Proceedings of the 2nd IFAC

Workshop on Fractional Differentiation and its Applications

Porto, Portugal, July 19-21, 2006

ANALYSIS OF FRACTIONAL - ORDER ROBOT AXIS DYNAMICS

J.M. Rosario1, D. Dumur2, J.A. Tenreiro Machado3

1
Laboratory of Automation and Robotics - Faculty of Mechanical Engineering
UNICAMP - University of Campinas, 13083-970 Campinas- SP, Brazil
rosario@fem.unicamp.br
2
Department of Automatic Control
SUPELEC, 91192 - Gif-sur-Yvette, France
didier.dumur@supelec.fr
3
Department of Electrical Engineering
Institute of Engineering, Polytechnic Institute of Porto, 4200-072, Porto, Portugal
jtm@isep.ipp.pt

Abstract: Robots are complex mechatronics systems where several electric drives are
employed to control the movement of articulated structures. In industrial environments
they must perform tasks with rapidity and accuracy in order to produce goods and
services with minimal production time. These procedures require the use of flexible
robots which can act in a large workspace, thus subjected to important parameters
variations and nonlinear dynamics effects. This paper investigates the fractional order
dynamics during the evolution of trajectories of three robotic joints, considering the
complete system dynamics. Copyright 2006 IFAC

Keywords: Mechatronics, Robotics, Simulation, Dynamics, Control, Fractional calculus.

1. INTRODUCTION A Virtual Robot Environment was implemented that

permits the analysis of the axis positions and
The number of robots working in industry has velocities, electrical currents in the motors, and
increased significantly due to their operation spatial displacement of this device. Therefore, the
capabilities in terms of flexibility, rapidity and study presented here can assist in the design of the
accuracy. The integration of robots and mechatronic control system architecture to be used.
devices into a Flexible Manufacturing Cell (FMC)
are essential to achieve this performance leading to a This paper presents a fractional system perspective in
reduced production time and small energy the study of the mechatronic device: a PRR robot
consumption (David, et al. 1998; Pimenta, et al. with 3 DOF, as well as the dynamical modelling and
2001). Therefore, resourceful control algorithms simulation of this system, particularly emphasizing
must be implemented, which can cope with the development and implementation of robotic joint
important parameters variations and vibration. position controllers.

The study of feedback fractional order systems has Bearing these ideas in mind, this paper is organized
been receiving considerable attention (Machado, as follows. Section 2 describes the mechatronic
2003, Lima et al. 2005) both due to the fact that system, including kinematics, dynamic and actuator
many physical systems are well characterized by modelling. Section 3 presents the axis control
fractional-order models, and with the success in the structure implemented under the RST form. Section 4
synthesis of fractional-order PID controllers that is dedicated to the results obtained within a virtual
have been applied in a variety of dynamical robotics environment. Finally, section 5 draws the
processes (Barbosa, et al. 2004). main conclusions and points out future work.
 2. MECHATRONIC SYSTEM where = (d 2 3 ) is the vector of joint positions

The mechatronic system is a FMC particular

(
and x = p x p y p z is the vector of the )
application based on the coordination and integration table positions.
of two industrial robots and a PRR robot (prismatic-
revolute-prismatic joints) developed for accurate Function f enables to calculate the movement of the
welding work purposes (Fig.1). This device can end-effector resulting from the movement of the
assist in tasks for which traditional manipulators joints. This function is nonlinear and so (1) has a
have difficulties to reach some parts of the piece. For nontrivial analytical solution. This relation may be
that, the table is synchronized with the manipulators expressed mathematically by a matrix connecting the
allowing them to carry out complex tasks without system of fixed coordinates in the base of the robot
adjustments. with a system of coordinates associated to the robot.
This matrix, the so called homogeneous transition
matrix, is obtained through the product of the
homogeneous transformations matrix A i,i 1 , linking
the system of coordinates of element i with the
system of the previous element i 1 , that is:

[n T
]
s T a T p T = Tn = A 0,1 A 1, 2 A 2,3 (2)

[ ]
where p = p x p y p z is the position vector and
[ ]
n = n x n y nz , s = sx s y sz , [ ] [
a = ax a y az ]
represent the orientation vector.

