Mechatronics 21 (2011) 285297

Contents lists available at ScienceDirect

Mechatronics
journal homepage: www.elsevier.com/locate/mechatronics

Transmission control for power-shift agricultural tractors: Design

and end-of-line automatic tuning
Mara Tanelli a,, Giulio Panzani a, Sergio M. Savaresi a, Carlo Pirola b
a
Dipartimento di Elettronica e Informazione, Politecnico di Milano, Piazza L. da Vinci, 32, 20133 Milano, Italy
b
SAME Deutz-Fahr Group, Viale F. Cassani, 15, 24047 Treviglio (Bergamo), Italy

a r t i c l e i n f o a b s t r a c t

Article history: This paper addresses the analysis and design of the transmission control system for a high-power
Received 24 May 2010 power-shift agricultural tractor. Specically, all the criticalities involved with the correct management
Accepted 14 November 2010 of both single clutch and double clutch gear shifts are investigated, and a control system capable of
Available online 8 December 2010
providing good shifting performance in all operating conditions is proposed. Further, to comply with
components tolerances and spreads in the production line, an automatic procedure for the end-of-line
Keywords: tuning of the transmission control system is proposed to objectively classify the quality of the gear shift
Power-shift transmission
and automatically optimize it. The suitability of the proposed approach is thoroughly tested on an instru-
Agricultural tractors
Automotive systems
mented vehicle.
End-of-line tuning 2010 Elsevier Ltd. All rights reserved.

1. Introduction and motivation
Agricultural vehicles have to cope with working conditions
which are more complex and demanding than those experienced
by other ground vehicles, [10]. In fact, agricultural vehicles are
essentially designed to work at low speed while providing large
traction forces. Moreover, their ease of moving on uneven soil
makes them suitable also for heavy trailers transportation. To en-
sure the maximum exibility of use at each speed and to exploit
the maximum engine power available in all working conditions,
nowadays agricultural vehicles are often equipped with a so-called
power-shift transmission. This kind of transmission has a large
number of gears available (typically from 9 to 30) and it allows
to perform a gearshift with no (or at least with a minimum) loss
of power from the engine to the driving wheels.
Usually, a power-shift transmission is characterized by the
presence of two or more (depending from the number of gears
and the overall mechanical architecture of the gearbox) wet
clutches connected to an hydraulic circuit, whose pressure can be
regulated by a proportional solenoid valve. Considering the large
number of gears available and the fact that to achieve an optimal
gear shift it is necessary to correctly manage several control vari-
ables, this kind of transmission needs to be properly controlled.
The design of such a control system is not a trivial task. In the
scientic literature, some works dealing with power-shift or dual
Corresponding author. Tel.: +39 02 2399 3621; fax: +39 02 2399 3412.
E-mail address: tanelli@elet.polimi.it (M. Tanelli).
clutch transmissions control for ground vehicles are available,
see e.g., [38,15], but very little has been done on specic solutions
for agricultural tractors. This is mainly due to the fact that agricul-
tural vehicles have very specic performance specications due to
the very broad range of working conditions and variability of the
vehicle load, which make the gear shift optimal performance de-
nition different from that of ground vehicles. As a matter of fact,
the main constraints are the repeatability of the manoeuvre and
the comfort of the driver on all working grounds, which vary from
asphalt roads to rough off-road terrains. Also the load distribution
in tractors is much different than for other vehicles, due to the fact
that it might be due to either front or rear additional loads due to
the various working instruments that need to be employed for
differen tasks. Finally, note also that the variation of the operating
conditions is most often non measurable via on-board sensors, and
thus asks for robust and easily tunable gear shift controllers. These
facts make the problem of ensuring an optimal and repeatable gear
shift on an agricultural tractor a very challenging task.
To design an effective transmission control system, rst of all
the most signicant variables which inuence the gear shift quality
must be identied, see e.g., [2,16]. Further, the gear shift control
system has to optimally manage the trade-off among the following
conicting requirements:
(i) yield comfortable gear shifts;
(ii) guarantee that no loss of power to the driving wheels occurs
during gear shifts;
(iii) cause a minimum wear and tear of mechanical components
over the life of the vehicle transmission.

