Vous êtes sur la page 1sur 15

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART C: APPLICATIONS AND REVIEWS, VOL. 37, NO.

6, NOVEMBER 2007

1137

Modeling, Identification, Design, and


Implementation of Nonlinear Automotive
Idle Speed Control SystemsAn Overview
Zhengmao Ye, Senior Member, IEEE

AbstractAutomotive idle speed control (ISC) is one of the most


challenging aspects in engine control fields. Essentially it is a highly
nonlinear, time-varying, complicated, and uncertain dynamic control problem. As the typical automotive transient operation, the
quality of the ISC has a significant impact on fuel economy, emission levels (HC, CO, and NOx), combustion stability, transient
response, and noise, vibration, and harshness characteristics. This
paper presents a thorough review of various dynamic control technologies which have been successfully applied to ISC systems. In
particular, practical implementations on a variety of different engine types are provided, which cover broad areas of control, including classical control, modern control, and intelligent control.
Over 90 selected papers published during the last two decades
are reviewed and then summarized from a control point of view.
These control approaches can be generalized to the control of other
automotive, electrical, mechanical, and aeronautical systems.
Index TermsAdaptive control, artificial intelligence, fuzzy control, intelligent control, internal combustion engines, neural networks, nonlinear systems, optimal control.

A/F
BDC
BIBO
BP
GA
GDI
IAC
ISC
LQ
LQG
LQI
LQR
LSE
MAP
MIMO
MRAC
NVH
PID
PFI
RBF
RLS
RPM

NOMENCLATURE
Air to fuel (ratio).
Bottom dead center.
Bounded input bounded output.
Backpropagation.
Genetic algorithm.
Gasoline direct injection (engine).
Idle air control.
Idle speed control.
Linear quadratic.
Linear quadratic Gaussian.
Linear quadratic integral.
Linear quadratic regulator.
Least squares estimation.
Manifold absolute pressure.
Multiple input multiple output.
Model reference adaptive control.
Noise, vibration, and harshness.
Proportional, derivative, and integral.
Port fuel injection (engine).
Radial basis function.
Recursive least squares.
Revolution per minute.

Manuscript received January 9, 2006; revised March 20, 2006 and July 6,
2006. This paper was recommended by Associate Editor P. Horacek.
The author is with the Department of Electrical Engineering, Southern University, Baton Rouge, LA 70813 USA (e-mail: zhengmaoye@engr.subr.edu).
Digital Object Identifier 10.1109/TSMCC.2007.905810

SSR
STR
SVD
TDC
TWC

Sum of squared residues.


Self-tuning regulator.
Singular value decomposition.
Top dead center.
Three-way catalyst.

I. INTRODUCTION
UTOMOTIVE idle speed control (ISC) is a multiobjective
and multivariable control issue. The selection of target idle
speed corresponds to a tradeoff among fuel consumption, idle
operation stability, and emission levels. The idle speed control
mechanism of different engine types varies. For a typical port
fuel injection (PFI) engine, the ISC system uses a controlled
bypass valve to regulate the flow rate of bypass air around a
closed throttle, using a solenoid, stepper motor, or controlled
duty-cycle valve. The throttle bypass valve is used to adjust the
intake manifold air flow around the primary throttle plate at idle.
Its secondary function is to prevent stalling and produce smooth
throttle tip in and tip out by providing extra air during idling
and acting as an electronic dashpot during sudden deceleration.
Another category of ISC systems uses a direct throttle actuator. The air volume at idle comes through a slight throttle
opening, which is controlled by a dc motor, stepper motor, or
solenoid. On average, about 30% of fuel consumption in a city
driving is spent at idle [36], and it continues to increase with
increasing traffic levels.
A reasonable target idle speed needs to be maintained in
order to overcome mechanical frictions, misfiring, and load disturbances to prevent engines oscillation, vibration, hunting, and
stalling under a variety of circumstances. At idle, both rich air
to fuel (A/F) mixture and incomplete combustion are present
due to in-cylinder residual gas fraction and low engine speed,
giving rise to higher levels of HC and CO emissions. Basic
requirements for ISC systems are listed as follows.
1) Low idle speed set point is needed for good fuel economy.
2) Low idle speed set point is needed for the reduction of HC
and CO emissions.
3) The ISC stability is of the upmost importance. Regarding
target idle speed, if it is too low, stalling may occur; if it
is too high, harsh gear engagement may occur.
4) ISC nonlinear delays need to be compensated for.
It is clear that some requirements are conflicting. As a result,
multiobjective modeling, control, and optimization for dynamic
ISC systems are necessary. From literature surveys, virtually all
existing control theories have been successfully implemented to

1094-6977/$25.00 2007 IEEE

1138

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART C: APPLICATIONS AND REVIEWS, VOL. 37, NO. 6, NOVEMBER 2007

solve the ISC dynamic problems. The ISC is so fundamental that


nearly all areas of control fields are encompassed. Therefore, a
well-prepared control engineering researcher should be familiar
with the practical aspects of classical control, modern control,
and intelligent control; a well-prepared automotive researcher
should be familiar with idle speed control problems of PFI
engines, gasoline direct injection (GDI) engines, diesel engines
and potentially natural gas engines, and hybrid vehicles.
To optimize engine operations during idling, miscellaneous
load disturbances must be taken into account. In general, there
are four basic types of external disturbances: air conditioning,
power steering, gear shifting between neutral and drive positions, as well as the alternator.
1) The accessory load from air conditioning is about 10 Nm
for most midsize vehicles, which roughly equals to 10%
of the maximum torque or is around 20% of torque at idle.
2) With the presence of power steering, it acts as another
major disturbance. The peak steering wheel load occurs
at low-vehicle speed, especially when a steering wheel
reaches its full travel limits.
3) The alternator provides the power, which supplies all
electrical components (e.g., ECU, electric fan, defrosters,
power locks, power windows, headlights, radio, etc.) on a
vehicle, while battery charging is needed to store energy.
Alternator load is normally 24 Nm [24].
4) During transient engagements of the transmission, gear
shifting between neutral and drive induces an additional
load disturbance to engine ISC dynamic control systems.
5) Moreover, an ISC system should be robust against varying
ambient temperature, barometric pressure, fuel composition, lubricant type, air humidity, cycle-to-cycle combustion variations, manufacturing tolerances, and component
aging. ISC systems should also be applicable across different engine types [1].
Engine coolant temperature is a dominating factor in the target
idle speed set point determination. The analysis of an ISC system
can be conducted either on a time basis or on a crank-angle basis.
For instance, the differential equation for intake manifold filling
dynamics can be formulated in a time domain, while engine
rotational dynamics can be formulated as a function of either
time or crank angle.
A sound former survey of ISC dynamic systems has been
made [33], with a focus on applications of idle speed control
problems to PFI engines at Ford Motor Company using many
modern control theories. In view of the diversities of both engine types and control methodologies, its scope, is however,
relatively limited, especially compared with the rapid development of engine and control technologies.
An orderly sequence is conformed in this review of over
90 publications. In terms of engine types, a survey on ISC of
PFI engines is conducted at first, followed by GDI engines,
diesel engines, and natural gas engines. In terms of control theories, methodologies are discussed among which are classical
control, modern control, and intelligent control, respectively.
Classical control is based on transfer function methods; modern
control is based on state space methods and intelligent control includes fuzzy logic control, artificial neural networks, and

genetic algorithms. Fuzzy logic is rule based, whose knowledge base consists of fuzzy ifthen rules characterized by a
set of membership functions. Neural networks employ learning
algorithms to modify weights on the connection links among
multiple layers. In most cases, hybridization has to be used to
enhance control performance, such as in ISC fuzzy PID control. Sometimes there might also be no distinctive borderlines
between different control types. As an example, neither optimal
control nor adaptive control can be grouped as classical control
or modern control exclusively.
II. PFI ENGINE IDLE SPEED CONTROL APPROACHES (AZ)
It is inevitable that engine dynamics is related to nonlinear
phenomena, such as fuel transport lag, intake charging delay,
and induction-to-power stroke delay. A well-known fact is the
wall wetting effect. These phenomena are induced by both periodic speed fluctuations and aperiodic load disturbances. Periodic speed fluctuations can be examined by the intake manifold
filling dynamic, combustion dynamics, and engine rotational
dynamics. Idle speed quality degradation can arise from the
combustion roughness and aperiodic loads imposed by air conditioning, power steering, alternators, and neutral-drive transitions [6], [55].
A. ISC Feedforward Control
An ISC system with the feedforward compensation has been
designed at Mitsubishi [35]. With the presence of alternator load
current, the feedforward compensation method is derived based
on a linearized engine model. It is used to suppress the idle
speed fluctuation generated by some electric load disturbances.
This ISC dynamic model consists of alternator dynamics, intake manifold filling dynamics, and engine rotational dynamics.
In feedforward control, a voltage regulator is used to change
the field current by adjusting the intake air flow rate. All control parameters of this ISC system are identified using sensing
information from the airflow sensor, hall element load current
sensor, and so on. As a result, idle speed fluctuation has been
suppressed. Under measurable disturbances from air conditioning and power steering, the ISC feedforward controller can also
be used for load disturbance rejection [17], [28], [87], [88].
Feedforward control is simple to implement at a low cost. As
an open loop approach, its control authority is limited. The ISC
feedforward controller is usually combined with some other control approaches, such as optimal control and neural networks, to
further improve the quality of an ISC system.
B. ISC Fundamental PID Control
Engine speed oscillations are caused by the pulsating nature of the engine torque [27], due to fuel detonation in the
cylinders. In idle regime, these oscillations are of special concern when the inertia of the vehicle is not connected to the
engine. Damping of idle speed oscillations is considered using
a reversible alternator. This reversible alternator applies control
torque to the crankshaft to improve the uniformity of the speed.
Practical parameterizations of controllers are conducted by

YE: NONLINEAR AUTOMOTIVE ISC SYSTEMSAN OVERVIEW

Fig. 1.

