Proceedings of the 2006 American Control Conference ThB14.

3
Minneapolis, Minnesota, USA, June 14-16, 2006

Development of an Engine Idle Speed and Emission

Controller

Masaaki Nagashima, and William S. Levine, Fellow IEEE

 compared their performance with that of the existing Honda

Abstract—This paper deals with the simultaneous control of
idling and emissions. Three candidate controllers were
developed and are compared, via simulation, with Honda's
current production controller. The simulation utilizes a
detailed engine model made up of an airflow dynamics model, a
combustion model, a fuel injection model, and a catalytic
converter model. Data from the engine model is compared
with measurements from a production engine in order to test
and evaluate the model.
I. INTRODUCTION
controllers which contain separate idle, air and fuel ratio,
and emissions controllers. This comparison is based on the
simulation.
This is a heavily abridged version of a report on this work
[5]. Please see the report for more detailed information
about every aspect of this work.
II. ENGINE MODEL
The engine model consists of four main components:
airflow dynamics model, model of engine dynamics, fuel

I n this paper, we describe two candidate MIMO controllers

for an internal combustion engine in the idling condition.
When the vehicle is idling, it is difficult to stabilize engine
speed because the engine speed is very low and is sensitive
to changing loads due to the Air Conditioner (A/C), Power
Steering (P/S), and Alternator (ALT). Meanwhile, United
States federal law requires improved regulation of the
exhaust emissions. Thus, it is also essential to reduce the
emissions. Research upon these problems has been
conducted. Most treat the engine speed control problem
and the emission reduction problem separately [1], [2], [6].
injection model, and catalytic converter model.
A. Airflow dynamics model
Fig. 1 gives an overview of the airflow dynamics. Note that
there are three inputs, the throttle command from the driver,
the manifold pressure, and the purge flow. The purge flow
is the result of inhaling evaporated gas into the cylinder in
order to prevent unburned gasoline from escaping to the
outside air. The airflow function is, as indicated, a nonlinear
function of its two inputs. The manifold fill in is a
representation of the last term in Eq. (1)
dPb ( t )

The combined objective of idle speed and A/F ratio control Aout
c (t ) Ain ( t )  A purge ( t  t purge )  ( ( dt )
R air ˜Tm ( t ) ) Vm 760
has been studied in [3]. (1)
We combined these problems—emission reduction and idle where
speed regulation—into one problem, thereby possibly A’out(t) Air mass flow from intake manifold (g/s)
avoiding interference between the two individual Apurge(t) Purge air mass flow rate (g/s)
controllers. tpurge Purge flow delay (s)
In order to test the proposed controllers, we developed a Rair Gas constant 0.28703 (J/gK)
detailed model of the Acura RL 3.5L V6 engine, 2005 Vm Intake manifold volume (m3)
model year and its emissions. In addition, a model of Tm(t) Manifold temperature (deg)
Honda’s production controller for this engine was used to 1 u(t) f(u(t),pb(t)) ka
control the model. The accuracy of the model was then Throttle
command Delay Rate Limiter Airflow
s+ki
Transfer Fcn
tested against measured data from a production engine. To function

evaluate the performance of the new controllers, we 2

pb(t)
d/dt kb
Manifold
fill in
1
Manifold A'out(t)
pressure Gain

3
Manuscript received September 15, 2005. This work was supported in Purge
part by Honda R\&D Tochigi, Japan. flow

M. Nagashima was a visiting Faculty Research Assistant at the Institute

for Systems Research at the University of Maryland, College Park, MD
20742, USA and is now an engineer at Honda R\&D Co., Ltd. Tochigi
Center 4630 Shimotakanezawa Haga-machi Haga-gun Tochigi, JAPAN
321-3393; (e-mail: Masaaki_Nagashima@n.t.rd.honda.co.jp).
W.S. Levine is with the Department of Electrical and Computer
Engineering, University of Maryland, College park, MD 20742, USA (e-
mail: wsl@eng.umd.edu).
Figure 1 Block diagram of the airflow dynamics
Because the controller under consideration only operates
after the vehicle is fully warmed up, Tm is a constant.
The output of the airflow function is filtered, as shown in
the block diagram. This greatly improves the accuracy of

