Beruflich Dokumente
Kultur Dokumente
Series Editor:
G.M.L. GLADWELL
Solid Mechanics Division, Faculty 0/ Engineering
University a/Waterloo
Waterloo, Ontario, Canada N2L 3Gl
ISBN 978-94-010-4802-6
Edited by
H.S.TZOU
University of Kentucky,
U.SA.
and
T.FUKUDA
Nagoya University,
Japan
Preface IX
1. Introduction 1
2. Principle of Electrostatic Micro Actuator 3
3. Modeling and Dynamics of 3 DOF Electrostatic Micro Actuator 4
4. Control Method of 3 DOF Electrostatic Micro Actuator 12
5. Position Control Simulation of 3 DOF Electrostatic Micro Actuator 14
6. Structure of Prototype 3 DOF Electrostatic Micro Actuator 15
7. Experimental Results of Controlling 3 DOF Electrostatic Micro Actuator 16
8. Design and Structure of 6 DOF Electrostatic Micro Actuator 16
9. Control of 6 DOF Electrostatic Micro Actuator 18
10. Experimental Results of Controlling 6 DOF Electrostatic Micro Actuator 21
11. Optical Actuator 21
12. Conclusions 22
1. Introduction 39
2. Three Dimensional Micro-Mechanical Structures 42
3. Electrostatic Actuation 49
4. Fluid Bearings for Platform Levitation 50
5. Mobile Micro Robots in Fluids 53
6. Prototype for a Mobile Micro-Robot in a Liquid 57
7. Summary 57
1. Introduction 74
2. Description of the VR Spherical Motor 76
3. The Variable Reluctance Model 77
4. Magnetic Flux Prediction Using Finite Element (FE) Method 84
5. Results and Discussions 87
6. Conclusions 92
• Appendix: Overlapping Area Between Two Poles 96
VI
1. Introduction 112
2. Dynamic Model 114
3. Motion Control Strategy 120
4. Implementation of Inverse Model 127
5. Conclusions 132
1. Introduction 140
2. Thin Film Deposition, Processing, and Properties 144
3. Circuit Integration 147
4. Microsensor Applications 149
5. Microactuator Applications 156
6. Conclusions 159
1. Introduction 176
2. Distributed Neural Sensation 177
3. Distributed Muscle Actuation 181
4. Case-I: A Hemispherical Shell 184
5. Case-2: A Cylinder with Distributed Neurons and Muscles 188
6. Case-3: Distributed Convolving Ring Sensors 191
7. Case-4: Segmentation of Distributed Sensors/Actuators 202
8. Summary and Conclusions 213
Appendix: Mechanical Forces and Moments 217
1. Introduction 219
2. Basics Fiber Optics 221
3. Fiber Optics in Industry and Robotics 227
4. Optical Fiber Sensors for Robots 231
5. Fiber Optic Lighting and Imaging Systems in Machine Vision 273
6. Conclusions and Future Trends 296
vii
Design and Dexterous Control of Micromanipulator with Multiple Degrees of
Freedom
Toshio Fukuda, Motohiro Fujiyoshi and Fumihito Arai 299
1. Introduction 300
2. Dexterous Teleoperation Control 302
3. Design of Micromanipulator with Multiple Degrees of Freedom 312
4. Conclusions 326
1. Introduction 346
2. Kinematics 348
3. Prototype Micro-Motion Manipulator 353
4. Special Configuration 360
5. Conclusion 362
1. Introduction 375
2. A Miniature Mobile Robot Using Piezo Vibration for Mobility in a Thin
Tube 377
3. A Wireless Mobile Robot Driven by Piezo Vibration in a Thin Tube 403
4. A Miniature Mobile Robot for Use on a Wire 408
5. A Miniature Actuator for Plate Movement 415
6. Other Miniature Mobile Robots Using Piezo Elements 418
7. Conclusion 420
1. Introduction 426
2. Piezoelectricity Theory 428
3. Distributed Structural Identification Theory 429
4. Distributed Vibration Control Theory 437
5. Case Studies 447
6. Summary and Conclusions 464
PREFACE
ABSTRACf
Since micro manipulation is necessary for the handling of the micro order small
objects, actuators suitable for a micro manipulator have long been desired. This paper
proposes a simplified structure of the new electrostatic actuator with multi degrees of
freedom(3 DOF and 6 DOF), which is different from the conventional ones. For the
dexterous manipulation, versatile skillful motions are required and this can be easily
accomplished by using the proposed micro actuator. In this study, a prototype of this micro
electrostatic actuator is produced. In this paper, the nonlinear dynamics of this actuator is
modeled, then the position control simulations are carried out with the proposed nonlinear
feedback control method. Design method and experimental position control results of the
prototype micro actuator are also shown for the discussion. Moreover, we introduce the
optical actuator which has advantages of (i) non-contact control and (ii) control signal
transmission. These advantage will become important for the next actuators for the high-
precision micro systems.
1. INTRODUCTION
H S Tzou and T Fukuda (eds), PreClslOn, Sensors, Actuators and Systems, 1-37
© 1992 Kluwer AcademIC Publzshers
2 T. FUKUDA AND F. ARAI
conventional relatively large sized manipulator with a precise control system, and the other
way is to miniaturize a manipulator itself, adapting itself to the small objects. Comparing
these ways with each other, we came to conclude that a small sized manipulator is
convenient in the point that can be used in a narrow space. Therefore, we have miniaturized
manipulators itself and have made two different types of manipulators; one with a
piezoelectric actuator/5/ and the other with an electrostatic actuator/8,15,16/.
As an actuator, a piezoelectric type is appropriate to the stabbing control of
penetrating small stick type tool to the membrane of cells/4f, because of its quick response.
But in the point of miniaturization, it has the structural limitations around millimeter
range. To achieve further miniaturization of a manipulator, the actuator itself must be
miniaturized. The electrostatic actuators, which have already been reported/8/, are
appropriate in this point of view.
This paper proposes a simplified structure for the new actuator with multi degrees
of freedom(3 DOF/8,15/ and 6 DOF/16!), which is different from the conventional
types/2,3,4/. For the dexterous manipulation, versatile skillful motions are required and this
can be easily accomplished by the proposed structure. In this study, a prototype of this micro
electrostatic actuator is produced. The structure of this actuator is suitable for the photo
etching process, and it has the possibility to be miniaturized much smaller, in the
future!9,12,13/.
In the study fields of the microactuator, there are quite few study work on a model
based dynamic control method. This paper describes the modeling of the proposed
microactuator. The position control simulations are carried out with the proposed nonlinear
feedback control method. Design method and experimental position control results of the
prototype micro actuator are also shown for the discussion.
Most of these actuators utilize electric energy to actuate and control itself. Recently,
optical actuator, which can be operated by optical energy sources, have been reported. This
kinds of actuators have advantages as follows; (i) Non-contact control is possible, and (ii)
It is free from electromagnetic noises. Especially, an optical actuator using optical
piezoelectro element, which has photostrictive phenomena, attracts us very much, because
(iii) it has characteristic of transforming optical energy to mechanical displacement
directly. In this paper, we introduce some characteristics of this actuator.
NEW ACTUATORS FOR HIGH-PRECISION MICRO SYSTEMS 3
(1)-A
F: electrostatic force
E : dielectric constant
When a couple of conductive plates are placed and shifted by X(Fig. I(B», thrust
force Ft arises between the plates in addition to the attractive vertical force. Thrust force is
calculated by integrating the potential energy and differentiating it with x. The first
approximation is shown in eq. (1)-B.
(l)-B
where W is the width of the electrode.
To make a small actuator with mUltiple degrees of freedom, the structural design of
the microactuator is important. Here we propose the structure of the microactuator in Fig. 2
to meet this requirement.
Figure 3 shows the moving modes of this microactuator. The tip of the probe moves
in the different direction in accordance with the applied voltage at each electrodes; no. 1,2,
3 and 4. It is easily understood that the three different types of basic moving modes can be
generated. "Mode I" is a rotary motion of the probe tip around the Y axis. This can be
attained as follows. At first, voltages are applied to the electrodes No. 1 and No. 2
equally. Then these voltages are switched to be applied at the electrodes No.3 and No.4.
Then, switching these applied voltages again to the electrode No.1 and No.2 and repeating
this process, we can produce Mode 1. "Mode 2" is a rotary motion around the x axis.
Likewise, applying voltages to the electrodes No.2 and No.3 equally, and switching these
applied voltages to the electrodes No. 1 and No.4, we can produce Mode 2. In the same
way, applying voltages to the all electrodes equally, and switching them to the different
level, a traveling motion can be generated in the z direction, which is called "Mode 3".
NEW ACTUATORS FOR HIGH-PRECISION MICRO SYSTEMS 5
3.2 Characteristics
The model of the actuator is shown in Fig. 4. A base coordinate system is set at the
center of the stator electrode. The coordinate system of the movable electrode is set parallel
to the Z direction at the distance of c from the base coordinate system.
Now, we consider the relationship between the electrostatic force Fv and the
restoring force Fk of the suspension spring which holds the movable electrode. Figure 5
shows the relationship between the displacement x and the generated electrostatic force Fv
on each supplied voltage, and the relationship between the displacement x and the restoring
force Fk of the suspension spring.
Here we think the case that the primary distance between the stator electrode and the
movable electrode is 100 micrometer, and the applied voltage is less than 300 v, and as in
Fig. 5, the electrostatic force curve ( Fv curve) comes in contact with the restoring force line
(Fk line) at the applied voltage of 300 v.
In Fig. 5, the intersection of the Fv curve and the Fk line means that at that voltage
and displacement, " Fv = Fk " is realized and these forces are balanced. For example,
when the applied voltage is 200 v, there are 2 points, that is , the points around x = 10
11m and 67 11m where these forces are balanced. In this case, these points are apart from
each other, so they do not affect each other. But in the case that the applied voltage is
around 300 v, the stable points around x = 33 11m are close to each other, and they have
possibility to affect each other at that applied voltage. This will degrade the stability of the
position control. Moreover, if the voltage is applied over 300 v, the system becomes
unstable. We should apply the voltage less than 300 v.
In addition, the difference between the Fv curve at 300v and the Fk line implies
the maximum attraction force by the electrostatic actuator at that displacement.
Therefore, the electrostatic actuator cannot exert any force to the objects
around the applied voltage of 300 v.
Next, we calculate the elastic constants of the spring which supports the movable
electrode (Fig. 2-3). In Fig. 5, the gradient of the Fk line implies the elastic constant. When
we set the Fv curve contact with the Fk line at the possible maximum voltage, we can
make the working range of the actuator wide while preserving the capability to exert force
6 T. FUKUDA AND F. ARAI
to the objects. So here we design the elastic constant satisfying the conditions denoted in
3.2(Fig. 5), that is, we design the Fv curve comes in contact with the Fk line at the maximum
voltage of (300 v).
Figure 6(a) shows the configuration of the spring. When the spring is loaded, it is
deflected as in Figs. 6(c) and 6(d). The relationship between the displacement and the load
is given by the following equation.
(3)
where P: load
1: length of the spring
E: Young's modules
b: width of the spring
h: thickness of the spring
In this case, the elastic constant of the spring is given by the following equation.
(4)
Cause we design the Fv curve comes in contact with the Fk line at the maximum voltage of
300 v, here we determine the elastic constant equal 188.8 [N/m]. Then, we design the
spring as follows: The thickness h is 50 [Ilm], the width is 0.49 [mm] and the length is
8.0[mm].
Subsequently, we calculate the mechanical characteristics of the actuator. The
torsional elastic constant around the Y' axis is calculated as follows. Here we suppose the
rotary angle i (See Fig. 7) is written as follows,
. dy Ml
1 dx 16EI tana (5)
and suppose a is small and tan a =a , then the torsional elastic constant around the Y'
axis is written as follows.
NEW ACTUATORS FOR HIGH-PRECISION MICRO SYSTEMS 7
(6)
lee M 16EI
e 1 (7)
The torsional elastic constant around the X' axis(Fig. 8) is calculated as follows. by the
assumption that <II is small.
M=2X(lwx(lwsin<l> 16Ebh3))
2 2 213
, , , 4E,, ,b, , h,-,3w
!!....2-ssin<l>
13 (8)
(9)
(to)
The inertia moment about the X' axis I~ (Fig. 9) is calculated as follows.
(11)
The inertia moment about the Y' axis Ie (Fig. 10)is calculated as follows.
(12)
(13)
3.4 Dynamics
(14)
(15)
where
wcp=cos'l'cosS<I>+sin~
w9=-sin'l'cosS<I>+cos'l'S
w1jF'sinS+\jT (16)
Using K and P, we can derive the dynamic characteristic equations by the Lagrangian
method. The dynamic characteristic equations of this actuator are written as follows.
-
T9=-}<~ Ia)sin2\jfcos9 $+(I~sin 2\jf+Iacos2\jf)e
. .
+1 ~ -Ia)cos2\jf-I", }cos94>\jI+(~ -Ia)sin2\jf9\j1
·2
·1·{I~cos2 \jf+Iasin2\jf-I",)sin 294> lea9
..
T",=I",sin94>+I",'¥+1 (Ia-I~)cos2 \jf+~ }cos94>9
·2 ·2
·1·{I~-Ia)sin2 \jfcos294> ~~-Ia)sin 2\jf9 (17)
(18)
Here we derive the force and torque at each axis generated by the applied voltage.
At fIrst, we are to calculate the distance h between the point A on the movable electrode in
Fig. 11 and the stator electrode. The transformation from the point 0 in the basis coordinate
system to the point A is expressed by the matrix T1 written as
or T1 is written as follows.
10 T. FUKUDA AND F. ARAI
aCaCljrbCaC",
0
CaC", -CaC",
a(S~SaC",+c~S",)
S~SaCV+C~S", -S~SaS",+C~C", -S~Ca -b(S~SaS",+c~C",)
Tl
-C~SaC",+C~S", C~SaS",+S~C", a(-C~SaC",+s~S",)
C~Ca
-b(C~SaS",+s~C",)+z
0 0 0
1
(20)
where
S~=sin <I>,C~=cos <I>
Sa=sin e,Ca=cos e
S",=sin 'If,C",=cos 'If (21)
(22)
(23)
The differential calculus of h with respect to time is given by the following equation.
(24)
The torque and the force around X' axis generated by the small capacitor at the point A are
given by the following equation.
NEW ACTUATORS FOR HIGH-PRECISION MICRO SYSTEMS 11
(25)
(26)
dF - Ee£V2dxd
z 2h2 Y (27)
By integrating the whole area, we can obtain the torque and force as follows.
(28)
~ JJ"Ee£y2nX
Ta=- £oJ 2 dxdy
n=l 2h (29)
Fz=- ±
n=l
Jj"Ee£;dxdY
2h (30)
(31)
(32)
(33)
12 T. FUKUDA AND F. ARAI
(34)
(35)
(36)
(37)
Based on the dynamics of the microactuator derived here, we can develop the control
scheme in the next chapter.
In order to sense the position and the attitude of the movable electrodes, the
positions of the four corners of the movable electrodes are measured by the 4 electrostatic
capacitance displacement sensors. After the target position( Zo, <1»0, 80 ) and the attitude of the
movable electrode is given by the operator, the target positions of the 4 corners of the
movable electrode are calculated. The applied voltages are determined with the
appropriate feedback gains according to the differences between the target positions and
the positions of the 4 corners of the movable electrode.
The relationship between the input voltage and the generated force of the
electrostatic actuator is nonlinear. Nonlinear effects must be controlled skillfully based on
the dynamics. In this paper, we propose the nonlinear feedback control method for the
electrostatic actuator with 1 DOF.
The model of the electrostatic actuator with 1 DOF is shown in Fig. 12. The equation
of motion of this actuator is written as follows.
NEW ACTUATORS FOR HIGH-PRECISION MICRO SYSTEMS 13
.. k(Z-ZstdrCZ-U.L-
mz=- .\. A,,2
z2 (38)
where
A_EoES
2 (39)
(40)
where
kp: position feedback gain
kv: velocity feedback gain
(41)
then
- - -kp (*
Ay2 Z -z)k·
- vz=u
z2 (42)
mz=-k(z-ZstcV-d+u (43)
actuator is used as a dexterous manipulator, force control is required. In such cases, the force
sensor like a strain gage will be made at the bottom of the end-effector with the photo
fabrication technique, and the clamping force can be sensed.
Based on the equations derived in the chapter 3, we simulated the position control of
the actuator. The control rule is given as follows.
(44)
where
u:5:0
u=_Ay2
z2 (45)
The input voltage applied at each electrode is determined by the following equation.
(46)
Simulations of the position control of the electrostatic actuator with 3 DOF are
carried out by using the Runge-Kutta method. Parameters of the simulation model are
specified as follows.
These parameters are the same as those of the prototype micro electrostatic actuator. The
NEW ACTUATORS FOR HIGH-PRECISION MICRO SYSTEMS 15
The electrostatic actuator is shown in fig. 14. The actuator consists of five parts. (See
Fig. 2)
The stator electrode is made of a print circuit board by the photo etching process.
The stator electrode is separated into four parts, each of them are wired to the pads located
at the reverse side of the stator electrode. The electrodes for the displacement sensors are
made at the 4 corners of the stator electrode at the same time.
(4) Spacer
The spacer is made of a 100 micrometer thick polyester film. The distance between
the two electrodes is adjusted by the spacer.
16 T. FUKUDA AND F. ARAI
(5) End-effector
The end-effector is made of a 0.5 [mm] thick FRP. The thickness of the spacer is 100
micrometer, so that the maximum displacement on each axis is Zmax=lOO [Jlm],
$max=+-0.8 18[Deg], 9max=+-0.573[Deg], individually.
In this study, we produced a prototype devise which has the "micro "scale
working range and resolution. Because the structure of this actuator is suitable for the photo
etching process, it has the possibility to be miniaturized much smaller.
Figure 15 shows the system for the experiment. We carried out the experiments of
the position control using the prototype 3 DOF electrostatic actuator with the following
equation.
(47)
The result of the position control experiment is shown in Fig. 16. The position
indicated by the perforated line in the figure of zl to z4 are the target positions of the 4
comers of the movable electrode. Each position of the 4 comers of the movable electrode
moves to each target position, respectively. In the experiment, because of the lack of the
calculation speed of the computer and the accuracy of the displacement sensor, each
electrode is not balanced sufficiently. In this case, the desired position was set around z=90
(x=lO) micrometer, where the stabilizing force is easily generated. The experiment system
must be improved in the future.
8.1 Structure
micro actuators into a new one in the shape of a triangular prism. This micro actuator
consists of fixed electrode, movable electrode, spiral spring, and axis The structure is
shown in Fig. 17 and the prototype is shown in Fig.18. The inner electrode, labeled as the
movable electrode hangs in the midair of the outer part, labeled as the fixed electrode by the
two spiral springs. The stick in the middle of the movable electrode is the center axis. Six-
degree-of-freedom is attained at the tip of the axis.
The fixed electrode is fabricated by the photoeching process on the 0.8 mm thick print
circuit board (PCB). This process is summarized as follows; first the PCB made of glass
epoxies with both sides of copper leaf, are coated with photo resist, contacted with photo fllm
firmly, and printed. Next after development, the resisted PCB is etched. Each fixed electrode
unit is made of four rectangular ones. This is a basic 3 DOF electrostatic micro actuator. In
this structure, the voltage can be applied to the electrodes independently through four holes
made at the low part. Hexagon or triangular prism type can also be considered for the
construction. The structure like triangular prism can be built by combining three units.( Fig.
19 ).
Each movable electrode unit shown in Fig. 20 is fabricated by the same photo etching
process as the fixed electrode. The whole movable electrode is constructed with three units.
A hexagon structure can also be considered. The shape of the electrode is the triangular
prism, in this case. It is worth to note that the length toward the metal part axis of the
movable electrode is shorter than that of the fixed electrode. The voltages are supplied with
the pads fabricated at the upper part through the axis put in the movable electrode. The
triangular prism of the movable electrode is inserted into the fixed one, and hangs in the
midair of the outer part by the two spiral springs, keeping some distance between the two
electrodes (prototype is lOO[llm]). The outer and inner electrodes form parallel condensers.
Of course, each electrode is electrically insulated.
18 T. FUKUDA AND F. ARAI
The movable electrode and the axis can not move freely on the previous boardlike
spring. So we fabricate the spiral like spring ( in Fig. 21) using the same process with the
50[~m] thick brace leaf. The diameter is 4 [mm]. Voltage is applied through this spring. This
design technique is effective in making the movable electrode and axis move smooth without
any extra wiring.
8.5 Axis
(1) Movement along the axis X: If the quad electrode BI, B2, B3, B4,CI and C2 are
applied voltage ( the distribution of voltage is owe to the other exact calculation),
the tip moves toward +X. If AI, A2, A3, A4, C3 and C4 are applied voltage, the tip
moves toward -X.
(2) Movement along the axis Y: If CI, C2, C3, C4 or AI, A2, BI, B2 are applied
voltage, the tip moves toward +Y or -Y.
(3) Movement along the axis Z: If AI, A2, BI, B2, CI, C2 or A3, A4, B3, B4, C3, C4
are applied voltage, the tip moves toward +Z or -Z.
(4) Rotation around the axis X : If A4, B3, CI, C2 or A2, BI, C3, C4 are applied
voltage, the tip rotates around the axis X.
(5) Rotation around the axis Y : If AI, A2, C2, B3, B4, C4 or BI, B2, CI, A3, A4, C3
are applied voltage, the tip rotates around the axis Y
(6) Rotation around the axis Z : If AI, A2, BI, B2,Cl,C2 or A3, A4, B3, B4, C3, C4 are
applied voltage, the tip rotates around the axis Z.
Here we suppose that nine sensors are attached at the outer (i.e fixed) electrodes. In
this case, it will be possible to control the position and pose of the micromanipulator. The
sensor is, for example, laser or capacitance displacement sensor which has high resolution
and which measures the distance between the fixed electrodes and the movable electrodes.
The entire arrangement of sensors are described in Fig. 23. The arrangement of the sensors is
described in Fig. 24( top view) and Fig. 25( bottom view). Eight sensors out of nine are
arranged at the electrodes from Bl to B4 and from Cl to C4 of the side faces. Rest one is set
at the bottom of the manipulator. Let the sensor which is placed at the No. 1 electrode ( see
Fig. 24 ) of the side B be shorten into SiB for example. A sensor placed at the bottom
20 T. FUKUDA AND F. ARAI
labeled as Sz measures the displacement along the axis Z. In Fig. 26, A point P is on the axis
included in the bottom triangle, a point Q is in the top triangle and a point R is on the tip. The
distance between the point P and Q is h, the distance between the point Q and R is d. So P is
(0,0, °) ,Q is (0,0, h ) and R is (0,0, h+d). The small displacements of dx, dy, dz and the
small rotations of d9x, d9y, d9z are calculated by eq. (48).
- .~(dS~+dS~+2dS:+2d~)
2,3
h+dSz
-[~]
dSz
(48)
where" dS1B, dS 2B, ,dSz" stand for each displacement calculated from the output of the
sensors SIB, S2B, ,Sz, respectively. Ds is the distance between S Ie and s2c. These
equations indicate the better possibility to control the pose of the tip by the sensor outputs.
NEW ACTUATORS FOR HIGH-PRECISION MICRO SYSTEMS 21
In this paper, the position of the tip is measured on the TV monitor using the
microscope with a CCD camera system. The rotation angle 9 is calculated by eq. (49),
assuming that the rotation arises at the mid of the manipulator(Fig.27).
(49)
where I is the distance between the mid of the manipulator and the tip. ~t is the displacement
of the tip.
The static characteristics are shown in Figs. 28, 29 and 30. Figure. 28 indicates the
static characteristics of movement along the axis Y. Fig. 29 is the relation between the
voltage and the movement along the axis Z. The rotation along the axis X is shown in Fig. 30.
An applied voltage of 350v results in 3.15 [~m] movement along the axis Y and 0.7 [~m]
movement along the axis Z and 1.1xlO- 2 [Deg] rotation around the axis X. The movement
along the axis X has the same moving mode, so the characteristic is similar to that of the axis
Y.
In recent years, many kinds of actuators for the micro-electric mechanical systems
have been investigated. The conventional actuators such as, electrostatic actuator,
piezoelectric actuator and SMA have problems of transmission of both energy and
information. Recently, optical actuator, which can be operated by optical energy sources,
have been reported. This kinds of actuators have advantages as follows; (i) Non-contact
control is possible by transmitting energy and information, and (ii) It is free from
22 T. FUKUDA AND F. ARAI
electromagnetic noises.
In this paper, we introduce (Pa, La)(Zr,Ti)03 ceramics (PLZT 3/52/48) as an
optical piezoelectric element. This actuator is driven by the applied U.V.(ultraviolet) light
signal which has peak. point at 365 [nm]. From the experimental results(Fig.31), the photo
response characteristics of this actuator is characterized by the following tree different
effects; (i) photostrictive effect, (ii) pyroelectro effect, and (iii) thermal effect. Total
response of the actuator is complex but characterized by the applied U.V. light source (
strength, power spectrum, distribution) and the environmental condition (tempemture).
Based on the experiments, we modelled photo response of this optical piezoelectric
actuator(Fig.32). The simulation results of this model(Fig.33) agreed with the experimental
results. The optical piezo actuator has characteristics of both energy and information
transmission. The pyroerectric effect can be used as the control signal transmission, while
the photostrictive effect can be used as the driving energy transmission. These two functions
can be realized by single light beam emitted to this micro-actuator. This kind of actuator is
prospective as the next actuator for the high-precision micro systems.
12. CONCLUSIONS
We have shown the structure and the position control method of the newly
developed multi DOF electrostatic micro actuator. The proposed microactuator can be
applied for the micromanipulators. We derived the modeling and the dynamics of the
microactuator. Based on the model derived here, we proposed the design strategy of the
electrostatic actuator on the point of the position control, and demonstrated the position
control by simulation and experiment using the prototype actuator. A problem is that this
type of micro actuator has rather a few work space. This actuator will be combined with
roughly movable devices in a pmctical use.
We also introduced the optical actuator. Optical piezoelectro element, which has
photostrictive phenomena, attracts us very much, because it has characteristic of
transforming optical energy to mechanical displacement directly. The pyroerectric effect
can be used as the control signal transmission. These two functions can be realized by single
light beam emitted to this micro-actuator. This kind of actuator is prospective as the next
actuator for the high-precision micro systems.
NEW ACTUATORS FOR HIGH-PRECISION MICRO SYSTEMS 23
REFERENCES
[11] H. Fujita and A. Omodaka, "Electrostatic Actuators for Micro mechatronics", Proc.
of IEEE Micro Robots and Teleoperators Workshop, Hyannis, MA, 9-11 Nov.,
1987.
[12] William C. Tang, Tu-Cuong H. Nguyen, Roger T. Howe, "Laterally Driven
Poly silicon Resonant Microstructures", Proc. of IEEE MEMS, pp.53-59, 1989.
[13] Mark G. Allen, Martin Scheid!, Rosemary L. Smith, "Design and Fabrication of
Movable Silicon Plates Suspended by Flexible Supports", Proc. of IEEE MEMS,
pp.76-81, 1989
[14] Richard P. Paul, "Robot Manipulators: Mathematics, Programming, and Control",
The MIT Press, 1981, chapter 6, pp. 152-188.
[15] T. Fukuda and T. Tanaka, New Mechanism and Dynamic Control Method of Micro
Electrostatic Actuator with Three Degrees of Freedom, Proc. 1991 IEEE Int. Conf. on
Robotics and Automation, (1991), pp.1610-1615.
[16] T. Fukuda, M. Fujiyoshi, K. Kosuge" and F. Arai, Electrostatic Micro Manipulator
with 6 D.O.F, Proc. of Intelligent Robots and Systems (IROS) '91, Osaka,
(1991), pp.1169-1174.
Fv
,/999,;?2 l
,/Q QQ7- T
UFv
/~ Ft~
6
a
x ====/ :;=::=:i53-
~Ft
8 ~
I ~ I
I I
: I I .OO~.
.0..
~ ~ ~
I . I I I
~~
~ : ~'-/
8~~0
~;~
~0
~
rJ
'II ~ 1~1 ,j
~
HodeJ
~
lien:!.'
==::::::JII ~ I CI=~
LC:=== ==~
1.3 ON OfF' I. ... ON
~
1Iod.1
x y
t
__
J
:J L
§
L7 74z ~ "h
bl!
(a) (b)
(d) (c)
, ,
X X
,
Z
z'
x'
,
z
X
Xslu _
v
o
-- ': !."
. HJJ .'
t:...::. ' ::
I::I}-__ = -.:-':..:-,-.-.:u.s.~!-,J
(I
eol-~
. t [sec ] tl~e~] ~~I~
.-V~ 100.'
t 1s~?,:: (\_jl;~l'~IL_-_-__~-~------'j.t"'';=''~J:b;C;'l)
!dH •
c
12 Dr 12bll Electrostatic
/ ~-I---l Displacement
00 LSensor
-_ _ _ _-----'
p ~
X4
u ~ffl
TllJ
I-Ugh
Voltage
Amplifier
1-_-'
~
X4
;~'C
'~J,,., "~ , ".-----..--- ;~T'=---~,--~-.-..-~-,,-'--""''''-
=
~~r=.=::'~;"~"" "." ;~F~·.,;--,.,:<,:-·;:·c;.,l" ,':
o
~= l.~ 0 I. thee
~ Coilt:d Spring
Fh:e.d EJ.'~ctrodc.
No.101E 58
1
2
~----HI- X
3
Side C
S~ Sf
S~
Side A
Side D
S~
S,
'--+-x
[
3
C
Q)
2
E
Q)
u
c;l
'0.
.~ 0 100 200 300
0
Voltage V
~ 1J
oj
....t::
L
0
E
~
u . ~\
-.Cd
p..
V) ~
,
100 200 300
is
Voltage V
~
.s 10
e
b....... 0
...
X
<!
~
00 0
c
~
Voltage V
electrode UV
strain gauge
PLZT
1-0 I- L
-Wet) W -he(L.t l
3~---------------------'
\
~ 0~--------~~(--------1
-1
-2
t ••
( c)
3,----------------------r
, .
t (8ec)
R S. Feanng
Dept. of EE&CS
Umv. of Callforma
Berkeley, CA 94720
Abstract
There are many advantages to shnnkmg robots and mechamcal actuators to the
same SIZe as the parts to be mampulated. Extremely dehcate forces can be apphed, robots
can be readIly parallehzable, and the relative accuracy requITed can be markedly reduced.
One of the major dIfficulties m bUlldmg mIlhmeter scale mIcro-robots IS overcommg
forces due to fnctIon and wmng. FnctIon forces can be reduced by usmg flexures mstead
of rotary or hnear shdmg Jomts, and usmg flUld lubncatIon, such as an aIT-beanng. Thls
paper consIders two ImtIal Ideas towards Implementing practical sub-mIlhmeter robotic
systems, and then exammes a natural analog to a mobIle mIcro-robot: a smgle celled
ammal. The first mIcro-robot Will use sIhcon structures that can be folded out of the
plane of the wafer. The second mIcro-robot system Will be bUllt from an arr-beanng-
supported planar electrostatlc motor, made by sIhcon processmg, and has 3 degrees-of-
freedom. MobIle deVIces on a sub-mIlhmeter scale working m a flUld medIUm could be
useful for mampulatIon and testing of small bIOlogIcal or electro-mechamcal com-
ponents. The power requITements for such robots working at low speeds are very favor-
able.
1. Introduction
In the not too dIstant future, mobIle mIcro-robots, such as depIcted m FIg. 1, may be
batch fabncated usmg sIhcon and photo-hthographIc techmques. The sIhcon may be
used for on-board mtelligence, and m addItion for electro-mechamcal sensor and actuator
systems. Thls mlcro-robot-on-a-chlp has been populanzed by Brooks and Flynn [1989]
These robots may see Wide apphcatIon m micro-tele-operatIon for very small
39
H S Tzou arui T Fukuda (eds), PreclslOn, Sensors, Actuators arui Systems, 39-72
© 1992 Kluwer Academlc Publzshers
40 R S FEARING
maccesslble areas, and m the massIvely parallel handlIng of small bIOlogIcal or elec-
tromechamcal elements.
Another feature of mlCromaclumng IS that extremely small and senSItIve tools can
be made, and phYSIcal scahng laws can be taken advantage of. One scahng law that IS
helpful for mIcro-actuators IS the mcreased break down field strength of very small gaps
due to the Paschen effect. Thus for small SIzes, the obtamable electrostatIc forces can be
stronger than electromagnetIc forces [Tnmmer and Jebens, 1989] Small tools ImpractI-
cal to fabncate by other methods, for example, an electrostatIcally dnven gnpper able to
handle 10 Il111 dIameter parts [KIm et aI, 1991], can be made by surface and bulk
mlCromaclumng of sIhcon. AnIOng the many other novel deVIces made pOSSIble, are
planar rotary mIcro motors [Fan et aI, 1989], and mImature gyroscopes [GreIff, et al
1991].
MICRO STRUCTURES FOR IMPLEMENTING SUB-MILLIMETER ROBOTS 41
Small objects, such as SIngle cells, are very dIfficult to handle manually. SIngle clnp
systems wlnch Incorporate actuators for movIng objects and pOSItIon sensors WIll greatly
SImplIfy the transport and mampulatIon of very small objects. Due to large frictIon
forces on small dry parts, a flUId medIUm and lubncatIon (eIther gas or hqUId) IS com-
monly used. For example, whole SIlIcon wafers on a gas bearmg can be dnven electros-
tatIcally [H. TokIsue, et al, 1991]. AlternatIvely, leVItatIon due to the MeIssner effect
[FUJIta, 1989] can be used to aVOId problems WIth frictIon.
To be useful, SIlIcon mIcro-robots will need tools to Interact above the plane of the
wafer, not Just In It. One promlSlng approach, called "SIlIcon ongamI" [Slumoyama,
1992] or mIcro-lunges [plster, et a11991] allows planar fabncatlOn, followed by 3 dImen-
SIOnal assembly of structures. ThIs paper examInes some of the new tool makmg capabil-
ItIes obtaInable WIth thIS process.
An Integrated system for mampulatIng dry parts In the plane USIng multIple mobIle
mampulatIon umts was proposed by PIster et al [1990]. ThIs deVIce conSIsts of a 1 cm 2
substrate WIth an arr beanng to support IndlVldua11mm 2 platforms (see FIg. 2). The IndI-
VIdual platforms are dnven In the plane by electrostatIc forces, and could carry gnppers,
probes for sensIng, or tools for processIng. By IncorporatIng capaCItIve pOSItIon sensIng
42 R S FEARING
of the platforms, an mtegrated mIcro system for parts handlmg could be made on a smgle
chIp. ThIs paper explores the capabIhtIes of these sImple planar robots, and suggests pos-
sIble alternatIve desIgns, mcludmg a method to bUIld autonomous mIclO-robots on an arr
beanng.
Current mtegrated mIcro systems are lImIted by the mostly planar mIcro-machImng
techmques avaIlable. For some apphcatIOns, It will be necessary to be have sensors and
actuators that extend far beyond the surface of the devIce, e.g. mIcro-robots. For exam-
ple, a sensor or probe may need to be sufficIently far from the sensor surface to aVOId
boundary layer effects.
One way to bUIld three dImensIOnal devIces IS to use new fabncatIOn techmques, for
example, Laser-assIsted ChemIcal Vapor DeposItIon (LCVD). In LCVD, a focussed laser
beam actIvates chemIcal specIes whlch can eIther locally deposIt or etch a structure
dependmg on the gas medIUm used m the reactIon chamber. ComplIcated structures such
as a boron spnng [Westberg et aI, 1991] and stepped pItS WIth controlled slopes [Bloom-
stem and Ehrhch, 1991] have been demonstrated WIth LCVD. Volume resolutIon of
111m3 has been obtamed.
Wlnle LCVD IS very flexIble for machInmg three dImensIOnal structures, It IS
mherently a senal process, and hence much slower than photo-lIthographlc fabncatIon.
An alternatIve approach IS to perform conventIonal two dImensIOnal plocessmg, and then
to assemble (post-process) three dImenSIOnal structures from planar components. These
components could be bonded to each other usmg, for example, weldmg techmques
[Fedder and Howe, 1991].
WIre anemometer, assembled by foldIng 3 orthogonal lunge structures out of the plane.
(The anemometer structure extends approXImately 200 ~m above the substrate). One
advantage to the hInged 3 dImenSIOnal structures, compared WIth LCVD, IS that all the
planar hthograpluc resolutIOn IS maIntaIned.
Standard surface mIcromaclumng, has lugh planar resolutIon, low vertlcal resolu-
hon, and hmIted vertlcal range (tYPICally less than 5 ~m). These charactenshcs make
surface mICromaclumng an excellent chOIce for planar apphcahons, but have hmIted uhl-
Ity for three dImenSIOnal deSIgns. We present a process In wluch structures are fabn-
cated USIng surface mIcromaclumng, and then rotated out of the plane of the wafer on
Integrally fabncated lunges. The resulhng structures have lugh resoluhon In both the
planar and vertlcal drrechons, and have a vertlcal range from 10 ~m to more than a mIl-
hmeter. TIns lunge-based method allows the benefits of lugh resoluhon surface hthogra-
phy wlule prOVIdIng access to the thrrd dlffiensIOn wIth lugher vertIcal resoluhon than
preVIOusly pOSSIble. Drawbacks of hInge-based deSIgns Include the need for post-process
assembly, and IncompahbIhty WIth typIcal MOS processes.
2.1. Process
The SImplest verSIOn of the lunge fabncahon process IS a three mask, double layer
POlysIhcon process WIth OXIde sacnficlal layers (FIgure 3). A sacnfiClal phosphosIhcate
glass (PSG) layer IS depOSIted on a bare substrate, followed by an undoped polysIhcon
layer (poly]), and a dopIng PSG layer. All deposIhons are by low pressure chemIcal
vapor deposIhon (LPCVD). The POlysllIcon IS patterned In a plasma etcher. TIns
POlysIhcon etch defines the maJonty of the structural components, IncludIng the lunge
'pInS', about WhICh most structures WIll rotate. A second sacnficIal PSG IS deposIted
and both the first and second sacnficlal OXIdes are patterned In a plasma etcher. TIns etch
defines contacts between the second POlysIhcon layer and the substrate, as well as con-
tacts between the two POlyslhcon layers. A second layer of polyslhcon (poly2) IS depo-
SIted and patterned. Tlus second POlysIhcon etch defines the 'staples' wluch tie the first
POlysIhcon layer to the substrate, as well as formIng addItIonal structural components.
FInally, the sacnfiClallayers are removed In a concentrated HF etch, the wafers are nnsed
In deIOllized water, and arr dned at room temperature. PerforatIons are used In the larger
structures to allow complete release In the 1 mInute release etch.
44 R S FEARING
TYPICal film thIcknesses for the sacnficial and dopmg PSG layers are between 0.5
and 2.5 !lm. POlysIhcon layers are tYPIcally between 1 and 2 !lm thIck. If the structures
are mtended to be e1ectncally actIve (e.g. the anemometers dISCUSSed below), the sub-
strate IS passIvated WIth a 0.5 !lm thermal oXIde and 0.1 !lm LPCVD mtnde before the
first sacnficial oXIde IS deposIted.
DImples can be added to the POlysIhcon layers by patternIng part way through the
sacnficial layers WIth BHF. These dImples are not stnctly necessary, but help prevent
the POlysIhcon layers from adhenng to the substrate and to each other after release. Por-
tIons of poly] can be anchored to the substrate, If desrred, by etchIng contacts m the first
sacnficial PSG before the poly] depOSItIon.
If the total thIckness of the two sacnficial oXIde layers IS greater than the thIckness
of the first POlysIhcon layer, then the pm of the hInge WIll be able to slIde between the
two legs of the poly2 staple. TIns sort of 'play' m the hInge IS not generally desrrable,
and can be elImmated by usmg a tImed BHF etch ImmedIately folloWIng the patternIng
of poly]. TIns tImed etch undercuts the sacnficial oXIde under the hInge pm, and due to
the poor step coverage of low temperature LPCVD oXIde, mdrrectly reduces the second
sacnficial oXIde thIckness near the pm as well. TIns results m a partIally encased poly]
pm, under (and almost mSIde) a poly2 staple. Smce the hInge locatIon IS now determmed
by the poly1 pm rather than the poly2 staple contacts, we refer to thIs as a 'self-alIgned'
pm.
