F. Perini a, P. C. Miles b, R. D. Reitz a · Chamfered piston bowl shape DERC piston bowl shape. University of Wisconsin -- Engine Research Center ... -Area opposed to flow is not

University of Wisconsin -- Engine Research CenterERC Seminar – Madison, WI – Nov 19, 2013

slide 1

IMPORTANCE OF INTERNAL FLOW AND

GEOMETRY MODELLING IN THE GM 1.9L LIGHT DUTY ENGINE

F. Perinia, P. C. Milesb, R. D. Reitza

aUniversity of Wisconsin-MadisonbSandia National Laboratories

ERC Seminar – Madison, WI – Nov 19, 2013

Acknowledgements:

This research is funded by the Sandia National Laboratories


slide 2

Outline

Motivation and challenges

Capturing internal flows in the Sandia light-duty optical Diesel engine

Code development

Engine Geometry representation

Adjustable swirl-ratio modelling

Fluid flow validation vs. PIV measurements

(Preliminary) Capturing the effects of flow, composition and thermal non-uniformities on HCCI combustion


slide 3

Motivation

The success of advanced combustion strategies heavily relies on local mixture preparation

Engine sector simulations incorporate geometrical simplification and azimuthal averaging An average-of-the-average

Extremely accurate at predicting global engine behavior, but can fail when local phenomena are relevant

KIVA, -5.0 CA

P1,

y a

xis

[cm

]

-2 0 2

-2

0

2

P2,

y a

xis

[cm

]

-2 0 2

-2

0

2

x axis [cm]

P3,

y a

xis

[cm

]

-2 0 2

-2

0

2

Experiment, -5.0 CA

-2 0 2

-2

0

2

-2 0 2

-2

0

2

x axis [cm]

-2 0 2

-2

0

2

0

0.5

1

0

0.5

1

0

0.5

1

1.5

-15 -10 -5 0 5 10 15 200

10

20

30

40

AHRR, pinj

= 860 bar, Rs sweep

crank angle [degrees ATDC]

appa

rent

hea

t rel

ease

[J/

deg]

KIVA, Rs = 1.55Exp, Rs = 1.55KIVA, Rs = 2.20Exp, Rs = 2.20KIVA, Rs = 3.50Exp, Rs = 3.50

Global performance

Local mixture formation


slide 4

Motivation

The success of advanced combustion strategies heavily relies on local mixture preparation

Engine sector simulations incorporate geometrical simplification and azimuthal averaging An average-of-the-average

Extremely accurate at predicting global engine behavior, but can fail when local phenomena are relevant

Can detailed engine modeling improve the simulation’s predictivenessand

provide a computational counterpart to the extensive set ofexperimental measurements carried out on this engine?


slide 5

Challenges

Even with an adaptive meshing methodology and local cell refinements only where needed, the total grid size for a complete engine facility easily adds up to > 500k cells

Code parallelization needed for practical simulations on multi-core computers

Solver numerics (Jacobi-preconditioned CR method) are outdated • Unmanageably large number of iterations per time step

• Convergence is not always guaranteed (e.g. at the valve openings)

Reaction mechanisms for multiple and multi-component fuels are quickly increasing in size, thanks to advanced chemistry solvers (e.g., SpeedCHEM)

The same number of species has to be advected by the fluid flow solver!


slide 6

Workflow

Build a reliable model for the Sandia-GM 1.9L light-duty Diesel engine to explore advanced combustion concepts

RANS approach is optimal for reproducing ensemble-averaged experimental measurements

Implementation of state-of-the-art numerics, spray and chemistry models on the remains of the KIVA solver

To do so, we need add up to the model the following bricks, in this order:

1. [Geometry] Accurate optical engine geometry

2. [CFD] Validate fluid flow predictions

3. [Numerics] Accurate, fast reactive flow solvers

4. [Combustion] Validate vs. HCCI ignition experiments

5. [Spray] Validate vs. local mixture formation Use the predictive tool to explore further combustion strategies

Thispres.


slide 7

Engine geometry modelling

Detailed combustion chamber

Intake and exhaust runners

Pressure vessels

605k cells at BDC305k cells at TDC

Improved wall boundary treatment


slide 8




Pressure vessels



Refined unstructured mesh- Crevice and near-liner region- Valve seats and stems


slide 9




Pressure vessels



- Modified KIVA code for multi-layered valves- arbitrary # of cell layers at the intake for discharge coefficient capturing


slide 10




Pressure vessels



Recesses

Non-tuned crevice height

Chamfered piston bowl shape DERC piston bowl shape


slide 11

Initial and boundary conditions

CylinderComposition: measured,

exhaustp, T: from pressure trace

Intake regionComposition: arbitrary fresh air + measured EGR comp

p, T: from intake transducers

Exhaust regionComposition: measured,

exhaustp, T: from transducers

InjectionsActual timing, duration, injected mass and fuel composition from the

bench data


slide 12

Swirl generation modelling

Different swirl ratios are obtained by throttling the intake ports

Adjustable throttles are mounted on the intake ports

High swirl: Tangential port open, helical port throttled

Low swirl: Tangential port throttled (7), helical

port throttled (15)

