OBM/OBD Anomaly Pinpointing for Euro 7

Abstract

OBM for Euro 7 needs high class emission models for cold start and a good anomaly pinpointing for OBD

Full text

OBM/OBD Anomaly Pinpointing
for Euro 7
Marco Moser, Jonas Köhne, Patrick Stracke, Philipp Brinkmann -03.2023

Motivation
IAV 03.2023 -OBM/OBD Anomalie Pinpointing for Euro 7 -Marco Moser2
Tampering includes modification
of hardware AND software
"utmost importance" of tampering
protection
Specific objective: “Improve
control of real-world emissions”
OBD through sensors "does not
detect accurately and timely"
Euro 6: "emissions are not
adequately monitored or limited
over entire lifetime"
source: COM_2022_586_1_EN_ACT_part1_v8 (3).pdf
Potential Euro7 OBM
requirements:
overall: detection of high
emitters (2.5x limits)
Emission monitoring: last trip,
last 1000km, lifetime
Sensor-based NOx and NH3
(other emissions with OBD)
Model-based if sensor not
ready
RDE with very few restrictions
Emission limits for NOx, CO,
THC, NMHC, PM, PN10, NH3

Content
IAV 03.2023 -OBM/OBD Anomalie Pinpointing for Euro 7 -Marco Moser3
Evolution from OBD to OBM
OBM challenges
Emission Modelling Approaches
Anomaly Pinpointing
Causal Chain
Residual Finder
Autoencoder
Summary

Evolution from OBD to OBM
IAV 03.2023 -OBM/OBD Anomalie Pinpointing for Euro 7 -Marco Moser4
OBD OBM
Component failure detection
Fast detection mandatory
Emission/system failure detection
Multiple failures/tolerances possible
Define a component for RTB 23°C
which exceeds limit
Find this component in RDE
environment
Exceedance of limits can happen
through a set of different tolerances
Find the cause in any RDE cycle
OBD and OBM require precise models to assess sensor information and monitor emissions
OBM monitors emissions from sensors
and models
OBM uses OBD monitors for reliable monitoring
OBD monitors operate independently and
rely on models to check for deviations

TWC GPF / TWC
GAS. ICE
Heat
Sensors for OBM
IAV 03.2023 -OBM/OBD Anomalie Pinpointing for Euro 7 -Marco Moser5
NOX
DOC / LNT SCR / DPF SCR / ASC
DIESEL ICE
NOxNOxNOx
Heat
AdBlue AdBlue PM
1. Raw Emi. Model Model
(incl. Ageing)
Cold Start 1Hz: mg/s
Trip: mg/km
2. NOx Sensor
DewP. NOX11Hz: mg/s
Trip: mg/km
3. DewP. NOX21Hz: mg/s
Trip: mg/km
NOx Sensor
Model
(incl. Ageing)
Model
(incl. Ageing)
Model
(incl. Ageing)
Model
(incl. Ageing)
Model
(incl. Ageing)
Modell
(incl. Ageing)
4. DewP. NOX31Hz: mg/s
Trip: mg/km
NOx Sensor
NH3
NH3
1Hz: mg/s
Trip: mg/km
NOx/ NH3Sensor
PM -
PM Sensor
Modelling (includes ageing) mainly during Cold Start
Accuracy modelling (NOXsensor and SCR model)
OBD for detection of high emitters
Necessary sensors for NOXand NH3available
Assessment of models during cold start required enhancement if necessary

Computation Speed
Degree of Accuracy
Mathematical space ECU capability
Physico-chemical space
Reaction
Kinetics
(Low-Dimensional)
Science
informed
Machine
Learning
Machine
Learning
Map-
Based
Reaction
Kinetics
(Micro-, Macro-kinetic)
Raw emission models
already in serial application
(specific emission predicted for
control or OBD purpose)
EAT Model reduction for
catalyst and sensor feasible
(Simplification of models done at IAV
0D/1D model)
Modelling Approaches
IAV 03.2023 -OBM/OBD Anomalie Pinpointing for Euro 7 -Marco Moser6
Physico-chemical approach
Raw emissions Machine learning || EAT components Physico-chemical
Machine learning approach

Raw Emission Modelling –Science-Informed Model
IAV 03.2023 -OBM/OBD Anomalie Pinpointing for Euro 7 -Marco Moser7
Features Layer 1
Layer 2
Outputs
Feature Selection
System knowledge for
physico-chemical
relevance
Data Preparation
Feature and emission
synchronization
Emission and signal
plausibility
Time-based gradients
HC
CO
NOx
Model Training
Fuel specific compatibility check of
emission ratios (“COMPA” equation
based on Brettschneider-λ)
Post Processing
Compensation of Model
inaccuracies (e.g. dynamic
amplification, low pass)
Knowledge of the physico-chemical process is essential for good machine learning models
HC
NOx
CO
Already good emission prediction for
initial multi-target model
High potential of further improvement
(e.g. RNN)
Applicable to all emissions (e.g. PM)
WLTC

Raw Emission Modelling –Feature Engineering for Modularization
IAV 03.2023 -OBM/OBD Anomalie Pinpointing for Euro 7 -Marco Moser8
Emission
Model
Emission
Model Time-Based
Gradients
Physico-chemical
Submodels Humidity
Fuel
Feature
Replacem.
Sensor vs.
Model
System-specific feature engineering required
ML models for humidity and fuel derivative Proof of Concept
Fuel Blend Density Detection
Model of Absolute Humidity

