simulating integrated engine-emissions-doc-dpf performance · soot emissions model zphysically...
TRANSCRIPT
Simulating IntegratedEngine-Emissions-DOC-DPF
Performance
Acknowledgements
David FosterDave Kapparos, Indranil Brahma, Kushal Narayanawamy,
Stephen England, Andrea StrzelecERC University of Wisconsin-Madison
Yongsheng HeGeneral Motors Research & Development
Gurpreet Singh and Kevin StorkDepartment of Energy: FreedomCAR and Vehicle Technologies Program
Christopher RutlandEngine Research Center
University of Wisconsin-Madison
2
System Models: Current Literature
He, GM R&D (SAE 2007-01-1138)
– Integrated in SimulinkPischlinger, et al., FEV (SAE 2007-01-1128)
– In-house integration environmentGuthenke, et al., DaimlerChrysler (SAE 2007-01-1117)
– GUI based in-house integration environmentPiscaglia, et al., Polt. Di Milano (SAE 2007-01-1133)
– Simulink– DPF model to test ECU model
Aftertreatment Devices
Engine and Emissions
Exhaust flowNumerical parameters
• Detailed physics basedAT models
• Primary use: evaluate AT configurations
• No engine or emissionsmodels- Driven by laboratory data
3
Simulink
ERC Work: Integrated System Level Model
Aftertreatment Devices
Engine and Emissions
Exhaust flowNumerical parametersControl connection
Combustion Models
Emission Models
Engine Flow Models
LNT, etc. Models
DOCModel
DPFModel
4
System Modeling Approach
Develop an integrated system level computer model– Capture interactions:
• Device – Device• Device - Engine
– Transient– Conventional and LTC (HCCI, PCCI, etc.) diesel combustion
Develop new component models as needed– Examples: LTC heat release, emissions, heat transfer
Incorporate existing models/modules– Examples: DPF, DOC, LNT, …
Integrate components as modules in an overall system level environment– Environment: Matlab-Simulink
• Also allows controllers– Develop modular approach for efficiency and ease-of-use
Validation– Component level: experimental data– System level: experimental data (when available) and “sense checks”
5
Objectives and Results
Explore steady state and transient scenarios– Examine and help explain device interactions– Suggest ‘guidelines’ of operation
• Based on model results and analysis
Recent results– LTC combustion under load transients– Effect of DPF loading and regeneration on engine
operation– Compare different DPF regeneration techniques– Model prevention and control of runaway DPF
regenerations
6
Diesel Engine Model
WAVE or GT-PowerIndustry standardIncludes:– Heat transfer models– Flow models– Turbocharger model (VGT)– Simple Combustion model
• Heat release rates from experimental pressure
• Could also use built in heat release models
• Calibration requiredCommunicate with Simulink using sensors and actuators
Example of a WAVE engine model with sensors and actuators
Exhaust Fuel Injector (Ahead of DOC)
exhaust
Intake Throttle
7
Traditional Diesel Engine Model - Disadvantages
Simple combustion (heat release) model– Prescribed function, calibrated with engine data– Simple spray models
Simple emissions models
Sufficient for conventional diesel operation but not for new LTC type combustion technologies
Need improved combustion / heat-release models and emissions models for current application
8
User Defined Cylinder Models
Fuel Injection
Energy, SpeciesConservation
CombustionVaporization
IVC EVO
Heat Transfer
Emissions In-Cylinder
Models
GT-Power Engine Model
9
Multi-Zone Combustion Model
Combustion chamber initialized with multiple zonesThe zones include sub-models for vaporization, chemical kinetics (CHEMKIN), heat transfer, energy/species conservation
Schematic representation of zones
IVC
EVO
2 31 54
10
-40 -30 -20 -10 0 10 20 30 40400
600
800
1000
1200
1400
1600
1800
2000
22005
4
3
2
1
Zone 1 Zone 2 Zone 3 Zone 4 Zone 5
Tem
pera
ture
[K]
