See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/3415114

Gearshift control for automated manual transmissions

Article in IEEE/ASME Transactions on Mechatronics · March 2006

DOI: 10.1109/TMECH.2005.863369 · Source: IEEE Xplore

CITATIONS READS
211 7,039

4 authors:

Luigi Glielmo Luigi Iannelli

Università degli Studi del Sannio Università degli Studi del Sannio
264 PUBLICATIONS 4,439 CITATIONS 122 PUBLICATIONS 1,856 CITATIONS

SEE PROFILE SEE PROFILE

Vladimiro Vacca Francesco Vasca

Fiat Chrysler Automobiles Group, Pomigliano D'Arco, Italy Università degli Studi del Sannio
6 PUBLICATIONS 257 CITATIONS 186 PUBLICATIONS 3,625 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

HAEOLUS Project View project

Unmanned Aerial Vehicles View project

All content following this page was uploaded by Luigi Glielmo on 27 May 2015.

The user has requested enhancement of the downloaded file.

IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 11, NO. 1, FEBRUARY 2006 17

Gearshift Control for Automated

Manual Transmissions
Luigi Glielmo, Member, IEEE, Luigi Iannelli, Member, IEEE, Vladimiro Vacca, and Francesco Vasca, Member, IEEE

Abstract—A gearshift control strategy for modern automated startup operating conditions have been proposed: quantitative
manual transmissions (AMTs) with dry clutches is proposed. The feedback theory [5], model predictive control strategy [6], fuzzy
controller is designed through a hierarchical approach by dis- control [7], decoupling control [4], and optimal control [8], fur-
criminating among five different AMT operating phases: engaged,
slipping-opening, synchronization, go-to-slipping, and slipping- ther in [9], the authors propose a particular engagement tech-
closing. The control schemes consist of decoupled and cascaded nique. Problems and solutions related to the clutch engagement
feedback loops based on measurements of engine speed, clutch during the gearshift phase have been also considered in the liter-
speed, and throwout bearing position, and on estimation of the ature. In [10], an analytical procedure for computing the desired
transmitted torque. Models of driveline, dry clutch, and controlled engine speed during upshift and downshift is proposed. In [11],
actuator are estimated on experimental data of a medium size gaso-
line car and used to check through simulations the effectiveness of a model-based backstepping methodology is used to design the
the proposed controller. gearshift control in AMTs without the synchronizer. In [12], a
neuro-fuzzy approach is used by considering the driver’s inten-
Index Terms—Automated manual transmissions (AMTs), auto-
motive control, clutch engagement control, dry clutch, gearshift. tion and variable loads.
In spite of the extensive literature on AMT control, some
problems still need further investigation: the role of speed
I. INTRODUCTION feedback loops in the clutch engagement control, the definition
ARS with modern transmission systems exhibit high fuel of a controller architecture which can be exploited both during
C economy, low exhaust emission, and excellent driveabil-
ity. Recent reports on the future automotive market forecast that
vehicle startup and gearshift, the robustness of the solution
with respect to clutch aging, and uncertainties in the clutch
in 2010 the production of manual transmissions will have fallen characteristic. This paper tries to provide a contribution in this
below 50% while the modern automatic transmissions will have direction by proposing a new controller for gearshift and clutch
reached 25% of production [1], [2]. Among other responces, the engagement in AMTs.
automated manual transmissions (AMTs) represent a promising The paper is organized as follows. In Section II, models of
solution since they can be considered as an inexpensive add-on driveline, dry clutch, and closed-loop electrohydraulic actuator
solution for classical (in European and Latin countries) man- are considered and tuned on experimental data. In Section III,
ual transmission systems. Moreover, AMTs are also extensively five different operating phases of the AMT are considered: en-
used in racing cars and as a reconfiguration element in modern gaged, slipping-opening, synchronization, go-to-slipping, and
hybrid electric vehicles. slipping-closing. The controllers, designed through a hierar-
One of the most critical operations in AMTs is represented by chical approach with decoupled and cascaded feedback loops
the gearshift and more specifically by the clutch engagement. based on measurements of clutch speed, engine speed, and
In automotive drivelines, the goal of the clutch is to smoothly throwout bearing position, are presented in Section IV. The con-
connect two rotating masses, the flywheel and the transmission trolled AMT is simulated in the Matlab environment where the
shaft, that rotate at different speeds, in order to allow the transfer Simulink scheme corresponding to the current AMT phase and
of the torque generated by the engine to the wheels through the corresponding controller are selected by a Stateflow finite
the driveline. The automation of the clutch engagement must state machine. Simulation results showing the effectiveness of
satisfy different and conflicting objectives: It should obtain at the proposed approach are presented in Section V. Conclusions
least the same performance manually achievable by the driver that synthesize the results of the paper are reported in Section VI.
(short gearshift time and comfort) and improve performance in II. MODELING
terms of emission and facing wear. The engine and clutch speeds
during the engagement and at the lockup play an important role A. Driveline
both for comfort and friction losses [3], [4]. A driveline model suitable for parameter identification and for
In order to achieve the objectives of the clutch engagement au- the clutch engagement control design can be obtained assuming
tomation, several control approaches which deal with the vehicle the clutch speed ωc is equal to the mainshaft speed ωm , and
considering the mainshaft rigid (Fig. 1). Thus, when the engine
Manuscript received March 20, 2004; revised December 23, 2004. Recom-
flywheel and the clutch disk are in slipping operating conditions,
mended by Technical Editor H. Peng. the driveline model can be written as
The authors are with the Dipartimento di Ingegneria, Università degli
Studi del Sannio, 21-82100 Benevento, Italy (e-mail: [email protected]; Je ω̇e
[email protected]; [email protected]).
Digital Object Identifier 10.1109/TMECH.2005.863369 = Te − Tc (xc ) (1)

