Professional Documents
Culture Documents
net/publication/330541852
CITATIONS READS
0 316
1 author:
George-Madalin Chitaru
Technical University of Civil Engineering of Bucharest
8 PUBLICATIONS 14 CITATIONS
SEE PROFILE
All content following this page was uploaded by George-Madalin Chitaru on 22 January 2019.
MASTER’S THESIS
Project coordinator,
Conf. Dr. Ing. Costin Ioan Coșoiu
Author,
Ing. CHITARU George-Mădălin
Bucharest
2018
MINISTERUL EDUCAȚIEI NAȚIONALE ȘI CERCETĂRII ȘTIINȚIFICE
UNIVERSITATEA TEHNICĂ DE CONSTRUCȚII BUCUREŞTI
FACULTATEA DE INGINERIE A INSTALAȚIILOR
Master Efficacité énergétique des installations techniques du bâtiment
Project coordinator,
Conf. Dr. Ing. Costin Ioan Coșoiu
Author,
Ing. CHITARU George-Mădălin
Bucharest
2018
DECLARAȚIE DE ONESTITATE
EN: I hereby declare that the master’s thesis "Numerical study of a harsh
environment testing facility" is written by me and has never been presented to another
faculty or institution of higher education in the country or abroad.
Student,
Ing. CHITARU George-Mădălin
CONTENT
Table 4.1: Results from battery pressure drop simulations for the 3D model with Refined
Mesh 1 grid .................................................................................................................... 33
Table 5.1: Side-by-side comparison of the 5 grids ......................................................... 46
Table 6.1 Mass flow boundary conditions for each injection method .............................. 60
LIST OF SYMBOLS
FOREWORD
This thesis has been the result of a 2-year collaboration with the Department of
Hydraulics and Environmental Protection from the Technical University of Civil
Engineering of Bucharest and it continues the work presented in Georgescu (2017) and
Georgescu et al (2017).
The theme of this paper was proposed as a result of a research partnership between
the Technical University of Civil Engineering and the National Institute for Aerospace
Research "Elie Carafoli". The research grant under which the collaboration between the
two entities was concluded was registered under number C244 / 2016. The objective of
this collaboration was to conduct studies to characterize and improve the performance of
one of the newest INCAS facilities, namely the harsh environment testing facility used
to conduct experiments regarding the exposure of aviation equipment to high
concentrations of dust or sand, combined with air temperature levels specific to the desert
areas. This harsh environment facility was built primarily to test new equipment developed
under the AFLoNext project, which is supported through the European Framework
Program for Research and Technological Development (FP7) and which aims to improve
airplane performance by reducing fuel consumptions and noise levels during take-off and
landing procedures. However, after a thorough verification, the same facility could be
used to test and certify different types of equipment.
From the Technical University of Civil Engineering of Bucharest, the research team
responsible for the C244/2016 contract was composed of 3 master’s students (Ing.
George-Mădălin Chitaru, Ing. Matei-Răzvan Georgescu and Ing. Gelu Muscă),
coordinated by Conf. Dr. Ing. Costin Ioan Coșoiu.
The experimental data was collected at INCAS, and the research was carried out in
the Department of Hydraulics and Environmental Protection at the Aerodynamics and
Wind Engineering Laboratory "C. Iamandi ".
Finally, I would like to thank the National Institute for Aerospace Research for giving
us the opportunity to work on this project. I would also like to thank all those in the
Hydraulics and Environmental Protection Department for having provided us with all the
necessary equipment and simulation software, as well as technical support and advice
offered.
ABSTRACT
The combined effects of dust, sand and volcanic ash is of high interest for the aircraft
industry due to a plethora of technical problems that these can cause for airplane
systems, from the obstruction of openings or penetration into cracks, crevices, bearings
and joints, to erosion and corrosion of different airplane components. Due to the many
potential problems that can arise, aviation authorities have imposed standard testing
procedures in order to ensure that any airborne equipment would be tested to various
environmental conditions, one of these conditions being the exposure to blowing dust and
sand at moderate speeds. To assure the safety of the airships, airline companies are
continuously trying to reduce the effects of particle ingestion and component deterioration
for key aircraft equipment that is usually exposed to dust & sand while flying. The
preferred choice for making these kinds of tests are special harsh environment testing
facilities (special atmospheric boundary layer wind-tunnels) that can recreate the same
exterior conditions as the dust heavy environments. One of these testing facilities can be
found at the National Institute for Aerospace Research “Elie Carafoli” in Bucharest
(INCAS) where new airplane equipment for the AFLoNext European project is being
tested.
The main objective of our work is to study the multiphase flow of dust with air and
sand with air inside this harsh environment testing facility and to determine if there are
ways of improving the current set-up. Due to the advantages that virtual modeling is
offering nowadays, the study has been done mostly numerically, using a commercial
computational fluid dynamics (CFD) simulation software in order to analyze the particle
behavior during the standard experimental procedure. A series of CFD simulations have
been conducted to determine the distribution of the granular secondary phase within the
test section of the tunnel.
In order to compute the multiphase flow, the Eulerian multiphase model, along with
the realizable k-ε viscous model and with the dispersed multiphase turbulence model
have been chosen to run the steady state simulations. Lacking an experimental validation,
a grid independence study has been realized to verify the robustness of the simulation
model. Results from various injection methods have been collected and compared with
regards to particle cross section distribution, actual particle velocity and also particle
trajectory for the case when gravity is accounted for.
The findings have shown that improvements can be made to the reference injection
system in order to obtain a more even distribution for the granular phase, therefore
offering a better reproduction of the real-world phenomena. Moreover, the numerical
model could be used to approximate the correct positioning of the equipment inside the
test section, as well as determining the required dust mass flow at the injection points so
that the concentration of the particulate matter is kept within the limits imposed by the
standard test procedure.
This work provides a better understanding of a particle-laden flow inside a wind-
tunnel and also suggestions of possible modifications to a real-life test facility which can
improve the accuracy of the experiments carried therein.
CHAPTER 1 - INTRODUCTION
Airborne particles represent a potential problem for the aeronautical industry that is
usually associated with the flight of aircrafts in dust and sand loaded areas like the deserts
or areas which suffer droughts for longer periods of time [Lekas et al (2011)]. Due to the
atmospheric circulation, the particles can get airborne by the effects of saltation and
suspension and form dust or sand storms. In situations where there is significant
atmospheric instability the particulate matter can reach as high as 6000 m and it can
travel to farther regions where there are usually no such phenomena [Ghosh (2014);
Middleton and Kang (2017)]. Still, most of the time the aircrafts are affected by sand and
dust storm activity within the atmospheric boundary layer where higher concentrations of
particulate matter are present. These particles could represent a hazard for airplanes after
repeated take-off and landing procedures and they can represent an even more serious
problem for aircrafts that fly only at low-altitude regime, like the helicopters in dust loaded
areas. [Bojdo and Filippone (2014)]
Furthermore, the frequency of sand and dust storms (SDS), has increased
dramatically over the last decades, especially in arid or semi-arid zones that have been
affected by desertification, land degradation and deforestation. It has therefore become
more common to encounter an SDS on flight paths that pass through these regions.