Fig. 1: Flexible Manufacturing Cell.

The modelling and simulation of the three degree-of-

freedom (PRR robot), leads to the design of a virtual
environment simulator adopting electric and
mechanical libraries blocks using SIMULINK.

The main elements of these robotics joints are

brushless DC motor drives, axes inertia, gears and
control blocks. Other elements of the manipulator
(including loads) are represented by three nonlinear
models, one for each motor drive.
Axis i di i ri
The control system itself consists, essentially, in a 1 0 d 0 0
cascade of control loops (for each axis). The inner
speed and torque control loops are part of the drive 2 0 0 2 a
model; only the position loop is explicitly modeled.
In fact, the position control of the manipulator can be 3 3 0 0 b

implemented through the control feedback of each

isolated joint (David, et al. 1998), requiring the Fig. 2: PRR robot - Denavit-Hartenberg parameters.
model of each joint.
The kinematics description of this robot is performed
The simulator also includes a path generation through the Denavit-Hartenberg procedure, after
module, providing the joints with the axis trajectories isolating the four parameters { i , ri , d i , i } (Fig.
as reference signals to the control parts. Finally, a 2). These coefficients enable representing the
graphical interface is available, showing results of different positions of the parts of this mechanical
joints movements obtained through typical
trajectories.
device. The position vector p x p y p z and the ( )
orientation angles ( ) for an object with
2.1 Kinematics mode ( )
dimensions L x L y L z placed in the centre of the
The geometrical model of a 3 DOF robot provides table are given by:
( )
the position p x p y p z and orientation (, , )
p x = L x c 2 c3 L y s1 + L z c 2 s 3
with respect to a coordinate system fixed at the
centre of the table, as a function of its generalized p y = L x s 2 c3 + L y c 2 + L z s 2 s 3 + d (3)
coordinates joints, that is: p z = L x s 2 + L z c 2 + (a + b )
x = f () (1)
 ( ) = rot (x , ) rot ( y , ) rot (z , ) where Tm (t ) is the motor torque, m (t ) the angular
c c c s s s c c s c + s s (4) position of the motor, i(t ) the motor current, L, R
= s c s s s + c c s s c c s respectively the inductance, resistance of the motor,
s c s c c J eq the inertia of axis load calculated on the motor
side, resulting in the block diagram of Fig. 3.
nZ ny
= ATAN 2 , = ATAN 2
c n x + s n y n x Position + 1 1 1
KT N
controller R+ L s Bm + Jeq s s
+
s a x c a y _ _
= ATAN 2
s s x + c s y KE
nx = c2 ny = s2 nz = 0
sx = s2 c3 sy = c2 c3 sz = s3
ax = s2 s3 ay = c2 s3 az = c3
Fig. 3: Block diagram of the joint axis.

