280 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 15, NO.

2, APRIL 2010

Dynamic Modeling of a Two-Axis, Parallel,

H-Frame-Type XY Positioning System
Klaus S. Sollmann, Musa K. Jouaneh, Senior Member, IEEE, and David Lavender

Abstract—XY positioning is an important task in industrial ap-

plications. This paper addresses the dynamic modeling of a belt-
driven, parallel-type XY positioning system constructed in the form
of a capitalized H. The system uses one long timing belt to transmit
the rotation of two stationary motors to end-effector motion. Due
to less moved masses, the studied H-frame system is potentially
capable of fast acceleration, and therefore, faster positioning than
traditional stacked systems. The use of an elastic transmission ele-
ment also causes the biggest disadvantage of the system, which is
an uncertainty of end-effector position due to stretching in the belt.
Thus, the objective of this paper is to develop a dynamic model that
can capture the response of this system. Using Lagrange’s method,
an eighth-order lumped parameter dynamic model of the stage
motion is derived. The effect of nonlinear friction in the pulleys
and cart motion is added to the model. The response of the model
was simulated in MATLAB Simulink, and the model prediction is
compared with real data obtained from the developed system. The
results show that the model can accurately predict the dynamics of
the developed H-frame positioning system.
Index Terms—Belt drive, dynamic modeling, H-frame, XY posi-
tioning system.

I. INTRODUCTION Fig. 1. General layout of the H-frame positioning system.

HE XY positioning systems are widely used in industry to
T position a part or a tool within a two dimensional rectangu-
lar area. These systems are typically used for cutting, welding,
marking, or for pick-and-place applications. Current implemen-
tations of XY positioning systems typically consist of two linear
guides, each with their own independent timing belts or ball/lead
screws and motors [1], [2]. One guide is stacked on top of the
other, and oriented perpendicularly to the other. Although the
current configurations of XY positioning systems are widely
used, these systems tend to be bulky and they are not very suit-
able for low-profile applications.
High-speed systems are desirable in manufacturing because
they can increase productivity. To provide a lightweight, low-
cost, compact positioning system, a single belt-driven H-frame-
type XY positioning system can be used. Such a system consists
of two guides that are parallel to each other, and a third guide
perpendicular to the first two to form a shape resembling a
capitalized H (see Fig. 1). The most unique aspect of the device’s
Manuscript received October 3, 2008; revised February 1, 2009. First
published May 15, 2009; current version published March 31, 2010. Recom-
mended by Technical Editor M. E. Benbouzid.
K. S. Sollmann is with Volkswagen, 30179 Hannover, Germany (e-mail:
klaus_sollmann@yahoo.de).
M. K. Jouaneh is with the Department of Mechanical Engineering, University
of Rhode Island, Kingston, RI 02881 USA (e-mail: jouaneh@egr.uri.edu).
D. Lavender is with General Dynamics-Electric Boat, Groton, CT 06340
USA (e-mail: davelavender@verizon.net).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TMECH.2009.2020823
1083-4435/$26.00 © 2009 IEEE
configuration is that it only uses one belt and the motors are
stationary. The use of a single belt allows the device to be low
profile because all the guides are in the same plane.
A survey of the literature revealed no work describing a
similar parallel drive XY positioning system as the developed
H-frame system. Most XY positioning systems described were
either stacked ball screw systems or stacked belt drive systems.
The control of a stacked ball screw XY positioning system is
described in [3]. The paper considers that the ball screw has
a finite stiffness, and therefore, leads to a difference between
end-effector position and motor position. A torsional displace-
ment feedback control was proposed to improve the tracking in
this work. A positioning system consisting of two stacked belt
drive axis was studied in [4] and [5]. The use of elastic trans-
mission elements (the belts) lead to the described uncertainty in
end-effector position. A sliding mode control strategy is used
in both works to address this problem. The earlier papers em-
phasize the need for an accurate model of the system to exist in
order to successfully implement the control strategies. They also
identify elastic transmission as well as nonlinear friction as the
two major challenges in accurately controlling XY positioning
systems.
Further literature review dealing with nonlinear friction and
elastic transmission elements for only one-axis systems has
also been carried out. Methods proposed to control one-axis
belt drive systems found in the literature include adaptive
PID control [6], PID control plus acceleration and friction
SOLLMANN et al.: DYNAMIC MODELING OF A TWO-AXIS, PARALLEL, H-FRAME-TYPE XY POSITIONING SYSTEM 281

compensation [7], fuzzy logic [8], sliding mode control meth-

ods [9], [10], and feedforward compensator under maximum ac-
celeration and velocity constraints [11]. Adaptive control strate-
gies have been proposed in [12]–[16] to address the problems
caused by nonlinear friction in mechanical positioning systems.
Again, the majority of these control strategies required the de-
velopment of an accurate model of the system to be controlled,
which is the main emphasis in the present work.
The remainder of this paper is organized as follows. In
Section II, the design and kinematic characteristics of the de-
veloped H-frame positioning system are presented. Section III
discusses the development of a lumped-parameter model of the
positioning system. Section IV presents the response of the
model with and without nonlinear friction added to the model.
The effect of belt stiffness is discussed in Section V. The con-
cluding remarks are given in Section VI.