2.2. Design
GIVen the three mask process above, It IS possIble to make several dIfferent types of
hInges, as Illustrated m FIgure 4. The SImplest of these IS the 'substrate hInge', whIch
conSIsts of a poly] plate and lunge pm constramed by a poly2 staple. The staple IS
attached to the substrate at two contact pomts, and the plate IS free to rotate a full 180
degrees off of the substrate Note that the freedom of the plate to rotate may be lImIted
by the geometry of the pm and staple. If the WIdth of the pm IS greater than the sum of
the thIcknesses of the poly] and sacnficial oXIde layers, then the pm WIll be unable to
rotate a full 90 degrees wIthout contactIng the substrate and staple.
MICRO STRUCTURES FOR IMPLEMENTING SUB-MILLIMETER ROBOTS 45
The substrate lunge IS used to hmge poly1 plates to the substrate To lunge plates to
each other requrres a dIfferent type of lunge. Two poly1 plates can be lunged together
usmg a 'scIssor hmge'. poly2 strIpS are attached between mterdIgltated poly1 fingers,
preventmg the two poly1 plates from pulhng apart, and allowmg the plates a relatIve
rotatIon of roughly 180 degrees. Tlus type of lunge can only fold 'concave-down'. A
sImIlar SCIssor lunge, Illustrated m the figure, lunges two poly2 plates together and folds
'concave-up' .
Unhke substrate lunges, there IS no 'pm' m a SCIssor lunge. GlVen the typIcal film
tlucknesses used, a substrate lunge WhICh IS mtended to rotate 90 degl ees must have a pm
wluch IS no more than 2 !lm Wlde. SClssor lunge geometrIes are not constramed by the
film tluckness of the structural or sacnficial layers. As a result, SClssor hInges can be
made wIth all geometrIes much Wlder than 2 !lm, makmg them much stronger than sub-
strate lunges.
2.3. Assembly
After the release etch, the structures are rotated mto therr final pOSItIons. TIns IS
currently accomphshed at a probe statlon usmg standard electrIcal pi obmg eqUIpment to
rotate the structures mto pOSItIon. A sharp probe tIp IS shd under a released structure and
raIsed to 11ft the structure off of the surface of the wafer, and rotate It mto the desrred
pOSItIon. By mter-Iockmg two lunged structures, the final pOSItion of the structures can
be accurately controlled. The hmges can be pennanently locked mto pOSItIon by an addI-
tIonal depOSItIon step. For example, a 0.3 !lm PECVD OXIde depOSIted on assembled
structures enables substrate lunges to WIthstand torques m excess of 10 nNm (= 1 mIlh-
gram force at 1 mIllImeter) Wlthout damage.
Assembly IS a labor mtensive process (e.g. the gnpper requrres roughly 10 mmutes
for assembly), however we are workmg on deSIgns wluch requrre no manual assembly.
HydrodynamIC forces may prove to be very useful m automatIng the assembly process.
We have observed that many structures rotate 90 degrees or more dunng the post release
nnse, and a drrected stream of arr from a capIllary tube has a SImIlar effect on released
structures. Expenments have shown that depOSItIng a tenSIle film (e.g. 0.1 !lm sputtered
Molybdenum, 1.5xl0 11 Pa tenSIle stress) on the post-release structures makes them much
more susceptIble to hydrodynamIc forces, due to the resultIng curvature of the plates.
46 R S FEARING
We hope to explOIt thIS behavIOr by desIgmng structures whIch WIll have a tendency to
'snap' mto place when they are close to theIr desIred final pOSItIons. In thIs way, an
entIre wafer could be batch assembled by VIgOroUS agItatIon In a DI nnse, or IndIVIdual
structures could be assembled sequentIally USIng dIrected streams of aIr.
WIth one exceptIon (the gnpper) the deVIces presented here do not use the hInges
for actIve rotatIon. Rather, the hInges are used only dunng assembly to rotate the struc-
tures mto pOSItIOn. Thereafter, the hInges are deSIgned to remam locked In place.
MotIon of the structures IS then obtamed by USIng the bendIng of beams. In thIS way, the
hIgh fnctIon of the hInges IS an advantage dunng assembly, and IS not a hIndrance dunng
use. WIth these Ideas m mInd, we present a few examples of hInged structures.
Another out of plane sensmg capablhty provIded by the hInge structure IS shown m
FIgure 5. A probe for electncal testIng can be made WIth an Intnnslc spnng, and electn-
cal contact to substrate through hInges. ThIS structure has not been tested.
Recently, planar (roughly 2 Ilm thIck) micro-gnppers have been fabncated WIth a
gnppIng range on the order of 10 Ilm, [KIm, PIsano, Muller, 1991] The hInge technol-
ogy offers the opportumty to produce micro-gnppers of a scale dIfficult to obtam WIth
prevIOus mIcro-machInIng processes. Gnp surface dImenSIOns and gnpper opemngs
measured In hundreds of mIcrons are possIble, whIle actuatIng resolutIon IS on the order
of mIcrons. The structure In FIgure 6 IS a parallel-plate gnppe1 conSIStIng of four
separate pIeces. The two Jaws of the gnpper are folded up separately and locked In place
at one end by another plate WIth two slots In It. The Jaws are suspended at the end of 400
Ilm-Iong beams, each of whIch IS 20 Ilm-WIde. A 1 mm-Iong tendon travels from each
Jaw and locks mto the vertIcal handle. When the handle IS pulled back, the tendons pull
the Jaws open. When the handle IS released, the spnng force of the support beams closes
the Jaws. In the 'closed' (rest) pOSItIon the Jaws are actually 100 Ilm apart. Over 100
cycles of up to 0.5 mm opemng have caused no damage to the gnpper
MICRO STRUCTURES FOR IMPLEMENTING SUB-MILLIMETER ROBOTS 47
As prevIOusly mentIoned, the rotary hmge Jomts have very hrgh fuctIon, and elastIc
JOInts, such as a cantIlever beam, wIll be much eaSIer to dnve. There are several optIons
for actuatIng these elastIc JOInts, mcludmg shape-memory-alloy [Ikuta, 1990], elec-
tromagnetIc [Wagner and Benecke, 1991], and electrostatIc dnves. Although the forces
are very small, low voltage electrostatIcs IS perhaps the eaSIest to Implement on the lnnge
process WIth few addItIonal masks. FIgure 7 shows a smgle degree-of-freedom actuator,
wlnch consIsts of a movable plate supported cantIlever beam, and a fixed plate.
A useful range of motIon for thrs actuator IS 0° to 10°. (Assume that the hmge Jomt
IS locked III place). When a potentIal dIfference IS apphed between the fixed and movmg
plate, the plates are attracted to each other. To estImate the order of magrutude of tills
force, we can assume that the plates are approXImately parallel WIth a gap of lO!lm. A
typIcal plate SIze would be 200 !lm ( =a) square, With plate tlnckness 2 !lm ( =h). Then
the electrostatIc force nonnal to the plate (Fz) for 10 volts appbed between the plates, IS
gIven by
(1)
The actuator IS umdrrectIonal Without the restonng force of the cantIlever. The can-
tIlever spnng should be soft for the plate to be "pulled-m", yet strong enough to support
the weIght of the plate. Usmg a 2 !lm thrck polysIhcon layer, a 200 !lm long beam of
48 R S FEARING
square cross sectIon will have appropnate comphance. The approXImate spnng constant
for such a polysIhcon beam WIll be about 4xlO-2Nm- I [Lm et al, 1991] The mass of the
plate IS mplate =PSzlxwxd , where PSz IS the densIty of sIhcon, 2.3xl<fKgm-3 • For a
square plate , 200 Ilm on each edge, 2 Ilm thIck, the mass of the plate IS approxImately
2xlO- lOKg, WIth weIght of 2xlO-9N. The dIsplacement of the cantIlever beam due to the
weIght of one plate IS:
(2)
whIch IS neglIgIble, only 0 1% of the unactuated gap. Because of the mverse square rela-
tIonshIp between F z and the gap, the 10 V potentIal WIll be sufficIent to dnve the plates
together.
There are many problems to bUIldmg thIs type of stacked structure usmg the hInge
process. A recent paper by PIster [1992] proVIdes some strategIes for bUIldmg thIs dev-
Ice. ThIn and fleXIble POlysIhcon nbbon cable can be fabncated to WIfe up all the plates
to external connectIons. These connectIons would loop around the SCIssor hInge and not
exert any force when operatIon. Self-assembly catches can be added so that structures
fold up and latch mto place dunng suffiCIently eXCIted, for example, by a nnsmg step
POlyslhcon plezo-reslstIve stram gauges can be used to measure force or pOSItIon of the
Jomts (the bendmg of the cantIlever), and thIn film tranSIstors can be added to the plates
to control plate addressmg. Many problems remam to be worked out, but there IS hope
for an Implementable stacked actuator deSIgn.
MICRO STRUCTURES FOR IMPLEMENTING SUB-MILLIMETER ROBOTS 49
3. Electrostatic Actuation
ConsIder the planar vanable capacItance motor shown m FIgure 11. The platform IS
leVItated above the substrate by an arr beanng (see next sectIon) at a heIght h. The elec-
trode gnd pattern on the platform has a ratIo of 3:1 m conductor to dlelectnc pads m the
x and y dImenSIOns. The stator has a gnd pattern ratIo 4:2 m conductor to dlelectnc pads.
If the mobIle platform has ItS own voltage source, such as a battery, the entIre stator can
be grounded, and the platform can dnve Itself around, by puttmg appropnate potentIals
on ItS electrodes. FIgure 12 shows the capaCItance dlstnbutIon between the platform and
substrate electrodes, and FIgure 13 gIVes the eqmvalent ClTcmt.
Consldenng Just a row of 3 electrodes, the total capaCItance can be found from the
Co
geometry and the crrcmt usmg the parallel plate capaCItor approXImatIon as - where
3
CO IS the capaCItance of an mdlvldual element. For an apphed voltage V the force m the
tangentIal drrectIon can be found from:
F
a Co 2 V2
---V =--E -
a
x
=
ax 6 6 0 h
(3)
a Co 2 V2 2 1
F =---V = - - E a - .
z 6 az
6 0 h2
(4)
50 R.S FEARING
where a IS the length of a sIde of the platform electrode. It IS mterestmg to see the ratio
of normal to tangential force:
a
=h (5)
For tYPIcal values, say h =51-1m and a=10-3m , the normal force IS 200 times larger than
the tangential force. TIns requrres a very stiff and stable bearmg to aVOId stIckmg. The
platform has 3 degrees of freedom, as IS shown by showmg eqUIhbnum posItions for
vanous potential dIstnbutIons m Figurel4.
Electrostatic actuation may have problems With charge accumulation m the dielec-
tnc, see for example [Anderson and Colgate, 1991]. Another problem IS that electrostatic
dnve won't work m a conductive flUid medmm such as water. Magnetic actuation may
thus be an attractive alternative, see for example [Busch-Vishmac, 1991]. It IS pOSSIble
to make the platform from illgh permeabIhty or magnetIc matenal, and embed COlIs In
stator. The forces requrred for translation of the platform are very low, so even low
strength magnetic fields may work well. MagnetIcs may also be attractIve when USIng an
Internal hIgh current low voltage source hke a SIngle cell battery, SInce the call wIll usu-
ally be farrly low Impedance.
A stable and relatIvely stIff flUid beanng IS essentIal to ensure smooth operatIon of
the mobIle platform. If the rotatIonal stIffness IS too low, the platform could tIlt and
touch down, probably stIckIng to the surface. As we have seen, the normal force Into the
substrate may be 100 tImes greater than the tangentIal dnve force. The more common
approach to an arr beanng uses an external arr supply. However, for small dImenSIOns
and narrow arr gaps, a form of self-pressunzatIon can be obtaIned by USIng a squeezed
film lubncatIon. Tills seCtIon examInes the feasIbIhty of USIng the squeeze film to support
the platform.
MICRO STRUCTURES FOR IMPLEMENTING SUB MILLIMETER ROBOTS 51
In a squeeze film beanng, illgh frequency transverse OSCIllatIons of one of the bear-
mg surfaces proVIdes a pumpmg actIon The osclliatory squeeze motIon results m a
tIme-averaged pressunzatlOn effect pnmanly due to the compressIbIlIty of the gas
film, and the degree of thIS pressunzatIOn mcreases monotomcally WIth the amplI-
tude of the oscIllatIOn relatIve to the average gap
The mtUItIon beillnd a squeeze film beanng can be Improved by the followmg
example ConSIder two square flat plates WIth a velY tilln layer of alI m between them If
these two plates are brought together slowly, the alI between them wIll escape out the
SIdes of the plates, and the two plates wIll touch If the two plates are brought rapIdly
together, the gas WIll not all have tIme to escape, and wIll be compressed m the central
regIOn, mcreasmg the flUId pressure there
smce the gap IS now smaller, mcreasmg the pressure under the platform, and forcmg It
back up to Its eqUIhbnum heIght.
Assummg constant flUId densIty, Reynold's equatIon for compressIble gas film
lubncatIon IS [Sherman, 1990]:
where h IS the heIght, p IS pressure, and U IS velocIty. If the vanatIons m pressure '6p
are small compared to Pa' the ambIent pressure, then m lmeanzed form [Blech, 1983]:
The force response can be charactenzed by the non-dImenSIOnal squeeze number 0".
For a square plate of area A =a 2, dnven smusOIdally by eocosrot, the squeeze number IS
gIVen by
(8)
where mean film thlckness IS hm . For a 5 !lin truck alI beanng, W- 6m 2 area platform,
and dnve frequency of ro =W5s- l , the squeeze number IS approXImately (J:::; 10. For
large squeeze numbers, the flUId layer has mcreased reSIstance to flowmg out between the
plates, mcreasmg the effectIve stIffness of the alI beanng.
(9)
(10)
MICRO STRUCTURES FOR IMPLEMENTING SUB-MILLIMETER ROBOTS 53
The cutoff squeeze number (Jc :::2rc2 occurs when the spnng and dampmg force are equal.
The spnng force IS proportIonal to the mverse 5th power of the plate separahon:
(11)
whlle the dampmg force IS proportIonal to the mverse cube of the float heIght.
F
a
= -lhCV v2 1
e az 2 ::: - -2E 0
a 2- .
h2
m
(12)
For a 5 Ilm gap wIth V::: 10 V, the normal force IS approxImately 2xlO-5N. Thls force
eo 3
IS obtamable from the arr spnng usmg a relahve oscIllahon of only - ~ 10- . (The per-
hm
turbahons m platform heIght due to thls mduced vIbrahon should be mSIgrnficant.) WIth
thls oscIllahon amplItude, h m wIll be a stable eqmlIbnum pomt. Smce f k IS mversely
proportIonal to the fifth power of the mean platform heIght, thls beanng WIll be very shff,
but well damped due to the dampmg force f d. WIth (J) m the hlgh audIO range, a VOIce
cOlI actuator can easIly generate the requrred osciliahon, as shown m FIgure 17.
Each mIcro-robot could be 1000 ~m x 1000 ~m m area, and only 100 ~m thICk. The
mIcro-robot would rest on a flUId film (an arr beanng) and never move willIe m fnctIOnal
contact. The robot could be controlled externally, or It could contam a mIcro-computer
for on-board mtelhgence, a power source such as a solar-battery, and a layer of elec-
trodes to dnve the robot electro-stahcally. Commumcahon mIght be done ophcally. To
detenmne whether tills Idea IS feasIble, we need to look first at how much power IS
needed to dnve robots m a flUId medlUm.
(13)
A first order model for the tangenhal dnve force IS glVen by an overlappmg parallel plate
capaCItor model:
(14)
. V 2£ 0
X=--. (15)
2a~
Wlth a V = 10 volts, and the length of the platfonn a =1O-3m, the tennmal velOCIty IS
about 10-2ms -1. The power requrred to overcome tills drag force IS 10- 12 watt.
R = pVT
e 11
where p IS the densIty of the flUld, T IS the radIUS of the sphere, II IS the VISCOSIty m
N m-2 s, and v IS the velOCIty of the sphere. For a 1 cm radIUS robot m water at room
temperature (p = 103Kgm- 3 and 11 = 1O-3Nsm- 2 ), the transItlon to turbulent flow wIll
occur when v IS greater than about Ims- 1. For sImphcIty, tills paper only conSIders
slower speeds, and lammar flow condItlons. For lammar flow, the drag force on a sphere
IS gIVen by Stoke's Law [Prandtl and TIetjenS, 1934]:
V 2
P drag =F·v =6nT 11- ,
11
where 11 IS the effiCIency of converSIOn from electrIcal power to mechamcal power. Now
conSIder the drag-lImIted velOCIty for the mIcro-robot WIth the assumptlon of eIther an
mternal battery or a surface battery as power sources, and that all power goes to propul-
SIOn. Assume that for an mternal battery, the energy per umt volume wIll be constant,
mdependent of the SIze of the robot. For a constant hfe of the lobots power source
(mdependent of robot SIze), the power denSIty per umt volume must be constant. Thus,
for an mternal battery, we have:
Thus the maXlffium velOCIty of the robot will scale hnearly WIth the robot radIUS:
v max = -(2PVD / - )
T 11 ~
3 11
For a surface battery, for example a photo-voltaIc source, there wIll be constant flux per
umt surface area:
P
area
= ~2 = 6nTIlv22 = 31l2V2
4nT 114nT 11 T
56 R S FEARING
Thus
V max
=.r,:-
r
( 2Parea 11 )\6 .
3~
Thus for a surface battery, the maXImum speed scales only as the square root of the
radIUs of the robot.
ThIs sectIon calculates the maXImum velocIty for several example robots and power
sources. The drag force for a sphere of appropnate SIze WIll be used to obtam an order of
magrntude estImate. VISCOSIty for water IS assumed to be 1O-3Nsm- 2 and VISCOSIty for
arr IS assumed to be 2xlO-5Nsm- 2. Two power sources WIll be consIdered. The first WIll
be a lIthIum battery, WIth an energy densIty of approxImately 200OJcm-3 [HorOWItz and
HIll, 1989]. The second power source IS a photo-voltaIc array, whIch has an output of
10mW cm-2 (assummg one sun IllummatIon, and 10% effiCIency), or eqUIvalently
1O-lOW~m-2.
For the second example, consIder a 1 mm square platform, 0.1 mm thIck, whIch
floats on an arr beanng, WIth velocIty of Ims -1. The drag force F drag :::J 10-7N, and
P drag :::J 1O-7W. The avaIlable power from a photovoltaiC source IS
0.01cm 2 x 10mWcm-2 or lO-4W. Here the avaIlable power IS 3 orders of magrntude
more than reqUITed to overcome drag reSIstance, even at very hIgh velocItIes.
For the thrrd example, consIder a mIcro-robot the SIZe of the smgle-celled CIlIate,
paramecium, whlch has an apprOXImate length of 300~m, and SWIms at 1O-3 ms -i. Its
drag force IS F drag :::J 10-9N, and power consumptIon would be Pdrag :::J 10- I2 W. The area
for a photovoltaic array IS easIly :::J103~m2. Thus the avaIlable electncal power IS 105
tImes greater than the requrred mecharncal power. Very mefficient propulsIOn methods
are qUIte reasonable for thIs SIZe of mIcro-robot.
MICRO STRUCTURES FOR IMPLEMENTING SUB-MILLIMETER ROBOTS 57
These ammals exillblt the phenomena of galvanotaxIS, whereby they SWIm towards
the cathode when placed m a current m a conductmg flUId. Usmg VIsual sensmg of the
ammal's pOSItIon, sImple closed-loop trajectory followmg m the plane has been shown
[Feanng, 1991]. FIg. 21 shows the trajectory follOWing capabIlIty of the ammals for
trackmg the penmeter of a box willch IS approXilllately 5 mm on a SIde. POSItIon accu-
racy to ±lmm was acilleved usmg proportIonal control. Future work WIll examme the
cooperatIon of multIple cells at movmg neutral densIty objects, as In FIgure 22.
7. Summary
For the two proto-type mIcro-robots presented, the arr-beanng and water types, the
reqUITed dnve power IS 0.001 % to 0.1 % of reasonably avaIlable electrIc power. One hm-
ltatIon IS that the mIcro-robots should not try to move extremely qUIckly. However, the
power requITed at speeds of several body lengths per second In flUId IS very small for
these mIcro-robots. The opportumty presents Itself of developIng completely new pro-
pulsIOn methods, that would be too horrendously IneffiCIent to conSIder USIng at the
macro-scale. For example, In the mIllImeter SIze range, Jet propulSIOn or even thermal
gradIent/convectIve power may prOVIde enough force. These IneffiCIent actuators may
have the advantage of beIng easy to fabncate, and ease of fabncatIon IS much more
Important than effiCIency of operatIon for these flUId mIcro-robots.
Acknowledgments
PartIal fundIng for tills research was proVIded by the SemIconductor Research Cor-
poratIon under grant 91-DC-008, the Berkeley Sensor and Actuator Center, and the
58 R. S. FEARING
National Science Foundation under grants NSF-IRI-9157051. The author thanks K.SJ.
Pister for his valuable contributions and collaboration.
References
[1] K.M. Anderson and J. Edward Colgate, "A Model of the attachment/detachment
cycle of electrostatic micro actuators", ASME Micromechanical Sensors, Actua-
tors, and Systems, DSC-vol. 32, pp. 255-268, Atlanta, GA Dec. 1-6, 1991
[2] IJ. Blech, "On Isothermal Squeeze Films", ASME Inl. of Lubrication Technol-
ogy, vol. 105, pp. 615-620, 1983.
[3] T.M. Bloomstein and DJ. Ehrlich, "Laser Deposition and Etching of Three
Dimensional Microstructures" 1991 Int. Can! on Solid-State Sensors and Actua-
tors (Transducers '91), June 1991, San Francisco, CA, pp. 507-511.
[4] R.A. Brooks and A.M. Flynn, "Rover on a Chip", Aerospace America, October
1989, pp 22-26.
[7] L. Fan, Y. Tai, and R.S. Muller, "IC Processed Electrostatic Micromotors", Sen-
sors and Actuators, 1989, v. 20, p.41-47.
[9] G.K. Fedder and R.T. Howe, "Thermal Assembly of POlysIhcon MICroStruC-
tures", IEEE Micro Electro Mechanical Systems", pp. 63-68, Nara, Japan, Feb.
1991
[10] A. Flynn, "Gnat Robots, (and How They WIll Change RobotIcs)," IEEE Solid
State Sensors Workshop, Nov. 9-11, Hyanms, MA 1987.
[17] P. HOroWitz and W. HIll, The Art of Electronics, 2nd &htIon, Cambndge Umver-
Sity Press, 1989.
60 R. S. FEARING
[18] K. Ikuta, "Micro/Miniature Shape Memory Alloy Actuator", IEEE Int. Conf. on
Robotics and Automation, pp. 2156-2161, Cincinatti, OH, May 1990.
[19] CJ. Kim, AP. Pisano, and R.S. Muller, "Overhung Electrostatic Microgripper"
1991 Int. Can! on Solid-State Sensors and Actuators (Transducers '91), June
1991, San Francisco, CA, pp. 610-613.
[20] L. Lin, AP. Pisano, and AP. Lee, "Microbubble Powered Actuator", 1991 Int.
Can! on Solid-State Sensors and Actuators (Transducers '91), June 1991, San
Francisco, CA, pp. 1041-1044.
[21] C.H.T. Pan and P.H. Broussard, "Squeeze-Film Lubrication", Proc. Gas Bearing
Symposium on Design Methods and Applications, pp. 12/1-12/35, Univ. of
Southampton, April 1967
[23] K.SJ. Pister, R.S. Fearing, and R.T. Howe, "A Planar Air Levitated Electrostatic
Actuator System", IEEE Workshop on Micro Electro Mechanical Systems, Napa
Valley, CA Feb. 12-141990.
[24] K.S.J. Pister, M.W. Judy, S.R. Burgett, and R.S. Fearing, "Microfabricated
Hinges: 1 mm Vertical Features with Surface Micromachining", IEEE Transduc-
ers '91, San Francisco, CA June 1991.
[25] K.S.J. Pister, "Hinged Polysilicon Structures with integrated CMOS TFTs", to
appear, Hilton Head 92, Solid State Sensor and Actuator Workshop.
[26] L. Prandtl and O.G. Tietjens, Applied Hydro-and Aeromechanics, Dover, New
York: 1934.
[28] W. Tnmmer and R. Jebens, "Actuators for MICro Robots", IEEE Int. Con! on
Robotics and Automation, Scottsdale, AZ, May 1989, pp. 1547-1552
Probt
Sen&in5 circ.uiuy
" Pl .. ,fo rm
Air hole
MICRO STRUCTURES FOR IMPLEMENTING SUB-MILLIMETER ROBOTS 63
Figure 3. Polysilicon beam with polysilicon staple-type hinge folded out of plane.
Figure 5. The hinge process sequence At top is shown a cross section after poly 1 has
been patterned and the second layer of PSG deposited. Following this, contacts are
etched through both layers of PSG. Next, poly-2 is deposited and patterned. Finally, the
oxide is removed in a sacrificial etch, and the poly-! ler is free to rotate out of the plane
of the wafer.
Figure 6. Three basic hinge types (A) A substrate hinge, which is used to hinge released
structures to the substrate. (B) A 'concave down' scissor hinge, used to hinge released
structures to each other. (C) A 'concave up' scissor hinge.
• Poly·1
III Poly·2
• Contact
MICRO STRUCTURES FOR IMPLEMENTING SUB-MILLIMETER ROBOTS 65
Figure 8. A parallel plate gnpper. The gnpper IS normally closed, WIth a gap of 100 ~m
between the plates. Pulling the vertIcal bar (left sIde) causes the Jaws to open. Opemng
of 0.5 mllhmeter IS
possIble WIth no damage to the deVIce.
R. S. FEARING
66
beam
MICRO STRUCTURES FOR IMPLEMENTING SUB-MILLIMETER ROBOTS 67
v
Figure 13. Equivalent Crrcuit for Capacitor Motor
l
T
68
R. S. FEARING
D
D
DD D Move Left Move Down
DD D
Move Up Rotate
MICRO STRUCTURES FOR IMPLEMENTING SUB-MILLIMETER ROBOTS 69
(side vie w)
platform
/
(top view)
Platform
P(x,y)
Substrate
R. S. FEARING
70
mobile platforms
deb
Permanent
Magnet
Inspect Ion
statlOn welding station
light source
for power
Micro Robots
moving parts
MICRO STRUCTURES FOR IMPLEMENTING SUB-MILLIMETER ROBOTS
71
Platform
v
Substrate
."
72 R. S. FEARING
Paramecia Tracking
1 pixel = 50 urn
275.0
C/l i,
rrt,--
\ ..,..:---;:.:..",-----,;'----;-,
. .~-\
(1)
x '". I··;:····· .. !; ;-
.0.
225.0 l:.
t:: ., Start I.. ,
o 1./'
:;:; I""'!
. iii
o " .....
a. "ji~~:~~~'
>- 175.0 '; ,
,I'"
.. ~'
L
:1
'
' - - - - - ;.--....-... -
. ::----""- '.,..
"""" ...-- ......
'-::::-::-
:. - " ' ('
-j •• -
125.0
150.0 200.0 250.0 300.0 350.0
x position, pixels
Mobile Vehicles
Electrode
DESIGN. MODELLING, AND CONTROL STRATEGIES
OF A THREE DEGREES-OF-FREEDOM VR SPHERICAL MOTOR
PART I: RELUCTANCE FORCE CHARACTERIZATION
Kok-Meng Lee
Associate Professor
The George W. Woodruff School of Mechanical Engineering
Georgia Institute of Technology
Atlanta. GA 30332-0405
EMAIL: KMLEE@GTRIOl.GATECH.EDU
PHONE: (404)894-7402
FAX: (404)894-8336
Abstract
The research presented here is to establish a basis for the design and control of a
variable-reluctance (VR) spherical motor which presents some attractive possibilities by
combining pitch, roll, and yaw motion in a single joint. For clarity, this paper has been
organized in two parts: The first part describes the characterization of reluctance force for
design. The second part addresses the dynamic modelling and control of the spherical
motor. The reluctance force which is the driver of a VR spherical motor is characterized by
using both the finite-element method and the lamped-parameter approach. The objective of
this paper is to provide a good understanding of the magnetic fields and forces at play for
realizing effective design and control of a VR spherical motor. The permeance-based model,
which is commonly used in the stepper motor community to model the reluctance force of a
step motor, was developed to model the torque of a VR spherical motor. Since the success of
the permeance-based model depends on the assumed shape of the magnetic flux tubes,finite
element methods were used in this study to provide physical insights of the magnetic flUX
patterns and to examine the reluctance force completed using the assumed flux shape. Two
models were computed using the finite-element methods for illustration of the operational
principle of a VR spherical motor.
73
H.S. Tzou and T. Fukuda (eds.), Precision, Sensors, Actuators and Systems, 73-109.
© 1992 Kluwer Academic Publishers.
74 KOK-MENG LEE
1. INTRODUCTION
The research efforts described in this paper have been focused on the creation of an
interesting three degrees-of-freedom (DOF) variable-reluctance (VR) spherical motor which
presents some attractive possibilities by combining pitch, roll, and yaw motion in a single
joint. Due to the ball-joint-like structure without the use of a speed reducer, the VR spherical
wrist motor results in a relatively simple joint kinematic and has no singularities in the
middle of the workspace except at its boundaries. The VR spherical wrist motor has potential
applications such as laser and plasma cutting where high-speed, smooth, isotropic
manipulation of the end-effector is required.
A particular form of spherical induction motor was originally designed, built and
successfully tested by Laithwaite el at. [1-3]. The concept of a spherical motor was later
employed in the design of a rotodynamic pump [4] and gyroscope applications [5-6]. A
spherical induction motor was conceptualized by Vachtsevanos et al. [7] for robotic
applications and the detailed analysis was given in reference [8]. However, realization of a
prototype spherical induction motor remains to be demonstrated. The mechanical design of a
spherical induction motor is complex. Laminations are required to prevent movement of
unwanted eddy currents. Complicated three phase windings must be mounted in recessed
grooves in addition to the rolling supports for the rotor in a static configuration. These and
other considerations have led Lee et at. to investigate an alternative spherical actuator based
on the concept of variable-reluctance stepper motors which are easier to manufacture [9].
Recently, several other design concepts of multi-DOF wrist motor have been reported.
Among them, Hollis et at. [10] has developed a six DOF direct-current (DC) "magic wrist"
as part of a coarse-fine robotic manipulation. An alternative DC spherical motor design with
three DOF in rotation was demonstrated by Kanedo et at. [11], which can spin continuously
and has a maximum inclination of 15·. Although the DC spherical motor is characterized by
its constructional simplicity, the range of inclination and the torque constant are rather
limited. Foggia et al. [12] demonstrated an induction spherical motor of different structure,
which has a range of motion characterized by a cone of 60·. Since the control strategy of the
induction motor [12] has not been reported, the ability of the motor to realize any arbitrary
motions remains to be demonstrated. As compared with its DC counterpart, a VR spherical
DESIGN AND CONTROL OF A VR SPHERICAL MOTOR: PART-l 75
motor has a relatively large range of motion, possesses isotropic properties in motion, and is
relatively simple and compact in design. The trade-off, however, is that sophisticated control
scheme is required.
Variable-reluctance motors are actuated as a result of electromagnetic attraction between
the rotor and the stator poles. The magnetic attraction or the reluctance force is created as the
system tries to minimize the energy stored and reduces the reluctance of the magnetic path.
Lee and Kwan [13] have presented the design concept and developed the theory based on the
local interaction between the adjacent stator and rotor poles to illustrate the concept
feasibility of the VR spherical stepper motor. To allow for a relatively few but evenly spaced
coils for smooth motion control of a VR spherical motor, Lee and Pei [14] developed a
method to examine the influences of the design configurations on motion feasibility.
A good understanding of the magnetic fields and forces at play are necessary to realize
an effective design and control of an innovative VR spherical motor. For this reason,
research efforts have been directed towards the characterization of the reluctance force of a
VR spherical motor. The analysis presented in this paper was performed by using both the
finite-element method [15] and the permeance-based model [16]. The permeance-based
model, which is commonly used in the stepper motor community to model the reluctance
force of a stepper motor, was developed using a lumped parameter approach to allow ease of
implementation of the torque model. The permeance-based model has potential uses in
design optimization, dynamic modeling, and motion control of a VR spherical motor. Unlike
the previous study [16] where the perrneance-based model has been developed for a single-
axis VR stepper, a detailed study on the reluctance force characterization is presented here for
a three DOF VR spherical motor. Extends to one and two DOF VR motors are relatively
straight forward. Since the success of the permeance-based model depends significantly on
the assumed shape of the magnetic flux tubes, finite-element methods were used in this
investigation to provide additional insights of the flux patterns on the torque generation and a
means to examine the reluctance force computed by using the assumed flux path. As it will
be demonstrated later, unlike spherical motors of other types where the solutions to the
forward and inverse dynamics are generally unique, the VR spherical motor is characterized
76 KOK-MENG LEE
by its multiple solution to the inverse dynamics. The flexibility to control the multiple coil
excitations allow an optimal input vector to minimize a pre-selected cost function.
The remaining paper is organized as follows: Section 2 describes the basic structure and
the operational principle of a VR spherical motor. Several methods of modeling the
reluctance of a VR spherical motor are discussed in Section 3 and 4. Section 5 presents the
results of several finite-element models and the comparison between the permeance-based
model and the finite-element method. The conclusions are made in Section 6.
f =dWm=_~1~12dR
dx 2 R dx (VI - V2)
2 (1)
78 KOK-MENG LEE
where R is the reluctance at the airgap. Equation (1) illustrates that the reluctance force
prediction of VR motors requires the reluctance model and its derivative with respect to the
displacement x for a given potential difference to be characterized.
Figure 3 illustrates a differential flux tube of cross section ds and length I between the
two equipotential surfaces of magnetic potentials, V 1 and V 2' The magnitude of the field
intensity H is
(2)
where I is the length of the flux tube. The differential flux d~ is /loHds where /lo is the
permeability of air. The total flux ~ flows through the airgap I between the two
equipotential surfaces is the integral of d~ over the entire equipotential surface S:
(3)
(4)
Jds
110 s f
(5)
p = 110 I y.
S
ds (6)
DESIGN AND CONTROL OF A VR SPHERICAL MOTOR: PART-I 79
Equation (6) shows that the penneance is a function of geometry and that the computation
requires the knowledge of the flux tube (i.e., S and I).
When the airgap is much shorter than the dimensions of the adjacent pole faces, the
magnetic flux ~ is constrained essentially to reside in the core and the airgap, and is
continuous throughout the magnetic circuit. Thus, the permeance of the airgap can be
approximated as
J.loAo (7)
p=-
g-
where Ao is the overlapping area and g is the shortest distance between the two overlapping
poles respectively. The assumption implies that a zero overlapping area corresponds to a
zero flux, and that the flux density distribution in the overlapping area is unifonn.
The penneance model given by Equation (7) does not have a continuous derivative at
x=O and thus, it poses some difficulties in computing the torque at x=O. In practice, the
magnetic field lines bulge outward somewhat as they cross the airgaps. The effect of the
fringing fields is to increase the effective cross-sectional area Ao and results in a higher
permeance than that computed using Equation (7). Typical experimentally permeance
models generally display a continuous derivative at x=O [13]. Equation (7), however,
provides an effective means of estimating the permeance without a detailed knowledge of
flux pattern, when the airgap is much smaller than the dimensions of the overlapping area.
Typical plots comparing Equation (7) against the experimentally detennined permeance and
its derivative are given in Figs. (4) and (5).
One of the most commonly used techniques to account for the fringing effects in
modeling the permeance is to assume the shape of the flux tubes in Equation (6). A typical
flux path for a given rotor position with respect to the stator coordinate frame is shown in
Fig. (6) where the poles of a spherical motor are shown in conical shape for the simplicity of
illustration. The flux path from one potential sUlface is connected to the nearest surface by a
straight line and/or a circular arc such that the flux enters into or emerges from the iron
surface perpendicularly and does not cross other flux paths.
80 KOK-MENG LEE
In order to compute for the permeance of the VR spherical motor, the airgap that
separates the rotor from the stator is divided into elements; namely, M elements in 9
direction and N elements in 1/1 direction. The area of the (I, k) element on a spherical surface
where I and k denote the indices along the 9 and 1/1 respectively, at the location defined by
the spherical polar coordinate (9, .p, D(2) [18] is given by
(8)
where S1/I denotes the trigonometric sine function of the angle 1/1, and 01/1 and 09 are the
incremental angular displacements in the 1/1 and 9 direction respectively. For the flux path
assumed in Fig. 4, the permeance of each element is calculated by using Equation (6), which
yields
(9)
where I 0 and I 1 are the arc lengths between the airgap element and the nearest stator and
rotor poles, respectively. Let C/k (9,1/1) and C rj (9 rj , 1/Irj) be the position vectors of a
particular air gap element and the jth rotor pole respectively. The angle between these
(10)
The position vector of the jth rotor coil with respect to the stator coordinate frame is defined
by
1 JXYZ 1 123
where [T] is a homogeneous transformation desclibing the rotor frame 123 with respect to the
stator frame XYZ, and Crj describes the position vector of the jth rotor pole with respect to
the rotor frame. Similarly, the angle between the position vectors of the stator pole and the
(12)
where C si ( 8 si' 1/1 si) is the position vector of the ith stator pole. Since the stator poles and the
airgap elements are fixed with respect to the stator frame of reference, only the arc length 11
varies with the rotor position.
The arc length I 0 between the airgap element and the nearest stator pole is given by
Equation (13):
eo 1 g 1t sgn( ~s - as) + 1
D =2" (1 + 0)(2 + ~s - as) tan (~s - as) 2 (13)
where as is the half cone-angles of the stator. Similarly, the arc length 11 between the
airgap element and the nearest stator pole is given by Equation (14):
fl 1 1t sgn(~, - a,) + 1
D ="2 (2 + ~s - a,) tan (~, - a,) 2 (14)
3. No magnetic flux leakages between the adjacent stator coils, between the adjacent
rotor coils, or in the system. The assumption implies that the spacing between
any adjacent rotor (or stator) poles is assumed to be much larger compared to the
airgap.
82 KOK-MENG LEE
4. The coil excitations are such that there are only attraction between the rotor and
the stator poles. Repulsion between the stator and rotor poles generates
significant leakage fluxes. This assumption is also a necessity in order for the
assumption (3) to be reasonably stated.
As discussed in the previous subsection, the airgap that separates the rotor from the
stator is divided into MxN elements. Since the flux paths through the element are assumed
known, the permeance of each airgap element may be calculated from Equation (9). The flux
flowing through the airgap is determined by using magnetic circuit analogy. Note that the
flux is assumed to flow through the shortest path between the two overlapping pole faces.
The assumption implies that the permeance of the airgap where the two poles are overlapped
is given by Equation (7) and the permeance of the fringing flux is accounted for by using
Equation (9). A complete derivation of the overlapping area between any two overlapping
circular poles in a spherical coordinate frame is given in Appendix A.
An equivalent magnetic circuit of a VR spherical motor is given in Fig. 7. The magnetic
flux il>ij flow through the airgap between the ith stator pole and the jth rotor pole can be
determined from Equation (15):
where Pij is the permeance between the ith stator and /h rotor poles; Msi and M rj are the
magneto-motive-forces (mmf's) of the ith stator coil and jth rotor coil respectively, and the
magnetic potential at the rotor core. Since
m n
L L <1>··=0, (16)
i=l j=l IJ
the magnetic potential V can be derived by substituting il>ij from Equation (15) into Equation
(16), which lead to
m n
.L .L Pij (Msi + Mr)
V = 1=1 J=1 (17)
m n
L L p ..
i=l j=1 IJ
DESIGN AND CONTROL OF A VR SPHERICAL MOTOR: PART-l 83
Thus, the magnetic core potential V can be determined once the permeance of each element
is computed. The flux flowing through each of the airgap elements can be determined by
using Equation (15) where ~ij and Pij are replaced by ~I k and PI k' respectively.