ci = cos(i ) si = sin(i ) (likewise for , , ) 2.3 Dynamic model

where the function ATAN2 to mean that the arc of
tangent is reckoned according to an algorithm talking As previously mentioned, the control of each joint is
into account that the values of both sine and cosine considered independently. To take coupling effects
are known, and hence the quadrant of the angle may into account, and to solve the trajectory problem, the
be deduced. dynamic control involves the determination of the
inputs, so that the drive of each joint moves its links
The elaboration of references in angular coordinates, to the position values with the required speed. The
referring to the tasks defined in the Cartesian space, dynamic model of a robotic joint can be derived
is expressed mathematically by the numerical through the Euler-Lagrange formulation that
inversion of the kinematics model using the Jacobian expresses the generalized torque (David, et al 1998).
function, which is: The manipulator dynamic behavior is described by a
group of differential equations called dynamic
= J 1 (x ) (5) equations of motion. For an N DOF rigid
manipulator, the equations are:
2.2 Control structure including kinematics
( )
i (t ) = J i ( (t )) &&i (t ) +C i (t ),&(t ) + Qi ( (t ))
(6)
For many operations the operator defines the tasks,
i = 1,L,3
or reference trajectories of the controller, in relation
to a coordinate system that is fixed to the end- where i (t ) is the generalized torque vector, i (t )
effector of the robot (in the Cartesian space).
Nevertheless, the desired movements (expressed in the generalized frame vector (joints), J i (t ) the
angular coordinates) and the control laws are in i ( )
inertial matrix, C , & the non-linear forces (for
different coordinate systems, requiring the example centrifugal) matrix, Qi ( ) the gravity force
implementation of algorithms for the inversion of the
kinematics model and the generation of the reference matrix. Fig. 4 shows the complete dynamical model
trajectory in angular coordinates. In this way, its considering for analysis, 2 robotic axes only.
trajectory is defined through a set of angles
associated to the angular movement of each degree
of freedom of the robot. After interpolation, these
angles will act as reference signals for positioning
controllers located at each joint, that compare the
signals deriving from the position sensors of the
joints (Pimenta, et al 2001).

2.4 Actuator model Fig. 4: Transfer Function and Dynamical Model.

Each robotic joint commonly includes a DC motor, a J T 1 1 + BT 1 1 + Tr1 = Tm 1 J T 2 2 + BT 2 2 + T r 2 = T m 2
gear and an encoder. Considering the DC motor, the
with
three classical equations are the following:
J T 1 = A2 + K1 + A3c1 , BT 1 = A3 2 s 2
d i (t ) d m (t )
u (t ) = L + R i (t ) + K E 1
1 2

dt dt Tr 1 = ( A3 + A2 c 2 ) 2 + ( A2 2 s 2 ) + A4 c 1 + A5 c( 1 + 2 )
2 2
d 2 m (t ) d m (t ) JT2 = A3 , BT2 = 0
Tm (t ) = J eq + Bm (7) 2
2 dt 1 1
dt Tr 2 = ( A3 + A2 c 2 ) 1 + A2 1 s 2 + A5 c(1 + 2 )
2 2
Tm (t ) = K T i (t )
The controller makes the corrections taking into a ball screw transmission (only for axis 1) and
account the robots dynamic model developed above. incremental encoders (Table 1).
These corrections are transmitted to the manipulator
through the actuator described in the next subsection, Table 1: Motor Parameters.
including a gearbox characterized by their ratio,
inertia and stiffness and damping of input and output Inertia (kgm2) 0.71 10-3
shafts. The gearboxes output shafts are connected to Weight (kg) 8
the other parts of the robot structure, which results in
Mechanical time constant (ms) 1.94
the effective torque reflected to each joint. For each
of the three joints, the other links effects are globally Voltage constant (V/rad/s) 0.807
considered as a single load inducing to the joint a Torque constant (Nm/A) 1.33
torque composed of three terms (Eq. 6). Inductance (mH) 14.7

Resistance ( ) 1.44

3. AXIS CONTROL ARCHITECTURE

4.2 GPC tuning parameters
One advantage of the virtual environment is the
possibility to implement and test advanced axis The axis controllers are designed independently
control strategies, in particular Predictive Control, a following the mechanism developed resulting in
well known structure providing improved tracking three RSTs, considering the same axis motor but with
performances. This philosophy aiming at creating an different inertia on the motor side due to different
anticipative effect is using the explicit knowledge of geometrical features for each axis (Fig.6).
the trajectory in the future (Boucher, et al 1995).
Four tuning parameters are required: N1 the
3.1 RST form of the controller
minimum prediction horizon, N 2 the maximum
The minimization of the cost function results in the prediction horizon, N u the control horizon and
predictive controller derived in the RST form the control weighting factor. The parameters given in
according to Fig. 5 and implemented through a Table 2 have been chosen to provide good stability
difference equation: and robustness margins (Boucher, et al. 1995).

