Downloaded from SAE International by University Of Newcastle, Wednesday, August 08, 2018

SI Engine Controls and Mean Value Engine

Modelling
Elbert Hendricks
Inst. of Automatic Control Systems

COPYRIGHT /?75"
SOClf N OF AUTOMOTIVE
The Technical University of Denmark

Spencer C. Sorenson
ENGINEERS, INC. Lab. for Energetics
The Technical University of Denmark

ABSTRACT
domain control theory and static measured
Many existing classical electronic engine maps. The actual regulators used
control systems (speed-throttle, speed- are also classical in a control sense: P,
density, MAF (mass air flow)) are based on PI or PID controllers have been used for
quasistatic engine models and static f. ex. idle speed control and relay
measured engine maps. They are thus time control techniques have been applied to
consuming to adapt to new engine types, steady state lambda control [ I , 21 . These
are sensitive to dynamic sensor errors and first generation control techniques have
in general have undesirable dynamic been successful in reducing engine
characteristics. One of the main reasons emissions and fuel consumption; they have
for the characteristics of these also inaugurated the wide spread use of
strategies has been the lack of a precise, microprocessors. Second generation
systems oriented, equation based, dynamic controllers, using modern control theory
engine model. Recently a compact dynamic (state variable engine models,
mean value engine model (MVEM) has been multivariable optimal control and
presented by the authors which displays observers) have also been studied [ 3 , 41 .
good global accuracy. A mean value model Third generation engine controllers,
is one which predicts the mean value of employing specialized sensors (in-cylinder
the gross internal and external engine pressure sensors and/or optical combustion
variables. This paper shows how the sensors), are an area of intensive
engine model can be applied to the development [ 5, 6I .
systematic design and analysis of While second and third generation
classical electronic engine control control systems will offer significant
systems. performance increments, they will also
One of the main aims of the paper is require hardware which is not yet
to eliminate the use of cut and try available in sufficient quantity or
methods in designing dynamic engine quality. Second generation controllers
controls. The goal is also to improve also require more microprocessor power
transient drivability and emissions and/or programming convenience than is
performance. The main subjects treated currently available while inexpensive and
are steady state and transient fueling convenient third generation sensors are
strategies, lambda control and idle-speed still in the development laboratory. Thus
control. There is also a comparison of performance improvements necessary to meet
conventional speed density and MAF control drivability and 1994-95 emissions
systems including the limitations imposed requirements must be obtained using first
by signal conditioning filters and generation strategies, microprocessors and
available sensors. sensors. This will be a difficult task if
no new techniques are made available for
1 . INTRODUCTION engine controller designers.
Fortunately a very useful tool has
By now electronic engine controls are recently been constructed which shows some
nearly a generation old. These ways in which classical engine controllers
controllers might be termed classical can be improved without significant
controllers (or strategies) as they are hardware changes. This tool is the mean
based on the use of classical, frequency value engine model (MVEM) [ 7 1 . Mean value
 Downloaded from SAE International by University Of Newcastle, Wednesday, August 08, 2018

engine models predict the mean values of

the gross external engine variables (crank
shaft speed, manifold pressure, etc.) as
well as the gross internal engine
variables (thermal and volumetric
efficiencies, etc.) . The model is
physically based and has a global accuracy
which is comparable to standard
dynamometer data ( 2 2 % ) in spite of its
simple, compact form. It is also easy to
adapt to different SI engines. Generality
and flexibility are thus basic attributes
of the model and inappropriate and
unnecessary empiricism is avoided. The
basic modelling approach is also adaptable
to other engine types and several
different levels of complexity, depending
upon the application intended [8, 9 , 1 0 1 .
The main advantage of the mean value
modelling approach over conventional
methods is that it gives a good dynamic
description of engine behavior while its
steady state accuracy is basically the
same as an engine map. That the time

IT*,^' &I
development of important engine variables
can be accurately predicted means that
control systems can be designed which take EFI: ---
advantage of this knowledge. This implies
that transient fuel consumption and
emissions can be reduced while at the same
time improving vehicle drivability. Figure 1 . Nonlinear block diagram of the
The purpose of this paper is to show mean value engine model (MVEM) of a CFI or
some of the uses of the mean value engine EFI SI engine. The difference between the
model in improving conventional speed- two different injection types in the model
throttle, speed-density and mass air flow is in the fuel film submodel as indicated.
(MAF) engine control strategies. These The dotted and thin paths indicate that
improvements are not obtained by cut and these interactions are relatively weak
try methods but by an analysis of the mean compared to those which are bold face.
value model itself. The cost of these
improvements to the user is a little extra the main structural support of this paper
software which is compatible with has been presented in some detail earlier
conventional microprocessors. In fact it [7] and a number of newer developments
is shown that some second generation (including the emissions submodel) are to
control techniques can be profitably be presented elsewhere [ I 21 . Here the
applied to first generation, classical model is presented in terms of a control
engine controllers without too great an block diagram in order to show its
execution time penalty and without structure. The block diagram makes
additional sensors. readily apparent the physical details
The design procedure to be used here which are important for controlling an SI
is of interest in itself. It might be engine. Temperature and transient heating
called design by simulation. Following effects are not explicitly taken into
the controller design, the strategies account here for the sake of simplicity.
found are tested using PC (Personal However these effects can be inserted into
Computer) simulation routines (principally the model as is required.
SIL [ I l l ) and finally on an engine a The overall block diagram of a
dynamometer. In this way much costly and central fuel injection (CFI), multipoint
time consuming testing can be avoided. injection (EFI) or sequential fuel
Because of the mathematical form of the injection engine (SEFI) is given in figure
model, global stability of the control 1. The only difference between the
systems found can be confirmed over a different injection forms is the fuel film
broad operating range with due model which is at the top of the figure.
consideration of modelling error due to
production tolerances and engine wear.
2. SI MEAN VALUE ENGINE MODEL (MVEM)
The SI engine model which provides
 Downloaded from SAE International by University Of Newcastle, Wednesday, August 08, 2018

These equations have a linear form where r, = 2 / (ncyl

(n/60)) is the mean time
but are nonlinear because X depends upon between fuel inlection and fuel ignition.
the other states (dependent variables) or While the fuel flow is the mean speed
inputs of the model. In the case of a CFI control of an engine, a secondary control
engine, X = X(a, n) where the n dependence can be applied via the spark advance, 8,
is weak. In a port injected engine, X = and more uncommonly, via the air/fuel
~ ( n ) with a weak n dependence. ratio, A . The spark advance and air/fuel
Interestingly, the time constant for ratio influence directly the indicated
evaporation of the fuel film is of the efficiency of the engine as it has been
same order for all types of injection shown that
systems [I3, 141 . It should be noticed
that the output of the fuel film submodel
is the fuel flow into the intake valves.
This is the fuel flow which is actually
available for combustion. Some comments
must be made on equations (1) through (3) by several authors [7, 191 .
above. Superficially these equations look At the bottom of the block diagram is
like those due to Aquino [I51 . They are the dynamic subsystem which shows how the
nevertheless quite different as they are air mass flow to an engine is controlled.
written in terms of the fuel mass flow The external control input here is of
instead of the fuel mass. In the form course the throttle angle, a . This
given above they are more directly portion of the model describes the rate at
compatible with the MVEM (and its control which air can be supplied to an engine
applications) and are more akin to the past the throttle plate while at the same
equations given in 1976 by Rasmussen [161. time the engine is consuming an air mass
The fuel film dynamic submodel above has flow ma, = (n/2)Vdp q It is the latter
been identified using statistical methods effect which causes the strong coupling of
in work which has been published [I71 and the crank shaft speed into the manifold
by a different method in a paper which has pressure state equation while coupling the
yet to be presented [I81 . These other way is quite small (via the pumping
experiments are of a much more direct type losses) . The manifold pressure state
than those used by Aquino in [I51 and are equation is
thought to represent the best information
currently available on the manifold fuel
dynamics. Incidentally equations (1) - - Vd "lvol
Pman - - Pman
( 3 ) cannot be derived from Aquino's 2 v
equations: they represent a different
formulation of the problem from a
different point of view.
After going through the fuel film
dynamics, the fuel goes through the intake
valves. The resulting power drives the
-
inertia of the crank shaft and load with Pman - - - Pman +
their associated losses. This is shown in 'man
the middle of the block diagram (the
output on the right is the engine crank
shaft speed, n) . The proportion of the
fuel energy which drives the crank shaft
and pumping plus frictional losses and the
load is of course the indicated where rman = 2 V/ (V, n qVol) -
efficiency, q,. It should be pointed out All three subsystems have a low pass
here that the unit conversion between the nature but with very different effective
angular speed in radians/sec and in rev time constants. For a typical small
/min is "hidde?" in the moment of inertia engine, r, is very close to being a
I as I = (2n/60) /I 000) (I(engine)+I (load)) . constant at given coolant and ambient
The crank shaft differential (or state) temperatures (typically 1 - 2 sec). T~ is
equation represented by the block diagram not constant but variable between say 10
is and 1 sec over the operating range of an
engine and is smaller at higher speeds.
rman varies between 150 and 0.5 msec and is
minimum at high speed. T,,, is so small
compared to the other time constants that
it is often neglected. This leads to a
number of unfortunate consequences as will
become apparent presently. The output of
the lowest section of the block diagram is
the manifold pressure. The quantity which
 Downloaded from SAE International by University Of Newcastle, Wednesday, August 08, 2018

