Sei sulla pagina 1di 7

Friction Compensation for Sensor-Less Force

Reection in Servo Manipulators


for High Radiation Areas
Surendra Singh Saini, Ushnish Sarkar, Debashish Datta Ray
Division of Remote Handling and Robotics
Bhabha Atomic Research Centre
Mumbai, India
{sainiss, ushinish, dray}@barc.gov.in
ABSTRACT
Master Slave Servo Manipulators are used to handle highly
radioactive and hazardous substances. These operations re-
quire a delicate and dexterous control e.g.in handling ra-
dioactive isotopes irradiated in tray rod facilities of high
ux research reactors. To achieve a dexterous man-in-the-
loop control, the interaction forces of slave arm with the
environment must be sensed and reected back to the mas-
ter arm. Due to very high radiation levels, normal Force-
Torque sensors cannot be used at the end eector of the
slave manipulator to sense these interaction forces. Hence
a sensor less force reection scheme is required. Sensor-less
force reection utilizing the joint motor current enhances the
eect of friction and results in a non-linear force reection
to the master arm causing more fatigue and poor dexterity.
This paper discusses a scheme for dexterous control using
friction compensated sensor-less force feedback in systems
with friction and backlash. The proposed scheme has been
implemented on a 1-dof master slave setup and the results,
showing the ecacy of the scheme, have been presented. A
novel method for sensorless estimation and control of joint
motor current (and hence torque) is also presented.
Categories and Subject Descriptors
I.2.9 [Robotics]: Manipulators
General Terms
Algorithms,Experimentation,Verication
Keywords
sensor-less force reection, master slave servo manipulator,
sensorless motor current measurement, friction compensa-
tion, man-in-the-loop control
Permission to make digital or hard copies of all or part of this work for
personal or classroomuse is granted without fee provided that copies are not
made or distributed for prot or commercial advantage and that copies bear
this notice and the full citation on the rst page. Copyrights for components
of this work owned by others than ACM must be honored. Abstracting with
credit is permitted. To copy otherwise, or republish, to post on servers or to
redistribute to lists, requires prior specic permission and/or a fee.
Request permissions from Permissions@acm.org.
AIR 13, July 04 - 06 2013, Pune, India
Copyright 2013 ACM 978-1-4503-2347-5/13/07 ...$15.00.
http://dx.doi.org/10.1145/2506095.2506138.
M
Master
Slave Joint Controller
(Position Servo)
Current Feedback
Position Demand
Current Demand
Position Feedback
Master Joint Controller
(Torque Servo)
M
Slave
Figure 1. Sensor-Less Force Reection Scheme Us-
ing reection of the Slave Motor Current to the Mas-
ter.
1. INTRODUCTION
Sensor less force reection, in principle, can be achieved
easily by measuring motor current of each joint of the po-
sition controlled slave manipulator and reecting the same
to the corresponding joint of the torque controlled master
arm (Fig. 1). However, in practice, due to friction, the mo-
tor current measurement contains an error representing the
frictional load on the motor along with the actual interac-
tion forces. This frictional component in the measured force
has a nonlinear prole. This results in nonlinear force re-
ection at the master. In case of a man-in-the-loop master-
slave control, this nonlinearity in reected force results in