1083-4435/$20.00 © 2006 IEEE

18 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 11, NO. 1, FEBRUARY 2006

eration pedal positions and vehicle speeds. The clutch torque

has been estimated by an offline inversion of (1):
T̂c = Te − Je ω̇
ˆe (7)
where ω̇ ˆ e is obtained through the so-called dirty derivative,
i.e., a filtered incremental ratio on the measured engine speed.
The model parameters have been identified by using the least
square method and the corresponding results have been found as
follows: Jc + Jeq = 0.004 kg·m2 , Jw = 133 kg·m2 , βtw = 295
Fig. 1. Driveline scheme. N·m/(rad/s), and ktw = 6200 N·m/rad. Note that in the sum
Jc + Jeq , the dependence of Jeq on the gear ratio is negligible.
It can be also verified that, as typical for the type of driveline
[Jc + Jeq (ig , id )]ω̇c and car under investigation, the first resonance in the frequency

 
1 ωc response of the model appears at a few hertz.
= Tc (xc ) − ktw ∆θcw + βtw − ωw (2)
ig id ig id
B. Dry Clutch
Jw ω̇w
  From a physical point of view, the dry clutch consists of two
ωc
= ktw ∆θcw + βtw − ωw − TL (ωw ) (3) disks (the clutch disk connected to the mainshaft and the fly-
ig id wheel disk connected to the engine) covered with a high friction
∆θ̇cw material and a mechanism which presses the disks against each
ωc other (clutch closed or engaged) or keeps them apart (clutch
= − ωw (4) open or disengaged). During an engagement phase, the clutch
ig id
disk is moved towards the flywheel disk until the friction due
where J’s are inertias; ω’s speeds; T ’s torques; xc the throwout to their contact allows the torque transmission. The throwout
bearing position; and the subscripts e, c, t, and w indicate engine, bearing position xc determines the pressure between the fly-
clutch, transmission, and wheels, respectively. Moreover, ig is wheel disk and the clutch disk and, therefore, the transmitted
the gear ratio, id is the differential ratio, Jeq (ig , id ) = Jm + torque during the slipping phases. The nonlinear characteristic
(1/i2g )(Js1 + Js2 + (Jt /i2d )), Js1 and Js2 are the inertias of the Tc (xc ) that relates the throwout bearing position xc to the torque
two disks connected to the synchronizer, Jm is the mainshaft transmitted by the clutch is not easy to model. The clutch wear
inertia, TL is the load torque, θ’s are angles, k’s are elastic drastically influences such a characteristic and thus the torque
stiffness coefficients, and β’s are friction coefficients. When the transmission. Moreover, the clutch characteristic is also influ-
clutch is engaged, the engine speed ωe and the clutch disk speed enced by the dependence of the friction coefficient on both tem-
ωc are equal. The corresponding engaged model can be obtained perature [14] and slip speed. In particular, the negative variations
by adding (1) and (2) with the assumption ωe ≡ ωc . For more of the friction coefficient with slip speed can induce torsional
details see [13], where a more complex model that also takes self-excited vibrations of the driveline [15], [16].
into account the flexibility of the mainshaft has been derived. The nominal nonlinear characteristic Tc (xc ) has been identi-
A further reduction of the model can be obtained by also fied from the experimental tests described above. By represent-
considering the driveshaft to be rigid. By assuming ωc ≡ ωm ≡ ing the estimated values of the clutch torque obtained through
ig id ωw and by reporting the vehicle inertia to the mainshaft, one (7) as a function of the corresponding values of the signal xc ,
obtains the following model: the set of points reported in Fig. 2 (top diagram) is obtained.
Je ω̇e = Te − Tc (xc ) (5) The absolute value of the clutch torque has been modeled by
  using the interpolation curve reported in bottom diagram of
ωc Fig. 2 (curve b), and its variations (curves a and c) which model
Jv (ig , id )ω̇c = Tc (xc ) − TL (6)
ig id different clutch wear and have been used to test the controller
robustness. Note that the wear changes the bearing position, say
where Jv (ig id ) = Jc + Jeq (ig id ) + (Jw /i2g i2d ). The corre-
x̄c , at which the two disks come in contact and the transmitted
sponding engaged model can be obtained by adding (5) and
torque becomes different from zero.
(6) with ωe ≡ ωc .
The model (1)–(4) provides a good compromise between
C. Clutch Actuator
description of the driveline dynamics and model complex-
ity. The parameters of (1)–(4) have been tuned from exper- In AMTs, the electrohydraulic actuator is mainly composed
imental data carried out on a FIAT STILO 2.4 gasoline car of a hydraulic piston connected to a system of springs that
with Je = 0.2 kg·m2 , id = 3.94, and the set of gear ratios keep the clutch closed when the piston does not apply any force
ig = [3.08, 2.23, 1.52, 1.16, 0.91] from the first gear to the fifth (see Fig. 3). The piston is controlled by a three-port electrovalve
gear, respectively. The signals ωe , ωc , ωw , and Te have been that regulates the oil flow through the hydraulic circuit and then
acquired with a sampling frequency of 100 Hz during tests in determines the force on the mechanical actuator. In standard
which a series of upshifts were carried out with different accel- commercial applications, the electrohydraulic actuator is used
GLIELMO et al.: GEARSHIFT CONTROL FOR AUTOMATED MANUAL TRANSMISSIONS 19

