1. Topic 1 – Internal flow
Presenter: Marco Arienti, Sandia National Laboratories
Support by Sandia National Laboratories’ LDRD (Laboratory Directed Research and Development) is gratefully acknowledged. Sandia National Laboratories is a multi-program laboratory managed and operated by Sandia Corporation, a wholly owned subsidiary of Lockheed Martin Corporation, for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-AC04-94AL85000.

2. Spray C/D (4 contributors)
Politecnico di Milano - OpenFoam:
    Ehsanallah Tahmasebi,  Tommaso Lucchini and Gianluca D'Errico
ANSYS-FLUENT: Saeed Jahangirian, Aleksandra Egelja-Maruszewski, and Huiying Li
Università di Perugia - Converge: Michele Battistoni
CMT - CavitatingFoam (OpenFoam) Pedro Martí

3. Spray C
Spray D
Common rail fuel injector
Bosch 3-22
Fuel injector nominal diameter
0.20 mm
Nozzle K factor
K=0
Nozzle shaping
5% hydroerosion
Flow with 10 MPa pressure drop
200 cc/min
Number of holes
1 (single hole)
Common rail fuel injector
Bosch 3-22
Fuel injector nominal diameter
0.186 mm
Nozzle K factor
K=1.5
Nozzle shaping
Hydroerosion to Cd=0.86
Flow with 10 MPa pressure drop
228 cc/min
Number of holes
1 (single hole)
Wireframe of the tangentially-averaged interior wall of the sac
Radius
Axial coordinate

4. Zwart-Gerber-Belamri
Institution/Code
Uni-PG
Converge
ANSYS-FLUENT
Polimi
OpenFOAM -
cavitatingFoam
CMT
Cavitation Model
Homogenous Relaxation
Zwart-Gerber-Belamri
Homogenous
Equilibrium
Inclusion of turbulent viscous energy generation Y
Turbulence
LES
Dynamic sgs
RANS:
SST k-ω with compress.
SST k-ω
- k-epsilon
- SST k-ω
Spatial discretization
2nd order
- QUICK for void fraction
- 2nd order
Solver
PISO
Steady-State Coupled
PIMPLE
Yl = 1.46e-6 s2/m2
Yv= 3.43e-5 s2/m2
Instead of the classical PISO algorithm, a PIMPLE approach is used to support partial convergence of intermediate iteractions, it can be turned into a PISO or SIMPLE algorithms by selecting the right number of inner and outer loops. PIMPLE algorithm combines the loop structures of PISO and SIMPLE, including terms in equations but not limited by Courant number [OpenFOAM].
POLIMI
Details about the solver and its validity presented by (Salvador et al 2010)
For solving N-S equations , PIMPLE (Piso-Simple) algorithm with 3 correctors and 4 non-orthogonal correctors is used.
Adjustable time steps according to the maximum Courant number =0.5 and maximum Acoustic Courant number = 20 is applied.
Euler algorithm for time derivatives, and Gauss algorithms are used for divergence schemes.
Results are presented at time = 0.002

5. EOS models Uni-PG Converge ANSYS-FLUENT Polimi OpenFOAM -
cavitatingFoam
CMT
K liquid bulk modulus.
Schmidt et al., Int. J. of Multiphase Flow (2010)
[1] Caudwell et al., Int. J. of Thermophysics 25(5) 2004
[2] To match Khasanshin, et al. Int. J. of Thermoph. 24(5) 2003
[3] Zwart et al. ICMF 2004
[1] Salvador et al., Mathematical and Computer Modelling
[1] Desantes et al., SAE l Paper
[2] Khasanshin, et al. Int. J. of Thermophysics 24(5) 2003

6. Fixed fully open needle configuration
Institution
Code
Uni-PG
Converge
ANSYS-FLUENT
Polimi
OpenFOAM
CMT
Inlet boundary
P = 150 MPa T = 343 K
Outlet boundary
P = 20 MPa T = 303 K
Fixed fully open needle configuration
CMT: Time varying static pressure?
Low of the wall?
Slip vs. no slip?

