Académique Documents
Professionnel Documents
Culture Documents
5~20
Abstract.
Mainly for reasons of cost and durability, an intake oxygen sensor is typically not available. An alternative way
of obtaining inlet oxygen concentration information is
to use dynamic observers. Such an observer would be
based on a model of the burnt gas fractions and mixing
dynamics of the fresh air and EGR in the engine. This
model happens to be linear (with time varying coefficients) and of second order. Unfortunately, however, it
is practically unobservable, excluding a straightforward
application of classical observer design theory. The estimation task is possible under one special situation: the
above dynamic model needs to be asymptotically stable.
It is fortunate that this stability requirement is met, allowing the estimation of oxygen concentration through
a simple simulation of the aforementioned model.
The estimation scheme will be complete if the coefficients of the linear model can be efficiently estimated
from measured quantities. We will show that if an inlet temperature sensor were available then we would be
able to implement a complete estimation scheme.
For the case where an inlet temperature sensor is not
available, we provide some empirical assumptions that
we have been able to verify on the available engine
model. This leads to a linear model of first order whose
coefficients depend on only one unmeasured quantity,
namely, the EGR maSS flow rate. This quantity has
been successfully estimated using neural network techimp1ementation for the
niquesi yielding a quite
Oxygen estimation scheme.
The model of the burnt gas fractions dynamics can simply be extracted from a mean-value model such as the
one given in Figure (1). Notations are summarized in
the table of the next page.
This control oriented model (1)for the diesel engine under consideration was described in [l]and other publications. It is a highly nonlinear model with seven dynamic
state components. Parts of the model are entirely empirical engine mappings and depend on the specific engine
P. E. Moraal: Ford Motor Co., FFA Control Systems, Aachen, Germany; e-mail: pmoraalBford.com.
I. V. Kolmanovsky, M. van Nieuwstadt: Ford Motor Co., FRL Control Systems, Dearborn, MI;
e-maik {ikolmano .mvanniel}@f ord.com.
S. Diop: Laboratoire des Signaux & Systkmes; Gif sur Yvette; France; e-mail: diopalss.supelec.fr.
The work of this author has been made possible, in part, by financial support from Ford.
0-7803-5446-XI99$10.00 0 1999 IEEE
852
-Fe
4
Fx
-mi
mx
N
Pi
fi
pz
3
(WexTe
VX
kg
kn
I rpm
I kPa
I kPa
[ kPa
pc
Dost-intercooler temDerature
(-Pc+71mPt) .
Ttc
Intake Manifold
Fuel injection:
kg/s
Vm
WO1
pi
Wxw
@.
kg/m3
Observability issues
853
While one may easily agree on this assertion by inspecting the engine model, its formal proof would, unfortunately, require tedious developments involving formal
manipulations of engine mappings in particular. It is
true that physical facts may be brought out to argue the
observability of 4 for instance. One such fact is that the
specific heat at constant volume of the air/combustion
gas mixture in the intake manifold is typically a function
of 4. The point is that the resulting dependence of the
supposedly measured quantities is very weak, leading to
an observability condition which is nearly singular.
Given the dynamics of mi and w, asserting the burnt
gas fractions to be unobservable from the dynamics
model ( 2 ) amounts to asserting that the following variables Wci, Wi,, We,, W,,, Wxt, Wri, and Wf are independent of 4, F,, and Fe.
Consequently, the burnt gas fractions dynamics become
linear
F = A(t)F B ( t ) ,
(5)
(4 -
-3 Wri 1
+ (F, - g,)'
2%
The unobservability implies that the only way to provide an estimate of these quantities will be under an
asymptotic stability condition of a dynamic model of 4
and F,. Since the concentration quantities 4 and F,
range between 0 and 1 no matter how the engine o p
erates, the stability issue is how the control variable of
the model should be constrained in order for the resulting submodel of 4 and F, to be both consistent and
asymptotically stable.
Exponential stability
m,
>I / ( W
holds for all t 2 to.
The above symmetric matrix inequality means that
Q ( t ) - 1/(27) is positive definite. Here I stands for the
unit matrix of order 2.
i = A(t)$ + . B ( t ) .
/
(6)
854
V ( e )= e?
+ e;
m, = Pxvx
-m,-
we,
Implementation options
that is,
fi = A(t)P + B ( t ) .
(9)
pi, N,and
WO assume the quantities Wd,Wf,
avnil~bleonline.
Ti to be
vx120
- -vd N
v, 120
vd N
m, = W=--
8 = A(t)P+ B(t)
provided that condition (8)holds and an intake temperature sensor is available.
Now the point is that such an impiementation has two
major drawbacks. First, it assumes an inlet temperature sensor which is not necessarily desirable. Second,
the uncertainty on the volumetric efficiency may induce
large errors on the total engine flow rate Wi,.
Assumption 4
Fe = F,
This c\ssumption is quite realistic considering the relatively small volume of the exhaust manifold and the high
gAs velocities associated with the blow-down of exhaust
gas out of the engine int.0 the exhaust syst.em. Simulation results shown in the figure below also support this
Assumption.
855
where
Fi,(t) = (1
L2J
........
..................,.........,.....__.__.................................................I
.
.
........................................
.........
.
.
i
1
... . . . I ................... .........i. ........i_ .........I_ ........ + ........i ......... :........
..
.
..
.:
:.
lo,
0
J.
..............
U*
0.1
Qs
@(t) evaluates to
0.6,
.
..
...
..
.
...............................................
.
.
+hOtolO+d.
;........I
; j
SNinMamda
Figure 2:
Fe
By Equation (3)
1--
t H exp
,o
Wri
wex
(-
X(s)ds)
4 = - X ( t ) (4 - F,,(t)) ,
im
........................................................................................
.
.
.
. . . . . . ..i. . . . . . . . . . . .
............ i. . . . . . . . . . . :. . . . . . . .j. . . . . . . . . .j
vd
- 2m
q 120
mi
..........................
0
200
hence
and
vd N
X(t) = - -
5.3
120
+ Wf + wri77v,
. . . . . . :. . . . . . . : . . . . . . . . .: . . . . . . . . .
.............:...............;...............:..............
BM
400
800
1200
1Mx)
...............
..............................................
Wci +Wf
wci
;,
......
*
.......
......
...............:._.
os
Finally we obtain the following equation for the estimation of the intake manifold burnt gas fraction:
......
..............;. ...............................
......................
1
1.5
2.6
timsh.konds
(- Jot
Figure 3: exp
X(s)ds) . In about 3 seconds, the
estimation initial error &(O) is devided by el0.
5.4
5.5
T am,
0,015
0.01
0.m
Simulation
We finally present the simulation of the burnt gas fraction in the intake manifold that our estimation scheme
allows to obtain. Note that for the EGR estimation, the
above described neural network was used, whereas the
mean value engine model against which the estimator is
compared, utilized the orifice flow equation.
2 0.~1
..............
24.05
10
15
20
25
30
IimetnlrmC
Conclusion
...............>............ :.....
References
............... ...............
.............:. ............. :................
10
16
................
20
25
30
857