1139

Idle speed PID controller implementation [31].

choosing parameters to satisfy the performance specifications




1 T 2
1 T 2
v (t)dt;Jq =
q (t)dt.
(1)
Jv =
T 0
T 0
PID and Youla parameterization are used to evaluate the performance metrics, such as the mean square of velocity and the
mean square of control torque in (1). Engine idle speed oscillations can be substantially reduced using this method. It also has
the potential to reduce engine size and weight by using a lighter
flywheel. The major advantages of PID parameterizations lie in
the simplicity and the low dynamic order, while the stability is
not guaranteed. Youla parameterization provides the guaranteed
internal stability, while the selections of rational stable functions
are not unique. Thus, software package has been developed to
evaluate the performance on the design charts.
PID tuning systems are widely used with no need to obtain a dynamic model [31] (see Fig. 1). Continuous-action
reinforcement-algorithm-based online PID tuning has been
tested on a Ford 1.8-L engine. It is used to simultaneously
tune the parameters of the three-term controller to minimize its
performance objective. The cost is evaluated online according
to time and speed error during transient operations. This technique does not require a priori knowledge of system dynamics.
It can provide the optimized control to the complex nonlinear
systems. PID control of the ISC systems can be improved by
combining with fuzzy control [79]. To avoid engine stalling
and to suppress engine speed fluctuations upon switching of air
conditioning (on/off), a fuzzy PID controller has been applied
onboard to a 1.3-L PFI engine. The magnitude of the speed
fluctuation is substantially reduced. The ISC PID control is regarded as a reliable approach, where low control complexity is
involved. On the other hand, there is no general rule to follow
for parameterizations of the three-term coefficients.

where e(k) is engine speed difference, u(k) is the increment


of u(.) from its nominal operating point, and is a positive constant. The LQI tracking controller [70] contains an observer by
which state variables are estimated from this observer dynamic
model, thus control variables are determined in turn. This simple
design is easy to implement with no extra sensor requirement. At
transient states that are subject to various torque disturbances,
the prompt response tracking can still be maintained. For a
complex LQ engine model with five state variables, additional
spark advance feedback loop is used [89]. Spark timing control
improves idle speed stability by compensating for the delay in
torque generations [2]. It has been indicated that the performance of both LQ optimization and fuzzy control is superior
to that of regular PID controllers in practical implementations.
An experimental comparison of the ISC PID controller, linear
quadratic regulator (LQR), inputoutput linearization, and sliding mode controller is made via an electronic throttle control
system [39]. PID controller is a reliable approach, which provides the proper tuning time and adequate control performance.
LQR controller gives rise to better fuel economy than that of
PID control. Both PID and LQR are subject to the disturbance
rejection and reduction of speed variations. That is, the reason
that nonlinear approaches are introduced where inputoutput
linearization is presented to solve this problem. Continuous
sliding mode control gives rise to good fuel economy and disturbance rejection capability against additional load disturbances
from power steering pumps and air-conditioning compressors.
However, the induction-to-power delay and finite response time
problems still need to be solved. LQ control approach is not
difficult to design from a control point of view. It also results
in relatively good control performance, while its performance
depends on the accuracy of the ISC modeling and performance
index formulation.

C. ISC Linear Quadratic (LQ) Optimization


Another traditional ISC scheme is the LQ optimization. In
modern control theory, an exact model is required, so control
methodologies can be very complex. Accordingly, a relatively
simple linear quadratic integral (LQI) control approach has been
proposed. Its performance index is defined as

[e2 (k) + u2 (k)]
(2)
J=
0

D. ISC Optimal Control and L1 Optimization


Engine dynamic systems at idle can be modeled as nonlinear
ordinary differential equations with some variable-time-delay
elements (e.g., time constant in a wall-wetting model). The ISC
nonlinear optimal control problem has been formulated [69],
which employs a quadratic function as the performance index
in terms of engine speed, throttle angle, and spark advance, with

1140

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART C: APPLICATIONS AND REVIEWS, VOL. 37, NO. 6, NOVEMBER 2007

different weighting functions


 T
2 }dt
J=
{(N N 0 )2 + ( 0 )2 + ()

(3)

where N is engine speed, is spark advance angle, is throttle


angle, N 0 and 0 are set points. Mathematical programming is
proposed based on the projected Newtons method. Backward
Euler scheme is used for difference equation approximation on
the differential algebraic systems. This optimal approach can
bring about suitable idle speed control design based on engine
models. Compared with the PID controller, the settling time is
reduced to a half using the optimal control. Robustness of this
controller should be investigated considering uncertainties of
engine models. Feedforward control and feedback control are
usually applied simultaneously [17]. For an application on a
4.6-L V8 Ford engine, L1 feedforward control is selected
for the transient tracking, while the linear quadratic Gaussian
(LQG) optimal feedback control is designed for the steady-state
tracking. The set-point tracking and closed loop stabilization depend on the LQG design and integral feedback controller. This
L1 optimization seeks for a transfer function that minimizes the
tracking term (L1 norm) of the closed loop transfer function. L1
optimal design method uses an internally stabilizing controller
to reduce the peak speed excursion by load disturbance rejection. The coordination of spark ignition control and A/F ratio
control will enhance the dynamic performance. The frequency
response of the L1 feedforward control functions as a lead compensator for time delays of ISC systems. It is shown that a L1
optimal feedforward controller satisfies the transient tracking
requirement.
E. ISC H Control
Compared with the LQ optimization, the H synthesis can
provide an accurate modeling via system identification. Appropriate weighting functions are crucial to the H control
design. This ISC H design approach [77] has been applied
to a typical V6 engine. Control variables, such as the coolant
temperature, mass air flow, and engine speed, are monitored to
determine the actual A/F ratio, mass fuel flow, spark advance
angle, and bypass air flow. The aim of this H controller is
to achieve desirable performance, where the controller is robust against the parameter perturbations. This H design cycle
involves the specification of weighting functions, synthesis of
controller, controller model reduction, and closed-loop system
evaluation. The key point for the H design is the iteration of
weighting functions for bypass valve control, ignition control,
and control sensitivity formulation. The rule of weight changing is to provide a broad bandwidth by penalizing the ignition
timing less at low frequencies and penalizing the idle air flow
less at high frequencies. Integral action is introduced into the
idle valve control loop, so as to keep steady-state spark timing
for the engine combustion. The simple reduced model consists of only three state control loops; however, performance
and robustness are guaranteed. Another ISC H method [90]
requires trial-and-error weight adjustments to penalize the excessive actuator actions and sensitivity terms . Its controller

consists of a bypass valve frequency-shaped PI controller and a


spark advance PD controller. The focus of H controller is the
robustness of ISC systems. From testing results, this ISC H
controller has produced a slightly faster response than the LQ
controller.
F. ISC Synthesis Control
The feasibility of a synthesis robust ISC design has been
investigated on a 2.9-L Scorpio engine model [32]. synthesis
control acts as a natural extension of H control. The inputs of
bypass valve opening and spark advance are used to maintain
the desired idle speed. Its basic idea is as follows: for an ISC
nominal plant and a perturbed plant with several delay terms,
transfer functions of the spark advance bypass valve opening
and sensitivity are designed in advance
u = diag{1 Ir 1 , . . . , r Ir r , . . . , 1 , . . . , F }.

(4)

By using the singular value decomposition (SVD) method on


the statespace representation of error signals between nominal
and perturbed functions, all singular values can be determined.
Then, both structured uncertainty matrix and controller matrix
are subject to examination. Particularly, structured uncertainty
matrix is expressed in (4), where () denotes maximum singular value of , i represents repeated scalar complex uncertainties, i represents full complex block uncertainties. The
overall matrix consists of the scalar repeated complex uncertainties (individual diagonal matrices) and full complex uncertainties (individual block matrices). For each individual matrix,
the maximum singular value is calculated. Afterward, every
maximum singular values (upper bound) from each individual
matrix are computed, where is defined as the reciprocator
of the minima among all maximum singular values. The optimal is determined via iterations to guarantee the robustness.
Hence, the synthesis technique overcomes the influences of
perturbations and uncertainties through the combination of separated models. The superiority of an ISC synthesis controller
over a H controller is that being a robust design, synthesis control also leads to the low idle speed set point. In
the mean while, its formulation is more complicated, however
worthwhile.
G. ISC Robust Control
Uncertainties of ISC nonlinear systems are classified into two
categories: internal varying factors (slowly and rapidly) and external disturbances. The robustness analysis of the ISC stability
could be conducted using analyzing tools [1], where the robust
stability issue based on the polynomial formulation is converted
to a problem whether the roots locate at the left half complex
plane or not. The block diagram of a linearized engine model
contains intake manifold filling dynamics, combustion dynamics, and engine rotational dynamics. The regular engine model
is simplified using the signal flow graph analysis and the Mason
rule, which results in two inputs (throttle valve duty cycle, spark
angle) and two outputs (engine speed, intake manifold absolute
pressure). Considering the uncertainty issue, the Fiat intermediate value models are used to coordinate with the engine models.

YE: NONLINEAR AUTOMOTIVE ISC SYSTEMSAN OVERVIEW

The integral feedback action on the speed error signal is used


to ensure the zero-steady-state deviation while, the derivative
compensator for the spark advance is to ensure that fast corrections are made with no state offset. A closed-loop characteristic
polynomial is used for this robust stability analysis. With the
presence of a digital controller, the hybrid system is generated
by combining the continuous engine models, thereby sampling
rates are determined. A recent study of the robust idle speed
regulation is based on the directly identified nonlinear inverse
of a stable system [56]. Identification is conducted for ISC problems using linear robust control techniques in parameter spaces.
Genetic algorithms are then introduced into robust control systems to evaluate error performance functions. The ISC stability
is maintained by robust control approaches against uncertainties and load disturbances, where classical control theories have
also been applied to improve other aspects of the ISC system
performance.
H. ISC Least Squares Estimation (LSE)
Suitable idle speed controllers depend on the proper engine
modeling. Models should not only represent highly nonlinear
and complex engine dynamics, but also be simple enough for
the control synthesis. In a nonlinear identification model, the
intake air actuator position and spark advance angle are inputs
while intake manifold pressure and engine speed are outputs.
The LQ technique is applied to the model synthesis of a 1.4-L
commercial engine [54]. Assuming that N observations have
been measured, a regressive LSE approach has been used so
that the variable predictions can be made simultaneously. The
sum of squared residuals (SSR) will be minimized for the optimal estimation. This method saves up to 50% of the tuning
time of the controller gain. It is shown that this approach can
be extended to the more robust nonlinear models. For an ISC
adaptive fuzzy control design, its direct adaptive law is formulated using the least mean squares, from which satisfied results
are obtained. The linearized engine ISC model can be identified using a generalized least squares method with respect to
the nominal operating points [29]. The identified parameters are
then used as initial values of a linear quadratic optimal controller. This optimal controller is applied to a Chrysler 2.0-L
neon engine. The idle speed set points have been selected as
840, 780, and 720 rpm, respectively. Experimental results indicate that the optimal controller has the capability of effectively
adjusting engine idle speed under various types of load disturbances, such as from a cooling fan and power steering. Tracking
errors are substantially reduced via optimal control, but it takes
longer time. The least squares estimation is simple, and there
is no computation complexity for applications in high-order
cases. It can also be used for prediction control. The restriction of the LSE applications is that ISC systems are highly
nonlinear, in most cases the linearization can not be reached
directly.
I. Model Following Control and Gain Scheduling Control
Due to its wide adaptability and robustness under various
environmental conditions, model reference adaptive control

