0,29 0,02 0,05 0,09 -0,22 0,01 0,02 0,09
V
P3
-0,75 0,21 -1,29 0,01 0,54 0,21 -1,05 0,02 0,30 0,21
U
P4
-0,58 0,14 -0,4 0,01 0,18 0,14 -0,39 0,01 0,19 0,14
U
P5
-0,6 0,16 -0,42 0,01 0,18 0,16 -0,42 0,02 0,18 0,16

a) Ra = 3x10
7

Experiment Numerical simulation (Fluent) Numerical simulation (SOLVSTR)
Variable D U
D
S U
SN
E U
V
S U
SN
E U
V
T
12
27,23 0,24 27,27 0,02 0,04 0,24
T
16
6,76 0,18 6,58 0,03 0,18 0,18
Not available for this calculations because heat
transfer was not solved within the walls
T
7
25,51 0,18 25,40 0,02 0,11 0,18 25,64 0,09 0,13 0,20
T
10
24,40 0,21 24,69 0,04 0,29 0,21 24,57 0,11 0,17 0,24
T
15
25,08 0,33 24,82 0,02 0,26 0,33 Not available
U
min
-1,12 0,76 -1,01 0,01 0,11 0,76 -1,23 0,04 0,11 0,76
U
max
0,97 0,76 1,01 0,01 0,04 0,76 1,23 0,04 0,26 0,76
V
min
-6,11 1,16 -3,65 0,05 2,46 1,16 -5,29 0,06 0,82 1,16
V
max
6,19 1,16 3,65 0,05 2,54 1,16 5,29 0,06 0,90 1,16
V
P1
-4,55 1,59 -2,39 0,01 2,16 1,59 -3,03 0,02 1,52 1,59
V
P2
-3,58 1,28 -2,19 0,02 1,39 1,28 -2,53 0,07 1,05 1,28
U
P3
-0,55 0,24 -0,36 0,02 0,19 0,24 -0,36 0,02 0,19 0,24
V
P3
-1,98 0,75 -1,68 0,01 0,30 0,75 -1,97 0,06 0,01 0,75
U
P4
-0,94 0,45 -0,48 0,01 0,46 0,45 -0,52 0,01 0,42 0,45
U
P5
-1,04 0,40 -0,49 0,01 0,55 0,40 -0,58 0,02 0,46 0,40

b) Ra = 1.5x10
8

Variables which were compared: T
12
, T
16
,T
7
,T
10
,T
15
point temperature measurements (see Figure 1),U
min
, U
max
,
V
min
, V
max
extreme values of horizontal and vertical velocity components in the central cross-section of the
cavity, U
P3
, U
P4
, U
P5
, V
P1
,V
P2
,V
P3
values of horizontal and vertical velocity in selected points in the right bottom
corner
D experimental value(mean value based on the series of velocity fields)
U
D
uncertainty of experimental value estimated by calculation the estimator of means dispersion
S computational value
U
SN
uncertainty of numerical simulation estimated on the basis of Richardson extrapolation
E comparison error, E=|D S|
U
V
validation metric, U
V
= SQRT(U
SN
2
+U
D
2
)
Values for which validation condition (E<U
V
)

does not hold were distinguished (yellow color).

Temperature values in points depicted in Fig.1, minimal and maximal values of velocity components
in central cross-section of cavity as well as velocity values in several points in the bottom right corner
Tomasz Michaek


were used for validation purposes (see Table 2). Validation error, defined as absolute value of
difference between experimental data and computational results, were less than value of validation
metric for most of compared values for both computational results. Examples of velocity field for two
of considered configuration are presented in Fig.6. But the value of validation metric which determines
the level of validation was high, especially in the case of velocity values. It was due to described
previously spikes, which increase the value of the standard deviation as well as dispersion of the mean
velocity. Condition for positive validation was not satisfied (validation error was grater than value of
validation metric, see Table 2) for horizontal velocity components in points located along the bottom
wall of the cavity.

