Professional Documents
Culture Documents
IMECE2013
November 15-21, 2013, San Diego, California, USA
IMECE2013-65627
Shahla Eshraghi
Payame Noor University of Tehran
Dept. of Biological Science
Tehran, Iran
Sh.eshraghi@gmail.com
(a)
swirler section
SUBSCRIPTS A A dp
Air Core
Vortex
A Acceptor
Ls Ds
D Donor ,U
l liquid swirl
chamber
orr orifice x,Ux section r,Ur
p Inlet port- Primary phase
a Air
r Radial direction Lo
exit
do orifice Lp
sw Swirl Chamber section View A-A
t Tangential Direction - turbulent
Tangential Direction
z Axial Direction Thin, hollow, conical liquid sheet
1 Original condition (3-D)
2 Equivalent condition (2-D) (b)
INTRODUCTION
Pressure swirl atomizers are widely in use because of their ability
to produce fine sprays in a process commenced by liquid sheet
break up. Formation of the very hollow cone sheet of liquid is a
result of swirling motion of fluid inside the swirl chamber and the
orifice of the atomizer which is referred to as the internal nozzle
flow [1,2]. As illustrated in Fig. 1, the swirling motion will Figure 1: Internal Nozzle Flow shown in a 2-D Schematic (a)
account for the formation of an air core inside the atomizer in a and Image of the Flow by Ma [2] (b).
central negative pressure region. The boundary between the liquid
turbulence model instead of laminar flow stipulation
and the surrounding gas (interface) has been studied by many
Some researchers made useful comparisons between
researchers such as Rizk and Lefevbre [3], Ashraf and Miliand [4]
laminar and turbulent flow modeling to get deeper
and Nonnenmacher and Piesche [5].
understanding of the nature of the flow they were dealing with
Location of this interface at the tip of the nozzle orifice and in and investigated the performance of existing models. Hansen et
the ambient gas is an indicator of the liquid film thickness and al [10] used Large Eddy Simulation (LES) and laminar model
primary spray cone angle respectively. Another indicator or spray in conjunction with the VOF and the Eulerian multiphase
parameter related to the performance of atomizers is the discharge models for a three-dimensional simulation and concluded that
coefficient, which is a measure of the resistance to flow due to laminar-VOF was the best combination in terms of
existence of air core and frictional losses along the nozzle. Simplex computational time and accuracy. History and detailed
atomizers are expected to possess low discharge coefficients due to description of VOF model and different tracking schemes could
the presence of the air core within the nozzle [6]. be found in publications of Hirt & Nichols[11], Rider and
To date, valuable numerical works have relied on various Kothe [12], Gueyffier et al. [13] and Gorokhovski and
models and techniques to successfully capture a sharp and accurate Herrmann [14].
interface and evaluate the dependency of spray parameters on Numerous discussions could be found in the literature about
boundary conditions such as inlet pressure and swirl strength. In methods and algorithms of capturing the interface between fluids
fact, existence of features like turbulence and interface between in different phases. Simpler schemes such as the Implicit or the
phases complicate the modeling of this type of flow to a great Euler-Explicit method available with the VOF model received
extent. Nonnenmacher and Piesche [5] and Chu et al. [7] used a attention from researchers such as Yeh [15] who presented closely
sophisticated analytical-numerical approach for laminar flow to matching results for a set of turbulent models. Simplicity stems
solve for the film thickness, velocity profile and some other from the fact that such schemes rely fundamentally on an
parameters. Mandal et al [8] used the laminar flow assumption in averaging process and hence fail to represent sudden jumps for
their numerical study to extend their simulation from Newtonian to variables like density and viscosity. An unfortunate consequence
Non-Newtonian fluids. A similar study was conducted by Xue [9] of this approach is smearing of the interface according to Hirt &
who improved the modeling through the use of a sophisticated Nichols [11]. Besides this, such simple schemes allow for
turbulence model instead of laminar flow stipulation. unbounded diffusion of the scalar volume fraction (). Thus, a
The computational domain in this study is simplified to a 2-D 0.95 2.34 2.01 1.21 3.2 406 651
axisymmetric domain to reduce the computational effort (Figs.
2,3). This simplification is in accord with the work of Ashraf and
Miliand [4] and Nonnenmacher and Piesche [5]. In order to In the present study, implicit scheme is used for verification of
maintain similarity between the real flow and the simplified case simulation results in the first step. For this scheme, two grids with
(annular inlet), the following set of equations has to be considered: 7641 and 24087 quadrilateral elements are used (Fig. 3) to ensure
about grid-independence of results. Finer grid size in this
simulation was wisely selected to a size close to the grid size
m 1 = m 2 , (1) used by [8] who concluded that 26682 grid cells was
sufficiently fine to obtain valid numerical results for the same
Vin1 2 = Vin2 2 ,
1 1 flow geometry. Though, the grid was constructed in such a way
(2)
2 2 to avoid heavy and unnecessary computational load based on the
Vin1 Rsw -Rp =Vt2 .Rsw ,
fact that grid was refined in one step and only in zones where
(3)
interface between phases was expected to happen in the domain.