3.3 Torque Prediction
Knowing the permeance and the flux, the reluctance force components at each of the
airgap elements can be determined from Equation (18)
(I8a)
and
(I8b)
where F tjJ and F 8 are along the directions of tjJ and 8 with respect to the local coordinate
attached at each of the elements, respectively. Note that the rotor is constrained by the stator
and thus, the radial component of the reluctance force does not contIibute to the motion. The
computational procedure of the reluctance force is summarized in the flow chart given in Fig.
8.
The resulting torque contributed by all the airgap elements with respect to the stator
coordinate frame of reference is
(19)
Unlike the conventional single-axis stepper motor, which is constrained physically to rotate
about one axis and thus only one force is necessary to actuate motion in that direction, the
spherical motor has infinite number of rotational axes and has three degrees-of-freedom.
Two non-colinear torque components, the spinning torque, T s and the rolling torque, Tr
are required to control the three degrees-of-freedom orientation of the VR spherical motor.
The rolling torque Tr allows a point on a rotor surface to be actuated in any direction on the
inner surface of the stator and thus, provides two DOF motion control. The third DOF
motion control is provided by the spinning torque to result in the spin motion about the rotor
84 KOK-MENG LEE
shaft. These torque components are given by Equation (20) with respect to the rotor
coordinate frame:
(20)
and
ITsl = IT. ~1.
ITrl = IT-Tsl,
/\ Tr d/\
er = ITrl an ey = !b
A A
where e'Y which is pointing away from the center along the rotor shaft and e r perpendicular
A
to e'Y'
V.B=O, (21)
VXH=J, (22)
where B is the flux density; H is the magnetic field intensity; and J is the current density.
Equation (21) states that the magnetic flux lines are sourceless at any point in the field.
Equation (22) states that the circulation of the magnetic field at a point is due to the existence
of current with the density J at that point. In addition to these equations, the constitutive law
that describes how the physical properties of the materials affect the field and source
quantities is given by
B = fJ. H. (23)
DESIGN AND CONTROL OF A VR SPHERICAL MOTOR: PART-l 85
where II- is the permeanability of the material. The magnetic field can be numerically solved
from the Maxwell's equations using finite element method. In order to do so, the Maxwell's
equations are formulated into a Poisson equation. Two particular methods are used in the
finite element formulation; namely, the two-dimension (20) vector potential formulation or
the three-dimension (30) reduced scalar potential formulation. In either case, the solution
process must be iterative since II- is a nonlinear function of the magnetic field strength H.
vX A =B, (24)
and (25)
Using the constitutive law defined in Equation (23) and the definition given in Equation (24),
v X [~ VX AI = J. (26)
In the two-dimensional analysis, where both the current density J and the magnetic vector
(27)
distinguished as the fraction due to the current sources, H s ' and the fraction due to the
(28)
or
v X Hs = J,
and
Equation (28) suggests that Hm is curl free and therefore can be expressed as the gradient of
a "reduced" scalar function V as
(29)
The term "reduced" comes from the fact that V is defined only by H m , not the full field
intensity H which can be expressed correspondingly as
which is in the form of a Poisson equation. The left hand side term Hs can be directly
calculated from the current sources by applying the Biot-Savart law as follows
1 r Jxr (32)
Hs = 41t JD 1?1 dv.
where the domain of integration D is the entire current carrying body, r is the vector pointing
from the differential current carrying body to the point where the Hs is to be calculated, and J
is the current density at this differential body.
4.3 Reluctance Computation
The finite element solution yields the nodal potential values and the average elemental
flux density values. With the knowledge of the fields at hand, the flux through a surface can
(33)
DESIGN AND CONTROL OF A VR SPHERICAL MOTOR: PART-I 87
where !!i' !2i and toS i are the flux density at the centroid, the unit normal vector, and the
surface area of the ith element. The reluctance is computed by using the finite-element
method with Equation (5). Once the reluctance model is determined, as a function of the
displacement, the reluctance force between the stator and the rotor poles can be computed by
using Equation (1).
determines the input currents (or mmf's) required to obtain the desired torques are needed.
The prediction of the forward torque is given by Equations (18)-(20), for a given set of coil
excitations, Msi and M rj , where i=l, 2, .. , m andj=l, 2, .. ,n. The inverse model which is
rewritten on the basis of Equations (18)-(20) consists of three non-linear quadratic equations
in the form:
m n
T] L L (Xij Ii I j (34a)
i=! j=!
m n
(34b)
T2 = L L ~ij Ii I j ,
i=! j=!
m n (34c)
T3 = L L YiJ I, Ij ,
i=! j=!
where the coefficients (Xij' i3ij and 'Yij are functions of the permeances and their derivative.
The coefficients (Xij' i3ij and 'Yij are functions of displacements for a given geometrical
structure.
For m stator coils and n rotor coils, the solution to the inverse model may be determined
from the three non-linear quadratic equations with m+n unknowns. This interesting feature is
distinctively different from the popular three-consecutive-joint wrist based on traditional
single-axis-motors on the spherical motors of other types, which are typically characterized
88 KOK-MENG LEE
by their unique solutions to both their forward and inverse torque models. The nature of the
multiple solutions to the inverse model allow an optimal choice minimize a specified cost
function. To illustrate this unique feature and to provide additional physical insights to the
operational principle of a VR spherical motor, a 2D model as shown in Fig. 9 where the
depth is infinite was formulated by using the vector potential method. As it will be illustrated
later, the computational results provide some physical insights to the design and control of
VR motors without the 3D geometrical complications, that are inherent to the VR spherical
motor. In addition, the 2D vector potential method explicitly computes the flux lines of the
magnetic model and provides an effective means to validate the reluctance force computed
using the assumed flux shapes.
The magnetic model is shown in Fig. 9(a) and 9(b) and the finite element model is given
in Fig. 10. The nodal potentials at the boundary planes which are treated at infinity are set to
zero. The static force per unit depth between the stator and the rotor for a given coil
excitation is computed using both the finite-element method and the permeance-based model.
To make the two methods comparable, the product of the current density and the coil cross-
sectional area is chosen such that it yields the desired magneto-motive-force (mm£). That is,
the current density, J = (mmO/A, where J in Ampere-turns/m2 ; mmf is in Ampere-turns; and
A is the cross-sectional area of the coil in m2 . In the simulation, the cross-section area of the
coil on each side is 20mm x 3mm. Thus, 100 Ampere-turns are equivalent to 1.66E+6
Amperes/m2 . Two different values of iron permeanability were used, namely, JJ./ JJ.o = 1E+3
and 1E+7.
TABLE 1 summarizes the computation result where the percentage error of the
reluctance forces listed are relative to that computed using the 2D finite element method. A
few selected flux pattern computed using the finite-element method is given in Fig. 11. The
excitation is indicated as positive if the mmf is directed toward the air gap. The magnitude of
each excitation is 100 Ampere-turns. Except for Case (5), the permeance-based model yields
relatively good approximation when the magnetic materials has a very high permeability.
There are three major assumptions which may accumulate significant errors in the
permeance-based model:
1. The model assumes no reluctance in the iron core.
2. The model neglects leakage paths.
DESIGN AND CONTROL OF A VR SPHERICAL MOTOR: PART-l 89
j=6.25
.The contribution of the first source may be inferred by running a modified finite element
model using an iron core of very high permeability. As shown in TABLE 1, the relative
errors in Case (1) are 35.3% and 15.5% with the iron permeance of lE+3 and lE+7,
respectively. The decrease in the relative error can also be observed in all cases, which is
consistent with the assumption made in permeance-based model that the reluctance of the
iron is negligible or the permeability of the iron is infinite. The relative contribution of the
second and the third sources to the error is not as obvious. However, the assumption of no
magnetic leakages in permeance-based model implies that the flux would generally flow
through the excited coils and returns through the remaining poles on both side of
electromagnetic structure. As shown in the flux pattern computed for Case (5) demonstrates
that the inaccurately assumed flux path may result in a relative error over 90%. Thus, the
magnetic flux path should be selected to reduce un modelled flux leakages. Repulsions
between the stator and the rotor poles, which generate significant leakages, should be avoided
in permeance-based model. To comply with the assumption of no repulsion between poles,
there is a need for a procedure to check the relevance of the solutions in the permeance-based
model of a VR spherical motor. Alternatively, the rotor poles may be designed to be
constructed of magnetic materials of high permeanability but to have no electromagnetic
coils or permanent magnets. The elimination of coil excitations in the rotor also reduces the
complexity of the design and simplifies the inverse model for real-time control of the VR
spherical motor.
As illustrated in case (4) where the input excitation is tripled, the reluctance force
increases by a factor of four as compared to that of case (1). The significant increase of the
reluctance force is a direct result of a well-shaped magnetic flux path which not only utilizes
all the rightward force generating airgaps, but also effectively eliminates the magnetic flux
from flowing through the airgaps conu'ibuting to the generation of leftward forces. The VR
spherical motor allows the input power to be distributed among several poles each of which
contributes a fraction of the total mmf and thus requires a relatively low CU1Tent per coil but a
large surface area for heat dissipation.
90 KOK-MENG LEE
It is interesting to note that there are infinite combinations of coil excitations for a given
power input. Thus, the solution to the forward torque model may have multiple solutions if
only a total input power is specified. This interesting feature is demonstrated in cases (1), (2)
and (3) for generating the rightward force and cases (5) and (6) for generating the leftward
force. As an example, case (1) results in a much higher reluctance force than that of cases (2)
and (3) for the same power input. Thus, one would expect that there exists an optimal set of
coil excitations which offers a specified torque with minimum power. In other words, an
optimal solution may be generated by minimizing the input electrical power needed to
generate the desired torque, which is proportional to the square of the current. The optimal
solution to the inverse problem is thus, a standard problem of constraint extrema.
Introducing three Lagrange multipliers A!, ! = 1,2,3, the optimal input is the solution to the
system of (m+n+3) simultaneous equation of the fonn:
(35)
where k=1,2, ... ,m+n
and M N
T = L L (ex 'J I·' I)
, . 1 . 1 J
1= J=
where the energy cost function E is given by
M 2 N 2
E = L I, + L Ij , (36)
i=l j=l
and Ii and Ij are the input currents to the ith stator coil and the jth rotor coil respectively.
5.2 Three-dimension Finite-element Model
The 3-D reduced scalar potential formulation, takes into account of the flux distribution
in the third dimension and thus gives a more accurate solution than the 2D model. Thus,
despite the complexity of modelling, a 3D model was used to solve for the fringing flux
distribution and a typical permeance model. Note that the leakage flux has no influence on
the permeance model which is only a function of geometry.
The following assumptions are made in order to reduce the complexity of in the 3D
finite-element computation:
1. The pole sizes are assumed to be small as compared to the spherical surface of the
rotor such that the pole face can be assumed to be planar.
DESIGN AND CONTROL OF A VR SPHERICAL MOTOR: PART-1 91
2. The ratio of the distance between any two adjacent pole boundaries to the airgap
spacing is of the order of ten and thus, a local magnetic model which consists of a
rotor pole and a characteristic set of stator poles may be treated as isolated from the
rest of the spherical motor.
As an example for the design configuration given in Fig. 2, the equivalent planar model
may consist of a rotor pole and three stator poles for the finite-element analysis. The 3D
model is shown in Fig. 12, where the location of the three stator poles form an equilateral
triangle, the shape corresponding to one of the faces of an icosahedron. Since the magnetic
model is assumed to be isolated from the rest of the VR spherical motor, the cylindrical
boundary surface satisfies the Dirichet boundary condition and the nodal potentials on the
boundary are zero. The top and bottom boundaries of the model vary from one excitation
configuration to another.
The stator coils were replaced by permanent magnets in the finite-element analysis in
order to simplify the air gap reluctance calculation. Permanent magnets, however, have
different magnetic properties from the core material, the magnetic field in the air gap would
be different from that of an air gap bounded by two iron cores. Thus, a 3-millimeter-thick
iron cap is placed on the magnet as illustrated in Fig. 12 so that the magnetic field solution
around the air-gap is independent of the type of excitation.
Three different cases of excitation denoted as up-none, up-up, and up-down are
illustrated. In all three excitations, stator pole 1 was energized such that the flux through
stator pole 1 would flow towards the rotor and stator pole 3 was unexcited. The excitation of
stator pole 2 and the boundary conditions of the top and bottom surfaces for the three cases is
given as follows:
Up-none: Stator pole 2 is not excited. The nodal potentials of the bottom and top
boundary surfaces are set to zero, in which case the flux flows
perpendicularly to the bottom and top sUifaces.
Up-Up: Stator pole 2 is energized to have the same polarity as stator pole 1.
Boundary conditions are set as the case of up-none.
Up-down: The polarity of stator pole 2 is in the opposite direction of stator pole 1. As
the flux is expected to circulate between stator poles 1 and 2 via the rotor
pole, the fluxes in the back irons of the stator and the rotor are assumed to
be parallel to the bottom and upper boundary surfaces. Thus, the top and
bottom boundaries satisfy the Newmann boundary conditions.
92 KOK-MENG LEE
The computed results are tabulated in TABLE 2 where the useful flux is integrated over
three regions. Region 1 is the overlapping area between the rotor and stator pole and the flux
flowing through is denoted as 4>1' Region 2 is the rest of the planar surface of the rotor pole,
and region 3 is the cylindrical surface surrounding the rotor pole. The flux flowing through
regions 2 and 3 are denoted by 4>2 and 4>3 respectively. The fringing flux is the sum of the
flux flowing through regions 2 and 3, i.e. 4>f = 4>2 + 4>3' The above finite-element
computation leads to the following observations:
1. The magnetic flux flowing through the overlapping area is uniform and the
corresponding reluctance of region 1 can be closely estimated using Equation (7).
2. The fringing flux accounts for over 25 percent of the useful flux in the air gap.
Therefore, the negligence of the fringing flux under-estimates the flux and the
reluctance of the airgap.
3. The percentage of the fringing flux was found to increase from 26.7% to 31.7% as the
air-gap spacing was doubled. However, it was determined the increase of the rotor
pole length from 6 mm to 30 mm (corresponding to an aspect ratio defined as the
length-to-diameter ratio of 0.2 to 1) did not have significant effect on the fringing flux.
This is because the fringing flux mainly distributes around the end of the rotor pole.
4. The finite-element computation has shown that when the pole-separation to air-gap
ratio is greater than 10, the fringing flux in an air-gap is relatively insensitive to the
excitation of the adjacent stator pole. The result reasonably justifies that the magnetic
model as shown in Fig. 12 may be treated as an isolated section from the rest of the VR
spherical motor in the finite element analysis.
5. The finite-element results have shown that the magnet surface is not equipotential
without the iron cap which apparently serves to provide an unifonn flux distribution at
the air gap. The influences of the iron cap on the magnetic flux and reluctance are
compared in Table 3, where the average potential value has been used in computing the
air-gap reluctance. Due to the non-uniform potential at the magnetic stator pole face,
the reluctance of the overlapping area is about 7% smaller than that computed using
Equation (7) under all three excitations. The percentage of the fringing flux, however,
is not remarkably sensitive to the presence of the iron cap.
6. CONCLUSIONS
The theoretical basis of modeling the reluctance torque which is essential to the design
and control of a VR spherical motor has been presented. Several methods of modeling the
penneance on the reciprocal of the reluctance have been discussed.
DESIGN AND CONTROL OF A VR SPHERICAL MOTOR: PART-l 93
ACKNOWLEDGEMENT
This research is supported under the National Science Foundation grants DMC-8810146 and
DDM-8958383. The author wishes to thank Uri Gilboa and lianfa Pei for their
computational assistance.
94 KOK-MENG LEE
REFERENCES
1. William, F., Laithwaite, E., and Piggot, L., "Brushless Variable-Speed Induction
Motors," Proc. IEEE, No. 2097U , June 1956, pp.1 02-118.
3. Laithwaite, E., "Design of Spherical Motors," Electrical Times, Vol. 9, June 1960, pp.
921-925.
4. Laing, I. and Laing, N. Patent U.S. 4352646, Rotodynamic Pump with Spherical Motor,
October 5, 1982.
7. Vachtsevanos, G., Davey K., and Lee, K.-M., "Development of a Novel Intelligent
Robotic Manipulator," Presented at the 1986 IEEE International Conference on
Systems, Man, and Cybernetics, Atlanta, GA October 10-17, 1986. Also in Control
Systems Magazine, June 1987, pp. 9-15.
9. Lee, K-M., Vachtsevanos, G. and Kwan C-K., " Development of a Spherical Stepper
Wrist Motor" Proc. of the IEEE 1988 International Conference of Robotics and
Automation, Philadelphia, PA, April 24-29, 1988. Also in Journal of Intelligent and
Robotic Systems 1 (1988) 225-242.
10. Hollis, R. L., Allan, A.P., and Salcudan, S., "A Six Degree-of-freedom Magnetically
Levitated Variable Compliance Fine Motion Wrist," Proc. on the International
Symposium on Robotics Research, Santa Cruz, August 1987.
11. Kaneko, K., Yamada, I. and Itao, K., "A Spherical DC Motor with Three Degrees-of-
freedom," ASME Trans. Journal of Dynamic Systems, Measurement and Control, Vol.
111, Number 3, September 1989, pp. 398-402.
12. Foggia, A., Oliver, E., Chappnis, F. and Sabonnadiere, J., "A New Three Degree of
Freedom Electromagnetic Actuator," Conference Record - lAS Annual Meeting, Vol.
35, No.6, Published by IEEE, New York, NY, USA; 1988, pp. 137-141.
DESIGN AND CONTROL OF A VR SPHERICAL MOTOR: PART-l 95
13. Lee, K.-M. and Kwan c.-K., " Design Concept Development of a Spherical Stepper
Wrist Motor," IEEE Journal of Robotics and Automation, Vol. 7, No.1, February 1991,
pp.175-181.
14. Lee K.-M. and Pei, J. "Kinematic Analysis of a Three Degrees-of-freedom Spherical
Wrist Actuator," Proc. of Fifth International Conference on Advanced Robotics, Pisa,
Italy. June 20-22, 1991.
15. Sylvester, P. and Ferrari, R. L., Finite Elements for Electrical Engineers, Cambridge
University Press, New York, 1986.
16. Chai, H. D."Permeance Model and Reluctance Force Between Toothed Structures,"
Theory and and Applications of Step Motor edited by Kuo, B.C.; West Publishing Co.
1973.
17. Smith, D. E., Essentials of Plane and Solid Geometry, Wentworth Smith Mathematical
Series, 1923.
APPENDIX A
[T(e)] = [1 0 0]
0 cos e -sin e (A.i)
o sin e cos e
In the following derivation, the coordinates are written with respect to the X-Y -Z frame.
Using the transformation matrix given in Equation (A.1) the curve e2 can be described by
The position vectors of the two intersecting points are PI and P 2 which are symmetrical
about the YZ plane and can be derived by solving Equations (A.2) and (A.3) simultaneously.
R
-:--9 [(sin Iftl sin 9)2 - (cos ~ cos 9 - cos 1ft2i]lf2
sm
(AA)
and point P2 is a mirror image of point Pl' If a plane is defined to pass through PI and P2
and the origin of the sphere, the plane would divide the overlapping area S into two parts, S 1
and S2' Let the intersecting contour be denoted by the curve e3 and the angle between plane
oPIP2 and the XZ plane be p. The intersecting contour e3 is derived with respect to X-Y-Z
frame as
(A.5)
where
From the projection of the curves ei' i = 1,2, and 3 on the XY plane given in Equations
(A.2), (A.3) and (A.S), the overlapping area is computed from the following integral:
s, -- JD, J-J 2
R
2 2
dx dy i=I,2 (A.7)
R -x -y
where Si and Di are the areas bounded by ei and e3 on the spherical surface and the
corresponding projections on XY plane respectively. By carrying out the integration, the
overlapping area is found to be
2
S = L S, (A.S)
1=1
where
S, = R2{[1 + Sgn(cos ~ cos 9 - cos ~+I)] (1 - cos "',) 1t + 2 Sgn(cos ~ cos e cos "',+1) (A.9)
[.1, cos if, - sin -I (cos r, sin .1,)] }
98 KOK-MENG LEE
Ll; = tan-I (
-l(Sin V'i sin 9)2 - (cos if; cos e- cos Yt;+li) (A.ll;
cos if; cos e - cos 'f;+1
where "'3 = "'1 and the function sgn(x) is 1 if x is positive or -1 if x is negative.
* E(l) and E(2) are the relative error of calculated reluctance forces using
permeance-l:iased model with respect to that computed using finite-element model
with Il/Il a = 1E+3 and 1E+7 respectively.
DESIGN AND CONTROL OF A VR SPHERICAL MOTOR: PART-l 99
Shaft - ____ ~
Rotor
Slid i ng Block
X-guide Stator
Stator
Coil
Bearing Ro t or Pole
S I rder
I t1X)
St Bto~ Pole
1 '"
0,4 1
-,
" , I
0,5 L ~
1 -"-- "
OA ~
0
".
i
". 1
0.3 ~ J
i I
0,2 1- I
I
0.1 .-
"
°o~------~----------------~--------~--~~----------
0,2 0.4 0,6 0 ,8 1 1.2
XIO·S
.o;~
E
i
.::
.~
..:~
'::l \
~
u
u
~
.2:t \
,r
t)
E -3.5 -l
~
-4 :..
_ _ Experimental data I
Derivative or
-4 ..5 - Equation (7)
-5 ~·-------------~----~~------~----~
o 0.2 0.4 0.6 0.8 \.2
/
/
/
/
/
V
V
I
J;S1 L S2 IL si ~sm
T T 'T T
~r1 ~r2
'ij~~ ~rn
~ Mrj
-I T T T
I
Fig. 7 Magnetic Circuit of the Spherical VR Motor
...
F10d the nearest stator COli to each arrgap
element and compute the flux path length to
that COli
~
Measure rotor displacement and f10d the
coordmates of each rotor COli
...
For each alrgap element find the nearest
stator and compute the flux path length to
that COli
~
Solve rotor equauon of For each mrgap element compute the
mouon permeance and the permeance denvatlve
I I
t
Define COlis eXCllauons Construct
eqUivalent magneuc CirCUli to find flux
denslty 10 each element
+
Compute force of each element Sume
mgap forces to find total force
----- ROTOR
6mm -
STATOR
90mm - - - - - - - - - -
':1~
~ :11
:V
«l 0 , 8 .... 1
II
) 6 ~I
: • -I
:": 1L
' _ - ' - -_ __ _ -:---::--'"---::;-_-;:---1-
~ ~ ~ ~ ~ a
" (.IlOIlO AI
STATOR ROTOR
14
'I 11
6 t.\
j-
I~
I
I
~
1=
II
It
5 10
" " ~
I ~
1==
~
10
lO
r:=
4 i=
q 9
Iq
IF=
I
I
I
3 I
1 't
8
Ii - - -
11
6
2 t===-
~
5
16
f=.=
I -- :::
I
, - 7
1 6
LINEAR
STEPPING MOTOR
IIIIIII II IIIIIII IIII II IIII III III IIII II IIII III IIIIIIIII III IIII ! IIIIII I!I lill lll :11
(a) Case 1
:: I!' : :1. '''1 ! 1 i 'I'I !!! I IIII ~ l'I' I I, !' :' II IIII1 II1 I'1 V : 'I'I !:I I I I I I I ~
:,-, ~ - ,~I ,-
, - ii
'
, i ' I I
; I! I '
-~ - ---- - '-,-, ur.;
I
,I,I
:
;,
11
"I T''
I
! : ! 1
~ -- -- I
I
II NIII II
"--------'
1llllll lll lI iil lllllllll lllllllll lll! lilll!11 111111 1111 111I 1111IIII IIINll lil liV
(b) Case 2
II I N1 1 1 1 11 ~ !l1I II! II I! II II! II III! I!III! I! II IIIIIIIIIIII IIII IIII ! 111 111 ! IIII ~ I !
(c) Case3
II: II IIIIIIIII :IIIII II;IIII IIIIII ! I:1 :III :Ill, Illt llllll q ill I:1' Ii 111111 N !'
II! ..I:.H'(~~=:::::
I I~II
1 ~ !1I 1 11 !II 1'1111 ! IIII ! I!I!I! I!11 111111111 111111111 I'! Ii III I ! IIII I~ I !
III N! 1111
(d) Case 4
(e) Case 5
11I1 1 11 1 1 1111 1 1 1 1 : 1 ' llI , llll i l l l l l ilI Il I I L"lI ; I : I I I N II I I III I I III V1ItI Il I I : i ! lI ~
I I I I I:I~§~~~~¥J
1111 11
IIIIIIIIIIIII II!I IIIIIIII IIII II III !IIIIN III!I:III IIl IIII IIII III! III! IIIII
II I!'I ':! '
---- ----_ . . - - ... . .. .. . _ .. . . . . .... .. . __.. __.. .... ..... .. _- _. ...... . . .. .. ... .
(0 Case6
I11111111111 1111 1111111 111111111 ~ I IIII I I III I II IIIII I I I I I I I II IIII I i 'II 'III , 1II1 1I ,
11111
~';';';';'~:::::::::::':':::':::....:.J. !:I
!11 11I 1/!1 !l i! III!III IIIII II IIIII I!I! I!I! I! I! !l!III !I!II I! 11I! I!!I I!I !II ! I!II!:! '
(g) Case 7
rotor bock ,f on
I
i l
A
permane n t
magne t
b c
by
Kok.Meng Lee
Associate Professor
The George W. Woodruff School of Mechanical Engineering
Georgia Institute of Technology
Atlanta, GA 30332-0405
Abstract
This paper presents the dynamic modeling and the control strategy of an interesting three
degrees-offreedom (DOF) variable-reluctance (VR) spherical motor which presents some attractive
possibilities by combining pitch. roll. and yaw motion in a single joint. Both the forward dynamics
which determine the motion as a result of activating the electromagnetic coils and the inverse model
which determines the coil excitations required to generate the desired torque are derived. The model
represents the first detailed study on the inverse dynamics. and yet. permits a spectrum of design
configurations to be analyzed. The solution to the forward dynamics of the spherical motor is unique
but the inverse model may have multiple solutions and therefore an optimal choice is required. The
multiple-coil excitations allows an optimal control vector to be chosen to minimize a specific cost
function. This characteristic significantly differs from that of a popular three-consecutive-rotational-
joint wrist based on the traditional single-axis motor or spherical motors of other types. which are
typically characterized by having an unique solution to both the forward and inverse dynamics. The
control strategy of a VR spherical motor consists of two parts .. namely. the control of the rotor
dynamic in terms of the actuating torque. and the determination of the optimal inputs for the required
torque. An optimal choice is determined from an unconstrained optimization problem. The
implementation issues in determining the optimal control input vector in real-time are also
addressed.
III
H.S. Tzou and T. Fukuda (eds.), Precision, Sensors, Actuators and Systems, 111-138.
© 1992 Kluwer Academic Publishers.
112 KOK-MENG LEE
1. INTRODUCTION
An increasing need for high performance robotic applications has motivated several
researchers to direct their investigation efforts to new actuator concepts to improve the
dexterity of robotic wrists. Examination of the existing mechanical joints reveals that the
ball-joint-like spherical actuator is an attractive alternative to the three consecutive-rotational
joint configuration. The interest in spherical motor as a robot wrist is triggered because of its
ability in providing the roll, yaw, and pitch motion in a single joint, isotropic in kinematics
and kinetics, and its relatively simple structure. Also, it has no singularity in the middle of
workspace except at the boundary. The elimination of gears and linkages enables both high
positioning precision and fast dynamic response to be achieved by a properly designed
spherical motor. These attractive features have potential applications such as high-speed
plasma and laser cutting where the orientation must be achieved rapidly and continuously
with isotropic resolution in all directions.
Recently, several design concepts of spherical motor were proposed. A spherical
induction motor was conceptualized in [1] for robotic applications and the detailed analysis
was given in [2]. However, it is difficult to realize a prototype of its kind because of its
complexity in mechanical and winding design and manufacturing, which requires inlaying all
three transversing windings on the inner spherical surface of the stator. Laminations are
required to prevent movement of unwanted eddy CUlTents. Complicated three phase windings
must be mounted in recessed grooves in addition to the rolling supports for the rotor in a
static configuration. These and other considerations have led Lee et al. [3] to investigate an
alternative spherical actuator based on the concept of variable-reluctance (VR) stepper motor
which is easier to manufacture. Hollis et al. [4] has developed a six DOF direct-culTent (DC)
"magic wrist" as part of a coarse-fine robotic manipulator. An alternative DC spherical
motor design with three DOF in rotation was demonstrated by Kanedo et al. [5], which can
spin continuously and has a maximum inclination of 15·. Although the control of a DC
spherical motor is relatively simple, the range of inclination and the torque constant are rather
limited. Foggia et al. [6] demonstrated an induction type spherical motor of different
structure, which has a range of motion characterized by a cone of 60·. Since the control
DESIGN AND CONTROL OF A VR SPHERICAL MOTOR: PART-2 113
strategy of the induction motor [6] has not been reported, no results were given on the ability
of the motor to realize any arbitrary motions.
As compared with its DC counterpart, a VR spherical motor has a relatively large range
of motion, possesses isotropic properties in motion, and is relatively simple and compact in
design. The trade-off, however, is that sophisticated control scheme is required. For this
reason, in this paper we discuss both the dynamic model and the control strategy of a VR
spherical motor. The contributions of this paper may be briefly summarized as follows: (1)
An analytical dynamic model of an unique, potentially useful design of a three DOF ball-
joint-like VR spherical motor has been described. The model represents the first detailed
study on both the forward and the inverse dynamics of a VR spherical motor. Yet, the model
permits a spectrum of design configurations to be analyzed; (2) The analysis offers some
interesting insights to the design and control of VR spherical motors. For motion control of
the VR spherical motor, both the forward dynamics which determine the motion as a result of
activating the motor coils and the inverse model which determines the coil excitations
required to obtain the desired torques are needed. The solution to the forward dynamics of
the spherical motor is unique but the inverse model, however, have multiple solutions. The
multiple-coil excitations allows an optimal control vector to be chosen to minimize a
specified cost function. This characteristic significantly differs from that of a popular three-
consecutive-rotational-joint wrist based on the traditional single-axis motor or spherical
motors of other types which typically have unique solutions to both the forward and inverse
dynamics and limited the flexibility of controller designs; (3) The paper also represents the
first attempt to address the control strategy of the VR spherical motor. The formulation of an
unconstrained optimization problem from a standard problem of constraint extrema is
interesting. Along with the discussions of the implementation issues, the application of the
optimization technique to the VR spherical motor is illustrated.
The remaining paper is organized as follows: Section 2 begins with the derivation of the
torque prediction equations. Section 3 presents the motion control strategy of the spherical
motor, which consists of the control of the rotor dynamics and the determination of the
optimal input vector. The real-time implementation of the optimization method is addressed
in Section 4. The conclusions are given in Section 5.
114 KOK-MENG LEE
2. DYNAMIC MODEL
The VR spherical motor referred to in this paper is a ball-joint-like device similar to that
conceptualized by Lee and Kwan [7].
2.1 Structure of a VR Spherical Motor
The structure of the VR spherical motor is shown in Fig. 1 and an exploded assembly
view is given in Fig. 2. The VR spherical motor consists of basically three mechanical
assemblies; namely, a spherical rotor, a hollow spherical stator, and an orientation measuring
system. The spherical rotor is constrained but allows to roll on the bearing gimbals which are
mounted on the inner surface of the stator.
The coils with ferromagnetic cores are evenly located on the rotor and the stator, and
each coil can be energized individually. In order to maintain geometrical symmetry for
simplicity in control, the stator poles and the rotor poles are of circular shape. The rotor
poles meet at the center of the rotor, and the stator cores are connected by the magnetic
conductor layer in the stator shell to form a magnetic circuit with the airgap. For simplicity,
it is desirable that the poles are evenly spaced on the stator and the rotor following the pattern
of regular polyhedrons. Each vertex of the polyhedron corresponds to the location of one
pole. Pythagoras and Plato [8] have shown that a complex polyhedral angle must be made up
of at least three faces and must be less than 360· to form a closed polyhedron. Using these
principles, it can be shown that the maximum number of coils which can be evenly spaced on
a sphere is 20, the figure corresponding to the number of complex angles of a dodecahedron.
The influences of the design configuration on motion feasibility and the methodology of
selecting the patterns can be found in reference [9].
In the operation of the VR spherical motor, the stator coils are energized individually
using the control circuitry. A magnetic field is established which creates magnetic energy in
the airgap. The created energy is the function of the relative position of the rotor and the
stator. The motion of the spherical VR motor is thus generated as the rotor tends to move to
a position such that the energy in the airgap is minimized.
A means of measuring the orientation is shown in Fig. 1. The mechanism consists of
two circular sliding guides mounted on two orthogonal axles attached to the stator. The
DESIGN AND CONTROL OF A VR SPHERICAL MOTOR PART-2 115
sliding guides confine a sliding block which houses an encoder for measuring the spin angle
of the rotor, 9z. When the rotor rotates relative to the stator, the shaft and hence the sliding
block positions the sliding guides. The corresponding angles rotated by the x- and y- sliding
guides are measured by the encoders as 9 x and 9 y . The detailed kinematic relationship
which describes the orientation of the rotor as a function of the three encoder readings can be
found in reference [9].
2.2 Rotor Dynamics
As shown in Fig. 1, a base Cartesian coordinate frame XYZ is fixed at the center of the
spherical stator with the X and Y axes pointing towards the x- and y- encoders and the Z axis
pointing toward the open-end of the spherical socket. Similarly, a coordinate frame 123 is
assigned to the center of the spherical rotor with the 3 axis pointing along the rotor shaft. As
it will be discussed later, the structure of the spherical motor has certain symmetric properties
with respect to Z-Y-Z Euler angles, which are greatly exploited in solving the optimal control
input for a specified torque. Thus, the orientation of the spherical motor is specified using
the Z-Y-Z Euler angles. Given an actuating torque T = [T l' T 2' T 3] T, the rotor dynamic
equations described in terms of Z-Y-Z Euler angles, q=[ Ijt, 9, qJ] T, is given as follows:
~]
Iz
,
1=lx=ly and Iz are the moments of inertia about the principles axes,
and S(.) and C( • ) denote the trigonometric sine and cosine functions of the angle ( • ),
respectively.
116 KOK-MENG LEE
(2)
the magnetic potential V can be derived by substituting ~ij from Equation (2) into Equation
(3), which leads to
DESIGN AND CONTROL OF A VR SPHERICAL MOTOR: PART-2 117
m n
L L P ij (Msi + M rj )
i=l j=l
v m n
(4)
L L p ..
. 1 . 1 1)
1= J=
Governing equations of energy conversion
The torque generated by the electro-magnetic system is derived by using the principle of
conservation of energy.
.
Em(t)
.
= Ee(t) - T(t) • ro(t) (5)
where
Since
(6)
where dl/>x' dl/>y and dl/>z are the infinitesimal changes of angles with respect to the stator
frame. Using the result from Equations (5) and (6) and noting that the differentials of I/>x' I/>y'
and I/>z are independent of each other, the torque generated by the magnetic system is given
by
(7)
where
and i, j, k are the unit vectors along the X, Y, Z axes of the stator coordinate, respectively.
The electrical power input to the system is given by
• m n
Ee = L L (Msi + M rj ) <i>ij (8)
i=l j=1
1 m n 2
E =- ~ ~ <l>ij Rij (9)
m 2 i=! j=!
From Equation (8) and the time-derivative of Em obtained from Equation (9):
• • 1 m n •
Ee - Em =- ~ ~ (Msi + M rj - V)2 Pij . (10)
2 i=! j=!
Combining Equations (1), (6), and (10) and noting that IPx' lP y , and IP z are independent, the
torque T can be represented by Equation (11).
m n
T =-}l'--~! ~ (Msi + M rj - V)2 V Pij ·
(11)
j=!
3. The value of P(x) at the origin (i.e. when the poles are fully overlapped) can be
reasonably well-determined by
For a given geometry, a typical permeance CUlve which satisfies the above conditions is
given as follows:
P(x) (13)
P(x = 0)
DESIGN AND CONTROL OF A VR SPHERICAL MOTOR: PART-2 119
N
P(x) = a" + L ak cos k x , (14)
k=l
where the coefficients (aO' aI' ... , aN) can be determined from experimental data or from
numerical computation. Note that P(x) is an even function and therefore the sine terms
vanished.
2.5 Torque Prediction Model
The torque prediction model determines the torque generated by the spherical motor for
a given set of input currents applied to the electromagnetic coils. For a specified geometry,
the permeance between any pairs of adjacent stator and rotor poles is a function of the angle
between the position vectors characterizing of ith stator and the lh rotor poles, iflij' Hence,
(15)
Let Csi(xsi' y si' zsi) and CrjCxrj' Yrj' Zrj) be the position vectors of the ith stator and the jth
rotor poles, respectively. The angle between any pairs of stator and rotor poles can be
determined from the dot (inner) product of the position vectors C si and C rj ; that is
(16)
where R is the mean radius of a spherical surface separating the pole faces of the stator and
rotor. The position vector of the jth rotor coil with respect to the stator coordinate frame is
defined by
120 KOK-MENG LEE
(17)
where [T] is a homogeneous transformation describing the rotor frame with respect to stator
frame, and Crj describe the position vectors of /h rotor pole with respect to the rotor frame.
From Equations (7) and (11), it can be shown by using differential geometry that the torque is
given by
(18)
where eij is an unit vector perpendicular to the position vectors Csi and C rj and can be
written as
(19)
where Csi x C rj denotes the vector cross-product of Csi and Crj" Thus, Equation (18), along
with Equations (2) and (19) and a permeance model given by Equation (14), defines the
torque generated by the spherical motor for a given set of inputs in terms of the magneto-
motive-forces (mmf's) of the coils.
where V = [V 1/1' V 9' VI/>] T is the control vector. Using the feedback law given in Equation
(20), the closed-loop dynamic equation becomes
Since M(q) > 0 V 9 ;06 0 and therefore M- 1 (q) exists V 9 ;06 0, we have
q= V. (21)
Equation (21) represents a linear system with three decoupled second-order subsystems under
the control vector V and hence linear control theory can be applied to each of the coordinates
separately. As an example, the control vector V may be chosen as
(22)
where
As O":i:- 0, we have
x = ['/I, 0, $, ;, e, $]
'/I
.
(24)
e.
$
f(X, T) = [I z erico + ~) - 2HHco - T) Ccp + T2 Scp] -
ISo
[(I - Iz) ~ (;Co + $) So - ; SSo + T)Scp + T2 Ccp*
• • 2 •• • 1 1 T3
[(Ie '/I (1 + Co) - Iz 0 ('/I Co + $)] !Sa + (T) Co - T2 Scp) Co ISo + 1;-
(25a;
e
';+$
f (X, T) = 1 (25b)
I (T) Scp + T2 Ccp)
1 T3
Iz
If the control torque T is determined by Equation (20), then the right-hand side of Equation
(23) is continuous on D cR 6 x R, a closed, connected and bounded region. Furthermore, it
also satisfies a local Lipschitz condition on D. Then for any (x o' to) fD, where Xo = x(to)' to
is the initial time, there exists a unique solution x( ., to' xo) defined over some interval a < to
< b with (x o ' [a, b]) cD. Moreover, the solution depends continuously on to and xo' This
argument is based on the Picard's Existence and Uniqueness Theorem [12].
DESIGN AND CONTROL OF A VR SPHERICAL MOTOR: PART-2 123
The mapping field f in Equation (23) is not continuous as 9 = 0 and hence, local
Lipschitz condition is not satisfied. In fact, the rotor motion can essentially be described by
the two independent variables 9 and "'+<p at 9 = 0 where the precession and the spin axes are
aligned and measured about the same axis. The vector field f degenerates from R6 in
Equation (24) to R4 in Equation (25). Thus, multiple solutions exist at the singular point of
the Z-Y-Z Euler angles as there are two independent equations with three variables.
