Content uploaded by Marco Moser
Author content
All content in this area was uploaded by Marco Moser on Jun 16, 2022
Content may be subject to copyright.
IAV 10/2021 Marco Moser -Holistic OBM/OBD/Controls
Holistic Approach for OBM,
OBD and Controls for Diesel
Emissions
Marco Moser, IAV, Berlin
Model PN
Model CH2O
Motivation
Challenges … Possible EU7 Exhaust Treatment OBM
IAV 10/2021 Marco Moser -Holistic OBM/OBD/Controls
2
Model HC
Model CO
Engine DOC DPF SCR SCR ASC
NOx
Model PM
Model NOX
TTCH4
HC
CO
NOX
NH3
N2O
dp
PM
CH2O
NOx TNOx
PM T
PN ?
Sensor-based emission monitoring difficult due to missing and none continuous measuring sensors
Potential: sensor-adapted model-based approach
OBD limits for EU7 and OBM tolerances so far not defined OBM tolerances main focus and enabler
NH3NH3
Model CH4
no HC sensor
no CO sensor
PM sensor not
continuous
NOXsensor readiness and
cross sensitivity
no CH4sensor
no CH2O sensor
currently not all
models available
new
new
new
new
new
Path to model
proposals for
EU7 / OBM
particles from
tires / brakes
Content
IAV 10/2021 Marco Moser -Holistic OBM/OBD/Controls
3
Engine
5. Nitrogen Oxides Model
4. Particulate Matter Model
1. Engine Out Emission Model
2. Engine Out Validation
3. Carbon Oxidation Model
6. Tailpipe Emission Validation
Emissions
Model PN
1. Engine Out Emission Models
Physico-Chemical / Machine Learning
IAV 10/2021 Marco Moser -Holistic OBM/OBD/Controls
4
Model HC Model CH2O
Model CO
Engine DOC DPF SCR SCR ASC
Model PM
Model NOX
Model CH4
Engine Output
1. NOx
2. CO
3. HC
4. Soot
5. T31
6. pmi
7. …
Excitation
1. nM
2. mFuel
3. p2
4. SOI
5. EGR
6. pRail
7. Swirl
DoE-Model
(Volterra or
Gauß etc.)
Map-based / Physico-Chemical Machine Learning
Expert Knowledge
Features
and / or
NH3NH3
CH4
HC
CO
NOX
NH3
N2O
PM
CH2O
PN
SciML –scientific machine
learning
physical loss function
physics-informed neural
network
1. Engine Out Emission Models
Machine Learning from Expert Knowledge
IAV 10/2021 Marco Moser -Holistic OBM/OBD/Controls
5
*.dat
*.dat
*.dat
*.dat
*.dat
*.dat
*.dat
*.dat
*.dat
*.dat
*.dat
*.dat
*.dat
*.dat
*.dat
*.dat
*.dat
*.dat
Proven in series: Machine learned emission models through engineering from physico chemical knowledge
Environment
Knowledge
Measurements
Data
Analysis
Emission
Models
Meta
Model
Cross
Correlation
Machine
Learning
Risk: Sensor tolerances
Humidity sensor
physico chemical models for
well-known correlations
Environment O2
from sensor or
high-class model
boosts accuracy
boosts accuracy
1. Engine Out Emission Models
Machine Learning – Examples (NOXengine out)
IAV 10/2021 Marco Moser -Holistic OBM/OBD/Controls
6
Higher accuracy compared to "state of the art" approaches < ± 20% @ 2σ (95,45%)
•for altitude: 0m … 4000m // ambient temperature: -30°C … +50°C
General approach, applicable for different emissions or sensors Proven in series application
Emission measurements on
Altitude-Climate-Test-Bench:
Environment variation:
altitude
temperature
humidity
WLTC
1. Engine Out Emission Models
Machine Learning – Examples (CO, Soot)
IAV 10/2021 Marco Moser -Holistic OBM/OBD/Controls
7
Machine learned emission models for all emissions possible (HC, CH4, CH2O, CO2, O2) accuracy requirements might be challenging
CO Soot
WLTC WLTC
dp
Model PN
2. Engine Out Validation
Sensors for Tolerance Detection
IAV 10/2021 Marco Moser -Holistic OBM/OBD/Controls
8
Model HC Model CH2O
Model CO
Engine DOC DPF SCR SCR ASC
NOx
Model PM
Model NOX
Model CH4
T
fault detection of engine
and emission adaption
Engine-out sensors are essential to detect engine specific tolerances and correct raw emission models
NH3NH3
CH4
HC
CO
NOX
NH3
N2O
PM
CH2O
PN
2. Engine Out Validation
Sensors as Fix Point for Model Correction
IAV 10/2021 Marco Moser -Holistic OBM/OBD/Controls
9
Emission []
λ[]
NOX
Soot
CO
HC
1 6
Cross correlation
•change in NOXchange in HC, CO, soot, CH4,
CH2O … to adapt models
Exact engine out models required
raw emission models created for norm case
tolerances of engine and surrounding
components not completely modelable
Reliable sensor layout
what is needed to detect tolerances and failures
NOXthe only available emission sensor to detect
engine failures but needs confirmation:
Alternatives
2 NOXsensors (or 3?)
NOXand PM (only if for high soot
ready) … dpDPF sufficient sensitivity?
NOXand lambda
lambda probe with sensitivity
to more than O2?
Model PN
3. Carbon Oxidation Model
Modeling Carbon Oxidation
IAV 10/2021 Marco Moser -Holistic OBM/OBD/Controls
10
Model HC Model CH2O
Model CO
Engine DOC DPF SCR SCR ASC
NOx
Model PM
Model NOX
Model CH4
TT
Conversion
HC,CO,CH4,CH2O
fault detection DOC
OBM adaption
TT
fault detection ASC
OBM adaption
Conversion
HC,CO,CH4,CH2O
NH3NH3
Accurate temperature sensor recommended for exothermic modeling carbon compound oxidation
CH4
HC
CO
NOX
NH3
N2O
PM
CH2O
PN
dp
3. Carbon Oxidation Model
Modeling Carbon Compound Oxidation
IAV 10/2021 Marco Moser -Holistic OBM/OBD/Controls
11
Adaptation of carbon compound models:
Extended Kalman Filter for parameter
estimation (e.g. estimation aging factor)
https://dieselnet.com/tech/catalyst_methane_oxidation.php
1D-Flow-through catalyst model
Measurement of exothermic
Technical knowledge
closed loop
different species different behavior
Model PN
4. Particulate Matter Model
Soot Mass Model
IAV 10/2021 Marco Moser -Holistic OBM/OBD/Controls
12
Model HC Model CH2O
Model CO
Engine DOC DPF SCR SCR ASC
NOx
Model PM
Model NOX
Model CH4
TTdp
fault detection DPF
OBM adaption
Conversion
PM,PN
TPM T
fault detection DPF
OBM adaption
NH3NH3
PM sensor detects defect DPF if soot is measured, DPF is defect
dp sensor for soot mass measurement HC/CO emissions during DPF regeneration
CH4
HC
CO
NOX
NH3
N2O
PM
CH2O
PN ?
4. Particulate Matter Model
Filtering Model
IAV 10/2021 Marco Moser -Holistic OBM/OBD/Controls
13
1D Filtering model simplified
Flow pattern: Flow through
Simplified pressure drop model
Simplified soot deposition model and axial distribution
Simplified soot reactivity
Adaption of filter efficiency
(PM and PN)
Measurement of
-pressure drop
-particle mass
Technical knowledge
•dp sensor for detecting defects and soot
cumulation
•PM sensor measures non-continuous
interval-based diagnostics
Model PN
5. Nitrogen Oxides Model
Modeling Nitrogen Oxides
IAV 10/2021 Marco Moser -Holistic OBM/OBD/Controls
14
Model HC Model CH2O
Model CO
Engine DOC DPF SCR SCR ASC
NOx
Model PM
Model NOX
Model CH4
TT
Conversion
NOX,NH3,N2O
dp NOx TPM TNOx
fault detection SCR
OBM adaption
NH3NH3
models required for all SCR components
CH4
HC
CO
NOX
NH3
N2O
PM
CH2O
PN
5. Nitrogen Oxides Model
SCR Modeling Approaches
IAV 10/2021 Marco Moser -Holistic OBM/OBD/Controls
15
Physical models, data driven models and smart combinations can be used for SCR modeling
Reaction Rates: Modeled Reactions:
=, exp /( ) 1 NH3adsorption
=, exp ,(1 )/()NH3desorption
=, exp /(
), Standard SCR
=, exp /(
), Fast SCR
=, exp /(
), Slow SCR
=, exp /(
),NH3 Oxidation
State Equations:
1)
,
=
2)
= 0
3)
= 0 =
, +
4)
= 0 = , 0.5
5)
= 0 = , 0.75 0.5
Low-Dimensional Physical Models
•models derived from first order physical and chemical principles
•all essential phenomena are modeled
•real-time capability is often achieved by 0D-modeling or quasi
1D-modeling (multi-brick models)
Gelbert, MTZ 2/2017