0957-4158/$ - see front matter 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mechatronics.2010.11.006
286 M. Tanelli et al. / Mechatronics 21 (2011) 285297
Moreover, in the industrial context, once the control design
phase is accomplished and the control system is implemented into
nal products, an end-of-line tuning phase is usually scheduled to
deal with constructive tolerances and production spreads which
cause the nal system to be different from the prototype one used
for control validation and testing. Hence, this phase is tailored to
optimize the controller parameters so as to guarantee that the
expected gear shifting performance is achieved on all vehicles.
Usually, this phase is carried out by human testers, who tune the
controller parameters based on personal driving preferences and
experience. Thus, is it clear that end-of-line tuning is a crucial
and difcult phase to deal with. As a matter of fact, since no objec-
tive indexes to evaluate the gear shift performance and comfort ex-
ist, a gear shift can be qualied as comfortable by one operator, but
not by another one: this means that the nal tuning can lead to
very different gear shift behaviors on different vehicles of the same
type. Note that, as the vehicle handling qualities, of which the gear
shift characteristics are a signicant component, is often consid-
ered as a trademark of the single manufacturer, the ability of deliv-
ering vehicles with identical manoeuvre features can be a key to
achieve customers satisfaction and to promote customers loyalty
to the brand. Moreover, another signicant advantage of the pro-
posed approach is that of reducing the industrial costs associated
with end-of-line tuning by reducing the number of gear shifts
needed to tune each vehicle and by making the process automatic,
thus not requiring highly experienced operators to perform it.
It is worth noting that the approach presented in this work,
even though tailored to a specic application, has a validity which
goes beyond the considered problem, as the aforementioned
design steps constitute a working paradigm which can be applied
in many different production contexts. As a matter of fact, this pa-
per is one of the rst contributions which aims at formalizing the
end-of-line tuning of industrial applications endowed with control
systems, proposing a systematic approach to the considered prob-
lem. In this respect, the results in [13,16] offer other applications of
the proposed methodology and address the problem of quantifying
of the driving style and safety via measured data, and of designing
and objectively tuning the motion inversion control of an agricul-
tural tractor, respectively.
Although being different problems with respect to that consid-
ered herein, both these works share (all or part of) the systematic
approach presented in this work, which is constituted by the fol-
lowing steps:
 an evaluation of the characteristic features which dene the
quality of the considered system;
 an experimental sensitivity analysis to single out the relation
between the features to be optimized and the measurable
variables;
 the denition of the cost functions;
 the design of the control algorithm and of the procedures for its
end-of-line tuning grounded on the cost functions optimization.
This methodology makes the results in the present paper of
general interest for all those applications in which a control system
must be designed and tuned while dealing with the dispersion
coming from production spreads and tolerances which make the
underlying plant (i.e., the nal vehicle) different from that used
for design purposes. The resulting research area requires tools both
of control theory and optimization, combined with the specic
application-domain knowledge.
The presented results are based on a joint research work
between the Politecnico di Milano and the R&D Department of
the SAME Deutz-Fahr Group (SAME, Lamborghini, Deutz-Fahr,
Hrlimann, Adim Diesel and Deutz AG). The work has been focused
Fig. 1. The tractor employed in this work.
on a power-shift transmission designed for high-power (200 HP)
agricultural tractors (see Fig. 1).
The rst effort has been devoted to dene appropriate cost
functions which allow an objective evaluation of gear shift comfort
and quality. Then, an accurate analysis of all relevant gear shift
dynamics has led to design a simple but effective transmission
control strategy. Further, to obtain the best possible gear shift per-
formance on every production vehicle, an automatic tuning phase
is proposed which guarantees satisfactory and repeatable gear shift
performance.
The structure of the paper is as follows. Section 2 provides a
description of the power-shift transmission system, both from an
hydraulic and a mechanical viewpoint. Section 3 is focused on pre-
senting the performance indexes which have been selected to
judge the gear shift quality. In Section 4, the proposed control
strategy for power-shift gear shifts is described, both for the case
of single clutch and double clutch gearshifts, together with the re-
sults of an experimental sensitivity analysis of the performance in-
dexes with respect to the controller parameters. Finally, Section 5
is devoted to describe the end-of-line self-tuning procedure and to
present related experimental results.
2. System description
The overall mechanical layout of the considered power-shift
transmission is depicted in Fig. 2. As can be seen, the power ows
from the engine (on the left of Fig. 2) towards the driving wheels
through two different gearboxes: the High-Mean-Low (HML)
group, composed of three different gears and the 123 group, which
also comprises three different gear ratios. The transmission is com-
pleted with two other components, namely the motion inverter
and the mode selector. The motion inverter (see [11,16]) is an elec-
tro-hydraulic system, constituted by two clutches, which allows to
perform an automatic motion inversion, i.e., it takes the vehicle
from a, say, forward speed to a reverse speed with the driver sim-
ply acting on a lever. The mode selector allows to choose among
three different working modes: creep, work and transport, which
can be varied only when the tractor is at standstill. In what follows,
we concentrate on the control of the gear shift, and consider the
two gearboxes only, assuming that no motion inversion is occur-
ring (note, in passing, that during a motion inversion the driver
cannot command a gear shift), and that a xed mode has been
engaged.
As the two gearboxes are in series, nine transmission ratios be-
tween the engine and the driving wheels are available (disregard-
ing the nal differential, whose ratio is xed). Conceptually,
although mechanically different, the two gearboxes can be treated
equally for control design purposes. Each gear is associated with a
 M. Tanelli et al. / Mechatronics 21 (2011) 285297 287

HML Motion 123 Mode

gearbox Inverter gearbox selector

Fig. 2. Schematic view of the power-shift transmission.

wet clutch: to select a particular gear the corresponding clutch

must be completely engaged, so that the torque coming from the
engine can be completely transferred via the clutch itself. The
wet clutches handled in this work are multi-plate wet: in order
to be engaged (and hence to select the associated gear) the surfaces
of the plates must be in close contact and the normal force they ex-
change must be large enough to develop a friction force which
guarantees that no relative slip occurs between them.
Fig. 3 shows a schematic view of the physical relationship be-
tween normal force and clutch oil pressure. For gear shift analysis,
three different zones must be considered. Starting from zero (i.e.,
from atmospheric pressure), an increase in the pressure brings
no changes in the normal force between the plates, which remains
negligible.
When the so-called kiss-point pressure is reached (see Fig. 3),
this distance that separates the plates has been covered and the
surfaces come into contact. From here over, the normal force in-
creases proportionally to the pressure. During this phase, the
friction force between the surfaces allows to transfer a certain
amount of incoming torque through the clutch but, as there is a
non-zero relative slip between the plates, the gear ratio is
indenite.
Once the engage pressure is reached (refer again to Fig. 3), the
normal force is large enough to annihilate any relative slip be-
tween the plates of the clutch and a precise gear ratio can be
dened.
Fig. 4 shows a schematic view of the hydraulic scheme, which
allows to understand how the pressure in a clutch can be con-
trolled. There are six ONOFF directional valves which connect
each clutch to the master hydraulic pressure, regulated by a pro-
portional solenoid valve. Such hydraulic architecture yields the fol-
lowing clutch pressure behavior: when a directional valve is
switched ON the pressure in the clutch is equal to the master pres-
sure. Being this a one-way valve, note that the pressure can only
increase, even if the master pressure decreases. Conversely, when
the directional valve is switched OFF the pressure in the clutch is
zero. As such, the available control variables are the following:
H M L 1 2 3
Fig. 4. Simplied hydraulic scheme of the considered transmission.
(1) the master hydraulic pressure. Note that, as no pressure sen-
sors are available, the real control variable is the current
driving the proportional valve. Such a variable can be linked
to the output pressure via a static map. In what follows, we
will regard the pressure as control variable, keeping in mind
that the aforementioned conversion from current to pres-
sure has to be performed;
(2) the onoff status of each directional valve.
To execute a gear shift with a power-shift transmission, the
outgoing clutch must be brought to zero pressure, whereas the
incoming clutch must be brought to maximum pressure. Note that
a non-power-shift gear shift would disengage the outgoing clutch
and then engage the incoming one. In so doing, there is a time
interval in which the vehicle is in a neutral state, and no engine
torque can reach the driving wheels. In agricultural vehicles the
neutral state must be avoided, as the large load forces would cause
the vehicle to stop. Thus, it is of utmost importance to ensure a
continuous torque transfer to the driving wheels during the gear
shift, which is the main characteristic of a power-shift gear shift.
To conclude the system description, Table 1 shows the nine
available gears together with the associated engaged clutches. As
can be seen, usually a gear shift requires to change only one clutch
(i.e., the one belonging to the HML gearbox). We refer to such gear
shifts as single clutch gear shifts. However, when the 34 and 67
gear shifts are considered, two clutches must be changed (one

Table 1
Normal force [N ] Available gears and respective engaged clutches.