Fig. 2. H-frame positioning system.

II. DESIGN AND KINEMATIC MODELING
In the H-Frame positioning system shown in Fig. 1, there are
two parallel tracks along which a bridge is lead through on linear
ball bearing blocks. On the bridge there is a third track mounted,
which is perpendicular to the first two tracks, on which a cart
slides. These three tracks form a capitalized H. On each end of
the two parallel tracks sits one pulley, where the ones at the lower
end are directly attached to the motor shaft. On each end of the
track on the bridge, there are also two pulleys. An open timing
belt is guided around these eight pulleys including the motor
pulleys. The open ends are both attached to the cart, which runs
on the bridge. The system forms a parallel drive configuration,
meaning that the actuator drive system is not an open kinematic
chain. This parallel drive setup enables the rotational motion of
the two stationary motors to transform into a linear x-motion of
the bridge and a linear y-motion of the cart relative to the moving
bridge. The overlapping of these two linear motions creates the
XY-motion of the end-effector.
To relate the rotation of the two motors to the XY-motion of
the cart, consider Fig. 1 again. Turning only one motor while
keeping the other one still results in a linear motion of the end-
effector in a ±45◦ angle toward the XY-coordinate system. A
positive rotation of motor 1 while holding motor 2 still results in
a motion in negative x-direction and negative y-direction, while
a negative rotation of the same motor would cause a motion
in positive x-direction and positive y-direction. Mathematically,
this can be written as
r∆ϕ1 = −∆x − ∆y (1)
where ∆x, ∆y, and ∆ϕ1 are the change in x-, y-, and ϕ1 -
direction, respectively, and r is the radius of the motor pulley.
Note that the positive direction of rotation for each of the
motors is defined as the mathematical positive rotation around
the z-axis pointing outward from the paper plane, so the motor
torques shown in Fig. 1 are considered positive torques.
Similarly, a positive rotation of motor 2 while holding motor
1 still would cause a motion in positive x-direction and negative
y-direction, and a negative rotation of the same motor would
cause a linear motion of the end-effector in negative x-direction
and positive y-direction. Mathematically, this can be written as
r∆ϕ2 = ∆x − ∆y (2)
where ∆ϕ2 is the change in the ϕ2 -direction. Solving (1) and (2),
for ∆x and ∆y, respectively, we obtain, the following kinematic
relationship between the axes:
1 1 