[Ganor (2010); Sissakian et al (2013); Notaro et al (2015); Krasnov et al (2016)] Moreover,
considering that air traffic in these particle laden environments has been continuously
increasing and that the desertification and pollution have caused an overall augmentation
in atmospheric particle concentration, the aircrafts are becoming more and more exposed
to the effects of particle ingestion and deposition.
Due to the increased global awareness of this phenomenon [UNEP (2016);
International Conference on Combating Sand and Dust Storms, Tehran (2017)],
governments and agencies like the European Aviation Safety Agency (EASA) have taken
measures to assure that any airborne equipment would be tested to various
environmental conditions, one of these conditions being the exposure to blowing dust and
sand at moderate speeds. At at the moment of writing this paper, there are 3 standard
test procedures that are specified in the EASA “Certification Specifications and
Acceptable Means of Compliance for Engines (CS-E)” [EASA (2007)]:
1) EUROCAE ED-14G
2) RTCA/DO-160, Section 12, Category D
3) MIL-STD-810
Besides airplane rerouting, flight cancellation and airport operation difficulties, more
important problems that are associated with dust & sand loaded areas are tied to the
mechanical impact of the particles upon the aircraft’s exposed components, like: erosion
and corrosion of turbine blades or other vulnerable components, failure of systems that
rely on the pitot-static probes, failure of conditioning packs due to blockage or, in severe
cases, even engine flame out in flight. [Reagle (2012); Khalifa (2016)] In order to avoid
these problems, adequate modifications should be made to improve the design of the
engine or other exposed components based on the particle mass ingestion tolerance and
resistance. Companies which are involved into such research & development projects
need to test these parts in a safe and controlled environment. The preferred choice for
making these kinds of tests are special wind tunnels that can recreate similar conditions
as the sand & dust laden environments.
Figure 2.2: Photo of the inlet section, where the fan and the heating battery are placed
After the battery, a honeycomb has been introduced to straighten the flow and just
after that there is a convergent-divergent section where the injection mechanism is
installed and where the sand and dust tests are taking place. As it can be seen in Figure
2.1, the section of the tunnel gradually reduces from 700mm diameter to 400mm diameter
through a constriction element and right after the contraction there is the injection
mechanism which is the highlight of our CFD simulations. A part of the injection
mechanism can be seen in Figure 2.3 below. The tube presented in the photo is
connected to a dust recipient and to an air compressor. The high-pressure air enters the
Figure 2.4: Photo with the uncovered test section and the visible support for the actuator
EXPERIMENTAL DATA GATHERING
Before starting the first numerical simulation stage, it was necessary to collect an
experimental dataset to determine the operating parameters of the installation and how
these vary. Therefore, a series of experimental tests was carried out using a single-phase
flow, by introducing only air through the inlet section of the aerodynamic tunnel without
injecting any dust or sand particles through the dedicated mechanism. What we aimed to
The first step in any CFD project is to create or procure the geometry of the domain
that needs to be studied. In this case, it was necessary to create the virtual geometry from
scratch because there was no CAD model of the facility already available. The geometry
has been created in one of the ANSYS products, ICEM CFD. As seen in Figure 3.1, the
3D virtual model used for the CFD simulations does not contain the entire harsh
environment testing facility, as we were interested to study only what happens in the test
section and what is the influence of the upstream components. Omitting the cyclonic
separators and the last few meters of the tunnel has offered us a great advantage
because we spared precious time not modeling their complex geometry and it significantly
reduced the time required to run the simulations due to a much lower number of cells.
𝑉 = cell volume;
𝑚̇𝑝𝑞 , 𝑚̇𝑞𝑝 = mass transfer between phase q and phase p;
∆𝑡 = time scale.
In the above equation the qth phase is the dust and the pth phase is the air. It is worth
mentioning that, with default settings, FLUENT only calculates the secondary phase
volume fraction, estimating the volume fraction for the primary phase from the condition
expressed in (3).
∑𝑛𝑞=1 𝛼𝑞 = 1 (2)
This means that, in our case, only the volume fraction for the dust will be computed
using (1) and the volume fraction for the air will be considered equal to the difference:
𝛼𝑎𝑖𝑟 = 1 − 𝛼𝑑𝑢𝑠𝑡 (3)
Another thing worth mentioning is that, for the above equation (1), the source term
𝑆𝛼𝑞 is zero.
𝜕
(𝛼 𝜌 𝜈⃗ ) + 𝛻(𝛼𝑠 𝜌𝑠 𝜈⃗𝑠 𝜈⃗𝑠 ) = −𝛼𝑠 𝛻𝑝 − 𝛻𝑝𝑠 + 𝛻𝜏̿𝑠 + 𝛼𝑠 𝜌𝑠 𝑔⃗ +
𝜕𝑡 𝑠 𝑠 𝑠
∑𝑛𝑙=1(𝐾𝑙𝑠 (𝑣
⃗⃗⃗⃗𝑙 − ⃗⃗⃗⃗)
𝑣𝑠 + 𝑚̇𝑙𝑠 ⃗⃗⃗⃗⃗⃗ 𝑣𝑠𝑙 + (𝐹⃗𝑠 + 𝐹⃗𝑙𝑖𝑓𝑡,𝑠 + 𝐹⃗𝑤𝑙,s + 𝐹⃗𝑣𝑚,𝑠 + 𝐹⃗𝑡𝑑,𝑠 )
𝑣𝑙𝑠 − 𝑚̇𝑠𝑙 ⃗⃗⃗⃗⃗⃗) (5)
Where:
𝑔⃗ = the gravitational force;
𝛼𝑞 , 𝜌𝑞 , 𝛼𝑠 , 𝜌𝑠 = the volume fraction and the density for the qth and sth phase;
𝑣𝑞 ⃗⃗⃗⃗
⃗⃗⃗⃗⃗, 𝑣𝑠 = velocity of the qth and sth phase;
∇𝑝 = the pressure gradient shared by all phases;
𝑝𝑠 = the solids pressure for the sth phase;
𝐹⃗𝑞 , 𝐹⃗𝑠 = an external body force acting on the qth and sth phase;
𝐹⃗𝑙𝑖𝑓𝑡,𝑞 , 𝐹⃗𝑙𝑖𝑓𝑡,𝑠 = the lift force acting on the qth and sth phase;
𝐹⃗𝑤𝑙,𝑞 , 𝐹⃗𝑤𝑙,𝑠 = the wall lubrication force acting on the qth and sth phase;
𝐹⃗𝑣𝑚,q , 𝐹⃗𝑣𝑚,𝑠 = the virtual mass force acting on the qth and sth phase;
𝐹⃗𝑡𝑑,𝑞 , 𝐹⃗𝑡𝑑,𝑠 = the turbulent dispersion force acting on the qth and sth phase;
𝐾𝑝𝑞 , 𝐾𝑙𝑠 = interphase momentum exchange coefficient;
4.1.3 CONSERVATION OF ENERGY
To assure the energy conservation for multiphase applications, once again two
separate equations have to be calculated, but they both have the same formulation
[ANSYS Fluent Theory Guide Chapter 17]:
𝜕 𝑑𝑝𝑞
(𝛼𝑞 𝜌𝑞 ℎ𝑞 ) + ∇(𝛼𝑞 𝜌𝑞 𝑢
⃗⃗𝑞 ℎ𝑞 ) = 𝛼𝑞 + 𝜏̿𝑞 : ∇𝑢
⃗⃗𝑞 − ∇q
⃗⃗ q +
𝜕𝑡 𝑑𝑡
𝑆𝑞 + ∑𝑛𝑝=1(𝑄𝑝𝑞 + 𝑚̇𝑝𝑞 ℎ𝑝𝑞 − 𝑚̇𝑞𝑝 ℎ𝑞𝑝 ) (6)
Where:
𝛼𝑞 , 𝜌𝑞 , 𝜏𝑞 = have been previously mentioned in equations (1), (4) and (5);
ℎ𝑞 = the specific enthalpy of the qth phase;
𝑞⃗𝑞 = the heat flux;
𝑆𝑞 = the user defined source term;
𝑄𝑝𝑞 = the intensity of heat exchange between the pth and qth;
ℎ𝑝𝑞 , ℎ𝑞𝑝 = the interphase enthalpy;
4.1.4 VISCOUS AND TURBULENCE MODEL
Turbulence modelling has been one of the more difficult tasks for CFD applications.