Gear L M H 1 2 3
1  
2  
3  
4  
5  
6  
7  
Kiss-point Engage Pressure [bar] 8  
9  
Fig. 3. Oil pressure in the clutch as a function of the normal force.
288 M. Tanelli et al. / Mechatronics 21 (2011) 285297

belonging to the HML and one to the 123 gearbox), making the de-
(a) 10
sign of the gear shift controller more complex, as will be shown
subsequently. We refer to such gear shifts as double clutch gear 9.5

Speed [km/h]
shifts. 9

8.5
3. Gear shift quality assessment
8
As discussed in Section 1, dening an objective gear shift quality 7.5
assessment yields the following advantages: 0 2 4 6 8 10
Time [s]
 it provides a unique and objective indication of gear shift per-
formance, helpful to compare different vehicles and/or different (b) 8.5
control algorithms. 8

Speed [km/h]
 it makes the end-of-line tuning phase easier and cheaper by 7.5
relying on the automatic optimization of appropriate perfor-
7
mance indexes.
6.5
The crucial issue to deal with in dening the most suitable cost 6
functions is that of determining meaningful relationships between
5.5
measured signals and gear shift comfort and quality. Several stud- 0 2 4 6 8 10
ies have been carried out in the Automotive context, showing good Time [s]
results in evaluating comfort via acceleration measurements, see
e.g., [9,14]. For the type of vehicle considered in this work, it is easy (c) 11
to understand that this kind of signal is not suitable, as soil irreg- Speed [km/h]
10.5
ularities cause measurement noise which shadows the actual gear
shift contributions to vehicle accelerations. Moreover, accelerome- 10
ters are not standard sensors to have on-board of agricultural trac- 9.5
tors. Thus, we concentrated on investigating the relationships
9
between gear shift quality and vehicle speed, whose measurement
is commonly available via wheel encoders. As discussed in [16], 8.5
0 2 4 6 8 10
this signal can be exploited to provide satisfactory comfort
evaluation. Time [s]
To understand the rationale behind the quality index design, Fig. 5. Different gear shifts and respective quality evaluation: (a) good; (b)
Fig. 5 shows the time histories of the vehicle speed in three differ- medium; (c) bad.
ent gear shifts, whose performance was judged by an expert driver:
the rst one (Fig. 5a) has been classied as good, whereas the last
two (Fig. 5b and c) as medium and bad, respectively. The speed v ref 1
v m treq ; 3
behavior in the three considered gear shifts is as follows: the speed
where treq is the time instant at which the gear shift is requested by
always starts from a constant value and increases (up-shifts have
the driver. The reference speed in the last part of the manoeuvre is
been performed in all cases) until it reaches a higher nal value,
also constant and computed as
which depends only on the nal gear ratio as the engine speed is
kept xed and constant during the gear shift. What really makes v ref xeng treq r sInc: ; 4
end
the gear shifts different is the smoothness with which the speed in-
creases. Note, in fact, that while in the good gear shift in Fig. 5a the where xeng(treq) is the engine speed at the beginning of the gear
speed increases with a smooth ramp, in the medium quality gear shift (recall that the engine speed is xed and constant during the
shift the speed increase is only piecewise linear (see dotted oval gear shift), r is the average wheel radius and sinc is the transmission
box in Fig. 5b) and shows a signicant initial undershoot. Finally, ratio of the incoming gear (also known when the gear shift is issued
the bad gear shift is characterized by a quite irregular speed behav- by the driver).
ior and large oscillations (see dotted oval box in Fig. 5c). The reference speed evolution in time between these limiting
Based on these considerations, the performance index has been speed levels, which denes v ref2 , is chosen as linear, yielding
dened as
v ref  v ref1
  v ref t v ref1 end
t  t1 ; 5
J Var v m t  v ref t ; t2  t1
2
1

where vm(t) is the measured wheel speed and vref(t) is a reference where the time instant t1 is dened as
signal to be designed, which describes the speed behavior in an  
t1 : jv m t 1  v ref1 j P 0 ^ fjv m t  v ref1 j < 0 8t > t 1 g: 6
optimal gear shift. The measured vehicle speed vm(t) is computed as
Namely, t1 is the last time instant at which the measured speed is
1X 4
lower than the initial reference speed v ref1 , while the time instant
v m t xi tri ; 2
4 i1 t2 is dened as
 
where xi(t), i = 1, . . . , 4 are the wheel rotational speeds measured t2 : jv m t 1  v refend j < 0; 8t < t2 ^ fjv m t2  v refend j
via the wheel encoders and ri, i = 1, . . . , 4 are the wheel radii. P 0 8t P t 2 g 7
The reference signal vref(t) has been designed as composed of
three different parts (see Fig. 6). The rst is dened by the constant thus representing the rst time instant at which the measured
speed value at the beginning of the manoeuvre, i.e., speed reaches the nal reference speed value v refend .
 M. Tanelli et al. / Mechatronics 21 (2011) 285297 289

8.5

Gearshift request Gearshift end

8
vref 2 (t )
7.5

Speed [km/h]
v refend
Measured
7 Reference
v ref1

6.5

J1 J2 J3
5.5
0 t req 1 2 t1 t2 3 4 t3 5 6

Time [s]

Fig. 6. Reference signal (red) and measured speed (blue) for gearshift quality assessment.

Note that each of the three phases of the gear shift highlighted
in the reference signal captures a different issue related with the
manoeuvre quality and comfort. Specically, there are three main
problems to be managed by the control system (see also Fig. 6):
(1) the speed undershoot at the beginning of the gear shift;
(2) the oscillations in the acceleration phase;
(3) the overshoot and the settling phase at the end of the gear
shift.
Each of these issues is taken care of by means of an appropriate
tuning of different control parameters. To do this, it is crucial to be
able of capturing the three phenomena via objective performance
indexes which can be computed based on the measured signals.
To this end, three different cost functions J1, J2 and J3 are consid-
ered, one for each part of the manoeuvre, dened as
J 1 Varv m t  v ref1  8t 2 treq ; t1 
J 2 Varv m t  v ref2 t 8t 2 t1 ; t 2 ;
J 3 Varv m t  v refend  8t 2 t2 ; t 3 ;
where the time interval of xed duration over which J3 is computed
has been dened via a dedicated sensitivity analysis. Specically,
the time instant t3 to be used in (10) has been selected as
t 3 arg maxjJ 3good t 3  J 3bad t 3 j
t3
namely by maximizing the distance between the J3 index values ob-
tained for a good and a bad gear shift, whose classication was also
conrmed by the expert drivers rating. Actually, the drivers were
asked to judge the manoeuvre as a whole. Then, among those rated
as bad, those gear shifts with a value of J3 exceeding a pre-dened
threshold were selected to be employed for the tuning of t3 via (11).
Based on the obtained results a time window of 1.75 s has been
selected to evaluate J3. To conclude the discussion on gear shift
quality evaluation, Table 2 reports the Ji, i = 1, 2, 3 index values
for the three gear shifts shown in Fig. 2. As can be seen, the chosen
cost functions correctly capture the manoeuvre quality and have
been shown to match the expert drivers rating.
4. Gear shift controller design
This section presents the gear shift control system design. To
motivate the proposed approach, it is worth highlighting that the
transmission architecture presented in Section 2 poses some
intrinsic limitations which must be taken into account, see also
[1,11,12]. Specically:
(1) it is not possible to modulate the pressure of the outgoing
clutches, as when the ONOFF valve is in the OFF position
it directly connects the clutch to zero pressure. Moreover,
the exact time instant at which pressure loss starts (that is
the clutch opening time) is not known, even though the
8 pressure dynamics can be considered very fast;
(2) being the ONOFF valve one-way, the pressure in the clutch
9 can only increase;
(3) no measurements of the oil pressure in the clutches are
10 available. Thus, the pressure value is estimated via a static
relationship relating the proportional valve input current
to the clutch pressure. Considering that this map is unique
for all vehicles and that each clutch behavior is indeed inu-
enced by several yet not measurable factors (i.e., temper-
11 ature, wear of the surfaces, plates rotational speed) no
reliable estimate of the actual driving torque can be derived,
see also [11,12]. This fact does not allow to design a genuine
closed-loop pressure controller to manage gear shifts.
Considering the aforementioned limitations, an open loop con-
troller will be designed. Thus, the core of the control problem is in
dening and optimizing an appropriate reference pressure prole
to command the proportional valve which allows to obtain the de-
sired gear shift performance.
4.1. Single clutch gear shift control

