Advanced Nozzle Flow and Spray Modeling...•Near nozzle spray features are better predicted by the...

Copyright © 2014 Convergent Science Inc., All rights reserved.

Advanced Nozzle Flow and Spray Modeling

• Able to predict effects of needle wobble • Able to predict hole-to-hole variations

!

!

!

• Start of injection simulations show gas injection before liquid • End of injection simulations show dribble and ingested gas in the sac

!

!

!

• Improved near nozzle comparisons with x-ray data • Spray jet shock wave generation

End$of$Injec+on,

Liq.%Volume%Frac0on%

2200 µs

2250 µs

2300 µs

Dribble?%

3-4 weeks on 100 processors, ~20 million cells


Advanced Nozzle Flow Modeling

eccentricity (positive x-axis)

1

2

3

4

5x

y

0 500 1000 1500 20000

50

100

150

200

250

Nee

dle

mov

emen

t (µm

)

Time (µs)

lifteccentricity (modulus)

▪ Effect of Needle Side Motion on Hole-to-Hole Mass Flow Variation

*Battistoni et al., SAE 2014Collaboration with Argonne and Cummins


Advanced Spray Modeling

*Xue et al., SAE 2014

60 µm

30 µm

15 µm

X-ray measurements

Projected mass density

0.1 mm

2 mm

ECN Spray A

Collaboration with Argonne and Cummins


Advanced Spray Modeling

*Quan et al. SAE 2014

0.81 kg/m3 2.55 kg/m3 11.7 kg/m3

▪ Eulerian-Eulerian coupled with VOF ▪ Compressible liquid and gas ▪ Dynamic structure LES ▪ AMR

Collaboration with Sandia and Argonne


Advanced Turbulence ModelingRNG – 0.25 mm

The computational cost of grid-convergent LES (@ 0.0625 mm) is about 30-40 times compared to grid-convergent RANS (@ 0.25 mm)



Advanced Turbulence Modeling

Radial Distance (mm)

-10 -5 0 5 10

Axi

al V

eloc

ity (m

/s)

0

25

50

75

100

125

150

MeasuredInstantaneous LES (20 Realizations)


-10 -5 0 5 10

Axi

al V

eloc

ity (m

/s)

0

25

50

75

100

MeasuredEnsemble-Averaged LES 25 mm

25 mm 35 mm 45 mm


-10 -5 0 5 10

Axi

al V

eloc

ity (m

/s)

0

25

50

75

100



-10 -5 0 5 10

Axi

al V

eloc

ity (m

/s)

0

25

50

75

100


Sample Size

0 2 4 6 8 10 12 14 16 18 20

Per

cent

0

10

20

30

40

50

60

70

80

90

100

0.5 m/s1.0 m/s2.0 m/s5.0 m/s

Sample Size

0 2 4 6 8 10 12 14 16 18 20

Per

cent

0

10

20

30

40

50

60

70

80

90

100

0.5 m/s1.0 m/s2.0 m/s5.0 m/s

Sample Size

0 2 4 6 8 10 12 14 16 18 20

Per

cent

0

10

20

30

40

50

60

70

80

90

100

0.5 m/s1.0 m/s2.0 m/s5.0 m/s

*Senecal et al. ASME 2013


-10 -5 0 5 10

Axi

al V

eloc

ity (m

/s)

0

25

50

75

100

125

150

MeasuredPredicted Average, 20 InjectionsPredicted Average, 8 Injections

8 cycles are needed to match the 20-cycle average to within a tolerance of 2 m/s


• Each simulation requires 60 hours on 64 cores • Peak cell count: ~20 million


Grid Convergence

Time (ms)

0.0 0.2 0.4 0.6 0.8 1.0

Liqu

id P

enet

ratio

n (m

m)

0

10

20

30

40

50

60

70

80

Measureddx = 2.0 mmdx = 1.0 mmdx = 0.5 mmdx = 0.25 mmdx = 0.125 mmdx = 0.0625 mmdx = 0.03125 mm

•Great care has been taken to ensure grid convergence in the CONVERGE spray models, such that the results can be analyzed with confidence

•Cell sizes down to 30 microns have been analyzed (AMR)