degraded force transparency from the slave to the master,
thus making it unsuitable for delicate and dexterous con-
trol. The unwanted resistance forces reected to the master
arm due to friction also causes more fatigue to the operator
especially when large loads are being handled. Therefore
a friction compensated force reection is essential that will
reect only the true interaction forces of the slave arm with
the environment to the master arm and remove the nonlin-
ear frictional eects.
Simple single model based friction compensation can re-
duce the resistant force applied to the user but it may cause
oscillations [1] due to lags between the position input to
the friction compensator and real displacement of the fric-
tion contact surfaces causing over-compensation near zero
speed. Moreover, the simple model based approach is valid
only for systems without any backlash with xed load and
the model parameters have to be pre calibrated.
The presented scheme uses a dierent approach from a sin-
gle model based methodology; we calculate friction and true
load values using an on-line real time estimation of static
friction at zero velocity (for any dynamic load) and use this
value of the obtained load to dynamically update the general
kinetic friction model for all non-zero velocities.
2. SELECTION AND APPLICABILITY OF
FRICTION MODELS
Selection of a proper friction model is essential for the de-
velopment of a good compensation strategy. We shall briey
discuss some friction models for their applicability to our
Master Slave Servo Manipulator. But before that we shall
discuss some basic but important friction mechanisms.
2.1 Friction Mechanisms
A preliminary study of solid friction models has been dis-
cussed in [2]. At the microscopic scenario, protrusions in
the surface of contact (viz. asperities) deect in response to
the relative motion and act as shear stiness. The spring like
behavior of the deection of these asperities exert frictional
forces dynamically on the system. Also, the true contact
area depends on the closeness of these asperities, and is in
general dierent from the apparent area of contact. These
observations have enabled the justication of the physical
theory with the experimental results of friction [3], [4].
Some dierent friction mechanisms are - Static Friction,
Coulomb Friction [3], [5], Pre-sliding displacement, Ris-
ing Static Friction, Rate dependent breakaway [6], Stribeck
Eect [7], Viscous Friction [8], Stick Slip Limit Cycle, Fric-
tion Lag and Hysteresis [4], [6].
In our work, the Pre-sliding displacement and Rising Static
Friction mechanisms have been exploited and the Rate de-
pendent breakaway mechanism has been specially given due
care in designing a static friction compensation strategy.
These eects occur mostly at the discontinuity of the Static
and Kinetic Friction. Some other eects, though non-linear
in nature, occur in the kinetic friction regime and have one-
to-one relationship with velocity and hence can be compen-
sated using a model / lookup table based approach.
2.2 Static Models
In static models, the friction force depends statically on
the applied load and velocity. Static models consist of Coulomb
friction, static friction, viscous friction and Stribeck eect.
One such static model consisting of all these eects is Gen-
eral Kinetic Friction (GKF) Model [4] as represented by the
following equations.
F =