must be known if accurate lambda cont~ol

is to be realized in a CFI engine is ma,, mat
__f_
v
the air mass flow past the throttle plate.
In a port injected engine (EFI or SEFI)
;he air mass flow which must be known is
map, the intake valve or port air mass
flow.
The mean value engine model described
above has been extensively tested in the t o manifold pressure sensor
steady state and dynamically in the
specified operating range of several
engines[7]. The accuracy of the transient CFI speed-density
predictions of the model is on the same
order as that for steady state predictions
for all variables including the
instantaneous value of A .
3. CONTROL ISSUES
With the model in hand it is now a
simple matter to specify the main hot wire sensor
(stationary and dynamic) engine control
problems with precision. Because of the
structure of the model, these problems can EFI hot wire MAF
be divided into three nearly disjunct
classes, containing nearly disjunct
subclasses. Figure 2. An illustration of two popular
classical control strategies. V is the
1 . Air/fuel control volume of the intake manifold between the
a. compensation for fuel wetting throttle plate and the intake valves. The
(acceleration enrichment and top figure is for a CFI system with a
deceleration leaning) speed-density controller. Here it is
b. estimation of the air mass flow at necessary to know ma, (the air mass flow
the location of the injectors past the throttle plate) in order to
2. Idle speed control obtain a correct air/fuel mixture. The
a. load compensation bottom figure is for a EFI (or SEFI) hot
3. Spark control wire mass eir flow WAF) system. Here one
a. basic spark table must know map (the air mass flow past the
b. idle speed adjustment of spark intake port or valve) as it is this flow
advance that exists at the injector location. As
shown mounted,in the figure, the hot wire
la. It is widely known that when the will measure ma,. One should be aware that
throttle of an SI engine is opened rapidly both systems will suffer from transient
then extra fuel must be added to the measurement errors due to sensor lags
mixture in addition to that required for and/or anti-aliasing filters.
combustion. The main reason for the
excess fuel is that some of it is
deposited on the intake manifold and does placed at the injector locations but
not reach the engine until later. A either up stream or down stream from them,
secondary reason for excess fuel is that depending on the injection configuration
it is in general desireable to increase used. Moreover, the nearly unavoidable
engine power during acceleration. The mean use of signal conditioning and/or anti-
value engine model shows exactly why it is aliasing (low pass) filters will give the
difficult to fulfill the first requirement sensors a finite response time even if
exactly: the fuel wetting mechanism is they themselves have a zero response time.
basically a transient, nonlinear problem. It is of course immaterial how such low
It is not observable in the steady state pass filters are realized; they will still
and it is not an easily solved linear have a finite response time. A very thin
problem. fluctuation damping tube connected between
Ib. It is clear that the air mass flow the intake manifold and the manifold
must be known at the place where the fuel pressure sensor is equivalent to using an
is added in order to obtain correct electrical low pass filter after the
air/fuel ratio control (see figure 2). sensor.
Conventional classical engine control It will be observed here that the
strategies have failed to take into physical separation of point 1 . into
account the fact that the air flow or points I a. and Ib. has historically been
pressure sensors used are in general not made very difficult by the nearly
 Downloaded from SAE International by University Of Newcastle, Wednesday, August 08, 2018

universal use of engine maps and by a The models are nonlinear because of the
failure to understand the basically dependence of X and r on the engine inputs
nonlinear transient nature of the fuel and/or states.
wetting model (given its linear form). Formulation of the problems in
There has been a tendency to make do with electrical engineering terms immediately
an overall transient acceleration suggests a solution to the problem. This
enrichment "fix" (using large quantities is to make a linear frequency domain
of calibration and dynamometer time) and compensation filter based on simple
in general too rich a mixture. inversion of the transfer function form of
Now that problem 1. has been split up equations (1) to ( 3 ) . This is by no means
into two separate problems it is of course a new idea but surprizingly, no proper
a simple matter to attack them in detail. solution has yet been presented in the
The obvious approach is to first literature (see [I51 and 1201 1 . In other
compensate for the fuel film and second words one must realize a filter of the
compensate for the incorrect and/or tardy form
sensor (filtered) measurements. In common
language: the fuel must be placed where
desired at the instant that it is desired.
To do this, the air mass flow must be
accurately known at the same location.
2a. It is important to keep the engine between the fuel flow command and the
idle speed constant mainly because of the injection driver. rfe and Xe in equation
driver reaction to engine speed changes. (7) are experimentally determined
As automobile engines spend a large parameters of the intake manifold and
proportion of their lifetimes in idle, it engine which one desires to control. This
is important that good performance be means that the overall transfer function
maintained here. Using engine maps, it is between the fuel flow command and the
very difficult to characterize the intake valve fuel flow becomes
engine/control system's stability
characteristics. This is so because a
system's stability is a neighborhood (or
local) rather than a point attribute in a
mathematical sense. Hence an engine's
stability has been mysterious in the idle
speed region. This need not be the case
given a mean value engine model.
1 + (1 - X)ST,
1 + sr,
- I (8)
3a. Spark control is a steady state
problem which is quite variable from one
production unit to another and from wear
if Xe -
X and rfe rf. In other words if
the model is nearly perfect (and linear)
-
condition to another. Thus self- then the effect of the intake fuel film
optimizing adaptive spark controls have dynamics can be made to disappear.
been used to do extremal control in some As the fuel film model is neither
experimental investigations. Such exact nor linear only an approximate
algorithms are as yet too time consuming compensation will be possible.
for current microprocessors and will not Nevertheless, the algorithms which can be
be considered here. obtained are much simpler to implement and
Another interesting possibility to much easier to optimize than those in
appear recently is spark retardation conventional strategies. There is also a
during knock. This requires a reliable very useful power enrichment and
knock sensor which is sensitive to this deceleration leaning is "built into" the
condition alone. While some useful proposed scheme.
systems of this kind have been The transfer function formulation of
constructed, they are sensitive to details the fuel flow model is useful as a
of sensor placement, sensor frequency heuristic tool but it is not possible to
response and power plant vibrational realize a transfer function compensation
modes. There is thus not a clearly in a microprocessor. For this to be
apparent physically based method for using possible one must return to the
knock sensors at the present time and they differential equation formulation of the
will not be investigated in what follows. compensator. It can be written as

4. LINEAR FUEL FILM COMPENSATION SCHEME

The model for the intake manifold
fuel film (CFI engines) and for the intake
valve fuel film (EFI and SEFI engine) have
the same mathematical form: they are
quasi-linear pole/zero low pass filters.
 -
Downloaded from SAE International by University Of Newcastle, Wednesday, August 08, 2018

a (degrees)

-
Figure 3a. Fast throttle transient to
test the linear compensator. The rise and
fall times are 50 msec.
5 10 t (sec) 15

Figure 3b. Fuel flow command signal

corresponding to the throttle angle pulse 0.01~
of figure 3a. Ideal air/fuel control is
assumed in order to isolate the film fuel
control problem.
advantages of
This is one of the
simulation studies as
OaO'
I
opposed to direct experimentation. 6.~-31 1

Figure 3c. Fuel mass flow at the injector

after the compensator. The peaking on the
flanks of the pulse is of course caused by Osol.
the compensator. This corresponds to
transient acceleration enrichment but it
is obtained by direct calculation instead
0.
of measurement. I I I
0 5 10 t (sec) 15

I
Figure 3d. Port or intake valve fuel mass
flow. The function of the compensator is
to make this signal as much like figure 3b
as possible (which is the commanded fuel
flow). This is achieved to a reasonable 5 . ~ 4 ,
extent in this simulation.
0 5 10 t (sec) 15

Figure 3e. The final figure in this

series shows that a large tip-in lean
spike is being avoided and that there is a
slight acceleration enrichment inherent in
the compensator. The large lean spike on
deceleration means that there is also
inherent tip-out leaning built into the
system.