1141

(MRAC) acts as an effective ISC control approach. However,


complex control laws are required for MRAC implementation.
On the other hand, a model following control could be much
simpler by combining with a torque estimation model and an intake air dynamics model [75]. So the torque model and collector
filling model are proposed and included into a model following
ISC system. The opening of the air passage, which bypasses the
throttle valve, is controlled to suppress engine speed variations.
This ISC model with estimated torque signals can provide a
quick response to load disturbances, so that the feedback control
systems can follow the model output rapidly. The phase difference due to intake nonlinear dynamics has been compensated
for. Using model following control onboard, ISC performance
has been improved under different load disturbances of power
steering, air conditioning, or headlights.
The engine hunting phenomenon is related to vibrations at
a fundamental frequency [12], ranging roughly from 7 (800
rpm) to 50 Hz (6000 rpm). The ISC stability against harmonic
disturbances at a time-varying fundamental frequency is a direction of active control of engine vibrations. The standard state
feedback control can be employed in this case. A load disturbance can be modeled as the output of statespace model and
its time varying state observer is used to estimate states from
the actual model. The disturbance is then remodeled and used
as a control input. Thus, an alternative nonadaptive algorithm is
developed for the active vibration and noise control against load
disturbances at harmonic frequency. Time-varying frequency is
used as a scheduling variable. The stabilizing observer gain is
calculated from the observer gain scheduling mechanism. It is
shown from testing that the engine vibration is reduced significantly at idle.
It is not difficult to design the ISC model following control
and gain schedule control algorithms. However, the resulting
ISC performance strongly relies on the modeling accuracy.
J. ISC Fundamental Adaptive Control
The primary goal of an ISC system is to maintain idle speed
operations at a feasible minimum speed [51] without affecting
the ISC quality and emissions. By adjusting operating points
adaptively in response to operating conditions and environmental conditions, emission levels, fuel economy, and roughness
level (idle quality) can be improved. An ISC system in the
pioneering research consists of four feedback loops: load compensation loop, speed control loop, roughness control loop, and
adaptive roughness control loop. Engine speed is controlled by
adjusting the throttle position through a stepper motor. The initial roughness set point is modified by the adaptive roughness
loop to adapt to operating conditions and environmental conditions. The response from the inner adaptive roughness loop
is slow in comparison with other loops of the ISC system to
modify the roughness set point according to the slowly varying parameters, e.g., ambient temperature. The second loop is
to reduce the roughness level. The third loop is much faster,
which keeps the engine idle speed at a specified value. The
outer loop sends out the command to the throttle actuator for a
fixed step increment. Optimal operating points with respect to

1142

Fig. 2.

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART C: APPLICATIONS AND REVIEWS, VOL. 37, NO. 6, NOVEMBER 2007

Idle speed adaptive control model [51].

the roughness level and engine speed can be maintained despite


of the variations in operating conditions. In a recent study, spark
ignition control depending on the adaptive strategy has been developed to stabilize the idle speed fluctuations [76]. Adaptation
of the spark advance angle gives rise to a tradeoff between the
operating speed and estimation accuracy. The success of this
adaptive controller has been indicated. An adaptive controller
tracks the engine speed set point tightly, and it is robust to load
disturbances. The adaptive control flowchart (Fig. 2) shows a
well designed ISC control approach. It in fact reflects various
demands of ISC systems, whose physical mechanisms are also
straightforward. The drawback lies in that its complexity is increased and its optimality is not guaranteed. Nevertheless, this
approach has provided a general picture what actually an ISC
system is about.
K. ISC Controller and Observer Design
A group of uncertain time delays of an ISC system has been
investigated [10]. This dynamic control system is modeled as
an observer-based controller with delays. The linear matrix inequality is formulated in both the delay-independent and delaydependent cases. Sufficient conditions regarding the existence
of the controller and observer matrices are obtained to guarantee
the asymptotic stability in a former case. A standard toolbox is
used to find the solutions to the inequality in a latter case, when
the observer-controller design is applied to the ISC system
 h
Z T ()P1 Z()d.
(5)
V (z, t) = Z T P0 Z +
th

As a result, by means of the combined controllerobserver


design methodology, the ISC system is guaranteed to be asymptotically stable with the feedback using estimated states. Furthermore, dynamics of the state estimator is guaranteed to be
asymptotically stable. Its stability issue can be tested by defining
quadratic and radial unbounded Lyapunov functions (5), where
P0 and P1 are two positive definite matrices. The aforementioned design of observers and controllers will give rise to a

low nominal speed. This ISC formulation stems from modern


control theories, where the ISC performance depends on both
the controller design and the observer design.

L. ISC Design via State Feedback and Output Feedback


A nonlinear model is designed for the calibration of an ISC
system where the full state feedback is applied, considering
the nonlinearity of engine dynamics at idle [52]. The nonlinear
model is used for the ISC design and for prediction through
a broad range of the operating conditions. Under load disturbances, this ISC system takes advantage of both spark advance
adjustment and idle air actuating to reduce variations of the idle
speed. The intake manifold filling time delay is modeled as a
function of engine parameters, idle air actuator position, ambient pressure, and temperature. Engine torque is a function of
manifold absolute pressure (MAP), engine speed, spark, fuel,
and a time delay in combustion dynamics. Rotational dynamics results from the inertia of the mechanical systems. Each of
them is represented as the first-order system. All together the
ISC system is formulated as a nonlinear MIMO model. This
simple nonlinear engine model is applied to the optimal controller design to enhance the capability of disturbance rejection
during gear transitions (from/to idling). The prediction for the
idle speed on a 4.3-L GM V6 engine has been made. The range
of the idle air control (IAC) actuator position is from 30 to
60 steps. The range of the spark advance angle is from 0
30 . Using the full state feedback control, the results indicate
that the prediction error of the idle speed is very small (10
rpm) under different operating conditions. In a similar case,
an ISC control law for the real-time applications is investigated [26], where the finite time delays between the inputs
(MAP, spark advance angle) and output (torque) are counted
explicitly. To maintain a target idle speed against the nonmeasurable torque disturbances of power steering or air conditioning
using the output feedback, the bypass valve position, and spark
advance angle are used as control signals, where two time delays
(intake-to-power stroke delay, toque generation lag) are taken

YE: NONLINEAR AUTOMOTIVE ISC SYSTEMSAN OVERVIEW

into account. To allow for real-time applications, the control


law is discretized according to the sampling rate of the measuring signals. This linear time-invariant control law has shown
adequate performance. It is also robust against the parameter
perturbations.
In order to design the state feedback and output feedback,
the assumption is that an ISC system represents a linear system,
where all linear control theories can be applied. In fact, these
two approaches can only be applied to some simple cases via
linearization, although the state feedback and output feedback
have the accurate mathematical formulations.

1143

is chosen (7) according to the torque excursion and A/F ratio


excursion. This design is used to evaluate the dynamic programming approach in the nonlinear environment. Satisfactory transient performance of the ISC neural network controller can be
reached. In addition, the real-time adaptation capability allows
the neural network controller to be further refined and improved
by continuous learning and adaptation. The trained controller
has the potential to outperform the existing controllers. The ISC
performance optimization can be accomplished using discrete
dynamic programming in a finite time horizon. However, the
cost function being selected will affect performance of the ISC
system, which varies among different designers.

M. ISC Lyapunov Function Approach


A variable structure control scheme has been proposed for
ISC systems under the unknown time-varying disturbances [68].
Control laws for the throttle opening and spark advance are
designed individually to tune the idle speed. The throttle opening
is used as a primary source of control for the torque generation.
Spark timing control is to compensate for the intake-to-torque
generation delay
Je 2 1 2
+ ep .
(6)
2
2
The disturbance estimator is designed to reconstruct the
torque disturbance. The stability of this system is proved via
a Lyapunov function method. Its control objective is reached
by constructing a Lyapunov function candidate, consisting of
the quadratic terms corresponding to engine speed and engine
load (6), where is engine speed (rad/s), Je is engine inertia
moment, eP = (p/p0 ) (p/p0 ) , p and p0 are engine load and
atmosphere pressure.
A Lyapunov function serves as a candidate itself to analyze the
stability issue. It has been applied to a neural network controller
or a discrete adaptive sliding mode controller for the idle speed
regulation, which authorizes the controller to be adaptive to
the system parameter uncertainties and external disturbances.
In fact, the Lyapunov function can not be used independently
for ISC systems. It must be applied together with other control
approaches.
V =

N. ISC Nonlinear Dynamic Programming


Multiple objectives should be taken into account at the same
time for ISC systems [42], e.g., engine performance, emission
levels, and robustness under different operating conditions. Discrete dynamic programming can be applied to simple and small
scale control problems, whose cost function is defined as the
summation of a utility function from present to future. Using
testing data from a 5.3-L GM V8 engine, a neural network
model is developed, and then dynamic programming is combined with the neural network controller to obtain optimization
via self-learning
1
1
[T (t) T (t)]2 + [(A/F ) (A/F ) ]2 .
(7)
2
2
Both engine torque control and exhaust A/F ratio control are
conducted to track torque command and regulate A/F ratio. The
cost function of the engine torque and A/F ratio tracking control
U (t) =