Crank Angle [deg ATDC]
Tem
pera
ture
[K]
SOI = -128 ATDC; Injection Duration = 32.5CADStart of combustion predicted accuratelyCapture variations in combustion phasing due to thermal and charge stratifications: Useful for parametric LTC studiesLimitations: Requires specified fuel distribution (e.g. no spray)
No mixing between zones
Multi- Zone Model Validation
Temperature Distribution
-40 -30 -20 -10 0 10 20 30 400
10
20
30
40
50
60
70
80
Experiment Simulation
Cyl
inde
r Pre
ssur
e [b
ar]
Crank Angle [deg ATDC]
0
100
200
300
400
500
600
700
800
900
1000
1100
1200
1300
Hea
t Rel
ease
Rat
e [J
/deg
]
Cyl
inde
r Pre
ssur
e[ba
r]
HR
R [J
/deg
]
CAT 3401 SCOTE,821 rpm, 25% Load
11
CFD In-Cylinder Model
Accurate modeling of fuel injection, spray dynamics, mixing, vaporization, chemistry and emissions calculationsFine – coarse mapped grid
Detailed in-cylinder CFD model
(KIVA-Chemkin)
IVC
EVO
12
CFD In-Cylinder Model
Accurate modeling of fuel injection, spray dynamics, mixing, vaporization, chemistry and emissions calculationsFine – coarse mapped grid
IVC
EVO
13
-30 -20 -10 0 10 20 30 40 50 600.0
1.5
3.0
4.5
6.0
7.5
9.0
10.5
12.0
13.5
15.0
0
100
200
300
400
500
600
Model Validation
Simulation results agree very well for lower loadsCaptures cool flame ignition and main combustion for early injection operation
30% load, EGR = 72 %
-30 -20 -10 0 10 20 30 40 50 600.0
1.5
3.0
4.5
6.0
7.5
9.0
10.5
12.0
13.5
15.0
0
100
200
300
400
500
600
700
800
Cyl
inde
r Pre
ssur
e [M
P a]
Crank Angle [deg ATDC]
Experiment
Simulation
20% load, EGR = 71 %
Crank Angle [deg ATDC]
Cyl
inde
r Pre
ssur
e [M
P a]
HR
R [J
/ deg
]
Experiment
Simulation
SOI = -50 ATDC SOI = -50 ATDC
14
Run Times for Different Approaches
The modeling approaches can be used in multiple cycle, transient simulations
Single Zone user combustion model
4 minutes
5 zone external cylinder model 17 minutes
GT-Power with mapped grid approach in 2-D
~70 minutes
GT-Power with refined grid in 2-D (1052 cells at BDC)
~ 9 hours
GT-Power with mapped grid using 3-D sector grids
~90 minutes
GT-Power with refined 3-D sector grids (5460 cells at BDC)
~ 1 day
15
CA50 Controlled by Actuating IVC
Step transient induced in CA50Controller forces a delay in intake valve closure
IVCactuated
CA50 Controlled
Target CA50
Sensed CA50
Actuated IVC
CA
50 (
CA
D) IVC
(CA
D)
Time (s)
16
Dual Control – Control CA50 During Load Increase
Step change in load
IVC actuated to maintain combustion phasing
Target CA50
Sensed CA50
Actuated IVC
CA50 set to –6.6 deg
CA
50 (
CA
D)
IVC (C
AD
)
IVCactuated
Target IMEP
Sensed IMEP
Fuel Injected
IMEP
(bar
)
Time (s)
Fuel Injected (m
g/cycle)
IMEP Controlled
Fuelactuated
17
Diesel-PCCI Mode Change: Cooled EGR and IVC
Model transition: conventional – PCCI –conventionalConventional SOI:
–12 deg ATDCPCCI mode SOI:
IVC + 5CAD
IVC response for PCCI under different EGR conditions– IVC actuation during
the mode transition decreases with increase in EGR
IVC
[CA
D]
0 1 2 3 4-150
-145
-140
-135
-130
-125
-120
Time [s]
EGR 10%
EGR 27%
EGR 40%
Intake valve closes early with increase in cooled EGR
CA50 set to - 5.7 ATDCIMEP set to 5.3 bar
18
Emissions Models
Need accurate, fast emissions models to drive the aftertreatment device models
Focus: Simpler, faster engine combustion models (for now)Long run times for DPF filling
General approach:– Use good physically based phenomenological model as
starting point– Improve model using neural networks to replace model
coefficients
Models required– Soot– NOx– CO2– HC
completed
under development
19
Soot Emissions Model
Physically based neural network modelPhysical model: Bayer and Foster,SAE 2003-01-1070Phenomenological models for– Injection spray
• Predicts spray angle, liquid penetration, liftoff length, local equivalence ratio, temperatures, etc.