 − r r

∆x  2 2  ∆ϕ1
=  . (3)
∆y 1 1 ∆ϕ2
− r − r
2 2
The previous equation considers the belt to be noncompliant.
A photograph of the developed system is shown in Fig. 2. The
footprint of the positioning system is 850 mm × 740 mm and the
motion range is 560 mm along the x-axis and 380 mm along the
y-axis. The motors used for the system are Pitman dc-brush-type
geared motors with optical encoders (Model GM9236C534-
R2). The motor gear ratio is 5.9:1 and the incremental encoders
have 512 lines per one revolution of the motor shaft. These
motors are interfaced through a pulsewidth modulation (PWM)
servo amplifier (Model A12 from Advanced Motion Controls)
to a 12-bit D/A converter. The timing belt used has an MXL
pitch and a width of 0.953 cm (0.375 in.). The length of the belt
is 3.95 m (155.5 in.).
There are also two laser-displacement-measuring devices
(Model optoNCDT 1700–100 from µ-Epsilon). One laser mea-
sures the position of the head along the x-axis and the other
measures the position of the head along the y-axis. These sen-
sors measure the actual position of the head which is more
accurate than using the encoder readings to estimate the head
position. The laser sensors have a measuring range of 100 mm
with a standoff distance of 70 mm. The analog output of the
laser sensors is read by the A/D converter.
282
Fig. 3. Generalized coordinates introduced to the H-frame system.
III. DYNAMIC MODELING
A schematic of the two-axis positioning system is shown in
Fig. 3. To obtain a dynamic model of this system, we need to
assign generalized coordinates. The belt sections between the
pulleys are assumed to be springs, so all the pulleys can to
some degree rotate freely from each other. Therefore, general-
ized angular coordinates ϕ1 to ϕ8 are assigned to the pulleys.
Additionally, the cart can move in two directions, which adds
two linear generalized coordinates x and y. This gives a total of
10 generalized coordinates that correspond to 10 DOFs for this
system. This means that a full-order lumped-parameter model
of this system will be of 20th -order. Such a model was derived
by Sollmann [17], but such a large model will require a sub-
stantial computational effort. To simplify the modeling, we will
lump some of the elements of this system together to obtain an
eighth-order dynamic model.
In order to reduce the order of the system, we could assume
that the inertia of all the pulleys, not driven by a motor, is lumped
into the two motor pulleys. So that
JM i = 4Jp , for i = 1, 2
whereJM i are the lumped inertias of the motor pulleys and Jp
is the inertia of a single pulley. However, before doing this, a
close look at the actual motion of the H-frame positioning sys-
tem needs to be taken. If the motions in x- and y-direction
are compared, this assumption is only valid for motion in
y-direction. This is because if the system is solely moved in
x-direction, the cart itself does not move relative to the bridge,
and therefore, the four pulleys attached to the bridge do not
rotate either. This, in turn, means that the rotational inertia of
these pulleys does not affect the overall dynamics of the system
for motion in x-direction. Hence, the assumption made in (4)
would lead to too much rotational inertia for this kind of motion.
Hence, for motion in x-direction, only the inertia of the corner
IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 15, NO. 2, APRIL 2010
pulleys opposite of the rotating motors should be lumped into
the motor pulleys. So, (4) changes to
JM i = 2Jp , for i = 1, 2. (5)
The modified inertia of (5) better approximates the dynamics
for x-direction motion, but not for y-direction motion. This is
because for motion in y-direction, the bridge pulleys do turn,
and therefore, their inertias have to be considered. As a solution
for this, the missing inertia for y-motion will be lumped into
the cart mass, which moves solely in y-direction. Note that the
actual mass of the cart and the mass of the four bridge pulleys
are included in the mass of the bridge Mbridge , but we will only
change Mcart , and Mbridge will not be changed. In order to
convert the rotational parameter Jp into a translation parameter,
a conversion factor of 1/r2 has to be multiplied, where r is the
pulley radius. So, the new mass parameter of the cart is
1
Mcart8 = Mcart20 + 4Jp (6)
r2
where Mcart20 is the parameter of the full 20th-order model and
Mcart8 is the parameter of the simplified 8th-order model.
In a similar fashion, the friction at the individual pulleys has
to be lumped partially into the motor pulleys and into the cart.
Equations (7) and (8) show the new friction coefficients for the
motor pulleys and the cart of the simplified eighth-order model.
BM i = 2Bp , for i = 1, 2 (7)
1
bcart8 = bcart20 + 4Bp (8)
r2
where, again, BM i and bcart8 are the viscous friction parame-
ters for the motor, and cart, respectively of the new simplified
8th-order model, while Bp and bcart20 are the viscous friction
parameters of a pulley and the cart, respectively, of the 20th-
order model.
Now, since all pulleys except the two motor pulleys are ideal
pulleys on which neither inertia forces nor friction or external
forces act, they do not have to be considered with generalized
coordinates, since they do not affect the dynamic behavior of the
system. Therefore, only four generalized coordinates have to be
introduced to the system, two of which are angular coordinates
ϕ1 and ϕ2 describing the motion of the two motor pulleys. The
(4) remaining two are the linear coordinates x and y describing the
motion of the end-effector, as shown in Fig. 3.
The belt stiffnesses kl (left), kr (right), and kb (bottom) can
be determined as the effective stiffness of the corresponding
springs of the belt segments. These are given by
k1 k5 k7
kl =
k1 k5 + k1 k7 + k7 k5
k2 k6 k8
kr =
k2 k6 + k2 k8 + k6 k8
k3 k4 k9
kb = . (9)
k3 k9 + k3 k4 + k4 k9
After these preliminary considerations, the kinetic and poten-
tial energy terms can be derived.
SOLLMANN et al.: DYNAMIC MODELING OF A TWO-AXIS, PARALLEL, H-FRAME-TYPE XY POSITIONING SYSTEM 283

1) Kinetic energy where the motor torques are described by

Cart (only in y-direction) Kt Kt Ke
τM i = Viin − ϕ̇i , for i = 1, 2. (17)
1 R R
Tcart = Mcart8 ẏ 2 . (10)
2 Substituting (15)–(17) into Lagrange’s equation [(18) later],
where n = 4 and q1 through q4 are x, y, ϕ1 , and ϕ2 , respec-
Bridge (including cart) tively, gives the set of equations of motion (19) that describe the
1 dynamic behavior of this simplified system.
Tbridge = Mbridge ẋ2 . (11)

2 d ∂L ∂L
− = Qnj c , for j = 1, 2, . . . , n (18)
dt ∂ q̇j ∂qj
Motor pulleys 1 and 2
1
1 ẍ = [−(kr + kl + 4kb )·x − bbridge ·ẋ + (kr − kl ) ·y
Tp1 = JM 1 ϕ̇21 Mbridge
2
1 − (kl r + 2kb r) ·ϕ1 + (kr r + 2kb r) ·ϕ2 ]
Tp2 = JM 2 ϕ̇22 . (12) 1
2 ÿ = [−(−kr + kl )·x − (kr + kl ) ·y − bcart8 ẏ
2) Potential energy: The potential energy of each belt section Mcart8
is obtained from V = 1/2k∆s2 , where k is the spring − kl r·ϕ1 − kr r·ϕ2 ]
stiffness of the belt section and ∆s is the extension or