O. ISC Sliding Mode Control (Continuous, Discrete)


To solve the engine induction-to-power delay problem, a continuous sliding-mode control approach is presented for an engine control system with time delays, which has been applied
to the upstream side of the idle actuator of a 4.6-L Ford V8
engine [45]. A linear transformation is conducted in order to
transit a delayed system to a delay-free system, where all the
unstable poles of original systems are involved. Sliding mode
control is designed based on the delay-free system. A suitable
control law is selected to satisfy the global stability of a switching surface S = 0, where a Lyapunov function is defined as
V = ST S/2, whose derivative term is negative so as to be stable.
Its asymptotic stability is ensured by a delay-free system and
a delayed system under bounded and matched disturbances. A
state observer is also presented to estimate both intake manifold
pressure and engine torque. Solid disturbance rejection capability has been observed. Another adaptive sliding-mode control
design is applied to a discrete nonlinear system, where the three
layer feedforward neural networks are used to approximate unknown dynamics, which is tuned by a sliding surface approach
using gradient methods [43]. The control goal is to lower the
target idle speed and reduce speed variations against load disturbances. The target idle speed is set to be 611 rpm for the
concern of fuel economy instead of the production idle speed of
740 rpm. A discrete sliding surface is defined as a stable polynomial of system tracking errors. The sliding mode control laws are
designed as: s(k + d) = C(q 1 )e(k + d), where q 1 is the unit
delay operator; e(k + d) = y(k + d) r(k + d), which is the
speed tracking error and C(q 1 ) = 1 + C0 q 1 + +CL q L ,
which is the Hurwitz polynomial. The state trajectories of the
discrete mode are shown to converge within a small sliding sector, which actually deviates a little from the continuous sliding
manifold except for all sampling points. The state trajectories
are initialized along the sliding surface when the system is at
the sliding mode. When a disturbance is applied, the state trajectory returns to the surface after a short transient period and
remain there until the load is released. The time delay problem
is formulated as a high order system. This ISC controller has the
capability to maintain a low idle speed against load disturbances
ss < 0. |s(k)| . |s(k + 1)| < |s(k)| .

(8)

A discrete sliding mode controller based on the reachability


condition is also applied to a 2.4-L DaimlerChrysler engine.

1144

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART C: APPLICATIONS AND REVIEWS, VOL. 37, NO. 6, NOVEMBER 2007

System models are obtained via nonlinear state transformation


and local linearization [47]. Both the reachability condition and
control laws (8) should be satisfied. This design guarantees that
the system motion reaches the sliding manifold smoothly and
remains on it for a finite time. With regard to the speed tracking
performance and disturbance rejection, sliding mode control is
found to be superior to PID control and LQG control [37]. Considering the nonlinearity of ISC systems, a sliding mode controller provides more robust performance when it is combined
with nonlinear H robust control. For the robustness issue, the
ISC sliding mode control leads to satisfied results, which is better than most regular control approaches. On the other hand, its
application depends on other control methodologies.
P. ISC Nonlinear Describing Function Approach
Power steering is a common nonmeasurable external torque
disturbance in an ISC system. The desired idle speed of a Ford
V8 engine is set to be 600 rpm, whose allowable tolerance is
140 rpm. The input to the ISC system is a voltage signal. It is sent
to the bypass air valve that is bounded by 0.8 V. For steady-state
operations, inputs are not subject to saturation. For transient
states, saturation must be considered [30]. In this ISC system,
air control and ignition control are used in parallel to reach the
desired idle speed. The ignition control loop usually has a much
wider bandwidth than the air control loop. However, the control
authority of the air control loop is much greater than that of the
ignition control loop. Consequently, the bypass air valve that
regulates the air flow into the engine is applied to the ISC system. Also, actuator saturation is considered to achieve enhanced
regulation so as to satisfy stringent performance specifications
1
N (A) = {2 sin1 (B/A) sin[2sin1 (B/A)]}

4
+ (B/A)cos[sin1 (B/A)].

Q. ISC Adaptive Kalman Filter and Extended Kalman Filter


An adaptive Kalman filter has been proposed to design the
engine load torque estimator [19]. It can be extended to the second order nonlinear stochastic model of the torque, based on a
linear inertia model. Kalman filter has been used as a part of the
load torque compensator combined with the idle speed PI controller and linearized engine model. This adaptive Kalman filter
can provide the stable response and disturbance rejection. It has
the advantage in suppressing the engine speed fluctuation after a
sudden change of the engine load disturbance. Kalman filter has
also been applied to the recurrent neural network training [59].
An engine ISC model selects the throttle angle and spark advance as inputs. The measured outputs of this ISC system are
RPM and MAP, both of which are recorded four times per engine revolution. Outputs are easily corrupted by Gaussian noise,
so the objective of this ISC system is to maintain the idle speed
at 750 rpm against the unmeasured torque disturbances. The
extended back propagation is selected as the real-time recurrent
learning algorithm for neurocontroller training. The dynamic
gradients of the on line training algorithm are calculated by
the truncated back propagation to update both derivatives and
weights in a real time. The decoupled Kalman filter algorithm
has been demonstrated to be a satisfactory controller for this
nonlinear ISC problem. Kalman filter itself acts as an optimal
observer. The merit of the ISC Kalman filter design is shown
from the optimal estimations of outputs.
R. Hybrid Control (Continuous Time and Discrete Event)

(9)

The design is based on the pseudolinearization technique,


which uses the describing function analysis to enhance the regulating performance. In (9), A is the input amplitude to the
saturation element and B is the maximum output of a unity
slope saturation element. Under the large step disturbances, it
depends on the controller design, which saturates the actuator. It
operates linearly under the small step disturbances. By Nichols
chart analysis, a closed-loop system is bounded input, bounded
output (BIBO) stable
N (A)Gu GC = 1.

sides, its graphic analysis must be conducted offline that is not


appropriate for real-time ISC systems.

(10)

In the mean while, the limit cycle analysis has been applied,
where Gu is the transfer function of the engine model and GC
is the pseudolinearized open loop transfer function (10). It is
likely that the closed loop system can get away from the limit
cycle. This ISC nonlinear control approach has resulted in better performance than that of the regular linear control approach.
The describing function approach is an alternative to analyze
ISC nonlinear systems, which provides a shortcut between linear systems and nonlinear systems. It works in certain case, but
obviously this approach lacks accuracy due to saturations. Be-

The problem of the fast output torque delivery [91] is investigated by means of a throttle valve position controller with
the drive-by-wire electronics, based on a multipoint port injection manifold. This optimal control problem is solved using
the cyclic hybrid modeling and control approach, which is subject to the constraint that A/F ratio stays as close as possible
to the stoichiometric value of 14.64. Air dynamics is modeled
as a function of the quantity of air flow entering the cylinders
through intake manifold. At the same time, powertrain dynamics is modeled as a linear system, whose state variables are the
axle torsion angle, engine speed, and wheel revolution speed. A
single-state hybrid system reaches the dead point of a control
event whenever pistons reach either top dead center or bottom
dead center. The behavior of each cylinder is represented by
a finite state machine and a discrete event system for torque
generation. Every ith injection system is represented by a hybrid system. Then, the generalized model consists of four hybrid
systems representing behaviors of four cylinders, while the powertrain and intake manifold models are applied to each cylinder.
The dead-points are synchronous and one step delay is assumed
with respect to the previous cycle event. The optimal A/F ratio can be achieved by this hybrid control design. At idle, the
same approach is applied, while powertrain dynamics has been
simplified from a three-variable case to a two-variable case.

YE: NONLINEAR AUTOMOTIVE ISC SYSTEMSAN OVERVIEW

1145

S. Hybrid Control (Continuous Time and Discrete Time)

U. ISC Fundamental Fuzzy Control

The engine interactions are basically classified into discrete


phenomena (e.g., torque generation and spark ignition) and continuous operations (e.g., powertrain evolution and intake air dynamics). It is reasonable to build a hybrid engine model that
captures both types of interactions [8]. In order to achieve the
best fuel economy, the target idle speed must be chosen to be
possibly low as long as the acceptable combustion, emission
levels, and NVH characteristics are maintained. With the ISC
hybrid control closed loop system, the continuous and discrete
variables retain their own distinct natures. To satisfy the performance specifications, a maximal robust control invariant set is
determined and an optimal ISC controller is presented. It can
be used to locate the allowable maximum range of the torque
disturbance and the allowable minimum range of engine speed
under disturbances. The ISC hybrid control approach combines
the merits of the continuous time and discrete time control systems. On the other hand, it increases the complexity of the ISC
controller. As an engine system essentially includes various
types of dynamics coupled together, the practical implementation of this hybrid control approach needs to be tested case by
case.

An automatic rule generation design is formulated, where its


nonlinear engine ISC model has two states (speed, load) [73].
Two crisp finite constant values of throttle angle (5 35 ) and
spark advance (10 45 ) are fuzzified, whose finite points in
the state space are chosen to anticipate the trajectory from one
subspace to another. Data are collected to the rule base for the
feedback regulation. By applying the fixed control, a set of transitional fuzzy relations are obtained. By changing values of the
fuzzy control pairs, a finite number of the transition relations
are chosen. Then, the performance index is defined as the sum
of the squared error, squared control signal, and delay function in each cell group. Besides, the same end-up value always
serves as the source for the next transition, which eventually
leads to an invariant manifold (countless trajectory). The target cell group must be an invariant manifold to guarantee the
asymptotic stability, similar to the concept of the reachability in
modern control theories. The phase portrait assignment results
in suboptimal conditions for fuzzy clusters. By employing this
method on an ISC engine model with 56 fuzzy rules, the desirable stability and robustness under accessory load perturbations
have been shown. With the presence of heavy load disturbances,
i.e., sudden switching of air conditioning on/off, a fuzzy PID
controller [79] has been designed and implemented on a 1.36-L
EFI engine. The speed error signal away from its target idle
speed set point (750 rpm) is fed back into the fuzzy controller,
where proportional, integral, and derivative terms are calculated
via procedures of fuzzification, fuzzy inference, and defuzzification, the fuzzy PID controller then sends out a command to
the stepper motor, which then adjusts its pintle position accordingly, so as to change bypass air flow rate into the engine to
control the idle speed. The ultimate goal is to avoid the engine
stalling and idle speed fluctuation against the sudden load disturbances. On-board testing results indicate that overshoot and
undershoot are reduced considerably during the switching of air
conditioning (on/off).
In a similar case of idle speed control [3] against power
steering load disturbances using air valve duty cycle control,
a 64-rule based fuzzy PI controller leads to more stable engine
performance than PID controllers upon loading and releasing of
power steering. From earlier experimental studies (Fig. 3), the
magnitude of the idle speed fluctuation decreases greatly, while
the ripple frequency of the fuzzy controller increases slightly.
The ISC fuzzy control is the first artificial intelligence approach
successfully applied to the ISC system. It has a strong impact
on concepts of the ISC system used in the conventional control
approaches. It achieves the primary ISC goals of the low speed
set point and disturbance rejection. As fuzzy control is empirical
based, its overall performance compared with all other types of
ISC controllers is still subject to testing.