F() if = 0
F
e
if = 0 and |F
e
| < F
s
F
s
sgn(F
e
) otherwise
(1)
F() = F
c
+ (F
s
F
c
) e
|/
s
|

s
(2)
Where, F is the frictional force, is the velocity of the
interacting surfaces, F
c
and F
s
are the coulomb and static
friction forces,
s
is a shaping factor, F
e
is the applied force
and
s
is the Stribeck velocity (note that it occurs nonlin-
early in equation (2)). Equations (1) and (2) suggest that
the static friction models do not capture the eects of pre-
sliding displacement, varying breakaway force, accurate stick
slip onset conditions, and frictional lag.
2.3 Dynamic Models
There are dierent dynamic models e.g. Reset Integrator
Model is a widely used dynamic model based on a compu-
tationally feasible variation of the bristle model [9]. In this
model, instead of snapping any bristle, the bond between
the bristles is held constant, by curtailing the increment of
the strain at point of breakaway. This model uses an extra
state to estimate the strain. LuGre Model [10] is also re-
lated to the bristle model. Here, friction is related with the
average deection force of elastic spring like bristles. When
a tangential force is applied to the bristles, they will deect
like elastic springs. If the deection is suciently large, the
bristles will start slipping. The average bristle deection for
a steady motion is a function of velocity. The LuGre model
also incorporates the rate dependent friction phenomena e.g.
varying break-away force and frictional lag. A comprehen-
sive discussion on these dynamic models can be found in
[11] and [12].
2.4 Choice of a Model
For choosing a particular model for friction compensation
in force reection of our servo manipulator [13], we have to
discuss the applicability of the above mentioned static and
dynamic models to our system. The dynamic models are
valid for two surfaces that are always in contact i.e. there
should be no backlash in the system (otherwise there should
be some means to track the contact and breakup of the two
surfaces).In our system however, we have an unavoidable
backlash arising due to gears and hence we have to limit
ourselves to a static model. The static models on the other
hand cannot take care of the eects of friction at zero speed
e.g. the pre sliding displacement with backlash. The main
focus of our work is the development of an algorithmic model
that, on one hand, uses the General Kinetic Friction model
for steady nonzero speeds and on the other hand, an itera-
tive method to calculate true load and friction values at zero
speed. The separation of these two cases of zero and non-
zero velocities is important because the position sensors are
only on one side of the friction causing surfaces with back-
lash, i.e. only with the motor side (viz. resolvers). The
distinction between dierent regions of backlash is possible
either by having a relative position sensor between the two
contact surfaces or otherwise via an algorithmic estimate as
is explored in our case. It is important to note that, dur-
ing a monotonic motion (a non-zero speed), the two surfaces
(e.g. cogs of a gear) of the manipulator will eventually touch
each other in a stipulated small amount of time. To remove
the discontinuity caused by friction and backlash in load
measurement at zero speed, we measure the friction com-
pensated load by traversing through the zero speed discon-
tinuity and taking an average value (assuming the friction
model to be symmetrical in both directions, which we exper-
imentally found to be true to a fair approximation). Another
approach to tackle the friction discontinuity at zero speed
has also been attempted in [5] but, this also requires that
the surfaces should always be touching. The approach in [5]
is a simplication to remove the dead band caused due to
friction, but it does not consider the eects of backlash.
3. FRICTION COMPENSATION: THE
METHODOLOGY
Force reection to the master arm requires precise slave
load estimation. This can be achieved either directly by
using a force torque sensor or indirectly by measuring the
motor torques and compensating the eects of friction in the
measured torque. Motor torques can be measured from the
motor current. In this work, instead of using a current sen-
sor (e.g. shunt resistor), we have estimated and controlled
the motor current indirectly by establishing a relation be-
tween the motor current and the motor drives PWM duty
cycle. Also, the duty cycle has been analytically decom-
posed into two contributing factors viz. the actual load and
the lossy friction. Here, to estimate the motor current and
hence motor torque, we only need to measure the velocity of
the rotor and not the current through the armature wind-
ings. This scheme has facilitated the optimization of the
feedback path in our control loop. Though sensorless esti-
mation and control of motor torque has also been attempted
earlier, e.g. the disturbance observer based approach for in-
direct torque measurement as in [14]; however this method
relies on the observable disturbances in joint angles which
in our case will not change until we actually cross over the
friction discontinuity. The presented method on the other
hand is valid even for zero displacement of joint angles. The
aforementioned analysis and the actual friction compensa-
tion algorithms are systematically discussed in subsequent
sections.
3.1 Estimation of Slave Motor Current and its
Decomposition into its Load and Friction
components
Consider the voltage characteristic equation of the Slave
Permanent Magnet DC Motor:
I
slave
= (V
supply
V
bemf
)/R
ckt
(3)
Where, I
slave
is the slave joint motor current, V
supply
is
the source voltage of the PWM drive, is the instantaneous
duty cycle of the PWM controller, V
bemf
is the generated
back electro motive force (emf) that is equal to the speed
of the motor () multiplied by its back emf constant (K
e
)
and R
ckt
is the equivalent series resistance of the slave joint
motor. Now, the total torque developed by the slave joint
motor is equal to the sum of its load and frictional torques
as shown below:

motor
=
load
+
friction
(4)
Since motor current and torque are directly related by the
torque constant of the motor, we have:
I
motor
= I
load
+ I
friction
(5)
Therefore,
V
supply
V
bemf
R
ckt
=
V
supply