1-4244-0210-7/06/$20.00 ©2006 IEEE 3278

the model, as shown in Fig. 2.
Measured airflow
Airflow w/o filter
Airflow w/t filter
Figure 2 Comparison of measured and simulated airflow
B. Model of Engine Dynamics: airflow to torque to speed
Fig. 3 is an overview of the engine dynamics.
Figure 3 Block diagram of the engine dynamics
The airflow amount, ignition timing, and the air/fuel ratio
(Ȝ), control the engine torque. Lambda (Ȝ) is also an
extremely important factor for the emissions; on the other
hand, the effect of Ȝ on the engine speed has been neglected
in most research. After the engine is warmed up, Ȝ stays
close to stoichiometric, the ideal air/fuel ratio; therefore, we
can linearize at the stoichiometric air/fuel ratio.
To calculate the torque due to combustion, we assume
that the combustion torque can be linearized at a nominal
point. Temporarily ignoring the effect of the Internal
Exhaust Gas Recirculation (Internal EGR), the relationship
between the airflow amount and the engine combustion
torque is almost linear. This explains the left-most three-
input summer in Fig. 3. The effect of internal EGR is
accounted for in the block labeled Combustion Trq. There
3279
is a nonlinear curve relating the output torque to the
manifold pressure. It is assumed here that the internal EGR
does not affect the manifold pressure which makes the
manifold pressure an exogenous input.
We have so far accounted only for mean torque.
However, the combustion torque is only generated during
the explosion phase of the engine cycle. In addition we
should include a pumping loss which is incurred during the
induction phase. The block labeled Cylinder angle provides
the information necessary to compute both pumping loss
(the Friction Trq. Block) and the time dependence of
combustion torque (inside the Combustion Trq. Block). The
computations to do this are complicated. The details are
given in our report [5]. The combustion, friction and load
torques are then added, scaled, and integrated once to obtain
engine speed. Notice that the engine speed is fed back to the
Cylinder angle block to enable us to compute the cylinder
angle.
To test the model, we compare its response to a step
change in throttle angle to the response of a real engine to
the same input in Fig. 4. Note that the simulation jumps
very frequently, causing the signal to appear blurry in Figure
4.
Measured data
Simulation data
Figure 4 Response of engine speed to a throttle
perturbation
As can be seen, the simulation is reasonably, although not
perfectly, accurate. It does slightly underestimate the effect
of the throttle.
C. Fuel Injection Model
A block diagram of the model of fuel injection is given in
Fig. 5. The model is in discrete time because the control is
the injection duration. Note that the feedback loop between
the injection duration and the Air/Fuel function models a
phenomenon known as wall wetting. Some portion of the
injected fuel does not enter the cylinder immediately. It
instead adheres to the wall and the valves around the
injector. This fuel is absorbed into the cylinder within
several time steps. The piece of the block diagram between
the rightmost summers is a linear approximation to the EGR Rol(t) Relative oxygen level (between 0 and 1)
(Exhaust Gas Recirculation). Coxy Oxygen storage capability
Ȝpre(t) Pre-catalyst lambda
Ȝpost(t) Post-catalyst lambda
Exflow(t) Exhaust gas flow (g/s)
fro(Rol(t)) Relative oxygen absorption or release
function
The exhaust gas flow can be computed from the air mass
flow out of the intake manifold (A’out(t)—defined earlier) by
Figure 5 Block diagram of the fuel injection system
'
The specific engine modeled is a V6 which has two banks Ex flow (t ) Aout (t )(1  (O pre (t ) /14.5) (4)
and two air/fuel sensors. A comparison of the simulation Note that we have added a low pass filter (at the right in Fig.
with data from a real engine is shown in Fig. 6. Notice that 7) to improve the match to experimental data.
the simulation responds much faster than the real system. The pre- and post-catalyst lambdas determine the
emissions of each of the three important gases, NOx(t),
HC(t), and CO(t). This is done by an algebraic calculation
Measured data shown symbolically in Fig. 8.
Simulation data

Figure 6 Comparison of simulation with measured Figure 8 The nonlinear emissions computation
response A comparison of the emissions measured for the modeled
six cylinder engine with the emissions from the simulation is
D. The catalytic converter model shown in Fig. 9. These results are for a fresh catalyst. The
The dynamic model of a three-way catalytic converter results for a maximally aged catalyst (120,000 miles) are not
presented here is taken from [4]. A controller is given in as good but are still satisfactory. They can be seen in the
[6]. The model is in two parts, a dynamic oxygen storage report that is the basis for this paper [5].
model and a static feed-gas emissions and catalyst
purification model. The dynamics are given in Fig. 7.

Figure 7 Block diagram of the dynamic oxygen storage

model
The oxygen storage model can be written as Measured Emissions
Estimated Emissions
dRol (t ) (O pre (t )  O prebase )
(0.21 ( sat ( Ex flow (t ))(
dt Coxy O pre ) f ro ( Rol (t ))
(2)
'
O (t ) O pre (t )  f ro ( Rol (t ))(O pre (t )  1)
pos (3)
where Figure 9 Measured versus actual emissions