7. 3D, full axis-symmetric model
Institution
Code
Uni-PG
Converge
ANSYS-FLUENT
Polimi
OpenFOAM -
cavitatingFoam
CMT
OpenFOAM -cavitatingFoam
Dimensionality
3D, full axis-symmetric model
2D axis-symmetric
3D 5o wedge
Cell Type
- Cartesian cut cells
- Wall functions, y+ = 30
Hex mesh with 10 boundary layers (from 1 μm)
Hex & tets
quads
Cell count (total interior and exterior)
2.5 mm
20k (79k in adapted mesh)
51k (Spray C) 54k (Spray D)
Submerged N Y

8. Internal flow: sharp (spray C) vs. smooth (spray D) pressure decrease

9. Without cavitation, Spray D produces a slightly longer liquid core length and a narrower cone angle
Spray C
Spray D

10. This effect is recognized in new measurements of the spray width and length
From spray boundary contrast (threshold 0.37 KL) using the diffuse backlit illumination (DBI) technique:*
Diffuse backlit illumination is a technique for measuring soot volume fraction and liquid length.
KL is what we call the optical thickness and has no dimension. It is the natural logarithm of the transmission of light through the line of sight. The diffused nature of the light counteracts beam steering effects caused by gradients in the refractive index through the media and leaves only light attenuation from absorption and scattering. The dense spray region near the nozzle scatter the light and we use this physical interaction to identify the spatial location of the liquid phase of the spray.
*from Fredrik Westlye’s presentation

11. Comparison against measured mass flow rate
[g/s]
CONVERGE and FLUENT-ANSYS simulations are the only that capture the increase between spray C and D
In the aggregate, there is more variation amongst models for the same spray type than between the sprays for the same model

12. Comparison against measured momentum
CONVERGE and FLUENT-ANSYS simulations are the only to capture the increase between spray C and D (by a rather small margin)

13. Mass flow rate and momentum values
SPRAY C
Experiment(*)
Uni-PG
Converge
ANSYS-FLUENT
Polimi
OpenFOAM k-ω SST
CMT
OpenFOAM k-ε
Mass flow rate (g/s)
10.07±0.11
10.3
10.8
12.8
10.4
Momentum (N)
5.83±0.06
6.29
6.49
7.69
6.30
6.79
SPRAY D
Experiment
Uni-PG
Converge
ANSYS-FLUENT
Polimi
OpenFOAM k-ω SST
CMT
OpenFOAM k-ε
Mass flow rate (g/s)
11.72±0.15
10.9
11.3
11.6
10.2
10.5
Momentum (N)
7.03±0.11
6.41
6.62
6.27
6.24
6.69
(*) std. dev. from the CMT measurements on 5 different specimens

14. Spray C: noticeable differences in boundary thickness between simulations

15. Spray D vs. spray C at the exit orifice
Similar velocity/density profiles are obtained for spray D
Cavitation displaces mass flow toward the orifice axis in spray C
Spray C
Spray D

16. The effect of cavitation for spray C
Note the different models’ effectiveness in generating cavitation at the orifice’s wall
liquid core boundary

17. Conclusions
Relatively small variations in the amount of cavitation at the wall result in differences of mass flow rate and momentum for spray C simulations
Even when the variation is correctly predicted, its magnitude is underestimated
The trend in spray penetration/width from spray C to spray D is correctly captured by the only non-submerged simulation (UniPG with Converge)
Cannot quantify agreement for lack of averaged data
Passing pockets of vapor in the liquid core are shown in the only LES simulation (UniPG with Converge)
A frequency analysis of this feature is recommended

18. Topic 1.2 – Spray A needle transient opening
Presenter: Marco Arienti, Sandia National Laboratories
Support by Sandia National Laboratories’ LDRD (Laboratory Directed Research and Development) is gratefully acknowledged. Sandia National Laboratories is a multi-program laboratory managed and operated by Sandia Corporation, a wholly owned subsidiary of Lockheed Martin Corporation, for the U.S. Department of Energy’s National Nuclear Security Administration under contract DE-AC04-94AL85000.

19. Two of the remaining questions for Spray A from ECN3:
What is the exit temperature of the fuel?
Is the injection transient modeled realistically?