– Particulate formation and oxidation
Inputs are obtained from the engine model– Profiles of in-cylinder pressure, in-cylinder mean temperature,
mass flow rate of fuel through injector, heat release rate– Engine speed, percent EGR, global equivalence ratio
0.5 1.8sosf EE
RTs RTsf f so s
dm C m P e C m P edt
φ−−
= −
Small soot particles formedfrom rich initial combustion
Large soot particles in vortex head, mixing until eventually oxidized by diffusion flame.
Entrained Hot Air
20
Neural Net Soot Emissions Model
Brahma et. at., SAE 2005-01-1122
193 4( )ww wΦ+
50.5wP
6 7sf
f
Ew w
RTe−
⋅ +
20wfm
141.8wP
15 16so
o
E w wRTe−
⋅ +
Φ
Csf
mf
P
-Esf/RTf
CSO
ms
P
-Eso/RTo
YO2
6 72019 5
3 4
( )0.5( )
sf
f
Ew w
RTww wsf fw wx C m P e
−⋅ +
Φ+= ⋅ ⋅ ⋅ ⋅
15 1622 141.8
2
( )2
10 11( )So
os O
E w wRTw w
so sm Y Py C w w m e−
⋅ +
⋅ ⋅ ⋅= ⋅ +
23212 217 1 2 8 9 18( ) ( )s wwdm
w wx w x w y w y wdt
= + − + +⎡ ⎤⎣ ⎦ 2 2210 11( )s s
ww m wm+
Neural network weights added to the physical model
Neural network weights trained using experimental engine data for 8 modes of operation – Approximately 57% reduction in
error of predicted soot
Converted into an M-file– M-file version runs in 1/3 the
time of the original Simulinkversion
– Small time steps are no longer propagated to other component models
21
Soot Model and Time Steps
Each component / module in integrated system requires a certain time step, Δt
Overall system runs at ~2.5ms Δt (primarily set by engine model)
Soot requires 0.5 CA data ( ~ 0.05 ms Δt)– Model requires crank-angle resolved spray and heat release
profiles
Simulink manages time stepping– Probes models to determine required times– Adjusts overall simulation to smallest Δt in system
Soot model Δt requirement can severely limit overall system speed
22
Solutions for Managing Disparate Time Scales
Hide inner workings of model from Simulink– Use self-contained Matlab or Fortran/C routines– Use sub-cycling time integration inside routines– Use level 2 S-functions
• These can specify their own hit time
Use external file fordata transfer– Pass small Δt data
outside of Simulink
Implement periodic ‘triggering’– Skip calculation in model except every ‘n’ engine cycles– Use most recent solution until next calculation
Combustion Models
Emission Models
Engine Models
LNT, etc. Models
DOCModel
DPFModel
23
NOx Emissions Model
Physically based neural network model (Brahma and appendix of England et. al., SAE 2006-01-0263)Implemented as an M-fileInputs are obtained from the engine and soot modelsNOx prediction is based on maximum rate of formation
Trained using experimental engine data for 8 modes of operation 1 2 3 4 5 6 7 8
0.2
0.4
0.6
0.8
1
Operating Mode Number
Nor
mal
ized
ppm
NO
x
Model PredictionExperimental
Operating Mode Number
Nor
mal
ized
NO
x
Experimental NOx Compared to Model Prediction
1/ 222 2
_X 1 1/ 2
_
exp [O ] [N ]NO max Diffusion Flame
Diffusion Flame
KT
KT
⎡ ⎤⎛ ⎞−⎢ ⎥⎜ ⎟⎜ ⎟⎢ ⎥⎝ ⎠= ⎢ ⎥⎢ ⎥⎢ ⎥⎣ ⎦
24
DOC Model
Proprietary GM model (Bissett)– Includes kinetics for oxidation of CO, HC, and NO
to NO2
Implemented as a Fortran version– Faster than Simulink version
Fortran allows decoupling of integration time– DOC model can operate at large time steps (10
ms or longer)
25
0.0
0.5
1.0
1.5
2.0
2.5
3.0
3.5
0 50 100 150 200 250 300 350
