Академический Документы
Профессиональный Документы
Культура Документы
1, JANUARY 2011
Abstract—This paper presents the analysis, design and vali- of all, it can be employed to achieve a regularization of the dy-
dation of a gain-scheduled controller for an electronic throttle namic relationship between the gas command and the driving
body (ETB) designed for ride-by-wire applications in racing torque transmitted to the ground during acceleration maneuvers,
motorcycles. Specifically, the open-loop dynamics of the system
are studied in detail discussing the effects of friction based on thereby offering a smoother vehicle dynamic behavior which
appropriate experiments. Further, a linear time invariant nominal can significantly enhance the vehicle handling and driveability.
model of the system to be controlled is experimentally identified Further, the ETB is also employed as an engine protection mech-
via a frequency-domain black box approach, together with the un- anism. It ensures that the engine operates within a controlled
certainty bounds on the model parameters. Based on these results range, for example limiting the engine speed and regulating the
a model-based gain-scheduled proportional-integral-differential
(PID) controller for throttle position tracking is proposed. The idle speed.
closed-loop stability of the resulting linear parametrically varying From a more advanced vehicle dynamics control perspective,
(LPV) system is proved by checking the feasibility of an appro- moreover, the ETB offers a way to differently shape the air flow
priate linear matrix inequality (LMI) problem, and the state space rate behavior in the face of a given acceleration command, thus
representation of the closed-loop LPV system is experimentally providing a means to customize the vehicle dynamic response to
validated. Finally, the performance of the controlled system is
compared to the intrinsic limit of the actuator and tested under the drivers’ gas request. This feature also allows vehicle manu-
realistic use, namely both on a test-bench employing as set-point facturers to personalize the vehicle driving feeling by conferring
the throttle position recorded during test-track experiments and it either a performance-oriented or a comfort-oriented dynamic
on an instrumented motorcycle. behavior, which would be in principle dictated by its mechan-
Index Terms—Electronic throttle body (ETB), gain-scheduled ical layout, simply via a different tuning of the ETB electronic
control, linear parameter varying (LPV) model validation, motor- control system.
cycle dynamics. Finally, of course, an effective ETB control system is a
mandatory building block for the design of traction control
system both for four- and two-wheeled vehicles, e.g., [1]–[3].
I. INTRODUCTION AND MOTIVATION Note that, mechanically, a throttle is a simple system; it is
mainly comprised of one or more butterfly valves actuated by
an electrical motor through a reduction system. The throttle
HE electronic throttle body (ETB) is a mechatronic actu-
T ator devoted to the regulation of the air inflow at the en-
gine intake manifold. According to the drive-by-wire paradigm,
dynamic behavior is rendered complex by packaging, cost, and
reliability constraints. These constraints often translate into
dominant friction and backlash behavior in the transmission,
an accurate control of the ETB dynamics enables a correct and making the control of the valve difficult. In the scientific
optimized management of the air mass flow rate, which can be literature, several control strategies have been proposed for
managed independently of the rider’s request. The availability throttle actuation in cars with the common aim of achieving
of a properly controlled ETB provides several advantages. First good tracking performance in all working conditions and in
the face of parametric uncertainties and avoiding overshoots,
Manuscript received May 08, 2009; revised November 26, 2009; accepted which are the main source of discomfort for the driver (see,
July 14, 2010. Manuscript received in final form August 08, 2010. Date of pub- e.g., [1], [4]–[9]).
lication September 07, 2010; date of current version December 22, 2010. Rec-
ommended by Associate Editor C. Novara. This work was supported in part by Electronic throttle actuation in motorcycles is far less
MIUR Project “New methods for Identification and Adaptive Control for Indus- common than in cars; consequently, little has been published
trial Systems” and by Piaggio & C. S.p.A., Aprilia Brand.
M. Corno is with the Delft Center for Systems and Control (DCSC),
on this topic in the open scientific literature so far. In particular,
Delft University of Technology, 2628 CD Delft, The Netherlands (e-mail: in [10] a solution for the ETB control of two-wheeled vehicles
m.corno@tudelft.nl). is proposed employing a variable structure control strategy. It is
M. Tanelli and S.M. Savaresi are with the Dipartimento di Elettronica e In-
formazione, Politecnico di Milano, 20133 Milano, Italy (e-mail: tanelli@elet.