S (q 1 ) (q 1 )u (t ) = R(q 1 ) y (t ) + T (q ) w(t ) (11)

Table 2: GPC tuning parameters for each joint.
1
T (q) DAC Process ADC
S (q 1 )
w +
u
Axis N1 N2 Nu
- y 1 1 8 1 92
Polynomial RST
controller 2 1 8 1 107.3
R ( q 1 )
3 1 8 1 126

Fig. 5: GPC in a RST form.

The main feature of this RST controller is the non

causal form of the T polynomial, creating the
anticipative effect of this control law.

4. TESTS IN THE VIRTUAL ENVIRONMENT

Previous sections have described the whole virtual

environment related to the 3 DOF manipulator,
including motor drives, gear boxes, kinematic and
dynamic models, and predictive axis controllers,
designed with electric drives and SIMULINK Fig. 6: Simulink Diagram of joint control (1 axe).
libraries. This section will now present results
obtained within this environment. Simulations 4.3 Simulation scenario
described below consider 3D trajectories issued from
the path generation module. The scenario considers the rotational trajectory
around the z-axis followed by a rod of 100 mm
4.1 Actuator parameters length (L), located at the center of the table of the 3
axis robot, with a 30o inclination angle. To do that,
The system considered here, used for supervision and the second and third axes are operating with maximal
control, includes three DC motors, a 1:100 gear box, and 10% velocity respectively.
During the movement, small disturbances are added
at the axis 2 reference signal at the time interval
between 3s - 3.5s, with amplitude level, in order of
the 10% of maximal current.

The desired path is thus a revolution cone

considering the dynamical model of axes, as shown
in Fig. 7. The effects of disturbances on each joint
can be seen. The several signals are recorded with a
sampling frequency of fs = 10 KHz.

a) Joint 1.

Fig. 7: Spatial trajectory.

4.4 Simulation results

b) Joint 2.
The simulation results obtained using axis
controllers, considering a disturbance effects and the
error analysis between previous scenarios. The
fractional analysis is realized using the Fast Fourier
Transform (FFT), using the transfer function,
between the reference signal and motor current. For
analysis the difference of this signals is used,
considering without and with disturbances effects
applied at axis 2.

Fig. 8 shows the amplitude of the Fast Fourier

Transform (FFT) of the axis 2 motor current signal.
The spectrum was also approximated by trend lines
in a frequency range larger than a decade. Table 3
presents the main results obtained for 3 robotic joints
by fitting parameters using a power function type
c) Joint 3
|I2(f)| = a fb, a,b , with frequency range f1.
Fig. 8: Spectrum results for 3 axis motor current.
4.5 Simulation analysis
The results of Fig. 8 and table 3 shows part 1of | i(f) |
Tracking performances offered by GPC laws for each robotic joint with a trend line superimposed
presents very small tracking errors at the trajectories over the signal, with slope approximately b = 1.06,
evolution. Globally, the results shows that the that reveals, clearly, an integer order behavior.
anticipative effect of the GPC law can provide better Similarly, part 2 of the spectrum signal was also
performances, even if the controllers were designed approximated by a trend line in a frequency range
neglecting the coupling effect between each axis. In larger than a decade, with a slope approximately of
this direction, GPC is less sensitive to inertia b = 3.3 presenting, clearly, a fractional order
variations (appearing as every axis acts on the other behavior. Other important transfer functions involved
ones) than PID. This significant simulation shows the other robot axis coupling were studied, revealing also
robustness of GPC, so that the inertia variation can an integer behavior, but under disturbance and no
be considered as a disturbance performing on the disturbance.
system.
The 2 and 3 axis motor current for a limited GPC controller has shown to be an effective strategy
frequency range present also fractional order in many fields of applications, with good time-
behavior while the joint 1 the spectrum is not defined domain and frequency properties (small overshoot,
in a large frequency range, because he has dynamic improved tracking accuracy and disturbance rejection
effects occurred because the coupling between axis. ability, good stability and robustness margins), able
to cope with important parameters variations.
The position signals were studied, revealing also an
integer behavior, both under disturbance and no The study provides useful information that can assist
disturbance conditions. in the design of a control structure to be used in
eliminating or reduction the dynamical coupling
Table 3: Fitting Parameters of FFT|i(t)| = a fb. between joints. Further work will look at the
robustness of the GPC strategy against measurement
Joint 1 noise and parameters uncertainties, as well as real
Freq. (Khz) A B R2 validation of the developed control algorithms
0 < f1 < 0.15 2,269E-06 1,053 9,9400E-01 accomplished through an experimental
0.3 < f2 < 0.8 1,177E+04 -3,310E+00 9,99300E-01 implementation.