Recalling the gears and clutches pairings shown in Table 1, only

Table 2
Quality indexes values for the gear shifts shown in Fig. 2. two clutches are involved in a single clutch gear shift: the one of
the outgoing and the one of the incoming gear. Hence, in this type
Gear shift J1 J2 J3
of gear shift, the control algorithm consists in dening the master
Fig. 5a 1.02 22.36 73.7 prole pressure and in appropriately selecting the time instant at
Fig. 5b 12.86 37.53 85.1
which the outgoing clutch has to be disengaged. In Fig. 7, the pres-
Fig. 5c 35.38 102.98 74.4
sure prole for a typical single clutch gear shift is shown.
290 M. Tanelli et al. / Mechatronics 21 (2011) 285297
Pressure
KP
pressure
Overlap
Gearshift
request
t DO Time
Fig. 7. Single clutch gearshift control: master pressure prole (solid line) and
outgoing clutch disengagement (dotted line).
After the drivers gear shift request, the master pressure and
thus the pressure in the incoming clutch, whose ONOFF valve is
ON is increased stepwise up to the kiss-point (KP) value (how to
automatically identify such a value will be discussed in Section 5).
This preliminary phase is needed to ensure that the incoming
clutch is ready to transfer torque (recall from Fig. 3 that for pres-
sure values below the kiss-point one no driving torque can be
transferred to the wheels). This lling phase lasts for a xed
amount of time, experimentally tuned to 500 ms. Such value has
proved sufcient to reach the desired KP pressure for all the
clutches of the transmission. Then, the core of the power-shift gear
shift takes place: during the Overlap time interval the incoming
clutch pressure, as well as the torque that it can transmit, in-
creases. Further, at time t = tDO the outgoing clutch is disengaged
(see the dotted line in Fig. 7). Finally, the master pressure is
increased to its maximum value in order to allow the full engage-
ment of the incoming clutch.
Specically, the pressure increases along a ramp up to the max-
imum value to which it can be modulated, and then it is taken to its
maximum nal value stepwise. The nal step in the pressure
shown in Fig. 7 is due to the characteristics of the hydraulic circuit
of the transmission. Specically, the pressure interval over which it
is possible to actually modulate the pressure ends at the pressure
value corresponding to the end of the ramp in Fig. 7, while the last
step is in fact actuated by a safety control of the hydraulic circuit
(a valve that mechanically closes when the pressure reaches a
pre-dened value) that takes the pressure to its maximum value,
designed so as to ensure that the clutch cannot disengage under
any (possibly abnormal) operating condition. As for the pressure
values in the different parts of the gearshift control, consider that
the whole pressure interval goes from 0 to 20 bar; the rst step
to the KP pressure is of approximately 4 bar (consider this value
as an average over the different gears), the modulation interval
(i.e., the ramp) goes up to approximately 18 bar. Thus, the last step
in which the pressure is increased but without allowing for being
modulated is of approximately 2 bar, which corresponds to the
10% of the overall pressure variation.
Note that during the Overlap time interval the outgoing clutch is
still engaged, thereby loading the engine with an additional torque.
Thus, while a short Overlap time interval could cause an insuf-
cient amount of torque to be transmitted by the incoming clutch,
which may cause an undesirable and abrupt vehicle halt, a too long
Overlap time interval would cause the tractor to decelerate due to
the additional load torque, and this would bring discomfort to the
driver. Moreover, an excessive wear of the clutch surfaces could
arise. Similar considerations can be made for the KP pressure level.
In other words, the tuning of the two parameter values is crucial to
obtain optimal gear shifts. To select the best parameter values for
controlling the single clutch gear shift, the J1 index in Eq. (8) is em-
ployed. In fact, deviations of the controller parameters from the
optimal values reect in signicant decelerations at the beginning
of the gear shift, which results in undershoots in the forward vehi-
cle speed (i.e., the feature captured by the index J1, see Eq. (8)).
The optimal tuning has been carried out based on experimental
sensitivity analysis. Fig. 8 shows the obtained results, i.e., the val-
ues of the J1 index as a function of the deviation DP of the lling
pressure from the KP value (top plot) and of the Overlap time inter-
val duration (bottom plot) for 12 and 23 gear shifts. The
trade-off in moving the lling pressure away from the KP pressure
value (i.e., of having an increased DP) is apparent: as such, an opti-
mal gear shift can be obtained only if an accurate estimate of the
kiss-point pressure is identied.
For the Overlap time interval duration (bottom plot) the
trade-off is less evident: this is due to the fact that the experiments
have been carried out in low load conditions. In this case, then, the
optimal Overlap time interval duration is close to zero.
In heavy load gear shifts, instead, selecting a correct value of
this parameter is crucial: for this reason a scheduled controller
could in principle be considered, which selects the most suitable
value for the Overlap time interval duration according to the cur-
rent engine power (whose estimated value is available from the
engine controller).
To assess the controller effectiveness, Fig. 9 shows the time his-
tories of the measured vehicle speed in a single clutch shift with
non-optimized and optimized controller parameters. The corre-
sponding values of the J1 index are J1 = 12.86 for the non-optimized
manoeuvre and J1 = 1.48 for the optimized one. The optimized val-
ues for the gear shift shown in Fig. 9 are KP = 4.8 bar and
Overlap = 100 ms.
4.2. Double clutch gear shift control
In double clutch gear shifts, two clutches for each gear are in-
volved: ideally, the gear shift control logic could be the same seen
for the single clutch shifts. Actually, however, the hydraulic circuit
is such that the HML gearbox has a different dynamic behavior
from the 123 one. Specically, the time needed to disengage the
clutches of the two gearboxes is signicantly different, even if
the command is issued for both gearboxes at the same time
instant.
The difference in the two gearboxes is due to the fact that the
pressure values in the two is different because of the differences
in the size of the clutches of the two gearboxes. Clutches of differ-
ent size are necessary due to the need of managing different torque
levels. Furthermore, the location of the two gearboxes with respect
to the main hydraulic accumulator makes the length of the pres-
sure circuit different, and this also affects the pressure dynamics
in the two gearboxes. The gearbox with higher pressure is the
123 one. As such, to correctly perform a double clutch gear shift
we need to rst disengage the 123 gearbox (the one with higher
pressure levels) and then the HML, so as to take into account the
differences between the two. Specically, the 123 needs a longer
time to compensate for the higher pressure gradient it needs to
cover before the outgoing clutch is disengaged. The DelayHML time
interval has been tuned so that the two gearboxes nd themselves
at approximately the same pressure level once the HML group is
disengaged.
Therefore an additional control parameter needs to be intro-
duced. The new control parameter is called DelayHML, and it repre-
sents (see Fig. 10) the time interval between the disengagement of
the 123 gearbox (tDO,123) and that of the HML one (tDO,HML).
As Fig. 10 shows, in order to take into account the delay intro-
duced by the hydraulic plant, the disengagement of the outgoing
HML group is commanded with a slight delay, equal to DelayHML.
Since a new control parameter has been introduced, a dedicated
sensitivity analysis has been carried out to optimally tune it. In this
case, both the J1 and the J2 indexes are considered, see Eqs. (8) and
 M. Tanelli et al. / Mechatronics 21 (2011) 285297 291