where the yew state c has been introduced

?nd where m,, is the commanded fuel flow.
mflis the compensator output which must be
inserted into the injector pulse width
control algorithm. Equations (9a) and compensator a set of simulation
(9b) can easily be written as difference experiments was conducted. During these
equations which can be immediately be tests, large, fast throttle angle
realized in a microprocessor. At constant transients were applied to the engine
coolant and ambient temperatures, X, is in model with ideal control assumed. Using
general a table function of the relevant techniques reported earlier, a model of a
input and/or state variables. modified intake manifold of a 1 .I L Ford
4.1 SIMULATION STUDIES OF THE LINEAR FUEL CFI engine was constructed [ 1 7 1 . This
FILM COMPENSATOR - In order to test the identification process yielded the
basic concept behind the linear function and constant
 Downloaded from SAE International by University Of Newcastle, Wednesday, August 08, 2018

fe = 1 sec (lob)
This manifold fuel film model was
then used to test the linear compensator.
A typical set of results is shown on
figure 3. The first figure, 3a, shows the
fast rise time (0.1 sec) throttle angle
pulse applied. The following figure shows
the corresponding fuel flow command input
to the compensator. Figure 3c shows the
compensator pulses both for an imagined
ideal compensation (dashed) and for the
linear compensator. One can see that the
linear compensator is less sharp than the
ideal compensator. This is because of the Figure 4. Test of the linear compensator
nonlinearity in X(a). The resulting on a real engine. The compensator
intake valve fuel flows in figure 3d show algorithm was realized in a PC based
that while the ideal compensator gives a control system. The algorithm was placed
perfect square pulse, the linear between the injector driver control and
compensator overfuels the engine slightly the fuel flow command signal. The engine
at the start of the transient. The reason was set to run in the open loop at an
for this is that the air/fuel ratio has arbitrary operating point under load and
gone rich in the linearly compensated case the fuel flow was adjusted so that lambda
while the ideal compensator keeps h equal was one. A square wave perturbation was
to one (see figure 3e). Though results superimposed on the otherwise constant
have only been presented for one fuel flow command signal and the
particular throttle angle transient, modulation depth was about +lo% of the
equivalent simulations have been made over constant level. An NTK Air/Fuel Ratio
the entire operating range of the engine. meter was used to measure lambda during
In this way the linear compensator can be the experiment. If the compensator worked
shown to useful globally. as it should, then the square wave
In order to further test the perturbation of the fuel flow command
algorithm found on a real engine, the 1.1L signal should survive its transition
Ford CFI simulated above engine was fitted through the engine into the lambda signal
with a PC-based control system of at the exhaust without dis$ortion. The
indigenous design [ 2 1 1 . The compensator fuel flow shown above, mfi, is that
algorithm of equation (7) was realized in measured at the injector by demodulating
the PC and placed between the fuel the pulse width modulated injector pulses.
injector driver control and the fuel flow The effect of the compensator is apparent
command signal. The engine was set to run in the peaked response to the square wave
in the open loop at a selected operating command signal (not shown). The resulting
point under load and the fuel mass flow lambda signal after the engine is a square
was adjusted so that lambda was one. The wave as intended.
throttle angle was constant. A square
wave perturbation was superimposed on the arbitrary units of volts and are read
otherwise constant fuel flow command out directly from the data logging A/D
signal and the modulation depth was about converters. The fuel flow shown is that
+lo% of the constant level. A linear measured at the injector by demodulating
lambda sensor NTK Air/Fuel Ratio meter was the pulse width modulated fuel injector
used to measure the air/fuel equivalence pulses. The scale for the lambda signal
ratio during the test. If the compensator is arranged in such a way that A = 0.5 is
works as it should then the square wave zero volts, h = 1 is 2.5 volts and h = 1.5
perturbation of the fuel flow command is 5 volts. It is thus clear that lambda
signal should survive its transition is changing symmetrically around h = 1 as
through the engine into the lambda signal was intended. The square wave form of the
at the exhaust without distortion. fuel flow command is maintained and the
Figure 4 shows that this is in fact compensator is seen to fulfill its desired
the case. The signals shown are given in function to a good approximation. The
 Downloaded from SAE International by University Of Newcastle, Wednesday, August 08, 2018

unfortunate dynamics of the fuel film the basis of this simplification from the
subsystem are being made to disappear. mean value engine model equations.
A number of other tests at a number In replacing the differential
of different operating points and loads equation (6) with an algebraic equation
have confirmed that the compensator one is using a singular perturbation model
performs well globally. Less accurate reduction technique 1221. This can be
compensation is possible at very small explained by setting e = rman in equation
throttle angles due to the failure of the (6a). This reduces the manifold pressure
modelling assumptions. At small throttle state equation to
angles of course, absolutely correct
compensation is of less interest as no
power is required of an engine.
5. MEASUREMENT OF AIR MASS FLOW AND
CONTROL OF THE AIR/FUEL RATIO It has been shown earlier that the
manifold pressure large signal time
The basic principle of air/fuel ratio constant is very small compared to the
control which was set forth in section 3 crank shaft equation large signal time
is that the air mass flow must be measured
or estimated at the location of the
injector(s1. This is obviously true both
If one lets e -
constant [71. e is thus a small quantity.
0 then equation (11)
becomes an algebraic equation.
for CFI and EFI (and SEFI) engines. It
should also be obvious that sensor
deficiencies in accuracy and response time
must also be taken into account.
In conventional control systems the
air mass flow is nearly invariably
measured at a different location than the
injector(s) for practical reasons. In a
CFI speed-density system the manifold To make the limiting operation valid one
pressure is measured after the throttle
plate, which is the usual target for the
injector. See figure 2 (top). A hot wire
-
must notice that em,, remains finite for e
0. Equation (12) is the speed-density
equation which was the starting point for
MAF/EFI system commonly measures the air the derivation of (1 1 ) . It says that the
mass flow before the throttle plate even port and throttle air mass flows are
though the injectors are far down stream equal. This is only approximately true.
at the engine ports. See figure 2 Another way to state the result of
(bottom). In cases where sensor placement equation (12) is that if the manifold
is correct, often no account is taken of pressure reaches equilibrium rapidly then
sensor dynamics and anti-aliasing filters, one might as well use the steady state
leading to transient errors. solution of equation, (6b). This can be
Thus it is clear that whatever sensor obtained by setting p,,, = 0. This yields
complement or system configuration is the same result as equation (11) in this
selected, the air mass flow should be simple case. The reason for making this
estimated at the injection point with approximation is that it reduces the model
proper attention to sensor dynamics and order which is a common engineering
signal conditioning filters. This section practice. This reduced order model is
deals with the imperfections of current also the nearly universal foundation of
systems in this regard as well as some classical engine control system design
methods to improve performance based on practice and fits in well with
mean value engine models. conventional quasi-static models of how
5.1 SINGULAR PERTURBATIONS AND AIR/FUEL engines work. Unfortunately, this
CONTROL - The mean value engine model is simplification has also made it difficult
composed of three dynamic subsystems: the to see an important reason why classical
fuel film model, the crank shaft subsystem control systems do not function properly
and the manifold filling dynamics. The in transient operating modes.
fuel film compensation network of the 5.2 SPEED-THROTTLE SYSTEMS - A speed-
previous section makes it possible to throttle engine control system is a low
ignore the effects of the first sub-system cost system because it requires only a
so that only the last two states need to throttle angle sensor, a crank shaft speed
be considered here. It is well known that sensor and a manifold temperature sensor.
the air mass flow through the intake It is also basically an open loop (or feed
manifold finds its equilibrium much more forward system). For this reason the
rapidly than the crank shaft system. Thus injection system most often used with this
it is common to ignore the manifold filing kind of strategy is an inexpensive type:
dynamics in the construction of engine CFI . Because there is no sensor to
control strategies. It is easy to derive actively measure the air mass flow, the
 Downloaded from SAE International by University Of Newcastle, Wednesday, August 08, 2018

system is dependent on accurate functional description

calibration (or engine mapping) for proper
operation.
The speed-throttle control algorithm
has a purely phenomenological basis and is
difficult to derive in functional form.
Nevertheless, a valid starting point is
equation (12) for the throttle plate (or,
more correctly, the port) air mass flow where a = a, + ala, b = b, + bla, the
subscripted a's and b's are constants, no
Vd " 'lvol is a fourth order polynomial in the
throttle angle and qVolo is a constant.
This is actually the lower branch of a
hyperbola whose apex moves to the right
If ig,is known from equation (13) then the with n and up with increasing a. It is
requlred fuel flow for a given value of A purely empirical. The reason for its
is complicated behavior is that it is
expressed in terms of variables which do
not have a simple physical relationship to
the quantity expressed. In parenthesis it
should be remarked that this is a typical
example of inappropriate empiricism.
where 1 / A is selected to correspond to the The relationship between p,,, and a in
driving mode desired (three-way catalyst the steady state can be derived by
and steady state operation, acceleration inverting the mean value engine model
enrichment, etc). Equation (14) is of equation for the air mass flow past the
course easily converted to a form which throttle plate
facilitates calculation of the injection
pulse width, for example