load
V
bemf
R
ckt
+
V
supply

friction
V
bemf
R
ckt
(6)
Thus,

load
=
V
supply
(
friction
) + V
bemf
V
supply
(7)
Also, for = 0,

load
=
friction
(8)
Therefore, we use equation (7) to calculate true load torque
of the slave joint motor. We also use the same equation to
control the joint motor torque of the master joint motors.
3.2 Friction Compensation Technique
The interaction forces of the slave arm of the manipula-
tor can be inferred by measuring the joint motor currents.
As stated earlier, we will be using dierent approach for
zero and non-zero slave joint speeds for compensating fric-
tion. It is important to note that our zero speed friction
compensation algorithm works only when the slave joint po-
sition controller has reached its set point and has settled.
For example, let us consider two cases where the slave arm
is exerting forces on a rigid wall and when the slave arm is
holding a weight in free space. In the rst case, if the master
tries to push against a rigid wall, the slave position controller
can never reach its set point and hence the zero speed fric-
tion compensation algorithm wont run. This may lead to a
higher force being reected to the master arm while he tries
to push against a wall. This eect however may be compen-
sated by considering that: when the position of the servo
controller is stuck at one value for an adequate time and
the position control error is still higher than the allowable
dead-band, we can use the non-zero velocity compensation
method and consider the direction of the reaction force due
to friction away from the set point and towards the manipu-
lators current position. In the latter case when the slave is
holding a weight at its end tip in free space, the proportional
integral controller of the slave joint will eventually settle at
its position set point. Once the position controller of the
slave joint settles, the zero-speed compensation algorithm
will be invoked.
3.2.1 Steady Non-Zero Speed Compensation
For steady nonzero speeds of a joint, friction has a bi-
jective mathematical relation as given by the General Ki-
netic Friction Model (for non-zero speeds) as is evident
from equations (1) and (2). The parameters of this model
can be found by regression analysis of the data plotted in
Fig. 2 which shows the speed of the joint motor (with
zero/balanced external loads) as a function of the current
drawn by the joint motor.
To do so, we give constant current to the motor, the motor
will then accelerate through a constant torque (the torque
0
200
400
600
800
1000
1200
1400
1600
1800
2000
0 100 200 300 400 500
S
p
e
e
d

(
r
p
m
)

Current (mA)
0g 10g 20g
30g 40g 50g
60g 70g 80g
90g 100g
Figure 2. Speed in Revolutions Per Minute vs. the
applied Constant Current to the joint motor at dif-
ferent Balanced Load values (g).
applied by the motor is equal to the current drawn by the
motor multiplied by its torque constant). The speed of the
joint will saturate when the reaction torque due to friction
exactly balances the torque applied by the motor. Once we
know the relation of friction (balanced-load current) with
speed of the slave joints, it is easy to compensate friction for
steady speeds by subtracting the corresponding balanced-
load current (calculated from the calibrated GKF model)
from the measured value of the current before reecting it
to the master arm. It may be noted that for this work, since
we have earlier established a dependence of the slave current
and its friction component with the drives duty cycle, the
experimental data as shown in Fig. 3 is actually used to do
the regression analysis and hence calibrate the GKF model
for non-zero speeds.
0
50
100
150
200
250
300
350
400
0 200 400 600 800 1000 1200

f
r
i c
t
i o
n

Speed (rpm)
0g 10g 20g 30g
40g 50g 60g 70g
80g 90g 100g
Figure 3. Component of the Duty cycle of the motor
drive corresponding to friction vs. Speed in Revolu-
tions Per Minute at dierent Balanced Load values
(g).
3.2.2 Zero Speed Compensation
When the slave manipulator is exerting force on the envi-
ronment but the slave arm is not moving and the position
controller of the slave joint has settled within its dead-band,
the zero-speed compensation algorithm will come into eect.
To elaborate the eect of a non-compensated force reection
at zero speed, consider a single joint test setup with a bal-
anced arm with vertical plane of rotation. Now keeping the
load at the end tip constant, we gave the slave arm step
inputs of motion from zero degrees to 90 degrees measured
from the vertically downward axis and repeated this opera-
tion several times. With iteration, as the position controller
settles the joint to 90 degrees and balances the weight of
the load, the steady state current drawn by the motor is
dierent for dierent trials (Fig. 4). This can be explained
by the fact that the dierence of the torque applied by the
joint motor and by the external load is being balanced by
the friction of the joint and the frictional component has a
discontinuity at zero speed.
The zero-speed compensation algorithm measures the present
value of the joint motor current and reects it to the master.
We then increase the amount of current given to the slave
joint motor in small steps and observe if there is any motion
(in the pre-sliding displacement regime). If there were no
displacement, we shall increase the slave joint motor cur-
rent again and will continue to do so unless we achieve some
non-zero speed. We will measure the value of current at
which the motion just occurred (breakaway) and will call
this value I
max
. We shall now call the slave position con-
troller again and wait for it to settle to the previous position
0
100
200
300
400
500
600
700
800
900
0 5 10 15 20 25
C
u
r
r
e
n
t

(
m
A
)