20. Spray A (3 contributors)
CMT - OpenFOAM w/ Eulerian Spray Atomization Pedro Martí
Bosch - Cascade Technologies Edward Knudsen, Eric Doran (Bosch Research & Technology Center)
SNL - CLSVOF Marco Arienti
Initial transient was not the priority for Bosch

21. Institution: Code: Bosch Cascade Technologies CMT OpenFOAM ESA SNL
CLSVOF
Equation of State for the liquid phase
Peng-Robinson
Non-linear r(P,T)(Payri et al., Fuel 2011)
Tait eqn. calibrated for n-dodecane; new e(P,T)
Moving mesh N
N/Y (axial only) Y
Inlet
Static pressure increases from 0.5Pinj to Pinj at t = 0
Time-varying static pressure
Constant pressure
Turbulence
LES
Dynamic sgs model
RANS
SST k-ω
No turbulence model
Inclusion of turbulent viscous energy generation?
Spatial Discretization
2nd order
1srorder
1st / 2nd order
􏱡􏰳􏰵CMT inlet boundary condition: constant temperature time-varying velocity condition at injection pressure and temperature.
‪I use the standard mesh solver included in OpenFOAM. The input is the wall velocity and direction, and I set up the solver for cells deforming equally in all directions.

22. Spray A reference and actual laboratory conditions
At SNL and ANL, ambient density is matched at cooler, non-vaporizing conditions. From Lyle et al. SAE
Spray A
ANL
SNL
Liquid T [K]
363
333
343
Gas
0% O2 N2
Gas T [K]
900
303
440
Back-pressure [MPa] 6 2 3
Density kg/m3
22.8
Bosch*, CMT+
SNL+
Injection into a low-temperature environment is necessary at Argonne in order to prevent damage to x-ray transparent polymer windows.
A slightly elevated ambient temperature (440 K) at Sandia is required because of seal design for the high-temperature chamber. The ambient temperature in either facility is lower than the boiling point temperature for n- dodecane (489 K at 1 bar) such that minimal vaporization occurs.
Line-of-sight radiography measurements were made at different axial and radial positions.
*Tfuel,intern.< 363 K
+Tfuel,intern. = 343 K

23. Exit temperature predictions from ECN3
ANL
Converge
Sandia
CLSVOF
UMass
HRMFoam
CMT
ESA
IFP
C3D
Incompressible
Non-linear function of p,T
Const. compressibility
Stiffened gas EOS
turbulent viscous energy generation
Specific heat constant for CMT
DT = 0
DT ≅ 0
DT << 0
DT << 0
DT < 0
T [K]

24. Contributions to DT = Texit-Tinlet
Expansion through the orifice 
Viscous energy dissipation 
Heat transfer through injector’s wall DT 

25. Liquid phase compression
Peng-Robinson
Calibrated Tait
100% C12H26
Tc = 658 K
rc =226 kg/m3
p = 2000 bar
r = r0(T ), p0 = 1 bar
Liquid phase compression
CMT - Density and speed of sound as function of pressure and temperature were calculated by Khasanshin et al. [22].
[Caudwell et al., Int. J. of Thermophysics, 2004]

26. Isentropic expansion upper bound:
Dp = bar DT = -22 K from calibrated Tait EOS DT = 0 K from isobaric EOS
DT = -217 K from adiabatic p.g. EOS (g =1.4)
787 kg/m3 363 K bar
646 kg/m3 341 K
60 bar
adiabatic p.g.: g = 1.4
Density[ kg/m3]
Temperature [K]
Crosses r = 600 at Mpa
Gets to 145 K

27. SNL results show limited temperature increase with adiabatic walls
Temperature [K]
Density [kg/m3]
Adiabatic w.
Constant TW = 383 K
Adiabatic w.
Constant TW = 383 K
DTL,exit = +3 K
DTL,exit = +18 K
rL,exit = 716 kg/m3
rL,exit = 720 kg/m3

28. CMT results also show small DT except near the wall
Temperature [K]
Density [kg/m3]
Adiabatic 343 K
Constant TW = 363 K
Adiabatic 343 K
Constant TW = 363 K

29. The viscous dissipation of turbulent energy is the main source of temperature increase
Orifice cross-sections:
273 K
303 K
323 K
Adiabatic
343 K
363 K

30. However, the opening transient displays a bulk temperature increase
Simulation with moving needle Tw = 383 K
Interpretation: the fuel heats up while passing through the narrow gap between needle and injector
This effect disappears once the passage is fully open

31. Independent study: transient and non-isothermal modeling of cavitation with GFS*
Steady-state temperature field
Variation of the outlet temperature in one injection cycle
350 K
500 K
Minimum gap: 5 mm (with standard wall function)
Minimum cell size
Dx = mm
City University in London
*By Salemi, McDavid, Koukouvinis, Gavaises, and Marengo, in ILASS 2015

32. Conclusions on DT = Texit-Tinlet
Expansion through the orifice:
Moderate but constant during injection
Potentially under-estimated depending on EOS
Viscous energy dissipation:
Potentially large but transient
Puts under scrutiny the choice of standard wall function in micron-size gap

