Professional Documents
Culture Documents
ThB01.6
I. INTRODUCTION
Turbocharged gasoline engines are becoming more and
more common in the marketplace due to the promise of
improved fuel economy at equivalent performance. It is
challenging for automakers to meet fuel economy, emissions,
driveability and performance objectives while keeping cost
affordable to consumers. They must rely heavily on the
control strategy to deliver the expected benefits of this
complex system.
Knowledge of current system behavior can be very beneficial for control. However, the turbocharged engine is a
harsh environment for sensors, particularly when the variable
of interest is in the exhaust path where temperatures can be
extremely high. Therefore it is not always possible to reliably
and/or affordably measure desired variables. This paper
focuses on estimating typically unmeasured characteristics of
the turbocharged gasoline engine: exhaust manifold pressure
and turbocharger speed.
An estimate of exhaust manifold pressure (pe ) may be
useful for many reasons, but it is particularly important
for its role in emissions control. Emission control systems
for gasoline engines rely heavily on feedforward air-fuel
ratio (A/F) control to meet regulations. Since fuel flow rate
can be regulated much more quickly than air flow rate,
the feedforward fuel command is slaved to an estimate of
the mass of air that will be in the cylinder at the time
This work was supported by Ford Motor Company
J. Buckland and M. Jankovic are with the Powertrain Controls Research
and Advanced Engineering Dept. at Ford Motor Company,Dearborn, MI
48128, USA, jbucklan@ford.com, mjankov1@ford.com
J. Grizzle and J. Freudenberg are with the Department of Electrical Engineering and Computer Science, University of Michigan,
Ann Arbor, MI 48124, USA, grizzle@eecs.umich.edu,
jfr@eecs.umich.edu
2714
Authorized licensed use limited to: University of Michigan Library. Downloaded on July 22, 2009 at 10:05 from IEEE Xplore. Restrictions apply.
pi
pb
z
Turbocharged
Gasoline
Engine
+
Ki
Gasoline
Engine
Wastegate
Actuator
Fig. 1.
N tc
pe
Turbine
Wwg
c0 (uwg ) + c1 (uwg ) We ,
(1)
f(x, u, w)
x =
[pi pb pe Ntc]
u =
w =
y =
[ uwg ]
N
T
[pi pb ]
(2)
T
T qbk ,
Ax + Bv
Cx
T
[u w]
Compressor
K fb
Throttle
Intercooler
(3)
25.23
10.97
4.36
0
0.89
105.26
0
147.67
,
A=
208.47
8.20 204.6
9.29
2.68
56.65
7.40
86.52
75.60
0
0.47
23.79
0
0
,
B=
0
41.66 31.51
0
0
0
C = [I22 022] ,
where the system states, inputs and outputs have been scaled
by their maximum values. 3 This linearization is used to
develop observer concepts and motivate singular perturbation
arguments in Sections III and IV.
III. BENCHMARK: THE TRADITIONAL
OBSERVER
The system (3) has four states, two of which are measured,
pi and pb . The remaining two states, Ntc and pe, can be
estimated using a traditional reduced order linear observer.
This approach is pursued to provide a benchmark to evaluate
performance of more practical observers in Section IV.
Following the approach described in [5], the system is
partitioned as follows
T
x1
[pi pb ]
x2
[pe Ntc ] ,
such that
x =
y
A11
x 1
=
x 2
A21
x1 = [I2x2
A12 x1
B1
+
v
A22 x2
B2
x1
02x2]
.
x2
x
2
=
=
Ar + Br
Cr + Dr
2715
Authorized licensed use limited to: University of Michigan Library. Downloaded on July 22, 2009 at 10:05 from IEEE Xplore. Restrictions apply.
10
12
14
16
18
20
12
14
16
18
20
12
14
16
18
20
12
14
16
18
20
72
71
70
0
10
Time (s)
120
110
100
0
10
Time (s)
185
180
175
0
10
Time (s)
Fig. 3. Actuator inputs and sensor outputs used for transient simulation of
estimators.
176
pe (kPa)
[02x2 L] .
174
172
NL Sim
2nd Order Observer
170
168
4.5
5.5
6.5
6.5
Time (s)
Relative Estimation
Error (%)
Time (s)
uwg (%)
Dr
18
17
0
pi (kPa)
[u y]
[A22 LA12 ]
[(B2 LB1 ) (A22 LA12) L + (A21 LA11 )]
[I2x2]
pb (kPa)
=
=
=
=
(deg)
Ar
Br
Cr
0.3
0.2
0.1
0
0.1
4.5
5.5
Time (s)
2716
Authorized licensed use limited to: University of Michigan Library. Downloaded on July 22, 2009 at 10:05 from IEEE Xplore. Restrictions apply.
Laplace Transform,
Ntc (krpm)
138
NL Sim
2nd Order Observer
136
pb (s)
=
(s)
s+
134
10
12
14
16
18
20
12
14
16
18
20
Relative Estimation
Error (%)
Time (s)
0.02
0
0.02
0.04
0
10
Time (s)
Fig. 5.
Transient prediction of Ntc using a traditional reduced order
observer.
the fast model and the slowest state defines the slow model.
By assuming that the fast states change instantaneously from
the viewpoint of the slow state, the slow model can be used
to construct a first order observer for both pe and Ntc .
With this approach, the number of calibration parameters
is reduced substantially to 14. The estimates are easily
separated for implementation at different execution rates, but
10 parameters are required to estimate either pe or Ntc alone.
The possibility of further reduction in calibration complexity
is explored with a different partitioning.
