closing the loop in engine control by virtual...

Closing the loop in engine controlby virtual sensors

Luigi del Re

Johannes Kepler University of LinzInstitute for Design and Control

of Mechatronical Systems

2Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems

Message

Actually obvious:• Closing the loop requires measuring the target variables – no commercially

available sensors do or will in the next future

• Virtual sensors offer substantial advantages in terms of price and reliability

Logic, but less obvious:• The design of a virtual sensor for engine control is not trivial

Unexpected:• The real problem are the setpoints

• Going from heuristical tuning to systematic optimization requires models. Virtualsensors are just this – and already in the right form.

• Mean value models seem to be the wrong approach for control system design


Overview

• Engine control and sensors

• Virtual sensor design

• Sensors vs. virtual sensors

• Optimization by virtual sensors (still in work)

• Conclusion and outlook


Framework

• Work performed in projects of the Linz Center of Competence in Mechatronics and Austrian Science Foundation

• Support by BMW Motors

• Tests on a dynamic engine test bench (AVL APA) under PUMA Open and ISAC, BMW M47Ü2 Diesel Engine, test vehicle 320d


Engine control and sensors

• The single most important engine control problem is a nonlinear optimal problem with constrained inputs and finite horizon

0

20

40

60

80

100

120

0 200 400 600 800 1000 1200

Zeit [s]

Ges

chw

indi

gkei

t [km

/h]

1220

( ), [0,1120]0

min ( )fuelu t tq t dt

∈ ∫

Under the boundary conditions:

1220

0

( ( )) ( )

( ) ( ) ( )

( )

ref

Xi Xi

M u t M t

u t u t u t

q t dt Q

=

≤ ≤

≤∫

Torque necessaryto track the speed

All manipulatedvariables

Bounds on pollutants


Numerical solution seemspossible but

1220

( ), [0,1120]0

min ( )fuelu t tq t dt

∈ ∫

Under the boundary conditions:

1220

0

( ( )) ( )

( ) ( ) ( )

( )

ref

Xi Xi

M u t M t

u t u t u t

q t dt Q

=

≤ ≤

≤∫

~ MeasuredDependency on uunknown

Can be estimated

For production: no idea!

Furthermore: computing intensive, slow


Daily solution

Transforms theoptimization probleminto a time reference

Forces the value to follow the references

Estimates unmeasuredquantities

Restating it as a tracking problem

ReferenceControl

EngineActors Combustion

Estimator

J v(t) u(t) w(t) y(t)

z(t)

h(t)

GeneratorTracking

Calibration


Motivation Basic idea of the standard solution

• Single SISO control loops (sometimes with feedforward) are not optimal

• Reference chosen empirically, usually interpolating steady stateoptimal points (in the sense of the optimal problem). How do you getthe right ones?

CombustionFF

C-

+

α w* yVGT

Inj

Thumble

EGR

-

+ C

Air path control


Tracking with original setpoints

185 190 195 200 205 210 215

200

400

600

800MAF

measured MPCreferencemeasured ECU

185 190 195 200 205 210 215800

900

1000

1100

1200MAP

measured MPCreferencemeasured ECU

185 190 195 200 205 210 2150

20

40

60

80

100

EGR VGT position

egr MPCvgt MPCegr ECUvgt ECU

185 190 195 200 205 210 2150

10

20

30

40FUEL

time [s]

fuel MPCfuel ECU

Faster in transients

Source: Ortner, del Re: IEEE T-CST May 2007


Tracking with reachable setpoints

0 5 10 15 20 25 30 35 40 45350

400

450

500MAF

[mg/

st]

MPCreferenceECU

0 5 10 15 20 25 30 35 40 451070

1075

1080

1085

1090MAP

[hP

a]MPCreferenceECU

0 5 10 15 20 25 30 35 40 450

20

40

60

80

100EGR position

[%]

MPCECU

0 5 10 15 20 25 30 35 40 4520

25

30

35

40

45VGT position

time [s]

[%]

MPCECU

Source: Ortner, del Re: IEEE T-CST May 2007


Effect on emissions

0 100 200 300 400 500 600 700 800 900 10000

50

100

150NOx x AirMassFlow

[ppm

* k

g/h]

MPCECU

0 100 200 300 400 500 600 700 800 900 10000

20

40

60

80

100OPACITY

time [s]

[%]

MPCECU

660 665 670 675 680 685 690 695 7000

50

100

150NOx x AirMassFlow

[ppm

* k

g/h]

MPCECU

660 665 670 675 680 685 690 695 7000

20

40

60

80

100OPACITY

time [s]

[%]

MPCECU

Explicit MPC: NOx: 110 Opacity: 50compare: original (non reachable) setpoints

Reference ECU: NOx: 100 Opacity: 100

Explicit MPC: NOx: 136 Opacity: 56


Commercial NOx sensors

Problems: cross sensitivities (but in some cases, like SCR, can beexploited) and especially start delay and magnesium poisoning


Virtual sensors

• Essentially models of the unmeasured quantity

• Two approaches– Simple models with error feedback (e.g. Kalman filters)

– Simulation models (if possible, FIR to avoid integratror effect)

• Error-feedback approach requires appropriate measurement

• Simulation models require model: where from?– First principles?

– Data?