where Mv is the spool mass, yv is the spool position, Fm is

the force caused by the solenoid current I, bv is the friction
coefficient, Kv is the effective spring constant, and Fk o is the
mechanical spring force due to the preload of the spring. The
oil force Foil is the Bernoulli’s force defined as [18]

Foil = ϕCv wyv ∆p cos(ϑ(yv )) (9)

where ϕ is the discharging coefficient, Cv is the velocity

coefficient, w is the control port width, ϑ(yv ) is the jet an-
gle described by a polynomial function of the spool posi-
tion (ϑ(yv ) = C3 yv3 + C2 yv2 + C1 yv + C0 ), and the variation
of pressure in the three-port electrovalve is defined as

pL − pA for L-A opened (yv > 0) → Clutch open
∆p =
pA − pT for A-T opened (yv ≤ 0) → Clutch close
where pL is the line pressure, pT is the tank pressure, and pA is
Fig. 2. Set of points (xc , Tˆc ) obtained from experimental data (top diagram) the oil pressure in the actuator chamber. The variation of the oil
and clutch characteristics (bottom diagram) for positive slip speed with different pressure pA can be described by
wear. Curve a: new clutch. Curve b: medium wear. Curve c: high wear.
E
ṗA = [Q(yv , ∆p) − Ap ẋc ] (10)
Ap xc + Vt

where E is the fluid bulk modulus, Ap is the piston cross sec-

tional area, xc is the actuator position (or, equivalently, the
throwout bearing position), Vt is the minimum actuator cham-
ber volume (corresponding to xc = 0), Q(yv , ∆p) is the oil flow
in the servocylinder computed as shown by the equation at the
bottom of the page, where d is the underlap to the port, Clk is
the leakage coefficient, and ρ is the oil density. The last equa-
tion of the hydraulic actuator model describes the motion of the
servocylinder piston as

(Mp + Mc )ẍc = −bp ẍc − Fspring (xc ) + Ap pA (11)

Fig. 3. Hydraulic actuator scheme corresponding to the clutch engaged.
where Mp and Mc are respectively the piston and the clutch
mass, bp is the friction coefficient, and Fspring is the non-
with a feedback control on the throwout bearing position. Usu- linear spring force of the diaphragm spring modeled through
ally a further inner control loop on the current is implemented. a static characteristic. For further details on the model,
By identifying the parameters of the detailed actuator model see [17].
proposed in [17], we now show that, for the goal of this paper, Some considerations on the model are needed. First of all,
the actuator with a position feedback loop can be satisfactorily several parameters (e.g., bulk modulus and other oil parameters)
approximated by a first-order linear system. depend on temperature and operating conditions. Furthermore,
The actuator model consists of a set of equations describing some hydraulic dynamics due to the pipeline and the actuator
the dynamics of the electrovalve spool, the servocylinder piston, chamber are difficult to model. So the model does not catch
and the pressure variation in the mechanical actuator chamber. all dynamics and second-order phenomena of the actual sys-
The motion of the valve spool is described by a force balance tem, yet it is sufficiently detailed for showing that, when the
equation actuator is controlled by using a feedback loop on the posi-
tion, the overall control system is quite robust with respect to
Mv ÿv = Fm (I, yv ) − Foil (∆p, yv ) − bv ẏv − Kv yv − Fk o parameter variations and uncertainties as well as unmodeled
(8) dynamics.