Catalyst Modelling –Low Dimensional Kinetics (Physico-Chemical)
IAV 03.2023 -OBM/OBD Anomalie Pinpointing for Euro 7 -Marco Moser9
Simplified equations, less reactions and quasi-stationary assumptions for low-dimensional model
EU6
EU6
EU6
(HC+ NO
x
)
Temp . [°C]
T
out
Sim
Temp in
T
out
Exp
CO [mg/km]
CO
in
CO
out
Exp
CO
out
Sim
HC [mg/km]
HC
in
HC
out
Exp
HC
out
Sim
NO
x
[mg/km]
time [s]
0
400
800
1200
1600
NO
x, in
NO
x, out
Exp
NO
x, out
Sim
Calibration of parameters
•Chemical reactions are calibrated
•Reaction rates are defined by an Arrhenius-
type equation
calibration of
•factor
Ai
•activation energy
Ei
   =
 


•e.g., for CO Oxidation: CO + O2 CO2
R =
 
 [CO][O2] /
G( )
TWC

Content
IAV 03.2023 -OBM/OBD Anomalie Pinpointing for Euro 7 -Marco Moser10
Evolution from OBD to OBM
OBM challenges
Emission Modelling Approaches
Anomaly Pinpointing
Causal Chain
Residual Finder
Autoencoder
Summary

Pinpointing -Concepts
IAV 03.2023 -OBM/OBD Anomalie Pinpointing for Euro 7 -Marco Moser11
System Complexity
Modular OBD
Global understanding of interrelationships
Special knowledge of every OBD monitor
Expert Knowledge of Subsystems
Residual Finder
Understanding of interrelationships
System behavior with cascading switches of
model vs. sensor
Autoencoder
Understanding of boundaries of the system
Classic OBD
EU7
EU6

Pinpointing 1: Modular OBD -Causal Chain
IAV 03.2023 -OBM/OBD Anomalie Pinpointing for Euro 7 -Marco Moser12
System diagnosis based on collected component informations
All existing information are used to generate the decision
The coordinator gives a reasoned result to the error logic, not the fastest one
results on hold t
OBD symptom A (e.g. AFM)
OBD symptom B (e.g. Injection)
OBD symptom C (e.g. T sensor)
OBD symptom D (e.g. EGR)
Diagnosis with higher quality: every
diagnosis can use/observe the
whole drive-cycle, no inhibition
Depending results are released when
all involved diagnosis got a result
with the possibility to
discard them
OBD symptom E (e.g. NOxsensor)

Pinpointing 2: Residual Detector
IAV 03.2023 -OBM/OBD Anomalie Pinpointing for Euro 7 -Marco Moser13
Injector model (A) Lambda model (B)
Sensor
Model
Model Residual
Fingerprint of a fault
property rights: DE 10 2020 100 158.4 // DE 10 2021 127 196.7
Switching model inputs
between sensor / model

Pinpointing 3: Autoencoder Overview
IAV 03.2023 -OBM/OBD Anomalie Pinpointing for Euro 7 -Marco Moser14
1. Autoencoder trained to
reproduce the „nominal“
system behavior
2. Real time data feed
the autoencoder
3. Output is
reconstruction of inputs
4. Deviations of
input to output
5. Pinpointing from
analyzing anomalies with
appropriate metric
(classification)
Knowledge of nominal system and classification of anomalies
Main applications for
autoencoders are
anomaly detection and
feature extraction.

Pinpointing 3: Autoencoder
IAV 03.2023 -OBM/OBD Anomalie Pinpointing for Euro 7 -Marco Moser15
Output of Autoencoder is deviation between signal and reconstruction
https://www.assemblyai.com/blog/introduction-to-
variational-autoencoders-using-keras/
What is an autoencoder?
Feedforward artificial neural network
Persists of an encoder and a decoder
Deviations
input to output

Pinpointing 3: Autoencoder Pinpointing
IAV 03.2023 -OBM/OBD Anomalie Pinpointing for Euro 7 -Marco Moser16
Autoencoder computes a
global health monitor
Pinpointing to faulty
signals
Logic or Classification Algorithm for Pinpointing Metric

Summary
IAV 03.2023 -OBM/OBD Anomalie Pinpointing for Euro 7 -Marco Moser17
Evolution of Monitoring
OBD
Specific monitors for
defect components
OBD
OBM
Global pattern recognition in
signals for complex systems
e.g. Autoencoder
Pattern recognition in existing
OBD monitors with enhanced
models
e.g. Residual detector

Contact
Marco Moser
IAV GmbH
Carnotstrasse 1, 10587 Berlin (GERMANY)
marco.moser@iav.de
www.iav.com
Jonas Köhne, jonas.koehne@iav.de
Patrick Stracke, patrick.stracke@iav.de
Philipp Brinkmann, philipp.brinkmann@iav.de
Concept Cars: Light-duty gasoline and diesel demonstrators
IAV/AECC/IPA Gasoline Zero-Impact Demonstrator IAV Diesel Zero-Impact Demonstrator