closing the loop in engine control by virtual...
TRANSCRIPT
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
3Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
Overview
• Engine control and sensors
• Virtual sensor design
• Sensors vs. virtual sensors
• Optimization by virtual sensors (still in work)
• Conclusion and outlook
4Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
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
5Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
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
6Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
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
7Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
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
8Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
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
9Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
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
10Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
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
11Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
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
12Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
Commercial NOx sensors
Problems: cross sensitivities (but in some cases, like SCR, can beexploited) and especially start delay and magnesium poisoning
13Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
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?
14Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
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!
15Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
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
16Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
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
17Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
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
18Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
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
19Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
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
20Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
Sensor equations
Results can look horrible
NOx
PM
21Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
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
22Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
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
23Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
Comparison virtual sensor vs ANN
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 ]
24Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
Comparison virtual sensor vs ANN
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
25Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
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
[%]
26Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
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
27Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
Possible uses of virtual sensors
• Cold start
• Fall back sensor
• FDI
• Resampling
• Model for optimization
28Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
Validity test: MPC results arereproduced correctly
29Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
• 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
30Johannes Kepler University LinzInstitute for Design and Control of mechatronical Systems
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