ExperimentalModel
Modified version of the MTU 1-D model (based on work of Konstandopoulos and Johnson)
Fortran code linked to Simulink using an S-function (level 2)
Input and output modified for dynamic use in Simulink and time integration improved
Calibrated against experimental data
Darcy law pressure drop for the wall and soot layer
Device sub models describe:– Particle deposition inside wall
and on wall surface– Flow and temperature fields– Catalytic and thermal oxidation
of soot
DPF Model
Time [min]
Pre
ssur
e D
rop
[kP
a]
Comparison of Experimental DPF Pressure Drop and DPF Model Prediction
Deep Bed Filtration
Cake Filtration
Tran
sitio
n
26
Combined DPF-LNT
DPF
LNT
+
SNT
=
PlugExhaust gases in
Exhaust gases out(higher CO, CO2
levels)Soot ‘cake’ deposit
Leanstorage
NOx storage
Exhaust gases in
Exhaust gases out
Soot depositFlow chronology
NOx
NO
x
27
Controllers and Validation
Controllers implementedin Simulink for:– Engine load (fueling)– Engine speed– EGR– Turbo boost– DPF regeneration
The component models tested and validated individually
The integrated model is tested to verify proper component interaction– DPF loading and regeneration simulations to test the model
and component interaction– The simulation results are consistent with experimental
results presented by Singh et. al. (SAE 2006-01-0879)
Combustion Models
Emission Models
Engine Models
LNT, etc. Models
DOCModel
DPFModel
28
DPF Loading and RegenerationSimulate a light duty, turbo charged, CIDI engine– Common rail direct fuel injection– Variable geometry turbine
Regeneration: modes 3 and 5– DPF initially loaded to 3 [g/l] of soot and regeneration starts at 3.2 [g/l]– Mode 5: Fuel injected ahead of the DOC– Mode 3: Fuel injected ahead of the DOC with intake throttling assistance
Regeneration tests were done with mode switching (mode 5 to 2) during regeneration to investigate prevention/control of regeneration runaway
ModeNumber Load
EngineSpeed[RPM]
ApproximateFueling Rate
[kg/hr]
EGR [mass %]
2 Very Low 970 1.42 62.143 Low 1120 4.035 37.965 Moderate-Low 1520 9.365 19.158 Moderate 2320 18.85 10.50
29
DPF Regeneration (Mode 5)
Fuel injected in exhaust before DOCIncreases reactions in DOC and exhaust TControlled to achieve a desired DPF inlet temperature of 600 °C
0 10 20 30 40250
300
350
400
450
500
550
600
650
Engine OutDPF OutDOC Out
Time [min]
Exha
ust T
empe
ratu
re [
°C]
Exhaust Temperatures
DPF Regeneration
0 10 20 30 40250
260
270
280
290
300
Total BSFCEngine BSFC
Time [min]
BSF
C [
g/kW
-h]
Fuel Consumption
DPF Regeneration
30
DPF Regeneration (Mode 5)
Simulation results are consistent with the experimental results given by Singh et. al. (SAE 2006-01-0879)
0 10 20 30 400
5
10
15
20
25
0 10 20 30 40
2
4
6
8
10
12
14
Time [min] Time [min]
Mass of Soot Trapped in the DPF
Trap
ped
Soot
[g]
Pressure Drop Across DPF
DPF RegenerationPr
essu
re D
rop
[kPa
]
DPF Regeneration
Note long regeneration
time
31
Throttle Assisted Regeneration (Mode 3)
Engine exhaust temperatures too low for fuel injection before DOC (> 300 C required) Throttle engine intake air flow– Increases equivalence ratio and engine exhaust T (ΔT = 44 °C)
Higher exhaust T allows fuel injection in exhaust before DOC
0 10 20 30 40
240
260
280
300
320
340
Total BSFCEngine BSFC