1
contraction of the belt section. For the left belt section, ϕ̈1 = − (kl + 2kb ) r·x − kl r·y − (kb + kl ) r2 ·ϕ1
a rotation ϕ1 of motor 1 results in an extension of the JM 1


bottom end of this belt section by ϕ1 r. Also an x and y Kt Ke Kt
displacement of the cart results in an extension of the other − BM 1 + ·ϕ̇1 + kb r2 ·ϕ2 + V1in
R R
end of this belt section that is attached to the cart by x +

1
y. Thus, the net extension of the left belt is ϕ1 r + x + y. ϕ̈2 = (kr + 2kb )r·x − kr r·y + kb r2 ·ϕ1 − (kb + kr )r2 ·ϕ2
In a similar fashion, one can obtain an expression for the JM 2
extension for the right and bottom belt sections. Thus, the

Kt Ke Kt
potential energy of the left and right belt sections is written − BM 2 + ·ϕ̇2 + V2in . (19)
R R
as
The previous set of equations of motion can be transferred
1
Vl = kl (ϕ1 r + x + y)2 into state-space form, as shown in (20)–(24) next
2
1 x1 = x x2 = ẋ
Vr = kr (−ϕ2 r + x − y)2 . (13)
2 x3 = y x4 = ẏ
(20)
For the belt section in between the motor pulleys (kb ), we get x5 = ϕ1 x6 = ϕ̇1
1 x7 = ϕ2 x8 = ϕ̇2 .
Vb = kb (−ϕ1 r + ϕ2 r − 2x)2 . (14)
2 The matrices for this model are as follows.
From these terms, the Lagrangian L is assembled as System matrix
 
1 1 1 1 0 1 0 0 0 0 0 0
L = T − V = Mcart8 ẏ 2 + Mbridge ẋ2 + JM 1 ϕ̇21 + JM 2 ϕ̇22 a
2 2 2 2  21 a 22 a23 0 a25 0 a27 0 
 
1 1
− kl (ϕ1 r + x + y)2 − kr (−ϕ2 r + x − y)2  0 0 0 1 0 0 0 0 
 
2 2  a41 0 a a a 0 a 0 
A=  (21)
43 44 45 47
1  0 0 0 0 0 1 0 0 
− kb (−ϕ1 r + ϕ2 r − 2x)2 . (15)  
2  
 a61 0 a63 0 a65 a66 a67 0 
 
The next step is to consider the nonconservative forces Qnj c .  0 0 0 0 0 0 0 1 
The motor torques τM i have to be treated as nonconservative a81 0 a83 0 a85 0 a87 a88
forces, and also the friction forces of the cart and the bridge have
to be treated in the same way. As discussed earlier, the friction where
torques in the pulleys only occur in the motor pulleys and the kl + kr + 4kb
a21 = −
friction coefficients are the lumped frictions coefficients BM i Mbridge
as derived earlier. Therefore, the virtual work of the nonconser-
bbridge
vative forces acting on this system is given by a22 = −
Mbridge
δW n c = −ẏbcart8 δy − ẋbbridge δx + (τM 1 − BM 1 ϕ̇1 ) δϕ1
kl − kr
a23 = −
+ (τM 2 − BM 2 ϕ̇2 ) δϕ2 (16) Mbridge
284 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 15, NO. 2, APRIL 2010

kl + 2kb Output matrix

a25 = − r
Mbridge

0 0 0 0 1 0 0 0
kr + 2kb C= . (23)
a27 = r 0 0 0 0 0 0 1 0
Mbridge
kl − kr Feedthrough matrix
a41 = −
Mcart8

0 0
kl + kr D= . (24)
a43 =− 0 0
Mcart8
bcart8 For comparison purposes, we list next the equations for the
a44 = −
Mcart8 20th-order model from [17]
kl
a45 = − r 1
Mcart8 ẍ = [− (k3 + k4 + k5 + k6 ) x − bbridge ẋ
Mbridge
kr
a47 = − r − k3 rϕ1 + k4 rϕ2 − k5 rϕ3 + k6 rϕ4 − k3 rϕ6
Mcart8
kl + 2kb − k5 rϕ7 + k6 rϕ8 ] (25)
a61 = − r
JM 1 1
ÿ = [− (k7 + k8 ) y − bcart20 ẏ + k7 rϕ7 + k8 rϕ8 ]
kl Mcart20
a63 =− r
JM 1 (26)

kl + kb 2 1 Kt Ke
a65 = − r ϕ̈1 = − k3 rx − (k1 + k3 ) r2 ϕ1 − B1 + ϕ̇1
JM 1 J1 R


BM 1 Kt Ke Kt
a66 =− + + k1 r ϕ3 − k3 r ϕ5 +
2 2
V1in (27)
JM 1 RJM 1 R

kb 2 1 Kt Ke
a67 =
JM 1
r ϕ̈2 = − k4 rx − (k2 + k4 ) r ϕ2 − B2 +
2
ϕ̇2
J2 R
kr + 2kb 
a81 = r Kt
JM 2 + k2 r2 ϕ4 − k4 r2 ϕ6 + V2in (28)
R
1
kr ϕ̈3 = [−k5 rx + k1 r2 ϕ1 − (k1 + k5 )r2 ϕ3 − B3 ϕ̇3 − k5 r2 ϕ7 ]
a83 = − r J3
JM 2
(29)
kb 2
a85 = r 1
JM 2 ϕ̈4 = [k6 rx + k2 r2 ϕ2 − (k2 + k6 )r2 ϕ4 − B4 ϕ̇4 − k6 r2 ϕ8 ]
J4
kr + kb 2
a87 = − r (30)
JM 2