Modeled using a layer of cells,

deactivated and with their faces set

as a solid wall

HT

HelicalTangential


slide 13

Intake throttle generation

Cells are identified, rotated, accordioned and deactivated

From streamwise cells

From cross-sectional layerTangential pin = 19, Helical pin = 5

T

H

Example

Tangential pin = 19, Helical pin = 15


slide 14





T

H

Example


This model is not perfect!- No throttle stem- At least one cell layer per side - Area opposed to flow is not exactly the correct one when cross-sectional layer is used


slide 15





T

H

Example


This model is not perfect!- No throttle stem- At least one cell layer per side - Area opposed to flow is not exactly the correct one when cross-sectional layer is used

However!

Achieving a more accurate geometry would pose significant modelling problems on a hexahedral mesh


slide 16


0 20 40 60 80 100-1

0

1

2

3

4

tangential port throttle angle [deg]

max

sw

irl r

atio

dur

ing

inta

ke [

-]

Helical pin = 11Helical pin = 15Helical pin = 11KIVA 11KIVA 15KIVA 19

0 20 40 60 80 1001

2

3

4

5

6

7

helical port throttle angle [deg]

max

sw

irl r

atio

dur

ing

inta

ke [

-]

UW ProductionUW single cylinderGM productionKIVA

Helical port throttlingHelical port throttling Tangential port throttlingTangential port throttling

* Measurements from R. Opat, Master Thesis, UW-Madison, 2006

Max swirl ratio during intake is the closest configuration to the swirl meter bench

helical = open

helical = pin 11

helical = pin 15


slide 17


0 20 40 60 80 100-1

0

1

2

3

4


max

sw

irl r

atio

dur

ing

inta

ke [

-]


0 20 40 60 80 1001

2

3

4

5

6

7


max

sw

irl r

atio

dur

ing

inta

ke [

-]


Helical port throttlingHelical port throttling Tangential port throttlingTangential port throttling



helical = open

helical = pin 11

helical = pin 15


slide 18


0 20 40 60 80 100-1

0

1

2

3

4


max

sw

irl r

atio

dur

ing

inta

ke [

-]


0 20 40 60 80 1001

2

3

4

5

6

7


max

sw

irl r

atio

dur

ing

inta

ke [

-]


Helical port throttlingHelical port throttling

Match deteriorates when the throttle model lacks of resolution: 1) Throttle almost closed2) Throttle angle between 45 and 70 degrees

Tangential port throttlingTangential port throttling



helical = open

helical = pin 11

helical = pin 15


slide 19

Tangential velocities 3mm below fire-deck vs. PIV (Petersen et al, 2011)

Fully open (“Rs = 2.2”), different crank angles

CA = 35° bTDC, different swirl ratios

0 1 2 3 4 50

2

4

6

8

10

12

14

radius [cm] from cylinder axis

tang

entia

l vel

ocity

[m/s

]

CA = -50.0 CA = -35.0 CA = -25.0

(marks) experiments from [XYZ](lines) KIVA, full mesh


0 1 2 3 4 50

2

4

6

8

10

12

14


tang

entia

l vel

ocit

y [m

/s]

KIVA-4

CA = -50.0 CA = -35.0 CA = -25.0

(marks) experiments from [XYZ](lines) KIVA, Bessel fit α = 2.20

0 1 2 3 4 50

5

10

15

20

25


tang

entia

l vel

ocity

[m/s

]

Rs = 2.2 Rs = 3.5 Rs = 4.5


0 1 2 3 4 50

5

10

15

20

25


tang

entia

l vel

ocit

y [m

/s]

Rs = 2.20 Rs = 3.50 Rs = 4.50

(marks) experiments from [25](lines) KIVA, Bessel fit α = 2.20

FULLSECTOR


slide 20




0 1 2 3 4 50

2

4

6

8

10

12

14


tang

entia

l vel

ocity

[m/s

]

CA = -50.0 CA = -35.0 CA = -25.0



0 1 2 3 4 50

2

4

6

8

10

12

14


tang

entia

l vel

ocit

y [m

/s]