Figure 6: Computed velocity fields colored by velocity magnitude (program SOLVSTR)



a) Ra =3x10
7
b) Ra =1.5x10
8


The major discrepancies between experimental data and computational results were lack of spikes and
unsteadiness in computation results and difference in thickness of kinetic boundary layer. The most
probable source of these discrepancies are due to non symmetric and non uniform thermal boundary
condition along horizontal Plexiglas walls. Such non-uniformity in temperature boundary conditions,
especially along the bottom wall, could cause development of Rayleigh Benard convection.
Observed fluctuations can be driven by hot spots arising near the bottom wall and are further stabilized
in thin boundary layer. Our further research is aimed to resolve observed discrepancies and involve
application of non symmetric thermal boundary conditions in Direct Numerical Simulations.
4 Conclusions
Application of the sensitivity analysis in order to identify basic parameters and estimate necessary
precision of measurements for validation purposes was presented. Experimental data concerning
natural convective flow in differentially heated cavity were shown for Rayleigh numbers up to 10
9
.
Analysis of velocity fields was done in the central cross-section of the cavity as well as for its
fragment near the bottom right corner. Statistic characteristics were estimated on the basis of images
recorded for four considered configurations. Transition from steady to unsteady flow regime was
observed for lower value of Rayleigh number than expected theoretical critical Rayleigh number. The
difference was attributed to non-uniformity and asymmetry of thermal boundary conditions assumed
in computational model. Further research will be concentrated on application of proper boundary
conditions in order to obtain similar spikes like those observed in experiments.
EUROTHERM 82, September 13-16, 2005, Gliwice-Cracow, Poland


Acknowledgements
This work was supported by Polish Scientific Committee (KBN Grant No. 4TO7A00726).
Author is grateful to Prof. T.A. Kowalewski for comments and suggestions concerning this paper.
References
[1] F. Stern, R. V. Wilson, H. W. Coleman, E. G. Paterson, Comprehensive Approach to Verification
and Validation of CFD Simulations Part 1: Methodology and Procedures, Journal of Fluids
Engineering, 123, (2001), pp. 793-802.
[2] P. J. Roache, Verification and Validation in Computational Science and Engineering, Hermosa
Publishers, Albuquerque, New Mexico, USA, (1998).
[3] W. L. Oberkampf, T. G. Trucano, Verification and validation in computational fluid dynamics,
Progress in Aerospace Science, 38, (2002), pp. 209-272.
[4] P. J. Roache, Code verification by the method of manufactured solutions, Journal of Fluids
Engineering, 124, (2002), pp. 4-10.
[5] M. Garbey, W. Shyy, A least square extrapolation method for improving solution accuracy of
PDE computations, Journal of Computational Physics, 186, (2003), pp. 1-23.
[6] M. M. Sindir, S. L. Barson, D. C. Chan, W. H. Lin, On the Development and Demonstration of a
Code Validation Process for Industrial Applications, AIAA Paper 96-2032, 27 AIAA Fluid
Dynamics Conference, New Orleans, Louisiana, (1996).
[7] D. P. Aeschliman, W. L. Oberkampf, Experimental Methodology for Computational Fluid
Dynamics Code Validation, SAND95-1189, Sandia National Laboratories, Albuquerque, New
Mexico, (1997).
[8] T. A. Kowalewski, Experimental validation of numerical codes in thermally driven flows, Adv. In
Computational Heat Transfer, Begel House Inc., New York, (1998), pp. 1- 15.
[9] T. A. Kowalewski, M. Rebow, Freezing of Water in a Differentially Heated Cubic Cavity,
International Journal on Computational Fluid Dynamics, 11, (1999), pp. 193-210.
[10] T. Michaek, T.A. Kowalewski, Numerical Benchmark Based on Natural Convection of Freezing
Water, Progress in Computational Heat and Mass Transfer, Proc. of 4
th
International Conference
on Computational Heat and Mass Transfer, Cachan, Paris, 1, (2005), pp. 7 11.
[11] T. Michaek, Metoda oceny wiarygodnoci symulacji numerycznych przepyww lepkich i
termicznych, PhD Thesis, in polish, IPPT PAN, Warszawa, (2005).
[12] T.A. Kowalewski, Particle image velocimetry and thermometry for two-phase flow problems,
Visualization and Imaging in Transport Phenomena, Annals of the New York Academy of Scs.,
972, (2002), pp.213-219.
[13] P. Le Quere, Accurate solutions to the square thermally driven cavity at high Rayleigh number,
Computer and Fluids, 20(1), (1991), pp. 29-41.
[14] T. Michaek, T.A. Kowalewski, B. Sarler, Natural Convection for Anomalous Density Variation
of Water: Numerical Benchmark, Progress in Computational Fluid Dynamics, 5 (3-5), (2005), pp.
158-170.

Vous aimerez peut-être aussi