1
a88 =−
BM 2
+
Kt Ke
. ϕ̈5 = [−k3 rx − k3 r2 ϕ1 − (k3 + k9 )r2 ϕ5 − B5 ϕ̇5 + k9 r2 ϕ6 ]
JM 2 RJM 2 J5
(31)
1
Input matrix ϕ̈6 = [k4 rx − k4 r2 ϕ2 + k9 r2 ϕ5 − (k4 + k9 )r2 ϕ6 − B6 ϕ̇6 ]
J6
  (32)
0 0
 0  1
 0  ϕ̈7 = [−k5 rx + k7 ry − k5 r2 ϕ3 − (k5 + k7 )r2 ϕ7 − B7 ϕ̇7 ]
 0  J7
 0 
  (33)
 0 0 
 
  1
[k6 rx + k8 ry − k6 r2 ϕ4 − (k6 + k8 )r2 ϕ8 − B8 ϕ̇8 ].
B= 0 0 . (22) ϕ̈8 =
 Kt  J8
 0 
 RJ  (34)
 M1 
 0 
 0 
 Kt  Note that J1 = J2 = · · · = J8 = JP and B1 = B2 = · · · =
0 BP .
RJM 2
SOLLMANN et al.: DYNAMIC MODELING OF A TWO-AXIS, PARALLEL, H-FRAME-TYPE XY POSITIONING SYSTEM 285

TABLE I
PARAMETERS FOR THE H-FRAME EIGHTH-ORDER MODEL

IV. LINEAR AND NONLINEAR MODELING RESULTS

The parameters for this simplified model according to the ap-
plied assumptions are displayed in Table I. To verify the assump-
tions that were made, open-loop step response tests were per-
formed on both the real and the simulated eighth-order model.
These response plots were also compared to the response of
the 20th-order model that was developed in [17]. The response
plots for y-motion shown in Fig. 4 show that the simplified 8th-
order model has almost the exact same dynamic response as the
20th-order model and also matches the experimental data well.
Similar agreement between 8th-order model, 20th-order model,
and experimental data was also found for x-motion data [17].
In Fig. 4, there are some minor differences between exper-
imental and simulation data. In Fig. 4(a), a small (≈0.2 mm)
initial oscillation of the end-effector in x-direction can be seen,
which is not reflected in the model by the same amount. De-
viations like that could be caused by the assumption that the
belt is a massless linear spring. Since the belt in the real sys-
tem does have a mass, and therefore, its own dynamic behavior,
lateral vibration could occur that lead to these initial oscilla-
tions. It should be noted though that these initial oscillations
are comparably small effects, and therefore, the assumption to Fig. 4. Open-loop test results. (a) End-effector displacement. (b) Motor angles
displacement.
model the belt as massless linear springs remains valid. Also,
the end-effector does not return to a perfect x = 0 position after
this oscillation. This is probably an effect of nonlinear friction,
which lets the end-effector not return to the same equilibrium travelling in y-direction, the laser is not able to report the actual
position. position after it left the measuring range. A further effect, which
For the y-axis, there is no oscillation in the response, and the can be observed in the upper plot of Fig. 4(a), due to the rela-
simulation and experimental data have almost perfect match. tively small x-scale, is the random measuring noise. It seems that
However, the y-position of the end-effector of the experimental the end-effector oscillates with a very small amplitude around
data saturates at a certain value. This can be explained with its steady-state value; however, since the measurement signal
the measuring range (limited to 10 cm) of the laser measuring gets perturbed by noise, a statement to whether there is an oscil-
device used to collect the end-effector position data. lation with such a small amplitude or not cannot be made with
When the end-effector leaves the limited measuring range of certainty.
the measuring device, the laser keeps reporting the last valid The earlier results lead to the conclusion that the assumptions
measured displacement. Even though the end-effector keeps made in the previous section concerning the simplified lower
286
Fig. 5. Open-loop test results under −6 V/−6 V input. (a) End-effector dis-
placement. (b) Motor angles displacement.
order model are reasonable and the model response captures the
dynamic behavior of the H-frame positioning system.
The earlier eighth-order model is a strictly linear model, and
we would like to see how the experimental and simulation data
would match if the plots in Fig. 4 would be redone for different
input voltages with the same set of parameters shown in Table I.
To investigate this, open-loop simulations were carried out for
different input voltages. The input voltages that were used are:
(“motor 1 input”/“motor 2 input”) −2 V/2 V, −4 V/−4 V, and
−6 V/−6 V, and these voltages were selected to give different
x and y motions. Figs. 5–7 show the results. The linear model
matches the experimental data only for the amount of input
voltage for which it was tuned for (−4 V/−4 V and −4 V/4 V);
however, it deviates from the experimental responses for other
IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 15, NO. 2, APRIL 2010
Fig. 6. Open-loop test results under −4 V/−4 V input. (a) End-effector dis-
placement. (b) Motor angles displacement.
inputs. This is due to nonlinear friction that is present in the real
system. Hence, we need to modify the linear model developed
earlier to incorporate the friction nonlinearities that are present
in the system.
The first step to do is to rewrite the equations of motion into a
form in which the friction force/torque is expressed separately.
Equations (35) show the equations of motion in that form
1
ẋ2 = [(−x1 + x3 + rx7 ) kr + (−x1 − x3 − rx5 ) kl
Mbridge
+ (−4x1 − 2rx5 + 2rx7 ) kb − Ff x ]
1
ẋ4 = [(x1 − x3 − rx7 )kr + (−x1 − x3 − rx5 )kl − Ff y ]
Mcart8
SOLLMANN et al.: DYNAMIC MODELING OF A TWO-AXIS, PARALLEL, H-FRAME-TYPE XY POSITIONING SYSTEM 287