By the physical nature of ball-joint-like spherical motor, the solution to Equation (1)
should be continuous. To ensure a smooth motion at (9 = 0), a generalized vector field must
be constructed at (9 = 0) such that the vector field of the closed-loop system is continuous for
the rotor dynamics given by Equations (24) and (25). Since the possession and spin angles at
(9 = 0) can be arbitrarily chosen provided that their sum is uniquely determined by the
location of the body, the control vector V in Equation (21) in state-space representation can
be treated as a generalized vector field for the state feedback control law defined by Equation
(20). With Equation (25a) replaced by Equation (21) at (9=0), the vector field is continuous
over the whole range of motion and satisfies Lipschitz conditions and hence, the unique
solution is ensured. Typical simulation results for a rotor of 76.2 mm diameter with design
model. The solution to the inverse problem is to compute a set of coil excitations, which is
denoted here as a control input vector U, that is required to generate the desired torque T.
Unlike the forward torque prediction model which yields an unique torque vector for a
specified set of coil excitations, there are generally infinite solutions to the inverse problem
of the torque prediction model of a spherical VR motor for a specified torque. For clarity in
illustrating the inverse torque model, the following additional assumptions are made:
1. Only current sources are used and the mmf's of the coil are treated as system input
variables.
2. In practice, it is desired to have no wiring in the moving pmts and thus, only simple iron
cores with no excitation coils are assumed as rotor poles (i.e. M 1j = 0, j = I ,... , n).
124 KOK-MENG LEE
(26)
Hence, using the notations defined by Equations (26) to (28), the torque can be written in
I = 1, 2 3 (29)
where
where (uI' 1=1,2,3) is an unit vector along the axes of the rotor body frame. The matrices
designed larger than three, there are generally infinite numbers of solutions to the inverse
problem.
It is of interest to determine an optimal solution by some guidelines or a criterion, such
as one minimizing of the current amplitude or the consumed power. In other words, the
inverse modal is essentially an optimization problem which may be formulated as follows:
m
Minimize .L lu,l p where p > 0 (32a)
1=1
proceed within the feasible region, the GRG method is not suitable for the problem with
equality constraints. Therefore, the inverse problem (32a) is reformulated as an
unconstrained problem. The equality constraint problem can be converted to the
unconstrained problem in two ways; namely the use of Lagrange multipliers and the addition
of panel terms.
Formulation I
m 3 I
Minimize feU) = I Iud P + M L (2 ur [AI] U - Ti (32c)
i=1 1=1
where the weighting factor M > 0 is generally a factor very large real number. It has been
shown that under some very general conditions, the solution to the problem formulation II
approaches the solution to the original inverse problem (32a) as M .... co. The problem
formulation I is unbounded since A's can be chosen such that the objective function has
arbitrarily large amplitude with minus sign. Therefore, the gradient-based method would fail
to find the stationary points. On the other hand, the problem represented by the formulation
II is bounded below and a global minimal solution exists. The minimization of the functional
Equation (32c) is an unconstrained problem. It has been numerically found that the GRG
method works well in minimizing the functional represented by Equation (32c).
lIIustrative Example
An example is illustrated here by using a design configuration where the stator and the
rotor are arranged at the vertices of an icosahedron and a tetrahedron, respectively. The
coordinates of four vertices of a tetrahedron for an unit sphere are described in TABLE 1.
Similarly, the twelve coordinates describing the vertices of an unit icosahedron are listed in
TABLE 2. However, to allow for the motion of the rotor shaft, only eleven stator poles are
used in the design and pole 0 is irrelevant.
126 KOK-MENG LEE
The characteristic dimensions of the VR spherical motor using in the following example
are summarized as follows: The mean radius of the spherical surface separating the stator
and the rotor pole faces are 38.Imm. The radius of the stator and the rotor poles is I2.7mm
and the airgap separating the stator and the rotor pole faces is 1mm. The permeance model as
a function of the relative displacement between two circular poles was obtained
experimentally. The value Xo in Equation (13) was determined to be I8mm. With the
permeance model and the given pole coordinates, the matrices (AI' 1= 1, 2, 3) are formed.
To compute for an optimal input mmf's for a specified torque at a given orientation, an initial
input mmf's vector is estimated and a local optimal solution is computed by the GRG
algorithm. The global optimal solution is then searched by comparing the objective values of
local optimal solutions.
As a numerical example, the optimal input mmf's which generate the torque T = 1 u3
(N-m) at the rotor orientation at (0,0,0) are computed for two cases. In both cases, p is set to
be 2 so that the electric power is minimized and by choosing M = 10 8 the constraint
equations are satisfied with the relative accuracy of 10- 5. In case (1), all the eleven input
mmf's are independently excited. The values of the optimal input vector is tabulated in
TABLE 3. The minimized objective value is 17.8601 and that the absolute value of the
maximum input mmf is 3.78I6x10 3 Amp-turns.
In case (2), the coil excitations are grouped in pairs so that the number of power
amplifiers are reduced. In each grouping, the coils pointing towards each other along a
diameter are connected in series, i.e. ui = -ui+5' i = 1,2, ... ,5. The optimal solution of case
(2) is compared to that of case (1) in TABLE 3. The corresponding objective value is
46.2825 and the absolute value of the maximum mmf is 3.8087xI0 3 Amp-turns. Clearly,
the additional constraints introduced in case (2) substantially increase the objective value as
compared to that in case (1).
It is worth noting that the input mmf's can be effectively lowered by reducing the
airgap. If the airgap is reduced to O.Olmm, the current amplitudes for the same coils are
about 1/10 of the the about results.
DESIGN AND CONTROL OF A VR SPHERICAL MOTOR: PART-2 127
(33)
where ex, (j and A are constant coefficients to be determined for a specified torque. By
substituting Equation (33) into Equation (30), we have
I = 1, 2, 3, (34)
128 KOK-MENG LEE
where
~ [u:~,u,
T
UtAtUz U:A,U, ]
n1,J 0 thA t U3
T
U3AzU t U3AzUz 0
~ [u::,u, U;A,U, ]
UtAZUz
[B,J 2 UZAZU3
T
U3AzU t U3AzUz 0
[B,J ~ [u~:,u,
U3A3U t
T
U t A3UZ
T
0
U3A3UZ
U t A3U3
,
U;:,U, . ]
The coefficients (ex, (j, 'Y) are computed from a set of three nonlinear simultaneous equations
(34) and thus the required control input vector U can be solved from Equation (33). By
storing the three input vectors U I , U2 , and U3 for a given orientation in the precompiled
table, the torque variables are eliminated and the memory size required in the look-up table is
6nN 3 bytes.
It is worth noting that the need to compute the coefficients (ex, (j, 'Y) from a set of three
non-linear simultaneous equations can be eliminated if six additional constraints are imposed
as follows:
(35)
for I, K * J where I, J, K, = 1, 2, 3
which make all cross terms of [B I ] equal to zero. Therefore, without solving for (ex, (j, 'Y), U
can be directly written as
(36)
DESIGN AND CONTROL OF A VR SPHERICAL MOTOR: PART-2 129
t/J, 8, ~ are the precession, nutation, and spin angles, respectively. If the resolution of the
range of the parameters is r points/radian, the memory size of 2n(r1l")3 bytes would be
required if two bytes are used to represent a real number.
The rotor has four evenly spaced poles arranged at the apices of a tetrahedron. The
position vectors of the rotor poles for an unit sphere are listed in TABLE 1. As shown in Fig.
5, since the three rotor poles, j =1,2,3, are evenly spaced at 211"/3 radians apart at a plane
perpendicular to the axis of the fourth pole (indicated as b in Fig. 5), an input vector U(~)
would generate the same torque about the z-axis of the rotor as that would be generated by
U(~ ± 211"/3) for any particular (t/J,8). In other words, U(~ ± 211"/3) = U(~) for a specified
torque to spin the rotor about its z-axis. Thus, the range of the spin angle required in the
formation of the look-up table is 0 :s; ~ :s; 211"/3.
Fig. 6 shows the location of the eleven stator poles located at the apices of an
icosahedron of unit radius. The position vectors of the stator poles are listed in TABLE 2,
which are evenly spaced at 211"/5 radians apart in two circular planes perpendicular to the axis
of the eleventh pole. Fig. 7 shows the line projections of the stator pole axes on the xy plane.
The space bounded between two adjacent projections is denoted as Sk (k = 1, ... , 10) in Fig.
7, where b is the projection of the end-point of the 4th rotor pole on the xy plane.
Define the notation U(k) to be an input vector U required to generate T when b e Sk.
The range of the precession angle required in the table formulation is such that b f SI. When
b is outside the region S l' the input vector U(k) to generate the desired torque T can be
130 KOK-MENG LEE
deduced from the tabulated input vector U(l) for the same torque by means of the
transformation:
The determination of the transformation [R k ] is separated into two cases; namely, an odd and
an even number of k.
k is odd
Let (1/1, 9) denotes the position of b when b f S 1. The corresponding position of b in S3
is indicated as b' in Fig. 7, which can be written as [1/1 + (2rr/5), 9]. To generate a desired
torque when the position of the fourth pole is at b' f S3 using the lookup table, the input
vector u(1) is determined from the lookup table for the same desired torque at b and the input
vector U(3) is then obtained by shifting the index of the stator coils of the input vector u(1)
in the counter-clockwise direction, or equivalently by means of the transformation [R3 ] = [R]
defined as follows:
0 0 0 o
1 0 000
[R] 0 1 000 (38)
0 0 1 0 0
0 0 0 1 0
In general, for any positions of b f Sk defined by [1/1 + (2rr/5)k, 9], the input vector U(k) to
generate the desired torque T can be deduced from U(l) for the same torque using the
Equation (35) where the transformation [Rk ] is given by
k is even
Similarly, the input vector U(k) required to generate the desired torque T when b e Sk
(k is even) can be deduced from u(10) for the same torque by means of Equations (35) and
(36) or
(40)
DESIGN AND CONTROL OF A VR SPHERICAL MOTOR: PART-2 131
However, for any point b with its position denoted by (1/1, (J, tP) f S l' the point b" is a mirror
image of b about the x-z plane. The position of b" can be written as (-1/1, (J, ~) or (27r--w, (J,
27r~) f SlO' Since the symmetry between Sl and SlO is mirror-like, the mirror image of
the required torque when the position of the fourth pole is at b" f S 10 can be generated in
terms of the input vector U(1) when the position of the fourth pole is at b f S l' The mirror
image of the torque can be produced by the input vector u(10) = [RIO] u O) where
transformation [RIO] = [R'] is defined as follows:
1 0 000
0 0 0 0 1
[R'] 0 0 0 1 0 (41)
0 0 1 0 0
0 1 0 0 0
Denote the mirror image of the required torque T (when b" f SlO) as Te (when b f S1)
which is given by
0]
olOT. (42)
o -1
In general, for an even number of k,
(43)
The ranges of the Eulerian angles required in the formation of the table are reduced to
precession: o => 1/1 => 7r/5,
mutation: o => (J => 7r/4, and
spin: o => tP => 27r/3.
For the icosahedron/tetrahedron configuration, the required memory size is n(r7r)3/15 bytes,
which represents 1/30 as the original range. It is expected that the memory size required by
the table can be reduced to the order of 100 Kbytes.
132 KOK-MENG LEE
5. CONCLUSIONS
The dynamic model and the control strategy of an innovative three degrees-of-freedom
VR spherical motor have been given. The dynamic model of the VR spherical motor consists
of the rotor dynamics and the torque prediction. The torque prediction model has been
derived as a function of the electromagnetic coil excitations and a permeance model as a
function of the relative position between the rotor and the stator. The model represents first
detailed study of the inverse dynamics of the VR spherical motor, and yet, permit a wide
variation of design configuration to be analyzed.
The inverse model of a VR spherical motor, which determines the coil excitations for a
specified torque, is characterized by its infinite solutions. It has been shown that for a current
controlled spherical motor, the relationship between the output torque and the input currents
are algebraic and quadratic. The torque prediction model of a current controlled VR
spherical motor is decoupled from the dynamic equations of the system, and therefore allows
the determination of the optimal electrical inputs to be separated from the motion control of
the spherical rotor. Unlike the conventional motor design where the solution to the forward
and inverse is unique, the multiple-coil excitation allows an optimal control vector to be
chosen to minimize a specified cost function in the control of a VR spherical motor.
Along with the formulation of input vector optimization, the method of designing a
lookup table for the practical implementation of the optimal solution in real-time has also
been discussed. It has been shown that the memory size of the lookup table can be
effectively reduced by parameter elimination and by making use of the symmetry property of
the pole configuration.
Acknowledgements This work is supported by the National Science Foundation under grant
numbers DMC 8810146 and DDM-8958383. Partial support from CIMS program is
acknowledged. Comments from Professor Y -H Chen of Mechanical Engineering at Georgia
Tech are greatly appreciated.
REFERENCES
1. Vachtsevanos, G., and Davey K., and Lee, K.-M. "Development of a Novel Intelligent
Robotic Manipulator," IEEE Control Systems Magazine, June 1987.
DESIGN AND CONTROL OF A VR SPHERICAL MOTOR: PART-2 133
3. Lee, K.-M., Vachtsevanos, G. and Kwan C-K., " Development of a Spherical Stepper
Wrist Motor," Proceedings of 1988 IEEE International Robotics and Automation.
Philadelphia, April 25-29, 1988. Also in Journal ofIntelligent and Robotic Systems,
225-242 (1988).
4. Hollis, R. L., Allan, A.P. and Salcudan, S., "A Six Degree-of-Freedom Magnetically
Levitated Variable Compliance Fine Motion Wrist," Proceedings of the Fourth
International Symposium on Robotics Research, Santa Cruz, August 1987.
5. Kaneko, K., Yamada, I., and Itao, K., "A Spherical DC Servo Motor with Three
Degrees-of-Freedom," ASME Trans. on Dynamic Systems, Measurement and Control,
Vol. III, No.3, pp. 398-402, September 1989.
6. Foggia, A., Oliver, E., Chappnis, F. and Sabonnadiere, J., "A New Three Degree of
Freedom Electromagnetic Actuator," Conference Record - lAS Annual Meeting, Vol.
35, No.6, Published by IEEE, New York, NY, USA; pp. 137-141, 1988.
7. Lee, K.-M. and Kwan C-K., " Design Concept Development of a Spherical Stepper
Wrist Motor," IEEE Journal of Robotics and Automation, Vol. 7, No.1, pp. 175-181,
February 1991.
9. Lee K.-M. and Pei, J., "Kinematic Analysis of a Three Degrees-of-Freedom Spherical
Wrist Actuator," Proceedings of the Fifth International Conference on Advanced
Robotics, Pisa, Italy. June 20-22, 1991.
10. Pei, 1., "Methodology of Design and Analysis of Variable-Reluctance Spherical Motor,"
Ph.D. Thesis, School of Mechanical Engineering, Georgia Institute of Technology,
November 1990.
11. Spong, M. and Vidysagar, M., Robot Dynamics and Control, John Wiley and Sons,
1989.
13. Wolfe, P., "Methods for Linear Constraints," Nonlinear Programming, North Holland,
1967.
15. Abadie, J. and Carpentier, 1., "Generalization of the Wolfe Reduced Gradient Method to
Case of Nonlinear Constraints," Optimization, Academic Press, 1969.
16. Lasdon, L.S., and Warren, A.D., "GRG2 User's Guide," University of Texas at Austin,
1989.
pole x y z
0 0.0000 0.0000 0.0000
1 0.8944 0.0000 0.4472
2 0.2764 0.8507 0.4472
3 -0.7236 0.5257 0.4472
4 -0.7236 -0.5257 0.4472
5 0.2764 -0.8507 0.4472
6 -0.8944 0.0000 -0.4472
7 -0.2764 -0.8507 -0.4472
8 0.7236 -0.5257 -0.4472
9 0.7236 0.5257 -0.4472
10 -0.2764 0.8507 -0.4472
11 0.0000 0.0000 -1.0000
Table 3
Rotor --h1---~-
Stator
x
x-encoder
[3XI ElLWec
--eV 0
[5X) RetB:
(5X)
RIm
CQ
EE
511der c:"
Q.JtDUt
. 9f t
Arms
(ax)
J
J;S1 LS2 ~sm
T T T
~r1 ~r2 ~rn
-I T T
I
8 0.6
6 esired
'",.c: 0.4
4 :0
E
.: 0.2 ac tua l
2 ...
esired
00 00
0.5 1 1.5 2 0.5 1 1.5 2
15 0.03
5
Z 0.02 ", T)
.:
5 ... 0.01
"/
"
~
~ 0
-5 -0.01 0
0 0.5 1 1.5 2 1 1.5 2
.. z
3 2
4, b
3 2
Dennis L. Poll a
Department of Electrical Engineering
University of Minnesota
Minneapolis, Minnesota USA 55455
1-612-625-4873
Abstract
The use of ferroelectric thin films for microsensor and microactuator applications is
circuits for robotics applications. The chemical compositions of the ferroelectric thin
films have been selected for both piezoelectric and pyroelectric applications. The thin
films most extensively studied are lead zirconate titanate (PZT) and lead titanate (PT).
These films are combined with micromechanical structures which implement either an
easily deformable membrane (for force sensing) or low thermal mass (for heat sensing)
ferroelectric films are deposited on these membranes form ultrasensitive pressure and
force sensors, precision positioning devices, stepper motors, and infrared imagers.
1. Introduction
significant advantages of compactness, low system weight, and in many cases low power
Piezoelectric materials formed as a capacitor have often been used for sensing
applications due to their property of significant production of charge when stressed and
inherent low electrical noise characteristics. Conversely, when a voltage is place across
test chip demonstrating the compatible integration of several smart functions is shown
in Fig. 1.
sol-gel deposited lead zirconate titanate (PZT) and lead titanate (PT) thin films on
1.1. Background
Although piezoelectric materials have been used for a long time as sensors, surface
acoustic wave devices, and mechanical movers, this paper will focus only on more recent
focused on the use of zinc oxide (ZnO) and lead-zirconate-titanate [Pb(Zrx Ti 1-x)03 or
or pyroelectric and zinc oxide (ZnO) thin film technology. These microsensors include
detectors (6), acoustic sensors [7-10), and tactile force sensors [11). The ability to
integrate ZnO microsensors with on-chip electronics offers distinct advantages of high
Although many of the zinc oxide based sensors have served as fruitful concept
demonstration vehicles, the inherent low piezoelectric activity in this material has
ferroelectric material PZT [7) has been successfully integrated with silicon based
piezoelectric activity of approximately 15 times that of the best reported ZnO thin
films.
incorporate circuits on the same monolithic chip with integrated circuits or adopt
142 D.L. POLLA
deposition. The ideal deposition technique should therefore produce thin film properties
which are identical to naturally occurring bulk material properties, are conveniently
patterned with the precision set by integrated circuit design rules, and use processing
procedures which are compatible with the formation of other microstructures and
circuits on the same substrate. Only limited forms of integration of these microsensors
with on-chip MOS electronics have previously been possible due to the fabrication
difficulties associated with piezoelectric thin film technology. The key difficulties can
Microstructure sensing elements with piezoelectric thin films have mainly been
fabricated in the form of thin membranes and cantilever beams by silicon anisotropic
wet chemical etching techniques[13-14). This often utilizes the toxic silicon etching
have often made this form of integrated microsensor fabrication impractical. The
MICROMACHINING AND MICROACTUATORS FOR ROBOTICS APPLICATIONS 143
piezoelectric thin film deposition with silicon integrated circuit technology is a critical
concern for the manufacture of integrated sensors where both high performance and high
machining. This processing technique uses a spacer material and conformable deposition
substrate.
coupling of the piezoelectric sensor to the gate input of an MOS amplifying circuit. An
Both force sensing and micromechanical actuation are physically governed by the
D =ESE + eS (1)
(2 )
strain in the material, T is the externally applied stress, and cE is the elastic stiffness
in the presence of constant or zero electric field. Eqn. (1) relates electric flux density
to the electric field strength (E) and considers induced polarization resulting from
applied stress. Eqn. (2) is a form of Hooke's Law relating stress to strain and considers
The pyroelectric constant pO' is defined as the differential change of polarization (P)
with temperature (T) in the case of uniform heating, constant stress (s), and low
(3 )
This section describes the electrical and material properties of PZT and PT thin
films. Of these two thin film materials, PZT has mainly been optimized for force sensing
and mechanical movement applications and PT has been optimized with respect to its
piezoelectric activity than that observed in ZnO. Its use in integrated microsensor and
derived films must be deposited by spin or dip coating. Several sol-gel deposition
techniques have been previously reported [15-17). To date, only sol-gel deposition
techniques using alkoxide precursors have been used for microsensor [18) and
PZT films with various ratios of zirconate to titanate have been prepared by sol-gel
methods in our laboratory. The device work discussed below used films of 54%
zirconate to 46% titanate. This ratio lies near the morphotropic phase boundary and can
MICROMACHINING AND MICROACTUATORS FOR ROBOTICS APPLICATIONS 145
be expected to exhibit strong piezoelectric properties, although the validity of the bulk
material phase picture for these thin film samples remains an open question. These
films have exhibited excellent microstructure [21] and good ferroelectric properties
(Table 1), although some variation with deposited film thickness has been observed. The
ability to achieve proper crystallization of the piezoelectric films is critical and is aided
by the use of a platinum nucleation electrode formed by sputter deposition prior to sol-
gel deposition.
Considerable study has been carried out in our laboratory on the dependence of film
microstructure on the sol-gel precursors [22] and on optimizing the sol-gels to obtain
dense films. Films between 0.2 and 1.0 llm have been deposited on platinum substrates,
and this thickness range is adequate for sensor applications. The high piezoelectric
coefficients available in this family of films suggests that they may be suitable for
Test samples of the 54/46 PZT were fabricated using a platinum lower electrode and
gold upper electrode. Piezoelectric properties (see Fig. 3) were measured by a load cell
technique on film samples without the top electrode. Pyroelectric properties were
substrate underneath the film test area was etched away. A plot of the pyroelectric
coefficient vs. temperature is shown in Fig. 4. These values demonstrate the viability of
thin film forms of PZT for sensing schemes based on charge detection. The piezoelectric
voltage developed across the film thickness is directly proportional to the piezoelectric
coefficients, with contributions from both the d33 and d31 coefficients. For the
particular sensor geometry discussed below, the d31 component is the more important.
The actual voltage developed is inversely proportional to the capacitance per unit area of
the film. This suggests that, particularly for the case of pyroelectric detection, sensor
constant insulator. A high quality insulator would also alleviate the problem of charge
2.2 Micromachining
substrate or a sacrificial layer from the surface of the wafer. Of the two approaches, we
believe surface-micromachining holds more promise for VLSI systems for the following
are used, 3) precise dimensional control can be achieved, and 4) the mechanical
Ferroelectric fabrication begins with either MOS circuits processed up to the point of
with Si3N4/Si02 covered silicon substrates if off-chip electronics are to be used. A 0.3
and a 0.8 JJm-thick layer of phosphosilicate glass (PSG) are first deposited at 800 0C
and 450 0C, respectively. The silicon nitride layer forms an encapsulation layer to
protect the almost finished CMOS circuitry from subsequent processing of the on-chip
sensors and the PSG serves as the sacrificial oxide spacer used in the formation of sensor
membrane structures. The PSG is patterned and chemically etched to form anchor
deposition (by LPCVD). Sensor membrane regions are defined and anisotropic ally
sputtered over the entire wafer. The platinum serves the dual purpose of providing an
MICROMACHINING AND MICRO ACTUATORS FOR ROBOTICS APPLICATIONS 147
adhesion/nucleation surface for the subsequently deposited PZT of PbTI03 thin film and
described The ferroelectric films are then patterned by either chemical etching or Ion
beam sputter etching Photolithography IS then used to protect the ferroelectric thin
films In carrying out a lateral sacrificial etching step of the PSG layer This step IS
are released uSing 48 wt% hydrofluonc aCid to undercut the PSG spacer layer The
force sensing or a low thermal mass structure necessary for pyroelectnc detection For
pressure sensors, the open side walls of the mlcrobndge are sealed by the directional
A top electrode to the ferroelectriC thin film IS then deposited by TI/Au evaporation
Contact openings to both the CMOS transistors and sensor bottom polysllicon electrode
are then chemically etched This IS followed by sputtenng aluminum (2% SI) to form
3. Circuit Integration
plezoelectnc thin film depOSItIOn techniques With a high performance CMOS process
[31], and 3) the ability to carry out necessary mlcromachlnlng techniques while
from being drained from the capacitor A Simple integratIOn technique for PZT thin film
148 D. L. POLLA
sensors is shown in Fig. 11. An induced stress will produce a charge due to the
piezoelectric effect or a change in temperature will produce an induced charge due to the
pyroelectric effect. This charge (or voltage) is transduced to the gate of an on-chip
NMOS transistor. The amplifiers on our test chip had one input gate connected to a
configuration can be used to cancel unwanted signals by fixing a second ZnO element to
change in capacitor voltage produced due to the piezoelectric effect is given by [31]
A e!1S
!1V = Z=n=O____
Pieze C +C
ZnO Gate (4)
where CZnO is the capacitance of the sensing element, AZnO is the the sensor cross-
sectional area, CGate is the input MOSFET gate capacitance, e is the piezoelectric stress
constant, and as is the change in the average macroscopic strain of the piezoelectric
sensor.
The change in capacitor voltage produced by the pyroelectric effect in the infrared
AD!1T
!1V = __=Zn,-",a~
Pyre C +C
Zno Gate (5)
Because these voltages are directly transduced to the gate of an on-Chip MOSFET
amplifier, the small-signal drain voltage or change in dc operating point voltage can be
conveniently measured for both time-varying and steady-state sensing of stress and
MICRO MACHINING AND MICROACTUA TORS FOR ROBOTICS APPLICATIONS 149
temperature change depending on the specific sensing application. In either case, for
pyroelectric coefficients.
Fabrication of several CMOS integrated microsensor chips have been carried out using
a planar process based on 2- and 3-Jlm rules with both ZnO [23] and PZT sensor
technology [24], respectively. Sensor and n-well CMOS circuit fabrication was carried
out on 18-22 D.-cm, p-type wafers. CMOS signal processing circuitry is first
fabricated to the point of source-drain contact hole etching in a conventional seven mask,
polysilicon gate, process. Both n- and p-channel source and drain regions are formed by
The overall sizes of the diagnostic microsensor structures range from 20x20 Jlm2
to 200x500 Jlm 2 . Using a 15-mask process, test circuits and microsensors based on
the piezoelectric and pyroelectric properties of ZnO thin films have been fabricated for
the detection of infrared radiation, tactile force, pressure, and acceleration. Some
4. Microsensor Applications
Integrated piezoelectric/pyroelectric thin films have been used for tactile force
Tactile force sensors formed of ZnO and PZT thin film capacitors have been fabricated
normal to the array surface. The sensor voltage Vpz is given by the scalar product of the
150 D.L. POLLA
piezoelectric-strain constant d33 and the incident compressive force (T) divided by the
(6 )
These array structures have served the dual purpose in aiding in the measurement of
micromachining process. This eliminates the need for front-to-back side alignment,
does not weaken the mechanical integrity of the silicon wafer through bulk-anisotropic
etching, and does not complicate the fabrication of on-chip circuitry. Second, silicon
pressure reference cavity. This eliminates the need for high temperature oxidation or
theory of deformable plates. Our analysis [10, 25) is based adaptations of that
sensor. The coordinate axes and dimensional variables used in this derivation are shown
in Fig. 6. The deflection, W, of a rectangular plate with all edges built-in can be obtained
g
D
(7)
where
MICRO MACHINING AND MICROACTUA TORS FOR ROBOTICS APPLICA nONS 151
Eh 3
D=~---- (8 )
12(1-u 2 )
and q is the loading force intensity assumed to be uniform across the plate, E is the
modulus of elasticity, u is Poisson's ratio, and h is the thickness of the plate. In our
analysis we have assumed small elastic bending of a flat plate. This assumption may not
be applicable to the sealed cavity structure fabricated due to the steady state difference
in pressures on each side of the membrane. This assumption is currently being assessed
more completely.
From the solution for deflection, the normal stress components, (Ix and (ly, are given
by
-Ez- ( d 2w d 2w
(I - + u-~) (9)
x = - 1 _u 2 dx2 dy2
and
( 1 0)
The principal equations for the direct piezoelectric effect in ZnO can be expressed as
follows [9]
(Ix
(ly
Pxl
[ Py =
roo
0 0
0 0 d15 0 l
0 d15 0 0
(lz
( 11)
Tyz
pzJ Ld31 d31 d33 0 0 0 J Tzx
Txy
For bulk crystalline ZnO, the piezoelectric coefficients are d15 = -8.3x10- 12 GIN,
d31 = -5x10- 12 GIN, and d33 = 12.4x10- 12 GIN. The components of polarization are
represented by Px - Pz, and (I and T represent the normal and shearing stress
Eqn. (9) and (10) over the area of the membrane and dividing by the total membrane
area
a/2 bl2
axav = !J Ja xdxdy ( 1 2)
and
a/2 bl2
ayav = !J Jay dxdy ( 1 3)
where a and b are the length and width of the membrane, respectively. The average
stress components in the piezoelectric film, a'xav and a'yav, are then calculated by
integrating equations (5) and (6) over the thickness of the film.
-(h/2 - t z )
a'xav f- f axav dz ( 1 4)
z -h~~
and
- (h/2 -tz)
a'yav ~ fayav dz . ( 15 )
z -h~~
The average polarization is then obtained from
Since the electrodes are formed normal to the z axis, the induced charge on the
electrodes is nearly independent of az. From Eqn. (10), the unamplified sensitivity of
this device is
(17)
where Cz is the capacitance per unit area between the upper and lower electrodes. The
details of the solution reveal that the sensitivity is linearly dependent on the side length
variation over the range 200 Hz to 40 kHz. Some representative data is given in Fig. 7.
4.3. Accelerometers
cantilever beams of 1.5 iJm-thickness which serves as the structural support for a 1.0
iJm-thick ZnO thin film (see Fig. 8). The free standing composite structure oscillates in
the ZnO electrodes. This charge is converted to an output voltage. Basic piezoelectric
sensitivity is given in Eqn. (3). The unamplified response of a 100x1 00 iJm 2 device
infrared detector elements. This paper describes the integration of sol-gel deposited
micromachined membranes for pyroelectric sensing. The unique aspects of this work
are 1) the sol-gel deposition of PbTi03 thin films for infrared detector applications, 2)
the integration of PbTi03 thin films on low thermal mass polycrystalline silicon
Lead titanate (PbTi03) is a ferroelectric oxide ceramic which has both a high
natural advantages as a pyroelectric material are it's high sensitivity and wide operating
temperature range. In particular, depoling does not occur through normal use for
Successive 700 A-thick spin castings were used to build desired film thicknesses in
the range of 0.2-0.6 ~m. Firing for crystallinity was done at 550-600 DC for 30
minutes in air. The details of the materials processing procedure are described
elsewhere. The X-ray diffraction analysis with CuKa radiation has shown these films
have perovskite structure. The intensity of (001) peak in the X-ray pattern increases
with annealing temperature. However for the annealing temperature higher than 650 DC,
ratio of (001) peak to (100) peak is in the range of 0.6-0.8. The average grain size for
properties were evaluated by the hysteresis loops using a modified Sawyer and Tower
circuit. For the 0.5 ~m-thick PbTi03 films deposited on polysilicon, the remanent
polarization of 32-42 ~C/cm2 and a coersive field of 120-150 kV/cm have been
low thermal mass and low thermal conductivity to the underlying substrate. The basic
structure on which the active pyroelectric thin film is supported is shown in Fig. 9. A
patterned 0.8 ~m phosphosilicate glass (PSG) film was used to define the air gap of the
structure shown. A heavily doped 1.0 ~m-thick polycrystalline silicon film was
deposited and patterned over the PSG with anchor points to the substrate on two ends.
270x280 ~m2. The polycrystalline silicon layer also serves as the lower electrode of
Lateral etching of the PSG layer using hydrofluoric acid was carried out to form the
thermal micro-bridge structure of the infrared detector. PbTi03 thin films were
hydrolyzed alkoxide based starting solutions. The top electrode for the pyroelectric
MICROMACHINING AND MICROACTUATORS FOR ROBOTICS APPLICATIONS 155
capacitors with lateral dimensions from 10x20 to 260x270 ).1m 2 was form by gold
evaporation.
uniform heating, constant stress, and low electric field in the crystal. The measured
signal measurements.
structures using both a chopped blackbody radiation source and a modulated infrared
laser at a wavelength of 950 nm. A low frequency lock-in amplifier was used to
measure voltage response to a calibrated incident power. The measured 900 K blackbody
voltage responsivity, Rv, for active area of 7x10-4cm2 is shown in Fig. 11. The voltage
The low-frequency noise was found to be dominated by a thermal noise mechanism. The
The measured voltage responsivity and noise behavior was used to calculate a
Rv fAt
D
•
= Y~-
A ---- ---
Vn (18)
where A is the device area and df is the amplifier bandwidth. The frequency dependence
of the normalized detectivity 0* is also shown in Fig. 11. The 0* value at 297 K and 50
Hz is 1.0x109 cm-Hz1/21 W .
The response time of the pyroelectric detectors was studied using a pulsed infrared
laser at 950 nm. The pulsewidth of the laser beam used varies in the range of 0.5-
50).ls, the rise time measured by the pulsed-laser technique for variable area detectors
156 D. L.POLLA
previously described was in the range of 2-20 )1s. In general, the response time
measured is shorter for smaller active area. For the sample with area of 200)1m 2, the
PbTi03 thin films were also directly integrated with polycrystalline silicon
microstructures on silicon wafers containing analog CMOS test circuits. The direct
integration of pyroelectric elements with on-chip electronics avoids parasitic losses due
to external wires and leads to a high performance system on one monolithic chip.
Because the PbTi03 thin film capacitor structure has a high resistivity (approximately
10 7-10 8 o-cm), direct electrical interconnection with the gate of a CMOS amplifier is
through direct connection of the detector element with an amplifier stage with a voltage
gain of 9.5 dB. A 64 x 64 element infrared detector array shown in Fig. 12 has been
fabricated and will be reported at the 1992 VLSI Circuits Symposium [24].
In summary, sol-gel deposited PbTi03 thin films have been integrated with
polycrystalline silicon membrane for reduced thermal mass and thermal isolation
than other IC-compatible thin film materials and is compatible with CMOS integrated
5. Microactuator Applications
5.1. Micropositioning Devices
Novel macro- and micro- positioning actuators has been fabricated based on the
series and electrically in parallel [27). As shown in Fig. 13, each bar is electroded on
two opposing faces parallel to the length of the bar with electrical connection made to the
alternated to achieve linear expansion and linear contraction in adjacent bars. Because
both ends of the meanderline are anchored to a silicon substrate, the center of the
meander line experiences a forward displacement equal to N times the change in length of
microfabricated chip.
The positioning force obtained from the meanderline has been derived by Robbins, et.
F =2Ed wV
pz 31
where V is the applied voltage, w is the width of the piezoelectric bar, d31 is the
actuation, has been constructed using PZT [28). This motor has been constructed as a
mm x 1.6 mm connected between a glider base and an attached load (Fig. 15).
Displacements are controlled by the application of PZT extension voltages ranging from
158 D.L. POLLA
(see Fig. 16) over a 5.0 cm length of travel. Glider velocities ranging from 5.7 - 476
).1m s-1 for forward movement and 2.2 - 294 ).1m s-1 for backward movement were
caused a linear decrease in the glider velocity (Fig. 18). An attached load of 30-35 gm
the bottom of the glider is used with an electrode on the top of the trench to implement an
in four steps. First, the electrostatic clamp is activated with a large voltage (500 volts
max.) to hold the glider in place (Fig. 15a). Second, the PZT element is extended by an
applied voltage which moves the center of mass of the entire structure, glider and
attached load, forward (Fig. 15b). Third, turning off the electrostatic clamp then
releases the glider (Fig. 15c). The voltage across the PZT is held constant during this
step keeping the piezoelectric material extended and the overall center of mass in the
same position as at the end of the previous step. Fourth, the piezoelectric material is
finally snapped back to its original size by turning off its applied voltage. This pulls the
The results of our prototype stepper motor are encouraging and suggest that this
techniques. Integration of this type of device on a silicon wafer should result in even
generate large atomic forces while electrostatic motors are relatively limited in their
6. Conclusions
Piezoelectric and pyroelectric ZnO and PZT integrated sensors have been fabricated in
an entirely planar MOS processes. The ability to fabricate integrated thermal detectors
and integrated force sensors in an MOS process is attractive for both high-performance
and low-cost sensor systems. Polysilicon surface microstructures which support thin
films of pyroelectric and piezoelectric ZnO, PZT, and PbTi03 thin films have been
integrated with simple on-chip signal processing electronics. Because these surface
microstructures are formed by a combination of sacrificial oxide etching and dry etching
process may improve both yield and reliability of ZnO integrated sensors. The ability to
sensor systems with higher analog and digital capabilities, or smart sensors.
Acknowledgements
The author would like to acknowledge contributions and discussions from his
colleagues and students at both the University of Minnesota and University of California.
Robbins have strongly contributed to many of the device concepts presented in this work.
References
2. T. Tamagawa, P. Schiller, H. Yoon, and D.L. Polla, "Micromachined Zinc Oxide Thin
Film Sensors," Abs. 762, 1990 Meeting of the Electrochemical Society, Seattle,
1985.
Wave OSCillator," IEEE Trans. Electron Dev., vol. ED-35, 735 (1988).
Francisco, CA 1984.
7. T. Tamagawa, D.L. Poll a, and C.-C. Hsueh, "Lead Zirconate Titanate (PZT) Thin
Dev. Lett., vol. EDL-7, 254(1987) and E. S. Kim, M.S. Thesis, University of
Pressure Sensors," IEEE Solid-State Sensors and Actuators Workshop, Hilton Head,
11. D. L. Polla, W.T. Chang, R.S. Muller, and R.M. White, "Integrated Zinc Oxide-on-
Electron Dev. Lett., vol. EDL-7, 254 (1986). Reprinted in Microsensors, (R.S.
Muller, R. T. Howe, S.D. Senturia, R.L. Smith, and R.M. White, eds.) IEEE Press,
Anisotropic Etching of (100) and (110) Silicon, IEEE Trans. Electron. Dev., vol.
ED-25, 1178(1978).
15. R. W. Vest and J. Xu, "PbTi03 Films for Metalloorganic Precursors," IEEE Trans.
16. G. Yi, Z. Wu, and M. Sayer, "Preparation of Pb(Zr,Ti)03 Thin Films by Sol Gel
2717 (1988).
17. S. K. Dey and R. Zuleeg, "Integrated Sol-Gel PZT Thin-Films on Pt, Si, and GaAs for
18. T. Tamagawa, C. Ye, C.-C. Hsueh, and D.L. Poll a, "Sol-gel Derived Lead Zirconate
21. D. L. Polla, C. Ye, P. Schiller, and T. Tamagawa, "Application of PZT and Related
Thin Films in Piezoelectric Microsensors," Mat. Res. Soc. 1991 Fall Meeting,
Derived PbTi03 and La-PbTi03 Thin Films," Mat. Res. Soc. 1991 Fall Meeting,
25. P. Schiller C. Ye, T. Tamagawa, and D. L. Polla, "Design and Process Considerations
28. J. Judy, D. L. Polla, and W. P. Robbins, A Linear Piezoelectric Stepper Motor with
1990.