Virtual sensor design

• ANN seem a sensible choice (universal approximators)

• However, universal approximators are not a sensible choice (noteven for interpolation)

• Goal is not to approximate, but to discover!


Leap of faith

• Identification = hypothesis testing

””Leap of Faith: Search for global patterns in observed data to Leap of Faith: Search for global patterns in observed data to allow for dataallow for data--driven interpolationdriven interpolation”” (Ljung)(Ljung)

• In other words: use physical knowledge, but otherwise expandthe number of hypothesis

• Problem: curse of dimensionality

• Solution: balance act between generality and specificity

• Idea: data regularization

• Hardly expressable in mathematical terms, therefore algorithmicapproach


Why not simple approaches?

Time

log(

data

Tot[,

"opa

c"])

0 200 400 600 800

-2-1

01

Time

data

Tot[,

"opa

c"]

0 200 400 600 800

0.0

1.0

2.0

3.0

Time

tbin

d(re

s.m

odel

, res

.val

)

0 200 400 600 800

-10

12

3

Forward selection?

- Build a list of possible nonlinearvariables NARX like)

- Test correlations- Take the most important one- Orthogonalize the rest- Add the second important one- Does not really work, parallelTesting of many hypothesis withalmost equalfitness leads torandom solutions


Procedure

Pre-Analysis:Delay time estimation, system

order estimation, variable selection

Signal conditioning:Filtering, delay time removal

Data

FunctionalBasis

Evaluationfunction

OptimizationGP Forward Selection

Result:Compact analytical formula

Source:

D. Alberer, L. del Re, S. Winkler, P. Langthaler: Virtual Sensor Design of Particulate and Nitric Oxide Emissions in a DI Diesel Engine, ICE 2005.

L. del Re, P. Langthaler, C. Furtmüller, S. Winkler, M. Affenzeller: NOx Virtual Sensor Based on Structure Identification and Global Optimization, SAE 2005


GP life cycle

• GP produces automatically programs, formulas are programs

• Ist Lifecycle:Population

of programs(formulas)

-x+

xx*

x *xx

Select parents accordingto their fitness

-x+

x *xx

Generate new formulas

Evaluate formulas


Crossover und Mutation in GP

• mutation und crossover work (without direction) through theexchange of sub-trees of formulastructures

• selection step ist directional

-

*

x2(t-2)x3(t-2)

1.5

Parent2+

/

5.0

ln

x2(t-2)x1(t-1)

Parent1

*

x2(t-2)x3(t-2)

+

/

5.0 x1(t-1)

Crossover

Child1

*

x2(t-2)x3(t-2)

+

-

5.0 x1(t-1)

Child3

Child2

x3(t-2)

+

/

5.0 x1(t-1)

Mutation


Sensor equations

Results can look horrible

NOx

PM


Validation NOx

Comparison between measured and estimated data validation

50 60 70 80 90 100 110 1200

10

20

30

40

50

60

70

80

90

100

time [s]

NO

x* [

%]

Measured NOxEstimated NOx


Comparison virtual sensor vs ANN

• Only standard ECU signals

• Validation with FTP-75, warm engine

• Model inputs: engine turning speed, injection quantities and timing, boost pressure, aircoolant and exhaust temperatures, fresh air mass, environment pressure

• ANN: max. 2 hidden Layers, max. 25 nodes/Layer



6 0 0 6 1 0 6 2 0 6 3 0 6 4 0 6 5 0 6 6 0 6 7 0 6 8 00

5

1 0

1 5 2 0

2 5 3 0 3 5

4 0 4 5

5 0

Opa

city

[%]

M e a s u re m e n tV i r tu a l S e n s o r G PV ir tu a l S e n s o r A N N

t im e [s ]



400 450 500 550 600 650 700 750 800 850 9000

10

20

30

40

50

60

70

80

90

100

t [s]

Opa

zitä

t [%

]

Messung

Virtueller Sensor GP

Virtueller Sensor ANN


PM in ECE

0 20 40 60 80 100 120 140-10

0

10

20

30

40

50

60

70

80

90

time [s]

Opa

city

[%]


Error distribution

-100 -80 -60 -40 -20 0 20 40 60 80 1000

1000

2000

3000

4000

5000

6000

7000

Estimation Error [% Opacity]

Qua

ntity

GPANN

• Standard ECU-Measurment signal

• FTP-75, warm engine

Error distribution for GP based approach almost symmetrical


Possible uses of virtual sensors

• Cold start

• Fall back sensor

• FDI

• Resampling

• Model for optimization


Validity test: MPC results arereproduced correctly


• Virtual sensors are feasible also for very complex systems

• They will hardly work alone but cross coupling with other sensors canbe exploited to track system changes

• They can be used to close the loop from two sides:– They can reduce the requirements on hard sensors, in particular, they can

overcome critical operating phases (like engine warm up)

– They can be used to optimize the strategy

– Final tuning will be still necessary, but not indispensable

Conclusions


The right way?

• Plausibility check: what is the alternative? CAMEO

• CAMEO is converging to the same kind of models…

• This can lead to a change of paradigms:– Instead of making the best out of your parts

– make the best out of your system

– to this end, start by describing in a control suitable way the mostdifficult part (combustion)

– and then use them to optimize the control system design

closing the loop in engine control by virtual...

Documents