 (yv + d)ϕw 2|∆p|
− Clk (∆p), for yv > d
 ρ sign(∆p)

2|∆p| 2|∆p|
Q = (yv + d)ϕw ρ sign(∆p) + (yv − d)ϕw ρ sign(∆p), for − d < yv < d



 (y − d)ϕw 2|∆p|
v ρ sign(∆p), for yv < −d
20 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 11, NO. 1, FEBRUARY 2006

TABLE I formance of the overall engagement control strategy may be

ACTUATOR PARAMETERS, KNOWN (K) OR IDENTIFIED (I)
relevant. However, the actuator is typically equipped with a lo-
cal feedback control on the bearing position which provides
robustness to the closed-loop system. In order to show that, a
position feedback control loop with a classical regulator (PI) has
been added to the identified open-loop model. Some simulations
have been carried out by considering the nominal model and the
perturbed one. In the bottom diagram of Fig. 4, the outputs
(bearing position) of the controlled system are reported in the
nominal (solid line) and perturbed (dashed line) cases: The two
curves are almost equal, thus showing the robustness provided
by the position feedback control. Therefore, for the purposes
of evaluating the performance of an AMT control strategy, an
equivalent model of the controlled actuator not dependent on
the actuator parameters can be enough. An identification pro-
cedure has been then carried out in order to approximate the
complete closed-loop actuator model with a simpler one. The
input of the model to be identified is now the reference bearing
position and the output is the actual bearing position. So as jus-
tified by the simulation results (dotted line) reported in Fig. 4, a
first-order linear system has been found to be a good approxima-
tion for the closed-loop actuator model. This equivalent model
has been used in the gearshift simulations.

III. AMT OPERATING PHASES

In order to determine a possible classification of the AMT
operating phases, let us consider the engine speed and clutch
speed signals during the gearshift reported in Fig. 5. Five dif-
ferent phases can be identified: engaged, slipping-opening, syn-
chronization, go-to-slipping, and slipping-closing (see Fig. 6).
During ordinary operating conditions the AMT is in the en-
gaged phase, the clutch is locked up, and the engine torque is
directly transmitted to the driveline. A gearshift request from
the driver corresponds to the start of the clutch opening phase:
The throwout bearing position is decreased and the clutch disk
Fig. 4. Clutch bearing positions in open loop (top diagram) and closed loop starts slipping with respect to the flywheel disk, although the two
(bottom diagram) in the nominal case and the perturbed case. In the bottom masses rotate at approximatively the same speed (see Fig. 5).
diagram, the position obtained with an equivalent first-order system (dotted When the clutch is fully opened, a new gear can be engaged and
line) is also reported.
the synchronization phase starts. The clutch disk speed moves
quickly toward the speed value corresponding to the vehicle
The parameters of the model have been identified by us- speed reported to the mainshaft through the new gear ratio (dur-
ing the available experimental data on a controlled actuator: ing the synchronization phase the vehicle speed can be assumed
the solenoid current I and the throwout bearing position xc . to be constant because of the short duration of the phase with
A backstepping approach has been used to identify the servo- respect to the vehicle dynamics). During the synchronization
cylinder parameters, whereas a least square approach has been phase the engine speed also starts reducing (see Fig. 5). Once
used to identify the electrovalve parameters. The results of the the new gear is fully engaged, the flywheel disk and the clutch
identification procedure are reported in Table I. disk must be connected again and the go-to-slipping phase starts.
The actuator behaviour is nonnegligibly dependent on its In that phase, the bearing position increases and should move
model parameters. To verify this, the identified model has been as quickly as possible toward the value x̄c for which the clutch
simulated in open loop with an increased (+20%) value of the disk and the flywheel come into contact. For xc > x̄c the AMT
bulk modulus. In the top diagram of Fig. 4, the experimental is in the slipping-closing operating phase; the speed regulation
clutch bearing position is reported (solid line) compared with in this phase is very important for the overall performance of the
the clutch bearing position corresponding to the same input AMT. When the clutch and the flywheel reach the same speed,
signal but with the perturbed parameter (dashed line). Thus, they are locked up and a new engaged operating phase of the
the influence of the actuator parameter’s variations on the per- AMT starts.
GLIELMO et al.: GEARSHIFT CONTROL FOR AUTOMATED MANUAL TRANSMISSIONS 21

Fig. 5. Engine speed and clutch speed signals during a gearshift; the five operating phases are highlighted.

system switches between two different configurations: one in

which ωe and ωc are different velocities and one in which they
coincide. Such a switch excites the system dynamics and it is not
difficult to realize that a rough measure of the extent of excita-
tion is given by the discontinuity ω̇c (t̄+ ) − ω̇c (t̄− ). To compute
this discontinuity, let us assume, for simplicity’s sake that the
engine torque is continuous at t̄. Now, we add (1) and (2) and
use the fact that ωe ≡ ωc after lockup to obtain
1
ω̇c (t̄+ ) = [Te (t̄+ ) − Ω(t̄+ )] (13)
Je + J˜c
where

 
1 ωc
Fig. 6. Finite-state diagram of the AMT operating phases. Ω= ktw ∆θcw + βtw − ωw . (14)
ig id ig id