Fig. 8. Nonlinear simulation model.

Fig. 7. Open-loop test results under −2 V/2 V input. (a) End-effector dis-
placement. (b) Motor angles displacement. Fig. 9. ϕ 1 -axis nonlinear friction block diagram.

1
ẋ6 = [(−rx1 − rx3 − r2 x5 )kl + (−2rx1 − r2 x5 + r2 x7 )kb block diagrams and stored in subsystems. These subsystems
JM 1
are then assembled to form the nonlinear H-frame model, as
+ τM 1 − τf M 1 ] shown in Fig. 8. Two of these block diagrams represent the two
1 rotational motor axes, which have additional inputs Vin1 and
ẋ8 = [(rx1 − rx3 − r2 x7 )kr + (2rx1 + r2 x5 − r2 x7 )kb Vin2 to compute the motor torques. The remaining two block
JM 2
diagrams represent the linear axes. The block diagram for one of
+ τM 2 − τf M 2 ] (35) the rotational axes is shown in Fig. 9. The friction force/torque
where Ff is the friction force and Tf M is the friction torque. in each of these axes is computed with embedded MATLAB
The friction force and torque are given by functions with code similar to that shown in Table II.
Table III shows the set of friction parameters that were ob-
Ff = bv + sign(v)Fc (36) tained to generate the nonlinear simulation results shown in
τf M = BM ω + sign(ω)τcM . (37) Figs. 5–7. The viscous friction parameters differ from those
shown in Table I. For the linear model, the chosen viscous fric-
The first term in the previous two equations is the viscous tion parameter was only valid for a certain voltage input (certain
friction term, while the second term is the Coulomb friction velocity). So to assume that there was only viscous friction was
term. The four equations of motion can now be expressed as shown to be wrong. Instead, the friction force is a combination
288 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 15, NO. 2, APRIL 2010

TABLE II but the length of the belt section l is a function of the position
CODE FOR NONLINEAR FRICTION FORCE/TORQUE
of the end-effector in x and y. Thus, we can see that the belt
stiffness is a function of the XY position of the cart. In this
section, we study how belt stiffness varies with the XY position
of the cart.
The first step to do so is to quantify the dependency of every
single stiffness (k1 through k9 ) on x and y. The stiffnesses k1
through k9 depend on x- and y-position of the end-effector, as
described in the following equations. Here, A is the cross section
of the belt without the teeth, E is the elasticity (E) modulus of
the belt and l1 through l9 are the belt section length measured
TABLE III
SET OF FRICTION PARAMETERS FOR H-FRAME NONLINEAR MODEL
at the zero position of the end-effector located in the center of
the H-frame system.
Stiffness constant with respect to x and y
AE
ki = , for i = 1, 2, 9. (39)
li
The lengths of the belt sections in between the left motor
pulley and the left stationary corner pulley (k1 ), as well as the
one in between the right motor pulley and the right stationary
corner pulley (k2 ), and also the one along the bridge (k9 ), which
is not attached to the cart, do not change while performing
of Coulomb friction force and viscous friction force. In order to motion in x- and y-direction.
make the linear model fit for one input voltage, the viscous fric- Stiffness changing only with respect to x
tion parameter had to be chosen to be large so that the friction
force resulting out of it matches the sum of the Coulomb force AE
ki = , for i = 3, 4 (40)
and the actual viscous friction force. li − x
In Figs. 5–7, we see that the linear model matches the exper- AE
imental response only for 4 V input, for which the parameters ki = , for i = 5, 6. (41)
li + x
were verified. However, for 2 and 6 V input, the linear simula-
tion clearly deviates from the experimental results. In contrast, The belt sections in between the stationary corner pulleys
the simulation of the nonlinear model matches the experimental and the bridge (k5 and k6 ) get longer through a positive motion
data equally well for these two voltages. in x-direction. At the same time, the belt sections between the
Figs. 5(a) and 6(a) show similar differences in the amplitude motor pulleys and the bridge (k3 and k4 ) get shorter through the
of starting oscillations between simulation and experimental same motion.
response. This leads to the conclusion that these oscillations Stiffness changing only with respect to y
observed on the real system are caused by an effect other than AE
nonlinear friction not represented in the dynamic model. In k7 = (42)
l7 + y
Fig. 7(a), a slight motion in y-direction of the experimental data
AE
can be seen. These are effects not represented through the model; k8 = . (43)
however, they are relatively small (submillimeter) compared to l8 − y
the effects of nonlinear friction. Only dependent on a change in y are these two belt sections
This shows, as expected, that nonlinear friction is an impor- along the bridge that are connected to the cart (k7 and k8 ).
tant factor in the dynamic response of the H-frame system and The earlier stated equations can now be plugged into (9) in
need to be modeled to get an accurate dynamic response. Thus, order to obtain the stiffnesses kl , kr , and kb , which are used
the goal of developing a model that can be used to simulate a in the simplified eighth-order model of the H-frame system as
well-matched response for different sets of input voltages was follows:
achieved.
k1 k5 k7 AE
kl = =
k1 k5 + k1 k7 + k7 k5 (l1 + l5 + l7 ) + x + y
V. INFLUENCE OF END-EFFECTOR XY POSITION
ON BELT STIFFNESS k2 k6 k8 AE
kr = =
k2 k6 + k2 k8 + k6 k8 (l2 + l6 + l8 ) + x − y
All simulations, so far, have assumed that the stiffness values
of the belts are constant, which is only valid for limited range k3 k4 k9 AE
kb = = . (44)
motion. The stiffness of each belt section is given by k3 k9 + k3 k4 + k4 k9 (l3 + l4 + l9 ) − 2x