where a, is a constant (throttle plate

angle at closed throttle position) and
is the actual fuel charge per revolution.
This quantity must then be further sub-
divided, depending on the number of engine
cylinders and the injector configuration
used.
Returning now to equation (13), it
can be seen that the air mass flow per
revolution is dependent only on qVol p,,,.
Only one engine state variable, n, and
only one input variable, a, are known to
the control system at any time. Thus the
quantity qvol p,,, must be found in terms of Given the form of these equations, it is
the throttle angle and the crank shaft easy to see that they are for the air mass
speed. qvol is known from the mean value flow of a compressible fluid past a
engine model to be an algebraic equation constriction whose area is A, = 1 -
in the engine states qvol = qvol(n, p,,,) cos(a - a,) and with a constant flow
while p,,, is determined by a differential coefficient [ 7 1 . Equation ( 1 7 ) can, in
equation. Thus the equation qvol pman= f (a, principle, be inverted to yield
n) can only be correct in the steady
state.
There are several ways to determine
f (a, n) above. In general f is found by
~apping the air charge per revolution,
mat/(460) (or the air charge per cylinder if kat(a,p,,,) is known. This is how one
per revolution) as a function of a and n. might use the mean value model equations
This is also convenient as it saves to construct pman(a)for a speed-throttle
microprocessor memory capacity. It is strategy. p,,,(a) can also be obtained by
instructive however, to look at qvol(a, n) engine mapping. Figure 5 (bottom) is a
and pma,(a). qvOl(a n) is plotted for a plot of p,,, (a) with n as a parameter. This
typical engine in figure 5 (top) as a plot shows that the quantity £(a, n) will
function of n with a as the parameter. It be poorly characterized at large throttle
is a rather complicated mountain landscape angles. The reason for this is that there
which nevertheless has a fairly simple is a very small pressure drop across the
 Downloaded from SAE International by University Of Newcastle, Wednesday, August 08, 2018

throttle plate close to W.O.T. Thus, all

of the constant crank shaft speed curves
become very flat at large throttle angles.
This means that in order to express f ( a ,
n) in a microprocessor table, one has to
be very careful in cutting up the throttle
angle axis into table entries. The cuts
must be small at small throttle angles to
obtain good accuracy. It is also clear
that a must be measured by the control
system with good accuracy - especially at
small throttle angles.
In other words, the throttle plate
angle when the throttle is closed must be
carefully controlled on the production
line and the throttle position
potentiometer must be of very high
quality. This implies, of course, that
the throttle body and its sensor will be
expensive. Another important point is
that measurement noise both during
calibration and operation will play an
important role in degrading the quality of
control attainable. Proper, frequent and
qualified maintenance are thus primary
requirements.
As may be seen from the analysis
above, the correct operation of a speed-
throttle control system in the steady
state is not without problems. This is
even more apparent when the transient mode
of operation is considered. The main
reason for this is that the air mass flow
which is found is the port air mass flow.
As most speed-throttle systems are
intended for CFI systems, it is the air
mass flow past the throttle plate, iat,
which is the quantity that must be known a (degrees)
for proper air/fuel ratio control. As has
been pointed out earlier, this mass flow
can only be found by solving a Figure 5. Analysis of a speed-throttle
differential, not an algebraic equation control strategy. top: ,,q, (n, a) plotted
(see equation (6)). Thus a speed-throttle with the throttle angle as a parameter for
control system cannot solve the problem of a small SI engine. In order to find the
correct dynamic determination of the air function f (a, n) this function must be
mass flow rate at the injector position. multiplied by p,,,(a) in the figure at the
A number of "fixes" can be conceived. bottom. bottom: plot of pma,(a) with the
An obvious one is to make the enrichment crank shaft speed as a parameter. There
factor, y = ( l / h ) , dependent in part, on is very little variation of the function
the rate of change of the throttle angle. at large throttle angles ( > 5 0 degrees).
While this can do a great deal to This is because the throttle plate ceases
alleviate transient leaning, in general, to be effective when it is well open. It
it is always the case that too much fuel
is used to solve the problem. This in a will thus be difficult to express f (a, n)
sense is exactly the problem with accurately in a microprocessor table.
carbureted engines and one could say that
conventional speed-throttle systems are For this reason no further consideration
closer to being "electronic" carburetors will be given to this system here.
than true electronic engine control 5.3 SPEED-DENSITY CONTROL STRATEGIES - For
systems. a number of reasons, speed-density control
The inherent problems with the speed- systems are the most intuitively obvious
throttle system suggest that it will not control systems which have been made and
be a viable strategy for the mid 1990's. they are in fact also the oldest. In
The lack of robustness, with respect to addition to the throttle angle, crank
modelling errors and inaccurate shaft speed and manifold air temperature
adjustment, as well as poor dynamic sensors of the speed-throttle systems,
response suggest that it will not survive. they only require a manifold pressure
 Downloaded from SAE International by University Of Newcastle, Wednesday, August 08, 2018

sensor. In theory, they should work a (degrees)

ideally with the now very popular port 50.
injected engines (EFI, SEFI) with very (-
simple and compact algorithms.
Unfortunately, their development has been 40.
dogged by a lack of understanding of basic
engine dynamics and poor quality pressure
sensors. 30. L
Using the equations of the mean value 1 I I
engine model it is very easy to state the 0 5 10 t (sec) 15
basic conventional EFI (or SEFI) speed-
density control algorithm. The port air
mass flow is

If qvo, is known (from f. ex. engine

mapping) and n , and pman are measured
accurately, then ma, ca? be calculated with
good precision. And map is the air mass
flow at the injector location so that if Figure 6. Results of a simulation
correct compensation for the fuel film experiment assuming a sensor having a
dynamics can be achieved, ideal air/fuel finite response time (rise time: 1 msec).
ratio control should be immediate. The top figure (6a) shows the throttle
Unfortunately, pressure sensors (even angle transient applied to a mean value
those currently available) exhibit finite engine model of a small engine provided
response times and thus any calculation of with a conventional speed-density control
equation (19) will be behind the fact by algorithm. The lower figure (6b) shows
at least 1 to 5 milliseconds. Older the resulting lambda response of the
sensors, in particular those using simulated engine. The air/fuel ratio
frequency modulation techniques, will be becomes lean enough for the engine to
even slower in responding; say 100 to 400 misfire even with a very fast pressure
msec at best. In such a case performance sensor.
will be even worse and it is thought that
several manufacturers abandoned an expected . Engine performance can be
otherwise good strategy at an early date expected to be poor as the mixture becomes
because of this problem. lean enouqh for the engine to misfire.
In order to show that even a pressure There will of course be strong emissions
sensor response time on the order of 1 penalties, especially in HC and CO. NOx
msec is not sufficient to avoid a large will be reduced as the combustion
lean spike, a set of simulation temperature will decrease.
experiments using a MVEM were performed Again, one must remember that the
assuming a sensor response time of 1 msec. simulated sensor's rise time is 1 msec.
The results are displayed in figure 6. This is quite fast but obviously not fast
The engine modelled in these experiments enough. It is impossible to make a sensor
was a small capacity engine, V, = 1 .I this fast in a real system because of
liter. An ideal fuel film compensator was sampling time limitations in the control
assumed for the sake of clarity. Figure microprocessor. The sensor bandwidth
6a shows a large and fast tip-in and tip- cannot be made greater than 1/5 to 1/2 of
out throttle angle transient. The the sampling frequency. For a sensor
rise/fall times are about 70 msec. A response time of 1 msec this implies that
standard speed-density control strategy the sampling frequency must be between
was implemented using the algorithm above. above 2kHz to 5kHz. This will require a
The mean engine speed varies between 3300 rather expensive microprocessor. One must
and 4500 rpm during the simulation. degrade one's sensor with a filter if a
Because of the sensor response time, the practical system is to be made, but there
sensor gives a very poor picture of the is no option if aliasing is to be avoided.
transient air mass flow. Figure 6b shows There is no doubt that this problem
the resulting lambda transients. It is commonly seen on real engines. In fact
reaches A > 1.2 which is far too lean for in his classic SAE paper on transient
proper operation. In fact, only towards engine problems, Aquino shows that both
the middle of the tip-in pulse does the conventional speed-density and
speed-density system find the correct air conventional mass flow control algorithms
mass flow. There is again difficulty at give lean spikes on fast throttle
the tip-out transient as might be transients. This is figure 13 of [I51 .
 Downloaded from SAE International by University Of Newcastle, Wednesday, August 08, 2018