Event Number
Figure 4. Steady state current for dierent trials
for a step input of 90 degrees for a xed load of 100
grams.
set point. Once the position controller has again settled, we
will start decreasing the current given to the joint motor
and shall continue to do so in small steps until the motion
just starts. At the point where the motion just occurs in the
reverse direction, we will note the current value again and
call that as I
min
. The algorithm assumes that the eect of
friction is isotropic for movement of the slave joint in both
directions. Now we calculate I
Load
(Friction Compensated
Torque Load at zero speed) and F
Load
(The magnitude of
the Static Friction corresponding to the weight taken by the
joint axis) using the following equations.
I
Load
= (I
min
+ I
max
)/2 (9)
F
Load
=
|I
Load
I
min
| +|I
Load
I
max
|
2
(10)
The techniques discussed in subsections 1) and 2) are illus-
trated through the ow chart in Fig. 5
4. RESULTS AND DISCUSSIONS
The algorithms discussed above have been implemented
on a 1-dof master slave setup (Fig. 6). Results have been
shown for both zero and non zero speed cases.
4.1 Friction Compensation Results at Zero
Speed
The eectiveness of our algorithmic friction compensation
scheme at zero-speed was tested with various load values. A
particular case is when we attach a load to the slave arm
when it is hanging vertically downwards, i.e. the angle of the
arm () from the vertically downward axis is zero degrees.
In this case, the gravity load of any external weight will not
cause any torque load to the joint (since the force due to
gravity is in radial direction) and hence the reected torque
sensed by our compensated reector should be zero.
As the plot in Fig. 7 shows, we have been able to compen-
sate the eects of frictional discontinuity and reect the true
load. This would not have been the case if we just reect the
instantaneous motor current without compensation. With-
out friction compensation, the reected load can take any
value between the Non-compensated minima and maxima
values of the current for the corresponding weight. This is
because the Servo Position Controller may lock itself at any
value between these limits of current values. This behavior
can be justied by the fact that: as long as the magnitude
of the dierence, between the Load Torque and the Motor
Start
Load
Look-Up-Table
Set
Current FLoad = 0
Calculate Slave
V = d()/dt and
Run Slave PID
Is V = 0
For > last 20 PID
cycles
Calculate ILoad
using V and LUT
for Current FLoad
Read Slave
Joint Angle ()
Read Master
Joint Angle
Send ILoad
to Master
Increase I by one step
Read Slave
Joint Angle ()
Has Slave Joint
angle Increased
Store I as Imax and
Decrease I by one
step and Toggle
mode to Min
Read Slave
Joint Angle
()
Has Slave Joint
angle Decreased
Store I as Imin and
Increase I by one
step and Toggle
mode to Max
Calculate
Iload = (Imin+Imax)/2
Set Current FLoad =
(|Iload-Imin|+|Iload-Imax|)/2
No
No
Yes
is mode = Max
Yes
Yes
Decrease I by one step
No
No
Yes
Figure 5. Flow chart for friction compensation pro-
cess.
Torque, is less than the maximum value of static friction,
the reaction due to friction is exactly equal and opposite to
the said dierence. Therefore, in the aforementioned case,
the position of the joint (and hence its position sensors) will
not change until either the actual load or the motor current
changes by a magnitude sucient to overcome the static
friction. Hence, in a general case, the residual steady state
current in the slave motor, even without any external tan-
gential force on the slave arm is not zero, but can be any
value within the limits of the static friction.
At any angle other than zero, we shall have a net ex-
ternal tangential force i.e. an external torque on the slave
arm. Therefore, as we increase the weight on the slave arm
tip, both the static friction force (F
Load
), and the external
torque (I
Load
) on the slave arm increases. Without a good
compensation strategy, the amount of force reected to the
master arm could again be any value between the ranges
of static maximum friction for that load. In this work, the
boundary conditions for the joint angle () viz. 0 and 90
degrees have been taken to experimentally contrast the ef-
fect of uncompensated and compensated load and friction
values.
The results of the performance of compensation algorithm
for the trial using dierent loads are tabulated below in Ta-
ble I. Fig 8 shows the same information for the convenience
of visualization. Fig 9 shows the compensated Slave load
current value for dierent loads at = 90 degrees along with
Figure 6. The 1 - Degree of Freedom Master Slave
Test Setup.
-300
-200
-100
0
100
200
300
0 20 40 60 80 100 120
C
u
r
r
e
n
t

