1. Institute for Internal Combustion Engines and Thermodynamics
Analysis of options for a transfer function for emission correction between moderate and extended driving conditions Overview on the status of work at TU Graz (based on work plan proposed at ) S. Hausberger, S. Lipp, M. Grössing
Based on the ppt file from here Analysis on 4 vehicles is shown to elaborate the design of a transfer function. EMROAD results not yet available for 3 of 4 vehicles.
Stefan Hausberger
RDE test procedure 1

2. Idea of a “Transfer Function”
Working hypothesis: a “Transfer Function” in addition to the existing normalisation tools could reduce variabilities in emission results due to effects not covered yet.
Several options for the method exist, below one of them is shown as schematic example.
Since WLTP conditions are defined exactly, a function is elaborated, which shows changes compared to WLTP conditions. The function can be translated to any other centre point.
“Severity” = function of: *95 percentile of v x apos, *Altitude to be overcome (m/km), *Weight of loading, *Ambient temperature, * Maximum velocity * Average velocity * Engine speed
Parameters can be eliminated from the equation later easily.

3. Options for Transfer Functions
The change of emissions (NOx at the moment) is analysed as function of change of the “severity parameters”.
Change in the CF can be computed from the change in emissions on demand by dividing the result by the corresponding emission limit (e.g. 40mg/km effect on NOx means 40/80 = 0.5 change in CF).

4. Parameters to be considered in a transfer function
Statistical analysis showed later gave same statistical quality for regression when differences or ratios are used as parameters. Differences are selected for all parameters since “difference” on x-axis remains the same for any new center point while the ratio depends on the denominator used (WLTP values versus corner values of “normal driving”). From physical point of view also differences are reasonable (e.g. [gNOx /kg loading)
 transfer function looks as follows
𝐶 𝑡𝑓 = 𝑁𝑂𝑥 𝑡𝑟𝑖𝑝 − 𝑁𝑂𝑥 𝑊𝐿𝑇𝑃 = 1 𝑛 [ 𝐴 𝑖 × 𝑃 𝑖, 𝑡𝑟𝑖𝑝 − 𝑃 𝑖, 𝑊𝐿𝑇𝑃 𝑖 ]
With Ai....Constant coefficients to be elaborated
Pi....Value of severity parameter i in trip - Value of severity parameter i in WLTC Ctf... Emission value [g/km] in trip - Emission value [g/km] in WLTC
Severity Parameters Pi relevant for NOx:
“Severity” = function of: *95 percentile of v x apos, *Altitude to be overcome (m/km), *Weight of loading, *Ambient temperature, * Maximum velocity * Average velocity * Engine speed
Parameters can be eliminated from the equation later easily.

5. Actual status (06.10.2015) Simulation:
Raw exhaust gas map for NOx prepared
Model for transient effects on raw NOx prepared (NOx change as function of O2 concentration which is simulated by boost pressure. Boost pressure simulated as function of dynamic load change)
After-treatment model prepared with NOx conversion as function of space velocity and temperature. Available for SCR and SCR&NSC.
Measurement:
Analysis of 4 vehicles finalised (overview given on following slides) with exception of EMROAD results (EMROAD evaluation done for one vehicle only so far)
PEMS data for one additional vehicle available, not yet included.
Source
# trips
Vehicle
NOx in WLTP [mg/km]
Technology
Emission class (tbd)
TUG 28
C-segment
264
EGR + NSC
EU 6
Audi 3
SUV 90 ?
EU 6c Stage 1
Bosch 4
C segment 38
EU 6c Stage 2 VW 5
D segment
n.a.
Daimler
Analysis ongoing 
Figures use a = (v(i+1) – v(i-1) )/(2x3.6) in all cases

6. Actual status ( )
Measurement: overview on dynamics in the trips from the 4 vehicles analysed later
*….PEMS avg 18 vehicles is data shown March 2015 from data collection PEMS tests **..TNO data 95 percentiles read from a TNO diagram
Figures use a = (v(i+1) – v(i-1) )/(2x3.6) in all cases

7. Results vehicle 1 (TUG)
28 trips with variability in all relevant severity parameters.
Basic evaluation (Measured, CLEAR, EMROAD).
Note: EMROAD evaluation shall be checked by JRC. In actual results EMROAD increases the standard deviation compared to direct measured values, CLEAR reduces standard deviation by 44%
Measured
CLEAR
EMROAD
Avg. NOx [mg/km]
455
357
525
Std dev. NOx [mg/km]
230
129
266
Ries route agg. in Winter, 95P( v*a pos) = 28, Ta = 4°C
NEDC hot
PEMS Graz standard route summer
Ries route standard Summer 95P( v*a pos) = 12, Ta = 26°C
WLTP hot