By assuming the state and the engine torque to be continuous at

IV. GEARSHIFT CONTROL lockup one obtains
A. Controller Objectives
1
During a gearshift, the clutch engagement should be carried ω̇c (t̄+ ) = [Te (t̄− ) − Tc (t̄− )]
Je + J˜c
out as quickly as possible and with strict constraints on driv-
1
ing comfort and engine operating conditions. The fundamental + [Tc (t̄− ) − Ω(t̄− )]
constraint on the clutch engagement is the so-called no-kill con- Je + J˜c
ditions, i.e., one must avoid the engine stall. This condition can Je J˜c
be modeled by imposing that = ω̇e (t̄− ) + ω̇c (t̄− ) (15)
˜
Je + Jc Je + J˜c
ωe (t) ≥ ωemin ∀t. (12)
which leads to
A further important condition to be satisfied during the en- Je
gagement is the so-called no-lurch condition. A nonsmooth en- ω̇c (t̄+ ) − ω̇c (t̄− ) = ω̇sl (t̄− ). (16)
Je + J˜c
gagement process determines a mechanical oscillation of the
power train, which should be avoided in order to preserve the Thus, the engagement smoothness is somehow related to the
passengers comfort. To gain some insight into the phenomenon, slip acceleration at lockup and, hence, we will consider this
we first point out that at the time of engagement, e.g., t̄, the quantity as a quantitative performance criterion.
22 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 11, NO. 1, FEBRUARY 2006

Fig. 7. General closed-loop scheme during the slipping-closing phase. Except for the gearshift controller, the scheme is also the same for the other phases.

Another control objective consists of maintaining as low as

possible the energy dissipated during the engagement, which
can be written as
t¯
Ed = ωsl (t)Tc (t) dt. (17)
0
In our approach, the control objectives are met by ensuring
that the engine speed ωe and the clutch speed ωc (or the slipping
speed ωsl ) track desired reference signals. In what follows, the
controllers of the different AMT phases are described, starting
from the critical slipping-closing phase, which is also the most
important phase during the vehicle launch, i.e., starting from a
standstill. The input signals for the controllers are measurements Fig. 8. Block diagram of the gearshift controller during the slipping-closing
of the engine speed and clutch speed, and the estimation of phase.
the transmitted torque during the slipping phases. At a lower
TABLE II
hierarchical level, which is only ideally modeled in this paper, GEAR-RATIO-DEPENDENT PARAMETERS OF THE PI CONTROLLERS
measurements of the throwout bearing position and estimation
of the engine torque are also needed.

B. Slipping-Closing Controller
In this phase, a new gear has already been engaged, the
throwout bearing position has reached x̄c , and the clutch has
started transmitting torque to the driveline. The general architec-
ture of the controlled system is reported in Fig. 7. The driveline
to the model (1)–(4) classical single-input single-output linear
model to be considered is (1)–(4). The controller output signals
controller design techniques. Furthermore, the controllers’ pa-
are the reference engine torque Teref and the reference throwout
rameters are designed (and then scheduled at run-time) for each
bearing position xref
c , and are generated from the gearshift con- gear so as to obtain the same closed-loop decoupled transfer
troller on the basis of the reference speeds ωeref and ωcref . The
matrix and, hence, the same performance. By assuming for C2
signal Teref is actuated by the engine control unit, which is here
and C3 the transfer function form KP + KI /s, the resulting pa-
assumed to be ideal, i.e., Te = Teref . The reference signal xref
c is rameters are KI = 1.9 and KI = 25, respectively, for C2 and
actuated by the closed-loop electrohydraulic actuator approxi-
C3 , and proportional gains, are reported in Table II.
mated by the first-order system presented in the previous section.
An inner feedback loop on the clutch torque is introduced
Finally, xc is converted to the transmitted torque Tc through the
in order to compensate for the uncertainties on the static torque
clutch characteristic.
characteristic (see Fig. 2). The nonlinearities in the clutch model
A block diagram of the gearshift controller is shown in Fig. 8.
Tc (xc ) make it difficult to design the controller C4 through an
The controller C1 realizes a feedforward action obtained by
analytical procedure. Here, C4 has been chosen as a PI controller
computing the left-hand side of (1) after replacing the actual en-
whose parameters have been manually tuned (KP 4 = 0.01 and
gine speed with the corresponding reference signal. The feed-
KI 4 = 0.5). Robustness of the closed-loop system with respect
back loop on the clutch speed provides the reference clutch
to different C4 parameters have been checked in simulations
torque Tcref . The loop on the engine speed, together with the
showing that such parameters are not critical for the closed-
corresponding feedforward compensation, provides the desired
loop performance.
difference between the engine torque and the clutch torque.
By adding to this signal the transmitted torque [estimated by
C. Engaged Controller
using (7)], the reference engine torque Teref is obtained. This
scheme can be viewed as a decoupling controller (see [4]) and In this phase, the clutch is fully engaged and the transmitted
thus the two controllers C2 and C3 are PI designed by applying torque is directly supplied by the engine to the mainshaft. The
GLIELMO et al.: GEARSHIFT CONTROL FOR AUTOMATED MANUAL TRANSMISSIONS 23

driveline model is obtained from (1)–(4) by adding (1) and

(2) with the assumption ωe ≡ ωc . Since the clutch is completely
closed, i.e., not controlled, the only control signal is the reference
engine torque, which is regulated by the engine control unit.