Ideally, the system would be partitioned using the physical
variables, as this will keep physical intuition intact and
avoid the computations that come with transformations. To
this end, it is clearly seen that the first eigenvalue, 1 ,
corresponds to the state pe since the dominant component
of e1 is in the direction of the third state. Association of
the remaining three eigenvalues to physical variables is not
evident from the eigenvectors. From physics, however, it is
known that the turbocharger speed responds on the order
of 4 due to its large inertia and intake manifold pressure
responds on the order of 3 due to the pumping action of
the engine and its relatively small volume. This would imply
that 2 corresponds to pb .
At first glance, this doesnt make sense physically since
the boost volume, which extends from the compressor outlet
to the throttle inlet, is significantly larger than the intake
manifold. Closer examination of the model, however, reveals that the response to throttle is governed by very fast
dynamics. Consider the special case where the throttle is
controlled perfectly to maintain a constant intake manifold
pressure and turbocharger speed changes relatively slowly so
it can be assumed to be constant. Then throttle is effectively
a disturbance to the boost volume and
Wthr
Wthr
Wc
RTb
pb +
,
pb =
Vb
pb
pb
Pressure ratio
132
0
RTb
Vb
Wthr
.
Wc
Wthr
pb pb
Fig. 6.
= [pi Ntc]
xf
= [pb pe]
(4)
such that
As,1 As,2 xs
Bs
x s
=
+
v
x f
Af,1 Af,2 xf
Bf
xs
y = xs = [Cs Cf ]
.
xf
As suggested by singular perturbation theory [6], assume
that the fast variables change instantaneously from the viewpoint of the slow states,
xf
x s
A1
(Af,1 xs B2 v)
i
h f,2
As,1 As,2A1
f,2 Af,1 xs
i
h
+ Bs As,2 A1
f,2 Bf v
2717
Authorized licensed use limited to: University of Michigan Library. Downloaded on July 22, 2009 at 10:05 from IEEE Xplore. Restrictions apply.
Plant
G(s)
Kpi (s+a)
(s+b)
1
(s+b)
tc
wg
Kwg
+
+
KN
Observer
G(s) = G 3 (s)N + G1 (s)
Gi (s) = C(sI-A) -1 B i , i=1,2,3
Kpi
Kwg
KN
26.5
21.1
0.92
-79.7
0.7
0.95
Fig. 7.
134.5
tc
N (krpm)
135
134
133.5
4.5
4.7
4.9
5.1
5.3
5.5
Time (s)
Ntc (krpm)
138
NL Sim
nd
136
Order Observer
st
1 Order Observer
134
132
0
10
15
20
Relative Estimation
Error (%)
Time (s)
2nd Order Observer
0.2
0.1
0
-0.1
10
15
20
Time (s)
The resulting observer gain corresponding to the pb measurement is 0. Therefore the system is simplified by partitioning as follows,
T
v
A,2 x + [B A,1] ,
Ntc
2718
Authorized licensed use limited to: University of Michigan Library. Downloaded on July 22, 2009 at 10:05 from IEEE Xplore. Restrictions apply.
176
136.5
NL Sim
1st Order Observer
Simple Observer
136
135.5
175
174
pe (kPa)
Ntc (krpm)
135
134.5
134
173
172
133.5
133
171
132.5
170
NL Sim
nd
2 Order Observer
st
1 Order Observer
132
131.5
0
10
15
169
4.5
20
pi (kPa)
115
113
111
109
107
4.9
5.1
5.3
5.5
5.7
5.9
6.1
6.3
5.5
6.5
Time (s)
Time (s)
6.5
pb (kPa)
182.5
VI. ACKNOWLEDGMENTS
181.5
Benchmark
State FB w/ simple N obs
180.5
tc
Ntc (kRPM)
179.5
4.9
5.1
5.3
5.5
5.7
5.9
6.1
6.3
6.5
R EFERENCES
135.5
134.5
Benchmark
State FB w/ simple obs
N est, simple obs
133.5
132.5
4.9
tc
5.1
5.3
5.5
5.7
5.9
Time (s)
Fig. 10. Nonlinear system response with state feedback and simplified Ntc
observer.
[1] Per Andersson and Lars Eriksson, 2004. Cylinder Air Charge Estimator in Turbocharged SI-Engines. Proceedings of the Society of
Automotive Engineers World Congress. SAE-2004-01-1366.
[2] J. Buckland, J. Grizzle, J. Freudenberg and M. Jankovic, Estimation
of Exhaust Manifold Pressure in Turbocharged Gasoline Engines with
Variable Valve Timing. Proceedings of the ASME 2008 Dynamics
Systems and Control Conference, Oct. 2008, Ann Arbor, MI USA.
[3] A. Karnik, J. Buckland and J. Freudenberg, 2005. Electronic throttle
and wastegate control for turbocharged gasoline engines. Proceedings
of the American Control Conference, pt 7, Vol. 7, pp. 4434-9.
[4] J. Jensen, A. Kristensen, S.Sorenson, N. Houbak and E. Hendricks,
Mean Value Modeling of a Small Turbocharged Diesel Engine, Society of Automotive Engineers International Congress and Exposition ,
SAE-910079, February 1991.
[5] John S. Bay, Fundamentals of Linear State Space Systems (McGrawHill, Inc. 1999), pp. 125-126.
[6] P. Kokotovic, H. Khalil and J. OReilly, Singular Perturbation Methods
in Control: Analysis and Design (Academic Press, 1986).
2719
Authorized licensed use limited to: University of Michigan Library. Downloaded on July 22, 2009 at 10:05 from IEEE Xplore. Restrictions apply.