---
Piezoelectric Coefficient
d33 (pC/N) 190 20 240
---
Pyroelectric Coefficient
(nC/cm 2 K) 50 -70 75 - 96 ---
Dielectric Constant 800 -1100 80 --- 110
SAW Chemical
Vapor Sensors
Piezoelectric
Microactuator
Valves
Diagnostic
CMOS Devices
and Circuits
Cantilever
Accelerometer
Structures
Piezoelectric
Integrated
Pressure Sensors
/\
p
p-substrate
300
.- I I I I
Z
""-
0
-
E: 250 f0-
PZT (54146) 4~
c
-
•
~
--.,
.! 200
•
t-
U
0 150 t- -
0
~
... 100 I- -
U
!!!
u
0
N
U
50 t-
-
~
ii:
J I I I
0
0 0.2 0.4 0.6 0.8 1
Zr (%)
Pb(Zr, TI)03
-
~ 1000
N
E
-
u
o
--
c
cCl):
100
a a a a .. a .. .. D D D
• • • •
D D D
1&
•
•
•
•
• •
•
...
.. . . ..
--
PZT
u . ... . .. ...
Cl)
o
u 10 • • • • • • • • •
• • • • • • •
u • •
-
....
o
Cl)
Cl)
....o
~
1
120 160 200 240 280 320
n.
Temperature (oK)
~
~
<3
;:0
1 I ~PSG~ C/O
Si JubS-trlte Si l ubstnue
r Si sub, ... ,. I~
1';~'"b,tnI'c~'J 2l
;:0
;:0
ot:O
pol)' or tunlsten
~~~1?;Ii!i ...,o
n
C/O
Si substnlc ~
~~ " ~~ 5i s\Jbstnte
Si,ub,,,..,c ~,< ., > ~
n~
;:: al uminum l
...,
• EX , J E±f¥EM
Ai . ! I oz
C/O
Si lu bslTlte Si S.Ubllr'llc:
Aluminum
silicon substrate
100
"..
>=.
-• ••
-
II • • •
-
ca S • 228.89 ~VI ... b.. • • l i S . 151.23 ~Vlaab ..
.-II 501150 t"m) II ~ ~_"O(Jl.m)
•
-
10 ~.
••
....
0
> • •
II ~
••
- •
I:
C7I
•••
• II
~
VI II
II
U
•
II
•
•• ••
't
.-
II
•
-•
II S· 42.34 .. VIIIbat'
~
••
10_20 (11m)
I
•
- •
0
N
II
D- .1 -3
•
10· 10- 2 10. 1 10°
lower
el9C1lode
c:::::::::::3 PbTlO J
__ Polysllioon
SI
(D)
[S\\\SSSS,{SSs\sSSS\ssSS#SSNSSJ
;t 290
3.0
--
l::
l ::
Q) dT/dt, on left scale 280 g
....
....
:J 2.0 / (x10 Klmin) .,...
o
...
2-
'0
,/'
.» _X-....:
w-)(.-x ...... ,
""
\ 270 -.,...
:J
I'd
C1I I D..
1.0 E
C1I f. 260 .,
....o ~
>- I Ipy
Q. on left scale
0.0
250
-1.0
o 2 4 6 8 10 12
Time (min)
Time (min)
......
.......
('oj
N
o :J:
"tJ
C
;:
cu c
>
~
--
>
a:>
Rv ' left scale
102 ~~~~W-~~~~--~~~L-~~~W-~~~~10-8
10 1 02 1 03 1 04 1 05 1 06
Frequency (Hz)
Frequency (Hz)
Fig. 11. Voltage responsivity, noise voltage, and detectivity versus chopping
frequency for a PbTi03 infrared detector element.
MICROMACHINING AND MICROACTUATORS FOR ROBOTICS APPLICATIONS 171
+ ".
...---
_-
' .'
+. 1 f + .j. ++
E
4 •
..=:
I-
Z
3 •
w
:::liE
w
0
c(
2 •
..J
Q.
(J)
C •
0
0 20 40 60 80 100
1.2
,,
----G-- 25Hz
--+- 100Hz -+-- 100Hz
--
--{}- 400Hz --+- 400Hz I
1.0 ----t---- 800Hz ~ 800Hz
E
..::,
BaCKward Forward
G.I
I/)
0.8
'5 \ . - - - - Theoretical
0.. \
i 0.6
\
'E \1
~
E
~
u 0.4
III
Q.
I/)
is
0 .0 +-~+-~+-=~I---f......-if--.---l---+--I
-300 -200 -100 0 100 200 300 400
Voltage (V)
50 0
--
--
25Hz (Forward) /+J
--
4 501-- - - - - 100Hz (Forward)
100 f - - -
400Hz (Forward)
800Hz (Forward)
II
25Hz (Reverse)
100Hz (Reverse)
3 50 f--
-
~ ---0-- 400Hz (Reverse)
u
G.I 800Hz (Reverse)
J
I 1/(+)
I/) 30 0
E
2;
(oj
1\
II
25
>- \
~
o
200
(.)
'\ )
f\ I.j.
G.I 15
>
\~
0
/) ~
~+)
. (0) .....
50
0
(.) ~ [:\ (
k ~ ~(+)
·400 ·300 ·200 "00 0 , 00 200 300 400
Vo ltage (v)
Fig 17 Forward (+) and backward (-) gilder velOCities versus applied
piezoelectriC voltage
174 D. L. POLLA
300
250 --
--e- SOV
100v
120V
--., 14c>.t
-
u 16c>.t
CII 200
1Sc>.t
E 20c>.t
-
:1-
150
22c>.t
=
>- 24c>.t
U
0
'ii 100
>
50
0
0 5 10 15 20 25 30
Load (gm)
H. S. Tzou
Department of Mechanical Engineering
Center for Robotics and Manufacturing Systems
University of Kentucky
Lexington, Kentucky 40506~046
ABSTRACT
capabilities for the elastic continuum. In this paper, artificial thin-layer distributed
piezoelectric neurons and muscles are integrated with an elastic thin shell and new
distributed neural sensation and actuation theories are derived. Distributed neural
signals are formulated based on induced elastic membrane, shear, and bending strains
of the shell continuum. System dynamic equations of the shell with integrated
piezoelectric muscles are derived and their control applications discussed. The derived
shells, cylindrical shells, conical shells, zero-curvature shells, non-shell type continua,
shell (Case 1) and 2) a cylinder with distributed shell neurons and muscles (Case 2).
Distributed convolving ring sensors are designed and their performance studied (Case
175
H.S. Tzou and T. Fukuda (eds.!. Precision, Sensors. Actuators and Systems, 175-218.
© 1992 Kluwer Academic Publishers.
176 H.S. TZOU
1. INTRODUCTION
Artificial distributed neurons and muscles for distributed sensing and control of
elastic continua, distributed parameter systems, have long been interested and
neurons and muscles for elastic shell continua are proposed and new fundamental
theories are derived. There are many novel piezoelectric devices in a variety of
& Fukuda, 1991; Sessler, 1981; Mason, 1981). Tzou and Gadre (1989) derived a
multi-layered distributed actuator theory. Lee and Moon (1990) proposed distributed
modal sensors and actuators theory for plates and beams. Collins, et al. (1990)
designed distributed convolving sensors for beam structures. Tzou and Zhong (1991)
derived a distributed convolving shell sensor theory and applied to modal sensing of
ring structures. Tzou (1991) also proposed a distributed sensing and vibration control
theory for thin shell continua in which bending effects dominate the oscillation.
proposed and new neural sensation and muscle actuation/control theories for shell
continua are derived. The new theories include both in-plane membrane effects and
out-<lf-plane bending effects of the shell continua. Applications of the generic theories
to other common geometries (e.g., plates, cylinders, spheres, etc.) by using four system
parameters, two Lame's parameters and two radii of curvatures of in-plane coordinate
and a cylindrical shell coupled with distributed shell neurons and muscles are
THIN-LAYER DISTRIBUTED PIEZOELECTRIC NEURONS AND MUSCLES 177
orthogonal functions are proposed and performance evaluated (Tzou & Zhong, 1991).
are surface coupled, or embedded, with an elastic shell continuum, Figure 1. The
elastic shell is assumed thin so that the Kirchhoff-Love theory can be applied. A
Information
(In)
\
Fig.1. A generic shell continuum with thin distributed neurons.
178 H.S.1Z0U
It is assumed that the k-th piezoelectric neuron has an effective area A~, either
segmented or fully distributed, and a thickness tk which is much thinner than that of
the elastic shell continuum, i.e., t k « tS. Note that the superscript n denotes
piezoelectric neurons and s for the shell continuum. Distribution of shell neurons forms
a neural system which can be used for structural monitoring and diagnosis.
strain variations induced by structural vibration and generate output signals. The
(1)
(2)
(3)
where Al and A2 are Lame's parameters (defined by geometrical configuration) and IRI
and 1R2 are the radii of curvatures of two in-plane axes O!I and 0!2, respectively. u· is
1
the displacement and d k is the distance measured from the shell neutral surface to the
mid-plane of the piezoelectric neuron. S11 and S22 denote the in-plane normal strains
and S12 the shear strain in the k-th piezoelectric neuron. The mechanical strains can
be further divided into two major components: 1) membrane strains (all terms in the
THIN-LAYER DISTRIBUTED PIEZOELECTRIC NEURONS AND MUSCLES 179
first brace) and 2) bending strains (all terms in the second brace). Note that the
membrane strains are induced by membrane forces on the neutral surface. The
bending strains are induced by bending oscillations. These strain calculations are also
valid for any embedded piezoelectric neuron. (Note that in a previous study, only
bending effects were considered (Tzou, 1991).) These mechanical strains stimulate the
direct piezoelectric effect in the piezoelectric neuron. Based on the Maxwell principle
and Gauss theory, one can derive an expression for an electric potential ¢k as
(4)
where h .. are piezoelectric constants and {333 is the dielectric impermeability constant.
1J
The electric displacement D 3 in transverse direction can be rewritten as
(5)
Note that Q3 = 1AD3 dA where Q3 is the electric charge and A the surface area. An
(6)
Note that the above expression provides a surface averaged output signal. Substituting
strain expression into the above neural signal equation gives a generic output equation
180 H. S. TZOU
+ d~[i-;a&[~-i-;~] + ~[~-k~H~~]l
+ h 36 [[A2
AI Gai A2 + A2
8 [u2] Al 0cG AI + dn[A2
8 [U1]\ 8 [[u2
k AI Gai lR2 - A2 8u3]/A]
1 0cG 2
If the piezoelectric neuron has a finite area, i.e., lim(A~) --l O. The above
* * where (al,a2)
expression becomes a local (point) neural signal ¢n (al,a2,t) * * denotes the
location. Thus,
(8)
Signal from each neuron indicates a (distributed) local dynamic state and
collection of all neural signals constitutes a distributed neural sensation map of the shell
continuum. Note that it is assumed that these small neurons are physically separated
and each signal can be extracted separately. These two neural signal equations,
averaged and discrete, are very generic. Using four system parameters: two Lame's
THIN-LAYER DISTRIBUTED PIEZOELECTRIC NEURONS AND MUSCLES 181
parameters Al and A2 and two radii IRI and 1R2, one can simplify these equations to
many other specific geometries, such as spheres, cylinders, plates, beams, etc (Tzou,
1991). Demonstration examples, a hemispheric shell and a cylinder, are shown in case
accommodated in the equation. For example, the in-plane piezoelectric constant h36 is
class C6v = 6mm: piezoelectric ceramics). Some other materials may be specially
made to strengthen the in-plane twisting effects while to minimize the membrane
effects.
bonded with the elastic shell and forces/moments can be directly transmitted to the
elastic shell continuum. Injecting a high voltage induces strain/stress, due to the
Piezoe l ectric
Forces Muscle
The induced control forces N~. and moments M~., per unit length, can be estimated
1J 1J
as
m
NIl = d 31 Y m if;m3 , (12)
m mm
N 22 = d 32Y if; 3 , (13)
(14)
(15)
(16)
THIN-LAYER DISTRIBUTED PIEZOELECTRIC NEURONS AND MUSCLES 183
where ym is Young's modulus; and dy: is the moment arm (distance) measured from
the neutral surface to the mid-plane of piezoelectric muscles. The high voltage ¢T
could be either a reference signal (open-loop) or a neural signal (closed-loop), which
feedback control. Besides, control algorithms also affect the amplitude and/or pattern
of feedback voltages. Including all induced control forces and moments in the system
equation, one can derive a set of system equations in three principal directions.
1 [_8(NllA2)_8(N21Al)_NI28Al+N228A2]
AlA2 aCt 1 aCt 2 aa2 001
Q13*
s.. m
-1Ifl+ptUl=Fl+FI, (17)
* are defined as
Qij'S
(21)
Note that Nij and Mij are (elastic) mechanical membrane forces and moments (see
Appendix A). p is the elastic shell mass density, t S the shell thickness, Fi the external
mechanical excitation. All control components induced in the distributed muscles are
184 H. S.1Z0U
(25)
(26)
Again, the above generic expressions are for a generic elastic shell coupled (or
reduction procedures can be used to apply the theory to other geometries. Note that
the piezoelectric neural signals can also be processed in a "brain" (e.g., a computer),
muscle theories are very generic. Using four system parameters: two Lame's
parameters A I and A2 and two radii of curvatures IRI and 1R 2, one can easily simplify the
fundamental form; and the radii can be directly observed from the coordinate system
defined for the geometry.
THIN-LAYER DISTRIBUTED PIEZOELECTRIC NEURONS AND MUSCLES 185
illustrates the hemispheric shell with a distributed shell neuron covering from
assumed that the piezoelectric neurons and muscles are made of symmetrical hexagonal
RSin~
Rd 1IJVd 5
(27)
186 H.S.1Z0U
where ¢ and B define the neutral surface of the hemispheric shell; and IR is the radius.
Substituting these four parameters into the thin-layer distributed shell neuron and
muscle equations, one can derive the neural signal equation and the system equation
with the thin-layer distributed muscle. The neural signal can be estimated by
cps_
k - 27f
t~ Jor¢lrBI[h
Jo 31
{[ 1 f)U¢+u 3] +dn[l
II( 0ifJ 11(
f)
¢ lR 0ifJ
[U¢
-r - I I1( ----;JifJ
f)U 3 ]]}
1 Ou
+ h 32 {[~0fJ+~u3] + d B[ ~07J
B 1 f) [U 1
II<-~----OU
n
0u3] B
Note that d~ = dO for a piezoelectric sensor layer with uniform thickness. Note that
the total output signal is contributed by two strain components: membrane strains and
bending strains. All terms after d~ terms contribute the bending output component
and the other two terms contribute the membrane output component. For shells with
only experiences membrane motion, in-plane motion, the bending component should
be minimum. Note that all terms inside the largest brace { } multiplied by t~, i.e.,
without the surface integration and average, represent a local voltage amplitude if the
THIN-LAYER DISTRIBUTED PIEZOELECTRIC NEURONS AND MUSCLES 187
N N
A A
00 ] 2 . ,I, s ..
+ IR 2.
sm,/,
,I, [ "' 1/11/1 + ----or
---or- + 1R sm,/, p t U3 = IR 2sm,/,
. ,I, F 3· (31 )
N.. and M.. are the resultant forces and moments, i.e., N..
IJ IJ 11
= N.. + N~. and
11 I I
Mechanical forces N11.. and moments M11.. of the hemispheric shell can be derived in a
similar way by using four system parameters. The derived mechanical forces and
moments are
_ 1 8u3]+cot1/1[u1/1_1 BU 3 ]]}
~ 071 -r -nr- II< 0ifJ '
(35)
188 H.S.1Z0U
{[ Y(t S)3] / [12(1-1t 2 )]}. And It is Poisson's ratio. Substituting these into the
transverse system equation yields a final equation in terms of ui's and feedback
forces/moments.
The derived distributed neural sensation and muscle actuation theories are for a
generic shell continuum. Applying the theories to other general geometries needs four
system parameters: two Lame's parameters Al and A2 and two radii of curvatures IRI
cylinder with a partially covered neuron layer and a fully covered muscle layer is
illustrated in Figure 4. Note that it is assumed that the active directions of the
symmetrical hexagonal piezoelectric layers are aligned with the coordinate axes x and
O. Other assumptions were also previously discussed. The fundamental form of the
cylinder is
(38)
THIN-LAYER DISTRIBUTED PIEZOELECTRIC NEURONS AND MUSCLES 189
where R is the radius of the cylinder. Thus, Al = 1 and A2 = R. Besides, IRI = (I) and
Piezoelectric
(39)
190 H.S. TZOU
Note that UI = Ux and U2 = u e. The system equation of the cylinder with distributed
muscle layers can be derived in a similar way. The electromechanical equation of the
where
Resultant forces and moments can be simplified from the equations listed in Appendix.
(43)
(44)
(45)
where K is the membrane stiffness, D is the bending stiffness, and J.L is Poisson ratio.
NI~\ MI~\ and Mrr2 are defined in Eqs.(13), (15) & (16). Substituting these into the
transverse system equation yields a final equation in terms of u's and feedback
forces/moments.
THIN-LAYER DISTRIBUTED PIEZOELECTRIC NEURONS AND MUSCLES 191
A closed elastic ring is a special case of arch, which has a constant radius.
(Note that a beam is a special case of an open ring with infinite curvature.) In this
section, spatially distributed convolving sensors for ring structures are proposed,
sensor of ring structures is then derived (Tzou & Zhong, 1991). Note that it is
assumed that the piezoelectric sensor layer has constant thickness in the later analyses.
curvatures are IRI = II) and 1R2 = R. The membrane and bending strains are simplified
as
(46-a,b)
(47-a,b)
L
II)
(49)
n=O
arbitrary phase angle; and An and Bn are constants. In the following derivation, it is
assumed that a reference point is defined so that the phase angle rp = O. For a circular
ring with free boundary conditions, the first mode (n = 0) is a breathing mode and the
mode and a circumferential mode appear. It should be noted that for a given integer n,
there are two modal frequencies wn and wn in which the former determines the
I 2
transverse mode, i.e., transverse oscillation dominates; and the later determines the
(50)
Similarly, for circumferential modes, the ratio between the transverse oscillation
Bn2 _1_
(51)
N
= n
An
2
Substituting the modal equations and strain expressions into the neural sensor
III
,1 o
21r
sgn[W(O)].W(O)cos(nO)dO, (52)
where hI and h2 are distances measured from the shell neutral surface to the bottom
and top surfaces of the piezoelectric layer; sgn(·) is a singum function denoting the
circumferential modes with amplitudes An and Bn ; and the second part, bending
2 2
194 H.S.1Z0U
strains, is contributed by transverse modes with amplitudes An and Bn. The modal
1 1
Eq.(52) denotes the distributed sensor equation for spatially shaped piezoelectric
respond to the modal amplitudes of selected modes without the spillover of other
residual modes. Thus, observation spillover problem can be prevented. Since the kth
mode shape is orthogonal to all other mode shapes, the shape function W( a) can be
W k( 0) = bcos(k 0) , (53)
v~ = - ~
£33 .l1k
sgn(cosnO) (tn(kA k
2
+ Bk 2)
+ -k-(h~ - hD(kA kl + k2Bk)) 7Jk(t) , (54)
where 7Jk( t) is the modal participation factor. (Note that Ak denotes the effective
electroded area and Ak is a constant for the kth mode.) kl denotes the transverse
mode and k2 the circumferential mode for n = k. Ak and Bk are respectively the
1 1
circumferential and transverse modal oscillation amplitudes of the kth transverse
natural mode with a natural frequency wk. Ak and Bk are the circumferential and
I 2 2
transverse modal amplitudes of the kth circumferential natural mode with a natural
frequency wk .
2
THIN-LAYER DISTRIBUTED PIEZOELECTRIC NEURONS AND MUSCLES 195
circumferential modal sensitivity can be defined. (Note that the membrane modal
modal sensitivity in ring sensors.) Each of the sensitivities can be further divided into
oscillation amplitudes. Note that these two amplitudes are coupled by a constant,
(55-a)
(55-b)
Bk is
2
(56-a)
amplitude Ak is
2
196 H. S. TZOU
(56-b)
These two sensor sensitivities for the distributed convolving ring sensors are to be
analyzed. Thus, for bending oscillation where transverse modes dominate, st should be
used to estimate the oscillation amplitude. On the other hand, for circumferential
oscillations, SC is used.
6.2. Results
A steel ring structure with a radius 50 mm, width 10 mm, and thickness 1 mm
(25 11m) was spatially shaped as distributed modal sensors. Material and geometric
studies for transverse and circumferential modes were analyzed and results presented in
this section. Contributions from two modal oscillation amplitudes, e.g., the in-plane
and compared. Variations of ring and sensor thicknesses were also investigated. Note
that for a free-floating ring, the first mode k = 0 is a breathing mode and the second
transverse and circumferential modes, with distinct natural frequencies, appear. Thus,
cosine shaped distributed piezoelectric sensors were primarily designed for k ~ 2 and
sensitivity analyses of the distributed sensors were also evaluated for k ~ 2. Spatially
distributed cosine shaped convolving sensors for k = 2, 3, and 4 modes are shown in
Figures 6-a,b,&c. Note that the sensors are defined from 0 to 2?r and it is cut at
C0 8.6xlO-10 F/rn
distributed cosine shaped convolving sensor were calculated and plotted for k ~ 2
modes. As discussed previously, there are two natural modes for each k value, i.e., a
transverse mode and a circumferential mode, with distinct natural frequencies. Each
two oscillation components are coupled by a constant. Thus, for an output signal
measured, the voltage can be used to estimate both the transverse oscillation
amplitude and the circumferential oscillation amplitude for a specific natural mode.
two component sensitivities, for the same modes are plotted in Figure 8. In each figure
2 .0
- 8 - CIRCUNFERENTIAL NODES (POSITIVE)
- - TRANSVERSE NODES (NEGATIVE)
1.8
.......
1.6
,,
::E
1.4 P
:::!!
'-.. ,
> 1 .2 ,
r
E-<
:;
1 .0
., ,,
~ 0.8 ,
iii
Z 0 .5 ,
W
,
Ul "
0 .4
n
0.2
-<>
0 .0
? .~ ~ 5 5 7 9 1()
MOOr.S
2.0
- .. - CIRCUNFEREIITIAL NODES (NECATIVE)
- - TRANSVERSE NODES (NEGATIVE)
1.8
1.5
::::s 1.4
::::s
'-..
> 1.2
r
E-< 1 .0
:;
~ 0 .8
iii
Z 0 .5
W
Ul
0.4
___ _-0.-----0----
0.0
2 3 4 5 6 7 8 9 10
MOD8S
Figure 7 suggests that for a measured signal, the inferred transverse oscillation
membrane effect for transverse natural modes. This is true because the transverse
Figure 8 also leads to similar conclusions for the circumferential natural modes.
Transverse and circumferential modal sensitivities were also evaluated when the
thickness of the ring and the sensor layer were changed. In general, membrane strains,
the circumferential component, in the distributed sensor should be the same regardless
the ring thickness. However, bending strains, the transverse component, increase when
the ring becomes thicker. Higher strains in the sensor layer generate higher output
signals. Test results showed that the transverse sensitivity increases when the ring
thickness increases. However, the circumferential sensitivities are identical because the
membrane strains remained unchanged. These results suggested that the membrane
modal sensitivity is a linear function of ring thickness. It was also observed that the
sensitivities increase when the sensor layer becomes thicker, provided the same
6.3 Discussions
In this case, cosine shaped piezoelectric convolving modal sensors were designed
and analyzed for ring structures. Transverse modal sensitivity for transverse natural
modes and circumferential modal sensitivity (equivalent to membrane modal sensitivity
in shells) for circumferential natural modes were defined and results plotted.
202 H.S. TlOU
Parametric studies suggested that the transverse modal sensitivity increases when the
ring structure becomes thicker because the bending strains in the sensor layer increase.
membrane strains were independent of the ring thickness. It was also observed that
both modal sensitivities increase when the piezoelectric sensor layer becomes thicker.
study, provide a better understanding of how the distributed sensors work. Proper
selections of piezoelectric sensor thickness, shape, and convolution can provide spatial
control systems.
The quest for superior structural performance has recently driven new research
distributed piezoelectric sensors and actuators are used. In this study, the modal
y
b
z b
A Four
actuator
I
films
o
Four
sensor
a fUms
x
Fig.9 A plate with four segmented distributed sensors/actuators.
fully and segmented, are compared based on theoretIcal denvatIOns (Tzou & Fu, 1992)
(57)
where T)mn IS the modal coordInate, (mn IS the modal dampIng ratIO, Wmn IS the mn-th
natural frequency, Mmn IS the modal feedback factor (velocIty feedback), Fk IS the
(58)
where Nk = .(~ U~mndxdy; U 3mn is the transverse mode shape function; and F 3 is the
transverse mechanical excitation. Tk denotes the coupling terms between the mn-th
mode and all other residual modes. It is assumed that modal filters are used with the
the plate from Xl to X2 and from Yl to Y2. The modal feedback factor Mmn can be
derived as
. -1 mb na m7rX! m7rX2
Mmn = ptNk(Mxmnna+ MymniiiD)(cos-a--cos-a-)
nny! nny2
. ( cos----o- - cos----o-) , (59)
where p is the mass density and t is the plate thickness. Note that the plate has a
and 9 is the feedback gain; Smn is the sensitivity, Smn = !: [h31r~(~)\ h32r~(Ell .
where
Note that the above two sensor/actuator configurations, the modal feedback
factors Mmn's are identical if both m and n are odd numbers, i.e., the four-piece
segmented sensors/ actuators has the same control effect as the single-piece distributed
sensor/actuator for all odd modes. The single-piece sensor/actuator has no control
effect for all even modes and the four-piece segmented sensors/actuators for all
quadruple modes. (In this study, only the four-piece segmented distributed
distributed sensors and actuators are certainly feasible.) Then, a controlled damping
(65)
which is a linear function. The natural frequency wmn for a simply supported plate is
Note that control moments induced in the segmented distributed actuator only
counteract the motion in half of the oscillation cycle and aggregate the motion in the
other half cycle in a displacement feedback control. However, the control moments
always counteract the oscillation in a velocity feedback control. Thus, only the
velocity feedback control is considered in the later analyses. It was also observed that
equivalent control effects for the odd modes. Thus, odd-mode control effects of a
configuration. Control effectiveness with different feedback gains are evaluated in this
section.
A square plate made of plexiglas and sandwiched between two thin piezoelectric
polymers is used as a physical model. It is assumed that all four edges of the plate are
simply supported. The top and bottom piezoelectric layers are further segmented into
four equally sized sub-areas, i.e., separated in the middle, resulting in four segmented
actuators and sensors. Note that the separation is infinitesimally small so that it is
Note that damping ratio is used as a comparison index. The first six natural
frequencies of the simply supported plate are calculated and summarized in Table 3.
Note that there are symmetrical modes for modes (1,2), (1,3), (2,3), etc
Modes Theory
m= 1, n=l 61.40
m= 1, n=2 153.51
m=2, n=2 245.62
m= 1, n=3 307.02
m=2, n=3 399.13
m=3, n=3 552.64
Feedback Gains
equal to the total surface area of the plate, i.e., fully covered. The inherent system
damping ratio is assumed to be 1% for all modes. This is not necessary to be true.
However, since only the variation of damping ratios induced by the distributed control
is the main concern, the inherent system damping is assumed within a reasonable
range. Figure 10 shows the damping variations for the first three distinct natural
modes. (Note that the first three distinct modes actually represent first four natural
modes because of the symmetry.) Table 4 presets the damping ratio variations of the
0.013
o 0.012
o
a::
01
c:
~ 0.011
o
c
Feedback Gain
~
Damping order m= 1 m= 1 m=2 m= 1 m=2 m=3
n=l n=2 n=2 n=3 n=3 n=3
~
h-eedback
p'nin
61.4
Hz
153.5
Hz
245.6
Hz
307.0
Hz
399.1
Hz
552.6
Hz
Time-History Analyses
In this section, time history responses of transient oscillations are studied and
compared. It was proved that basically the plate size does not alter the controlled
damping ratios. Thus, a plate with a dimension of 40cm x 40cm is used, which has
210 H.S.1£OU
lower natural frequencies so that time histories are better distinguished. Note that
only modal coordinates are plotted and compared. (For physical displacements of the
plate, this modal coordinate needs to be multiplied by the mode shape function of that
mode.)
Transient responses of the plate with and without feedback controls are
calculated and the envelops are plotted in Figure 11, the (l,l)-th mode, Figure 12, the
(1,2)-th mode, and Figure 13, the (2,2)-th mode. The controlled responses are
calculated based on a total feedback gain of 300. Their corresponding 10% settling
0.0012
E
feedback
.u
c:
.
0
Q.
U
a:: 0.0000
c:
u
E
u
u
.
0
Q.
Q
-0.0012
0 2 3 4 5
Time (sec)
0.0005
e Wi thout feedback
.
~
/
u
.
c:
0
Q.
U
a:: 0.0000 ---------
c:
u
E
u
u
.
0
Q.
0 -0.0005
0 2 3
Time (sec)
0.0003
E Wi thout feedback
..
u
c:
..
0
Q.
/
U
a:: 0.0000
c:
u
E
u
u
0
..
Q.
0
-0.0003
Time (sec)
~
Settle order m=l m= 1 m=2 m= 1 m=2 m=3
ume n=1 n=2 n=2 n=3 n=3 n=3
~Condition
15.4
Hz
38.4
Hz
61.4
Hz
78.8
Hz
99.8
Hz
138.2
Hz
t Plate size: 40cm x 40cm; tt Gain = 300; ttt System damping = 1%.
7.3. Discussions
signals and the same gain is applied to all modes. The damping ratios are proportional
to the feedback gains and higher feedback gains give better control effects. However,
variations due to induced damping are relatively minimum since the dampings are low.
There are a number of factors needed to be considered when examining the above
results. First, the free decay of a modal oscillation depends on the product of the
modal damping ratio and the frequency, i.e., e-Cmnwmnt. Hence, even if they have the
same active damping ratio, the higher modes decay faster. Second, with the same
different for each mode. Higher modes are hard to be excited so that the output
THIN-LAYER DISTRIBUTED PIEZOELECTRIC NEURONS AND MUSCLES 213
signals will be relatively weak. Hence, larger feedback gains would be needed to
enhance the control effects. Third, the output signals are different for different modes.
Even the same amplitude of oscillation is considered for all oscillating modes, the
sensor layer deforms more severely for the higher modes. However, higher strain lever
does not warrent a higher output signal because charge cancellations could occur on the
surface electrodes. Fourth, two identical shape and different size plates, the sensor
layer on the smaller plate yields higher output signals for the same lever of oscillation
(or embedded) with an elastic shell and used for structural sensation and actuation
were proposed. New distributed neural sensation and muscle actuation theories were
derived based on the Maxwell principle and Gauss theory. Electromechanics (sensor
analyzed. It was observed that the neural signal is contributed by membrane strains
and bending strains experienced in the neuron layer. Two sensor sensitivities: 1)
transverse natural modes and the membrane modal sensitivity for in-plane natural
modes. Proper design of distributed sensor shape and convolution can provide modal
Actuator mechanics of the thin-layer piezoelectric muscles were analyzed and the
induced membrane control forces and counteracting control moments were integrated
into the system equations, in three principle directions. The derived theories are very
214 H S nov
genenc, whIch can be easIly slmphfied to account for many other shell and non-shell
dIstnbuted systems, such as cylmdncal shells, sphencal shells, plates, beams, etc ThIS
slmphficatlOn can be carned out by usmg four system parameters two Lame's
parameters and two radn of curvatures A hemisphenc shell and a cyhndncal shell
and apphcatlOns can be proceeded from the denved system govermng equatlOns
analyzed for nng structures m Case 3 Transverse modal sensztzmty for transverse
natural modes and ctrcumJerentzal modal sensztzmty (eqmvalent to membrane modal
sensztzmty m shells) for cHcumferentlal natural modes were defined and results plotted
Parametnc studIes suggested that the transverse modal senSItIvIty mcreases when the
nng structure becomes thIcker because the bendmg strams m the sensor layer mcrease
membrane strams were mdependent of the nng tmckness It was also observed that
both modal senSItIvItIes mcrease when the plezoelectnc sensor layer becomes thIcker
Case 4 The analyses showed that segmentmg dIstnbuted sensor and actuator layers
the system The segmented actuator deSIgn Improves the observablhty / controllablhty
for even modes WIthout degradmg the control ments for odd modes of a SImply
supported plate In general, lower modes are more Important than hIgher modes m
structural momtonng and control Thus, only several lower modes are consIdered m
thIS study, although further segmentatlOn of actuators are possIble and mIght proVIde
THIN-LAYER DISTRIBUTED PIEZOELECTRIC NEURONS AND MUSCLES 215
control capability for distributed parameter systems. Applications of the devices are
ACKNOWLEDGEMENT
This research was supported by a grant from the National Science Foundation
(No. RII-8610671) and the Kentucky EPSCoR Program. A grant from the Army
the position or the policy of the government, and nor official endorsement should be
inferred.
REFERENCES
Collins, S.A., Miller, D.W., and von Flotow, A.H., 1990, "Sensors for Structural
Soedel, W., 1981, Vzbratwns of Shells and Plates, Marcel Dekker, New York.
ROC.
Piezoelectric Sensors and Actuators, Part-1 and 2," Symposium on Active Vibration
Tzou, H.S. and Fukuda, T., 1991, Pzezoelectnc Smart Systems Apphed to
Thin Shell Coupled with Piezoelectric Shell Actuators for Distributed Vibration
Tzou, H.S. and Zhong, J.P., 1991, "Sensor Mechanics of Distributed Shell
and Systems, 1991 ASME 13th Biennial Conference on Mechanical Vibration and
NOise, Miami, Florida, Sept.22-25, 1991; and ASME Journal of Vzbratwn and
1) Membrane Forces
(A-I)
(A-2)
(A-3)
where Kis the membrane stiffness and K= [Yts/(l-i)], Y Young's modulus, t S the
2) Bending Moments
(A-4)
218 H.S. TZOU
(A-S)
1 0Ci2
- 1\2 0u3 ]/A] Al a
2 + 1\2 ---00;
[[ 1Rj-1\l---aaJ
UI 1 aU3]/A]}.
I,
(A-6)
displacement in i-th direction, IR. the radius of curvature of i-th axis, A. Lame's
I I
Janusz A. Marszalec
ABSTRACT
The subject of this text and the lecture is presentation of the fiber optic sensors
and systems developed for robotics and different applications that exploit unique
characteristics of optical fibers. Chapter 2 introduces briefly basics of fiber optics,
constituting a background for the further discussion. Chapter 3 presents an overview of
the applications of fiber optics in industry and outlines the specific requirements to fiber
optic systems for robotics. Chapter 4, which is the most broad in the presentation,
provides an overview of the optical fiber sensors for robotics, presenting some basic
phenomena, sensor models, and their applications. Fiber optic lighting and imaging
systems for machine vision are the subject of Chapter 5. The last chapter presents general
conclusions and discussion of the further prospects for the applications of fiber optics in
robotic systems taking into account the current trends in both photonics and robotics.
1. INTRODUCTION
level of intelligence, to assure its understanding of the environment and to determine how
to use its own resources to reach a desired goal. The measurements of the state of the
environment and the state of the robot itself provide the necessary information to build
the behavioural intelligence of the robot using appropriate reasoning and decision making
methods. Robot sensors play a principal role in this structure. For the last ten years
different types of sensors and systems for vision, sensing of proximity, force, and tactile,
219
H.S. Tzou and T. Fukuda (eds.). PreClsion. Sensors. Actuators and Systems. 219-298.
© 1992 Kluwer Academic Publishers.
220 J. A. MARSZALEC
as well as for sensing of some specific environmental parameters have been developed
and applied. Vision sensing has received the most attention. There is much research on
the robot sensors carried out at the universities and research laboratories, but only a few
types of sensors are commercially available. It seems that robot sensors provide a
technological challenge much more difficult that one might have been supposed some
years ago.
Photonics is the technology of generating and using light and other forms of
radiant energy, whose quantum unit is the photon, in conjunction with electronics. It
plays an important role in the development of robot sensors. For the development of
robot sensors many photonic effects have been exploited and photonic-effect-based
components used. Because of the unique properties of the emitted radiation, lasers have
been used in the design of a number of robot sensors. New possibilities for robot sensing
are also offered by fiber optics. In some cases these two photonic devices, lasers and
fiber fiber optics combined produce powerful new devices, with the features provided by
the laser light source and the optical fibers as light transmitting and modulating medium.
Lasers and fiber optics offer to robotic systems a wide range of possibilities for
measuring environmental parameters and for transmitting the information within the
system. They also broaden the application area of the intelligent robots to adverse
The subject of this text and the lecture is presentation of the optical fiber sensors
and systems developed for robotics and different applications that exploit unique
the applications fiber optics in industry and outlines the specific requirements to fiber
OPTICAL FIBER SENSORS AND SYSTEMS 221
optic systems for robotics. Chapter 4, the most broad in the presentation provides an
overview of the optical fiber sensors for robotics, presenting some basic phenomena,
sensor models, and their applications. Fiber optic lighting and imaging systems for
machine vision are the subject of Chapter 5. The last chapter presents general conclusions
and discussion of the further prospects for the applications of fiber optics in robotic
systems taking into account the current trends in both photonics and robotics.
There is a number of different optical fiber systems, like optical fiber sensors,
optical fiber data transmission systems, and optical fiber local area networks (OFLANs).
Optical fiber imageguides and optical fiber illuminators, widely used in many industrial
connectors
light
detector
source
optical
fiber
shown in Figure 2.1. Basic components of the most systems are: optical fiber(s), a light
222 J. A. MARSZALEC
optical fiber systems we will take a look at the optical fiber as a light transmission
A dielectric optical fiber is a small, transparent fiber that guides optical energy in
the form of visible light or infrared radiation. This fiber consists of an inner transparent
silica (glass) core which is surrounded by an outer transparent material, cladding. The
DIAMETER
';;IC-_- CLADDING
DIAMETER
300 \1m
h· 1.00143
reflections occur because of a difference in the refractive index (n) between the higher-
index core and lower index cladding. In the fiber only these light rays will be propagated
that reflects in the fiber at angles less then the acceptance half-angle. Other rays passing
through the cladding because of their reflection angle is greater than the acceptance half-
OPTICAL FIBER SENSORS AND SYSTEMS 223
angle will be lost. Acceptance cone definition and the described mechanism of light
RAY OUTSIDE
ACCEPTANCE
CONE
NUMERICAL APERTURE NA
There are many types of optical fibers produced. In a glass optical fiber group we
have step index multimode fibers, graded-index multimode fibers, single-mode fibers,
and multicomponent glass fibers. Optical fibers are also made of a few kinds of plastics.
fiber at wavelength 1300 nm is -0,5 dB/km; higher values for plastic fibers are
obtained;
dispersion.
224 J. A. MARSZALEC
Depends on the type of optical fiber system as a light sources usually are used
light emitting diodes (LED), semiconductor lasers, and white light sources. There are two
types of LEDs used, both shown in Figure 2.4: a surface emitting LED (Burrus type -
Fig.2.4a) and an edge emitting LED (Fig. 2.4b). The first one radiates into all directions
(generates a nearly Lambertian emission with a large NA-O,9), and approximately half of
optical power is lost. In order to reduce losses a lens is used to reduce NA down to 0,2.
Surface emitting LEDs offers low coupling efficiency. The edge emitting LED offers
much better coupling efficiency than surface emitting LED. The best coupling efficiency
of optical power into fibers guarantee the semiconductor laser (Figure 2.5), which
IIctol
Zn Diffusion
Uelal
a b
Figure 2.4. Light emitting diodes; a) surface emitting LED, b) edge emitting LED
operating characteristics depend on doping and structure of the device, among other
parameters. But it is the most expensive semiconductor light source. Spectral width of
surface LEDs is 80-100 nm, edge LEDs is 75-80 nm, and semiconductor laser is 0,8-2,5
nm. Lifetime for LEDs is 1000 years and for lasers is 100 years.