Even though there are numerous models that have been proposed, none of them are
perfect for every situation. This is why a thorough documentation is necessary in order to
make the right choice regarding the turbulence model.
In this case, we have opted to use a Reynolds Averaged Navier-Stokes (RANS)
Two Equation Model because this has shown the perfect ratio of accuracy to complexity,
meaning that the results present a satisfying accuracy with lower resource demands (time
and hardware) [Ratnam and Vengadesan (2007); Fröhlich and von Terzi (2008)]. There
are two main categories of RANS Two Equation Models: k-ε and k-ω. Having compared
the limitations of both categories we have chosen to use the k-ε model which solves two
additional equations, one for the turbulent kinetic energy k and the other for kinetic energy
dissipation rate ε. The limitations of this group of models were mainly observed for
simulations of flows with severe pressure gradients, separated flows, recirculation caused
by adverse pressure gradients, or strong streamline curvatures [Wasserman (2016)].
Facultatea de Inginerie a Instalațiilor - UTCB 23
Ing. CHITARU George-Mădălin Master’s Thesis
Master Efficacité énergétique des installations techniques du bâtiment
Since for the studied case these physical phenomena are not present, or at least not in
the sections that we are interested in, the choice of the k-ε family seemed more in line
with our research. The advantages that it has given us (less computational resources,
robustness in calculation of fully turbulent flows, ease of implementation, better
convergence) far outweigh the disadvantages. All these aspects also make it better for
parametric studies with lots of scenarios like it was the case for this thesis.
After a proper analysis of the available options for the viscous model, the Realizable
k-ε set of equations has been chosen due to its advantages compared to the older
Standard k-ε approach. The Realizable model is an improved version of the Standard
model and has been proven to be more robust when solving rotating flows, recirculation,
jet flows, mixing layers and separated flows because of a new approach for calculating
the turbulence dissipation rate ε [Shih et al. (1995)]. For the turbulent kinetic energy k, the
Realizable model uses a similar method as the older Standard k - ε model. The transport
equations for the turbulent kinetic energy k and for the turbulence dissipation rate ε are
presented in equations (7) and (8).
𝜕 𝜕 𝜕 𝜇 𝜕𝑘
(𝜌𝑘) + (𝜌𝑘𝑢𝑗 ) = 𝜕𝑥 [ (𝜇 + 𝜎 𝑡 ) (𝜕𝑥 )] + 𝐺𝑘 + 𝐺𝑏 − 𝜌𝜀 − 𝑌𝑀 + 𝑆𝑘 (7)
𝜕𝑡 𝜕𝑥𝑗 𝑗 𝑘 𝑗
𝜕 𝜕 𝜕 𝜇 𝜕𝜀 𝜀2 𝜀
(𝜌𝜀) + (𝜌𝜀𝑢𝑗 ) = 𝜕𝑥 [ (𝜇 + 𝜎𝑡 ) (𝜕𝑥 )] + ρC1 𝑆𝜀 − 𝜌𝐶2 𝑘+ + 𝐶1𝜀 𝑘 𝐶3𝜀 𝐺𝑏 + 𝑆𝜀 (8)
𝜕𝑡 𝜕𝑥𝑗 𝑗 𝜀 𝑗 √𝜈∗𝜀
Where:
𝜂 𝑘 𝑘2
𝐶1 = max [0.43, 𝜂+5] ; 𝜂 = 𝑆 𝜀 ; 𝑆 = √2𝑆𝑖𝑗 𝑆𝑖𝑗 ; 𝜇𝑡 = 𝜌𝐶𝜇 𝜀
Figure 4.2: Grid representation of the near-wall boundary region in the test section [a minimum of 12
cells]
Note: The y+ value after the divergent cone are lower due to a flow separation.
In our case, we have considered the porous zone as a homogenous media since
the flow was mainly affected on the X direction, with negligible impact on the Y and Z
directions. Therefore, the only unknown values from equation (10) remain the viscous
resistance coefficient (or permeability) and the inertial resistance coefficient. These two
terms, needed to define the porosity, have been determined from the experimental data
collected on-site at the testing facility, as described in Georgescu et al (2017).
A regression curve describing the pressure drop variation with the airflow velocity
has been determined (11) from the aforementioned experimental data using the pressure
Figure 4.5: The pressure drop variation determined from experimental data points
Figure 4.6: The pressure drop variation per unit meter determined from experimental data
[Georgescu et al (2017)]
In order to determine the viscous and the inertial resistance factors, we have taken
equation (10) and we have divided it to the length on the X direction in order to have the
same form as equation (11). Putting both equations (10) and (11) side-by-side, we have
determined the coefficients needed to define the porosity within Fluent:
Where:
𝑆𝑖 𝑛 = the momentum sink per unit meter [Pa/m];
∆𝑛 = the length of the porous zone (the battery) [m]
∆𝑝 = the total pressure drop on the battery section [Pa]
∆𝑝𝑛 = the pressure drop per unit meter [Pa/m]
𝑣 = velocity of the fluid passing through the porous zone (battery) [m/s]
Knowing that ∆𝑛 = 0.8 𝑚, we determined the inertial resistance factor (13) and
viscous resistance (14) following the methodology used by ANSYS Fluent:
1 6×10−14 1
= = 3.35 × 10−9 [𝑚2 ] (13)
𝛼 𝜇
𝜌 −1 1
𝐶2 = 4.21 ( 2) = 6.8735 [𝑚] (14)
In order to verify the accuracy of the porous media impact on the flow, we have
conducted a series of simulations in which the velocity at the inlet was varied from 0.5
m/s to 7 m/s and the pressure drop on the battery section was registered for each case.
We have then compared the results obtained from the simulations with the curves
determined by the experimental data and we have considered that the accuracy is
sufficient enough if relative errors were under 1%.