(
m
A
)

Load (g)
Compensated Current
Non-Compensated Minima
Non-Compensated Maxima
Figure 7. Comparison of the Range of Uncompen-
sated Current and the Compensated Current for dif-
ferent loads at Zero Speed at = 0 degree.
the maxima and minima values of motor current without
compensation. Fig. 10 shows the compensated and uncom-
pensated reected loads for dierent trials using a constant
load.
Table 1. Zero Speed Compensation at = 0 degree
Actual Load (g) Reected Load (g)
0 -0.36
10 8.90
20 19.09
30 27.27
40 41.63
50 51.27
60 59.27
70 69.27
80 81.63
90 90.54
100 100.54
It is evident from Fig. 7 and Fig. 9 that the static friction
value (F
Load
) depends only upon the total weight taken by
the joint axis and is almost independent of the joint angle
(). Therefore this value is used to update the GKF model
for the non-zero speed compensation. On the other hand,
the external torque i.e. the actual joint torque (I
Load
) of the
joint depends on the external load as well as on the joint
angle (), and hence, this value is reected to the master
arm.
The precise estimation of the external joint torques (in
contrast to the total weight experienced by the joint axis)
is essential since the overall interaction forces and moments
at the end eector are calculated by using the relation F =
(J
T
)
1
T. Where, F is the matrix depicting the forces and
moments of interaction at the manipulators end eector,
(J
T
)
1
is the inverse of the transpose of the Jacobian of the
slave joint of the manipulator and T is the matrix of all the
external joint torques of the slave manipulator [15].
-20
0
20
40
60
80
100
120
1 2 3 4 5 6 7 8 9 10 11
L
o
a
d

(
g
)

Trial No.
Actual Load
Reflected Load
Figure 8. Zero Speed Compensation at = 90 de-
grees.
-100
0
100
200
300
400
500
600
700
800
900
0 20 40 60 80 100 120
C
u
r
r
e
n
t
(
m
A
)

Load (g)
Compensated Current
Non-Compensated Minima
Non-Compensated Maxima
Figure 9. Comparison of Compensated Current and
Uncompensated Current range at Zero Speed at
= 90 degrees for dierent loads.
0
10
20
30
40
50
60
70
0 2 4 6 8 10 12
L
o
a
d

(
g
)

Trial No.
Actual Load (g)
Compensated Reflected Load (g)
Uncompensated Reflected Load (g)
Figure 10. Comparison of Compensated Current
and Uncompensated Current at zero speed at =
90 degrees for a constant load.
4.2 Compensation for Steady Nonzero Speeds
Keeping the load constant, for varying steady state speeds,
a comparative study of compensated and uncompensated
reected load current has been presented in Fig. 11.
5. CONCLUSION
Friction compensation makes the task of manipulation
more dexterous by reducing the unwanted nonlinear fric-
tional force from being reected to the master arm. The
0
10
20
30
40
50
60
70
0 200 400 600 800 1000 1200
L
o
a
d

(
g
)