BSF
C [
g/kW
-h]
Time [min]
DPF Regeneration
Fuel Consumption
0 10 20 30 400
100
200
300
400
500
600
700
Engine OutDPF OutDOC Out
Exha
ust T
empe
ratu
re [
°C]
Time [min]
DPF Regeneration
Exhaust Temperatures
32
Throttle Assisted Regeneration (Mode 3)
0 10 20 30 400
1
2
3
4
5
6
7
0 10 20 30 40
5
10
15
20
Pres
sure
Dro
p [k
Pa]
Time [min]
Trap
ped
Soot
[g]
Time [min]
DPF Regeneration
DPF Regeneration
Mass of Soot Trapped in the DPF
Pressure Drop Across DPF
Note long regeneration
time
33
Throttle Assisted Regeneration (Mode 3)
Engine out NOx increased significantly during DPF regenerationIntake throttling required reducing EGR to maintain sufficient O2 for given loadReduced EGR resulted in higher engine out NOx
0 10 20 30 4030
40
50
60
70
80
90
100
Time [min]
NO
x [p
pm]
Engine-Out NOx
DPF Regeneration
34
Good Device Gone Bad
35
Runaway Regeneration (Mode 5)
Regeneration by fuel injection before DOCTry to reduce regeneration time by injecting more fuelResults in runaway regeneration
0 1 2 3 40
200
400
600
800
1000
1200
Engine OutDPF OutDOC Out
0 1 2 3 4250
275
300
325
350
Total BSFCEngine BSFC
Time [min]
BSF
C [
g/kW
-h]
Fuel Consumption
Exha
ust T
empe
ratu
re [
°C]
Time [min]
Exhaust Temperatures
DOC fuel injection stop
Temperature continues to rise after fuel shutoff
36
Runaway Regeneration (Mode 5)
0 1 2 3 40
5
10
15
20
0 1 2 3 40
2
4
6
8
10
12
Time [min] Time [min]
Pres
sure
Dro
p [k
Pa]
Trap
ped
Soot
[g]
Mass of Soot Trapped in the DPF
Pressure Drop Across DPF
Note “runaway”regeneration
time
37
Preventing Runaway Regenerations
Use ΔT predictive capability in the regeneration controller via simple energy balance:
Very simple energy balance will get the DPF inlet temperature within 10 [°C] of the target value
Use PI feedback control to adjust the predicted exhaust fuel injection rate to obtain the target DPF inlet temperature
Feedback PI controller alone (without predictive capability) is not recommended
exh exh exhfuel
fuel
m Cp TmLHV
Δ=&
&
38
Regeneration During Mode 5-2 TransientRunaway regeneration occurs when DPF is very hot and:
– Exhaust flow rate is suddenly lowered in mode transitions– Excess O2 is available in exhaust– Consistent with results from Koltsakis, et al. (SAE 2007-01-1127)
Prevent runaway regeneration by intake throttling to reduce available O2 in exhaust
0 200 400 600 800 10000
200
400
600
800
1000
1200
InletMidpointOutlet
0 200 400 600 800 10000
200
400
600
800
1000
1200
InletMidpointOutlet
Time [s] Time [s]
DPF
Wal
l Tem
pera
ture
[°C
]
Regeneration: Runaway after Mode Switch
DPF
Wal
l Tem
pera
ture
s [°
C]
Prevent Runaway with Intake Throttling
Mode 5 Mode 2
Mode 2Mode 5
DPF Wall Temperature DPF Wall Temperature
Regeneration
39
Regeneration: Runaway after Mode Switch
Prevent Runaway with Air Injection into DPF
0 200 400 600 800 10000
200
400
600
800
1000
1200
InletMidpointOutlet
Regeneration During Mode 5-2 Transient
Alternative method to prevent runaway:– Air injection into exhaust
Cool air removes heat from soot oxidation reactions
0 200 400 600 800 10000
200
400
600
800
1000
1200
InletMidpointOutlet
Time [s] Time [s]
DPF
Wal
l Tem
pera
ture
[°C
]
DPF
Wal
l Tem
pera
ture
s [°
C]
Mode 5 Mode 2
Mode 2Mode 5
DPF Wall Temperature DPF Wall Temperature
40
Controlling Runaway Regenerations