OPTICAL FIBER SENSORS AND SYSTEMS 225
White light sources are used in optical fiber sensors utilizing some form of
spectral filtering. Halogen and xenon lamps with an output overlapping spectral range of
Figurl
P-i-n silicon detectors are the most often used type of detectors in optical fiber
systems. There are three basic reasons for so widespread use of these detectors: they
have a wide sensitivity range (from picowatts to several miliwatts); they have good
linearity of output current response to the input light signal, and they have a good long-
term stability. A diagram of a silicon PIN photodiode is shown in Figure 2.6. This type of
detector is used in many data transmission systems and optical fiber sensors (working in
near infrared 850-900 nm). Other detectors used in optical fiber systems are made of
germanium, and multicomponent materials like InGaAsP or InGaP all for wavelengths
MeU I contact
(
n + substtate
Ru' c!ntact
.:: :; .";.';;
o
500 1000 1500 .-
Wavelength (nm)
A very Important pomt of fiber OptIC systems are fiber OptIC connectors (FIgure
because of the OptIcal power losses caused by bed couphng A few types of connectors
OPTICAL FIBER SENSORS AND SYSTEMS 227
are available on the market with coupling losses of 0,1-0,5 dB for coupling of multimode
optical fibers.
---
.301.1"'_1
--,
, oa
-.....
7
,
, """ ••
, "'"
10._
'tOIl- I G I -lGDil
, """ > ·
-
00 1
__ ~_;r
, 00
•,
,
_oG. ....
""" ~I"IQII
REFERENCE
Fiber optics has been introduced into the industrial environment in the form of
four main groups of devices and systems, namely, the optical fiber sensors for measuring
environmental parameters and for process control, optical fiber communication systems,
the industrial applications of the optical fiber devices and systems is presented in Table
3.1. Many types of sensors have been developed using optical fibers for measurements of
acceleration, rotation angle, electric and magnetic fields' parameters, and spectral
properties of materials [1, 2]. By combining the optical fiber sensors with optical fiber
the measurements and control in the explosive environments of the petrochemical plant
(Mizushima Oil Refinery of Nippon Mining Company, Figure 3.1 [3]). Most of the
optical sensors, presented in the diagram, for the measurements of the physical and
interconnecting the computers and the distributed stations used in this application.
Iccl
Figure 3.1. Diagram of the fiber-optics-based sensing and data transmission system of
information, transmission of high power laser radiation, and for illumination and image
a relatively short length of optical fibers (from several meters to over a dozen
meters for sensors, and to a few hundreds of meters for data transmission
systems),
(e.g. depending on the type of the optical fiber sensor PCS (plastic clad silica) fibers,
230 J. A. MARSZALEC
HCS (hard clad silica) fibers, or plastic fibers). The specific type of fiber to be used in a
requirements. For some types of sensors such types of fibers can be used, which possess
certain special features for sensing of the considered quantity (e.g. HiBi-highly
transparent fibers for Faraday-effect current). For most optical fiber illuminators a small
diameter glass optical fibers are appropriate as well as plastic fibers. It is important to
note, that such parameters of optical fibers as attenuation and dispersion, that are very
important for long distance communication are not as important for robotics applications.
Mechanical properties of the fibers and their protection against damage constitute
including their dynamic operation on the robot, and possibilities of being damaged by
man or machinery. The fiber itself should have good mechanical properties (including
fatigue strength), and should additionally be protected using plastic and/or flexible steel
jacket.
REFERENCES
1. Dakin 1., Culshaw B., Editors, Optical Fiber Sensors, Vol, 1: Principles and
Components, Vol. 2: Systems and Applications, Artech House, Norwood, MA,
1989.
2. Udd E., Editor, Fiber Optic Sensors, An Introduction for Engineers and
Scientists, A Wiley Interscience Publication, John Wiley & Sons, Inc., New York,
1991.
4. Marszalec E., Marszalec J., Intelligent Optically Powered Robot Gripper, Sensor
Fusion: Spatial Reasoning and Scene Interpretation, P.S. Schenker Edit., Proc. of
SPIE, Vol. 1003, pp.284-290, 1988.
omCAL FIBER SENSORS AND SYSTEMS 231
Optical fiber sensors are a relatively new type of measuring devices. They have
wide area of applications from medicine and biology to industry, military areas and space
exploration. Much research on the subject has being carried out at different laboratories
over the last decade, phenomena in optical fibers have being studied, and new sensors
have been developed. Some of the sensors are already comercially available from a
number of companies. Robotics has also been identified as a branch which can take
advantages of optical fiber sensor technology and use optical fiber sensors for robot
control in the execution of different tasks in industrial and other environments. In 1988
the market for optical fiber sensors in robotics reached $0,5 million and is predicted to
grow to almost $5 million in 1996, with an annual growth rate over 30% for this period
[1].
environmental parameters. However, one can define a general optical-fibre sensor, and
Figure 4.1 shows a schematic diagram of such a device [2]. A constant light source
launched into an optical fiber to a region in which the light is modulated in one of the
above mentioned constant properties. The light is then returned from the modulation zone
along another fiber to be detected and demodulated. The main element of optical fiber
sensors is the optical fiber itself. The dialectric optical fiber is a small, transparent,
usually glass fiber that guides optical energy in the form of visible light or infrared
232 J. A. MARSZALEC
radiation. The basis of guiding the light via the fiber is the total internal reflection effect.
Light entering one end of a fiber, within the fiber's numerical aperture (or acceptance
cone, as depictated in Figure 2.3), will undergo total internal reflection at the core -
cladding interface repeatedly as it propagates down the length of the fiber. This effect is
There are two basic types of optical fiber sensors. The first, extrinsic or externally
modulated sensor, in which the fiber serves as a light pipe to deliver light to a transducer
and to receive light modulated by the transducer. The modulation process takes place
externally from the fiber, usually by way of an attenuation process, which is modulated
by indirect processes.
light SQV<'C:.
-0- , .. d l b
i .r
I -- ----;'::i.t;,-;::urn l ib.r
light demodulator
feed
return
--'--'---
D
-1 I
modulator
mf'asuran
d
'=.
_
d
~ measurand
--,""'.',...--ur-n--.J...... modutator
., i thin
' ib. r
Figure 4.1. General features of an optical fiber sensor showing the distinction between
that in which the measurand interacts directly with the light guided within the optical
fibre. Phase, polarisation and intensity may all be modulated within the fiber, using an
commonly used to describe the latter class. In intensity - modulated - sensors the
the sensing region. Other commonly used approaches involve spectral modulation,
scattering, and reflectance. Intensity - based sensors have been developed for primarily
temperature, pressure, position, displacement, flow, liquid level, chemical sensing, etc.
shifts between two or more coherent optical fields, and are generally based on fiber
propagate in a common fiber. Interferometric optical fiber sensors offer the highest
sensitivity of a wide range of measurands. Sensors based on hybrid systems have also
received attantion. Phase - modulated sensors are usually high performance sensors for
inertial navigation systems, to measure such parameters as rotation, and acceleration, and
high sensitivity devices for magnetic fields, electric fields, acoustic fields, strain,
Based on the presented considerations and the requirements for robot sensors, it
can be stated that optical fiber technology offers a great variety of new sensing principles
234 J. A. MARSZALEC
in constructing sensors and measuring systems for robots. Although experiments on fiber
optics in robotics are relatively new, some solutions have already become evident. The
small size and low weight, especially in systems where multiple sensors and long
robot gripper, which is often exposed to the harmful action of the environment;
reqirements.
Considering the requirements stated by robotics there have been a few types of
optical fiber sensors developed for robotics applications, namely optical fiber proximity
sensors and multiple proximity sensing systems, optical fiber tactile sensors, slip sensors,
and colour sensors. Some other types of optical fiber sensors developed for other
applicattions can also succesfully be used in robotics systems. A list of such sensors
includes optical fiber gyroscopes, optical fiber gas sensors, optical fiber electric field and
magnetic field sensors, temperature sensors and others. The discussion of particular types
Optical fiber proximity sensors are intensity modulated reflective sensors. The
sensors consist of two optical fiber branches: a transmitting branch with a transmitter (an
ifrared LED or a laser diode with driving electronics) and a transmitting fiber, and a
measuring head of the sensor consists of the free endings of the transmitting and the
receiving optical fibers, which are kept together in a special housing in a certain
geometrical relation to one another. A diagram of the sensor is shown in Figure 4.2. In
the sensor, light emitted by the light source is led to the illuminating fiber, transmitted by
the fiber and then sent towards the surface whose distance from the sensor head is to be
measured. A part of the light reflected by the surface and located within the cone of the
numerical aperture of the receiving fiber is accepted by this fiber and transmitted to the
photodetector. Parameters important for the sensor design are: diameter of optical fibers
and their numerical aperture, separation of the fibers in the sensor head, and the angle of
photodiode
reference
phOlodiode
In the sensor, optical power losses occur in coupling the light source with the
fiber, the detector with the fiber, in the optical fibers, and in the sensitive volume of the
sensor (between the sensor head and the reflecting surface). Among the enumerated
losses in optical power transmitted in the sensor, only those in the sensitive volume are a
function of the distance between the sensor head and the reflecting surface, and their
distribution determines the shapes of the sensor characteristics [3, 4]. Other losses'
components are sensor design dependent and are constant for a given sensor structure.
A voltage output signal of the optical fiber proximity sensor can be expressed by
a simplified equation:
where
K - is a constant coefficient for a given design of the sensor and characteristics of the
reflective surface,
T(z)- is a nonlinear function of the distance, taking into account optical phenomena
A typical characteristic of the sensor (the generated voltage is a function of the distance
between object and sensor head) and a profile of its sensitive volume are shown in Figure
4.3. Depending on the sensor design (especially diameters of optical fibers used) the
absolute measuring range of the sensor can be from a few milimeters to about 200 mm.
OPTICAL FIBER SENSORS AND SYSTEMS 237
The optical fiber proximity sensor has a nonlinear characteristic with high
sensitivity front slope and lower sensitivity back slope. The characteristic curve reaches
the maximum value at a distance of a few milimeters. Note that the distance is a double -
valued function of the voltage output. For robotics applications the back slope of the
curve is more often used. This sensor provides a narrow cigar - shaped sensitive volume,
elongated in one direction. If focusing optics are used in the sensor head for the
transmitting and the receiving fiber the sensitive volume of the sensor is permanently
focused at a certain distance (typically a few centimeters) in front of the sensor head. The
length of optical fibers used in the sensor can vary from 1 meter to more than 10 meters,
depending on applications.
X[mml
THEORETICAL P~OFI LE
OF SENSITIVE VOLUME
A B
a profile of the sensitive volume (no focusing optics used in the sensor head)
238 J. A. MARSZALEC
Sensor heads of optical fiber proximity sensors have small dimensions (linear
dimensions of a few milimeters) and light weight. At the same time the sensitive volume
of a single sensor is relatively narrow and elongated in one direction, so a single sensor
ensures proximity sensing in a relatively small section of the space around the robot
gripper, and may therefore be used for limited kind of tasks. In order to broaden the
possibilities of this type of sensors to perform more complicated tasks, small size
multidirectional sensor modules have been designed [5]. It allows the placing of a
number of sensors into a robot gripper in the desired manner, and to design a multiple
optical fiber proximity sensing system with possibilities to measure distances in different
Sensors are distributed in the gripper as shown in Figure 4.4a, and are placed on both
palms, four sensors on each palm. Optical power is supplied to the sensors from an eight
- channel optical power supply system designed so that each channel (and, hence, each
sensor of the system) works in a different frequency band. This allows the eliminatation
of the possibility of a given sensor reading a signal reflected from an object to which
distances are measured, and sent by the neighbouring proximity sensor. The sensors used
accuracy of 3%. Optical fibers used in the sensor design are fiber bundles of diameter of
orientation in relation to the measuring head. This constraint is not so strong for diffuse
reflecting surfaces as for specular reflecting surfaces (see diagram shown in Figure 4.5).
The described system can be used for dynamic range data acquisition for pose estimation
, __RQBJlI_G81e?i:1i - ,-
I SYSTEM OF I
: SENSOR HEADS :
-
I
1::[ OFPS No.1 I z i---4--- 1'" -"
u 1--
f---
M ~-
-'
,<
In. OFPS No.2i x r::=;= ~ --1 2 f---
'I- ---13 f---
'a OFPS No.3 1 -x I-+- jLlJ 1- - 0: 0: 1
Iii: -tit- --I 4 f--- w <.:) HOISPLAY
I OFPS No.4 ' y I=i=
, X I-
LlJ Z
u;::[
, w I--
-'
C
_w
<
0:
>
I- ",LlJ
wI-
I Z u'"
~~
OFPS No.5, z !==i= ~ -""5 f--- -
n. 0
u
0>-
o.. '"
0:
I
I~
-W ~ 1--
t-
-'
,-' ~--I5 f---
OFPS No.5 1 x I=i= LlJ 0: ~-
;:)
I ff:.
&---17 f--- ::[
It;: OFPS No.7 1-x
!==i= sl:il ,..-
ILlJ
1-' OFPS No.8 I y I=i= tii!- ---I 8 c-- I -
----r- -- -.J
I
,
,- Y OPTICAL POWER
SYPPL Y S YSTEM DRIVES
OF ROBOT I
a) ARM A ND GR IPPERI
b)
Figure 4.4. A multiple optical fiber proximity sensing system with eight measuring
channels, a) a diagram of the system, b) a view of the laboratory model of the system
240 J. A. MARSZALEC
100
1.3: z 50 mrre
2,4:z - 100mm
/ ..... "'-,
10
1=
I [/
/
\ '-
"-
-=
I 4-
0.1
-20 -10 o 10 20 deg
Figure 4.5. Output signals of an optical fiber proximity sensor for different orientation
angles of reflecting surfaces (z - distance, 1,2 - for diffusely reflecting surface, 3,4 - for
In the areas of high technology, including the electronic industry, and production
of medical and food products those places requiring a particularly high degree of
cleanliness are installing clean rooms. In order to assure a high degree of cleanliness in
those rooms it is vitally important to measure the volume of dust leakage from the air
personnel as operations are performed high above the ground. A Japanese company
Hazawa - Gumi [7] has developed a leak check robot called Leak Robo, to robotize the
the position of the filter has been provided. The position measurements are performed by
two types of specially designed sensor units which use optical fiber proximity sensors in
their construction.
There are two types of filter being used in clean rooms, those with a big level
difference between the frame of the filter and the filter surface (Figure 4.6a), and those
with a small level difference (Figure 4.6b). For detection and tracing frames and filters'
OPTICAL FIBER SENSORS AND SYSTEMS 241
surfaces two types of position detection sensor units have been developed. Simphfled
diagrams of the units with an illustration of the ranges of sensors' detection are shown m
Figure 4.7.
\ \ ~ II
~ Mount ed metal
~<-~/ ;><,,~
0 35 -50"" L 0~ -I'
a t y p c tJ _ l Y P e
Figure 4.6. Two types of purification filters being used in clean rooms
-tlr D
1:"
~~;
~.....
y iew f roll ab ove
);;..~
~ -_.
( a t y p e ) ( b _ t y p e )
FIgure 4.7. PosItion detection sensors for two types of fIlters bemg used m clean rooms
242 J. A. MARSZALEC
For reliable detection of the filter position a layout of the optical fiber proximity
sensors in the detection sensor units has been determined by studying different drive
patterns of the detection unit with the robot. The layout of proximity sensors in units for
position detection of filters of types "a" and "b" is the same, but because of the
construction differences of the filters the proximity sensor have different protrusion from
For the detection of the position of the filter with a big difference in level between
the filter frame and the installation unit (as much as 35 to 50 mm) "a" type sensor unit is
used. This sensor unit can detect the position of the side surface of the installation unit
frame (joined perpendicularly to the filter surface), and trace the surface of the
installation frame by moving the boom in a set direction while maintaining a fixed
distance from the filter surface (about 10 mm), with one proximity sensor.
The "b" type sensor unit permits the tracing of the boundary line between the
filter frame and the filter surface with the possibility to detect the filter frame position
through response by two proximity sensors at the same time. When the central sensor and
one of the peripherial sensors straddles the filter frame, distances measured by these
sensors decrease. After unifying the detection of direction of the sensor, and comparing
the measurements taken at filter surface and filter frame the detection of the filter frame
is performed. Since the level difference between the filter frame and the filter surface is
not big (about 10 mm) the detection of the filter frame in this case is more difficult than
in case "a".
The detection operations described have been patterned and programmed. The
millimeters. The sensor unit is moved in the direction of the filter by stepping motor with
a resolution of 0,089 mm/step. Robot movement sequences when detecting filter position
are shown in Figure 4.8. In order to prevent contact of the sensor unit mounted in the
robot arm with the filter, and scratching the filter surface, each distance sensor is scanned
by the robot controller within 1 ms, and safe drive control has been achived.
Y AX is
Figure 4.8. Sequences of robot movements when detecting position of the filter
sensors in robotic systems has been presented. Other cases can be found in the technical
literature. It can be seen that optical fiber proximity sensors have also been used, for
example:
for dynamic range data acquisition for pose estimation, using multiple optical
It seems that these sensors, now also comercially available from a number of
manufacturers, made of different type of glass or plastic fibers, will find even more
Tactile sensors have been the object of intense research in many laboratories in
the world. It is increasingly evident that some limitations of vision systems could be
overcome by the cooperative use of tactile sensors, and multisensory systems with vision,
tactile and other sensing capabilities, are a good base for the development of future
generations of robots.
There have been tactile sensors developed which are based on direct measurement
force transducers, that is devices which measure the mechanical deformation produced by
a force acting directly on the transducer itself. Tactile sensors are based on 2-D arrays of
such measuring "cells". Like their human analogue, in fact, robotic tactile sensors work
by measuring the contact pressure between the object surface and the robot gripper. By
dividing the contact area into an array of a number of measuring points (for example 8x8
sensing" is herewith defined as the continous measurement of contact pressure within the
array of so called "tactels", or tactile elements, as distinct from "touch sensing" which is
only single-point measurements, are often adequate for sensing the presence of an object
and/or the force exerted on its surface. Optical fiber touch sensors developed for the
robot gripper, all are based on the reflective sensor principle described in the previous
section. In optical fiber touch sensors transmitting and receiving ends of optical fibers
which are placed in the sensor head are mounted together in a closed casing, and
reflecting surface is usually formed by an internal surface of a moving part of the casing,
touch/force sensor is illustrated in Figure 4.9 [11]. The approach is analysed on the
reflecting surface mounted on a spiral spring element with a certain stiffness. First, from
the characteristic of an optical fiber reflective sensor (shown in Figure 4.9a) the most
appropriate part for the touch/force sensing purposes is determined, forming the
measuring range of the touch sensor. The reflective characteristic of the sensor should be
for a real geometry of an optical part of the sensor. For the case described, in the sensor a
low diameter (1,5 mm) concentric optical fiber bundle is used, with a transmitting central
part and receiving external one, but the same approach is valid when single transmitting
246 J. A. MARSZALEC
and receiving fibers are used. To the determined measuring range of the sensor (i.e.
displacement of the reflective surface) the length of the spring and its stiffness should be
matched, in order to assure the required touch sensitivity and force measuring range.
Letters T and R indicate the transmitting and receiving parts of the fiber bundle
According to Hook's Law for metal springs the force F acting on the spring is
spring. So, taking into account the displacement measurements versus voltage in the
design approach one has universal results, which are independent of the spring
characteristics, and thus by the simple operation of exchanging springs in the sensor head
A reflective intensity modulated optical fiber touch sensor with a cantilever beam
as a reflector, has been developed for the control of a multifingered robot gripper [12].
The sensing principle and the sensor design are presented in Figure 4.10. White light
from a source is transmitted through an optical fiber to be reflected off the bottom surface
of the cantilever beam into a receiving fiber. A pin fixed to the beam protrudes up from
the cantilever beam. Due to the load applied to the pin, the cantilever beam is depressed,
and the amount of light reflected into the receiving fiber varies. By measuring the amount
determine the relationship between force applied (proportional to depression and the
beam stiffness coefficient) versus light received. The fibers used in the sensor prototype
are plastic fiber with a numerical aperture of 0,5 and a diameter of 1 mm.
OPTICAL FIBER SENSORS AND SYSTEMS 247
a)
o Imn
I MOVEiC TIP I
I I
I
8 NA I
b)
I
"
I
I I
L _ ____ _ _ _ _ _ _ _ J
FIgure 4.9. A design approach and a diagram of a reflective intensity modulated optIcal
fIber touch sensor; a) measuring range of a touch sensor determined from an optical fIber
reflective sensor characteristic, and geometry of the sensor head, b) a diagram of the
1 Top 12 Spacer 4
2 Baoa 13 Spacer 5
3 nan 14 ~111.1
4 Flbr...... embly 15 Dowel Pin
5 Spring 16 Laad !Dew ID vaty d
6 ALt>ber Tube 17 Load SO' .... 10 va')" e
7 e.am ...._y 18 Tlnaded Do..... Pin
8 Preuura PlaM Assembly 19 SodtOI Hoad Sao...
9 Spacer 1 20 Sal Setaw
10 Spacer 2 21 Oplical Fibf.
11 Spac:er 3 22 Rubbe< Pad
Figure 4.10. A modular optical fiber touch sensor with cantilever beam; a) schematic
fibers, q - angle between the fibers and the horizontal plane, h - the initial height of the
Based on the experiments with single - point sensor a four - point sensor was
designed. This sensor consists of a matrix of sensing sites, each using a small cantilever
beam to sense force locally. These beams are cut out of a large single sheet covering all
the sensing sites. The design is modular in the sense that different ranges of acting forces
can be measured by simply changing the cantilever beam sheet for another one with a
different stiffness.
Externally modulated optical fiber tactile sensors are sensors in which modulation
of light by the external force takes place externally from the fibers. For this type of
sensor the aplied modulation principle is the same as for the touch sensors described in
the previous section (4.3.1), i.e. reflection of light from the surface disturbed by an
external acting force. These sensors in the form of measuring cells are of small
dimensions and designed so, that their integration into the form of tactile matrix for
A tactile sensing concept using infrared light and optical fibers [13] is presented
in Figure 4.11. The sensor is built on a solid lattice or honeycomb frame. The top of the
frame is covered with rubber or other elastic material with a reflecting botton
surface.This cover material is the contact area between an object and the robot gripper.
Infrared light of constant intensity is transmitted into each cell by a thin optical fiber
pointed towards the elastic cover material from below. Another optical fiber is connected
to a photodetector. A small gap (typically a few milimeters) separates the ending of the
light emitting and receiving fibers from the bottom surface of the elastic cover material.
When the cover surface is pressed from above, it moves slightly into the cell. This causes
250 J. A. MARSZALEC
the amount of the light reflected back to the receiving fiber from the inner surface of the
elastic cover to change. The changing amount of reflected light generates a voltage signal
indicating that pressure is acting on the cell. The dimensions of single cells indicated in
Figure 4.11 were to investigate the physical parameters of the sensor. The size of a
sensitive cell can be smaller e.g. 2,5 by 2,5 mm. With a 1 mm separation between cells
TO' Vlr w
IJNOEflilllASTlC COVEJIII SIOE Y1lw
I-sm-J PAESSURE
TSmm 2mm
00 00 00 .0
_L ......r----L..----,-L
00 00 00 00 T
00 .0 00
MA rERIAL
00 FleER CA81.E
OR lICHT
PIPE CONOUCTING
REFLECTED
fl8ER CA. BLE
LIGHT TO SENSOR
OR LtGHr
PIPE FROM
LIGHT SOURce
lIGrd TO SENSOA
Figure 4.11. Tactile sensing concept using infrared light and optical fibers
The high density optical fiber tactile sensing system [14], presented schematically
in Figure 4.12 uses each fiber for both transmitting and receiving purposes. In this design,
light is directed from the source to a beam splitter. The reflected part of the light from the
beam splitter is directed to a specially designed low resolution coherent optical fiber
bundle. The fibers in the bundle are of 250 mm diameter. The light travels down this
bundle and is reflected by the elastic reflector back into the same bundle. The reflected
light from the bundle is again directed to the beam splitter and part of the light
transmitted through the beam splitter is deflected to a video camera where the tactile
OPTICAL FIBER SENSORS AND SYSTEMS 251
image obtained is processed. A camera lens is chosen so that the viewed bundle fills the
screen. The sensor developed comprised of 25 layers of 26 fibers in each layer, with a
spacing of 1 fiber per 1 mm. This yields a fiber density of approximately 100 fibers per
cm 2 .
Figure 4.12. A tactile sensing system using a low resolution coherent optical fiber bundle
sensors it should be stated that both constructions presented for the sensors are complex,
Internally modulated optical fiber tactile sensors are the sensors in which the
external force interacts directly with the light guided within the optical fiber. These
sensors seem to be the most promising type of optical fiber tactile sensors for tactile
252 J. A. MARSZALEC
imaging in robotics, because of small size of a tactile matrix and easy construction. The
developed internally modulated intensity optical fiber tactile sensors exploit two effects
in optical fibers, i.e. the microbending effect in optical fibers, and the scattering of light
at fiber bounadry.
Microbends are small radius bends, which, when imposed on fibers, tend to
couple high - order guided modes into radiation modes, causing a power loss that reduce
the intensity of light at the fiber end. Changes of the intensity of the transmitted light are
recorded by an optical receiver and analyzed. This effect is especially prominent when
several such bends are spaced periodically along a fiber. Figure 4.13 ilustrates the
principle of a microbending sensor in which the fiber is bent sharply when force is
applied to metal teeth or an array of pins. For a graded - index multimode fiber,
maximum microbend attenuation for a given distortion amplitude occurs for a distortion
Ac = 21t a n / NA (4.2)
where a is the fiber core radius, n is the peak index of refraction of the core, and NA is
the fiber's numerical aperture. For typical multimode optical fibers, Lc is in the order of a
Pressure
I JL I
~====~O""'-
:===:==j
Q Q
=F .,.,..., Core
I I
I_ Ae _I
developed for monitoring the force field distribution [17]. A diagram of the sensor is
shown in Figure 4.14. Intensity modulated light from LED source is launched into a
sixteen port star multimode optical fiber coupler. Each output port of the coupler is
followed by a fiber optic microbending sensor. Folowing the analysis for one branch, any
light coupled into the cladding of the fiber is removed in a cladding mode stripper placed
before the modulation device. This has been ackomplished by coating the fiber with an
index matching liquid (in this case glycerine), so that light coupled into the cladding
propagates out into the liquid. When the light passes through the mode stripping region
into the microbending sensor, the amount of light coupled into the cladding region
depends on how tightly the fiber is bent. As the fiber bending radius varies due to the
motion of the bending plate with the acting force, the light coupled into the cladding
varies in a similar way. The light beam then exits the microbending region and passes
through a second mode stripper, where light propagating in the cladding is removed. The
light that finally reaches the PIN detector is amplitude modulated by an amount that
corresponds to the motion of the pressure acting plate. The electrical output pulses from
the photodetectors are fed to a 16-channel analog multiplexer, and then through an 8-bit
analog - to - digital converter (ADC). The processor analyzes the output signals of each
sensor and displays the distribution profile of the force field affecting the sensor. In the
sensor graded - index fibers with a core diameter of 50 mm and a cladding diameter of
125 rum are used. The mechanical wavelength of each microbend sensor is 2,27 mm.
Static and dynamic tests have been performed on this sensor. In static tests, a
known arbitrary weight has been applied to each sensor of the array. The actual and
measured weight distribution profiles are shown in Table 4.1 [18]. In the dynamic tests
vibrations with frequency up to a few kHz have been applied to each sensor. By
254 J. A. MARSZALEC
analysing the AC signals obtained, the amplitude and frequency of each vibration has
Microbend Photo-
Sensors detectors
lx16
Fiber Pro-
LED Optic ADC
cessor
Coupler
~
EJ
Figure 4.14. A diagram of an optical fiber force field distribution sensor
actual values
Based on the microbending principle a four layer optical fiber tactile sensor array
has also been developed [19]. The principle of the sensor design is illustrated in Figure
4.15. Layers 1 and 4 serve as corrugations and layers 2 and 3 are the active sensor fibers.
Layer 4 fibers are mounted on a solid sensor base. The corrugation layers might, in
general, be made of material other than optical fibers. However, for the prototype sensor,
optical fibers were used because of the same diameter and dynamic response as the active
sensor fibers. When force is acting on a single intersection of layers 1 and 2 (to a pixel)
as shown, a single fiber in layer 2 is bent by the applied force. Two fibers in layer 3 on
either side of the applied force are also deformed due to the bending of layer 2 and the
corrugation of layer 4.
LAVER 3
(se<cnd.,.,)
LAVER, LAVER 2
(corrugahon) (P"" "'"
Figure 4.15. The design principle of a four layer microbending optical fiber tactile sensor
array
of the sensor. A diagram of the laboratory prototope is shown in Figure 4.16. The active
array was made up of five by six fibers. To ensure that active fibers near the edge of the
sensor react similarly to fibers in the middle of the device, inactive or dummy fibers were
256 J. A. MARSZALEC
placed on the outsides of the active fiber groups, making the array seven by eight fibers.
The horizontal spacing between all fibers was 2 mm. This spacing was chosen mainly for
ease of sensor construction and testing. The base of the sensor was an aluminium block,
and the fibers were mounted onto the base using fast-drying epoxy glue approximately I
cm from either edge of the sensor area. Vertical spacing between the layers in the glue
I I
I I
.. "'--VIEWB
,
I
I
VIEW,.
VIEW A:
V, VE, ~I
COARUa.t.f
Fli!EAS
t,.AYEq"
COR.R UG~T1()N
F I 8E~S
VIEW B:
Figure 4.16. A diagram of the laboratory prototype of the tactile sensor array
The four-layer rnicrobend optical fiber tactile sensor array described is relatively
simple and inexpensive in construction, and its size and performance characteristics can
OPTICAL FIBER SENSORS AND SYSTEMS 257
easily be tailored to meet specific needs through appropriate choices of fiber type and
other design parameters. As the sensor is made of flexible fibers, it can also be made in a
skin will be developed to cover the sensor structure and to act as the sensor surface to
Another phenomena used for the development of the optical fiber tactile sensor
array came from the studies of mechanisms responsible for propagation loss in
communication optical fibers. Based on the observations it was found that some low-loss
bulk: glasses have many spherical inhomogeneties up to 1.0 mm. Such spherical
inhomogeneties might be stretched by the fiber drawing process and survive as very long,
very narrow dielectric needles precisely aligned with the fiber axis. From such long, but
finite, dielectric needle radiation transmitted within the fiber is scattered, with a certain
incident on such needle-like structures will couple into a receptor optical fiber along the
fiber's axis. The additional factors of spatial proximity, fiber diameter and relative
Figure 4.17. llustration of the principle of light scattered at a fiber boundary due to
surface roughness
258 J. A. MARSZALEC
orientation will detennine the overall coupling efficiency. To exploit this phenomenon
for tactile sensing the objective is to produce a sufficient amount of light from the
transmitting fiber in a direction normal to its axix in order to couple that light to another
The sensor based on the concept described uses optical fibers which have uniform
optical cross section - there is no core-cladding interface. Thus any abrasion to the
surface results in broadband phase-matched coupling between the propagating light in the
fiber and radiation into free space. The fibers are commercial plastic fibers with
diameters available from 150 mm to Imm. Total efficiency improves for smaller
diameter fibers because the effect of surface abrasion on the modal distribution of guided
light is greater. Furthermore, smaller fibers also offer higher spatial resolution.
The tactile sensor in the configuration developed and tested is shown in Figure
4.18 [20]. It consists of four optical fibers attached to a substrate plate to form a linear
array of equal spacing. A similar array of fibers is oriented perpendicularly to the first
set. A matrix of sensor sites is created by the row and column arrangement of the fibers.
The rows are separated from the columns by either a transparent elastomer or opaque
elastomer via holes machined to permit direct optical coupling between fiber crossings.
Force applied to a wear surface covering the composite structure compresses the
elastomer and increases the coupling between them at the sensing site by decreasing their
between upper and lower fibers and the light coupling between them.
The surface of each fiber was abraded near the points of crossing to facilitate the
coupling of light from one fiber to the other. Normally, no light would be radiated from
the fibers for the kind of lateral deformation experienced in this design.
OPTICAL FIBER SENSORS AND SYSTEMS 259
Figure 4.18. Configuration of elastomeric layer sandwiched between orthogonal rows and
A block diagram of the tactile sensor system designed is presented in Figure 4.19.
The prototype is a 4x4 matrix array covering 5x5 cm2 area. Optical fibers used are of the
diameter of 760 J.l.m. The input row fibers are excited by LEDs in a clocked sequential
manner. During the time interval in which one row fiber is illuminated, all output fibers
are sequentially polled by reading their respective photodetectors. The detected signal
level from each output photodetector is converted from analog to digital output by an
eight bit analog to digital converter (ADC).Theoretically this provides a 256 gray scale
resolution of force, or better than 0,5%. In practice, however, resolution is closer to 1%.
It is beneficial for the sensor performance if the elastomer used for the sensor
design has high shear compliance so that compression at one site does not induce an
CLOCK
f- DATA
~ -
--l f-L-
AID CO~'VERTER & MUX
f-r- - ADDRESS
f- -
CO~1PlrrER
f-
f-~
SOL'RCE
4x4 MATRIX HI-POWER LED SOURCES CO('.IROL
LOGIC
Figure 4.19. A diagram of the optical fiber tactile sensor system based on coupling of
New versions of the sensor with higher spatial resolution, faster sensor data
acquisition, increased array size and improved other parameters are under development.
The sensors will be used for tactile imaging, including 3-D mapping by touch probing,
Slip sensors detect and, if required, measure the relative movement between
sensor located in the gripper and an object. This sensor is particularly important in
applications where the system has no knowledge of the objects to be grasped and
sufficient force has to be applied to prevent movement. If insufficient force is applied, the
object may move under gravity. If an excess is applied and an object is of an irregular
OPTICAL FIBER SENSORS AND SYSTEMS 261
shape, it may be ejected from the gripper. Information on slip is also important for robot
control when opening the gripper and placing the object at a certain place after
transferring it. When the object touches the surface of placement its movement is in the
opposite direction, and the slip signal can be used to control the drive to open the gripper.
Apart from specially designed slip sensors the detection of slip can also be achieved by
In an optical fiber slip sensor the ability of some translucent elastomers to rotate
the plane of polarization of light passing through them when subjected to stress has been
exploited [21]. The sensor relies on the friction between its contact pad and the object to
detect its relative movement. As the object moves is carries with it the sensor's surface,
thus distorting it and creating a stress/strain region within it. At certain displacements the
restoring forces overcome friction between the two surfaces, releasing the stress/strain in
the sensor. This process repeats itself, enabling the detection of movement by monitoring
stresses. Two types of optical fiber slip sensors with photoelastic material are shown
A typical output characteristic of the optical fiber slip sensor is shown in Figure
4.21. From the characteristic it can be seen that the occurrence of slip can be easily
detected. It was also found that the slip signal depends on the condition of the slipping
surface, and the sensitivity of the sensor can be increased by artificially roughening the
surface area of the sensor. In addition to slip detection, the information from the sensor
can be used to determine the roughness of the grasped object's surface by measurement
of the vibration characteristics. The sensors developed have slip sensing areas of
approximately 1 mm2. The optical fibers used to guide light to and from the sensor are 1
mm diameter plastic fibers. The accuracy of the photoelastic slip sensors depends on the
262 J. A. MARSZALEC
stability of the photoelastic material, which may vary due to time, temperature, and
applied pressure.
object
lecto r
Figure 4.20. Optical fiber slip sensor with photoelastic material a) transmission type, b)
reflective type
-I
-2
-3
-4
10
T i rn~,!o
Figure 4.21. A typical output characteristic of the optical fiber slip sensor
OPTICAL FIBER SENSORS AND SYSTEMS 263
The perception of color constitutes one of the most valuable faculties of man's
senses. Information about color, obtained by means of sight, is utilized by man not only
in creating aesthetically attractive surroundings and individual objects, but also in other
domains, including automation and robotics. In robotics color information has been
found to be important for recognition and identification of objects and for designing the
machine vision systems employed for this purpose. As a recognition criterion color is
employed in two cases: (1) when the recognized objects have the same shape and differ
only in color, and (2) when the recognized objects differ in shape but the time of
criteria is too long (and does not allow real time control of the robot during task
execution).
the recognition and identification of the color of an object from a given, a priori
the recognition and identification of the color of any random colored object.
In the first case the color recognition system is at first informed about the set of
colors to be distinguished and then identification concerns only objects from this set. In
the other case the system recognizes objects of any color without any prior information
about them. The choice of the manner of color recognition depends on the kind of
operations to be performed by the robot, the universality of the job, the number of colors
264 J. A. MARSZALEC
to be identified, and on other factors. The realization of the first task is easier, the
systems for its performance is simpler, usually cheaper and the time of color recognition
intensity is modulated by the color of an object examined. Depending on the kind of the
object its color can be determined either by transmission or refelctive measurements. For
color recognition in most robotics applications, reflective color sensors are used.
However, transmission color sensors can be useful for some tasks in laboratory robotics,
where liquids or translucent materials are the subject of measurements (e.g. in chemistry,
biotechnology, etc.).
Universal systems for the determination of the color of objects are colorimeters
of the sample. Spectrophotometers are devices which do not measure color directly, but
perform measurement of spectral energy distribution from the sample. Based on the
software. Spectrophotometers and colorimeters are very good at determining both the
absolute color of an object and very subtle color changes from object to object. Some
head, which can be located at a long distance (up to several or dozens of meters) from the
light source and the receiver. Optical fiber heads are small and can be mounted in small
areas, for example in the vicinity of the robot gripper. However, colorimeters and
spectrophotometers or colorimeters. In such cases simple color sensors, which are based
Among a few color sensors developed for the recognition of colors of objects
sensor [22, 24]. The basis for the sensor design is a reflective spectrum analysis method
[22,26]. In the proposed method of recognizing colors, first an off-line spectral analysis
of the light flux reflected from the colored objects is performed for all the objects used in
the process, and based on the measurements only the most appropriate parts of the whole
visible spectrum are analysed by the sensor. The part of spectrum chosen for use in color
recognition is cut out from the whole spectrum reflected by the object by using an
A diagram of the optical fiber color sensor is shown in Figure 4.22. The sensor
fiber directing the radiation to the surface of the examined object. The receiving branch
interference filters and a set of photodiodes. The number of filters and photodiodes in the
receiving part is equal to that of the receiving branches and depends on the number of the
reference filter and a reference photodiode. The output signals from the measuring
photodiodes and the reference photodiode, after initial amplification and analogue-digital
conversion, are directed to the information processing system. The latter system
266 J. A. MARSZALEC
formulates decisions about the color of the identified object and sends them to the robot's
control system.
a.
L
<{
..J a::
Z UJ
UJ a.
~
a.
o o
..J ::r:: SENSOR
HEAD
* _I
<{ U
::r::
ILLUMINATING FIBRE
REFERENCE FBRE
L
L UJ
UJ l-
l- V)
V! >-
>-
V!
V)
~
..J z
o iii
a:: VI
I- UJ
z U
o o
u a::
oI- a.
FROM
SOURC::
I I \- TO
~''''"''
(a)
(e)
Figure 4.23. Optical fiber color sensor heads; a) for a small number of colors to be
Several variants of measuring heads have been designed for the sensor (Figure
4.23). In keeping with the 00 /45 0 condition, the illumination takes place in a direction
normal to the colored surface to be recognized, while the reception of the reflected light
is at an angle of 45 0 .
The color sensor discribed was tested in the application of recognition and sorting
colored objects [24]. A diagram of a robot with the sensor inserted in the gripper is
shown in Figure 4.24. The objects to be recognized are of the same shape and different in
color. Objects are always positioned in the same place. The recognition takes place when
the gripper detects the object between its palms and stops. Then grasping and color
recognition are performed at the same time and the identified object is transferred to the
desired place. Identification is performed quickly, operation of the robot is fast without a
t- -----~
light source
coloured object
1... _ _ _ _ _ _ _ _ •
Figure 4.24. Application of the optical fiber color sensor for sorting colored objects
268 J. A. MARSZALEC
designed color sensors, and machine vision systems that employ optical fiber bundles,
one important effect should be always considered. The transmission coefficient of optical
fibers depends on the wavelength of the transmitted light, and so the length of the fibers.