Here it is apparent that a lean spike of An observer is a mathematical model

the same order as obtained above is of a system which runs in parallel with
observed for a fast tip-in throttle ramp. the system in real time. If the
The lean transient pulse width (-1 sec) mathematical model is physically based,
corresponds to the throttle angle ramp then the observer uses apriori physical
time used as is also observed in the knowledge as a basis for control. The
simulation shown here. basic interest in using an observer is
That there is little general that it can make available in£ormation
recognition of the importance of the about a system which is not directly
response time of a pressure sensor in a accessible for measurement (or which will
speed-density system is shown by the require sensors which are too expensive or
results of an informal enquiry by the which are otherwise impractical).
authors during the 1989 SAE Annual A block diagram of the open loop
Meeting. At the exhibition booths of a observer control scheme is shown in figure
number of well known sensor and/or control 7. Here it may be seen that the state
system manufacturers, it was not possible estimates are inserted into the control
to obtain any information on the response algorithm which is used to control both
time of the available sensors. On the the control object (an engine) and the
basis of the simulation experiment mathematical model. It is being assumed
presented above, this is a remarkable here that the model is an accurate
state of affairs. representation of the system to be
5.3.1 Improving the Performance of Speed- controlled. This is of course an
Density Control Strategies - The main impertinent assumption which will exact
difficulty in realizing accurate control severe punishment if the system which is
of the port air mass flow is that the to be controlled is poorly modelled.
manifold pressure sensor cannot be made to Fortunately the manifold tilling model is
respond rapidly enough to avoid a tip-in known to be quite accurate once it has
spike. The matter of compensating for been properly put into place by the
inaccurate or missing sensors is a common selection of its constants. This has been
problem in modern control engineering. confirmed by the independent observations
The approach usually adopted is to use of the authors and numerous investigators
apriori knowledge of the physics of the [7, 15, 16, 241 . Moreover it is also
system in question to compensate for the known that the filling model is not very
lack of accurate measurements. This is a sensitive to wear and aging effects. It
logical starting point for a physically .is this fortuitous set of circumstances
based compensation for currently available which is to be used in the following.
manifold pressure sensors. An observer for the manifold filling
Accurate physical models of subsystem can be written
deterministic systems can be used in
control systems by building them into the
control algorithm. In this service such
models are called observers. Observers
can be either of the open or closed loop
varieties. The observer to be proposed
here is an open loop observer. The
interesting part of an engine as regards
A/F control is the manifold filling
dynamics. As the MVEM manifold filling where the hats indicate thaththe value of
submodel is quite nonlinear, a nonlinear p,,, found is an estimate and T,,, indicates
observer must be used in order to make an that this variable is to be evaluated at
accurate instantaneous estimate of the the measured value of n and the estimkted
manifold pressure. value of the manifold pressure, p,,,.
There are several reasons why an open Equation (6) and its observer version (20)
loop observer is being proposed as opposed have an interesting characteristic which
to a closed loop observer. The main can be used in an open loop observer.
reason is that the stability and This is the very small effective time
convergence characteristics of nonlinear constant for rising and falling throttle
closed loop observers for engines are angle transients. This suggests that the
still under investigation. While recent observer be used only on transients while
theoretical work suggests that these the manifold pressure measurement is used
problems can be solved [ 2 3 1 , it is never when the engine is operating in the steady
wise to propose control systems for state. This scheme avoids the response
practical systems with doubtful stability time deficiencies of the sensor/filter and
properties. In particular this is ill- allows a correction for system parameter
advised in a system which is so drift over long periods of time. The
potentially dangerous as an SI engine in criterion for changing from one mode of
an automobile. operation to another is the rate of change
 Downloaded from SAE International by University Of Newcastle, Wednesday, August 08, 2018

of the throttle angle. If this becomes

larger tha2 a certain threshold value, the control X

estimate p,,, is used. If it becomes

object
smaller than a certain value then the
measurement, p ,, is used in the fueling
r u
algorithm. For the sake of simplicity,
the observer output will always be called
control
-
h algorithm
pmanin what follows, independent of whether
or not the measurement or estimate of the ,L
4

X
manifold pressure is being used.
model
The integration of equation (20) must
be carried out on-line in the engine
microprocessor. This is not difficult as
the engine inputs and states in (20) are
already known in the microprocessor and
-
(20) can easily be converted to a simple Figure 7. Block diagram of an open loop
difference equation. There are a number observer embedded in a control algoritp.
of ways to do this but for the necessary If a correct model is available, then x =
accuracy, the computational and space x. One can the use the state estimates in
penalties are not great as the difference the control algorithm. Such a system
equation is in fact quite compact. might also be called a feedforward system.
It should be remarked here that for The overall system is controlled by the
the observer to work properly, the input r.
function P, (a) must be calculated or
measured. In general, a calculation based
on Harrington and Bolt's expression [ 2 6 1
will be sufficiently accurate. The wing" for obvious reasons. The throttle
quantity P, (p,,,) need not be calculated by effectiveness may be judged form the
itself but .can be included in a function partial derivative of the function with
or table mat(a, pman) . A plot of the respect to the throttle angle. It may be
seen here that it will be necessary to
function is given in figure 8. This
surface is sometimes called the "beta divide the manifold pressure axis up into
small increments in order to obtain
acceptable accuracy over the entire
surface.
5.3.2 Compensating for a Deficient
Pressure Sensor - With the fuel film
compensator and observer in hand it is not
difficult to improve both the CFI and EFI
(or SEFI) strategies. This hinges upon a
recognition of where the air mass flow
must be known.
In a CFI system,the air mass flow at
the throttle plate, mat,must be known at
the throttle plate as the fuel is in
general sprayed onto it. From equation
(20) it can be seen that hat is
automatically estimated in the observer
&
.
equation as mat(a, pman)
h

Proper control of EFI and SEFI

systems requires knowledge of the air mass
flow at the engine ports. As an
accurately, effectively instantaneous
measurement of pmanis available from the
Figure 8. Three dimensional plot of the P h
observer (p
,,,) it is easy to calculate the
surface ("beta wing") over the ( a , p,,,) required quantity.
plane. The slope of the surface along the
alpha direction shows the relative control
effectiveness of the throttle angle. When
the slope is shallow, the control
effectiveness is small. Not all of the
surface is normally accessible to the
engine as not all values of throttle angle
and manifold pressure are encountered in This of course requires, as before for the
normal operation of an engine. CFI system, that the observer be built
into the microprocessor.
 Downloaded from SAE International by University Of Newcastle, Wednesday, August 08, 2018

Figure 9a. The fast throttle angle

transient used to test the modified speed-
density strategy. The rise and fall times
are a little more than 0.1 sec. The
sensor was a 5 msec laboratory quality
semiconductor pressure transducer but the
anti-aliasing filter had an effective
response time of 0.05 sec.

Figure 9b. Time response of the engine

1'1 pman
(bar)

manifold pressure (2) and the observer

(1). It can be seen that the observer is
effectively "locked" to the manifold
pressure even during fast transients.

Figure 9c. Measured (2) and calculated

(1) air mass flow past the throttle plate.
It may be seen that the observer is
capable of accurate reproduction of the
independently measured air mass flow.
Some time shift has been introduced into
the measured response to compensate for
the distance between the measuring device
(a standard nozzle and pressure sensor)
and the engine.

Figure 9d. Lambda measured for the fast

tip-in/out transient from a linear NTK
Air/Fuel ratio meter. Here a linear
compensator of the type mentioned earlier
has been used together with the observer.
Reasonable performance is observed in
spite of the limited bandwidth of the
manifold pressure sensor.

5.3.3 Testing of the Improved CFI A/F observer in reducing the effects of the
Control Strategy - In order to test the finite rise time of the pressure sensor.
basic observer a series of tests were run 5.4 HOT WIRE (MAF) CONTROL STRATEGIES -
on an engine with fast throttle angle The hot wire (HW) MAF control system is a
transients. These time series were then very direct approach to the A/F control
inserted into a simulation routine and the problem. It is based on a direct
performance of the partial open loop measurement of the air mass flow to the
observer plotted against the measured engine. Assuming an ideal measurement of
manifold pressure. A summary of the the air mass flow, it is a trivial matter
results of this experiment are plotted in to determine the necessary fuel flow to
figure 9. The pressure sensor used here the engine for a given value of 1.
was a laboratory quality sensor having a
rise time of 5 msec. The engine was a
1 . 1 L Ford engine with a CFI injection
system. The control system was the PC
system used earlier [201.
The manifold pressure trace of figure where again y = (1/A) is again to be
9b shows that the observer is working as selected to fit the desired driving mode.
intended, resulting in a small lean This control algorithm is in a number of
transient in figure 9d. The results ways conceptually ideal as it is based on
obtained in the experimental series a sensor which is very fast (rise times of
documented here in figure 9 left little -1 msec are not common) and initially
doubt as to the effectiveness of the accurate. Unfortunately hot wire systems
 Downloaded from SAE International by University Of Newcastle, Wednesday, August 08, 2018

a (degrees)
50.