Figure 4.7: The pressure drop variation per unit meter [CFD Results vs Analytic Data]
As it can be seen in Figure 4.7 above, the results from the new 3D model are in line
with the results from the axisymmetric model described in Georgescu et al (2017). The
red circles represent the points calculated with the analytic formula from (8) and the
continuous line represents the CFD results. The largest relative error of 0,95% (Table 4.1)
was recorded for the 0,5 m/s injection velocity. This is of no concern, especially
considering that the pressure-drop values for velocities between 0 and 1 m/s are very
With this initial value we run a simulation for a limited number of iterations and then
we verify if the actual concentration of dust given by the simulation 𝐶𝑠𝑖𝑚 is close to what
we were aiming for (𝐶𝑟𝑒𝑞 = 5 𝑔/𝑚3 ). Most probably the value will not be exactly the one
that we search for, but we can increase or decrease the mass flow depending on the ratio
between the two concentrations and rerun the simulations. This way it should not take
more than 2 steps to get close to the value of 5 g/m3. Of course, the concentration might
not be exactly the same for each injection method because the forces acting on the
particles have a different impact level, and they can either slow down or accelerate the
flow of dust just as explained in [Chapter 6.3]. However, from the results presented in
Table 6.1 we can see that the differences are not that significant and can be neglected
considering the large interval in which we work in (3.5 g/m3– 8.8 g/m3). In the end, after a
series of simulations, we have decided that a total dust mass flow rate of 1.5 g/s was a
good all-around value since it provided a total dust concentration within the standard
procedure limits no matter the injection method that was used.
Another advantage of the “mass flow” type is that, if we change the inlet velocity
from 0.7 m/s to another value this means that we can recalculate the approximate
required mass flow just with the ratio between the new velocity and the old velocity in the
test section. For example: for a velocity of 1.4 m/s the airflow in the small section tube will
be twice as fast, therefore we can assume that we will need a dust mass flow rate twice
as high to maintain approximately the same concentration within the test section. In the
end the dispersion of the particles might be very different, but the total concentration
should have close-enough values.
A simplification that we considered for this stage was to specify the mass flow only
for the dust phase since we were not really interested to control the mass flow of air that
comes through the injection point. The fraction of air that is mixed with the dust can be
computed from the volume fraction specification like in the example below:
𝑔
Considering that a dust injection mass flow of 𝑚̇𝑑𝑢𝑠𝑡 = 1.5 , and a dust volume
𝑠
3
𝑚𝑑𝑢𝑠𝑡
fraction 𝛼𝑑𝑢𝑠𝑡 = 0.02 [𝑚3 ] have already been determined, we can calculate the portion
𝑡𝑜𝑡𝑎𝑙
of air that is injected 𝑚̇𝑎𝑖𝑟 with:
(1−𝛼 ) 𝑚 1−0.02 3 𝑘𝑔 𝑔
𝑚̇𝑎𝑖𝑟 = 𝑉̇𝑑𝑢𝑠𝑡 × 𝛼 𝑑𝑢𝑠𝑡 × 𝜌𝑎𝑖𝑟 = 5.66 × 10−7 [ 𝑠 ] × 0.02 [−] × 1.2 [𝑚3 ] = 0.033 𝑠 (15)
𝑑𝑢𝑠𝑡
Where:
𝑘𝑔 𝑘𝑔 0.0015 𝑚 3
𝑚̇𝑑𝑢𝑠𝑡 = 0.0015 ; 𝜌𝑑𝑢𝑠𝑡 = 2650 𝑚3 ; 𝑉̇𝑑𝑢𝑠𝑡 = 2650 = 5.66 × 10−7 [ 𝑠 ]
𝑠
Figure 4.9: Dust volume fraction representation with Phase Coupled SIMPLE scheme
Figure 4.11: Interfacial shape representations for 1st Order Upwind and Modified HRIC
[Walters and Wolgemuth (2009)]
As it can be noticed in Figure 4.12 below, due to the overly diffusive nature of the
1st Order Upwind scheme, the dust is actually more evenly dispersed across the
transversal section of the tunnel. This is caused by severe numerical diffusion determined
Figure 4.12: Contours of Volume Fraction on the longitudinal middle-plane for First Order Upwind and
Modified HRIC discretization schemes
On the other hand, the Modified HRIC scheme does not offer us the same good
results in terms of dust dispersion, but it actually describes better the motion of the
particles. In Figure 4.13 a close-up of the injection area is presented and here it can be
observed how the two discretization schemes differ one from another. The Modified HRIC
scheme uses a nonlinear blend of upwind and downwind differencing which takes into
account the presence of the velocity gradients and provides improved accuracy in
approximating the mass transfer between two adjacent cells. As a result, in Figure 4.13b,
we can see that the particles do not spread as much as for the Upwind case and that the
motion of the dust near the injection area affects the entire trajectory of the particles,
which tend to accumulate near the walls (Figure 4.12).
Figure 4.14: Results comparison between First Order Upwind and Modified HRIC
Left: Area Weighted Dust Uniformity Index Right: Mass Averaged Dust Velocity
In the end, for the Volume Fraction spatial discretization we used:
• Iterations 0 - 400: First Order Upwind
• Iterations 400 - 1000: First Order Upwind
• Iterations 1000 - 1500: Modified HRIC
While for the rest of the flow variables (Momentum, Turbulent Kinetic Energy,
Turbulent Dissipation Rate, and Energy) the spatial discretization has been done as
follows:
• Iterations 0 - 400: First Order Upwind
• Iterations 400 - 1000: Second Order Upwind
• Iterations 1000 - 1500: Second Order Upwind
A convergence criterion of 10-5 was set for all equations, but the model has rarely
converged to that level. Due to the small volume fractions of dust, there have always been
difficulties with secondary phase convergence, and also with secondary phase X velocity.
However, convergence is not the only indicator for the accuracy of the results. In our case
it was expected not to reach very low convergence levels due to the complexity of the
physics that are being computed (flow separation after divergent cone, particle dispersion,
particle collision, turbulence interactions, interphase exchanges etc.) therefore additional
factors were taken into consideration to evaluate the accuracy of the results.
One important aspect was to check the total mass imbalance of dust and air. An
acceptable error of 1% was considered as the limit for this factor. A second aspect that
we monitored was the fluctuation of key physical quantities or “quantities of interest”, how
they are called in a more CFD language. The physical quantities that were interesting for
us in this case were: the mass-averaged velocity in the volume of the test section, the
area-weighted uniformity index at the entrance in the test section, the volume-
averaged dust volume fraction and the maximum dust volume fraction on the test
section volume, the total dust mass in the test section volume and, of course, the mass
imbalance of dust between the inlets and the outlet.
- Outlet: The outlet has been defined as outflow with no additional changes.
- Solution set-up: The Coupled Pressure-Velocity scheme was employed, together
with the Pseudo Transient method. All simulations have respected the same spatial
discretization scheme mentioned in [Chapter 4.3.2]:
• Iterations 0-400: First Order Upwind for all variables;
• Iterations 400-1000: Second Order Upwind for Momentum, Turbulent
Kinetic Energy and Turbulent Dissipation Rate with First Order Upwind
for the Volume Fraction;
• Iterations 1000-1500: Second Order Upwind for Momentum, Turbulent
Kinetic Energy and Turbulent Dissipation Rate with Modified HRIC for the
Volume Fraction.
The 5 grids have been compared on the one hand with regards to equiangular
skewness, aspect ratio and ortho quality, and on the other hand with regards to
convergence level, dust velocity, particle distribution in the test section and mass
imbalance. All physical quantities specified in [Chapter 4.3.2] were monitored and
compared after each simulation.