D. Slipping-Opening Controller
When a new gearshift is requested, it is necessary to open
the clutch and, thus, a new slipping phase starts. The driveline
model is (1)–(4). The controller output signals are Teref and xref
c .
The reference torque is obtained with the engine speed control
loop of the scheme reported in Fig. 8. The signal ωeref is constant
for comfort reasons (see Fig. 5). The throwout bearing reference
position xref
c is selected in open loop as a decreasing ramp. This
phase is concluded when the engine flywheel disk and the clutch
disk are completely separated.

E. Synchronization Controller Fig. 9. Engine, clutch, and vehicle speeds for a sequence of upshifts with
medium wear of the clutch.
During this phase, the clutch disk is completely separated
from the flywheel disk and, therefore, Tc = 0. Since during this
phase the vehicle speed can be assumed constant, e.g., ω̄w , the TABLE III
PERFORMANCE FOR A SEQUENCE OF UPSHIFTS
driveline model can be obtained from (5) and (6) with Tc = 0
and the load torque replaced by the synchronization torque Ts .
The controller output signals are the reference engine torque
and the reference synchronization torque. Such control signals
can be obtained with two independent single-input single-output
feedback loops designed by using classical control methods on
the linear model described above. The reference clutch speed is
assumed to vary linearly from the speed at which the synchro-
nization phase starts and the speed corresponding to the wheel a state-dependent condition, i.e., ωe = ωc , whereas an external
speed transformed by the new gear ratio, i.e., ω̄c = ω̄w /(ig id ), event (the driver gearshift request) determines the commuta-
within a desired time interval. tion from the engaged phase to the slipping-opening phase.
Another state-dependent event (xc = x̄c ) determines the com-
F. Go-to-Slipping Controller mutation from the go-to-slipping phase to the slipping-closing
phase. The commutation among the different controllers is done
The main aim of this phase is to reach the bearing position
by setting their initial conditions so that bumpless transfer is
x̄c at which the friction between the flywheel and the clutch
ensured.
disk, and hence the torque transmission, starts. The driveline
Figs. 9 and 10 show the simulation results of the controlled
model is (1)–(4). Since in this phase xc < x̄c , the clutch torque
AMT for a sequence of consecutive upshifts. The behaviors of
is zero (see Fig. 2). The control signals are Teref and xref
c . The
the engine speed and clutch speed, and those of the throwout
reference engine torque is obtained with the same controller
bearing position and engine torque, are presented. In particular,
used in the slipping-opening phase. The throwout bearing refer-
Table III shows some performance indexes of the simulation test:
ence position xref
c is set (without clutch speed feedback) to the
dissipated energy, gearshift time duration, and slip acceleration
constant value x̄c . The value x̄c can be obtained by exploiting
at lockup are all within acceptable limits and testify the validity
the clutch torque estimator and averaging the throwout bearing
of the proposed control strategy. Dealing with driving comfort in
positions at which torque started to be transmitted in the most
particular, by assuming that the lockup phenomenon has a finite
recent gearshifts.
time duration, the left-hand side of (16) can also be interpreted
as an incremental ratio and, therefore, as an approximation of
V. SIMULATION RESULTS the jerk caused by the lockup of the clutch.
The simulation results have been obtained by implementing We wish to remark that the controller structure for the
the driveline model and the controllers of the different AMT slipping-closing phase can also be used both during gearshift
phases in the Matlab environment. The parts of the model that and when the clutch must be locked up from standstill. Fig. 11
change with the AMT phases are implemented with differ- shows that the control strategy also ensures a smooth clutch
ent Simulink schemes; the model corresponding to the active engagement during the vehicle startup, which is one of the most
phase is selected by a Stateflow finite state machine similar to critical situations for AMTs. It is apparent that the bearing posi-
that shown in Fig. 6. For instance, the commutation from the tion and the engine torque have similar shapes. This is due to the
slipping-closing phase to the engaged phase is obtained with fact that in the main part of the slipping-closing phase the clutch
24 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 11, NO. 1, FEBRUARY 2006

Fig. 10. Zoom of the upshift from gear 2 to gear 3 with medium wear of the clutch. (a) Engine speed (solid line), clutch speed (dashed line), desired engine speed
(dashed-dotted), and desired clutch speed (dotted line). (b) Throwout bearing position. (c) Engine torque.