Figure 10a. Fast throttle angle transient

used to test a conventional hot wire MAF
contol strategy. The rise and fall times
are 0.1 sec.

Figure lob and 10c. Response of the

simulated EFI engine and control system to
the fast input transient of figure 10a.
Note the very small lag (50 msec) due to
sensor response time. Plotted is the
actual air mass flow past the throttle
plate and the simulated sensor's
measurement of it.

Figure 10d. The finite sensor response

time and the manifold filling effects
together result in a very large error in
the exhaust lambda response. This is
unacceptable behavior in a system which is
so expensive to produce and difficult to
maintain.

are also delicate, so with the very common to measure the correct throttle air mass
EFI systems, they are in general placed in flow. It is however necessary again to
the wrong position: in front of tbe take into account the effects of anti-
throttle plate. In a? EFI system, map aliasing filters. This prevents one from
should be measured not ma, as in equation fully realizing the potential of the
(22). Another difficulty with hot wire sensor.
systems is that they are sensitive to As an example of this, figure 10 is
contamination and have a tendency to lose presented. Here are presented the results
their calibration over long periods of use of a simulation experiment in which an
[251. This difficulty will not be overall hot wire/filter time constant of
discussed further here as it is the A/F 0.02 sec is assumed. Ideal compensation
control problem which is the main interest for the fuel film dynamics is assumed for
here. simplicity. Again the mean value engine
As for the speed-density strategy, it model is presented with a fast large
is easy to show by simulation that the hot throttle angle pulse (figure 10a) and the
wire (HW) MAF system will in practice EFI engine is controlled with the standard
function somewhat less than ideally when HW A/F control algorithm. Figures lob and
provided with a sensor/filter having 10c show the measured (with the simulated
finite rise time. But here, in contrast sensor) (2) and the true manifold
to the earlier example, results are pressures and the equivalent air mass
somewhat dependent on operating point and flows. Here one can see that there is a
throttle opening speed. The reason for very slight lag in the response of the
this is that the sensors are fast enough pressure sensor and in the calculated air
 Downloaded from SAE International by University Of Newcastle, Wednesday, August 08, 2018

mass flows. This tiny difference is the 1.02. x

effect of the manifold filling dynamics.
Figure 10d makes it very clear however, 1.
that this apparently tiny effect gives an
unacceptably large lean spike ( A = 1 .I 5) .
The lean spike will become larger in 0.98
proportion with the volume V. If the
sensor/filter response time becomes as
small as 2 msec for the same engine and 0.96, I I 1
input as above, a 4% rich spike is 0 5 10 t (secl 1s
observed instead of a 15% lean one-. It is
for this reason that it is stated above Figure 1 1 . The use of an observer and
that HW systems are sensitive to details manifold filling compensation gives a
of the filtering system (as well as the useful improvement of the hot wire
engine air mass flow and throttle opening system's response to the fast input
time) . transient of figure 10a. The
This example shows that hot wire MAF sensor/filter overall response time is 20
systems do in general give better dynamic msec here.
A/F control characteristics than pressure
sensors. This is offset somewhat by
bandwidth/sampling time tradeoffs and the A
0 - Pman V
cost of the sensor. It is also apparent
that one is well advised to simulate very map - matm - -n rn

carefully the system which one desires to

build before any metal is cut (or any In order to show the effectiveness of the
microprocessor programmed). Otherwise one manifold filling the observer, a set of
can expect some very unpleasent and simulation experiments equivalent to those
expensive surprizes. above for the conventional HW system was
5.4.1 Modified How Wire MAF Systems - The performed. In figure 1 1 the operating
unfortunate state of affairs discyssed conditions and throttle input were
above can be corrected by measuring mat as precisely the same as for figure 10. The
is most often the case and correcting for sensor/filter time constant was set to 20
the effects of manifold filling to find msec. Only the lambda response will be
The mean value engine model's displayed. The estimated and actual
map manifold pressures and air mass flows are
derivation provides a hint as to how this
might be accomplished. The relationship indistinguishable on the scale which can
which is used to derive the manifold be presented here. Figure 1 1 shows that
pressure state equation is the modified system gives a very modest
1.5% lean spike and a 2% rich spike. Both
would probably be unobservable on a real
vehicle engine. Thus the observer is also
shown to be able to further improve an
already good sensor: a hot wire. The
where the expression on the far right is software costs are small.
obtained by differentiating the ideal gas
law. Using equation (23) it is easy to 6. CLASSICAL IDLE SPEED CONTROL
find
It is well-known that the main
Pman V difficulty involved in keeping the idle
speed constant is that secondary engine
loads may be applied effectively
instantaneously at any time without
where matmis now the measurement of the hot warning. This might be air conditioning
wire air mass flow meter. If the open orpowersteering loads as well as
loop observer is now written electrical loads (lighting, battery
recharging, etc). The reason that the
A
Pman= - -
I A
Pman +
R
- . '
m
a
n
matm (25a)
frequency response of the engine and its
controlling systems (fuel flow and
'man V manifold air mass flow) is so interesting
is that the bandwidth of the idle speed
equation (23) can be used to estimate controller must be made very large. It
A A
must be made large enough that nearly
both pmanand pman. An estimate for ia,can instantaneous loading or unloading of the
then be obtained from equation (24), using engine does not produce either crank shaft
the observer state and state derivative speed flare or droop. This cannot be
estimates of equation (25a). accomplished with the all-important
 Downloaded from SAE International by University Of Newcastle, Wednesday, August 08, 2018

unconditional stability requirement unless

the overall control object's frequency
response is known very accurately.
The main idle speed control problem
is to keep the crank shaft speed constant
at the desired idle speed using either a
subsidiary throttle angle actuator or an
air bypass valve. In order to apply
classical control methods to the control
of the engine speed it is necessary to
know the transfer function or frequency
response of the crank shaft subsystem. It
is also necessary to know whether or not
the manifold filling model and its
associated actuator will have an important
influence on the rate at which air can be
supplied to the engine around idle.
In order to find the requisite small
signal response the nonlinear engine model
should be linearized in the idle speed Figure 12. Linearized block diagram of
region of the engine's operating crankshaft/intake manifold portibn of the
characteristics. This has been observed mean value engine model. This subsystem
by many authors [26, 27, 281 . It is of forms the internal blocks of figures 13
course not possible to conveniently and 16 below. The block diagram is for
differentiate models which are based on the open loop engine alone: neither
engine maps and this was a significant air/fuel or idle speed control is being
difficulty in early work. applied. It is of course being assumed
The linearization method is well- here that some form of compensation for
known by now so ~t will not be reviewed in the fuel film dynamics is being applied.
detail. Instead, it is merely sketched Otherwise this input. system must be
here. It is assumed that the fuel film inserted in the input bm,.
compensation is used so that the system
order can be reduced from three to two. first order low pass filters. Of the
That is, only the crank shaft speed and cross coupling matrix elements, a,, is
manifold filing state equations need be small while a,, is large. This means that
considered. the crank shaft speed has a great
Carrying out the linearization influence on the manifold pressure (via
results in the block diagram of figure 12 the volumetric efficiency and the air
where the 6's indicate that the relevant pumping rate). Manifold pressure has only
variables are the incremental system a weak influence on the crank shaft speed
inputs and states. One should be aware (via the weak dependence of the pumping
that the block diagram presented is for
the open loop system without any controls
being applied: neither air/fuel nor idle
-
power on the manifold pressure) . Thus, a,,
0 and the system eigenfrequencies are
given by a,, and a,,. That b,, and b,, are
speed controls are being applied. The zero or small is shown in the same way
block diagram has now an interesting above. Of the system eigenfrequencies, a,,
structure which will be exploited in what is much larger than a,,, indicating that
follows. the manifold pressure reaches its
On the diagram, it ca? be seen that equilibrium much more rapidly that the
the incremental fuel flow, bm,, is the main crank shaft speed.
control of the incremental crank shaft Another important observation is that
speed, bn, while the incremental throttle close to idle with no load, the air flow
angle, ba, controls the incremental is nearly always choked (i-e. becomes
manifold pressure, bpman. The matrix sonic at the throttle plate) and become
formulation of the control problem is then independent of the manifold pressure.
This means that b,, is only a function of
the throttle angle. Moreover, even rather
large power requirements in idle ( 5 to 10%
where bx = [ bn, bp,,,] A = Vf,, of the maximum available) will not require
enough air for the air flow at the
bu = [ b&,(t-T,) , bal B = Vf, and the throttle plate to come very far from the
sonic region and b,, will remain
Jacobians are to be evaluated at the independent of the manifold pressure.
nominal desired idle speed operating This simplifies the control problem
point. It is important to realize that significantly.
both subsystems are first order lags or In order to use the linearized model
 Downloaded from SAE International by University Of Newcastle, Wednesday, August 08, 2018