•Allows for accuracy/run-time tradeoff*Senecal et al. ASME 2012

Grid-convergent AMR cell size of 0.25 mm for RANS

Up to 4 - 5 cells inside the nozzle diameter at finest resolution

O2 Concentration (%)

8 10 12 14 16 18 20 22

Flam

e Li

ft-of

f Len

gth

(mm

)

0

5

10

15

20

25

30

35

40

Measureddx = 1.0 mmdx = 0.5 mmdx = 0.25 mmdx = 0.125 mm

Collaboration with Argonne


Extending Grid Convergence to Engines

*Senecal et al. JEGTP 2014Collaboration with Sandia and Caterpillar

Cell size, dx (mm)

0.0 0.2 0.4 0.6 0.8 1.0

Flam

e Li

ft-O

ff Le

ngth

(mm

)

0

5

10

15

20

25

30

35

40

45

50

MeasuredPredicted, No Cell LimitPredicted, 300k Cell Limit




Measured CL of OH*



SOC (deg. ATDC)

-2 0 2 4 6

NO

x (p

pm)

0

50

100

150

200

250

300

350

400

MeasuredCONVERGE - Full Geometry

Crank Angle (deg. ATDC)

-40 -20 0 20 40 60

Pre

ssur

e (M

Pa)

0

2

4

6

8

10

12

14


-2SOC


-40 -20 0 20 40 60

Pre

ssur

e (M

Pa)

0

2

4

6

8

10

12

14


2SOC


-40 -20 0 20 40 60

Pre

ssur

e (M

Pa)

0

2

4

6

8

10

12

14


6SOC

OH* Chemiluminescence vs. OH iso-surface, 7.5 degrees after SOC

The grid-convergent methodology results in excellent agreement with global combustion behavior as well as flame lift-off length and flame location

2 million cells utilized



Collaborations with LLNL654 Species, 2827 Reactions Total Time (days)

SAGE 150

SAGE-MZ, Direct 60

SAGE-MZ, Iterative 8

SAGE-MZ-DMR, Iterative 2

GPU ???

Mechanisms with up to 1,000 species can now be simulated with CFD

Year

Tota

l Run

time

(Day

s)

0

50

100

150

200Diesel, 654 Species, 2827 Reactions

2010 2011 2012 2013

101 102 103 104

102

103

104

2-methyl-alkanes

before 2000 2000-2005 2005-2010 after 2010

biodiesel

(USC) JetSURFC8(ENSIC-CNRS) C8

C4 (LLNL)

(Konnov) CH4

C4 (LLNL)C2 (San Diego)

CH4 (Leeds)

MDC16

C14C12

C10

(USC) C4(USC) C2

PRF

C7

C7, C8 skeletal (Lu&Law)

DME (Curran)(Qin) C3GRI3.0

Nu

mbe

r of r

eact

ions

, I

Number of species, K

GRI1.2


Key Findings and Recommendations RANS Spray Simulations

LES Spray Simulations

Eulerian Spray Simulations

Nozzle Flow Simulations

•0.25 mm cell sizes provide reasonably grid converged results for non-evaporating, vaporizing, and combusting sprays

•0.125 mm or finer resolutions necessary (recommended to use 0.0625 mm) •Multi-realization averaging is necessary to mimic experimental data (at least 8 realizations need to be averaged)

•Near nozzle spray features are better predicted by the EE model •EE simulations require at least 15 micron resolution near the nozzle

•10-15 micron resolution is sufficient to capture bulk flow features •5 micron or finer may be needed to capture opening and closing transients •Needle wobble has a large influence on plume-to-plume variations


▪ Massively parallel simulations, CPUs and GPUs ▪ Improved chemical mechanisms and emissions models - predictive soot ▪ LES with grid-converged mesh resolution ▪ Eulerian spray modeling linked to nozzle flow ▪ Multiple cycles routinely ▪ Multiple cylinders routinely ▪ CHT routinely !

High Performance Computing is a significant enabler to helping us achieve these goals.

Future Directions - Engine CFD in 5 Years in CONVERGE

Advanced Nozzle Flow and Spray Modeling...•Near nozzle spray features are better predicted by the...

Documents

Transcript of Advanced Nozzle Flow and Spray Modeling...•Near nozzle spray features are better predicted by the...