T. ISC Hybrid Neuro-Fuzzy Control


A hybrid neural-fuzzy system for the vehicle ISC system has
been proposed, where the fuzzy controller is trained by neural
networks based procedure [58]. A nonlinear ISC model with a
nonignorable power delay is investigated and the testing results
are measured by sampling at controller update frequencies. The
neural network training is conducted by imposing various types
of torque disturbances, where an extensive recurrent layer of
eight nodes is used. Outputs are computed by using a sigmoid
activation function at output nodes. This approach can be regarded as one step toward the online training of ISC systems.
The properly designed controller extracts required useful information from available data. It also has sufficient control authority to train the weighting functions of the preprocessing layer,
so as to recover the priori state information [13]. Although the
hybrid neuro-fuzzy control could be a feasible control approach,
in most cases, fuzzy control and neural networks are applied
individually. It is possible that outcomes from two approaches
are similar. For instance, characteristics of three-way-catalyst
(TWC) at idle has been studied using both fuzzy control and
neural networks to predict the catalyst performance instead
of the onboard diagnosing. Emission levels of HC, CO, and
CO2 are calculated. The catalyst inlet-outlet temperature difference is used as an input to the fuzzy controller and neural network controller. If inputs and outputs of the catalyst
system are known, controllers can directly estimate catalyst
performance at steady states. Two types of intelligent control
approaches are based on expert systems or experiment data.
The hybrid neural-fuzzy control approach is promising. The
difficulty lies in the lack of an accurate engine mathematical
model.

V. ISC Fuzzy Sliding Mode Control


Due to mechanical frictions, misfiring, and sensitivity to load
disturbances, the ISC stability is difficult to maintain [15], especially at the low speed to satisfy requirements of emission levels

1146

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART C: APPLICATIONS AND REVIEWS, VOL. 37, NO. 6, NOVEMBER 2007

air control and the spark advance control has been formulated.
Using the ISC fuzzy control on a basis of equality relations, the
engine speed fluctuation has been reduced remarkably to prevent the engine oscillation against the load disturbances from
the air conditioning. This Mamdani fuzzy ISC system represents
a simple case of idle speed fuzzy control. There is no difficulty
essentially in its practical implementation; however its control
authority is limited.

X. ISC Genetic Algorithms (GA) Optimization

Fig. 3.

ISC fuzzy controller and nonlinear surfaces [3], [79].

and fuel economy. Some undesirable effects such as noise, jerk,


and vibration might occur frequently. Considering numerous
engine variables and time-varying processes, the fuzzy sliding mode controller is proposed, which employs the antecedent
knowledge to reduce the tuning time and to be robust against
load disturbances. It is indicated that the nonlinear control structure presented by this fuzzy controller is robust to the conditions
of the internal and external disturbances. However, the engine
constraints have to be omitted. For this reason, the sliding mode
control is combined with fuzzy control for this ISC system.
The sliding line is defined in the phase plane, which aims to
bring control variable back to this line, and remains there using
positive or negative control action: s = + = 0, where is
the engine speed error. Control laws are designed for a 16-valve
Renault engine with power steering applied, the idle speed is
found to react quickly back to the nominal set point. Control laws
for the saturated and nonsaturated areas are designed separately.
The fuzzy control inputs are the speed error and the derivative
of the speed error. The control variable of sliding mode fuzzy
control laws is equal to zero on the diagonal line and positive
(negative) above (below) this line. Fuzzy sliding mode control
has the advantages of both fuzzy control and sliding mode control. It becomes a new research direction. On the other hand,
the formulation of the sliding mode restricts the scope of fuzzy
control applications.

W. ISC Equality Relation-Based Fuzzy Control


Real-world ISC applications are not always coherent with
the understanding of control rules similar to some logical statements. In most cases, fuzzy control is viewed as an interpolation
of a partially specified control function in the vague environment [40]. A fuzzy controller is interpreted as vague descriptions
of crisp control actions on the crisp situation. It is a natural way
to represent the vague environment using a suitable interpolation
method. The ISC system of a Volkswagen engine has been presented, where Mamdani fuzzy model is selected along with the
design of the defuzzification strategies. This approach provides
a semantical base and a clear explanation of fuzzy control and
fuzzy rules. The flexible control mechanism for the idle bypass

The idle seed stability issue relates to the torque variations


among cylinders [38]. This mismatch affects the idle stability
by generating fluctuation ripples of engine speed. Genetic algorithms have been used to reduce engine speed fluctuation at
idle, which is based on neural network training of spark timing.
The resulting optimal spark timing is applied to the ISC system. Particularly, the neural network training of spark timing
will suppress the uneven combustion among cylinders, so as to
maintain uniform and steady engine speed. Genetic algorithms
are employed in a neural network structure with oscillatory neurons at sensor arrays in order to reduce engine speed fluctuations
efficiently. Additional efforts are necessary for this ISC control
approach when multiple inputs are considered, such as the throttle angle, airflow rate, and fueling rate. In another application,
a finite difference model is used to represent engine nonlinear
inverse dynamics between engine speed and throttle duty cycle signals [57]. An inverse model of a time shifting system is
identified directly, using inputoutput data when unstable zero
dynamics is taken into account via a stable compensator. This
method is conducted by minimizing the prediction error using
genetic algorithms-based optimal search method. The ISC stability is tested in the experiments using an electric dynamometer.
Genetic algorithms are the global optimization approaches. A
key problem is still pending how to select a stochastic fitness
function properly that can cover every aspect of engine performance under various constraints.

Y. Neural NetworksISC Back-Propagation Approach


A static and temporal multilayer neural network controller is
developed to track the desired engine speed and engine load,
where the throttle opening and spark advance angle are used
as control signals [61]. The neural networks are trained to learn
inputoutput relationship of ISC models to present relationships
of inverse functions. Back-propagation (BP) learning rules are
employed, where the second-order filter is used to adjust the
gain with an integrator in a series connection to minimize the
back-propagation error. Output tracking is made by the necessary control signals. This control structure is simple to filter out
state error signals. In a case mentioned in Kalman Filter approach [59], Gaussian noise of the observer can be estimated by
back-propagation algorithms. Dynamic gradients are computed
by the recurrent learning algorithm (extended back-propagation)
to train the neurocontroller. Because forward computation of the
dynamic derivatives has some practical difficulties, the truncated

YE: NONLINEAR AUTOMOTIVE ISC SYSTEMSAN OVERVIEW

1147

ward controller provides a good idle speed control performance


which also adapts quickly to the RBF network weights. Being
one of the artificial neural network methodologies, the RBF approach also possesses the online optimization capability. It can
be used as both the feedforward controller and one part of the
feedback controller. The RBF approach is successful in solving
some ISC problems. It also has the potential to create synthesis design with other typical controllers, such as PID control,
adaptive control, and fuzzy control. So far, the RBF approach
is the most promising ISC control approach among those based
on neural networks.
Fig. 4.

ISC online RBF neural network training [58].

back-propagation is used to approximate the dynamic derivatives. A truncated BP approach decouples the cycling time to
update both the derivatives and weights. Combined with the
Kalman filter algorithm, an effective neural network controller is
created. This ISC back-propagation approach is flexible, whose
main role is to filter out the error signals. Thus, it can serve as
either an idle speed controller itself or a potential part of an ISC
system.
Z. Neural NetworksISC Radial Basis Function Approach
ISC systems Fig. 4 are nonlinear systems that are subject to
disturbance rejection [58]. Engines typically are of high nonlinearity at idle. Two control variables presented differ in both
the operating range and temporal characteristics. Spark advance
gives a prompt action but it is of a limit control authority, while
the throttle opening has a dominating control authority but it is
of a slow response. These problems stem from intake manifold
filling dynamics and intake to power stroke delay. The spark
timing variable outside its effective range will have a negative
effect, so nonlinear controllers are needed. The objective of
feedforward control is to minimize the quadratic performance
index. Thus, nonlinear adaptive feedforward control is used to
makeup for external load disturbances via radial basis function
(RBF) networks. Optimization is then achieved using a weight
matrix of RBF multilayered sigmoidal neural networks. RBF
networks with fixed centers are linear in terms of weights and
its online training is a special case of adaptive control with linearly parameterized nonlinearities. The number of the nonlinear
computations in RBF networks is not related to the output dimension. This approach can also be extended to other cases
under various load disturbances, even though the ISC neural
network control is still in its initial stage.
The ISC issues can be formulated as the phenomenological
engine model [28] (see Fig. 4). An ISC system is highly nonlinear owing to delays in the control loop, thus an online parametric
optimization technique uses RBF networks to estimate nonlinear vector fields for an adaptive feedforward controller design.
The feedforward controller and feedback controller have been
combined. The combination compensates for unknown disturbances and improves the transient response quality in comparison with the pure feedback control. This neural network
architecture is trained online adaptively to be robust against
various engine delays. Simulation results show that the feedfor-