8. Example SPSS result, stepwise adding of most significant variables
Transferfunction (TUG)
Steps to set up transfer function: SPSS software used to run multiple regression.
Dependent variable: difference or ratio of NOx in each trip vs. NOx in WLTP (1)
SPSS looks for best fitting equation which explains dependent variable by function of other variables. (1)…. Any other trip can be used as reference if “differences” are analysed to get same slopes
Example input data
Example SPSS result, stepwise adding of most significant variables
R²=0.64
R²=0.83
R²=0.87
R²=0.89

9. Results vehicle 1 (TUG)
Transfer functions further reduce standard deviation between results if elaborated for differences in measured NOx and for differences in CLEAR-NOx. Parameters are statistically significant.
Measured
Measured + Ctf
CLEAR + Ctf
EMROAD + Ctf
Avg. NOx [mg/km]
455
192
225
Std dev. NOx [mg/km]
230 76 58
200
Ries route agg.
NEDC
WLTC
Note: function translates per definition to “WLTP” parameters. This is a reduction for most PEMS trips.
Depending on the setting of the center point different levels can be adjusted later!

10. Results vehicle 1 (TUG)
Transfer function does not work properly based on EMROAD results (JRC check of correct EMROAD handling by TUG open!)
Measured
Measured + Ctf
CLEAR + Ctf
EMROAD + Ctf
Avg. NOx [mg/km]
455
192
225
Std dev. NOx [mg/km]
230 76 58
200
Ries route agg.
NEDC
WLTC
Note: function translates per definition to “WLTP” parameters. This is a reduction for most PEMS trips.
Depending on the setting of the center point different levels can be adjusted later!

11. Results vehicle 2
CLEAR reduces standard deviation compared to measured Nox emissions
Transfer functions further reduce standard deviation for measured and for CLEAR NOx
Measured
CLEAR
EMROAD
Measured & Ctf
CLEAR & Ctf
EMROAD & Ctf
Avg. NOx [mg/km]
145
137 Na
110
127
Std dev. NOx [mg/km] 68 25 na 12 20
WLTC

12. Results vehicle 3
CLEAR reduces standard deviation compared to measured NOx emissions
No statistical significant function for measured Nox found. In general poor regression coefficients for this vehicle. Test with highest emissions has average dynamics and average temperature. Measurement error(s) or regeneration events?
Measured
CLEAR
EMROAD
Measured & Ctf
CLEAR & Ctf
EMROAD & Ctf
Avg. NOx [mg/km] 90 82 na 75
Std dev. NOx [mg/km] 49 35 25
WLTC

13. Results vehicle 4
CLEAR reduces standard deviation compared to measured NOx emissions
Transfer functions further reduce standard deviation for CLEAR Nox. No statistical significant function for measured NOx. In general poor regression coefficients for this vehicle
Measured
CLEAR
EMROAD
Measured & Ctf
CLEAR & Ctf
EMROAD & Ctf
Avg. NOx [mg/km]
246
198
 na
199
178
Std dev. NOx [mg/km] 82 35 50 26
No WLTC measured

14. No statistical significant transfer function
Relevant severity parameters identified
Note: only vehicle 1 has a sufficient number of trips and variability in the severity parameters to perform multiple regression with all parameters.
Other vehicles allow only rough assessment of those parameters which varied in the tests.
Analysis of EMROAD results open for vehicles 2, 3, 4.
Parameters with statistical significant influence on NOx which thus may be included in transfer function ( X.. high influence, x… influence)
95P (v x apos)
(+h/100km)
Loading
Amb. T [°C]
rpm [s-1]
v_avg [km/h]
Veh 1
  x
Veh 2
   x
Veh 3
No statistical significant transfer function
Veh 4
95Perc.(v*apos) is relevant
Ambient temperature seems to be relevant but insufficient data available yet
Loading and positive altitude gains less relevant if CLEAR is applied before but insufficient variability in these severity parameters for most vehicles.