The fibers act as a transmission filter. On the basis of a theoretical analysis and a
simulation of the effect for the typical optical fiber bundles it was shown [27], that
already from a few meters long optical fiber bundle coupled with a halogen lamp the
output light decreases the color temperature, changing also the spectrum of the light
radiated from the fiber. The effect per unit length is smaller for other types of optical
fibers as they have lower spectral attenuation per unit length. The effect described may
cause some deformation in color perception and errors in the recognition of color.
Apart from the optical fiber sensors discussed above, which have been developed
mainly for robotics, there are many other types of optical fiber sensors which can also be
used in some robot applications. Among such sensors one can enumerate optical fiber
temperature sensors, acoustic sensors, sensors specially developed for the electrical
power industry, optical fiber gyroscopes, optical fiber gas sensors, and other sensors
conditions.
During the past few years, several optical fiber temperature sensors have been
developed for use in different environments. Sensors now commercially available are
clasified into two groups: low-temperature sensors with a range of -lOOoe to 400 0 e
using specific materials such as phosphors, semiconductors, liquid crystals; and high-
OPTICAL FIBER SENSORS AND SYSTEMS 269
The possibilities for temperature sensing with optical fibers are very extensive. These
using optical fibers. Optical fiber temperature sensors can be located from a few meteres
Magnetic field sensors, electric field sensors, current sensors and voltage sensors
constitute a group of optical fiber sensors specially developed for the electrical power
industry. These sensors are based on a number of different magneto-optic and electro-
optic effects, e.g. the Faraday effect, Pockels effect, Kerr effect, electrogyration effect
and other effects in optical fibers and special crystals. These sensors are both of the
extrinsic and intrinsic types. Magnetic field, electric field, current and voltage sensors
environments, like power plants, power stations and laboratories for testing high voltage
equipment [28].
Optical fiber gyroscopes are interferometric sensors for rotation sensing. The
based on the Sagnac effect. Fiber optic gyroscopes offer a number of advantages over
mechanical gyroscopes, i.e. small size, wide dynamic range, potential low cost and the
elimination of moving parts from the sensing device. These devices may be used for the
navigation of mobile robots and platforms. Some models of optical gyroscopes are
Optical fiber gas sensors are based on absorption spectroscopy. In this method the
spectrum of the light passing through a gas-detection cell is analyzed, and changes of the
spectrum carry the information about the type of gas present and its concentration. Light
270 J. A. MARSZALEC
to and from the detection cell is carried by optical fibers. Robots operating in special
environmental conditions may be equipped with gas sensors for smoke monitoring and
the detection of the leakage of inflammable gases such as methane (CH 4), propane
(C3H 8), ethylene (C 2H4) or other gases, in order to avoid accidents. Gas sensors may, for
constituted by optical fiber smart structures and skin systems [29]. These are optical fiber
sensors (in different sensing schemes) embedded in composite materials for real time
evaluation of stress, vibration, temperature and other parameters. The sensors have the
evaluation techniques once a part has been produced, to form a sensor network for
controlled system architecture. At present such sensor systems are used for the aerospace
and aircraft industry, production of pressurized tanks and other systems which need state
monitoring. But it seems that one day such materials may also be used for building the
There are many advantages of using optical fibers and fiber sensors in robotics.
changes and, hence, the high quality of measurement and transmitted signals;
measurements;
from the gripper which is often exposed to the harmful action of the environment;
sparking safety of optical fiber systems, in eliminating electric current from the
sensing area;
References
1. Kales D., Fiberoptic Sensors Find Growing Niche in World Market, Laser Focus
World, August 1989, pp.65-66.
2. Culshaw B., Optical Fiber Sensing and Signal Processing, Peter Peregrinus,
London, 1984.
3. Marszalec J., Optical Fiber Proximity Sensor for Robotics Applications, Optical
Fibers and Their Applications IV, Szustakowski M.IRomaniuk R., Editors, Proc.
of SPIE, Vol. 670, pp.161-167, 1986.
6. Marszalec J., Heikkila T., Jarviluoma M., Dynamic Range Data Acquisition and
Pose Estimation for 3-D Regular Objects, Intelligent Robots and Computer Vision
IX: Neural, Biological and 3-D Methods, D.P. Casasent Editor, Proc. of SPIE,
Vol. 1382, pp.443-452, 1990.
272 J. A. MARSZALEC
8. Bejczy A.K., Smart Sensors for Smart Hands, Progress in Astronautics and
Aeronautics, Vol.67, pp.275-304, 1979.
9. Wada M., Nakanishi T., Kouda M., Inagaki I., Shield-Segment Erection Robot,
Proc. of 5th International Symposium on Robotics in Construction, June 6-8,
1988, Tokyo, Japan.
10. Kopola H., Nissila S., Myllyla R., Karkkiiinen P., Intensity Modulated Fiber Optic
Sensors for Robot Feedback Control in Precision Assembly, Fiber Optic Sensors
II, A.M.Scheggi Editor, Proc. of SPIE, Vol. 798, pp.166-175, 1987.
11. Marszalec E., Marszalec J., Optical Fiber TouchlPressure Sensor for Applications
in Robotics and Automation, Proc. of II National Conference on Robotics,
Wroclaw, 1988 (in Polish).
12. Ristic R., Benhabib B., Goldenberg A.A, Analysis and Design a Modular
Electrooptical Tactile Sensor, IEEE Transaction on Robotics and Automation,
Vol.5, No.3, pp. 362-368, 1989.
13. Bejczy A.K., Applications of Fiber Optics to Robotics, International Fiber Optics
and Communications, VoLl, No.6, pp. 40-42,1980.
14. Schnitter J.L., Sheridan T.B., An Optical Tactile Sensor for Manipulators,
Robotics & Computer-Integrated Manufacturing, Vol. 1, No.1, pp.65-71, 1984,
Pergamon Press.
15. Fields J. N., Attenuation of a Parabolic-Index Fiber with Periodic Bends, Applied
Physical Letters, Vol. 36, pp.3265-3267, 1980.
16. Lagakos N., Cole J. H., Bucaro J. A., Microbend Fiber-Optic Sensor, Applied
Optics, Vol. 26, pp.2171-2180, 1986.
17. Shadaram M., Discrete Measurements of Field Distribution Using Fiber Optic
Microbend Sensor, in Optical Testing and Metrology II, Proc. of SPIE, Vol. 954,
652-656, 1988.
18. Shadaram M., Fiber Optic Trends - Sensing with Fibers, Photonics Spectra, June
1989, pp.117-118.
19. Jenstrom D. T., Chin-Lin c., A fiber Optic Microbend Sensor Array, Sensors and
Actuators, Vol. 20, pp.239-248, 1989, Elsevier Sequoia.
20. Schoenwald J.,Thiele A., Gjellum E., A Novel Fiber Optic Tactile Array Sensor,
Proc. of IEEE Conference on Robotics and Automation, pp. 1792-1797, 1987.
21. Kvasnik F., at aI., Photoelastic Slip Sensor with Optical Fibers for Use in Robotic
Gripper, Proc. of Conference on Sensors and Their Applications, Southampton,
58-59,1984, Institute of Physics.
24. Marszalec E., Optical Fiber Sensor for Colour Recognition in Robotics, Optical
Fibers and Their Applications IV, Szustakowski M./Romaniuk R., Editors, Proc.
of SPIE, Vol. 670, pp.161-167, 1986.
25. Marszalec E., Reflectance Spectrum Analysis Method and Based on This Method
Optical Fiber Colour Sensor for In-Process Colour Recognition, In-Process
Optical Measurements, K.H.Spring Edit., Proc. of SPIE, Vo1.1012, 1989.
26. Marszalec E., Reflectance Spectrum Analysis Method and Optical Fiber Colour
Recognition System for Robotics and Automation, International Journal of
Optoelectronics, Vol.4, 327-341,1989, Taylor & Francis.
27. Marszalec E., Marszale 1., The Effect on An Optical Fiber Colour Sensor of the
Colour Temperature of the Radiation Source and of the Optical Fiber
Transmission Properties, Perceiving, Measuring, and Using Color, Proc. of SPIE,
Vol. 1250, pp.256-263, 1990.
28. Szczot F., Marszalec J., Romaniuk R.S., Research Towards the Optical
Equipment for an Autonomous Robot for MV Environment, Industrial Inspection,
Proc. of SPIE, Vol. 1010, pp.100-106, 1988.
29. Udd E., Fiber Optic Skin and Structural Sensors, Industrial Metrology, Vol. I,
pp.3-18,1990.
Machine vision is one of the most important and most mature sensor systems used
in robotics and industrial inspection. There are different types of vision sensors
software. There are, however, certain problems and limitations in applications of machine
vision, which come from special requirements depend on the specifics of an application.
placement of camera system in small areas, which requires that both the
placement of the camera system in heated areas, which requires that the camera
Fiber optics is a technology which offers not only new solutions to some of the
existing problems, but opens a great range of possibilities to create new methods for
lighting and imaging with machine vision at low cost. The basics of optical fiber
technology for machine vision and some applied methods constitute the subject of this
chapter.
Optical fiber bundles are assemblies of optical fibers, fixed at both ends using
epoxy glue or bonded together by other methods, usually having great flexibility between
the two ends, and used to transmit light. Bundles are made of a large number of optical
fibers. Bundles which have a circular cross section are of diameter from 1 mm to several
centimetres. There are two types of optical fiber bundles: noncoherent and coherent.
Different types of optical fibers are used for the manufacturing of bundles. In most cases
optical fiber bundles are made of special, thin clad glass fibers which have a diameter of
10 J..I.m to 30 J..I.m. However, this type of fibers has some limitations in manufacturing
technology, since the fibers can not be longer than 2 metres, or in some cases up to 10
metres. So, for some applications optical fiber bundles are manufactured of other types of
optical fibers, e.g. glass fibers of larger diameters (PCS fibers or HCS fibers), or plastic
OPTICAL FIBER SENSORS AND SYSTEMS 275
fibers of different diameters (in some cases up to 1 mm). The length of bundles in these
cases can be longer than bundles made of thin glass fibers. Bundles are protected from
both.
Noncoherent optical fiber bundles are bundles in which the order of placement of
fibers at both ends is random (Le. fibers in a bundle are not parallel). Noncoherent
bundles are used for illumination purposes. The filling coefficient of this type of bundle,
showing the percentage of the active area of the bundle's cross section, in well packed
bundles in over 90%. Coherent optical fiber bundles are bundles in which fibers are
arranged in parallel, usually in the form of a matrix, so the order of fibers at both sides is
the same (spatial coherence). This type of bundles is used for the transmission of images.
The light transmission principle with noncoherent and coherent optical fiber bundles is
shown schematically in Figure 5.1. Although in general noncoherent and coherent optical
fiber bundles are flexible, there are also rigid components for image and light transfer,
and their shape can be matched to the geometrical configuration of the systems where
NONCOHERENT
BUNDLE
COHERENT
BUNDLE
Figure 5.1. Transmission of light by noncoherent and coherent optical fiber bundles
276 J. A. MARSZALEC
methods are used in machine vision systems, but unfortunately quite often the
less expensive to improve lighting than it is to further process the image. In many cases
also a part of the image lost by nonappropriate lighting can never be restored. Another
important problem for the imaging process is the ability to control lighting.
machine vision. The advantages of illumination with noncoherent optical fiber bundles,
limited areas,
incident angles,
remoting a hot, high wattage light source from the illumination zone,
attenuated or filtered,
ease of optical interface, e.g. lenses, filters, diffusers, without heat stress,
The most important disadvantage of optical fiber illuminators is that they are impractical
in cases where illumination of large areas is required. Another effect in the transmission
of the transmitted light, what can be important for color vision systems. The latter
shown schematically in Figure 5.2. A light source is located at some distance from the
area to be illuminated and viewed by a camera. Light from the source is coupled into the
bundle and transmitted to an object. To make light coupling to the bundle more efficient
which access is often impossible with other means of illumination, optical fiber bundles
should have suitable manipulation and positioning possibilities. The input end of the
Light
Sourc\o!
-0-0
Figure 5.2. A configuration of an optical fiber illuminating system for machine vision
278 J A MARSZALEC
bundle usually has a circular cross section, when the output end can be of different
shapes, depending on the place where is mounted and the illumination requirements.
and are commercially available as off-the shelf or customer designed products. Some
examples of illuminators with different shapes of output end and a multibranch optIcal
fiber bundle (with one input and 6 outputs) are shown in Figure 5.3.
In most applications noncoherent optical fiber bundles are used for front lighting.
Lighting with optical fiber bundles provides intense, shadow-free illumination over the
area viewed by the camera. For front lighting applications multi branch optical bundles or
fiber optic ring lights seems to be the most appropriate. When a multibranch bundle is
used branches can be located (fixed) around the scene and the angle of incidence of light
depends on the reflective properties of the objects viewed by the camera. An angular
reflections from specular reflecting surfaces, thus reducing saturation of the detector area.
But in other applications specular reflections can be used as a source of information, for
example, in the detection of the presence of parts with a certain shape based on the glare
from their surfaces. Generally, however, glare is undesirable as it disturbs the viewing of
Fiber optic ring illuminators are illuminators in which the output side fibers are
integration with a camera. An example of such an illuminator placed outside the optics of
the camera is shown in Figure 5.4 [1]. The camera is used for acquisition of an image
from the area close to the gripper and is mounted at the robot wrist. The light to the
illuminator is coupled from the primary light source placed on the other side of the wrist.
Light from the fiber ring is directed downwards, illuminating the scene located in front of
the camera. As the whole vision system is mounted on the robot wrist it can be moved
according to the movements of the gripper, and perform dynamic vision sensing, which
C"m~r.
Fibr.-oplic
illlJmin~'kln
burKi/II
~m~r. Grippe,
ilKis 'Ki~
Figure 5.4. Robor gripper equipped with a camera and an optical fiber ring illuminator
In the design of a special vision sensor for dynamic sensing an optical fiber
ribbon is used for the illumination [2]. The ribbon consists of eight fibers, which bring
the light from an externally located He-Ne laser source to an array of eight graded
refractive index lenses (GRIN lenses), located in one palm of the robot hand (Figure 5.5).
The lenses produce eight parallel, collimated light beams between fingers, which are
monitored by a linear detector array mounted on the opposite palm. The signal-to-noise
ratio in a well-lit area was greatly increased by placing an interference filter tuned to the
wavelength of the laser (633 nm) in front of the photodetectors. The sensor detects the
presence of objects between the fingers when one or more of the light beams are
interrupted. The object shape information is comparable to that obtained from a much
larger tactile sensor pad. Protection of optical fibers against mechanical damage is solved
OPTICAL FIBER SENSORS AND SYSTEMS 281
Pnotodetector
array
Protective
tube
Finger
slots
foad ~nses
Optioal
fibf'rs
Figure 5.5. A dynamic vision sensor with a linear array of optical fibers and GRIN lenses
for illumination
(ill (bl
Figure 5.6. Two possible sensor configurations and two different scanning methods when
light beams arranged in a linear array; a) linear scan: orientate the light beams
horizontally and move the sensor vertically, or orientate the beams vertically and move
the sensor horizontally, b) cross-section scan: orientate the light beams horizontally and
so that optical fibers are brought up to the gripper through a robust protective tube.
Toextract information about an object the robot scans in two ways: linearly, in the plane
orthogonal to that of the light beams, or cross-sectionally, in the same plane as the beams,
depending on how the sensor is oriented and moved (Figure 5.6). An example of a image
obtained from the linear scan of an object is shown in Figure 5.7. In the cross-section,
scanning data are taken at six different angles (when light beams are rotated clockwise to
the angles 00 , 900 ,300 , 1200 ,600 , and 1500 ), and this gives a sufficient amount of detail
scanning time.
a b
Figure 5.7. The result of a linear scan of an object of smaller than 2 cm 2 in size; a) pawn
The sensor presented has smal size, can work in the dark, it ignores inadequate
lighting and background contrast, shadows and reflections, and it is mobile so it can
access widely scattered or hidden parts. The sensor's accuracy is limited only by the
width of the light beams and by the robot accuracy. Using dynamic sensing, images with
a 0,5 mm resolution can be reconstructed. The design of the sensor can be simplified and
property of such a sensor and an advantage over typical vision sensors is that it can
and when it is totally interrupted , blocked by a wide calibration square. The only
requirement for the sensor application is is that the palms of the robot gripper maintain a
Illumination constitutes a difficult problem for inspection tasks when dealing with
surfaces that have a dominant specular component of reflection. For a purely specular
surface, light is reflected so that the angle of incidence eguals the angle of reflection. To
have camera images free from bright points or highlights the structured highlight
technique with a large number of point sources to illuminate the inspected object has
Sun
Workstation
r\. Color
MonitOr
'\
)
I
SHINY
• O;splay
• Norm
"lnsped
Figure 5.8. A schematic diagram for a structured highlight inspection system - SHINY
284 J. A. MARSZALEC
(SHINY) that uses a hemispherical array of 127 point light sources for illumination is
shown in Figure 5.8. Each point source is generated using light emitting diodes coupled
to single optical fibers. The ends of the optical fibers fixed in the hemisphere create the
final lighting system. The properties of the point sources can easily be matched to the
requirements by choosing fibers with appropriate diameter and numerical aperture. The
system was tested for solder inspection and showed very good results, providing a basis
illuminators, or so called "lines of light". In this type of illuminators, the input side of the
device is made in the form of a rounded shaped noncoherent bundle, and the output side
is in the form of a matrix of optical fibers, in which the length of the matrix is many
times greater than its width. A view of two models of line illuminators is shown in Figure
5.9. Fibers in this type of illuminator are tightly packed and specially randomized to have
fibers means that one portion of the fibers from one part of the input side of the
illuminator is directed to a few different parts of the output side. The standard length of
line optical fiber illuminators is up to about 400 mm, with slit width from 1 to 2 mm.
from 1,5 to 3 meters or longer in length are manufactured using plastic optical fibers,
which have bigger diameters than glass fibers (sometimes up to 1 mm), and uniformity of
light coming out in this case is lower (about ±20%). A cylindrical lens can be attached to
the output side of the illuminator to project and focus the line source on the scene to be
imaged.
OPTICAL FIBER SENSORS AND SYSTEMS 285
a b
Figure 5.9. Two models of line optical fiber illuminators, a) made by Schott Glaswerke
camera, which directly sees the area of interest, being mounted above the scene. This
way of image acquisition can easily be appled in many tasks, where there is no restriction
to the space where the camera is located or if there are no obstacles between the scene
and the camera, which would make image acquisition impossible. A schematic
comparison of the conventional machine vision system with a vision system with a
coherent optical fiber bundle is shown in Figure 5.10. In the second type of vision system
one side of a coherent bundle is fixed in front of the CCD sensor of the camera, and the
other side, instrumented with the appropriate lens if required, is used for image
acquisition. As the bundle is flexible and has small dimensions (a diameter in the order
286 J. A. MARSZALEC
of a few milimeters) its image acquisition side can easily be located in places where the
camera could not be installed. Typical high resultion optical fiber bundles are made of
small diameter optical fibers (20-30 J.l.m) with a very thin cladding layer. There are,
however, coherent bundles made of glass fibers of larger diameter or low resolution
bundles made of plastic fibers. The shape of the cross section of coherent optical fiber
bundles is usally a circle, but there are also bundles with a square or rectangular cross
section.
t1ic 0 lens
Figure 5.l0. Machine vision system, a) conventional, b) with a coherent optical fiber
bundle
A vision system with a coherent optical fiber bundle has been developed for
automated visual inspection [6]. As it is shown in Figure 5.11, one side of the coherent
OPTICAL FIBER SENSORS AND SYSTEMS 287
bundle used in the system is positioned at the robot end-effector when the other is
coupled to a camera. The vision sensor is supported by an acoustic sensor that measures
distance to an inspected surface. The acoustic sensor is also located at the end-effector,
close to the fiber bundle (as shown in Figure 5.12). The task goal was the positioning of
the coherent optical fiber bundle, based on range measurements, to correctly acquire an
image. While the vision system provides scene gauging and object localization, the
acoustic ranger provides the data to determine the camera's correct focal distance and,
hence, the vison system's gauge scale. The importance of the correct positioning of the
bundle when acquiring the image is shown in Figure 5.13. When the fiber optic bundle is
not at the correct focal distance objects are not correcly imaged (Figure 5.l3a). When the
bundle has been moved to the correct focal plane objects are imaged correctly (Figure
5. 13b).
VAX
cdcptlvely generate
motion commands for robot
distance - z motion
part data- xy motion
mohon
commands
I to imaging system)
Camero'--_ _..... I
Figure 5.11. Functional diagram of a robotic system for visual inspection, indicating the
combination of a vision sensor with a coherent optical fiber bundle and acoustic ranger
t I
Figure 5.12. Robot end effector with coherent optical fiber bundle and acoustic sensor
Figure 5.13. Objects imaged by a vision system equipped with a coherent optical fiber
bundle; a) not correctly imaged, due to incorrect focal distance, b) correctly imaged, i.e.
In most robotics applications of machine vision for scene analysis the camera is
remote from the end effector of the robot, due to the payload limitations of the robot.
This position provides a global view of the scene. Such placement of the camera usually
results in some position error, due to accuracy limitations in the joint coordinate
inversion calculations and calibration of the camera coordinates in world space. In the
approach presented above, the vision system captures images through a coherent optical
fiber bundle positioned directly at the gripper. This releases the burden of additional mass
at the end-effector and provides images from the most dynamic frame of reference - the
Fiber optics provides another unique opportunity for machine vision, to split a
few images to one camera. This can be very useful in obtaining several views of the same
object or to obtain images with different magnifications. The devices to perform image
splitting are optical fiber image splitters. A 4: 1 image splitter is shown in Figure 5.14 [1].
The ends of the four legs of the splitter used to pick up images are mounted near the
object to be inspected. Each leg can be instrumented with a lens according to the
requirements of the image acquisition. At the output side of the splitter the output ends of
four legs are placed together, and this side of the splitter is directed at the camera.
Optical fiber ribbons are another type of optical fiber devices which can be used
for imaging purposes in automated visual inspection. In the system for visual recognition
of small industrial components fed from a bowl feeder, two optical fiber ribbons with one
CCD camera are integrated [7]. A diagram of the system is presented in Figure 5.15. Two
parallel light beams are directed to shine, one down vertically on the component in the
viewing station, and another one horizontally across the track. As a result two shadows of
the component are formed, a plan view on the track floor, and side view on the track side.
The fibers in the ribbons pick up component silhouettes and transduce images to a linear
camera mounted outside the station. An example of a component imaged by the system is
presented in Figure 5.16. The upper portion of the screen depicts the side view, and the
lower portion depicts the plan view. It can be observed in the display that the side and
plan views are slightly offset. This is a result of the viewing station. The lighting and
optical fiber ribbon for the side and plan views are not in the same plane. They are
displaced approximately 6 mm. This prevents lighting from one projection being
Direct ion of
movement
Fiore opllcs
detect
silhouett es
in two direchon
eeo
Figure 5.15. Application of linear arrays of optical fibers (ribbons) for image acquisition
OPTICAL FIBER SENSORS AND SYSTEMS 291
An interesting feature in the approach presented is that the two ribbons of optical
fibers share a single linear camera (a 256 x 1 CCD camera was used). The 256 elements
of the linear camera can be allocated in any proportion to the side and plan view, so two
views could use the same number of elements (128 each). However, if the component
under inspection has relatively little height compared to its width, then it is possible to
The detection tolerances of the system described above are not more than 5%
(mainly restricted by the repeatability of the feeding system, the imaging system is able
to resolve better than 1%). But by recognizing different orientations of components and
detecting faulty components the system is suitable for feeding and sorting components
Figure 5.16. An image of the object obtained by using a vision system with optical fiber
ribbons
292 J. A. MARSZALEC
Taking advantages of both noncoherent and coherent optical fiber bundles and
merging them togther into one system, with an external primary light source and an
externally located camera, one can create a compact optical tiber eye-in-hand machine
vision system (Figure 5.17). Important features of such a system are that the sensing head
of the system, which contains its own illuminator and image acquisition part, is of a small
size, illumination and image transmission paths can be the most appropriate geometrical
configuration, and the system can be applied for dynamic sensing when located at the
robot end-effector.
/
/
CCD camera
coherent optical
fiber bundle
Figure 5.18. A vision sensing head, located in the robot gripper, consists of a ring
illuminator with Fresnel lens, and a coherent oprical tiber bundle placed inside the ring
illuminator. A light source and a camera are mounted on the robot arm. The vision
sensing area is directed outside the gripper and limited by the optical properties of the
illuminating and imaging bundles.
OPTICAL FIBER SENSORS AND SYSTEMS 293
Anicufllrion of robot
firm begins above
cam~ra Ana lighr SOufC~
U9hr- Robot
wBight arm Lighr
robust sourctl
camer~
(solid·
S"ral~J
Figure 5.18. An optical fiber eye-in-hand vision system placed on the robot
Apart from the direct method of designing eye-in-hand vision system, described
above, the special solutions can be created, which in some conditions can serve certain
arranged in the sensing head into such pattterns, which are the most appropriate for fast
fiber geometries for imaging as well as for illumination in a sensing head for the sensor
system presented in Figure 5.19 are shown in Figure 5.20. The linear, cross, and circle
patterns can very quickly provide area, centre location and angle information,
respectively. This simple preprocessing can reduce the amount of processing needed,
294 J A MARSZALEC
thus improving the close loop response. In the experiments performed with a cross-
pattern for hole centre location a linear response with very good sensitivity was
demonstrated [8].
PllOl0diode
iJrray f lecuonics
fO
processor
Scanner
heBd
'-------'!~i_U_. . l---Li_......J1
. Workpiece
The integration of coherent optical fiber bundles into robot vision systems is a
good way to acquire images from the area of the gripper, however it should be noted that
for this application, bundles should be specially protected. Flexible optical fiber bundles
OPTICAL FIBER SENSORS AND SYSTEMS 295
cannot withstand continual flexing, particularly if the bending radius is small. Dark spots
can soon appear in the image, seen by the camera, as some fibers in the bundles begin to
break and do not transmit light as well as undeamaged fibers. The same effect, when it
appears in noncoherent optical fiber bundles, just decreases the intensity of the
developing continously (including material and methods for fiber protection against
mechanical damages), and the mechanical strength of bundles produced for applications
References
2. Beni G., Hackwood S., Rin L., Dynamic Sensing for Robots - An Analysis and
Implementation, in Robot Sensors, Vol. I - Vision, APugh Editor, IPS
(Publications) Ltd., pp. 139-146, 1986.
3. Nayar S.K., sandereon AC., Weiss L.E., Simon D.A, Specular Surface
Inspection Using Structured Highlight and Gaussian Images, IEEE Transactions
on Robotics and Automation, Vol. 6, No.2, pp.208-218, 1990.
7. Cronshaw AJ., Heginbotham W.B., Pugh A, A Practical Vision System for Use
with Bowl Feeders, in Robot Sensors, Vol. I - Vision, APugh Editor, IPS
(Publications) Ltd., pp. 147-156, 1986.
applied to a number of tasks in robot applications, should have provided the reader with
the feeling for the potential of fiber optics in this increasing industrial concern. The
researcher and industrial practitioner may find in certain sections some specific
suggestions how to solve his/her own problems. The examples of devices and systems
which have been described should convience the nonspecialist that this area of modem
environments.
The various optical fiber sensors and systems described are oriented towards the
increased sensing abilities and intelligence of the robot. Fiber optic sensors constitute a
new type of robot sensors, with very interesting characteristics and important advantages.
But most of the research has been carried out up to now using components developed and
manufactured primarily for communiaction applications of fiber optics. This is one of the
important reasons that only very small potential of fiber optics for sensing could be
exploited. Components developed specially for fiber optic sensors (including the most
important special fibers for sensors) are just corning to the market and the further growth
in the fiber optic sensors for robots seems to be very optimistic. Robot vision systems
with optical fiber illuminators and coherent optical fiber bundles can be built using
commercially available components, but they should be the object of increasing efforts of
transfer from the research laboratory to robotics practice in the years to come. Also
optical fiber communication systems will be beneficial for various applications of robots
paricular, are going to experience a substential and sustained development in the years to
come. Laser- and fiber-optics-based measuring systems represent one of the major fields
of application of these technologies, and their adoption to robotics may take place at a
faster pace in the future as they are in many cases simple and reliable tools to meet the
of the operations.
A wider review of problems of integration of lasers and fiber optics into robotic
systems can be found in the coming textbook:
Marszalec J., Marszalec E., Integration of Lasers and Fiber Optics into Robotic Systems,
Tutorial Texts in Optical Engineering, SPIE Optical Engineering Press (to be printed)
Design and Dexterous Control of
Micromanipulator
with Multiple Degrees of Freedom
Ab stract
1. Introduction
®
(D
J(m=C 0) p
J(.= (k. , 0)
Km= o 0 K.= o 0)0
(kpm 0) (k p, o kIm o .kl'
-em 0)
C.. -
_(CPO
C,- 0)
_(Cpm 0) C,-_(CP' 0) o CI" o CI'
Cm-
o 0 o0 @
@ p
J( .. = (k ..
p
J(,= (k , 0)
O
J( .. =Co k1m )
p
J(,= (k , 0)
o 0
0 k:J o kl,
-rpm 0) -r p,
C,- 0)
C.. _(Cpm 0) _(CP'
C.- 0) Cm-
o Clm o CI'
-
o 0 o CI'
@
Km. Ks. Cm and Cs are the feedback gain matrix and the
observation matrix of the master and slave. where m and s denote
the master and the slave. respectively. The subscripts p and [
represent position and force. respectively. Kc is the adaptive control
gain and effective only for the no. S type. while it is constant for the
other type. The types of the control method of no. I to no. 4 have
conventionally studied. In this study. the dynamic behavior is taken
into account on the no. S type. Among the above mentioned bilateral
control schemes. it is shown that No. S is the most general method. and
that the others are the special case of the no. S. by letting some
part of the matrix ele ment be zero. In no. 1. the case of
Cs=Ks=O corresponds to the nor mal position servo control syste m.
k[(*
1= 2I(*+k P=[(P (1)
Assuming that the actuator of the gripper is a motor with the output
Pm and the input Um. we can derive the dynamics of this model as
follows.
304 T. FUKUDA ET AL.
(2 )
R ( Bm+-V
J{t r2)Bt +J{",
ami R: Resistance of the motor
R( r2 ) Km: Counter power constant of the motor
ICt lm+-;;rMm
Kt: Torque constant of the motor
2
r J{ Jm: Inertia moment of the motor
-;;r Dm: Friction of the motor
amo= r2 MlO: Mass of the load
j",+--::r+ M",
11 TIl: Friction of the load
K: Spring constant of the overall model
2:.A
n m n: Reduction gear ratio
bm= U: Input voltage to the motor amplifier
R( r2 )
J{t j"'+-;;rM", r: Radius of the pinion
i",=A;"x", + b;"um
y",=C~TX ..
where
MICROMANIPULATORS WITH MULTIPLE DEGREES OF FREEDOM 305
i.,=A.,z .. +bmUm
y .. = C!Zm
(4 )
where
Am= ( 0 1),
'-ainu -aml
bm= ({3AI)
/3mo
model, the state equations of the master and slave are described as
follows:
(i) Model representation of the master
im=AmZm+ bmu ..
y .. = C~z ..
C~=(l 0), Z~:=(Zm i .. )
(5)
(ii) Model representation of the slave
i,=A,z.+b.u.
y,= Ch.
A.=( 0 1)
- a,o - a•• '
b.=(O)
P.
Cl = (1 0), zI = (z. i.)
(6)
(7)
where
(10)
A.y.= /3,oe(t)
(11 )
(p+~,)Am(p)e
=(P+ ~,)Am(t)(Ym - y,)
= (p+ ~,)Am(P)Ym - [A.(P)(P+ ~2) -I- ,t:J} y.
=(t) -I- ,\,)Am(P)Ym - (p+ ~2){3soe -,t:Jy.
308 T. FUKUDA ET AL.
(t 2)
where
e ~
()l/J .. o(P+~I)II .. -/J'0(P+~2)Ym
Am P
(P+~I
+(/J.o(P+ ~2) - ~J} Y']
The parameters yJ and y2 of Q(p) are determined so that the transfer
function becomes positive real, i.e.
( 14)
where
~ _.&oA2 ~ ~2-AJ)
~T = ( A1- n, - /3mo' /3.,0' /3mo' /3.,0
eT{l P. _I_
= Q(p) Um. Q(p) Ym, Q(p) Ym.
J...- 1 }
Q(p)Y" Q(P)Y'
(I 5)
MICROMANIPULATORS WITH MULTIPLE DEGREES OF FREEDOM 309
E(t)=e(t)- e(t)
={3mo W(S)((pT +J(T(t))W)l
= (3mo W(s) EJTe(t)
(16 )
where
(17)
where
e(t)=- rW)E(t)
board of 1 Imml thick is placed upon the top block by the joints
made of phosphorus copper. Owing to this joint, the bottom block
moves smoothly with 3 degrees of freedom (D.O.F.J; 1 D.O.F. of
extension and 2 D.O.F. of two rotations caused by the stretching motion
of each PZT. This bottom block changes the pose of the mid block on
it. The mid block has 2 D.O.F.; 1 D.O.F. of extension and 1 D.O.F. of
rotation. A stainless board is attached to the edge of the mid block,
and connected with No.6 PZT of the top block. The top block has
I D.O.F. of extension. Thus, this micromanipulator totally have 6 D.O.F.
Photograph of this micromanipulator is shown in fig. 15.
"an" is the length of the link, "dn" is the distance between the links,
"en" is the angle between the two links. The relation between the two
1 0 o0
o0 1 0
A1 o -1
o0
o d}
o1
]
C2 0 -S2 0
S2 0 C2 0
A'"r o -1
o0
0 0
0
]
316 T. FUKUDA ET AL.
C3 0 S3 a3C3
S3 0 -C3 a3S3
A'=[ o1
o0
0
0
0
1 1
o0 1 0
~=[ 1 0
o1
o0
o0
o c4
o1 1
Cs 0 Ss asCs
Ss 0 -Cs asSs
A'=[ o1
o0
0
0
ds
1 1
1 0 o0
o1 o0
A'=[ o0
o0
1 d6
o1
j
(18 )
where Sn stands for the value of sin(No. n) at joint No. n, and en stands
for cos (No.n).
The expression for the tip of the manipulator about the base coordinate
is given by as follows.
MICROMANIPULATORS WITH MULTIPLE DEGREES OF FREEDOM 317
T6=A1A2A3A4.AsA6 (20 )
STo= 0 0 0 (21)
I~
1 0 0
0 1 do
0 0 1
'T.= 0 Ss
ICsSs 0 -Cs Ss dOl
-Csdo (22 )
0 1 0 ds
0 0 0 1
·To= 1 0
IgsSs 0 Ss
-Cs
ds
Ssd.
-Csdo+d,
I
0
0 0 0 1
(23 )
2T6= I-C3SS
S.Ss C3 -s.c. C.ds+S. (-CSd6+d.) +a3C.,
S3 C.Cs S.dS-C3 (-Csd.+d.) +a.S3
Cs o S5 Ssdo
o o o 1 (24)
c,s.s.-s.c. c,c. -c,s.c.-s,s. c, (C.d.+S. (-C.d.+d.) +a.C.) -S,S.d.
s,s,s.+c,c. S,C. -s,s,c.+c,s. s, (C,d.+S, (-C.d.+d.) +a.C,) +C,S.d.
c.s, -s, -c.c, -S,d.+C. (-C.d.+d.) -a,S,'
o 0 0 1
(25 )
C,s,s.- s,c. c,c, - c,s,c.- s,s. C,( C,d.+ S,( - C.d, + a,S. + d.) + a,C,}- S,(S.d.+ a.C.)
'T.= C,S. -s, -C,C. C,{C,d.+S'(-C.d.+a.S,+d.)}-S,d.-a,S,
- s,s,s. - c,c. - s,c, s,s,c. - C,S. d,- S,( C,d. +S,(d. - C.d. + a.S.)+ a,C,}- C,(S.d. +a.C,)
o
(27)
318 T FUKUDA ET AL
If the joint axes are given, the position and pose of the tip can be
obtained by calculating eq. (27).
AI-IT6=IT6 (28)
Ar I AI-IT6=2T6 (29 )
(33 )
(36)
dr = J dq (39)
(40 )
( 42)
d8 2 = C2d(NS2)-S2d(NC2)
C3
d8 s = Csd(NSs)-Ssd(NCs)
dd l = C3S2d(N II)-d(N 12)(ds+arC3C2Px+S3Py+C3 S2Pz)
dd6 = Ssd(N6I)- dS S(N62)
(SS)2
(43 )
d(NS2) = - oz (44)
322 T. FUKUDA ET AL.
ddl = (dl+d2+d3)/3
dd4 = (d4+ds)12
des = (d4-ds)/h
dd6 = d6
(45 )
where 1 is the distance between the two sensors; 2x 1 0- 2 1m) here. L1tl
and L1t2 are the detected elongation at each sensor.
The control system and block diagram are shown in fig. 20.
Control outputs for each PlT actuators are through a DI A board and
are amplified by controller upto the desired voltage from 0 [V] to 100
[V]. The stable time of controller for step input is about 200 [microsec].
Pose of the tip can be detected by non contact gap sensors and force
applied at the tip can be measured by strain gauge. The
measurement results are sent to the CPU. Desired pose is
transformed to the joint torque by the inverse Jacobian matrix and
transformed to voltages corresponding to the outputs of sensors.
back control with the Jacobian matrix calculation for more complex
manipulation. Desired pose of 5 [micron] displacement along Y axis
and 0.03 [degree] rotation around Y axis is given at the same time.
This position control are carried out at the same time. The sampling
time of the controller is rather long ( 0.22 [sec] ), and the measure-
ment signals are contaminated with noises. Never the less, the
position control is carried out without serious position error.
4. Conclusions
objects, the adaptive law can adjust and stabilize the control
system.
Next we proposed a micromanipulator with 6 D.O.F. made of PZT
actuators which is small and has many D.O.F. in comparison with
conventional types. This type of micromanipulator has many
application fields of complex microma-nipulation. A proble m is
that this type of micromanipulator has rather a little work space, so
in some cases, it will be used with roughly movable devices for
practical use. In the future, it will be required to increase the
rotatable angle of this manipulator to improve the performance. If
the shape or the distance fro m an ob ject is detected by means
of an image processing technique, tip orbit control and
automatic micromanipulation 1151 will be realized in this system.
5. References
k K* k
StraIn
Gaugl'
MOTOR
5 M
L A
A p 5
V T
E U E
oJt-
Fig. 3 Structure of the micromanipulator system
based on the master-slave bilateral control
Strain Gauge
Oisplacment
Sensor
~ Micro Actuator
"A'Slt" I'IA'5If:ll1
SOl SOl
~ ~
c , •• •
-: ~SI•
\0 ~ I •
-
~ h •• • '5l"--t
" fI ...
SLIII,(
0
~'l ,,--- 0
0..
• TI • •
\0
. ,.
0..
•
~
\0
fl ..,
•
,.
I
"L
1I,(ItA I Dill 01"'[11111110"
.0
Z Z
c::=:-=••
• .0 ,D D
•
H .... "
T~,.
nnE
....
0
Th"
\0
I
....
T
•
~ '1_,,0 I
20
(a) ( b)
,.,
:; .1...,c,-~_"""::-- __
ll.. I
-<,, c"l~
Q 0 10
11_ I
20
n .... t
~ UI j~'I~
k .". . .
C~----,..--_~
LL.
10 10 10 10
II . . . . II "'t t
Z 10
z'O~
o 10 ~o Il 113 10
h..
:T~
I
"['~
o 'SUII'Wt
"-
10 10 a 10 ZO
I, .. , TI .... t
(a) (b)
.. __ s I (II:
~
"I ~ :;
"I• /
I.
I... -
0 10 HI
~
I fl • •"
• SI.R,(
MI
." \l iIIIov [
0
<l.
.~' /""'
z 10 I II~.
~I
10
,
apUlIA '0111
:0
~Il~
JI",.•• t
I.
.e ......
o
::r~'
10 10
TI ... ,
III 10
~ 10 1'0
II ... !. Ilfn ,
(a) ( b)
so,
'Ol~ c ,-"j
_ _ -..._ _~,
~ 0 10 :0 o 0 - \0 20
TI_ , TI--. t
~~Sl ~ ~ UI
SUly!