AE To show the change in stiffness due to the change in XY

k= (38) position, Fig. 10(a) and (b) shows the stiffness where one of the
l
SOLLMANN et al.: DYNAMIC MODELING OF A TWO-AXIS, PARALLEL, H-FRAME-TYPE XY POSITIONING SYSTEM 289

TABLE IV
BELT PARAMETERS

the results shown in Fig. 5. Similar investigation for x-motion

(−6 V, 6 V), where kb was doubled, and kr and kl were held
constant, show that the computed x-position only changed by
a similar amount. These results imply that one can neglect belt
stiffness variation as function of XY position for this system. For
reference, we have provided in Table IV the values needed to
compute the stiffness of the different belt sections. It should be
noted that the stiffness values shown in Table I are computed for
x = −0.2227 m and y = −0.1163 m, corresponding roughly to
the center of the rectangular motion zone studied in this paper.

VI. CONCLUSION
In this paper, an eighth-order lumped-parameter model for the
dynamics of a belt-driven, parallel-type, XY positioning system
constructed in the form of a capitalized H was derived. The
model incorporates nonlinear Coulumb friction in addition to
viscous friction effects. Furthermore, the stiffness of the belt
sections is shown to be a function of the XY position of the cart.
MATLAB simulation of the model response was compared with
the response of the real system, and the results show that the
model can accurately predict the response of the stage at least
within the limited range of the sensors that were used. The model
can be used in the design of closed-loop controllers to control
the motion of the system.
Fig. 10. Change of belt stiffness with respect to cart motion. (a) x-motion.
(b) y-motion.
REFERENCES
[1] K. Itoh, M. Iwasaki, and N. Matsui, “Robust fast and precise positioning
coordinates, x or y, is held constant and the other one is changed of ball screw-driven table system on machine stand,” in Proc. 8th IEEE
over the workspace of the H-frame system. Int. Workshop Adv. Motion Control, 2004, pp. 511–515.
The plots show that there is a notable change in stiffness [2] R. M. Dougans, “ABCs of x–y positioning,” Power Convers. Intell. Mo-
tion, vol. 12, no. 5, pp. 71–74, May 1986.
when the end-effector is moving through the workspace. While [3] H. Lim, J.-W. Seo, and C.-H. Choi, “Position control of xy table in cnc
moving in y-direction, kl and kr change by a factor of about machining center with non-rigid ballscrew,” in Proc. Amer. Control Conf.,
1.5 over the whole workspace. For a travel through the entire 2000, pp. 1542–1546.
[4] A. Hace, K. Jezernik, B. Curk, and M. Terbuc, “Robust motion control
workspace in x-direction, the change in kb is most significant. It of xy table for laser cutting machine,” in Proc. 24th IEEE Int. Conf. Ind.
is changed by a factor of about 3. To investigate the sensitivity Electron., Control, Instrum., Aug. 1998, pp. 1097–1102.
of the dynamic model to stiffness changes, we repeated the [5] A. Hace, K. Jezernik, and M. Terbuc, “VSS motion control for a laser-
cutting machine,” Control Eng. Practice, vol. 9, no. 1, pp. 67–77, Jan.
nonlinear simulation results shown in Fig. 5 (y-motion), but 2001.
with kl reduced by 25%, kr increased by 25%, and kb kept [6] W. Li and X. Cheng, “Adaptive high-precision control of positioning
the same from the values listed in Table I to correspond to the tables—Theory and experiments,” IEEE Trans. Control Syst. Technol.,
vol. 2, no. 3, pp. 265–270, Sep. 1994.
stiffness trend changes shown in Fig. 10(b) for y-displacement [7] W. Li and M. Rehani, “Modeling and control of a belt-drive positioning
of about 0.25 m. The results show that the computed y-position table,” in Proc. 22nd IEEE Int. Conf. Ind. Electron., Control, Instrum.,
at t = 0.25 s only decreased by only 0.1 mm compared with Taipei, Taiwan, Aug. 1996, pp. 1984–1989.
290 IEEE/ASME TRANSACTIONS ON MECHATRONICS, VOL. 15, NO. 2, APRIL 2010