14

12 1-2
2-3
10

J index
8

1
4

0
-1000 -500 0 500 1000 1500
P kiss-point pressure [mbar]

3
1-2
2.5
2-3

2
J index

1.5
1

0.5

0
0 100 200 300 400
Overlap [ms]

Fig. 8. Sensitivity analysis on single shifts controller parameters.

8.5

8
Speed [km/h]

7.5

6.5

6 Not optimized
Optimized
5.5
0 1 2 3 4 5 6 7 8
Time [s]

Fig. 9. Time histories of the measured vehicle speed in a single clutch shift: results with non-optimized (dashed line) and with optimized (solid line) controller parameters.

Finally, it is worth noticing that the engagement phase of the

incoming clutches occurs while the pressure increases along the
nal ramp phase for t > tDO,HML but the two incoming clutches
may not engage simultaneously. With the proposed control
approach, the fact that the engagement phase occurs in a correct
way is evaluated by means of the cost functions, therefore without
a direct tuning of the engagement instant.
The obtained results are reported in Fig. 11, which shows the
values of J1 and J2, respectively, as functions of Overlap and Delay-
t t HML. For the sake of conciseness, the analysis for the KP pressure
value is not shown, as the obtained results are similar to those dis-
Fig. 10. Double clutch gearshift control: master pressure prole (solid line) and
outgoing clutches disengagement (dashed lines). cussed for the single clutch gear shift.
By inspecting Fig. 11, some remarks can be made.

(9). In fact, J1 correctly captures the speed undershoots at the
beginning of the manoeuvre , which are mainly functions of the
KP pressure and the Overlap time interval duration, whereas J2 ac-
counts for the oscillations in the acceleration phase, which heavily
depend on the value of DelayHML.
 First of all note that each controller parameter, i.e., Overlap and
DelayHML, has a predominant effect over one single index.
Namely, Overlap mostly affects the J1 index, whereas DelayHML
the J2 one. This fact allows to decouple the optimization phase,
and to consider two successive single variable optimization
292
Fig. 11. J1 and J2 index values as functions of the controller parameters.
problems, which can more easily managed in view of the auto-
matic end-of-line tuning phase. Specically, as will be described
in more detail in Section 5, the optimal value for Overlap will be
found by minimizing J1, while DelayHML will be tuned accord-
ing to J2. Specically, to assess the correctness of the sequential
minimization of the performance indexes, one has to observe
that the function J2(, DelayHML) has the same shape for all val-
ues of Overlap. Therefore, once a value of Overlap has been xed
by optimizing J1, then the optimization of J2 done by varying the
value of DelayHML will lead to a nal value for J2 which is
approximately always the same (note also that the cost function
always decreases as the value of DelayHML increases indepen-
dently of the value of Overlap). Of course, the nal value of J2
will not be rigorously the same irrespectively of the value of
Overlap with which it is evaluated and which is determined
by the minimization of J1, but the small differences in the nal
value for J2 are not practically relevant as they yield no signi-
cant changes in the nal gear shift performance.
 Note further that the results in Fig. 11b seem to suggest that a
large value of DelayHML would bring good quality gear shifts.
Nonetheless, it is worth pointing out that, as DelayHML
increases, the clutches are left slipping for an increasing amount
of time. Thus, this parameter should be kept at the lowest pos-
sible value so as to prevent an excessive wear of the clutches.
Section 5 will better discuss how to effectively deal with this
issue.
 Finally, it is worth comparing the results obtained in the single
and double clutch gear shift controllers as far as the value of
Overlap is concerned. Specically, at the end of the previous sec-
tion we pointed out that in the single clutch gear shift at least
in low load conditions a non-zero Overlap would induce only
the detrimental effect of yielding a longer manoeuvre, without
introducing any potential increment in the performance. In
the double clutch gear shifts, instead, a trade-off arises, due to
the fact that one needs to manage the two gearboxes differ-
11
Speed [km/h]
10
9
8
2 3
Fig. 12. Time histories of the measured vehicle speed in a double clutch shift: results with non-optimized (dashed line) and with optimized (solid line) controller parameters.
M. Tanelli et al. / Mechatronics 21 (2011) 285297
ently. Therefore, while on the one hand allowing a non-zero
Overlap time interval has indeed the detrimental effect of yield-
ing a longer (hence worse) gear shift, on the other hand it offers
the advantage of taking the pressure level up to a value which is
appropriate for the correct disengagement of the two gearboxes
that work at different pressure levels. Finally, note that, as men-
tioned in Section 4.1, if a single clutch gear shift performance
must be handled in the face of additional large loads, then a
non-zero Overlap might be of help to optimally deal with the
increased inertia of the vehicle, thereby leading to an adapta-
tion of the Overlap time duration as a function of the load con-
dition. Up to now, however, the tuning of this parameter has
been optimized for the end-of-line test conditions, which
involve gear shifts carried out at nominal load on asphalt road,
and thus led to set it to zero for the single clutch case.
To assess the controller effectiveness, Fig. 12 shows the time
histories of the measured vehicle speed in a double clutch shift
with non-optimized and optimized controller parameters.
The corresponding values of the J1 and J2 indexes are J1 = 24 and
J2 = 102.98 for the non-optimized manoeuvre and J1 = 1.31 and
J2 = 35.38 for the optimized one, respectively. The optimized values
for the gear shift shown in Fig. 12 are KP = 5.2 bar, Overlap = 390 ms
and DelayHML = 210 ms.
4.3. Analysis of the acceleration phase
The last issue to be considered for gear shift control, which is
shared both by single and double clutch gear shifts, is the possible
overshoot and oscillatory behavior present at the end of the
manoeuvre, when the clutch of the incoming gear needs to be fully
engaged and the vehicle must be brought to reach the nal, steady-
state, speed value.
Before discussing this issue and presenting the proposed solu-
tion to limit the overshoot, it is worth pointing out that in the
Non optimized
Optimized
4 5 6 7
Time [s]
 M. Tanelli et al. / Mechatronics 21 (2011) 285297 293