III. GDI ENGINE IDLE SPEED CONTROL


A. GDI Engine Idle Speed Feedforward Control
GDI engines operate at dual modes. The stratified mode is
for lean burn operation, which results in good fuel economy.
The A/F ratio can reach as high as 50 : 1 at the stratified mode.
On the other hand, the homogeneous mode is concerned with
emissions and exhaust gas aftertreatment. The operation at the
stratified mode leads to the improved fuel economy and reduced
CO2 emission. Engine delays restrict the potential to improve
the transient response of the ISC system. Therefore, reference
feedforward control has been proposed [64], which is applied to
the ISC system of GDI engines during lean burn. Feedforward
control will improve the transient responses of engine speed
and A/F ratio under load disturbances. The speed response is
enhanced following an increased throttle airflow using the feedforward control. On the other hand, an increased cylinder flow
helps to adjust the in-cylinder A/F ratio to approach its set point
rapidly. With the feedforward control, the brake torque can be
nearly equivalent to the desired value in spite of the saturation
in the speed loop. It is also shown that the speed fluctuation
decrease more than 20% and A/F ratio control is improved as
well. This controller can reduce the constraint effects within the
allowable A/F ratio range, which in turn, adjusts the baseline
controller to regulate speed and A/F ratio. The nonlinear transient response of this ISC system is also improved considerably.
B. GDI Engine Idle Speed Nonlinear Hybrid Control
An accurate nonlinear hybrid model for the stratified mode
operation of GDI engines has been used to design a lowdimensional discrete-time linear system [4]. Due to the presence of predictable load disturbances, a hybrid command governor unit has been designed. It uses two switching LQ optimal
controllers to ensure the closed-loop stability and performance
in the linear regime. The idle speed control for GDI engines is
formulated as hybrid engine models for the constrained optimal
control problem, so that both load variations and constraints are
explicitly taken into account. The major objective is the fuel
consumption minimization. A suboptimal solution is also obtained by the command governor methodology from the hybrid
model. This approach proposes to use the prescribed constraints
at the design phase to avoid optimality degradation. The fuel
consumption at idle has been reduced to 50%, while the total

1148

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART C: APPLICATIONS AND REVIEWS, VOL. 37, NO. 6, NOVEMBER 2007

fuel consumption reduction is about 2% when the allowable constraints from the command governor algorithm are used. It can
also be formulated as a robust constrained optimization problem
for hybrid GDI engines. Its suboptimal controller is designed
by the command governor methodology for the discrete time relaxation of hybrid models, which reflects the multirate nature of
engine input signals. As a result, transient fuel consumption is
reduced compared with the LQ optimal controller. No constraint
is necessary for tuning the optimal LQ switching controller. Its
robustness is guaranteed in this case. The ISC hybrid control
design conforms to the physical mechanisms (both continuous
and discrete) of engines. That is, the reason that it has been
applied successfully to both PFI and GDI engines. However, its
computational complexity has been increased instead.
IV. DIESEL ENGINE IDLE SPEED CONTROL

troller, which will then adjust the controller algorithm accordingly. For the stability concern, recursive least-squares estimation (RLS) is used for parameter identification, whose samples
synchronize to the discrete events of diesel engine injection.
Both the minimum variance self-tuning and the pole placement
self-tuning regulator are developed. Simulations based on this
nonlinear diesel engine model are conducted by tuning the adaptive controller, which converges within 30 s. By comparisons,
the minimum variance algorithm results in a faster settling time
at the expense of the slightly higher overshoot via the adaptive
idle speed control, unavoidable parameter variations resulted
from manufacturing tolerances can be successfully reduced using self-tuning algorithms by adjusting the idle governor to the
engine set points. This example has shown the success of ISC
adaptive control applications on diesel engines instead of on
gasoline engines exclusively.

A. Diesel Engine Idle Speed Discrete Time Control


A discrete time controller is designed using the reference governors for the ISC system of a turbocharged diesel engine [41].
The supplemental torque source control has been presented.
The reference governor generates the torque during the transients caused by a load disturbance (e.g., air conditioning, power
steering) to keep engine speed and A/F ratio within the specified range. By enforcing these limits, engine stalling and visible smoke emissions can be avoided. The torque disturbance
of the engine crankshaft is expressed as a sum of the known
component and the unknown component. The reference governor is a first-order discrete-time nonlinear filter with a scalar
state variable and a scalar adjustable parameter. Engine rotational dynamics is concerned with the torque balance on the
crankshaft. Airflow dynamics is expressed as a second-order
model owing to turbocharger dynamics and intake manifold
filling dynamics. Then, the continuous time engine mean-value
model is discretized, so that a third-order discrete time system
is obtained. So, the general discrete time constrained system
is formulated as a combination of a third-order engine discrete
time system under constraints (A/F ratio and engine speed), the
known and unknown disturbances and a first-order nonlinear filter of the reference governor. The role of the reference governor
is to enhance the ISC robustness against large load disturbances
during transient operations. Engine speed fluctuation and A/F
ratio are restricted within allowable limits by increasing adjustable parameters gradually from zero to one. The nonstationary sequence permits the reference governor to take advantage
of decay rates, which results in a less conservative reference
governor.
B. Diesel Engine Idle Speed Adaptive Control
Two self-tuning adaptive algorithms are proposed for a heavyduty diesel engine, in order to tune the idle governor to the specified parameters [50]. Engine parameters, typically vary across
engines and vary over time due to aging, which causes negative
effects on idle speed performance. Both the Self-tuning Regulator (STR) and MRAC can be used for this complex problem.
The specific engine parameters are tuned by the self-tuning con-

V. NATURAL GAS ENGINE IDLE SPEED CONTROL


A. Natural Gas Engine Idle Speed PI Tuning Control
A multivariable PI tuning [25] is presented, whose major objective is the effective disturbance rejection rather than the set
point tracking. PI control handles interactions to achieve the better disturbance rejection instead of decoupling all interactions
within multivariable plants. To achieve less speed fluctuation
during transient operations at idle, this tuning technique has
been applied to speed control and A/F ratio control of the leanburn natural gas engine. The control design focuses on a regulation problem at the proportional part of the algorithm. Static
decoupling tests of the closed-loop plant are also conducted using the integral functions. The proportional matrix serves as the
gain to the state feedback of this multivariable system, while the
integral gain matrix is used for the fine tuning. The augmented
system with PI controller is stable, if a fine tuner is chosen
to stabilize the system with the proportional feedback. This PI
controller can be used to achieve better idle performance on natural gas engines using existing actuators by coordinating idle
speed control and A/F ratio control. The fast idle speed recovery
from sudden load disturbances has been indicated, based on the
linearized models of natural gas engines.
B. Natural Gas Engine Idle Speed Expert System Tuning
Expert system methodology [67] has been applied to the ISC
system of natural gas engines. The gain tuning method is designed for a multivariable nondecoupling PI controller. In regard
to the disturbance rejection problem, it is designed to improve
performance of one variable by trading off the other. The proposed gain tuning makes it possible to select control options
among engine variables with flexibility. Essentially, in order to
meet the specified performance, an expert system approach is
used numerically to find optimal control gains needed for the
closed loop system. This tuning method is desirable for engine
idle speed control and A/F ratio control. The performance varies
among different options of individual weighting functions. Actual influences from the undesirable variables in the desirable

YE: NONLINEAR AUTOMOTIVE ISC SYSTEMSAN OVERVIEW

ones can be minimized by the proper weight selections. PI controller is shown to be successful in the idle speed tuning of
natural gas engines.
VI. CONCLUSION
A comprehensive overview on nonlinear idle speed control
systems has been presented, involving various types of automotive engines. The selection of the idle speed set point takes
an important role in each aspect of ISC performance, i.e., idle
speed stability, fuel economy, emissions, nonlinear delays, vehicle NVH, robustness, and so on. A variety of idle speed control
applications are investigated, which have been applied successfully to PFI engines, GDI engines, diesel engines, and natural
gas engines. Over 30 types of the ISC controllers are discussed
in this article, covering virtually all practical aspects of classical control, modern control, and intelligent control. These ISC
approaches also have the potential on the hybrid vehicle application. Automotive idle speed control provides a typical example
to implement the cutting edge and diversified control methodologies on the complex engineering problems. There is no doubt
that these control technologies can also contribute to many other
challenging areas of automotive, electrical, mechanical, robotic,
aeronautical, and biomedical nonlinear systems.
ACKNOWLEDGMENT
The author would like to thank three anonymous referees and the Associate Editor for their constructive comments
and valuable suggestions. The author is also grateful to Prof.
S. Shi and Prof. D. Zheng, the two Chief Scientists of China on
Engine and Control, respectively, who have directed this author
into these two challenging fields of automotive engineering and
control science.
REFERENCES
[1] M. Abate, B. Barmish, C. Sanchez, and R. Tempo, Application of some
new tools to robust stability analysis of spark ignition engines: A case
study, IEEE Trans. Control Syst. Technol., vol. 2, no. 1, pp. 2230, Mar.
1994.
[2] M. Abate and V. Dinunzio, Idle speed control using optimal regulation,
Soc. Automotive Eng., Warrendale, PA, Tech. Rep. 905008, 1990.
[3] M. Abate and N. Dosio, Use of fuzzy logic for engine idle speed control,
Soc. Automotive Eng., Warrendale, PA, Tech. Rep. 900594, 1990.
[4] L. Albertoni, A. Balluchi, A. Casavola, C. Gambelli, E. Mosca, and
A. L. Sangiovanni, Hybrid command governors for idle speed control
in gasoline direct injection engines, in Proc. 2003 Amer. Control Conf.,
Denver, CO, vol. 1, pp. 773778.
[5] L. Albertoni, A. Balluchi, A. Casavola, C. Gambelli, E. Mosca, and
A. L. Sangiovanni, Idle speed control for GDI engines using robust
multirate hybrid command governors, in Proc. 2003 IEEE Int. Conf.
Control Appl., Istanbul, Turkey, vol. 1, pp. 237242.
[6] H. Ando and M. Motomochi, Contribution of fuel transport lag and
statistical perturbation in combustion to oscillation of SI engine speed
at idle, Soc. Automotive Eng., Warrendale, PA, Tech. Rep. 870545,
1987.
[7] B. Badreddine, A. Zaremba, F. Lin, and J. Sun, Active damping of engine
idle speed oscillation by applying adaptive PID control, Soc. Automotive
Eng., Warrendale, PA, Tech. Rep. 2001-01-0261, 2001.
[8] A. Balluchi, L. Benvenuti, T. Villa, H. Wong-Toi, and A. L. Sangiovanni,
Hybrid controller synthesis for idle speed management of an automotive
engine, in Proc. Amer. Control Conf., Chicago, IL, vol. 2, Jun. 2000,
pp. 11811185.