it is, of course, necessary to find the a';lat

1
6n
values of the matrix elements at the --
1 6mf

linearization point which is where n = linearized

n(id1e) . It is possible to obtain a very
regulator engine/intake
accurate analytic solution to the MVEM
manifold model "man
under the assumption that the manifold
pressure reaches its equilibrium rapidly. I
One also finds that for reasonable changes
of the crank shaft speed, the matrix
element a,, is very close to being
I actuator I 1

constant. Thus one has a problem which is Figure 13. Block diagram of an engine in
very close to being a linear control the idle speed region when it is equipped
problem with a well defined (and easy to with a D. C. motor actuator system and an
find) operating point in spite of the idle speed control loop. The block on the
rather serious nonlinearity of the top right is figure 12 above. Note that
underlying problem. It would, of course, it is necessary to assume a (ideal)
have been very difficult to obtain such a fueling algorithm in order to close the
reduction of the complexity of the problem control loop in this case. This is a
if the available model did not have a reasonably accurate assumption which
simple functional form. allows the control problem to be
The "actual idle speed control simplified.
reauires a control actuator whose dynamics
must be included in the block diagram
above. The dynamics of the actuator In order to close the loop it is
depends on its type. In general, two necessary to take into account the
types are in common use at the present air/fuel ratio control running in the
time : D. C. electrical motors and background. As it is assumed that nearly
electro-mechanical air bypass valves. The ideal lambda control will be possible, the
problems of controlling both these systems connection between the incremental air and
can be rationally attacked using the fuel flow is
linearized model derived above and in the
appendix. 1
6.1 D. C. MOTOR ACTUATOR SYSTEMS - A am, = - "at
common method of controlling the idle A Lth
speed operating point is to control the
position of the throttle plate in such a
way that the engine crank shaft speed is
held constant. The position of the plate
is found via a potentiometer connected to
the throttle plate shaft. That a throttle where L wi$l in most cases be set equal
position sensor is present implies that 'mat
this system would be used as a speed- to one. = -0 as the air mass flow
throttle or speed-density system. The D. apman
C. motor is connected to the throttle is nearly independent of the manifold
plate shaft through a gear box with a pressure at small throttle angles. From
reduction ratio Re. The actuation/sensor the mFan value engine model equations, in
system is then represented by the [71 mat(a, pman)= mat1 Pl(a)P2(~rnan) +
I

differential equations Thus

'me 'me J

where the first equation is for the D. C.

motor alone and the second represents the
integral effect of the position sensor.
qe is the load torque of the return
ap,(a)
- -
aa
-
aa
a
- (1 - cos(a - a,)) -
springs, functional losses, etc. It is
assumed here that the D.C. motor has a
small electrical time constant so that it
-- 2 x

360
(a - a,) (32)

can be neglected in comparison with the

electro-mechanical one, . The block where alpha is measured in degrees and
diagram for the actuator/sensor systems is small throttle angles are assumed.
given in figure 13. The block diagram of the overall
 Downloaded from SAE International by University Of Newcastle, Wednesday, August 08, 2018

system is then as shown on figure 13. A

regulator has been inserted into the loop
in the proper location. The frequency
characteristics of the regulator and the
loop gain must now be selected in order to
\ 1 log I ~ ( j w 1)

obtain the maximum overall system band D corr. netw.

widths.
From figure 13, it is obvious that
the control system will have a desirable
integral component from the position
sensor. This, together with the low pass \
filter characteristic of the engine itself \ corr. loop
will give a frequency characteristic with \
two dominant poles. As there is an \ \ gain
important time delay in the control path
to the engine this suggest that the
feedback system bandwidth will be less
than the engine's natural bandwidth. This
is undesirable as it will mean poor
recovery response to load fluctuations.
To increase system bandwidth it is thus
desirable to use a differentiation network
to increase the phase margin. This is
indicated on figure 14. The
differentiation network will increase the
system bandwidth by one or two octaves but
it will still be insufficient to avoid
idle speed changes during heavy, rapid
load transients. This suggests that
feedforward must be established to deal
with really heavy loads such as those due
to air conditioning and power steering. Figure 14. Schematic representation of
These power sinks may require up to 5% of the frequency characteristics of the D. C.
the available maximum power of the engine motor actuator driven, idle speed control
in a small vehicle. system. The loop transfer function,
6.1 .I A Practical Idel Speed Controller H ( j o ) , has two dominant poles due to the
with a D. C. Motor Actuator - In order to engine/inertia load and to the actuator
test the concepts above an idle speed itself. A differentiation network can be
controller was realized with a PC engine used to increase the system bandwidth as
control system mentioned earlier [21]. shown dashed on the figure.
Here the engine was being controlled by a
modified air/fuel ratio controller in change. This is possible as the load is
conjunction with a linear fuel film relatively well known and constant. While
compensation network. An idle speed
control system was designed using the
techniques outlined above and the engine
was a 1 . I L FORD engine with a CFI
injection system. The idle speed
controller was tested by applying a step
load of 1 .5 kW. The results of the
experiment are shown on figure 15.
It may be observed that the correct
asymptotic idle speed is attained due to
the integral term in the regulator. It is
also obvious that the load gave an
unacceptable large crank shaft speed droop
and flare as was expected. The overshoot Figure 15. CFI engine provided with a D.
shows that the regulator is slightly C. motor actuator during a step load test.
underdamped due to the presence of the The 1.5kW load is applied in the course of
time delay. This means that it would not about 0.5 sec. The model prediction is
be wise to increase the loop gain further the smooth curve while the noisy one is
to increase the controller bandwidth: this the experimental data. The speed droop
would simply result in instability. The below the target speed (950 rpm) is seen
implication is then clear; that it will be to be too great for an acceptable system
necessary to apply feedforward from the and the use of feedforward is indicated.
load in order to reduce the idle speed
 Downloaded from SAE International by University Of Newcastle, Wednesday, August 08, 2018

perturbing the spark advance can improve

matters somewhat [28], this requires the regulator
realization of a multivariable feedback
loop which is outside the scope of this linearized
paper.
6.2 AIR BYBASS IDLE-SPEED CONTROLLERS -
More modern idle speed controllers are manifold model
based on solenoid driven valves which are
placed in parallel with the throttle
plate. In order to obtain a proportional
control, these valves are often driven Figure 16. Block diagram of an engine (in
with a pulse width modulation (PWM) signal the idle speed region) when it is equipped
which controls their opening time. The with an air bypass idle speed control
reason for this change is that PWM valves actuator. The block at the top right is
can be made less expensively than D.C. the engine/manifold combination of figure
motor systems and the response time is 12 above. It is assumed here that the
effectively instantaneous. This allows an engine is under control of an ideal
idle speed control which has a higher air/fuel control algorithm here for
bandwidth and a greater tolerance to load simplicity. This is a reasonable
fluctuations. approximation for a common system which
PWM valves can be used in CFI or EFI allows study of the isolated stability
systems with either manifold pressure or characteristics of the control loop.
hot wire controlled systems. The idle-
speed control system can then be built
around the ordinary fueling systems. Thus
a system simplification is also possible 7. CONCLUSIONS
using these components.
Assuming for the sake of simplicity A review of the operating principles
that a closed throttle plate passes no air of several of the most common engine
and that the fuel film compensator is control systems has revealed that they are
used, a block diagram of an air bypass susceptible to large transient A/F errors
valve system idle speed control system is during fast throttle angle inputs. These
shown in figure 16. The air mass flow is errors are both engine and sensor related.
simply equal to the modulation depth of While they are of a short duration, they
its drive signal so that the fuel flow to can easily give rise to engine stumbling
the engine is given simply as and misfires. This will be connected with
unacceptable emissions performance. If
conventional forms of compensation are
used to solve these problems, then the
nearly unavoidable result will be over-
fueling and poor emissions performance.
where ma, is the air mass flow through the It has been shown by simulation and
bypass valve. The dynamics of the experiment that a solution to this common
actuator are ignored as its response time problem is to include physical models of
is much smaller than the engine. As in the relevant engine subsystems in the
the earlier system block diagram, a control system. This means basically
regulator is shown whose parameters must using feedforward to compensate both for
be selected for best performance. fuel wetting of the intake manifold (or
The block diagram shows that it is intake valve area) and for the filling of
much easier to control this system than the intake manifold. Interestingly, it is
that controlled by a D. C. motor. As it also possible to compensate for an air
is desired that the idle speed is held mass flow sensor which is some distance
constant, it is necessary to use a PI from its associated injector system.
regulator. The break frequency for this There is, built into the scheme, an
regulator should be selected to be the automatic compensation for sensor
same as that of the engine. This is shown transient response and for signal
on figure 17. One must however remember conditioning and anti-aliasing filters.
that there is still the problem of the The penalty associated with the revised
time delay and this will put strong limits strategies presented is a minimum of
on the obtainable closed loop bandwidth of software. No additional sensors or
the system. This means of course that one calibration are required.
is not in reality much better off with the From the simulation studies
by-pass air actuator than with the D. C. presented, it is clear that the techniques
motor and that feedforward compensation presented can be of great assistance to
for idle speed loads will again be engine control system designers. Given
necessary. the large costs associated with acquiring
 Downloaded from SAE International by University Of Newcastle, Wednesday, August 08, 2018

and operating large experimental

facilities, simulation can provide welcome
savings in time and money. All the
simulations presented here have been
carried out on PC-AT s. This means that
the necessary dynamic simulation tools are
accessible, inexpensive and convenient.