considered tractor the engine speed is regulated via a dedicated
engine control unit, the internal controller of which is not directly
accessible. The only interaction with the engine controller can be
realized via the specication of the engine speed reference signal
which should be tracked during the gear shift. The nominal choice
is to have a constant reference engine speed equal to the one
measured immediately before the gear shift request.
Within this setting, if we consider an up-shift, and we let xeng,
xoeng , xw and s be the engine speed, the engine reference speed, the
wheel speed and the transmission ratio of the incoming gear,
respectively, the evolution in time of these variables during the
gear shift can be schematically represented as in Fig. 13.
Specically, immediately before the gear shift the tractor pro-
ceeds at substantially constant engine (dotted line in Fig. 13) and
vehicle speed (the solid line in Fig. 13 is the scaled wheel speed
xw/s, which can be directly compared with the engine speed).
Once the gear shift is requested (in correspondence of the leftmost
solid vertical line in Fig. 13), the disengagement of all the clutches
associated with the incoming gear occurs, and there is a slipping
phase during which the system is characterized by two degrees
of freedom, as the engine and the tractor can proceed at different
speeds and they interact via the torque transmitted by the incom-
ing clutches, which are slipping.
Once the clutches are fully engaged (in correspondence of the
second solid vertical line in Fig. 13), the engine and the wheels
have the same speed (denoted by xE in Fig. 13), and the system
has only one degree of freedom. As can be seen from Fig. 13, during
the slipping phase the engine speed decreases even though the ref-
erence engine speed xoeng (horizontal solid line in Fig. 13) is kept
constant at the value at which the gear shift began. This is due to
the fact that the torque at the clutch in this phase is generated
along the direction of motion of the transmission output shaft, so
that it accelerates the vehicle while it opposes to the rotation
of the incoming shaft, thus decelerating the engine. During the
traction phase which follows the clutches engagement, the engine
controller, in view of the engine speed error, tries to compensate
for it by increasing the engine torque. Due to large vehicle inertia,
this acceleration phase has a long settling time, and the engine
controller dynamics (most probably endowed with an integral
action) is such that the transient is characterized by an overshoot
and subsequent oscillations, which cause discomfort to the driver.
To counteract this phenomenon, which is more critical in the
double clutch up-shifts, the idea is to appropriately modify the en-
gine reference speed (of course, an alternative may be to directly
act on the engine controller; in our case, this is not possible as
the engine control algorithm cannot be accessed or modied).
To this end, consider the modied engine speed reference
shown in Fig. 14: as can be seen in the slipping phase the engine
reference speed is decreased along a ramp up to the point at which
the incoming clutches are fully engaged, while, during the traction
phase, the reference is increased along a second ramp which takes
the engine speed back to the nal value which coincides with that
at the beginning of the gear shift.
The motivation for this choice is as follows: in the slipping
phase the engine is forced to decelerate in view of the torque gen-
eration mechanics. Hence, in order to limit the magnitude of the
tracking error and the resulting detrimental effect of the related
integral action of the controller, it is convenient to let the reference
speed decrease accordingly. This has also the additional benecial
effect of shortening the slipping phase, as the engine does not try

slipping phase traction phase

1700
1650
1600
Engine speed [rpm]

1550
1500
1450
oeng
1400 eng
1350 E
1300
w/
1250

3 3.5 4 4.5 5 5.5 6 6.5 7 7.5 8

Time [s]

Fig. 13. Schematic view of the time histories of engine and wheel speed in a gear shift with constant engine speed reference.

slipping phase traction phase

1700
1650
1600
Engine speed [rpm]

1550
1500
1450
eng
1400
oeng
1350 E
1300
w/
1250

3 3.5 4 4.5 5 5.5 6 6.5 7 7.5 8

Time [s]

Fig. 14. Schematic view of the time histories of engine speed, wheel speed in a gear shift with modied engine speed reference.
294 M. Tanelli et al. / Mechatronics 21 (2011) 285297

to accelerate opposing to the clutch engagement, thus limiting the
clutch wear and tear.
Then, once the clutch engagement is detected the engine must
be accelerated as quickly as possible while avoiding the overshoot.
To achieve this, a ramp set-point has proved appropriate, and an
appropriate (xed) value of the slope has been tuned for each gear
shift. The same tuning has been performed to dene the ramp dur-
ing the slipping phase, with the only difference that the point at
which the clutch is engaged is not known a priori, and thus needs
to be estimated. Specically, the clutch is dened to be engaged
when the engine speed xeng equals the scaled wheel speed xw/s.
To correctly compare the two speed values, however, one must
consider that, even during constant motion, these are not perfectly
equal due to the geometry of the transmission.
Based on a large set of gear shifts data, the residuals of the
equation:
xeng  xw =s 0
have been studied, so that it was possible to verify that they are
normally distributed and that the condence interval at 99% was
given by a spread of 30 rpm. Thus, the clutch is considered to be
fully engaged when the relation
jxeng  xw =sj 6 30 rpm 13
holds over a time window of Dt = 250 ms. This last condition is
needed to average out the measurement noise and avoid outliers.
To appreciate the effectiveness of the proposed approach Fig. 15
compares the time histories of the engine speed, of the engine tor-
que and of wheel speed measured in a double clutch gear shift with
constant (dashed line) and modied (solid line) engine speed refer-
ence. As can be seen, with constant engine speed reference the en-
gine controller makes the engine torque saturate to its maximum
value, and this causes of course a delay in the settling time and sig-
nicant oscillations at the end of he transient. This clearly reects
on the vehicle speed, the transient of which signicantly benets
from the engine reference modication, yielding a much more
12 comfortable gear shift. The positive effect can be objectively quan-
tied by means of the quality index J3 (see Eq. (10)): the gear shift
with constant engine speed reference yielded J3 = 720.4, while that
with the modied reference scored J3 = 73.4.