15. Example on transfer function design
Example for the case that 2 severity parameters (95 Percentile (v*apos)) and ambient temperature) are considered. Additional parameters can be added if data is available.
Note: values are rough estimates from actual data and are shall be seen as placeholders!
𝐶 𝑡𝑓 = 𝐴 1 ×[95𝑃 𝑣∗ 𝑎 𝑝𝑜𝑠 −95𝑃 𝑣∗ 𝑎 𝑝𝑜𝑠 centre ] + 𝐴 2 × [𝑡𝑎𝑚𝑏−𝑡𝑐𝑒𝑛𝑡𝑒𝑟]
“95𝑃(𝑣∗𝑎_𝑝𝑜𝑠 )centre ” … reference value where no correction due to transfer function occurs. Could be defined from existing data base, e.g. average of PEMS trip collection at TUG (see figure next slide).
A1….Parameter for influence of 95Perc(v*apos). First estimates from data for “best technology”: Urban A1~10mgNOx/1m²/s³, Extra urban A1~5mgNOx/1m²/s³
A2….Parameter for influence of ambient temperature Example: if tcenter = 0°C, add 6 mg NOx/°C This parameter considers cool down effects of exhaust gas after-treatment, but also reduced EGR at low temperatures. To which extent EGR reduction shall be considered has to be discussed.

16. Example on transfer function design
Example for the 95Perc(v*apos)
“Threshold V2” adjusted to preliminary TNO results and TUG data
Example with highest Diff: Diff 95Perc(v*apos) MW = = 6 Road = = 6 Urb = = 3
Diff NOx [mg/km]
MW = 6*5 = 30 Road = 6*5 = 30 Urb = 3*10 =30  weighted
Diff NOx = 30
Diff CF = 30/80 = 0.38
Alternative: use trip average values (tbd)
Work to do:
Elaborate generic coefficients Ai for main influencing severity parameters.
Elaborate centre line function (or value) for each parameter, e.g. “normal load = 120kg”, “normal t >0°C”
Test if separate handling urban/road/MW is necessary. May be relevant for 95Perc(v*apos) to avoid loop holes

17. Summary Transfer function seems to be feasible
“Differences” in emissions as function of differences in severity parameters seems to be best approach. CF changes can be directly calculated from such a function.
Analysis on 4 vehicles shows 95perc(v*apos), Temp., loading and altitude gains as relevant parameters with highest effect of 95perc(v*apos).
If the 95perc(v*apos) shall be used in a transfer function we suggest to add weighting function for trips with low time share with acceleration as suggested in last meeting (see “open to do’s”)
If Temperature shall be included needs to be discussed (extended CF is defined already for T<0°C. Shall transfer function replace the actual step function?)
Correct usage of EMROAD by TUG has to be verified by JRC
Elaboration of a generic transfer function from measurement data would need a lot of tests (relevant parameters need to be varied or fixed)
Test data shows behaviour of tested vehicle, not necessarily state of art
Simulation will be applied to produce data set to elaborate proposal for parameters in transfer function

18. Open To do from Sept. presentations
ToDo for Heinz Steven?: elaborate min share accel. From WLTP data base.
Dynamics of the driving style (95 percentile of v x apos)
+ Good correlation with NOx for unbiased driving
- Critical, for trips with short full load accelerations and long cruise control phases.
Such trips would result in high 95 v*a+ percentiles and thus strong emission reductions by the transfer function.  misuse would be very attractive!
Suggested solution:
Define minimum “normal acceleration share” for urban, road, motorway.
Correct 95 percentile of v*a+ with this share if acceleration time in the trip was lower.
Example (acceleration time % values to be validated by Heinz?)
Example Motorway: trip has only 60 seconds v*a with a above 0.1 m/s² and gives v*a+ = 30m²/s³
Correction with (157 – 60) = 97 seconds with v*a+ = 0:
V*a+Corr = (30 * * 0)/157 = 30 * 60/157 = (alternative: calc. 95 percentile from 157s sample)
Such a correction would be applied only if No. of seconds with a>0.1m/s² is below minimum value (e.g. 157 or 22% for MW). Thus the correction would not appear for “normal” driving.

19. Open To do from Sept. presentations
Elaboration of mileage shares for combinations of severity parameters as matrix. (e.g. % share of 95P(v*apos) between 20 and 22 with temperature between -5° and 0°C with positive altitude gains m and loading above 350 kg)
Calculate overall effect of transfer function on EU average CF

20. Suggested rough timeline vs. status
Topic
Who
until
Collection of measured Emission data (PEMS and WLTP) > 3 vehicles “Step 2 like” > 3 vehicles “Step 1 like”
ACEA ?
Analysis of these data to
validate feasibility of transfer function
elaborate best design of transfer function (relevant parameters, diff or delta, etc. by checking significances and correlation coefficients)
Status: Delayed due to missing data. Draft for one vehicle available.
TUG finalised
Set up PHEM models as shown before Status: Ongoing, first model results expected mid October
TUG
Analyse simulation results and elaborate preliminary transfer function. Status: open
Further improvements, discussions etc.
All ?