[8] S. Kulkarni and M. A. El-Sharkawi, “Intelligent precision position control
of elastic drive systems,” IEEE Trans. Energy Convers., vol. 16, no. 1,
pp. 26–31, Mar. 2001.
[9] A. Hace, K. Jezernik, and A. Sabanovic, “Improved design of vss controller
for a linear belt-driven servomechanism,” IEEE/ASME Trans. Mechatron-
ics, vol. 10, no. 4, pp. 385–390, Aug. 2005.
[10] A. Hace, K. Jezernik, and A. Sabanovic, “SMC with disturbance observer
for a linear belt drive,” IEEE Trans. Ind. Electron., vol. 54, no. 6, pp. 3402–
3412, Dec. 2007.
[11] T. S. Jayawardene, M. Nakamura, and S. Goto, “Accurate control of belt
drives under acceleration and velocity constraints,” Int. J. Control, Autom.,
Syst., vol. 1, no. 4, pp. 474–483, 2003.
[12] P. Vedagarbha, D. M. Dawson, and M. Feemster, “Tracking control of
mechanical systems in the presence of nonlinear dynamic friction effects,”
in Proc. Amer. Control Conf., 1997, pp. 2284–2288.
[13] M. Feemster, M. Vedagarbha, D. M. Dawson, and D. Haste, “Adaptive con-
trol techniques for friction compensation,” Mechatronics, vol. 9, pp. 125–
145, 1999.
[14] Y. Hong and B. Yao, “A globally stable high-performance adaptive robust
control algorithm with input saturation for precision motion control of
linear motor drive systems,” IEEE/ASME Trans. Mechatronics, vol. 12,
no. 2, pp. 198–207, Apr. 2007.
[15] C.-I. Huang and L.-C. Fu, “Adaptive approach to motion controller of
linear induction motor with friction compensation,” IEEE/ASME Trans.
Mechatronics, vol. 12, no. 4, pp. 480–490, Aug. 2007.
[16] C. Du, L. Xie, J. Zhang, and G. Guo, “Disturbance rejection for a data
storage system via sensitivity loop shaping and adaptive nonlinear com-
pensation,” IEEE/ASME Trans. Mechatronics, vol. 13, no. 5, pp. 493–501,
Oct. 2008.
[17] K. S. Sollmann, “Modeling, simulation, and control of a belt driven,
parallel H-frame type two axes positioning system,” M.S. thesis, Univ.
Rhode Island, Kingston, RI, 2007.
Musa K. Jouaneh (S’88–M’89–SM’03) received the
B.S. degree in mechanical engineering from the Uni-
versity of Louisiana, Lafayette, in 1984, and the Mas-
ter’s and Doctorate degrees in mechanical engineer-
ing from the University of California, Berkeley, in
1986 and 1989, respectively.
Since 1990, he has been with the Department of
Mechanical Engineering, University of Rhode Island
(URI), Kingston, where he is currently a Professor
and Director of the Mechatronics Laboratory. His
current research interests include mechatronics and
robotics with particular interest in motion control systems. He is the author or
coauthor of over 60 publications including two U.S. patents. He has been a
consultant to many companies in the Northeast U.S.
Dr. Jouaneh has received two College of Engineering Faculty Excellence
Awards and the URI Foundation Teaching Excellence Award. He is a member
of the American Society of Mechanical Engineers (ASME).
David Lavender received the B.S. degree in
aerospace engineering from Virginia Polytechnic In-
stitute and State University, Blacksburg, in 2003, and
the M.S. degree in mechanical engineering from the
University of Rhode Island, Kingston, in 2007.
He is currently an Engineer with General
Dynamics-Electric Boat, Groton, CT.

Klaus S. Sollmann received the M.S. degree in me-

chanical engineering from the University of Rhode
Island, Kingston, in 2007, and the Diploma Engi-
neer degree from the Technical University of Braun-
schweig, Braunschweig, Germany, in 2008.
He is currently an Engineer with the Gradu-
ate Programme of the Volkswagen Group, Commer-
cial Vehicles Section, Hannover, Germany. His cur-
rent research interests include integrated production
systems, vehicle final assembly, production mainte-
nance, mechatronics, and manufacturing systems