3280

III. PRODUCTION CONTROLLER
At present, Honda has separate controllers for idle speed and
emissions. The engine speed controller also consists of two
separate controllers, a PID controller of the air amount and a
P controller for ignition timing. There are also feed forward
compensators for both load and ambient pressure. For
example, when the air conditioner (AC) needs to turn on, a
short pulse of extra air is inserted into the cylinder in order
to avoid a drop in engine speed.
The emissions controller is quite sophisticated. It includes a
self-tuning regulator (STR) to control the pre-catalyst Ȝ and
an adaptive sliding mode controller to stabilize the post-
catalyst Ȝ. Block diagrams of these two controllers are
given in Figs. 10 and 11.
Figure 10 STR controller for pre-catalyst lambda
The STR controller has five parameters that are estimated in
real time. They are individually time-averaged to avoid
chattering. The shaded block in Fig. 10 is the actual engine.
The other blocks are the controller components.
Figure 11 PRISM controller
The adaptive sliding mode controller is called a PRISM
(Prediction and Identification type Sliding Mode) controller.
A block diagram for it is given in Fig. 11. Again the shaded
blocks are the actual physical devices.
IV. CANDIDATE CONTROLLERS
Three different MIMO controllers were developed. All are
3281
based on a linearization of the model given in Section II.
The controller obtained by pole placement was clearly
inferior to Honda’s production controller. We will not
discus it further. The other two controllers, a dead beat and
an LQR-based design, both outperformed the production
controller, at least in simulation. All of the MIMO
controllers are based on a 23 state, 4 control model with
state feedback. State feedback is certainly not feasible.
However, both controllers would work with a state estimator
albeit with degraded performance.
The dead beat controller was designed to make the state
vector, x(k+1)=0, regardless of the initial state. The results
of the simulations of the dead beat and production
controllers are given in the following figures. Notice, in
Fig. 12, that both controllers regulate engine speed similarly.
However, the dead beat controller does a substantially better
job of regulating emissions, as shown in Fig. 13. Finally,
Figs. 14 and 15 show that the dead beat controller
outperforms the Honda controller when a power steering
input perturbs the system. Note that there is one exception
to this. The Honda controller produces less NOx emissions,
as can be seen in Fig. 15. There is a tradeoff among the
emissions. It would be possible to reduce the NOx
emissions of the dead beat controller but this would increase
the other emissions.
Of course, the production controller uses only feedback
from physical sensors, a considerable handicap compared to
the dead beat controller.
Dead beat controller
Honda controller
Figure 12 Dead beat controller versus current Honda
controller
 The other candidate controller was designed using the LQR
Dead beat controller theory. It also outperformed the current production
Honda controller
controller, as can be seen in the following Figs 16-19.
Again, the LQR controller increased NOx emissions slightly
in comparison to the Honda controller. Notice however that
the LQR controller did place much greater demands on the
ignition timing. This could be ameliorated by adjusting the
weights if necessary.

The LQR controller

Honda controller

Figure 13 Emission control by dead beat controller

versus Honda controller

Dead beat controller

Honda controller

Figure 16 Simulated engine speed control for the LQR

and Honda controllers

The LQR controller

Honda controller

Figure 14 Engine speed response to an increase in load

due to power steering perturbation

Dead beat controller

Figure 17 Simulated emissions for LQR and Honda
Honda controller controllers

Figure 15 Emissions resulting from an increase in load

due to power steering

3282
 modeled as effectively as possible. We invite others to try
to design controllers that will maintain idle speed as low as
possible while minimizing emissions despite large
perturbations to the load due to air conditioning, power
The LQR controller steering, and the alternator.
Honda controller
REFERENCES
[1] U Kiencke and L. Nielsen, “Automotive control systems for engine,
driveline, and vehicle,” SAE International, 2000, pp. 3-126.
[2] H. Kitagawa, “L4-engine development for a super low emission
vehicle,” Honda R&D Co. LTD, Tochigi R&D Center, SAE 2000-1-
0887, 2000.
[3] M.A. Shah and M.A. Franchek, “Control of IC engines for emissions
and idle speed regulation,” Proc. ASME Dyn. Syst. and Cont. Div. ,
Vol. 61, 1997, pp. 685-692.
[4] E. P. Brandt, Y. Wang, and J. W. Grizzle “Dynamic modeling of a
three-way catalyst for SI engine exhaust emission control,” IEEE
Trans. on Cont. Syst. Tech., Vol.8, No.5, 2000, pp. 767-776.
Figure 18 Simulated response to power steering [5] M. Nagashima and W.S. Levine, “Development of an engine idle
speed and emission controller,” ISR Report TR2005-103, 2005.
perturbation for LQR and Honda controllers Available: http://techreports.isr.umd.edu/reports/2005/TR_2005-
103.pdf
[6] G. Fiengo, J.W. Grizzle, J.A. Cook, and A. Y. Karnik, “Dual-UEGO
active catalyst control for emissions reduction: design and
experimental validation,” IEE Trans. on Cont. Syst. Tech., Vol. 13,
No. 5, 2005, pp. 722-736.

The LQR controller

Honda controller

Figure 19 Simulated emissions for LQR and Honda

controllers in response to power steering perturbation

V. CONCLUSION
The results presented here suggest that it would be
worthwhile to combine the idle speed and emissions
controllers into a single MIMO controller. In addition to the
advantages in performance, there is an unobvious advantage
to the two MIMO controllers. They do not require
feedforward air compensation. This eliminates the need for
calibration, thereby saving cost.
We did not design an Hf controller, although this is an
obvious candidate, especially since the main objective is to
reduce the effects of perturbations. This is something that
certainly should be tried.
We believe the model developed here has potential utility
beyond what we have done with it here. It is as detailed a
model as we could create of a real engine operating in the
idle regime, a mode that is challenging to control. The
effect of the normal perturbations of the idle speed has been

3283