(a) 1600

1550
Engine speed [rpm]

1500

1450

1400

1350 Engine reference speed

1300 with modified engine reference speed

with constant engine reference speed

1250
0 5 10 15
Time [s]

(b)
Engine torque [% maximum torque]

100
with modified engine reference speed
90 with constant engine reference speed

80
70
60
50
40
30
20
10
0 5 10 15
Time [s]

(c) 30

29

28
Speed [km/h]

27

26

25

24 with modified engine reference speed

with constant engine reference speed
23
0 5 10 15
Time [s]

Fig. 15. Time histories of engine speed (a), engine torque (b) and wheel speed (c) in a double clutch gear shift: results obtained with constant (dashed line) and modied
(solid line) engine speed reference.
 M. Tanelli et al. / Mechatronics 21 (2011) 285297
15000
Pressure [mbar]
10000
5000
0
H L
Fig. 16. Kiss-point pressure identication results obtained on three different vehicles.
5. End-of-line automatic gear shift tuning
We now present the automatic end-of-line tuning algorithm de-
signed to optimize the gear shift controller parameter values which
is needed to ensure satisfactory and repeatable performance on
different production vehicles. The parameters that have to be cor-
rectly tuned are:
(1) the KP pressure for each clutch;
(2) the Overlap and DelayHML time intervals duration for double
clutch shifts.
Note that the prole chosen for modifying the engine reference
speed is not object of end-of-line tuning, as the fact that a closed-
loop controller is present to regulate the engine speed, combined
with the chosen reference signal adaptation strategy, yielded a ro-
bust solution not needing additional tuning.
As the kiss-point pressure has a clear physical meaning, a model
based procedure will be followed to identify its value.
For the double shift parameters, instead, an optimization
procedure is proposed, based on the quality indexes described in
Section 3.
5.1. Kiss-point pressure identication
We have previously discussed that, if a clutch is completely en-
gaged, the overlap of another clutch is seen by the engine as an
additional load torque. As the engine is itself controlled so as to
keep its speed constant (at a set-point imposed by the driver),
when this additional load torque acts on it more power is needed
to keep the engine speed constant. From this basic idea, the algo-
rithm to identify the kiss-point is as follows. While a clutch is en-
gaged, the clutch pressure whose kiss-point value is to be
identied is slowly increased. The engine power (available as
output by the engine control unit) is monitored: when its value in-
creases meaning that the overlapping clutch is actually transfer-
ring torque the corresponding pressure value is identied as the
kiss-point one.
As the sensitivity analysis showed (see Fig. 8), a precision of at
least 500 mbar is needed in the KP pressure identication to guar-
antee optimal gear shift performance.
Fig. 16 reports the KP pressure identication results obtained by
applying the above identication procedure for four times on three
different vehicles (each colored1 bar refers to a single vehicle and it
1
For interpretation of color in Figs. 118, the reader is referred to the web version
of this article.
295
Vehicle 1
Vehicle 2
Vehicle 3
M 1 2 3
Clutch
is divided into four smaller bars that represent different executions
of the automatic procedure proposed). The maximum standard
deviation of the identied KP values is of 90 mbar, which, com-
pared to the measured KP values, shows that the proposed algo-
rithm is strongly repeatable. It is also worth underlining that the
procedure can be carried out in a fully automatic way, as it does
not require the vehicle to move. In fact the tuning involves only
the HML and 123 clutches, and the procedure can be executed
while opening the nal clutch between the transmission and driv-
ing wheels, thus without transferring torque to the ground and
with the vehicle at standstill.
5.2. Automatic tuning of the double clutch gear shift controller
parameters
Unlike the kiss-point pressure, the two parameters involved in a
double clutch gear shift do not have a precise physical meaning.
Therefore, their optimal values will be found by experimental opti-
mization of the introduced cost functions. Namely, the Overlap
time interval will be tuned so as to minimize the cost function J1.
Notice, in fact, that for double clutch gear shifts this parameter
cannot be easily set to a xed value for all vehicles (as for single
clutch ones) as it changes signicantly from a vehicle to another.
As for DelayHML, recalling the discussion at the end of Section 4,
its tuning phase is performed by minimizing the following cost
function:
J HML J 2 ; DelayHML c1 J 2 c2 DelayHML 14
where c1 and c2 are constant scaling factors. The cost function JHML
takes into account both J2 and the value of DelayHML itself in order
to prevent excessive wear of the clutch surfaces. Specically, the
constant parameters c1 and c2 in (14) are selected so that the two
terms on the right-hand size of (14) have the same size. The neces-
sity of an equal weighting of the two components is motivated by
the following fact: as mentioned in Section 4.2 the optimal value
of J2 occurs for large values of DelayHML, which are not appropriate
in general, as they induce wear and tear of the HML clutches that
are left slipping for long times. Thus, one needs to weight DelayHML
so that its nal value is not too large. On the other hand, however,
DelayHML must not be too low either so not to lose the advantage of
different disengagement times for the two gearboxes involved in a
double clutch gear shift.
As all indexes can be computed in real time, the tuning proce-
dure is as follows:
1. dene a maximum number of gear shifts NMax and a lower-
bound J iMin on the performance index Ji
296 M. Tanelli et al. / Mechatronics 21 (2011) 285297