Fig. 11. Zoom of the startup phase with medium wear of the clutch. (a) Engine speed (solid line), clutch speed (dashed line), desired engine speed (dashed-dotted),
and desired clutch speed (dotted line). (b) Throwout bearing position. (c) Engine torque.

is operating in an approximately linear region of the transmitted The simulation results reported in Fig. 13 show the robustness
torque characteristic reported in Fig. 2 and the engine speed is of the proposed control strategy in the presence of uncertainties
varying slowly, from (1) Te ≈ Tc (xc ). The proposed controller or variations of the clutch static characteristics Tc (xc ). Dur-
is also effective for downshifts as shown in Fig. 12. ing the slipping-opening phase, corresponding to the first part
GLIELMO et al.: GEARSHIFT CONTROL FOR AUTOMATED MANUAL TRANSMISSIONS 25

Fig. 12. Downshift from gear 5 to gear 4 in the presence of a clutch with medium wear (curve b in Fig. 2). (a) Engine speed (solid line), clutch speed (dashed
line), desired engine speed (dashed-dotted), and desired clutch speed (dotted line). (b) Throwout bearing position. (c) Engine torque.

Fig. 13. Upshift from 2 gear to 3 gear in the presence of a clutch with high wear (curve c in Fig. 2). (a) Engine speed (solid line), clutch speed (dashed line),
desired engine speed (dashed-dotted), and desired clutch speed (dotted line). (b) Throwout bearing position. (c) Engine torque.

of Fig. 13, the engine speed has an overshoot due to the fact VI. CONCLUSION
that for this simulation x̄c is larger than in the previous case The analysis of the existing literature on AMT control strate-
(see curves b and c in Fig. 2). By comparing Figs. 10 and 13 gies shows that a generally recognized good solution to the
during the slipping-closing phase, analogous considerations jus- problem of the gearshift control in AMT with dry clutch is
tify the fact that the controller imposes larger values of xc in the still missing. In this paper, a solution based on cascaded and
latter simulation in order to obtain similar performance. decoupled speed and torque control loops is proposed. Dynamic
26 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 11, NO. 1, FEBRUARY 2006