KIVA-4

CA = -50.0 CA = -35.0 CA = -25.0


0 1 2 3 4 50

5

10

15

20

25


tang

entia

l vel

ocity

[m/s

]

Rs = 2.2 Rs = 3.5 Rs = 4.5


0 1 2 3 4 50

5

10

15

20

25


tang

entia

l vel

ocit

y [m

/s]

Rs = 2.20 Rs = 3.50 Rs = 4.50


CA↑↑↑↑ CA↑↑↑↑

Rs↑↑↑↑Rs↑↑↑↑

FULLSECTOR


slide 21




0 1 2 3 4 50

2

4

6

8

10

12

14


tang

entia

l vel

ocity

[m/s

]

CA = -50.0 CA = -35.0 CA = -25.0



0 1 2 3 4 50

2

4

6

8

10

12

14


tang

entia

l vel

ocit

y [m

/s]

KIVA-4

CA = -50.0 CA = -35.0 CA = -25.0


0 1 2 3 4 50

5

10

15

20

25


tang

entia

l vel

ocity

[m/s

]

Rs = 2.2 Rs = 3.5 Rs = 4.5


0 1 2 3 4 50

5

10

15

20

25


tang

entia

l vel

ocit

y [m

/s]

Rs = 2.20 Rs = 3.50 Rs = 4.50


CA↑↑↑↑ CA↑↑↑↑

Rs↑↑↑↑Rs↑↑↑↑

Swirl, bench Sector @ IVC Full @ IVC

Rs = 2.2 2.2 2.09

Rs = 3.5 3.5 3.76

Rs = 4.5 4.5 4.80

FULLSECTOR


slide 22

Effect of throttling strategy on flow field

- Vertical cross sections at the intake valves CA = 540


slide 23

Smaller in-bowl velocities when having intake-generated swirl smaller overall momentum than

the IVC, Bessel fit imposed swirl profile

1 2

In the spray jet targeting zone, the high velocity region appears to be set by the valve recesses on the head

Full vs. sector mesh velocities

Sector

Full meshRs = 2.2throttles

fully open * After full induction stroke calculation


slide 24

Smaller in-bowl velocities when having intake-generated swirl smaller overall momentum than

the IVC, Bessel fit imposed swirl profile

1 2

In the spray jet targeting zone, the high velocity region appears to be set by the valve recesses on the head

Full vs. sector mesh velocities

Sector

Full meshRs = 2.2throttles

fully open * After full induction stroke calculation

KIVA, -5.0 CA

P1,

y a

xis

[cm

]

-2 0 2

-2

0

2

P2,

y a

xis

[cm

]

-2 0 2

-2

0

2

x axis [cm]P

3, y

axi

s [c

m]

-2 0 2

-2

0

2

Experiment, -5.0 CA

-2 0 2

-2

0

2

-2 0 2

-2

0

2

x axis [cm]

-2 0 2

-2

0

2

0

0.5

1

0

0.5

1

0

0.5

1

1.5

The sector simulations are:

-Overpredicting jet deflection in the central part of the cylinder

- Underpredicting jet penetration deep intothe bowl


slide 25

Full vs. sector mesh average flow properties

400 450 500 550 600 650 7000

1

2

3

4

5

6

7predicted swirl comparison


swir

l rat

io [

-]

Rs = 1.5 Rs = 2.2 Rs = 3.5 Rs = 4.5

(solid) full mesh, (dashed+marks) sector mesh

- A significant amount of calibration is needed to match near-TDC velocities with a sector simulation

- Imposed momentum conservation- Swirl vortex axisymmetry- Absence of geometric details in-cylinder

- Even after calibration, prediction of in-cylinder turbulent kinetic energy and dissipation rate appear drastically underestimated

600 650 700 750 8000

0.2

0.4

0.6

0.8

1.0


turb

ulen

t di

ssip

atio

n [m

2 /s3 ]

sector simulation

full engine geometry

x 104

600 650 700 750 8000

2

4

6

8

10

12

14


turb

ulen

t ki

neti

c en

ergy

[m

2 /s2 ]

sector simulationfull engine geometry

600 650 700 750 8000

2

4

6

8

10

12


turb

ulen

t le

ngth

sca

le [

mm

]

sector simulationfull engine geometry

T [m2/s2] TLT [mm] TεεεεT [m2/s3]

full geometry sector

fullsector


slide 26

Swirl center identification The swirling vortex in the engine shows precession during

compression, and vertical tilting

Vortex identification

(Michard et al., 1997)

-10 -8 -6 -4 -2 0 2 4 6-2

0

2

4

6

8

10

12

14

x [mm]