RESULTS AND DISCUSSION
5.3.1 GRID PARAMETERS AND QUALITY RESULTS FROM ICEM CFD
First of all, a side-by-side comparison of the 5 grids quality metrics has been
conducted, in order to have a better idea about the overall quality of each mesh. These
metrics have been extracted from ICEM CFD which is the software used to create the
geometry and the grids. From the list of available quality metrics, the three most important
have been selected for the comparison: the orthogonal quality, the equiangular skewness
and the aspect ratio. These are also the metrics that ANSYS Fluent uses to report the
quality of the grid, although in Fluent the reported values are the ones corresponding to
the minimum quality and maximum aspect ratio. Because Fluent is ignoring the overall
mesh quality, we have shown in the following paragraphs that properly judging the quality
of a mesh means more than just checking the worst values in the domain. In addition to
these quality indicators, the average simulation time per iteration has also been recorded,
in order to see how each refinement affects the total simulation time.
From Table 5.1 it can be noticed that the first 4 grids present a rather low minimum
orthogonal quality and equiangular quality, but the mean values do reflect an overall high-
quality mesh. It can also be observed that the mesh refinement can have either a positive
or a negative impact on the minimum quality values, with very small improvements to the
mean quality results. As mentioned before, these low minimum values are generated by
a very small number of highly skewed elements (from 0.005% to 0.01% depending on the
mesh) that are present in the center of the grid, therefore their impact on the overall quality
should be negligible. This can also be observed in the histogram representation for the
Facultatea de Inginerie a Instalațiilor - UTCB 45
Ing. CHITARU George-Mădălin Master’s Thesis
Master Efficacité énergétique des installations techniques du bâtiment
equiangular quality and for the orthogonal quality distribution across the domain which
are presented in Figure 5.4 and Figure 5.5 below.
Table 5.1: Side-by-side comparison of the 5 grids
ICEM CFD Grid Quality Parameters
Mesh Type
Initial Refined Refined Refined Refined
Criteria Mesh Mesh 1 Mesh 2 Mesh 3 Mesh 1 HQ
No. of elements 2,264,744 4,180,032 5,065,620 6,944,708 4,180,672
Orthogonal Min 0.137 0.149 0.149 0.118 0.675
Quality Mean 0.929 0.948 0.949 0.955 0.948
Equiangular Min 0.134 0.111 0.111 0.083 0.480
Quality Mean 0.913 0.923 0.923 0.925 0.918
Aspect Max 66.23 86.94 81.67 110.86 63.93
Ratio Mean 8.24 8.59 7.21 7.48 8.83
Average time per iteration 26 s 36 s 60 s 155 s 36 s
The first histogram (Figure 5.4) shows a rather similar quality distribution between
the 5 grids, with somewhat different values for the initial mesh. For all grids, 99.9% of the
elements present a quality higher than 0.5, which is considered to be a very acceptable
value. This is also why there are no bars visible after the 0.5 mark, because the number
of elements was too small to be represented in the chart.
In the second histogram (Figure 5.5) it can be seen that the quality distribution is
very different between 0.8 and 1 but considering the mean values from Table 5.1 these
differences do not affect the overall quality of the grids. The result is even better than the
equiangular quality histogram, since there are almost no bars visible after the 0.7 mark.
That is because about 99,9% of the elements show an orthogonal quality higher than
0.65, the only exception being the initial mesh which has only 99.7% of the elements.
It is worth mentioning that a “Global mesh smoothing” process, that aims to improve
the overall quality of the mesh, was used for each of these grids. However, a certain
number of elements couldn’t be modified in any way. Therefore, these were the best
results that we could obtain in terms of mesh quality.
50%
Percentage of cells
40%
30%
20%
10%
0%
0.95 0.90 0.85 0.80 0.75 0.70 0.65 0.60 0.55 0.50 0.45 0.40 0.35 0.30 0.25 0.20 0.15 0.10 0.05 0.00
Equiangular Quality
Histogram of Quality
Initial Mesh Refined Mesh 1 Refined Mesh 2 Refined Mesh 3 Refined Mesh 1 HQ
70%
60%
Percentage of cells [%]
50%
40%
30%
20%
10%
0%
0.95 0.90 0.85 0.80 0.75 0.70 0.65 0.60 0.55 0.50 0.45 0.40 0.35 0.30 0.25 0.20 0.15 0.10 0.05 0.00
Orthogonal Quality[-]
Figure 5.5: Histogram of orthogonal quality across the domain
In addition to the orthogonal quality and equiangular quality, the aspect ratio is also
a determining factor in the evaluation of a mesh. As it can be noticed in Table 5.1, the
maximum aspect ratio increases with each transversal refinement and decreased with
longitudinal refinement. However, we can observe that the mean aspect ratio actually
decreases for the second transversal refinement.
Aspect ratio refers to the difference between the minimum and the maximum
dimensions for an element. In a simplified way, this means that, at a certain point, we
have a cell that is 66 to 110 times longer than its width. A more detailed explanation about
how the aspect ratio for a hexahedral cell is calculated can be found in [ANSYS ICEM
CFD Help Manual]. A maximum value of 40 is usually recommended for the aspect ratio.
Even though this value is not respected for all elements, after analyzing the results, the
percentage of cells that have an aspect ratio higher than 40 is between 0.4% and 1.2%.
Histogram of Aspect Ratio
Initial Mesh Refined Mesh 1 Refined Mesh 2 Refined Mesh 3 Refined Mesh 1 HQ
50%
45%
40%
Percentage of cells [%]
35%
30%
25%
20%
15%
10%
5%
0%
66 63 60 56 53 50 46 43 40 36 33 30 26 23 20 17 13 10 7 3
Aspect Ratio [-]
Figure 5.6: Histogram of aspect ratio across the domain
A detailed overview of the postprocessed data used for the table and the histograms
above can be found in [Annex 1].
Figure 5.9: Area Weighted Uniformity Index for the dust volume fraction at the test section entrance
Figure 5.10: Mass Averaged Dust Velocity in the test section volume
Like in the velocity graph, a similar tendency can also be noticed for the mass
averaged volume fraction at the vein entrance (Figure 5.11). All 5 grids present the same
solution behavior with too small variations to be taken into consideration.
Figure 5.11: Area Weighted Average of dust volume fraction at test section entrance
However, when analyzing the maximum dust volume fraction at the test section
entrance, we can notice small differences between the grids, but these differences are
Facultatea de Inginerie a Instalațiilor - UTCB 50
Ing. CHITARU George-Mădălin Master’s Thesis
Master Efficacité énergétique des installations techniques du bâtiment
negligible compared to the targeted volume fractions of 3.32e-06 and 1.32e-06 which are
over 50 times lower than these maximum values. Moreover, the differences can come
from the variation in the cells dimensions between the meshes, and not from actual
numerical difference. Since the cells become smaller with each refinement it is normal to
have higher maximum dust concentrations in the smaller cells because more particles
are packed together into a smaller volume. Therefore, these values cannot be considered
as a crucial factor in the mesh independency study, but it remains an important quantity
for the assessment of the dust distribution.