6. ACKNOWLEDGEMENT
Joint 2
Freq. (Khz) A B R2 The authors acknowledge support of the Brazilian
0 < f1 < 0.15 2,272E-06 1,061 9,9978E-01 Research Council - CAPES- Brazil, through a
0.3 < f2 < 0.8 1,167E+04 -3,310E+00 9,9972E-01 collaborative work between the ISEP, Portugal and
UNICAMP, Brazil and ARCUS cooperation program
with SUPELEC, France.
Joint 3
Freq. (Khz) A B R2 REFERENCES
0 < f1 < 0.15 2,271E-06 1,059 9,9998E-01
0.3 < f2 < 0.8 1,173E+04 -3,340E+00 9,9997E-01 Barbosa, R.S., Tenreiro Machado, J.A., Ferreira, I.M.
(2004). Tuning of PID Controllers Based on
Bodes Ideal Transfer Function, Nonlinear
5. CONCLUSIONS AND FURTHER WORKS Dynamics 38, pp. 305-321, Kluwer Academic
Publishers.
In this paper a study was conducted to investigate
Boucher, P., Dumur, D. (1995), Predictive Motion
several robot signals, in a fractional system
Control, Journal of Systems Engineering,
perspective using a Virtual Robot Environment and
Special Issue on Motion Control Systems, 5:
the case study for a 3 DOF manipulator. For that
pp.148-162, Springer-Verlag.
purpose, a complete modular virtual environment
was designed using SIMULINK framework. This Clarke, D.W., Mohtadi, C., Tuffs, P.S. (1987),
simulator includes the kinematics and dynamic Generalized Predictive Control, Part I "The
models as well as the axis controlled loops built Basic Algorithm", Part II "Extensions and
around the actuator model, i.e. DC motor with gear Interpretation, Automatica, 23(2):137-160.
box and ball screw for the translational joint. A David, S., Rosrio, J.M. (1998), Modeling,
trajectory generation module and a graphical Simulation and Control of Flexible Robots,
interface were developed to help the user in testing
CONTROLO98, pp. 532-539, Coimbra,
realistic manipulator configurations. To emphasize
Portugal .
the modularity of this virtual environment, electrical
drive and controller libraries were integrated as well; Lima, M.F.M., Tenreiro Machado, J.A. (2005).
additional user-defined plug-in modules can be Experimental Set-Up for Vibration and Impact
added very easily. Analysis in Robotics, WSEAS Trans. on
Systems, Issue 5, vol. $, May, pp. 569-576.
In this direction, the axis controllers were structured Pimenta, K.B., Souza, J.P., Rosrio, J.M., Dumur, D.
under the RST formalism, which corresponds to the (2001), Control of Robotic Joints with
generic framework for numerical control. From this
Generalized Predictive Control (GPC),
form, a GPC control law was implemented on each
RADD2001, Vienna, Austria.
axis, simply designed without taking into account
coupling effects, providing improved performances Tenreiro Machado, J.A. (2003), A Probabilistic
in terms of rapidity, cancellation of overshoot, Interpretation of the Fractional-Order
accuracy, disturbance rejection and robustness Differentiation, Journal of Fractional Calculus
towards inertia variations and non linearities. This & Applied Analysis, vol. 6, No 1, pp. 73-80.
last point is one of the main challenges of robot
control, mainly when large workspace is involved,
because inertia can present important variations. For
comparison with a classical control axis strategy., the