Steady State Speed (rpm)
Actual Load (g)
Compensated Reflected Load (g)
Uncompensated Reflected Load (g)
Figure 11. Comparison of Compensated and Un-
compensated Current for a constant load at dierent
steady state speeds.
scheme used here is best suitable for the requirements of
a Man-in-the-loop type medium-capacity electrical master-
slave system designed for use in high radiation areas with
more allowable limits of backlash but lower friction limits
and may not be suitable to standard industrial robots that
have very low amount of backlash (but higher joint friction)
to achieve better repeatability. In this work, we have veried
both zero speed and constant speed friction compensated
force reection. For cases involving a reversal in the direc-
tion of motion of these high inertia systems, there exists a
nite span of zero velocity regime before attaining a veloc-
ity of the opposite sign. During this span of zero velocity,
the zero speed compensation takes place, compensating the
eects of friction and backlash. For cases involving acceler-
ated motion, the pseudo force due to a non-inertial frame
of reference will be added to the actual load. Although,
the same compensation methodology was used for the non-
uniform motion and is theoretically valid for such a domain,
the experimental verication remains as a future work.
6. REFERENCES
[1] Mahavash M., Dokamura A. M. 2006. Friction Com-
pensation for a Forcefeedbacktelerobotic System. In
Proceedings of the IEEE International Conference on
Robotics and Automation. (Orlando, FL, 2006), 3268-
3273. DOI = 10.1109/ROBOT.2006.1642200
[2] Dahl P. R. 1968. A Solid Friction Model. Technical
Report TOR-0158(3107-18)-1. The Aerospace Corpo-
ration, El Segundo, CA.
[3] Armstrong-Helouvry, Brian. 1991. Control of ma-
chines with friction. Springer ISBN: 0-7923-9133-0
USA. DOI = 10.1007/978-1-4615-3972-8 2.
[4] Armstrong-Helouvry, B., Dupont P., and Canudas de
Wit, C. 1994. A Survey of Models, Analysis Tools
and Compensation Methods for the Control of Ma-
chines with Friction. Automatica 30,7 (July 1994).
1083-1138. DOI = 10.1016/0005-1098(94)90209-7.
[5] Makkar, C., Dixon, W. E., Sawyer, W. G., and Hu,
G. 2005. A new continuously dierentiable fric-
tion model for control systems design. In Proceed-
ings of the IEEE/ASME Int. Conf. on Adv. Intelligent
Mechatronics. (Monterey, CA, 2005), 600-605. DOI =
10.1109/AIM.2005.1511048.
[6] Misovec, K. M., and Annaswamy, A. M. 1998. Fric-
tion Compensation Using Adaptive Nonlinear Con-
trol with Persistant Excitation. In Proceedings of
the Amer. Contr. Conf. 3 (1998), 1483-1487. DOI =
10.1109/ACC.1998.707074.
[7] Stribeck, R. 1902 Die wesentlichenEigenschaften
der Gleit - und Rollenlager (The basic properties of
sliding and rolling bearings). Zeitschrift des Vereins-
DeutscherIngenieure, 46 (1902), 1341-1348, 1432-1438
and 1463-1470.
[8] Garcia, E., Gonzalez, P., and De Wit, C. 2002. Ve-
locity dependence in the cyclic friction arising with
gears. Int. J. Robot. Res. 21, 9 (2002), 761-771.
[9] Haessig, Jr, D. A., and Friedland, B. 1991. On the
modeling and simulation of friction. Trans. ASME, J.
Dyn. Syst. Meas. Control , 113, 3 (Sept. 1991), 354-
3962.
[10] Astrom K. J., and Canudas-De-Wit, C. 2008. Revis-
iting the LuGre Friction Model. IEEE Control Systems
Magazine. 28, 6 (2008), 101-114.
[11] Bliman, P. A., and Sorine, M. 1995. Easy-to-use Re-
alistic Dry Friction Models for Automatic Control. In
Proceedings of the 3rd European. Control Conference,
ECC95 (Rome, Italy, 1995), 3788-3794.
[12] Canudas de Wit, C., Olsson, H. K., Astrom, J., and
P. Lischinsky. 1995. A New Model for Control of
Systems with Friction. IEEE Transactions Automatic
Control. 40, 3 (Mar. 1995), 419-425.
[13] Ray, DD, and Singh, M. 2010. Development of
a Force Reecting Tele-robot for Remote Handling
in Nuclear Installations. In Proceedings of the 1st
InternationalConference on Applied Robotics for the
Power Industry (Montreal, Canada, 2010), 1-6. DOI
= 10.1109/CARPI.2010.5624456.
[14] Murakami, Toshiyuki, Yu, Fangming, and Ohnishi,
Kouhei. 1993. Torque Sensorless Control in
Multidegree-of-Freedom Manipulator. IEEE Transac-
tions on Industrial Electronics. 40, 2 (1993), 259-265.
[15] Craig, John J. 1989. Introduction to Robotics: Me-
chanics and Control. (2nd Edition), Addison-Wesley
Publishing Company ISBN:0-201-09528-9.

Potrebbero piacerti anche