Figure 5.12: Maximum value of dust volume fraction in the test section volume
Finally, the mass imbalance has been analyzed. As it was previously mentioned, a
maximum mass imbalance of 1% is recommended. Considering that the dust mass
introduced is 0.0015 kg/s, this means that an error of 0.000015 kg/s (15E-06 kg/s) would
be acceptable. In Figure 5.13 we can see that all 5 meshes respect the mass imbalance
conditions, with values well below the limit of ± 1%. The two red line limits have been set
at 5E-06 and -5E-06, which represents a ± 0.33%, in order to better show the stability of
the solution.
Figure 5.13: Dust mass imbalance between injection points and outlet surface
Figure 6.1: Airflow velocity plot through the center line of the test section entrance at 0.8 m/s inlet velocity
Figure 6.2: Air and Dust velocity plot on a line passing through the center of the test section entrance at
0.8 m/s inlet velocity
Facultatea de Inginerie a Instalațiilor - UTCB 55
Ing. CHITARU George-Mădălin Master’s Thesis
Master Efficacité énergétique des installations techniques du bâtiment
Figure 6.3: Air and Dust velocity plot on a line passing through the center of the test section entrance at
11 m/s inlet velocity
NOTE: This last graph (Figure 6.3) has been extracted from one of the simulations
used to assess the performance of the Frontal Grille & Radial Slit injection method
described in [Chapter 6.2]. The inlet velocity was kept at the same value of 0.8 m/s, while
the radial and frontal dust injection velocities were set to approximately 11 m/s. All other
settings remained the same.
To conclude this analysis of the different factors that need to be taken into
consideration when choosing the inlet and the injection velocity we have made a list of
constraints and remarks that should be accounted for, not only when running the CFD
simulations but also while conducting the physical experiments.
List of constraints:
1) When choosing the inlet velocity, the influence of the additional mass flow at the
injection system should be taken into account in order to keep the airflow velocities
within specified limits;
2) When choosing the inlet velocity, the influence of the near-wall boundary region
should be taken into account because it can accelerate the airflow in the center;
3) When choosing the injection velocity, special attention should be given to the frontal
injection nozzles which might produce an acceleration of the airflow passing
through the test section;
4) When choosing the injection method, a better dust dispersion tends to increase the
friction between the air and the particles and therefore reduce the impact of the
injection velocity. The increased friction with the air will diminish the inertial forces
of the particles, accelerating them if the velocity of the airflow is higher than the
velocity of the particles and decelerating them if the particles are introduced at a
higher velocity than the upstream airflow.
5) The cumulative effect of all the aspects previously mentioned should be taken into
account;
Considering all the highlighted factors specified in this subchapter, we have decided
to run all future simulations with an inlet velocity of 0.7 m/s, which should generate an
airflow of around 2.4 m/s if we consider that the calculated value is 2.1 m/s and that the
Facultatea de Inginerie a Instalațiilor - UTCB 56
Ing. CHITARU George-Mădălin Master’s Thesis
Master Efficacité énergétique des installations techniques du bâtiment
influence of the near-wall region adds, in average, an increase of 0.3 m/s to the airflow in
the center. In addition, the influence of the dust velocity was tried to be mitigated by using
appropriate frontal injection velocities so that a uniform airflow is formed at the test section
entrance.
As it can be observed in Figure 6.4: Airflow velocity plot through the center line of
the test section entrance at 0.7 m/s inlet velocity below, our assumptions about the 0,7
m/s were correct. With this inlet velocity, the average speed of the airflow in the middle is
about 2.4 m/s due to the near-wall effect. It should be noted that the small non-uniformity
in the middle of the line is caused only by the presence of the cylindrical injector since no
dust was injected for this simulation.
Figure 6.4: Airflow velocity plot through the center line of the test section entrance at 0.7 m/s inlet velocity
Dust Mass flow Nozzle 2 x 0.0003 [kg/s] 0.0012 [kg/s] 0.0008 [kg/s]
Dust Mass flow Nozzle 3 x 0.0003 [kg/s] x x
Dust Mass flow Nozzle 4 x 0.0003 [kg/s] x x
Dust Mass flow Nozzle 5 x 0.0003 [kg/s] x x
Area Nozzle 1 2.55 [mm²] 2.55 [mm²] 2.55 [mm²] 110.00 [mm²]
Area Nozzle 2 x 2.55 [mm²] 60.00 [mm²] 60.00 [mm²]
Area Nozzle 3 x 2.55 [mm²] x x
Area Nozzle 4 x 2.55 [mm²] x x
Area Nozzle 5 x 2.55 [mm²] x x
Dust VF Nozzle 1 0.1 [-] 0.0200 [-] 0.02 [-] 0.0011 [-]
Dust VF Nozzle 2 x 0.0200 [-] 0.003 [-] 0.0023 [-]
Dust VF Nozzle 3 x 0.0200 [-] x x
Dust VF Nozzle 4 x 0.0200 [-] x x
Low Velocity Test
d) full frontal & radial injection (FFRI) d) full frontal & radial injection (FFRI)
Figure 6.6: Normalized volume fraction contour Figure 6.7: XY Velocity plot on the center line of the
on the test section entrance for the low velocity test section entrance for the low velocity dust
dust injection case injection case
In Figure 6.8 below a more detailed representation of how the solid phase is injected
and how the dust flow is distributed on the small section tube is provided. We can notice
that, even if the particles are injected at around 2.2 m/s, the dust flow near the injection
area quickly loses kinetic energy and falls under 2 m/s (represented by the greenish
colors). However, while traveling along the small section tube, the particles are
accelerated by the airflow around it, eventually reaching the values seen in Figure 6.7
above. In Figure 6.8a, we can see the narrow angle to which the dust is spread for the
SPI method, while in Figure 6.8b we can notice the wider spread angle and a more
significant evolution of the dust flow along the experimental vein for the MPI method. We
can notice that the trajectory of the particles is not only affected near the injection point,
but also while moving towards the test section entrance. This improves the dispersion for
the MPI method but leaves us with that free dust zone right near the center of the tunnel.
As expected, in Figure 6.8c and Figure 6.8d the dispersion is less significant and the
particles tend to follow a straight path towards the outlet.
d) full frontal & radial injection (FFRI) d) full frontal & radial injection (Zoom)
Figure 6.8: Dust velocity contours in the experimental vein for the low velocity dust injection case
d) full frontal & radial injection (FFRI) d) full frontal & radial injection (FFRI)
Figure 6.9: Normalized volume fraction contour Figure 6.10: XY Velocity plot on the center line of
on the test section entrance for the medium the test section entrance for the medium velocity
velocity dust injection case dust injection case
In Figure 6.11 below we can see more clearly how the dust is being distributed for
each case. Nothing really changes for the SPI method compared to Figure 6.8a, just the
angle of dispersion which is a big larger. However, we now notice that the dust flow is
decelerated by the airflow around it and not accelerated like it was the case for the low
velocity injection. This is shown by the velocity contour which maintains a value past 2.6
m/s on the entire length of the small section tube, slowly reaching the 2.5-2.4 m/s mark
towards the end of the test section. The most interesting representation is for the MPI
method, in which we can see how the injected dust particles tend to stick to the wall. This
is caused by one of the aspects of turbulence, which implies that cross-stream motion
pushes the low momentum fluid towards the center of the stream, and the high
momentum fluid towards the boundary layer (wall). Since the dust is treated as fluid
continua (with the same viscosity as the air) and since it has a higher momentum than
the airflow, then it is expected for it to move closer towards the wall. This happens at least
until the dust velocity comes to the same value as the airflow velocity or even lower
because, theoretically, after that it should be pushed towards the middle. A good
representation of this phenomena is shown in the results for the high injection velocity
case where the particles present this exact behavior and are sent towards the center of
the tunnel after reaching a lower velocity due to friction with the tunnel walls. Therefore,
we should also take this factor into consideration when choosing the injection velocity.