0.5
(4) Next, adaptive grid refinement technique has been used to attain a
Vr2 = (V2 2 -Vt2 2 ) ,
sharp interface between phases along the nozzle. Comparison with
m 2
t = ,
experimental data is performed in each step during refinement and
Rsw 2 l Vr2 (5) numerical results are carefully observed for verification of
numerical modeling. Later, the capability of Adaptive Mesh
Refinement Technique in generation of a sharp interface between
Wall
Orifice wall
Axis of symmetry
Figure 3: Computational Domain With (a) 7641 Cells And (b) 24087 Cells
Where +++++
)* )* is the Reynolds stress term and k and p are Here, p represents volume fraction at cell centroid of the
inverse effective Prandtle numbers for k and respectively. computational cell while variables p,f and Vf are calculated at face
Default values of the k and p (k=p=1.393) has been used in centroids surrounding the very computational cell respectively.
the present work [19].
The model finds the effective viscosity ( eff. tur lam. )
through the following expression for quantity which is the Continuum Surface Tension Force (CSF) and Weber
ratio between effective turbulent viscosity and molecular Number
dynamic viscosity [19] In the CSF model, surface tension is represented by a
volumetric force calculated from the curvature of interface at
any instant of time and divergence of scalar volume fraction
2 k
d( )=1.72 (12)
3 -1+Cv
0.5 0.5
ki i
( )
Fv = ij (15)
0.5(i +j )
Here , , , K are density, dissipation rate, viscosity, and the
kinetic energy of turbulence respectively. Default value of 100
has been chosen for Cv as the model constant in the present Where k is curvature of the interface (=. n/ ), and n/ is the
paper [19]. normalized normal vector ( n=q ). Finally Weber number for
film defined based on the film thickness and average velocity
of the film at the orifice is introduced. This number is defined
as the ratio of inertia to surface tension for a droplet and
Multiphase Flow Modeling smaller ranges of Weber number indicate higher level of
In the VOF multiphase model, presence of phases is influence of surface tension on flow properties and interface
recognized through the scalar phase volume fraction (). This shape evolution. This number is defined as [ 21]
scalar varies from zero (full of the secondary phase) to one (full of
the primary phase) in each computational cell. Through this l U1234 L (16)
scalar, variables density and viscosity are averaged in the model Wfilm =
(avg = + g (1-) and avg = + g (1-) ). The model is
primarily concerned with continuity equation of scalar volume
fraction [19,20] Where , Ufilm, L and are the density of liquid, average
velocity and thickness of the liquid film at the exit of orifice ,
and surface tension between air and water respectively.
(p p ) (p p Vi )
+ =0 (13)
t xj Discharge Coefficient
Finally discharge coefficient is introduced as a flow
parameter used to compare the actual flow rate to the
Here, p , p , and V represent density, volume fraction, and
theoretical ideal flow rate. In our study, pressure difference is
velocity of the liquid phase in the computational domain. computed by the software and calculation of this parameter is
based on Eq. 22 introduced by Heywood [22]
Interface Calculation 5678.
Cd =
(Ao :2P/)
(17)
Implicit Schemes
Interface in the VOF model is represented by means of the
scalar volume fraction. Available schemes in the VOF model Here Qact, Ao, P, and are actual flow rate from the
control derivation of Eq. 13. The simple Implicit scheme experiment, cross section of orifice, pressure drop along the
discretize the Eq. 13 without setting any limitation on diffusion nozzle computed by Fluent and density of the liquid phase.
Figure 4: Contour of Volume Fraction for Air, Failure of the standard K-; model
solution.
Inlet Static Pressure (Psi)
Inlet Static Pressure (Psi)
(1) (2)
Inlet Static Pressure (Psi)
(3) (4)
Figure 5: History of Inlet Static Pressure, (1) RNG K-, (2) STD K-, (3) SST K-, (4) RSM Models
models in terms of computational preferability. Tablel 3 illustrates sligthly higher pressure in presence of
As Fig. 5 illustrates, The RNG K- model converges to this force in steady-state simulation results. To our surprise, no
steady state solution within the smallest number of remarkable difference has been observed between other
computational iterations (almost 30,000 iterations) and properties of the flow listed in Tab. 3. This result has
experiences a robust and stable course of computation. In encouraged the authors to investigate the influence of surface
comparison, the STD K- model in diagram (2) experiences tension under lower inlet velocities towards an effective Weber
few oscillations but the model experiences divergence before number further on this study. In this regard, the inlet velocity
converging to a steady state solution. has been reduced successively from 2.34 m/sec to 1.5, 1, 0.5
Similarly, growing instability in computation using SST K- and 0.25 m/sec. The results of the simulation for the first 200
is discernible from the diagram (3) in fig. 5. Divergence of time steps are available in Fig. 8.
the numerical modeling despite early obtained and appreciably
maintained stability in simulation for large number of time
steps was a result of unbounded increase of turbulent viscosity Itr. No. = 30,000 Itr. No. = 72,340 Itr. No. = 79,000
growing on the interface in the regions of the domain where
continuously refined grid cells were located.