worth noting that the aforementioned manufacturing constraints
polimi.it; savaresi@elet.polimi.it). become even more strict when the ETB is being designed for
L. Fabbri is with Piaggio & C. S.p.A., Aprilia Brand, 30033 Noale, Venice, two-wheeled vehicles, especially for racing motorcycles. Mass
Italy. and volumes optimization becomes critical since racing motor-
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. cycles are very sensitive even to small changes in the center
Digital Object Identifier 10.1109/TCST.2010.2066565 of mass, see, e.g., [11]–[13]. Furthermore, racing applications
1063-6536/$26.00 © 2010 IEEE
CORNO et al.: DESIGN AND VALIDATION OF A GAIN-SCHEDULED CONTROLLER 19
Fig. 3. Behavior of the throttle position in open loop in quasi-static tests: (a) opening and (b) closing.
set point generation and filtering. It is anticipated that the target the position one which controls the throttle movement. Unfortu-
ECU will have a sampling rate of 1 kHz and a 20 kHz PWM nately, this solution could not be adopted because of the limita-
carrier. The design of the final throttle position controller will tions in the clock speed of the target ECU and due to the fact that
be carried out considering this final hardware specification, but cost constraints prevented the use of additional current sensors.
for analysis purposes the full potential of real-time controller However, in the experimental setup, the current loop option be-
(i.e., up to 20 kHz of sampling frequency) can be employed. comes feasible if implemented on the FPGA. As the inner loop
Fig. 2 shows a block diagram representation of the throttle control better decouples the mechanical behavior and the elec-
control system. As can be seen, the electrical dynamics trical behavior, it will be employed to estimate the mechanical
have been decoupled from the mechanical ones, which are dynamics of the return spring, as it makes it easier to isolate and
described by the planetary gear, the return spring, a friction understand the analyzed phenomena. As already mentioned, the
term and the LTI throttle dynamics . The interconnection current dynamics will be left in open loop in the final controller.
between electrical and mechanical ETB components is due to It will be seen later that this contributes to increase the uncer-
the electromotive force (E.M.F.). Finally, the system is com- tainties affecting the system dynamical model.
pleted by the position control loop, (s), which regulates To analyze the nonlinear behavior of the throttle position in
the throttle position to a desired set-point . open loop refer to Fig. 3(a) and (b), where the throttle position
is plotted as a function of the input current during opening and
III. OPEN-LOOP SYSTEM ANALYSIS closing quasi-static tests, in which the current was increased
along a very slow ramp. A clear asymmetry is visible between
This section is devoted to analyze the open-loop system be- the opening and closing, the former possessing a fully on/off
havior, characterized by the electrical dynamics of the dc motor behavior, whereas the latter shows a sort of staircase descent to
and the mechanical spring characteristic of the throttle body. the fully closed position. Note, moreover, that the position value
The electrical dynamics of the dc motor can be described by the obtained for zero current varies significantly (from 0.2 to 0.38)
following equations: in the different tests and it does not correspond to a fully closed
throttle. This clearly confirms the criticalities of the system due
(1) to mechanical nonlinearities and friction effects.
Further, the spring characteristic has been identified. To this
where is the voltage applied to the motor, is the winding aim, a very low bandwidth proportional-integral (PI) position
current, and are the dc motor resistance and inductance controller has been designed, so as to stabilize the closed-loop
and is the electromotive force, which is proportional to the system dynamics and make the controlled ETB able to follow
motor rotational speed . The motor generates a torque which a very slow reference signal constituted by an ascending ramp
is proportional to the current , whereas the control variable is from 0 to 1 followed by a descending, symmetrical, one. Fig. 4
the applied voltage . System (1) shows that the relation be- shows the position-to-current map of the throttle measured in
tween motor voltage and motor torque depends on , and ; three different tests. By inspecting Fig. 4 it is apparent that the
this dependency introduces two critical phenomena. First, the system exhibits a nonlinear hysteretic behavior. Overall, the fol-
strong dependency that the resistance has on the temperature lowing three different phases can be outlined in the ascending
translates into uncertainties on the torque. Second, the electro- ramp:
motive force determines a coupling between the electrical and • from 0.05 to 0.2 the spring stiffness is constant and approx-
mechanical dynamics. In mechatronics, these issues are typi- imately ;
cally solved by designing an inner current control loop to reject • from 0.2 to 0.9 the spring stiffness decreases to a value of
these disturbances [14]. This solution yields better results when about ;
the inner control loop is run at a higher sampling frequency than • from 0.9 to 1 the spring stiffness increases to .
CORNO et al.: DESIGN AND VALIDATION OF A GAIN-SCHEDULED CONTROLLER 21
Fig. 6. Open-loop valve opening (solid line) and closing (dashed line) when
the maximum and the minimum voltage is applied.