33. The measured Rate of Injection (ROI) and Rate of Momentum (ROM) of Spray A
The experiments were intended to be "cold", but in fact we have a different ambient gas T distribution and the duty cycle on the injector is different. In addition, Argonne's data are an ensemble average as collected, rather than a collection of instantaneous data that is later averaged. There may be some jitter problems that make it harder to define the true start of injection.
In the lab, vibrations may transmit pressure waves in the hydraulic tube that are falsely associated as ROI
The virtual ROI has been tuned based on some the experimental data that are considered real—for example we have measured momentum that coincides with measured fuel pressure fluctuations (measurement is 7 cm from the injector inlet). We believe those.
Please also see the latest in SAE
Diagram from SAE

34. Initial conditions: injection delay
as a function of partially filled sac/orifice
Vgas = mm3 (1/3of the sac)
Tdelay = ( ) ms = 9 ms
Fully open fuel passage
Time of apparent injection
Tdelay = 3 ms (instantaneous opening)
Vgas = 4 mm3 (half orifice) At t < 0 the pressure in the sac is ~Pinj/2

35. Mass flow rate during opening transient*
Steady mass flow: 2.56 g/s
*After removing all injection delays

36. Momentum flow rate during opening transient

37. Jet penetration during opening transient
I think it is worth mentioning explicitly in the caption that bosch has a fixed needle.  As I mentioned to Sibendu, we looked a lot more at downstream penetration than at initial transients and we therefore didn’t tune the initialization to match experiments.  The transient profile that you plot is fine to show, but it would be helpful to make clear that its shape a strong function of our arbitrary initialization and the lack of needle movement.

38. A request: establish a common set of properties and reliable EOS correlations
T = 353 K
Pressure [MPa]
Speed of sound [m/s]
Example: speed of sound calculation for liquid n-dodecane
Khasanshin et al., Int. J. of Thermophysics, 24(5) 2003
Padilla-Victoria, Fluid Phase Eq. 2013
Note that D coefficient in Arient’s paper should be corrected as: D = 0.16+( )/12+2.5e-3*T-5.85e-6*T*P;

39. Backup

40. Note 3: Dependence of internal energy on pressure
P = 0.1 MPa
P = 20 MPa
P = 140 MPa
New fit:
NIST data:
P = 0.1 MPa
P = 20 MPa
P = 140 MPa
Supercritical
Equation of state for n-dodecane compiled from data; Khasanshin, et al., Int. J. of Thermophysics, Padilla-Victoria et al., Fluid Phase Equilibria 2013.
[JSAE SAE ]

41. Experiment set-up and reference parameters
Fuel
n-dodecane
Inlet pressure
150 MPa
Ambient pressure
6 MPa
Fuel Temperature
363 K
Thermodynamic properties from NIST web-book (for dodecane):
Vapor sound speed (m/s)
134.59
Liquid sound speed (m/s)
1037.8
Liquid saturation density (kg/m3)
697.13
Vapor density (kg/m3)
Saturation pressure (Pa)
12622
Liquid viscosity (Pa.s)
5.6 e-4
Vapor viscosity (Pa.s)
5.44 e-6
( *1e-6)/(150-6)=0.0415
mu = 2.5e-3 kg/m/s
Re C = 4*10.26/PI()/0.02^2*0.02/0.025=26k
Re D = 4*11.67/PI()/0.0186^2*0.0186/0.025=32k
Cav
0.042 Re
26k/32k 41 41

42. Details of mesh preparation

43. Meshing Institution Bosch New meshing tool by Bosch-Cascade
Flow Domain
Chamber:
45 mm Long
New meshing tool by Bosch-Cascade
Start from CAD surfaces
Seed domain with points
Build Voronoi diagram, connectivity
No sliver cells at boundaries
Face normals point to cell centers
Minimal cell skew
More ‘sampling’ than hexes
Institution
Bosch
Dimensionality 3
Cell Type
14-faced polyhedra
Cell count (total)
3x106
Voronoi Mesh

44. Institution CMT Institution SNL Dimensionality 2 Cell Type Quad
Cell count (total)
67.4K
Geometry
12x6 mm
Institution
SNL
Dimensionality 3
Cell Type
Cube
Cell count (total)
7x107 to 21x107
Geometry
1.7x1.7x15.3 mm
CMT: is it immersed flow?