The integrated model has been used to simulate how a runaway regeneration can be detected and controlled– Reactive control
Can the integrated model be used to detect unfavorable conditions before a runaway regeneration starts to prevent runaway proactively?– To investigate this possibility phase diagrams
have been generated
41
DPF Phase Diagrams
Phase Diagrams:– Max DPF wall T– Function of:
• Equivalence ratio (e.g. available O2)
• Exhaust flow rate
Example:– Mode 8 to mode 2
transient during regeneration
Conditions:– Initial DPF wall
temperature = 600 °C– Soot loading = 4 g/l– Inlet temperature =
162 °C (after switch to mode 2) Global Equivalence Ratio, φ
Exh
aust
Flo
w R
ate
[kg/
s]
Maximum DPF Wall Temperature [°C]Experienced By Time = 10 [s]
mode 2
mode 8
0.2 0.4 0.6 0.8 1
0.02
0.04
0.06
0.08
0.1
0.12
600
620
640
660
680
700
720
Global Equivalence Ratio, φ
Exh
aust
Flo
w R
ate
[kg/
s]
Maximum DPF Wall Temperature [°C]Experienced By Time = 15 [s]
mode 2
mode 8
0.2 0.4 0.6 0.8 1
0.02
0.04
0.06
0.08
0.1
0.12
600
620
640
660
680
700
720
Global Equivalence Ratio, φ
Exh
aust
Flo
w R
ate
[kg/
s]
Maximum DPF Wall Temperature [°C]Experienced By Time = 30 [s]
mode 2
mode 8
0.2 0.4 0.6 0.8 1
0.02
0.04
0.06
0.08
0.1
0.12
600
620
640
660
680
700
720
Global Equivalence Ratio, φ
Exh
aust
Flo
w R
ate
[kg/
s]
Maximum DPF Wall Temperature [°C]Experienced By Time = 60 [s]
mode 2
mode 8
0.2 0.4 0.6 0.8 1
0.02
0.04
0.06
0.08
0.1
0.12
600
620
640
660
680
700
720
Global Equivalence Ratio, φ
Exh
aust
Flo
w R
ate
[kg/
s]
Maximum DPF Wall Temperature [°C]Experienced By Time = 120 [s]
mode 2
mode 8
0.2 0.4 0.6 0.8 1
0.02
0.04
0.06
0.08
0.1
0.12
600
620
640
660
680
700
720
Global Equivalence Ratio, φ
Exh
aust
Flo
w R
ate
[kg/
s]
Maximum DPF Wall Temperature [°C]Experienced By Time = 240 [s]
mode 2
mode 8
0.2 0.4 0.6 0.8 1
0.02
0.04
0.06
0.08
0.1
0.12
600
620
640
660
680
700
720
safe region
unsafe region
42
Using the Phase Diagrams
Runaway regeneration occurs when DPF is very hot and:– Exhaust flow rate is low– Excess O2 is available
By using the phase diagram unsafe operating conditions can be identified and avoided by:– Intake throttling– Air injection into DPF inlet
Global Equivalence Ratio, φ
Exh
aust
Flo
w R
ate
[kg/
s]
Maximum DPF Wall Temperature [°C]Experienced By Time = 240 [s]
mode 2
mode 8
0.2 0.4 0.6 0.8 1
0.02
0.04
0.06
0.08
0.1
0.12
600
620
640
660
680
700
720
Intake Throttling
Regen
Mode Chan
geAir Injection
Mode 2
Mode 8
43
Summary
Integrated system model for diesel aftertreatment studies
Advanced engine combustion model for LTC studies
Accurate emissions models to drive the aftertreatment devices
Accurate, modular aftertreatment devices models
Implemented in Simulink for controls and numerical management
Integrated model used to explore DPF regeneration
– Phase maps to explore runaway regeneration recovery
44
Ongoing Activities
Development of additional submodels– CO and HC emissions models– Incorporation of an SCR device model
Continue improvements to integrated system– Improve computer run times– Improve modularity and ease-of-use
Continue to use the system model to study device interactions and appropriate operation for transient and regeneration scenarios– Develop ‘guidelines for operation’
45
Thank You