INT

AK

E

←← ←←

y [m

m]

→→ →→ E

XH

AU

ST

Swirl center precession, CA = -50, -40, -25 aTDC

Rs = 2.2

Rs = 3.5

Rs = 4.5

CA ↑

(solid) KIVA(dashed) exp

-15 -10 -5 0 5 10 15-15

-10

-5

0

5

10

15Swirl center tilt [3, 10, 18 mm below firedeck]

x [mm]

INT

AK

E ←← ←←

y

[mm

]

→→ →→ E

XH

AU

ST

Rs = 2.2

Rs = 3.5

Rs = 4.5


-5

0

5 radial [cm/s]

-5

0

5tangential [cm/s]

-5 0 5-5

0

5 vertical [cm/s]

exhaust ←←←← x [cm] →→→→ intake

-1500 0 1500

precession tilting

( ) ( )Ω

Ω∈

⋅

⋅∧=Γ ∫MM

M dSvPM

zvPM

SP r

rˆ1

maxmax

CA↑z↓↓↓↓


slide 27

Swirl center identification The swirling vortex in the engine shows precession during

compression, and vertical tilting

Vortex identification

(Michard et al., 1997)

-10 -8 -6 -4 -2 0 2 4 6-2

0

2

4

6

8

10

12

14

x [mm]

INT

AK

E

←← ←←

y [m

m]

→→ →→ E

XH

AU

ST

Swirl center precession, CA = -50, -40, -25 aTDC

Rs = 2.2

Rs = 3.5

Rs = 4.5

CA ↑


-15 -10 -5 0 5 10 15-15

-10

-5

0

5

10

15Swirl center tilt [3, 10, 18 mm below firedeck]

x [mm]

INT

AK

E ←← ←←

y

[mm

]

→→ →→ E

XH

AU

ST

Rs = 2.2

Rs = 3.5

Rs = 4.5


-5

0

5 radial [cm/s]

-5

0

5tangential [cm/s]

-5 0 5-5

0

5 vertical [cm/s]

exhaust ←←←← x [cm] →→→→ intake

-1500 0 1500

precession tilting

( ) ( )Ω

Ω∈

⋅

⋅∧=Γ ∫MM

M dSvPM

zvPM

SP r

rˆ1

maxmax

Velocities 3mm below firedeckreflect close presence of the valve

regions

CA↑

“Fully open”, Rs = 2.2 configuration (blue) captures values well, not only trends

z↓↓↓↓


slide 28

In-cylinder temperature stratificationRs = 2.2, IVC INTAKE EXHAUST Cross section

Rs = 2.2, TDC Cross section Temperature stratification is significant (> 30K)

at IVC highest temperatures within the

bowl, clue to less efficient removal of the

exhaust gases

Some temperature stratification (∼ 10K) survives within the bowl

even until the end of the compression stroke may be greater at lower swirl ratios

! Crucial for HCCI combustion and reaction mechanism validation


slide 29

On-going work

1) HCCI operation is achieved through dual port fuel injection

• A common-rail injector injects small amounts of n-heptane at low pressure

• A PFI injector injects iso-octane

Complete full-cycle HCCI

simulations with comprehensive flow, fuel injection, and combustion modelling

Tangential portBosch CRIP2.2

Helical portTFS 89055-1

2) KIVA solver improvement and parallelization- Mesh movement with automatic re-partitioning

using METIS- Replacement of the CR solver with a specific

accurate and fast solver for simulations with many species


slide 30

Conclusions The comprehensive engine model captures intake flows reasonably

well

Significant cold flow deviations are observed when comparing the full model with the sector mesh representation

Development of advanced numerics is preparing the path towards ‘real-world’ full engine simulations with detailed chemistry and spray

Future work

Understand the effects of detailed geometry and flow non-uniformities (temperature and composition too!) on HCCI Help to quantify initialization uncertainties in sector mesh simulations

Capture the effects of detailed geometry on spray jet-by-jet discrepancies and local mixture preparation


slide 31

Thanks for your attention!Questions?

Acknowledgements• U.S. D.O.E., Sandia National Laboratories • Paul C. Miles, Rolf D. Reitz• Dipankar Sahoo – equivalence ratio measurements• Adam B. Dempsey, N. Ryan Walker – model development and experiments on the DERC engine• Randy Hessel, Joshua Leach – computing infrastructure access and setup

F. Perini a, P. C. Miles b, R. D. Reitz a · Chamfered piston bowl shape DERC piston bowl shape. University of Wisconsin -- Engine Research Center ... -Area opposed to flow is not

Documents