For the RI and FFRI methods we can see similar representations with a noticeable
increase in dust velocity for the FFRI case due to the higher frontal dust volume fractions.
The increased velocity and volume fractions of the dust have a negative impact on the
results of the FFRI method. As it can be noticed in Figure 6.11c and Figure 6.11d, even
if the dispersion of the particles at the injection point is better than in the case of RI, the
dispersion at the test section entrance is actually almost the same. This happens because
for the RI the dispersion angle slightly increases while moving from the injection point
towards the test section.
d) full frontal & radial injection (FFRI) d) full frontal & radial injection (Zoom)
Figure 6.11: Dust velocity contours in the experimental vein for the medium velocity dust injection case
d) full frontal & radial injection (FFRI) d) full frontal & radial injection (FFRI)
Figure 6.12: Normalized volume fraction Figure 6.13: XY Velocity plot on the center line of the
contour on the test section entrance for the high test section entrance for the high velocity dust
velocity dust injection case injection case
In Figure 6.14 the evolution of the dust flow in the small section tube, for the high
injection velocity case, can be observed. Here we can clearly see that the SPI method
still shows a very poor dispersion angle compared to the alternatives. For the MPI case,
we can see in Figure 6.14b how the radially injected dust particles are sent towards the
wall where they quickly lose their kinetic energy, passing from red to orange and then to
green contours. As opposed to the medium velocity injection case, this process happens
much faster and it is amplified by the increased friction with the wall. Since the particles
lose their kinetic energy to the point where their velocity is lower than the airflow velocity,
the cross-sectional turbulent motion starts sending the dust particles back into the center
of the flow. In the same figure, the rotation of the dust can also be seen when looking at
the volume fraction contours relative to the XY and XZ planes.
Passing to Figure 6.14c we can immediately see the difference between the MPI
and the RI methods when looking at the travel distance of the particles relative to the
injection point. It can also be noticed, by looking at the close-up view to the injection point,
that this difference comes from the fact that for RI the dust comes out in all directions.
Since the particles are very evenly spread around the radial slit, the fluid-solid drag is
much increased and the particle-particle collisions are reduced. As mentioned before,
these two factors act against the inertial force of the dust which loses its kinetic energy
and is eventually taken away by the upstream airflow.
In Figure 6.14d, it can be noticed that the result for the FFRI method presents the
best dispersion, both near the injection area and along the small section tube. It can be
seen that the dust flow covers a significant area of the test section entrance, and that it
passes right through the middle, where usually the airplane equipment is positioned. This
should provide a full exposition of the equipment to the dust flow, with the only concern
being the velocity impact and the higher dust concentrations in the middle.
In the end, none of the 4 injection methods that were analyzed is flawless. However,
this comparison study has shown that the FFRI method has the best potential of providing
d) full frontal & radial injection (FFRI) d) full frontal & radial injection (Zoom)
Figure 6.14: Dust velocity contours at the injection point for the high velocity dust injection case
INFLUENCE OF THE GRAVITATIONAL ACCELERATION
For all previous simulations we have chosen to neglect the influence of the
gravitational acceleration in order to facilitate the comparison study between the different
injection methods and to reduce the number of variables that could influence the air and
dust flows. Since this is one important aspect of the physical experiments, it had to be
included in our numerical study. However, since adding gravity means adding an
additional level of complexity, we have chosen to focus our attention on only one of the
previous cases that was run without gravity. Considering the results from the comparison
study, the high velocity injection case was chosen, along with the full frontal & radial
injection (FFRI) method, because it was considered the best overall result from all
scenarios that were analyzed.
These being said, the boundary conditions, model set-up and solver set-up was
kept the same as presented in [Chapter 6.3] with the only difference being the addition
of the gravitational acceleration on the negative Y axis. As before, the simulation was kept
running until the residuals and the monitored physical quantities were stabilized.
As it can be seen in the figures below, the addition of the gravity has significantly
changed the results of the previous simulation where we ignored its influence. In Figure
6.15 it can be noticed that the dispersion level of the dust remains about the same, but
the particles are more evenly spread, which generates an increase in the uniformity index
from 0,198 to 0,212. Nonetheless, the most obvious difference is of course the position
of the solid particles when entering the test section. Due to their very high mass, the
injected particles have a descendent trajectory corresponding to the direction of the
gravitational acceleration. This change in trajectory is more or less noticeable depending
on the dust volume fractions. As we can see in Figure 6.15, the concentrated dust zone
from the middle has shifted more than the dispersed particles around it. This happens
because the more the particles are packed together, the less they are influenced by the
airflow which tends to keep them on the same trajectory as the air streamlines. The results
show us that the combined effect of the fluid-solid drag, lift and buoyancy forces, in this
case, is much smaller than the magnitude of the gravitational force. Moreover,
considering that buoyancy is not that significant in our case and that the lift force acts on
the particles only where there are meaningful velocity gradients, we can only assume that
the fluid-solid drag remains the most significant force to oppose the change in trajectory
caused by the gravitational acceleration. Therefore, we have concluded that the low
velocity of the airflow might not be sufficient to provide enough drag and entrain the heavy
dust particles, especially when these are packed together.
Figure 6.15: Normalized volume fraction contour Figure 6.16: XY Velocity plot on the center line of
on the test section entrance for the FFRI injection the test section entrance for the FFRI injection
method with a high dust injection velocity and method with a high dust injection velocity and
gravitational acceleration gravitational acceleration
Figure 6.17: Side view of dust velocity contours for the high velocity dust injection case with added
gravitational acceleration
Figure 6.18: Side view of air velocity contours for the high velocity dust injection case with added
gravitational acceleration
CONCLUSIONS
Considering the air traffic in the desert areas nowadays, where sand and dust
storms have an increased occurrence, many potential problems related to the flight of
airplanes in these harsh zones can arise. Therefore, knowing the particle ingestion
tolerance of the exposed equipment to such environmental conditions can have large
safety and financial implications.
A detailed configuration of the numerical model has been presented in order to
provide all necessary information for replicating the CFD experiment. During the
development of the project, the numerical model has been proven useful when trying to
approximate the required dust mass flow at the injection points in order to provide an
adequate concentration of dust in the test section. A simple iterative process, that uses
the CFD results, has been proposed in order to correlate the required dust mass at the
injector with a certain concentration value in the test section. Additionally, the porous zone
configuration that was used to take into account the variable pressure drop on the heating
battery has proven to offer very close results to the experimental data.