Fig. 7. Open-loop valve position (solid line) and normalized current (dashed line) for a maximum voltage opening (left) and closing (right) test.
(4)
(5)
Fig. 11. Adaptation laws of the integral time T and derivative constant T of the PID controller.
Fig. 13. Time domain validation results of the LPV model at a frequency of 1.4 Hz (upper plot) and of 6.25 Hz (lower plot).
based on the scheduling of the integral time and the deriva- chosen with dynamics compatible with thermal and lubrication
tive time of the controller as static functions of the expected effects. The variability ranges of the parameters vector can be
variation of the reference position signal com- described, in the parameter space, by a 3-D polytope.
puted via (6). By plugging the corresponding value of the parameters in
Once the scheduling law is implemented, the whole con- and in and closing the loop, a transfer func-
trolled system can be seen as an LPV system. Furthermore, the tion of the closed-loop system for each point in the parameter
LPV framework also accounts for structured uncertainties in space can be obtained. However, to resort to LPV techniques
the ETB dynamic model. As a matter of fact, the analysis of the for the closed-loop stability analysis, one needs to obtain a state
identified model obtained in different working conditions—see space representation of the closed-loop system. This step gives
Fig. 8—has highlighted that the system model is subject to a rise to two different issues. Specifically, as we start from local
certain amount of uncertainty which can be accounted for by models of the closed-loop system obtained by evaluating the
allowing a variability in the position of the first zero and in the parameter vector at fixed points of the polytope, one needs to
transfer constant of the nominal transfer function . interpolate the local models and this would in principle ask
Fig. 8 also shows the uncertainty boundaries when the position that all the state space realizations are in the same coordinate
of the lower frequency zero of is moved within the basis, [33]. Second, in LPV systems—which are a special class
interval Hz and the transfer constant varies in the of time-varying systems—the usual notions of equivalence be-
interval . As can be appreciated from Fig. 8, tween input/output (I/O) and state space representations which
the structured uncertainty describes the variability of the system hold for LTI systems are not valid anymore, unless a dynamic
in the frequency range of interest. According to the adopted variation of the parameters is permitted (see [34], [35]).
black-box approach, the choice of the uncertainty-modeling In general, the interpolation problem can be dealt with by
parameters has been driven by complexity considerations. resorting to balanced realizations [33], [36]. In the considered
Specifically, we have looked for the smallest set of parameters case, however, as both the system and the controller structure
that could account for the whole variability shown by the were known, an analytical state-space model both for the uncer-
experimental data in the frequency range of interest. Three tain ETB dynamics and for the gain-scheduled PID controller
time-varying parameters are therefore identified, so that the has been obtained by performing a symbolic realization of both
resulting parameter vector can be defined as . and . Based on such model, it is possible to
Further, note that all the parameters vary with respect to time, write the LPV closed-loop system as
with bounds on the velocity of their time variation. The time
variability accounts for dynamic variations both of the system
uncertainties, which are expected to vary as functions of the (7)
specific ETB, of the engine temperature, of the lubrication
conditions, and of the set-point, i.e., the gas request command. where is the position set-point and is
Specifically, the time derivatives bounds on have been set the measured throttle position. As for the equivalence notion be-
to . The bound on the time tween I/O and state space realizations of LPV systems, it should
derivative of the set-point variation has been determined by be pointed out that the standard LTI realization is only an ap-
analyzing several gas request profiles commanded by a profes- proximation in the LPV case. It is well known that for LTI sys-
sional rider on race circuits tests, while the velocity bounds on tems, if the state space model corresponds to the
the variation of the zero and of the transfer constant have been transfer function then all the state
26 IEEE TRANSACTIONS ON CONTROL SYSTEMS TECHNOLOGY, VOL. 19, NO. 1, JANUARY 2011
Fig. 14. Time domain validation results of the LPV model at a frequency of 8.5 Hz (upper plot) and of 12.5 Hz (lower plot).
space models defined by , where is perturbations with amplitude and frequency tuned according to
square and nonsingular, are equivalent to the original one, in the their respective magnitude and velocity bounds. As for the set
sense that they give rise to the same input-output behavior. In the point variation , it was computed via the measured set-point
LPV case, however, the above notion of equivalence class does position according to (6) and used as input for the LPV model
not hold anymore. This concept is better illustrated by consid- simulation.
ering the state space representation (7) and the parameter-de- The results of this validation step are shown in Figs. 13 and
pendent coordinate change . If the coordinate 14, where the simulated and measured closed-loop throttle posi-
transformation is applied to system (7) one obtains tions are compared, using a highly exciting sine sweep reference
signal spanning the frequency range from 0.5 to 15 Hz. Specif-
ically, to increase readability, Figs. 13 and 14 show four details
of the validation results, at four different frequencies within the
whole frequency span of the experiment.