1149

[9] A. Balluchi, L. Benvenuti, T. Villa, H. Wong-Toi, and A. L. Sangiovanni,


Engine idle speed control via maximal safe set computation in the crankangle domain, in Proc. 2002 IEEE Int. Symp. Ind. Electron., Sevilla,
Spain, vol. 2, pp. 618622.
[10] S. Bengea, X. Li, and R. Decarlo, Combined controller-observer design
for uncertain time delay systems with application to engine idle speed
control, J. Dynam., Syst., Meas. Control, vol. 126, pp. 772780, Dec.
2004.
[11] P. Bidan and L. Kouadio, Electrical assistance for S.I. engine idle-speed
control, Control Eng. Pract., vol. 6, no. 7, pp. 829836, 1998.
[12] C. Bohn, A. Cortabarria, V. Hartel, and K. Kowalczyk, Active control
of engine-induced vibrations in automotive vehicles using disturbance
observer gain scheduling, Control Eng. Pract., vol. 12, no. 8, pp. 1029
1039, 2004.
[13] P. Botsaris, D. Bechrakis, and P. Sparis, An estimation of 3-way catalyst
performance using artificial neural networks during idle speed, in Proc.
2004 Fall Tech. Conf. ASME Intern. Combust. Engine Div., Long Beach,
CA, pp. 123129.
[14] S. Boverie, B. Demaya, J. Lequellec, and A. Titli, Contribution of fuzzy
logic control to the improvement of modern car performance, Control
Eng. Pract., vol. 1, no. 2, pp. 291297, Apr. 1993.
[15] S. Boverie, C. Patrice, and M. Jean, Fuzzy sliding mode control application to idle speed control speed, in Proc. IEEE Int. Conf. Fuzzy Syst.,
Orlando, FL, vol. 2, Jun. 1994, pp. 974977.
[16] P. Bromnick, Development of a model predictive controller for engine
idle speed using Cpower, Soc. Automotive Eng., Warrendale, PA, Tech.
Rep. 199901-1171, 1999.
[17] K. Butts, N. Sivashankar, and J. Sun, Feedforward and feedback
design for engine idle speed control using l1 optimization, in
Proc. Amer. Control Conf., Seattle, WA, vol. 4, Jun.1995, pp. 2587
2590.
[18] K. Butts, N. Sivashankar, and J. Sun, Application of L1 optimal control
to the engine idle speed control problem, IEEE Trans. Control Syst.
Technol., vol. 7, no. 2, pp. 258270, Mar. 1999.
[19] P. Danijel and D. Hrovat, SI engine load torque estimator based
on adaptive kalman filter and its application to idle speed control,
Soc. Automotive Eng., Warrendale, PA, Tech. Rep. 200501-0036,
2005.
[20] J. Deur, M. Jansz, D. Pavkovic, and M. Jansz, Identification and speed
control of SI engine for idle operating mode, Soc. Automotive Eng.,
Warrendale, PA, Tech. Rep. 2004-01-0898, 2004.
[21] L. Feldkemp and G. Puskorius, Trainable fuzzy and neural-fuzzy systems
for idle-speed control, in Proc. 2nd IEEE Int. Conf. Fuzzy Syst., San
Francisco, CA, vol. 1, Mar. 1999, pp. 4551.
[22] R. Ford, Robust automotive idle speed control in a novel framework,
Ph.D thesis, Cambridge Univ., Cambridge, U.K., 2000.
[23] R. Ford and N. Collings, Dynamic effect of residual gas temperature on
combustion torque at idle, presented at the SAE Fall Fuels Lubes Conf.,
2001.
[24] R. Ford and K. Glover, An application of coprime factor based antiwindup and bumpless transfer control to the spark ignition engine idle
speed control problem, in Proc. 39th IEEE Conf. Decision Control,
Sydney, Australia, vol. 2, Dec. 2000, pp. 10691074.
[25] A. Gangopadhyay and P. Meckl, Multivariable PI tuning and application
to engine idle speed control, in Proc. Amer. Control Conf., San Diego,
CA, Jun. 1999, pp. 26782682.
[26] L. Glielmo, S. Santini, and I. Cascella, Idle speed control through output feedback stabilization for finite time delay systems, in Proc. Amer.
Control Conf., Chicago, IL, vol. 1, Jun. 2000, pp. 4549.
[27] C. GokCek and P. Kabamba, Damping of idle engine speed oscillations
using a reversible alternator, in Proc. Amer. Control Conf., Philadelphia,
PA, vol. 3, Jun. 1998, pp. 14001404.
[28] D. Gorinevsky and L. Feldkamp, On-line optimization of RBF network
feedforward compensation load disturbance in idle speed control of automotive engine, in Proc. 1996 IEEE Int. Conf. Control Appl., Dearborn,
MI, pp. 764769.
[29] M. Han, R. Loh, and L. Wang, Optimal idle speed control of an automotive engine, Soc. Automotive Eng., Warrendale, PA, Tech. Rep. 981059,
1998.
[30] P. Herman and M. Franchek, Engine idle speed control using actuator
saturation, IEEE Trans. Control Syst. Technol., vol. 8, no. 1, pp. 192199,
Jan. 2000.
[31] M. Howell and M. Best, On-line PID tuning for engine idle-speed control
using continuous action reinforcement learning automata, Control Eng.
Pract., vol. 8, no. 2, pp. 147154, Feb. 2000.

1150

IEEE TRANSACTIONS ON SYSTEMS, MAN, AND CYBERNETICSPART C: APPLICATIONS AND REVIEWS, VOL. 37, NO. 6, NOVEMBER 2007

[32] D. Hrovat and B. Bodenheimer, Robust automotive idle speed control


design based on -synthesis, in Proc. Amer. Control Conf., San Francisco,
CA, vol. 4, 1993, pp. 17781783.
[33] D. Hrovat and J. Sun, Models and control methodologies for IC engine
idle speed control design, Control Eng. Pract., vol. 5, no. 8, pp. 1093
1100, 1997.
[34] D. Hrovat and W. Powers, Computer control systems for automotive
powertrains, IEEE Control Syst. Mag., vol. 8, no. 4, pp. 310, Aug.
1988.
[35] H. Inoue and S. Washino, A performance improvement in idle speed control system with feedforward compensation for alternator load current,
Soc. Automotive Eng., Warrendale, PA, Tech. Rep. 900777, 1990.
[36] R. Jurgen, Automobile Electronics Handbook. New York: McGraw-Hill,
1995.
[37] M. Kajitani and K. Nonami, High performance idle speed control applying the sliding mode control with H robust hyperplane, Soc. Automotive
Eng., Warrendale, PA, Tech. Rep. 2001-01-0263, 2001.
[38] D. Kim and J. Park, Neural network control for reducing engine speed
fluctuation at idle, in Proc. 1999 IEEE Int. Conf. Syst., Man, Cybern.,
Tokyo, Japan, vol. 4, pp. 629634.
[39] L. Kjergaard, S. Nielsen, T. Vesterholm, and E. Hendricks, Advanced
nonlinear engine idle speed control systems, Soc. Automotive Eng.,
Warrendale, PA, Tech. Rep. 940974, 1994.
[40] F. Klawonn, J. Gebhardt, and R. Kruse, Fuzzy control on the basis of
equality relations with an example from idle speed control, IEEE Trans.
Fuzzy Syst., vol. 3, no. 3, pp. 336350, Aug. 1995.
[41] I. Kolmanovsky, E. Gilbert, and J. Cook, Reference governors for supplemental torque source control in turbocharged diesel engines, in Proc.
Amer. Control Conf., Albuquerque, NM, vol. 1, Jun.1997, pp. 652656.
[42] O. Kovalenko, D. Liu, and H. Javaherin, Neural network modeling and
adaptive critic control of automotive fuel injection system, in Proc. 2004
IEEE Int. Symp. Intell. Control, Taipei, Taiwan, pp. 368373.
[43] X. Li and S. Yurkovich, Discrete adaptive sliding mode control for idle
speed regulation in IC engines, in Proc. 2000 IEEE Int. Conf. Control
Appl., Anchorage, AK, pp. 237242.
[44] X. Li and S. Yurkovich, Neural network based, discrete adaptive sliding
mode control for idle speed regulation in IC engines, J. Dynam. Syst.,
Meas. Control, vol. 122, pp. 269275, Jun. 2000.
[45] X. Li and S. Yurkovich, Sliding mode control of delayed systems with
application to engine idle speed control, IEEE Trans. Control Syst.
Technol., vol. 9, no. 6, pp. 802810, Nov. 2001.
[46] M. Livshitz and D. Sanvido, Absolute stability of idle speed control
system, Soc. Automotive Eng., Warrendale, PA, Tech. Rep. 860412,
1986.
[47] M. Lu and R. Loh, Modeling, design and implementation of discrete
sliding mode control for an engine idle speed control system, in Proc.
Amer. Control Conf., Anchorage, AK, vol. 3, May 2002, pp. 1984
1989.
[48] C. Manzie and H. Watson, A novel approach to disturbance rejection in
idle speed control towards reduced idle fuel consumption, in Proc. Inst.
Mech. Eng., Part D: J. Automob. Eng., 2003, vol. 217, no. 8, pp. 677
690.
[49] A. Martinez and M. Jamshidi, Design of fuzzy logic based engine idlespeed controllers, in Proc. 36th Midwest Symp. Circuits Syst., Detroit,
MI, Aug. 1993, vol. 2, pp. 15441547.
[50] D. Memering and P. Meckl, Comparison of adaptive control technique
applied to diesel engine idle speed regulation, J. Dynam. Syst., Meas.
Control, vol. 124, pp. 682688, Dec. 2000.
[51] M. Mihelc and S. Citron, An adaptive idle mode control system, Soc.
Automotive Eng., Warrendale, PA, Tech. Rep. 840443, 1984.
[52] M. Livshiz, D. Sanvido, and S. Stiles, Nonlinear engine model for idle
speed control, in Proc. 33rd Conf. Decision Control, Lake Buena Vista,
FL, Dec. 1994, vol. 3, pp. 24492451.
[53] H. Mohamed, S. Munzir, M. Abdulmuin, and S. Hameida, Fuzzy modeling and control of a spark ignition engine idle mode, in Proc. IEEE
Intell. Syst. Technol. New Millennium, 2000, vol. 2, no. 2, pp. 586591.
[54] D. Nicolao, G. Rossi, R. Scattolini, and M. Suffritti, Identification and
idle speed control of internal combustion engines, Control Eng. Pract.,
vol. 7, no. 9, pp. 10611069, Sep. 1999.
[55] Y. Nishimura and K. Ishii, Engine idle stability analysis and control, Soc. Automotive Eng., Warrendale, PA, Tech. Rep. 860415,
1986.
[56] A. Petridis and A. Shenton, Linear robust control of identified nonlinear
inverse compensated SI engine, J. Dynam. Syst., Meas. Control, vol. 125,
no. 1, pp. 6973, Mar. 2003.