A mesh independence study has also been conducted and the results have
indicated that only a transversal refinement was necessary to obtain a grid independent
result. Additionally, it was shown that the highly skewed elements caused by the
cylindrical shape of the injector do not affect the accuracy of the simulations. However,
the mesh independence study has only taken into consideration one injection method and
one injection velocity. Individual grid comparison studies might be necessary for each
separate case.
A comparison between 4 injection methods using 3 different injection velocity
regimes has been conducted. The results have shown that changes to the reference
injection system can be made in order to obtain a more even distribution for the granular
phase. The last proposed injection method (FFRI) tries to overcome the drawbacks of the
previous ones, showing the best results when a high-velocity injection is employed. Still,
due to the high dust velocity in the middle of the tunnel, the uniformity of the flow is not
perfect. Further investigations will focus on mitigating the influence of the frontal injected
particles.
During the comparison study, some convergence stalling and fluctuations of the
physical quantities have been noticed for the medium and high injection velocity cases.
Analyzing the possible sources of the stalling, two possible solutions have been
envisaged for the next steps of the project: extending the virtual model in order to have a
fully developed flow at the outlet and changing the simulation from a steady-state to a
time-dependent solver.
Finally, it was shown that the gravitational acceleration is a strong influential factor
in our numerical study, its addition changing the dispersion level, the volume fraction
distribution and the trajectory of the particles. The results from this last simulation suggest
that the gravitational force is greater than the drag, lift and buoyancy forces and this allows
the particles to fall towards the bottom of the tunnel. This type of simulation could be used
to approximate the correct positioning of the equipment inside the test section so that it
is fully exposed to the dust flow. However, an experimental validation of the particle
behavior is required in order to support our claims.
All things considered, this work provides a better understanding of a particle-laden
flow inside a harsh environment testing facility that is used for dust and sand exposure
experiments. Suggestions of possible modifications to this test facility have been made,
BIBLIOGRAPHY
[1] Lekas, Th. I., Kallos, G., Kushta, J., Solomos, S., Mavromatidis, E. (2011) - Dust impact on
aviation. 6th International Workshop on Sand/Dust Storms and Associated Dust fall,
September 7-9, Athens, Greece;
[2] Ghosh, T. (2014) - Dust storm and it’s environmental implications. Journal of Engineering
Computers & Applied Sciences (JECAS) ISSN No: 2319-5606 Volume 3, No.4, April 2014;
[3] Middleton, N., Kang, U. (2017) - Sand and Dust Storms: Impact Mitigation. Sustainability
2017, 9, 1053; doi:10.3390/su9061053;
[4] Bojdo, N., Filippone, A. (2014) - Effect of desert particulate composition on helicopter engine
degradation rate. 40th European Rotorcraft Forum, Southampton, September 2014
[5] Ganor, E., Osetinsky, I., Stupp, A., Alpert, P. (2010) - Increasing trend of African dust, over
49 years, in the eastern Mediterranean. Journal of Geophysical Research: Atmospheres
Volume 115, Issue D7, April 2010, doi:10.1029/2009JD012500
[6] Sissakian, K. V., Al-Ansari, N., Knutsson, S. (2013). Sand and dust storm events in Iraq.
Natural Science 5 (2013), No. 10, 1084-1094, Article ID:37734, doi:10.4236/ns.2013.510133
[7] Notaro, M., Yu, Y., Kalashnikova, O. V. (2015) - Regime shift in Arabian dust activity,
triggered by persistent Fertile Crescent drought. Journal of Geophysical Research:
Atmospheres Volume 120, Issue 19 Pages: i-v, 9837-10568, October 2015, doi:
10.1002/2015JD023855;
[8] Krasnov, H., Katra, I., Friger, M. (2016) - Increase in dust storm related PM10 concentrations:
A time series analysis of 2001-2015. Environmental Pollution Volume 213, June 2016, Pages
36-42;
[9] United Nations Environment Programme UNEP (2016) - Global Assessment of Sand and
Dust Storms. ISBN: 978-92-807-3551-2, UNEP Job Number: DEW/1971/NA
[10] International Conference on Combating Sand and Dust Storms: Challenges and Practical
Solutions Tehran, the Islamic Republic of Iran (3-5 July 2017)
[11] European Aviation Safety Agency (2007) - Certification Specifications for Engines -
Acceptable Means of Compliance (CS-E Book 2);
[12] Reagle, C.J. (2012). Technique for Measuring the Coefficient of Restitution for Microparticle
Sand Impacts at High Temperature for Turbomachinery Applications. Virginia Polytechnic
Institute and State University, August 2012, Blacksburg, Virginia, USA;
[13] Khalifa, E.M. (2016). Conscious study of impact of dust storm on aviation and airport
management. International Journal of Science Research and Technology, Volume 2 Issue
2, p 51-57, 25th June 2016;
[14] Georgescu, M. R., Chitaru, G.M., Coșoiu, C.I., Brânză, I., Nae, C. (2017). Numerical study
of the secondary phase dispersion in a particle-laden flow. International Conference on
ENERGY and ENVIRONMENT (CIEM) 2017, doi:10.1109/CIEM.2017.8120851;
[15] EUROCAE ED-14G (2011), Environmental conditions and test procedures for airborne
equipment. Section 12 Sand and Dust;
[16] Georgescu, M. R., (2017) – Cercetări privind simularea interacțiunii particulelor în suspensie
cu suprafețele solide. Universitatea Tehnica de Constructii Bucuresti – Facultatea de
Inginerie a Instalatiilor;
[17] ANSYS Documentation R18.2 (2017), ANSYS Fluent Theory Guide, Chapter 17: Multiphase
Flows - 17.5.4. Conservation Equations, 17.5.10. Virtual Mass Force, 17.5.6.2. Fluid-Solid
Exchange Coefficient,
[18] Ratnam, G. S., Vengadesan, S. - Performance of two equation turbulence models for
prediction of flow and heat transfer over a wall mounted cube. International Journal of Heat
and Mass Transfer, Volume 51, Issues 11–12, June 2008, Pages 2834-2846,
https://doi.org/10.1016/j.ijheatmasstransfer.2007.09.029;
[19] Fröhlich, J. and von Terzi, D. (2008) - Hybrid LES/RANS methods for the simulation of
turbulent flows. Progress in Aerospace Sciences. 44. 349-377, DOI: 10.1016/j.paerosci.
2008.05.001
ANNEX
50%
Percentage of cells
40%
30%
20%
10%
0%
0.95 0.90 0.85 0.80 0.75 0.70 0.65 0.60 0.55 0.50 0.45 0.40 0.35 0.30 0.25 0.20 0.15 0.10 0.05 0.00
Equiangle Skewness
60.0%
50.0%
Percentage of cells
40.0%
30.0%
20.0%
10.0%
0.0%
66 63 60 56 53 50 46 43 40 36 33 30 26 23 20 17 13 10 7 3
Aspect Ratio
Histogram of Quality
Initial Mesh Refined Mesh 1 Refined Mesh 2 Refined Mesh 3 Refined Mesh 1 HQ
70%
60%
Percentage of cells
50%
40%
30%
20%
10%
0%
0.95 0.90 0.85 0.80 0.75 0.70 0.65 0.60 0.55 0.50 0.45 0.40 0.35 0.30 0.25 0.20 0.15 0.10 0.05 0.00
Equiangle Skewness