(8) As can be seen in Figs. 13 and 14 the simulated response
shows very good agreement with the measured one, thereby
From the above relations, it can be seen that the obtained re- confirming the validity of the state-space LPV model (7) ob-
alization (7) can be regarded as a good approximation of the tained for the closed-loop system. Once the LPV state space
LPV system, i.e., the state-space model can be considered suf- representation has been validated, it is possible to apply LPV
ficiently close to its I/O representation, only if the time varia- stability analysis techniques. Here reference is made to the fol-
tion of the underlying coordinate transformation, i.e., the term lowing result [23], [24].
is negligible, which corresponds to Theorem 6.1: The system (7) is stable if there is a matrix-
accounting for a static parametric dependence only in the I/O-to- valued function satisfying
state-space transformation. Unfortunately, a formal expression
for the approximation error as a function of the problem data is
very difficult to achieve, and this constitutes a challenging open
problem in the LPV modeling and identification context. How-
ever, it is possible to perform a validation step to experimen-
tally validate the LPV model (7). Note that this validation issue for all , where and is the bound
is rarely addressed in the LPV modeling and control literature, on the time derivative of the vector . The notation
even though it constitutes a crucial part in assessing the sound- indicates that every combination of and should be
ness of any LPV model and controller which is derived based included in the inequality.
on local models. The above problem is an infinite dimensional one. In partic-
Here, the validity of the state space closed-loop system (7) ular, the infinite dimensionality comes from the fact that
has been checked by simulating the LPV system (7) and com- is a function of and that the above conditions must hold for
paring the results with experiments carried out on the instru- all . Several techniques are available in the literature to
mented motorbike with the ETB controlled via the proposed reduce the problem to a finite dimensional one. In this context
gain-scheduled PID controller. To account for the parametric the parameter space gridding (see, e.g., [12]) has been preferred.
uncertainties in the ETB dynamics, in the simulations the two Namely, the following steps are performed:
uncertain parameters and were varied by applying sinusoidal 1) grid the set ;
CORNO et al.: DESIGN AND VALIDATION OF A GAIN-SCHEDULED CONTROLLER 27
(9)
Fig. 17. Plot of the driver request measured in a test track lap (dotted line) and
the measured throttle position (solid line).
Fig. 16. Comparison between the measured closed-loop system behavior (solid
line) and the intrinsic open-loop performance limits (dashed line): position (top)
and current (bottom).
[32] V. Verdult, “Nonlinear system identification: A state-space approach,” Mara Tanelli (M’05) was born in Lodi, Italy, in
Ph.D. dissertation, Fac. Appl. Phys., Univ. Twente, Enschede, The 1978. She received the Laurea degree in computer
Netherlands, 2002. science engineering and the Ph.D. degree in infor-
[33] M. Lovera and G. Mercere, “Identification for gain-scheduling: A bal- mation engineering with a thesis on active braking
anced subspace approach,” presented at the Amer. Control Conf., New control systems design for road vehicles from the
York, 2007. Politecnico di Milano, in 2003 and 2007, respec-
[34] R. Tóth, “Modeling and identification of linear parameter-varying sys- tively, and the Master of Science degree in computer
tems,” Ph.D. dissertation, Delft Center for Syst. Control, TU Delft, science from the University of Illinois, Chicago, in
Delft, The Netherlands, 2008. 2003.
[35] R. Tóth, F. Felici, P. S. C. Heuberger, and P. M. J. Van den Hof, “Dis- She is currently an Assistant Professor of auto-
crete time LPV I/O and state space representations, differences of be- matic control with the Dipartimento di Elettronica
havior and pitfalls of interpolation,” in Proc. Euro. Control Conf., 2007, e Informazione, Politecnico di Milano. She is also currently with the Dipar-
pp. 5418–5425. timento di Ingegneria dell’Informazione e Metodi Matematici, Università
[36] T. Kailath, Linear Systems. Upper Saddle River, NJ: Prentice-Hall, degli studi di Bergamo, Dalmine, Italy. Her main research interests focus
1980. on control systems design for ground vehicles, estimation, and identification
[37] J. Löfberg, “Yalmip : A toolbox for modeling and optimization in for automotive systems, control systems design for agricultural tractors, and
MATLAB,” presented at the CACSD Conf., Taipei, Taiwan, 2004. identification and control for active energy management of data centers.