[57] A. Petridis and A. Shenton, Inverse-NARMA: A robust control method


applied to SI engine idle-speed regulation, Control Eng. Pract., vol. 11,
no. 3, pp. 279290, 2003.
[58] G. Puskorius and L. Feldkamp, Automotive engine idle speed control
with recurrent neural networks, in Proc. 1993 Amer. Control Conf., San
Francisco, CA, vol. 1, pp. 311316.
[59] G. Puskorius and L. Feldkemp, Truncated backpropagation through time
and Kalman filter training for neurocontrol, in Proc. 1994 IEEE Int. Conf.
Neural Netw., Orlando, FL, vol. 4, pp. 24882493.
[60] A. Raffari, L. Y. Wang, and O. Daniel, Optimal H-infinity decoupling
of engine idle speed and emission subsystems, in Proc. Annu. Southeastern Symp. Syst. Theory, Baton Rouge, LA, Mar./Apr. 1996, pp. 7
11.
[61] F. Salam and A. Gharbi, Temporal neuro-control of idle engine speed,
in Proc. 1996 IEEE Int. Symp. Intell. Control, Dearborn, MI, pp. 396
401.
[62] R. Scattolini and C. Siviero, Modeling and identification of an electromechanical internal combustion engine throttle body, Control Eng.
Pract., vol. 5, no. 9, pp. 12531259, Sep. 1997.
[63] J. Scillieri, J. Buckland, and J. Fredenberg, Use of feedforward in idle
speed control for a direct injection spark ignition engine during lean
burn, in Proc. Amer. Control Conf., Anchorage, AK, May 2002, vol. 2,
pp. 14191424.
[64] J. Scillieri and J. Buckland, Reference feedforward in the idle speed
control of a direct-injection spark-ignition engine, IEEE Trans. Veh.
Technol., vol. 54, no. 1, pp. 5161, Jan. 2005.
[65] D. Shim, J. Park, P. Khargonekar, and W. Ribbens, Engine idle speed
control, in Proc. Amer. Control Conf., Seattle, WA, Jun. 1995, vol. 4,
pp. 25822586.
[66] D. Shim, J. Park, P. Khargonekar, and W. Ribbens, Reducing automotive
engine speed fluctuation at idle, IEEE Trans. Control Syst. Technol.,
vol. 4, no. 4, pp. 404410, Jul. 1996.
[67] A. Sivasubramanian and P. Meckl, Numerical solution for multivariable idle speed control of a lean burn natural gas engine, in Proc.
Amer. Control Conf., Boston, MA, Jun./Jul. 2004, vol. 1, pp. 138
143.
[68] A. Stotsky, B. Egardt, and S. Eriksson, Variable structure control of engine idle speed with estimation of unmeasurable disturbances, J. Dynam.
Syst., Meas. Control, vol. 122, pp. 599603, Dec. 2000.
[69] P. Sun, P. Barry, and H. Davor, Optimal idle speed control of an automotive engine, in Proc. Amer. Control Conf., Chicago, IL, Jun. 2000, vol. 2,
pp. 10181126.
[70] T. Takahashi and T. Ueno, A simple engine model for idle speed control,
Soc. Automotive Eng., Warrendale, PA, Tech. Rep. 850291, 1985.
[71] M. Thornhill and S. Thompson, A comparison of idle speed control schemes, Control Eng. Pract., vol. 8, no. 5, pp. 519530, May
2000.
[72] G. Vachtsevanos, Idle speed control of an automotive engine using systematic fuzzy logic methodology, in Proc. 12th Triennial World Congr.
Int. Fed. Autom. Control, Sydney, Australia, Jul. 1994, vol. 3, pp. 29
34.
[73] G. Vachtsevanos, S. Farinwata, and H. Kang, A systematic design method
for fuzzy logic control with application to automotive idle speed control,
in Proc. 31st Conf. Decision Control, Tucson, AZ, Dec. 1992, vol. 3,
pp. 25472548.
[74] Y. Wang, A. Stefanopoulou, and M. Levin, Idle speed control: An old
problem in a new engine design, in Proc. Amer. Control Conf., San Diego,
CA, Jun. 1999, vol. 2, pp. 12171221.
[75] S. Watanabe and M. Imamura, Development of model-following idle
speed control system incorporating engine torque models, Soc. Automotive Eng., Warrendale, PA, Tech. Rep. 920160, 1992.
[76] M. Wendeker and J. Czarnigowski, Adaptive control of the idle speed,
in Proc. ASME Intern. Combust. Engine Div. 2003 Spring Tech. Conf.,
Salzburg, Austria, pp. 559563.
[77] S. Williams, D. Hrovat, C. Davey, D Maclay, J. Crevel, and L. Chen, Idle
speed control design using an H-infinity approach, Soc. Automotive Eng.,
Warrendale, PA, Tech. Rep. 930770, 1993.
[78] H. Yamaguchi and S. Takizawa, Analysis on idle speed stability in port
fuel injection engine, Soc. Automotive Eng., Warrendale, PA, Tech. Rep.
861389, 1986.
[79] Z. Ye, Research on idle speed control of gasoline engine and its fuzzy
control implementation, M.S. thesis, Tsinghua Univ., Beijing, China, Jun.
1996.
[80] Z. Ye, Temperature impact on modeling and control of lean NOx trap,
J. Fuels Lubricants, vol. 112114, pp. 790795, Sep. 2004.

YE: NONLINEAR AUTOMOTIVE ISC SYSTEMSAN OVERVIEW

[81] Z. Ye, Automotive hybrid system optimization using dynamic programming, Soc. Automotive Eng., Warrendale, PA, Tech. Rep. 2003-01-0847,
2003.
[82] Z. Ye, GDI engine exhaust aftertreatment system analysis and oxygen
sensor based identification, modeling and control of lean NOx trap,
in Proc. ASME Intern. Combust. Engine Div. 2003 Spring Tech. Conf.,
Salzburg, Austria, 2007, pp. 713719.
[83] Z. Ye and M. Lai, Genetic algorithm optimization of fuel economy for
PFI engine with VVT-VCR, in Proc. 2004 IEEE Int. Conf. Control Appl.,
Taiwan, China, vol. 1, pp. 364369.
[84] Z. Ye and L. Li, Control options for exhaust gas aftertreatment and fuel
economy of GDI Engine Systems, in Proc. 2003 IEEE Conf. Decision
Control, Maui, HI, vol. 2, pp. 17831788.
[85] S. Yurkovich and M. Simpson, Crank angle domain modeling and control for idle speed, Soc. Automotive Eng., Warrendale, PA, Tech. Rep.
970027, 1997.
[86] S. Yurkovich and M. Simpson, Comparative analysis for idle speed control: A crank-angle domain viewpoint, in Proc. Amer. Control Conf.,
Albuquerque, NM, Jun. 1997, vol. 1, pp. 278283.
[87] B. K. Powell, J. A. Cook, and J. W. Grizzle, Modeling and analysis of
an inherently multi-rate sampling fuel injected engine idle speed control
loop, Trans. ASME, J. Dyn. Syst., Meas. Control, vol. 109, pp. 405409,
Dec. 1987.
[88] B. K. Powell, J. A. Cook, and J. W. Grizzle, Modeling and analysis of
an inherently multi-rate sampling fuel injected engine idle speed control
loop, in Proc 1987 Am. Control Conf., 1987, vol. 3, pp. 15431548.
[89] B. K. Powell and W. F. Powers, Linear quadratic control design for
nonlinear IC engine systems, in Proc. 10th Anniversary Int. Symp. Autom.
Technol. Autom., 1981, vol. 1, pp. 118.
[90] C. Carnevale and A. Moschetti, Idle speed control with H-infinity technique, 1993, SAE Technical Paper 930770.
[91] A. Balluchi, L. Benvenuti, M.D. D. Benedetto, S. Cardellino, C. Rossi,
and A. L. Sangiovanni-Vincentelli, Hybrid control of the airfuel ratio
in force transients for multi-point injection engines, in Proc. 38th IEEE
Conf. Decis. Control, 1999, vol. 1, pp. 316321.

1151

Zhengmao Ye (M03SM07) received the B.E. degree in thermal engineering from Tianjin University,
Tianjin, China, in 1992, the M.S. degree in automotive engineering from Tsinghua University, Beijing,
China, in 1996, and the second M.S. and Ph.D. degrees in electrical engineering from Wayne State University, Detroit, MI, in 1999 and 2001, respectively,
next to his Chinese Ph.D. being pursued under the
Academician Advisor Dr. Shaoxi Shi (Past President
of Tianjin University, Tianjin, China and the Chinese
Chief Scientist on Internal Combustion Engines) in
the School of Mechanical Engineering, Tianjin University.
He is currently an Assistant Professor in the Electrical Engineering Department, Southern University, Baton Rouge, LA, where he is the Founder and
Director of the System and Control Laboratory. In 1992 and from 1996 to 1997,
he was with the National Key Laboratory of Engines, Tianjin University, and
from 1993 to 1996 with the National Key Laboratory of Automotive Safety
and Energy, Tsinghua University where he was engaged in research activities.
He served as Departmental Chief Student Leader at both Tianjin and Tsinghua
Universities. His current research interests include modeling, control, and optimization with diverse applications on automotive, electrical, mechanical, and
biomedical systems, as well as signal processing and image processing. Among
very few cross-disciplinary researchers worldwide, he has the first author publications in broad fields covering all the leading control proceedings in three
most prestigious engineering societies, specifically, IEEE, ASME, and SAE,
such as IEEE (SMC, CDC, CCA, ACC, ISIC, FUZZ, IJCNN, CASE, ICCA,
ICCC, SOSE, and CCECE), ASME (Congress IMECE, ICES, JRCICE), SAE
(Congress USA, EAEC, JSAE). He was also the academic reviewer for 45
published articles submitted to the IEEE, the American Society of Mechanical Engineers, the Society of Automotive Engineers, and some international
journals.
Dr. Ye is the recipient of the Chinese National Fellowship (First Prize) at
Tianjin University, the USA Allied Signal Fellowship (First Prize) at Tsinghua
University, and Most Outstanding Faculty of Electrical Engineering at Southern
University.