models of the driveline, the static characteristic of the torque Luigi Glielmo (S’87–M’90) was born in 1960. He
transmitted by the dry clutch during the slipping phases, and received the Laurea degree in electronic engineering
and the Ph.D. degree in automatic control from the
an equivalent model of the controlled electrohydraulic actuator Università di Napoli Federico II, Naples Italy.
have, been tuned on experimental data and used for the con- Currently, he is a Professor of Automatic Con-
trollers design. The control strategy exploits the hybrid nature trol in the School of Engineering of the Università
degli Studi del Sannio, Benevento, Italy, where he is
of the AMT by discriminating among five different operating also Head of the Dipartimento di Ingegneria. In 1989
phases and by designing dedicated controllers for each phase. and 1990, he was a Visiting Scholar at the School of
Simulation results show the effectiveness of the proposed so- Aeronautics and Astronautics, Purdue University, and
a Visiting Scientist at the NET Team, Max-Planck-
lution during startups, upshifts, and downshifts. Performance Institut für Plasmaphysik, Germany, in 1990. In 2002, he was Visiting Professor
is evaluated in terms of duration of gearshift, comfort, dissi- at the Johannes Kepler Universität, Linz, Austria. His research interests include
pated energy, and robustness with respect to uncertainties of singular perturbation methods, analysis and control of uncertain systems, dy-
namic positioning of ships, plasma control in (Tokamak) fusion reactors, Kalman
the clutch characteristic. The low computational load needed filtering, nonlinear system analysis, automotive control, manufacturing systems
to implement the proposed controller allows its realization on simulation, modeling, and control of wine production. He has coauthored more
commercial electronic control units. than 70 papers published in international archival journals and proceedings of
international conferences, and edited Robust Control via Variable Structure and
Lyapunov Techniques (Springer-Verlag, 1996).
Dr. Glielmo has participated in scientific committees of various interna-
REFERENCES tional conferences and organized an international workshop. He proposed the
establishment of and chaired for four years the IEEE Control Systems Soci-
[1] Knibb-Gormezano and Partners. (2000) Trasmissions in Europe 2000– ety Technical Committee on Automotive Controls. He is a member of ASME,
2010: A decade of significant change. [Online]. Available: http://www. SIAM, and SAE and is an Associate Editor of the IEEE TRANSACTIONS ON
kgpauto.com/Initiatives/Transmissions%20Press%20Release.pdf AUTOMATIC CONTROL.
[2] Just-Auto.com Editorial Team (2001, Jan.) The global transmission mar-
ket. [Online]. Available: http://www.just-auto.com/features detail.asp?
art=417&c=1 Luigi Iannelli (S’01–M’03) was born in 1975 in
[3] Y. Lei, M. Niu, and A. Ge, “A research on starting control strategy of Benevento, Italy. He received the Laurea degree in
vehicle with AMT,” in Proc. FISITA World Automotive Congress, Seoul, computer engineering from the Università degli Studi
Korea, 2000. del Sannio, Benevento, Italy, in 1999, and the Ph.D.
[4] F. Garofalo, L. Glielmo, L. Iannelli, and F. Vasca, “Smooth engagement for degree in information engineering from the Univer-
automotive dry clutch,” in Proc. 40th IEEE Conf. Decision and Control, sità di Napoli Federico II, Naples, Italy, in 2003.
Orlando, FL, 2001, pp. 529–534. During 2002 and 2003, he was a Guest Researcher
[5] J. Slicker and R. N. K. Loh, “Design of robust vehicle launch control in the Department of Signals, Sensors, and Systems,
system,” IEEE Trans. Contr. Syst. Technol., vol. 4, no. 4, pp. 326–335, Jul. Royal Institute of Technology (KTH), Stockholm,
1996. Sweden. In 2003, he was a Research Assistant in the
[6] A. Bemporad, F. Borrelli, L. Glielmo, and F. Vasca, “Hybrid control of Dipartimento di Informatica e Sistemistica, Univer-
dry clutch engagement,” in Proc. European Control Conf., Porto, Portugal, sità di Napoli Federico II. He is currently an Assistant Professor of Automatic
2001. Control in the Dipartimento di Ingegneria, Università degli Studi del Sannio. His
[7] H. Tanaka and H. Wada, “Fuzzy control of clutch engagement for au- current research interests include analysis and control of switched and hybrid
tomated manual transmission,” Veh. Syst. Dyn., vol. 24, pp. 365–376, systems, automotive control, and applications of control theory to electronic
1995. systems.
[8] L. Glielmo and F. Vasca, “Optimal control of dry clutch engagement,” Dr. Iannelli is member of the IEEE Control Systems Society, IEEE Circuits
SAE Trans., J. Passenger Cars: Mech. Syst., vol. 6, no. 2000-01-0837, and Systems Society, and SIAM.
2000.
[9] A. Szadkowski and R. B. Morford, “Clutch engagement simulation: En- Vladimiro Vacca was born in 1975 in Naples, Italy.
gagement without throttle,” SAE Tech. Paper Series, no. 920766, 1992. He received the Laurea degree in computer engi-
[10] F. Amisano, G. Serra, and M. Velardocchia, “Engine control strategy to neering with orientation toward industrial automation
optimize a shift transient during clutch engagement,” SAE Tech. Paper from the Università di Napoli Federico II, Naples,
Series, no. 2001-01-0877, pp. 115–120, 2001. Italy, in 2002. He is currently working toward the
[11] J. Fredriksson and B. Egardt, “Nonlinear control applied to gearshifting Ph.D. degree in control engineering, sponsored by
in automated manual transmission,” in Proc. 39th IEEE Conf. Decision Carlo Gavazzi Space, in the Dipartimento di Ingeg-
and Control, Sydney, Australia, 2000, pp. 444–449. neria, Università degli Studi del Sannio, Benevento,
[12] K. Hayashi, Y. Shimizu, S. Nakamura, Y. Dote, A. Takayama, and Italy.
A. Hirako, “Neuro fuzzy optimal transmission control for automobile From July 2002 to June 2003, he was involved
with variable loads,” in Proc. IEEE Industrial Electronics, Control, and in modeling and clutch control for automated man-
Instrumentation Conf., Maui, HI, 1993, vol. 1, pp. 430–434. ual transmissions during a collaboration with the ELASIS research center. His
[13] L. Glielmo, L. Iannelli, V. Vacca, and F. Vasca, “Speed control for auto- current research interests include automotive control, satellite formation flying,
mated manual transmission with dry clutch,” in Proc. 43th IEEE Conf. wireless monitoring systems, and real-time control in avionics. He is currently
Decision and Control, Atlantis, Bahamas, 2004, pp. 1709–1714. working on the ESA project (Fluid Science Laboratory).
[14] F. Amisano, R. Flora, and M. Velardocchia, “A linear thermal model
for an automotive clutch,” SAE Tech. Paper Series, no. 2000-01-0834, Francesco Vasca (S’94–M’95) was born in Naples,
2000. Italy, in 1967. He received the Laurea degree in elec-
[15] D. Centea, H. Rahnejat, and M. T. Menday, “The influence of the in- tronic engineering and the Ph.D. degree in automatic
terface coefficient of friction upon the propensity to judder in automo- control from the Università di Napoli Federico II,
tive clutches,” Proc. Inst. Mech. Eng., vol. 213, pt. D, pp. 245–268, Naples, Italy, in 1991 and 1995, respectively.
1999. Since 2000, he has been an Associate Professor
[16] E. M. A. Rabeih and D. A. Crolla, “Intelligent control of clutch judder and of Automatic Control in the Dipartimento di Ingeg-
shunt phenomena in vehicle drivelines,” Int. J. Veh. Des., vol. 17, no. 3, neria, Università degli Studi del Sannio, Benevento,
pp. 318–332, 1996. Italy. His research interests include automated man-
[17] M. Montanari, F. Ronchi, C. Rossi, A. Tilli, and A. Tonielli, “Control and ual transmissions, energy management of hybrid ve-
performance evaluation of a clutch servo system with hydraulic actuation,” hicles, battery state of charge estimation, nonlinear
Control Eng. Practice, vol. 12, pp. 1369–1379, 2004. dynamics and control techniques for power converters, averaging of nonsmooth
[18] H. Meritt, Hydraulic Control Systems. New York: Wiley, 1967. systems through dithering, and formation control of multiagent systems.

View publication stats