ML-Models
•pure machine learn models are derived from data only
•better models can be created with additional usage of expert
knowledge
•also combinations of physical and ML-models can be useful
ML-Model 1: Temperatures
•NARX model for gas and
wall temperatures in SCR
ML-Model 2: NH3-Filling
Level:
•NARX model for NH3-
Filling Level Estimate
ML-Model 3: Gas
Concentrations:
•FNN models for NO, NO2,
NH3,…
Inputs
Outputs
März, Emission Control
Science and
Technology 6/2020
Model PN
6. Tailpipe Emission Validation
Sensors as Fix Point
IAV 10/2021 Marco Moser -Holistic OBM/OBD/Controls
16
Model HC Model CH2O
Model CO
Engine DOC DPF SCR SCR ASC
NOx
Model PM
Model NOX
Model CH4
TT
Conversion
HC,CO,CH4,CH2O
Conversion
NOX,NH3,N2O
dp
Conversion
PM,PN
NOx TNOx
PM
fault detection DPF
OBM adaption
fault detection SCR
OBM adaption
Conversion
HC,CO,CH4,CH2O
T
fault detection ASC
OBM adaption
NH3NH3
at least 1 sensor has to be redundant as a fixed point NOXsensor most sensible candidate
CH4
HC
CO
NOX
NH3
N2O
PM
CH2O
PN
New sensor diagnose
-NOXtailpipe concentration very low
-Using of sensor internal controls to prove sensor
accuracy (not available at the moment)
6. Tailpipe Emission Validation
Sensors as Fix Point
IAV 10/2021 Marco Moser -Holistic OBM/OBD/Controls
17
NOXsensor tailpipe as fix point
Sensor redundancy
2 NOXsensors (or 3 ?)
NOXand NH3using cross correlation
NOXand Lambda using cross correlation
NOXDosimeter (add)
source: https://www.cpk-automotive.com/
high accuracy at low NOX
in development with
different manufacturers
Subtask Status Assessment
Engine out models:
•Accuracy depends on data availability
•external measurement equipment
Carbon: HC, CO, CH2O, CH4
•Accuracy depends on different
conversion rates for the different
species
Particle: PM, PN
•PN modeling very difficult
Nitrogen: NOx, NH3, N2O
•Modeling of the different species very
challenging
•NOXtailpipe sensor helps
Assessment
IAV 10/2021 Marco Moser -Holistic OBM/OBD/Controls
18
DOCEngine SCR/ASC
DPF-SCR
DOCEngine SCR/ASC
DPF-SCR
DOCEngine SCR/ASC
DPF-SCR
T T
PM
PN
dP PM
CH2O
CO HC
NOX
NH3
NOx NOx NOx
N2O
CH4
TT
DOCEngine SCR/ASC
DPF-SCR
Summary
IAV 10/2021 Marco Moser -Holistic OBM/OBD/Controls
19
Subsystem specific assessment of model, available sensors and closed-loop approach
•Modelling of emissions from engine to tailpipe required Different approaches available for engine and EAT components
ECU
Hardware
EAT Sys. 1
Engine
S
Model 1 with
Tolerance Adaption
S
States
Emission
Aging
Raw Emission
Models
EAT Sys. 2EAT Sys. 3
NOxNH3
HC N2O
PM CH4
CO
PN
HC
HO
SS
Model 2 with
Tolerance Adaption
Model 3 with
Tolerance Adaption
SSensor(s)
•Continuous adaption of emission models based on sensors Open-loop models cannot cover full range of vehicle lifecycle
•Closed-loop control and adaption algorithms Compensation of system tolerances or model inaccuracies
Contact
Marco Moser
IAV Gmb H
Carnotstrasse 1, 10587 BERLIN (GERMANY)
www.iav.com
Philipp Brinkmann
Dr. Gregor Gelbert
Steve Kipping
Patrick Stracke
Paul Tourlonias
IAV 10/2021 Marco Moser -Holistic OBM/OBD/Controls