8. NOMENCLATURE
The following symbols are used in this
paper :
t time (sec)
Pamh ambient pressure (bar)
,T ambient temperature (degrees
Kelvin)
= throttle plate angle (degrees)
rate of change of mass in
ma
the volume V (kg/sec)
mat air mass flow rate past
throttle plate (kg/sec)
air mass flow rate into port or
intake valve (kg/sec)
mfi injected fuel mass flow
(kg/sec)
mf cylinder port fuel mass flow
(kg/sec)
?ff fuel film mass flow (kg/sec)
m£" fuel vapor mass flow (kg/sec) Figure 17. Schematic representation of
X fraction of the injected fuel the frequency characteristics of the air-
which is deposited on the bypass valve, idle speed control system.
manifold as fuel film Because of the speed of the actuator, the
Lth stoichiometric air/fuel mass system has one dominant pole at a,,. In
ratio for gasoline (14.67) order to obtain a constant idle speed, a
I theoretical air/fuel ratio = PI regulator can be used as shown. The
ma/(mfLt,) bandwidth of the system cannot however be
Pman manifold air pressure (bar) increased much beyond the engine/inertia
'man manifold air temperature cut off frequency due to the engine
(degrees Kelvin) injection/torque delay.
p
h exhaust pressure (bar)
spark advance angle (degrees)
indicated efficiency
lI i
volumetric efficiency (based on va D. C. motor anchor voltage
'lvol (volts)
intake pressure) J D. C. motor and gear train
n crank shaft speed (rpm) moment of inertia (kg m)
I total momentzof inertia loading
engine (kg m ) M,e D. C. motor load and frictional
torque (N m/ (rad/sec)
p , frictional power (kW)
pumping power (kW) R. overall reduction ratio of
PP D. C. motor actuator
Pb load power (kW)
Hu fuel heating value (kJ/kg) 9. REFERENCES
Vd engine displacement (m or
liters) 1 . Franklin, G. F. and Powell, J. David,
v manifold+port passage volume Feedback Control of Dynamic Systems,
(m3) Addison - Wesley Pub. Co. , Inc. , Reading
R gas constant MA, 1987, p 495.
K ratio of the specific heats =
1.4 for air 2. Sullivan, J. A. "European I
0 angular speed of D. C. motor Applications of Ford Control Fuel
me electro mechanical time Injection Engine Control Systems", S A E
constant qf D. C. motor and Technical Paper 840546, 1984.
load (sec-)
Km D. C. motor torque constant 3. Cassidy, J. F., Athens, M. and Lee W.
(Nm/amp) H., "On the Design of Electronic
 Downloaded from SAE International by University Of Newcastle, Wednesday, August 08, 2018

Automotive Engine Controls Using Linear "Transient A/F Characteristics for Cold
Quadratic Control Theory", IEEE Trans. on Operation of a 1.6 Liter Engine with
Automatic Control, Vol. AC - 25, No. 5, Sequential Fuel 1njection" , SAE Technical
1980. Paper 880691 , 1988.
4. Hendricks, E., Holst, J., Poulsen, N. 1 5. Aquino, C. F. , "Transient A/F Control
and Joensen, H., Characteristics of the 5 Liter Central
"Adaptive Minimum Energy Control of Large Fuel Injection Engine", SAE Technical
Diesel Engines", 2nd IFAC Workshop on Paper 81 0494, 1981 .
Adaptive Systems in Control and Signal
Processing Proceedings, Lund, Sweden 1986. 16. Rasmussen, I., Emission Fra Biler
and (Emissions from Automobiles in Danish),
Hendricks, E. and Poulsen, N., "Minimum Ph. D. Dissertation, Laboratory for
Energy Control of Large Diesel Engines", Energetics, The Technical University of
SAE Technical Paper 861 191 , 1 986. Denmark, 1976.
5. Hubbard, M., Dobson, P. D. and 17. Melgaard, H., Hendricks, E. and
Powell, J. D., "Closed Loop Control Using Madsen, H., "Continuous Identification of
a Cylinder Pressure Sensor", Journ. of a Four-Stroke SI ~ngine", Paper TP10,
Dyn. Syst. Meas. and Control, Dec. 1976. 5:30, 1990 American Control Conference,
San Diego, CAI 1990.
6. Patrick, R. S. and Powell, J. D., "A
Technique for the Real Time Estimation of 18. Hendricks, E., " ~ e a nValue Modelling
Air-Fuel Ratio Using Molecular Weight and Lambda Control",
~atios",SAE Technical Paper 900260, 1990. to be published, 1991.
7. Hendricks, E. and Sorenson, S. C., 19. Matthews, R. D. and Shizhi, M.,
"Mean Value Modelling of Spark Ignition "Combustion Efficiency of Spark Ignition
Engines", SAE Technical Paper 90061 6, Enginestt,Int. Hour. of Vehicle Design,
1990. Vol. 8, No. 3, 1987.
8. Jensen, J. P., Kristensen, A. F., 20. Nishiyama, R., Ohkubo, S. and
Sorensen, S. C., Houbak, N. and Hendricks, Washino, S., "An Analysis of Factors
E., "Transient Simulation of a Small Improving A/F Control Characteristics",
Turbocharged Diesel Engine", I Mech. E SAE Technical Paper 890761, 1989.
Seminar on Engine Transient Performance,
London, U.K., Nov. 1990. 21 . Lillelund, J. and Hendricks, E., "A
PC Engine Control Development System" , SAE
9. Hendricks, E. , "Mean Value Modelling Technical Paper, to be published, 1991.
of Large Turbocharged Two-Stroke Diesel
~ngines", SAE Technical Paper 890564, 22. Kokotovic, P. V., "Applications of
1989. Singular Perturbation Techniques to
Control ~roblems",SIAM Review, Vol. 26,
10. Blanke, M. and Andersen, J. S., "On No. 4, Oct., 1984.
Modelling Large Two-Stroke Diesel Engines:
New Results from Identification", IFAC 9th 23. Vesterholm, T. and Hendricks, E.,
World Congress Proceedings, Budapest, "Practical SI Engine Observers", to be
Hungary, 1984. published, 1991.
1 1 . Houbak, N., SIL - A Simulation 24. Harrington, D. L. and Bolt, J. A.,
Language, Springer - Verlag, Lecture Notes "Analysis and Digital Simulation of
in Computer Science, No. 426, NY., NY., , SAE Technical Paper,
Carburetor ~etering"
1990. 700082, 1970.
12. Hendricks, E., Sorensen, S. C. and 25. Sauer, R., "Hot-Film Air Mass Meter-A
Holmskov, J. , "Compact Modelling of SI Low Cost Approach to Intake Air
Engines", to be published, 1991. ~easurement", SAE Technical Paper 880560,
1988.
13. Aquino, C. F., and Fozo, S. R.,
"Steady State Transient A/F Control 26. Washino, R., Nishiyama, R. and
Requirements for Cold Operation of a 1.6 Ohkubo, S., "A Fundamental Study of the
Liter Engine with Single Point Fuel Control of Periodic Oscillation of SI
Injection", SAE Technical Paper 850509, Engine Revolutions", SAE Technical Paper
1985. 86041 1 , 1986.
14. Fozo, S. R. and Aquino, C. F., 27. Powell, B. K., Cook, J. A. and
 Downloaded from SAE International by University Of Newcastle, Wednesday, August 08, 2018

Grizzle, J. W., "Modelling and Analysis of

an Inherently Multi-rate Sampling Fuel
Injected Engine Idle Speed Control ~ o o p " ,
Paper No. FA2 10:00, 1987 American Control
Conf., Minn. MA, 1987.
28. Morris, B. K. and Morris, R. L.,
"Modern Control Applications in Idle speed
Control", Paper No. WA4 11:45, 1983
American Control Conf., NY, NY, 1983.