2. while N 6 NMax
i. perform a gear shift and compute the performance index Ji
ii. if J i 6 J iMin in the last three gear shifts go to Step 3
iii. else update the value of the parameter pk to be optimized via
a gradient-based optimization algorithm, i.e.,
pk1 pk  arJ i pk ;
where arJi(pk) is the step size which depends on the experimen-
tally computed gradient of the cost function (a is a positive real-val-
ued parameter used to scale the gradient);
3. save and store the current value of pk
Note that, to terminate the auto-tuning procedure, it is required
that the termination condition on the cost function value is met in
three successive manoeuvres, so as to account for possible varia-
tions in the working condition which can affect the single gear
shift. Furthermore, the tuning cycle is repeated while the number
of performed gear shifts is lower than the upper-bound NMax = 35.
This additional condition has been added to the tuning logic for
safety reasons, so as to ensure that the automatic procedure termi-
nates even in case of faults or unexpected system behavior. In all
the experimental campaign conducted up to now this limit has
never been hit and the procedure has always terminated in view
of condition 2.ii.
This procedure is applied for DelayHML and subsequently for
the Overlap time duration. The self-tuning procedure has been re-
peated four times on the same vehicle (whose controller parame-
ters have been manually de-tuned after each run) in order to
verify that the proposed algorithm allows to obtain repeatable gear
shift performances at the end of the tuning phase.
Fig. 17 reports the time histories of the two performance
indexes J1 and J2 and of the two parameters DelayHML and the
Overlap time duration obtained in a run of the automatic
procedure. As can be seen, the procedure starts with the tuning
of DelayHML, while Overlap is kept at zero. Then Overlap is
tuned, while keeping DelayHML at its optimized value. Note that
while Overlap is being tuned based on the cost function J1,
the value of the performance index J2 remains substantially con-
stant, therefore conrming the possibility of relying on a single
performance index at a time to perform the tuning of the two
parameters.
It is worth noting that the choice of tuning tune rst DelayHML
and then Overlap is due to physical considerations. In fact, if we
tuned rst Overlap with the same rationale used to tune DelayHML,
i.e., with DelayHML equal to zero, we would start the optimization
of J2 from large values of the cost function (see Fig. 11b), and this
corresponds to large bumps in the vehicle speed coming from a
non correct disengagement of the clutches, yielding very demand-
ing working conditions for the transmission. On the other hand, if
one would tune Overlap with large values of DelayHML, then the
clutches would be left slipping for long times, and this again would
signicantly stress the clutches and the whole transmission. As
such, the safest approach is to tune rst DelayHML and then
Overlap.
Finally, Fig. 18 shows the results obtained in the four different
tests, reporting the values of both J1 and J2 and the number of gear

200
J2

100

0
0 50 100 150 200
300
DelayHML

200
100
0
0 50 100 150 200

50
J1

0
0 50 100 150 200
600
Overlap

400
200
0
0 50 100 150 200
Time [s]

Fig. 17. Time histories of J2, DelayHML, J1 and Overlap in an automatic tuning procedure of a double clutch gear shift.

1st tuning
40
2nd tuning
3rd tuning
30 4th tuning

20

10

0
J1 J2 Number of gearshifts

Fig. 18. Results of the double clutch automatic tuning in four different tests.
 M. Tanelli et al. / Mechatronics 21 (2011) 285297
shifts needed to complete the procedure. As can be seen, the nal
gear shift performance is satisfactory and repeatable, thus allowing
one to regard the proposed tuning method as effective.
6. Concluding remarks and outlook
In this work the control system design for a power-shift agricul-
tural vehicle transmission has been presented, both for single and
double gear shifts. To do this, an objective classication of the per-
formed manoeuvre based on measured data has been performed,
which led to the denition of quality indexes employed to guide
the controller tuning.
Further, based on the output of the quality assessment phase
and on an experimental evaluation of the quality index sensitivity
with respect to the controller parameters, an end-of-line automatic
tuning phase has been developed in order to tailor the controller
parameters to each production vehicle. The end-of-line tuning
algorithm provides the needed adaptation of the controller param-
eter values which allows to achieve a pre-dened quality level as
the gear shift manoeuvre is repeated by the user. The proposed ap-
proach has been extensively tested on an instrumented tractor,
thereby conrming its suitability for industrial application.
Acknowledgments
This work has been partially supported by SAME Deutz-Fahr
Group S.p.A., and by MIUR project New methods for Identication
and Adaptive Control for Industrial Systems.
References
[1] Codec F, Tanelli M, Savaresi SM, Frignati D, Taroni F, Tascedda L. Management
of angular speed sensors for control at the power-zero condition of a
297
continuously variable transmission: a describing-function approach. Proc
Inst Mech Eng D J Automob Eng 2008;222(12):226982.
[2] Galvagno E, Morgando A, Sorniotti A, Velardocchia M, Amisano F. An objective
evaluation of the comfort during the gear change process. SAE Internationals;
2007 [No. 2007-01-1584].
[3] Genta G, Morello M. The automotive chassis. System design (mechanical
engineering series), vol. 2. Springer; 2009.
[4] Goetz M, Levesley MC, Crolla DA. Dynamics and control of gearshifts on
twin-clutch transmissions. Proc Inst Mech Eng D J Automob Eng 2005;19(8):
95163.
[5] Han W, Yi SJ. A study of shift control using the clutch pressure pattern in
automatic transmission. Proc Inst Mech Eng D J Automob Eng 2002;217(4):
28998.
[6] Jacobson B. Outline of a new control concept for power shifting of xed step
ratio automotive transmissions. Proc Inst Mech Eng D 2000;215(5):61324.
[7] Kiencke U, Nielsen L. Automotive control systems for engine, driveline, and
vehicle. Berlin: Springer-Verlag; 2000.
[8] Isermann T. Mechatronic systems fundamentals. London, UK: Springer-Verlag;
2003.
[9] ISO. ISO 2631-1:1997. Mechanical vibration and shock evaluation of human
exposure to whole-body vibration part 1: general requirements; 1997.
[10] Marchant JA. Mechatronics in agricultural engineering. Mechatronics
1991;1(1):118.
[11] Savaresi SM, Tanelli M, Taroni F, Previdi F, Bittanti S. Analysis and design of an
automatic motion inverter. IEEE/ASME Trans Mechatronics 2006;11(3):
34657.
[12] Savaresi SM, Taroni F, Previdi F, Bittanti S. Control system design on a power-
split CVT for high-power agricultural tractors. IEEE/ASME Trans Mechatronics
2004;9(3):56979.
[13] Savaresi SM, Manzoni V, Corti A, De Luca P. Estimation of the driver-style
economy and safety via inertial measurements. In: Proceedings of the 14th
international forum on advanced Microsystems for automotive applications.
AMAA 2010, Berlin, Germany, May 2010. p. 1219.
[14] Scarlett J, Prince JS, Stayner RM. Whole-body vibration: evaluation of emission
and exposure levels arising from agricultural tractors. J Terramech 2007;44(1):
6573.
[15] Schwab M. Electronically-controlled transmission systems current position
and future developments. Int Cong Transport Electron 1990:33542.
[16] Tanelli M, Savaresi SM, Manzoni V, Monizza F, Taroni F, Mangili A. Estimating
the maneuver quality of an automatic motion inverter for end-of-line tuning in
agricultural tractors. In: Proceedings of the 17th IFAC world congress, Seoul,
Korea; 2008. p. 1072631.