Finally, in the case of the RSM model, there has been a
trade -off between appreciable stability of the simulation and
comparably heavier computational burden needed to reach a steady
state solution.
Qualitative representation of simulation results with the (a) 3.4 mm (b) 3.3 mm (c)
nominated model, RNG K-, is illustrated in figure 6 of this
paper. In this figure, contour of volume fraction of air displays
the regions of the domain completely filled by air in red and 0 1
degree of sharpness of color gradient from red to blue Figure 6: Contours of volume fraction with RNG k- model
(emptiness of air) manifests quality of interface representation (a) 7641 (b) 24807 and (c) 18653 grids cell
along the interface. Thereby significant improvement in quality
of interface has been resulted through both approaches of
refinement. Surprisingly, substantially better quality
improvement has been observed through the more efficient SIMULATION OF ORIGINAL FLOW,
PRESSURE-ITERATION
adaptive refinement of the mesh in diagram (c). However 25000
though, despite the fact that remarkably less number of grid
cells are participating in representing the sharper interface 20000
compared to diagram (b), this quality improvement is offset by
Pressure (Pa)
15000
dramatic increase of computational burden up to almost 79,000
no CSF
number of iterations illustrated in Fig. 6. 10000 with CSF
Pressure (Pa)
Discharge coefficient 0.2184 0.2167 0.218 600
Spray angle 76o 76o 77o Weber No= 444
400
Film Thickness (mm) 3.2 3.2 3.06 no CSF (I) no CSF (T)
200 with CSF (I) with CSF (T)
Iteration
The immediate observation from Fig. 8 is that inclusion of 0
CSF model becomes influential within the first few time steps 0 1000 2000 3000 4000 5000
of simulation in plots (a) to (d). In fact, interaction of the
surface tension force with high curvature of the interface in
Static Pressure At Inlet For Vin=1 m/sec
circulation region in vicinity of the inlet port has led to 250
significantly higher inlet static pressure reported in each plot. 0 40 80 120 160 200
Besides, ascending difference between curves in successive plots Time step
200
at the same flow time (labeled as with/no CSF (T)) indicates a (b)
Pressure(Pa)
growing influence of the surface tension by reduction of the inlet 150
Weber No= 204
velocity in simulation. Indeed, higher curvature of interface under
lower inlet velocities has led to existence of stronger surface 100
tension force in this circulation region (Eq. 15). As a consequence no CSF (I) no CSF (T)
of this higher resistance to the flow, higher static pressure in the 50 with CSF (I) with CSF (T)
inlet port has been reported compared to situations without Iteration
presence of surface tension during the simulation. 0
Further comparison between plots a, b, c and d reveals a 0 800 1600 2400 3200 4000
trend of ascending difference between final iteration numbers
of the curves labeled with/no CSF (I) in successive plots. Static Pressure At Inlet For Vin=0.5 m/sec
This difference grows from 200 iterations in plot b to 900 75
0 40 80 120 160 200
iterations in plot c and 1800 in plot d. In fact, this growing 60 Time step
difference is not only indicative of higher convergence rate in (c)
each time step with presence of surface tension in modeling but 45
Pressure (Pa)
also increase of the convergence rates with reduction of inlet Weber No= 52.2
velocity. Again similar to the trend observed in previous 30
discussion, onset of the later trend has occurred with the inlet no CFS (I) no CSF (T)
15 with CSF (I) with CSF (T)
velocity of 1 m/sec which could be introduced as a threshold
inlet velocity for recognition of an influential surface tension 0 Iteration
force in our simulation. 0 900 1800 2700 3600 4500
Constant value of 0.072 N/m represents surface tension of
water in calculation of Weber number for each case in equation
16. This number is 1240 for the original flow and 444, 204, Static Pressure At Inlet For Vin=0.25 m/sec
40
52.2 and 12.2 for inlet velocities of 1.5, 1, 0.5 and 0.25 m/sec 0 40 80 120 160 200
respectively. Data for calculation of Weber numbers is (d) Time step Pressure (Pa)
illustrated on each corresponding plot in Fig. 8. Observation of 30
Weber No= 12.2
these trends reveal that inclusion of CSF model is important for
inlet velocities less than 1 m/s which leads to introduction of a 20
threshold Weber number of 204 associated with this inlet
velocity. 10
no CFS (I) no CSF(T)
with CFS (I) with CSF (T)
0
CONCLUSION REMARKS 0 1100 2200 3300 4400 5500
The VOF model has been introduced to 6 turbulence models Iteration
through the implicit scheme and performance of each model
Figure 8: History Of Inlet Static Pressure For Different Inlet
was successfully investigated in this study. Two out of six velocity Magnitudes: (a) Full Simulation (I) Refers To
models- Standard and Realizable K- - failed to capture a well- Iteration and (T) Refers To Time Step