[Online]. Available: http://control.ee.ethz.ch/joloef/yalmip.php Dr. Tanelli was a recipient of the Dimitri N. Chorafas Ph.D. Thesis Award and
1
[38] V. F. Montagner, R. C. L. F. Oliveira, and P. L. D. Peres, “Design the Claudio Maffezzoni Ph.D. Thesis Award for her Ph.D. thesis. In 2008, she
of H- gain-scheduled controllers for linear time-varying systems by and her coauthors received the Rudolf Kalman Best Paper Award for the best
means of polynomial Lyapunov function,” in Proc. 45th IEEE Conf. paper published in 2007 in the ASME Journal of Dynamic Systems Measurement
Decision Control, 2006, pp. 5839–5844. and Control.
[39] W. Xie, “H2 gain scheduled state feedback for LPV system with new
LMI formulation,” IEE Proc. Ñ Control Theory Appl., vol. 152, pp.
693–697, 2005.
[40] F. Wang and V. Balakrishnan, “Improved stability analysis and Sergio M. Savaresi (M’00) was born in Manerbio,
gain-scheduled controller synthesis for parameter-dependent sys- Italy, on 1968. He received the M.Sc. degree in elec-
tems,” IEEE Trans. Autom. Control, vol. 47, no. 5, pp. 720–734, May trical engineering and the Ph.D. degree in systems
2002. and control engineering from the Politecnico di Mi-
[41] P. Gahinet, P. Apkarian, and M. Chilali, “Affine parameter-dependent lano, Milan, Italy, in 1992 and 1996, respectively, and
lyapunov functions and real parametric uncertainty,” IEEE Trans. the M.Sc. degree in applied mathematics from the
Autom. Control, vol. 41, no. 3, pp. 436–442, Mar. 1996. Catholic University, Brescia, Italy, in 2000.
[42] R. Tóth, F. Felici, P. S. C. Heuberger, and P. M. J. Van den Hof, “Cru- After the Ph.D., he was a Management Consultant
cial aspects of zero-order hold LPV state-space system discretization,” with McKinsey & Company, Milan, Italy. He was
in Proc. 17th IFAC World Congr., 2008, pp. 4952–4957. a Visiting Researcher with Lund University, Lund,
Sweden; University of Twente, Ensende, The Nether-
lands; Canberra National University, Australia; Stanford University, Stanford,
CA; Minnesota University, Minneapolis; and Johannes Kepler University, Linz,
Austria. Since 2006, he has been a Full Professor in automatic control with the
Politecnico di Milano and is currently the Head of the “mOve” research team
(http://move.dei.polimi.it). He is an author of six patents, over 60 papers on In-
ternational Journals, and 150 papers on international conferences proceedings.
His main interests include the areas of vehicles control, automotive systems,
data analysis and system identification, nonlinear control theory, and control
applications.
Dr. Savaresi is an Associate Editor of the IEEE TRANSACTIONS ON CONTROL
SYSTEMS TECHNOLOGY, the European Journal of Control, and the IET Control
Theory and Applications. He is a member of the Editorial Board of the IEEE
Matteo Corno jointly received the Master of Science Control Systems Society.
in computer and electrical engineering from the Uni-
versity of Illinois, Chicago, and the “Laurea” Degree
cum laude and the Ph.D. degree cum laude with a
thesis on active stability control of two-wheeled ve- Luca Fabbri was born in Owo, Nigeria, in 1963. He
hicles from the Politecnico di Milano, Milan, Italy, in received the M.Sc. degree in mechanical engineering
2005 and 2009, respectively. from the University of Padova, Padova, Italy, in 1990.
During his Ph.D., he had a six-month internship at From 1990 to 1993, he was with Aprilia, working
Alenia Spazio (now Thales Alenia Space). In 2008, as a mechanical designer in the racing unit. From
he had been a Visiting Scholar with the University 1993 to 2006, he was responsible for vehicle devel-
of Minnesota, Minneapolis. In 2009, After a joint opment in the racing unit, where he led the design
post-doc position at Politecnico di Milano and Johannes Kepler University, and development of racing motorcycles for the
Linz, he joined Delft University of Technology, Delft, The Netherlands, as an categories 125cc, 250cc, 500cc, and SuperBike.
Assistant Professor with the Delft Center for System and Control. His current Currently, he is the Innovation Manager of the
research interests include dynamics and control of two and four wheeled Motorcycle Engineering section for the brand Units
vehicles, nonlinear estimation techniques, and LPV control. Aprilia e Derbi within the Piaggio Group.