.0 lO o '-'- /'----:"::---~
10 lO
h ... 11",.
00 s
OI'(ItRTOIt ZOO,
I
O'(RAIOIt
I\~\\~~'O'I~
z
o
~ 10 10
" h •• t
S'r'~
-~
o
....
10 10
II lilt ,
(a) (b)
so, '!lUIUI
,., 'UUI(II
/"
00 ,. c / 00 ,.
- U,•
~ .•-
0 0
To_
T, .. I I
'UIY, ~" n"'Yf
0 0
,. ,.
"-
"-
r ,. • /' ,. ,
Z
0
00,
h ....
O'(UIOII
Z
O.
•
I TI ..
.,,,.00-
· .·
~ ~ ,.
" " I" fl._" t
T~,. •
• I.~
O. I
.... '.~
Ii"'t
,. , ,.
(a) (b)
Method 1 2 3 4 5
Position
Error 0. 057 0.077 0. 041 0.223 1.00
Force
Error - - 1.00 0.920 0.093
Hybrid
Error - - 1.00 0.926 0.071
Stainless Sleel
(TIlickness ~ I mm) 10mm
PZT(pb(Zr.Ti)03)
5rnm
60
E
::t
....c 40
Q)
E
Q)
u
-a
cIj
CI)
Q 20
o 20 40 60 80 100 120
Voltage V
No4 NoS
"""
Mea.su ring
Poi n t
Nol
No2
~ '~
'f.
•.
i .
:iii:
rrr.etulllQt
"I.
~; .
[3~ !I I!
7~i;"\ ...
[J
Fig. 14 Structure of micromanipulator
x
Z·direclion
ROt31ion al'Ound X,V axis
X '~direction
Rotation around
Z'axis
X
, ~ Y"
2 92 -90 0 0 0 -I
-90 I
3 93 a3 0 0
4 d4 90 0 d4 0 I
5 95 90 a5 d5 0 I
6 d6 90 0 d6 I 0
non -contact
Q gap senso r
/
W
!
11' Q ~~
6.tl~l::.tl
i 6t2 T T
I
D
~!
@ ~-- ------IlI.tl
____ +
6t2
"" Error e
NB LO PB
~~~----1-----+---~
= NB 15 15 15 .
......................................
.8 1.5 1.5 1.5
.~ I---+------+-----+-------l
'i:: 20 10 20
<!) ZO .......................................
..B 1 2.5
~I---+------+-----+-------l
1
6 PB .....................................
15 15 15
.
1.5 1.5 1.5
upper: Kp
lower: Ki
F
Soft or Hard
S...mG'ge" '" Object
I
•....
L..-. . : :.; . ;= = =.
/
ip of Manipulator
:;"'T:::.I
·
\
Phosphorus Copper
(Thickness = O.3mm)
c CONTROLLER
ACTUAT OR
p ST RA I N NON-CONTACT
GAGE GAP SENSOR
j - - -- - , ,---------,
x, - - - - - ,Vs x
PI Controller
s 40
~
~
0
30
'H
+' 20 :-
I'd
bO
~ 10
0
......
0 I I I
,,-....
----
~ 15 - .
<1J
'U
N
I
0
.-i
X
'-'"
<1J
i r--,----.
rl
biJ 0 0,2 O,Ll 0.6 0,8 1.0
c Ti lll~ S
~
'"'
El
~ 5 .... ;v--./I.NI,..AJI.~JV\fWA-,,~N....\/V·0 W\l\J>AA/.,ftIJ\.A.)'··J....Aj
.......
0
t::
·rl
r
+'
<Il
bo
c:
0
rl
(
W
o 1 0 2 0 3 0
Tim e S
r-
OO
<1l
'0
o .0 3
,
N
0
......
x
.......
<lJ
rl
bO
c 0
~ o 2 0 3 0
Tim e S
l.0
Z
.... 0.8
'0
......
x 0.6
0
0.4
~
0.2
width mm
30
height mm 20
weight g 50
max
Movement Jlm 50
max
angle deg 0.3
max N 0.015
force
input V DC 100
A THREE DEGREES OF FREEDOM
MICRO-MOTION IN-PARALLEL ACfUATED
MANIPULATOR
Kok-Meng Lee
Associate Professor
The George W. Woodruff School of Mechanical Engineering
Georgia Institute of Technology
Atlanta, Georgia 30332-0405
Shankar Arjunan
Engineer
Cummins Engine Company
Columbus, Indiana 47201
Abstract
1. INTRODUCTION
actuated manipulator using piezo-electric elements. Due to the advantages of being fast
response, high resolution, high force generation per unit volume are obtainable from the
applications include the assembly of optical components and delicate mechanical devices.
seem to indicate the benefit of extending the concept to other fields [1-4] such as integrated
the measurement errors, poor rigidity, and non-linearities of the actuators. In the control
of a manipulator, the end-point position and orientation are generally fed back by
measuring the joint angles followed by a forward kinematic computation. There are two
the robotic manipulator requires time-consuming numerical computation. And second, the
feedback information is indirect and hence the dynamic effects of the manipulator on
individual joint must be normally accounted for. To alleviate this problem, the concept of
coarse-fine motion control strategy to enhance the robot accuracy was suggested by Sharon
et al. [11] and Hollis [12]. A planar XY robotic fine positioning device was developed for
end-point sensing [13] and successful application experiments on circuit inspection were
reported in [14]. Hollis et al. [15] has recently developed an interesting design for a six
A THREE DOF MICRO-MOTION IN-PARALLEL ACTUATED MANIPULATOR 347
variable compliance.
ranges of centimeters could be achieved using the piezoelectric effect in the design of a
linear stepping device. Scire and Teague [17] developed a linear micropositioning stage
using a piezoelectric driving element and flexure pivots to achieve a resolution of O.OOltLm
over a range of 50tLm. The advantages of flexure pivots, which are characterized by having
no backlash, friction, and bearing noise over bearings and sliding components, were
demonstrated. Umetani et al. [18] [19] presented the principle of piezoelectric micro-
manipulator. The design of a linear and a multi DOF motion actuators based on
piezoelectric effect has also been reported in [20] [21] [22]. The advantages of being fast
response, high resolution, and high force generation per unit volume are obtainable from
the multilayer piezoelectric actuator and have motivated the development of a series-
parallel micro-motion mechanism [23] as a wrist torque sensor and to enhance the
manipulator which has two orientation and one translation freedoms was conceptualized by
Lee and Shankar [24]. Recently, Hunter et al. [25] [26] reported the progress of a micro-
robot designed using dual actuator approach for manipulation and dynamic testing of a
sensor and actuator was performed by Tzou et. al [27] [28] for vibration control of a light
348 KOK-MENG LEE AND S. ARJUNAN
weight arm. A general discussion of actuators for micro robots has been presented by
manipulator were discussed by Lee and Shah [30] [31] and by Lee and Johnson [32]. In
particular, this paper focuses on the closed-form solution and experimental verification of
the forward kinematics for micro-motion control. To provide a rational basis for the
2. KINEMATICS
The manipulator consists of a base platform, three extensible links and a moving platform
which houses the driving mechanism of the gripper. The moving platform is connected to
the links by means of ball joints which are equally spaced at 120 degrees and at a radius r
from the center of the moving platform. The other ends of the links are connected to the
base platform through equally spaced pin joints at a radius R from the center of the base
platform. By varying the link lengths, the moving platform can be manipulated with respect
to the base platform. It has been shown by Lee and Shah [30] that the manipulator has
one translation and two orientation freedoms. A closed form solution of the inverse
kinematics has been derived in reference [30]. It is also useful to derive a linearized
and the rotations about the X and Y axis of the base coordinate for a given set of link
lengths.
The coordinate system is shown in Fig. 1. A base coordinate frame XYZ, with unit
vectors i, i, and k respectively, is fixed at the center of the base platform with the Z axis
pointing vertically upward and the X axis pointing towards pin joint, Pl' Similarly, a
coordinate frame xyz is assigned to the center of the movable platform with the z axis
normal to the platform and the x axis pointing toward ball joint, B l . The forward
kinematics are derived using the following procedure. The angle subtended between the ilh
link and the base platform, OJ, as shown in Fig. 1 is determined as a function of the given
link lengths. Next, the coordinates of the ilb ball joint are determined in terms of the link
lengths and the subtended angles. Finally, the position and orientation of the moving
platform are computed from the known coordinates of the ball joints.
As the distance between the adjacent ball joints is rJ3, the implicit relationship
between the link lengths and the sub tended angles is [30]
2 2 2
L. + L. + 3 - 3p + L.L. cosO. cosO. - 2L 1.LJ. sin O. sin O.
1 J lJ 1 J 1 J
- 3L. cosO. - 3L. cosO. = 0 (1)
1 1 J J
where i = 1, 2, 3 and j = 2, 3, 1, and where ~ = IJR and p = rfR. Equation (1) is of the
following form:
Since the motion is in the order of microns, a closed-form solution can be derived by
specified Lo, the corresponding operating angle 00 is determined from the geometry:
350 KOK-MENG LEE AND S. ARJUNAN
cos 8o • R (1 - p) / l 0
af. af.
1 1
a a C1
a8. a8 j
1 8, l 8, l
0 0 0 0
and
af.
1
aL. 8 , l
. af;
al.
a C2
1 J 8, l
0 0 0 0
where
c1= 3 l sin 8 [ - l cos 8 ]
o 0 o 0
c2
A8. + A8 ... - -- (AL.+ AL.) (2)
1 J C1 1 J
where AD = 8 - 80 and AL = L - 4 and the subscripts denote the respective links i = 1,2,3
and j = 2, 3, 1. For a given set of link lengths, Mi can be derived by solving the three
M.
1
= L tan 8
(3)
o 0
where tan 80 -+ O. The three linearized equations, i.e., Equation (3) with i = 1, 2, 3, are
decoupled. Also, in the linearized range, the change of the subtended angle of ilh link with
the base platform, M i , is linearly proportional to the change of the respective link length,
AI.;. It is noted that 80 corresponds to a singular point and when 80 = w/2, the change in
link length is accompanied with no change in the subtended angle.
A THREE DOF MICRO-MOTION IN-PARALLEL ACTUATED MANIPULATOR 351
Since the ball joints are placed at the vertices of an equilateral triangle, the
Cartesian position or the origin of the xyz frame is essentially the centroid of the triangle.
1 3 2
X = - ~- L.cos 8. cos [- (i-l)w] (Sa)
c 3 i =1 1 1 3
1 3 2
Y .. - ~- L. cos 8. sin [- (i-l)w] (Sb)
c 3 i =1 1 1 3
13.
I = 3 ~ L. Sln 8. (Sc)
c i=1 1 1
With the coordinates of the ball joints described in terms of the link lengths and the
sub tended angles, the unit vectors of the body axes xyz with respect to the base coordinate
-+ -+
... 08 1 - OC
(6)
u
x r
-+ -+
... 08 2 - 083
u (7)
y
J3r
... ... ...
u u X u (8)
z x y
352 KOK-MENG LEE AND S. ARJUNAN
where OB j is the line vector originating from the point 0 to the center of the ith ball joint,
and OC is the line vector originating from the point 0 to the center of moving platform C.
The orientation will be described in terms of rotation about the X and Y axes of
the base coordinate. The rotations about the X and Y axes are designated as er and p
xyz body axes with respect to the base coordinate XYZ, is given by
Hence, the unit vectors of the body axes have been derived in Equations (6) - (8),
the components of which are the elements of the following matrix [ROT]:
U • U • Uzi
Xl
[ROT] . U •
XJ
Yl
U •
YJ
U •
ZJ
(10)
uxk uyk uzk
From Equations (9) and (10), the angles er and p can be obtained by the following
relations:
(11)
(12)
The solution of the forward kinematics can be computed as follows: The change of
subtended angle, Mj , is determined using Equation (3). The Cartesian position of the ball
joints and the center of the moving platform can then be computed using Equations (4) and
A THREE DOF MICRO-MOTION IN-PARALLEL ACTUATED MANIPULATOR 353
(5). The angles of rotation about X and Y axes, a and /3, can then be computed from
Equations (11) and (12) where the unit vectors of the body axes are given in Equations (6),
The design of a prototype manipulator is shown in Fig. 2. The prototype links which
provide motion in the order of microns were actuated by the piezoelectric actuator as
shown in Fig. 2(a). All components were assembled and tightly fixed by means of solid pins
The link design consists of a cantilever structure which flexes at pivot points A
motion generated by the piezoelectric actuator located at Q. Unlike the flexure pivot
design in [17], the flexure pivot at A was designed to have a notch on only one side of the
beam to allow a strain gage attachment. This design enables the link motion to be inferred
from the strain gage measurement. The piezoelectric element was clamped between the
base and the compressor. The compressor is a rigid element designed to align and to
secure the actuator in place using pressure exerted by a pair of screws fastened into the
through· holes. The loader behaves as a rotational spring to ensure a positive contact
between the actuator and the link and thus eliminates the backlash. The stiffness of the
loader was designed to be much less than that of the pivot point at A so that the dynamic
The equivalent change in link length, AI.., and AU, of the prototype link design can
shown in Fig. 3 where point P is the initial position and point P' is the corresponding
position after the cantilever has been actuated by the piezoelectric actuator at Q. Similarly
the point B has been moved to Bl after the actuation. Hence the angle that is subtended
between lines PB and PB 1 is the angle of rotation about the virtual joint P. Length PB1 can
be derived by applying the law of cosines to the triangle PP'B 1 shown in Fig. 3:
displacement at Q
[ PpBBl] 2 = 1 + 2 -
bqY sin9 + (_q
b Y )2
ca 0 c a
where a, band c are the lengths PB, QP and AQ respectively. Using the technique of
series expansion and neglecting the higher order terms of (b/ac)Yq, the change in link
length is given as
(14)
The underlying assumption of Equation (14) is that the geometry (b/ac)Yq must be
much smaller than unity. Hence, the prototype link design has a virtual pin joint at P and
The function of the ball joint, as shown in Fig. 2(b), is to provide three rotational
degrees of freedom. The thin cylindrical cross-section at A allows bending about the x and
Y axes. The thin section at C enables rotation about the Z axis as seen in the figure. The
neutral axes are BN for bending about the x and Y axes and ND for the bending about z
A THREE DOF MICRO-MOTION IN-PARALLEL ACTUATED MANIPULATOR 355
axis respectively. The structure of the ball joint is such that the length of the vector BD
that connects point B and pin location D remains a constant. Hence, the structure has a
virtual ball joint with its center at B and an effective radius equal to BD. The ball joint was
rigidly attached to the movable platform at D, which was constructed in the shape of the
The parameters of the link and the specifications of the piezoelectric actuator are
listed in TABLE 1. The piezoelectric element used in the experiment was TOKIN 2x3x18
[33], where the numerical values are the physical dimensions in mm. The piezoelectric
element has a linear range of 0-15/Lm corresponding to 0-150 volts input. The multilayer
piezoelectric actuator has a relatively dielectric constant of 13.5/Lm per 100 volts as
compared to that of 0.22 /Lm/kV and 15 /Lm/kV reported in [17] and [34] respectively. The
multilayer piezoelectric actuator has a higher force generation than that reported in [22]
[25] [27].
Since the motion is very small, it is almost imperceptible to the human eye. To
verify the kinematics of the micro-motion manipulator and to demonstrate the concept
feasibility, a mirror was attached to the movable platform and a laser beam was directed
towards the mirror. The reflected beam is seen on the wall at a distance of about 9 meters
(30 feet). When the platform is oriented, the deflection of the laser beam on the wall is
measured. From the correlation between the beam deflection and the angle of rotation of
When a step input is applied to one of the links, the moving platform and hence the
mirror rotates about both x and y axes. As the measured deflection indicates a net rotation
of the moving platform, the following steps were taken to correlate the experimental data
to the forward kinematics. When the beam is directed to the center of the moving
the triangle CAIAz, where tP is the angle of rotation reflected by the laser beam. From the
geometry, the angle, tP, can be determined as tan-I(AIAz/CA I) where AIAz is beam
of 9 meters. It is known from the law of physics that when a mirror is turned through an
angle l/J, the reflected beam is rotated through 2l/J. In other words, since the reflected beam
travels through tP, the plate should have rotated through tP/2. Hence by measuring the
distance of the center of the beam before and after moving each link, the angle of rotation
can be inferred.
Since link 1 is located in the Y =0 plane, the platform will rotate only about the Y
axis if only the link 1 is actuated. Hence there is a direct correlation between the angle of
rotation about the y-axis, and the angle of rotation of the reflected beam is determined
experimentally. However, if only link 2 is actuated, the plate will rotate about the line
joining ball joints Bl and B3 • By constructing the vectors directed from the mid-point
between Bl and Bz, to the point C, before and after the motion of link 2, it can be deduced
that the angle between these two vectors is essentially the angle of rotation of the moving
platform. Similarly, if only link 3 is actuated, the plate will rotate about the line joining ball
Experimental Verification
The experiments were conducted by actuating one of the links at a time, and the
steady state positions of the beam on the wall before and after the actuation were tracked.
The beam's movement is shown in Fig. 4(b). For an input of 100 volts to one of the links,
the image of the beam moved a distance of 6.75mm (0.27 inch), 8mm (0.32 inch), and
6.25mm (0.25 inch), corresponding to the actuation of link 1, 2, and 3 respectively. The
points were half-way in each direction for an input of 50 volts and thus verified that the
motion is linear within the range tested. Although the kinematics of the micro-motion
manipulator is linear with respect to the link displacement, the hysteresis effect of the
piezoelectric actuator may introduce non-linearity to the overall system control. As the
beam deflection was rather insensitive to z-displacement, no attempt was made to measure
To determine the end-point position for the link actuation, the linearized forward
kinematic was employed. The changes in link length were determined from the strain gage
outputs and the rotations about X- and Y-axes were then determined from Equations (3)-
(7). For comparison between the analytical and experimental data, the angles of platform
rotation were calculated_ The results are summarized in TABLE 2. The analytical results
are in agreement with the experimental data. The steady state end-point
that the link dynamics have significant influence on the end-point accuracy of the moving
platform. The dynamic effects of the moving platform and the payload on the individual
link control can be considered as reaction forces acting at the ball joints. It is of interest to
determine the dynamic model of the link actuated by the piezoelectric element. The
dynamic model was obtained experimentally using the standard frequency response
The experimental set-up is shown in Fig. 5. The output signal of the HP analyzer
was fed to a KEPCO power amplifier having a gain of 10, which in turn drove the
piezoelectric actuator. The actuation would displace the link which is sensed by means of
a strain gage that is located at the flexure point A on the link. The strain gage is part of a
wheatstone bridge, the output voltage of which is proportional to the displacement of the
link. The voltage output from the bridge is fed into a ANALOG DEVICES 3B18 signal
conditioning amplifier for filtering and signal amplification. The conversion factor of
combined strain gage calibration and 3B18 amplification was 1.3 volt/mm. The
bandwidths of the power amplifier and the signal conditioning amplifier were 3 kHz and 20
kHz respectively.
The frequency response data were collected with the input voltage to the
piezoelectric element limited between 25-75 volts. The Bode plots obtained experimentally
are displayed in Fig. 6. With the link actuation approximated as a simple mass-spring
system where the stiffness of the flexural joint and the preload spring were 250 kN/m and
4.2 kN/m respectively, the natural frequency was found analytically to be 185 Hz. With the
A THREE DOP MICRO-MOTION IN-PARALLEL ACTUATED MANIPULATOR 359
aid of the curve-fitting function in lIP Analyzer, three pairs of complex zeros and poles
The curve fits well and approximates the experimental data up to the frequency of tOOO Hz
as shown in the comparison in Fig. 6. Hence, the link dynamics is characterized by the
y (s) 3
q
VTs) = K IT (15)
i=1
where
K = 5.93 E-5
aol = 1.97 E+6 llo2 = 2.24 E+ 7 ao3 = 6.63 E+ 7
au = 110 al2 = 2009 al3 = 1005
and where s is a Laplacian operator, V(s) is the input voltage to the piezoelectric element,
experimentally determined transfer function of the link in time domain. A step input of 75
volts was applied to the actuator, and the strain gage output data was sampled at 20 kHz
can be seen from Fig. 7 that the experimental data and the analytical data are in agreement
and exhibited all the dynamics involved. The two graphs do not match perfectly, but they
exhibit how the higher order frequencies are dominant in the response. The higher-order
360 KOK-MENG LEE AND S. ARJUNAN
frequency components are due primarily to the piezoelectric elements, which have also
4. SPECIAL CONFIGURATION
which implies R = r, Equation (3) indicates that a change in link length, ALi> is not
from the forward kinematic equations, the end-point position and orientation are
xc .. yc ""y" 0
(16a)
The linearized forward and inverse kinematics are given in Equations (16) and (17)
respectively. This special case, where 60 = 1f/2 radius, indicates the micro-motion control
of Q, p, and Zc can be achieved without any accompanying change in Xc. Yeo and"y. The
work-envelop is such that the maximum values of Q and P decrease from a maximum at
A THREE DOF MICRO-MOTION IN-PARALLEL ACTUATED MANIPULATOR 361
AZc = 0 to zero at AZc = ± Ymax. It is interesting to note that the kinematic is independent
From the principle of virtual work, it can be shown that the static force relationship
between the Cartesian moment-force vector acting at the moving platform, [Tv Ty> FJ and
Fl 0
2 1 T
x
3 3 If
1 1 1 T
F2 -3 3 Y ,(18)
.f3 If
1 1 1
F3 3 3 F
.f3 z
where Fj , i = 1,2,3, is the actuating force perpendicular to the moving platform through the
ball joint; Tv and Ty are the torques acting about X and Y axes and the force acting in the
actuated by the piezoelectric elements. Hence, the measurements of Fj allow the Cartesian
torques about X- and Y- axes and an actuating force along the Z-direction. Any external
moments about Z-axis, Tz, or external forces along X and Y axes, Fx and Fy must be
supported by the joints in the form of reaction forces. The external forces, Fxand Fy> would
result in bending moments equal to Fx20 and Fy20 on the links. However, as the range of
The motion of the individual link which is subjected to reaction forces at the ball
joints can be controlled using strain gage feedback. Although end-point sensing can be
achieved using a laser scanning technique, the special configuration with small or zero eo,
which would result a close proximity between the moving platform and the base platform.,
offers an alternative means of end-point sensing using miniature forceps between the
moving platform and the base platform. An example of miniature forceps designed on the
basis of micro-bending in fiber optics can be found in reference [35], which is currently
S. CONCLUSION
using piezoelectric elements for actuation has been examined and developed. The
would result in the micro-motion control of a, p, and Ze accompanied with no change in the
subtended angles and, thus, no changes in Xc, Ye, and r. As the kinematics are
working range, rigidity and bandwidth. The static force relationship between the joint
space and the task space is given, which is useful for three DOF force/torque sensing.
The dynamic model of the piezoelectric actuated link has been determined
experimentally, which provides a rational basis for the design and for prismatic joint force
A THREE DOF MICRO-MOTION IN-PARALLEL ACTUATED MANIPULATOR 363
control of the high-speed micro-motion manipulator. The studies have indicated that
ACKNOWLEDGEMENTS
REFERENCES
4. N_ Taniguchi, -Current status in, and future trends of, ultraprecision machining and
ultrafine materials processing," CIRP Annuals, Vol. 32/2, pp. 573-582,1983.
11. A Sharon, and D. Hardt, "Enhancement of robot accuracy using endpoint feedback
and a macro·micro manipulator system," Proceedings of 1984 American Control
Conference. (San Diego, California, June 6-8, 1984) pp. 1836-1842.
12. R. L. Hollis, "A Fine Positioning device for enhancing robot precision," Robots 9.
(Detroit, June 2-6, 1985) Pp. 6.28·6.36.
14. R. L Hollis et. al., "Robotic circuit board testing using fine positioners with fiber
optic sensing," Proceedings of International Symposium on Industrial Robots. Sept.
11-13, pp.315-322, 1985.
15. R. L. Hollis et. al., "A six degree of freedom magnetically levitated variable
compliance fine motion wrist," Proceedings of the 1988 IEEE International
Conference on Robotics and Automation. (philadelphia, PA, April 25-29, 1988) pp.
261-269.
17. F. E. Scire and E. C. Teague, ''Piezodriven 50 micro stage with subnometer," Rev.
Sci. Instrum., 49(12), Dec. 1978.
23. K-M. Lee and S. Arjunan, "Force/torque sensing and micro-motion manipulation of
a spherical stepping wrist motor," Proceedings of the 1988 American Control
Conference (Atlanta, GA, June 1988).
24. K-M. Lee and S. Arjunan, "A three degree of freedom micro-motion in-parallel
actuated manipulator," in Proceedings of the IEEE International Conference on
Robotics and Automation, Scottsdale, Arizona, May 14-19, pp. 1698-1703, 1989.
A THREE DOF MICRO-MOTION IN-PARALLEL ACTUATED MANIPULATOR 365
28. H. S. Tzou, G. C. Wan, and C. I. Tseng, "Dynamics and distributed vibration controls
of flexible manipulators: integrated distributed sensing and active piezoelectric
actuator," in Proceedings of the IEEE International Conference on Robotics and
Automation, Scottsdale, Arizona, May 14-19, pp. 1716-1725, 1989.
29. W. Trimmer and R. Jebens, "Actuators for micro robots," in Proceedings of the
IEEE International Conference on Robotics and Automation, Scottsdale, Arizona,
May 14-19, pp. 1547-1552, 1989.
30. K-M. Lee and D. Shah, "Kinematic analysis of a three degree of freedom in-parallel
actuated manipulator," Proceedings of the 1987 IEEE International Conference of
Robotics and Automation. (Raleigh, NC, March 31-April 3, 1987) Also in IEEE
Journal of Robotics and Automation, Vol. 4, Issue 2, pp. 354-360, June 1988.
31. K-M. Lee and D. Shah, "Dynamic analysis of a three degree of freedom in-parallel
actuated manipulator," IEEE Journal of Robotics and Automation, Vol. 4, Issue 2,
pp. 361-367, June 1988.
35. J. Winger and K-M. Lee, "Experimental investigation of a tactile sensor based on
bending losses in fiber optics," Proceedings of the 1988 IEEE International
Conference on Robotics and Automation. (Philadelphia, PA., April 26-29, 1987).
366 KOK-MENG LEE AND S. ARJUNAN
**************************************** •••
Ball joint
----
,-_",....::Moving platform
Extensible
link
'{
Base platform
--------..;.......
Compress
'Base or
A THREE DOF MICRO-MOTION IN-PARALLEL ACTUATED MANIPULATOR 369
Through-holes -&.~==~L-J--j---+-
Base
Piezoelectric Actuator
Platform End
Link End
AQ
1-=-- --
Yp
c --I b - 1
- - - - - <.....
361 ""
31118 slgnQI
Cl.MpIIfIIP,. __---StrQln QQgIP Input
MOdU1IP
HP slgnQI
QJlQl yZlPr
-20
-40
-60
-80 Experimental
Curve Fit
LoS Hz
60 100 1000
0
(/)
Q)
Q)
\...
C'l
Q) -90
0
Q)
(/)
'"
.<::
0.. -180
-270 LoS Hz
60 100 1000
0.1
0.09
e 0.08
""'
e
-=
'-'
Q.,
0.07
~
.;.:: 0.06
:3 0.05
...
0
-=<.>e 0.04
<.> + - _ Experimenta 1
("j
0.03
-a
~
+ Analytical
S'" 0.02
0.01
0
0 10 20 30 40 50
Time (msec.)
Abstract
1. Introduction
using piezo elements are also being studied, e.g. microvalves on silicon
wafers[3), precise positioning mechanisms utilizing rapid deformations
of piezoelectric elements[41, These piezo elements were also applied
to miniature mobile robots[5)-[10), precise positioning devices and
Scanning Tunneling Microscopes[6) to cause high vibration and micro
deformation.
In this paper, we focus on miniature mobile robots which employ
piezo elements and also describe their construction, principle and
characteristics. Then we mention possible future developments. First
we describe a miniature mobile robot that uses piezo vibration to can
move within a thin tube. The robot consists of a piezo bimorph with
elastic fins attached at an angle. The robot moves as a result of
differences between the forward and backward friction of the fins
against the tube wall induced by piezo vibration. After analyzing the
dynamics of the piezo elements, we analyzed the mobile mechanism of
the robot by extending Hamilton's principle using the dynamic results
of the piezo vibration analysis. Measurements of both mobile velocity
and tractive force of the experimental robot agree closely with
theoretical results. This indicates that the proposed dynamic mobile
mechanism accurately expresses robot motion within a thin tube.
Second we describe a wireless mobile robot that uses piezo vibration
to move within an enclosed space such as a circular tube. The robot
consists of a piezo bimorph with conductive elastic fins attached at an
angle. Electric power is applied to the piezo bimorph through these
fins from electrodes attached to the tube wall. The robot can be
applied to display devices, precise positioning devices and miniature
mechanisms. Third we mention a miniature mobile robot for mobility
on a wire and a miniature actuator for plate movement. These consist
of two clamp components and an expansion and contraction component.
A MINIATURE MOBILE ROBOT USING PIFZO ELEMENTS 377
ILz _ I L ,
U, U2
dtIE---"---{
Top -view
\ .,. -- - -'-J )
Side · view
Fig.2 Bending vibration.
Fig.1 Mobile robot in a thin tube.
0.5 y
o L
E 0.4
-I
S 0.3
o 0
c:
0
.;; o
() 0.2 o
Q)
;;:: 000 Mv
0
Q)
0.1
x
0
0 0.2 0.4 0.6 0.8 1.0
f (kHz)
the left side is smaller than that on the right side, so the robot moves
to the left.
Experimentally measured frequency characteristics of the piezo
bimorph are shown Fig.3. These imply that the mobile velocity and
tractive force of this mobile robot depend on the frequency of the
alternating current voltage applied to the piezo bimorph. This dynamic
characteristic analysis is therefore an important aid to understanding
the motion of this robot. We first analyzed piezo bimorph vibration,
determining the relationships between the structure (positions of
elastic fins, size of a piezo bimorph, etc), the vibration frequency, and
displacement. We then determined the mobile mechanism by applying
these vibration results to the extended Hamilton's Principle.
A MINIATURE MOBILE ROBOT USING PIEZO ELEMENTS 379
bimorph and the elastic fins. Figure 4 shows the analytic model. In
this model, elastic fins are replaced by equivalent spring elements,
which are joined at distances L1 and L2 from the left edge of the piezo
bimorph. Sinusoidal bending moment Mv as shown in FigA is applied to
the piezo bimorph. The following dynamic equation and boundary
conditions are obtained when we consider piezo bimorph damping.
The dynamic equation is
where
we assume that both ends (x=O, L) are free, we get the following
relations,[18]
M=Mv-Mo=Mv-EI (a 2y/at2) =0
F= aM/at=-EI (a 3y/at3) =0. (2)
380 S. AOSHIMA AND T. YABUTA
where
b: bimorph-width
h: element-thickness
L
k
Y= 8n(t) Xn(x) (5)
rFl
Xn(x)
=(cos AnL - cosh AnL)(coS Anx - cosh Anx)
equation.
ro 2 = An 4 (El/pS) (8)
We then multiply the dynamic equation (1) by Xndx and integrate from
Integrating by parts for the second and third term in equation (1) and
where,
an = pSln
bn = CIAn 4 1n
Cn = EIAn 4 1n
I, = f X<i' dx . (11 )
dH+c' g,
d" d" b,oo e,
-a,oo'
dn+c,
0
d" d.. Inoo g, e.
-a.ro'
..
. 0
d",,+c.
d" cL.. ----- .. - g. e.
-a.ro' b.oo
dll+c'
-bIro d" d" h, f,
-a,rJ
0 d,,-H:,
-Inoo dll d.. h, f,
-a.ro'
.
0
d" cL.. . _--_ .. da+c. h. f•
-b.oo -a.cJ
(13)
The terms gn and hn of equation (12) can be obtained by solving matrix
A MINIATURE MOBILE ROBOT USING PIEZO ELEMENTS 383
equation (13). Using gn, hn from eq.(13) and eqs. (5), (12),
displacement y is given by
k k
Y = (L, gnXn) cos rot + (L, hnXn) sin rot
n=1 n=1
k k
=[(L, gnXn)2+(L, hnXn)2J1/2
n=1 n=1
k k
esin[rot+tan- 1((L, gnXn)/(L, hnXn))]. (14)
n=1 n=1
k k
D( ro ) = [(L, gnXn(L1))2 + (L, hnXn(L1))2 ]1/2 . (15)
n=1 n=1
Table Parameters
K1,2 = 0.046 [N emm- 1]
C =300 [gemm- 1e s- 1]
L =38 [mm]
L1 =8 [mm]
L2 =30 [mm]
E =2.45E4 [Nemm-2]
384 S. AOSHIMA AND T. YABUTA
0 .5 .--- - - -- - - --..
o 4 , - -- - - - - - - - .
E 0.4
5
c 0.3
o
.~ 0.2
c~ 0.1
°o~~~~~~~~~~ - 40~-~--~-~-~e
0.2 0 .4 0 .6 0.8 1.0
x (mm)
f (kHz)
Fig.6 Numerical vibration deflection.
where
Vt(ro) = ro O(ro) tan (X ( 1B)
ro : angular frequency
O(ro) : maximum displacement of the joint part, and
Vs : steady velocity.
E = Ww + Woo ( 19)
>-
.~
g Vsr---~~-4----*----4
~ 0 I------/---t-'\c-- - t -- /--I-',.---I
t, tz
T T 0 T T
-2" -4 4 "2
Fig.? Fin tip displacement parallel Time
The work W1 done by elastic fins 1 and 3 on the inner wall is given by
equation (22).
(22)
W2=2 i T/4
-T/4
NO~2V2dt (23)
where No is the force of the elastic fin tip against the inner tube wall.
given as
elastic fin and the inner tube wall. The term ex is the angle between
A MINIATURE MOBILE ROBOT USING PIEZO ELEMENTS 387
the piezo bimorph and the elastic fin. W1 is obtained from equations
Wo = FVsT/2 (28)
dW1/dVs.
Equations (31) and (32) are obtained by using equations (20) and
VslVf = X (39)
FIND = f. (40)
A MINIATURE MOBILE ROBOT USING PIEZO ELEMENTS 389
= O. (41)
Substituting tl and t2 in equations (20) and (21) into equation (41), the
(42)
This makes X2=0.16 and cos- 1X=1.16. However, the approximate results
are X2 = 0 and cos- 1X (=1t/2 - X)=1.17 if eq.(43) is applied. If we ignore
errors of up to 20%, the assumption is acceptable. Substituting
Next, we obtained the tractive force F for the robot. Putting Vs= 0
in equation (45), then f=2( 111 (0)) - 1l2(0))). Therefore, the tractive force
F = fNo
= 2No( 111 - 1l2) (46)
2.4 Experiment
To determine the mobile mechanism of the robot, we have
theoretically analyzed the dynamic characteristics of both piezo
bimorph and robot motion. We now compare the experimental work
with this theoretical analysis. A pilot model of the mobile robot in a
thin tube (tube inner diameter 20mm) is shown in Fig.9. Figure 10
shows a mobile robot for an 8 mm inner diameter tube. In this
experiment, 60VAC was applied to the former robot. The frequency-
tractive force characteristics are shown in Fig .11. Tractive force
depends on the excited frequency and has a maximum value of 0.12 N at
the 600 Hz resonance frequency. Comparing eq. (46) and the
experimental results, we can infer that the equivalent dynamic
coefficient depends on the excited frequency. This is because it is
defined by both the dynamic coefficient and the contact force. Using
A MINIATURE MOBILE ROBOT USING PIEZO ELEMENTS 391
0.20 ~ 1.0r - - - - - - - - - - - - ,
3 0.15 'CIII 0c: 0.8
III
t.I :!;i ·fl
....0 0.10
:;:E 0.6
III
> 3"1ij
.;; ... .!:! 0 .4
j~
t.I
...'" 0.05
t- ~ ~0.2
·5"0
tT.... 0 L - - -____.l.------,-J
oa SOD 1000 woo 500 1000
Frequency (Hz) Frequency f (Hz)
eqs. (22), (25). and (46) together with the measured F, the equivalent
dynamic coefficient 110(0)) is obtained as
1.0,------ - -- - ,
i .
~ 5 0.8 -
:g 'B
~:E 0.6
813
~ 'E O.4F-C)..()..(:>-_
-to
o _-
.~ ~
g.~ 0.2
W 0 °°
~ 600 , - - - - - - -- -- - - - - - - -,
500 ~---------------' ~
~ 400
E
lil 400 ! 300
~
E
.5 300 ;; 200
>-
.~
<II
g 100
>>- 200
...
Qj
2.5 Discussion
In this section, we discuss the design of the mobile robot. When
both mobile velocity and tractive force are specified, it is necessary
to obtain optimum design parameters. We must consider size and
material constants for both the piezo bimorph and the elastic fins, as
well as the orientation of the elastic fins .
The mobile robot is designed using the following processes.
[1] Measurement of the coefficient of dynamic friction between the
elastic fins and the tube wall when changing sliding velocity.
[2] Calculation of the required pushing force No from the desired
tractive force by using the following eq. (48). Then determination of
the spring constants k1 and k2 of the elastic fins in consideration of
the pushing force No.
[3] Clarification of the design parameter range taking both the tube
diameter and material characteristics into consideration.
[4] Calculation of both the necessary deflection of the piezo bimorph
and the resonant frequency co by substituting the desired mobile
velocity and the tractive force into eq. (49). Then determination of the
structure parameters which satisfy the required deflection D and
resonant frequency co .
4r-----------------~ 700.---------------~
I
3 650 I
I
I
E ""'
I
I
5
N
I
2 .; 600 I
k=O.046N/mm ,
I
\
,
c
550 -- .... ,\
"-
\
\
\
OL---~--~~--~--~ OL~--~----~--~--~
o 5 10 15 20 o 5 10 15 20
La (mm) La (mm)
Fig .16 Deflection VS. the position Fig.17 Resonance frequency vs. the position
of the elastic fins. of the elastic fins (k=O.046 N/mm).
800~--------------~
750
700
N I
I
J: I
650
...
I
~ I
I
I
I
I
600 I
\
~
,, "
550 \
\
\
\
\
0
0 5 10 15 20
Le (mm)
Fig.18 Resonance frequency vs. the position of the elastic fins (k=3.92 N/mm).
2000.----------------, 2000,---------------~
~ 1500 ~ 1500
E E
! !
1000 1000
c~
- 500 c 500
-~
OL-__ ~ ___ C_ _ _ _
~ ~
O~--~---=--~--~
o 5 10 15 20 o 5 10 15 20
Le (mm) Le (mm)
Fig .19 frD(fr) vs. the position Fig.20 frD(fr) VS. the pOSition
of the elastic fins(k=O.046 N/mm). of the elastic fins(k=3.92 N/mm).
1.4 750.------------------.
1.3
700
E 1.2
~
E 650
:S 1.1
0
- 600
1.0
0
0 2 4 6 8 10
K (N/mm) O~~--~--~---L--~
o 2 4 6 8 10
Fig.21 Deflection vs. the spring constant K (N/mm)
825
N 800
J:
E
! 775
...
-
0.. 750
o 2 4 6 8 10
K (N!mm)
parts.
f (a'y/ax')Xodx
-f (a'y/ax')(dXo/dx)dx
-f (ay/ax)(d'Xo/dx')dx.
=<ll + f (y(d'Xoidx')n
400 S. AOSHIMA AND T. YABUTA
(A 1-1)
r2
Jtl
[<5L(q1, ••• , qn, q1', ••• , qn', t) + I Oi' <5qi]dt = 0
(A2-2)
L = T-U
= T. (A2-3)
t
J. 28Ldt = J.t2 oTdt = 0, (A2-4)
tl tl
t 2
J. (L ai' ox)dt = 0, (A2-5)
tl
t2
o J. L (ai' x)dt = 0 (A2-6)
tl
t2
d/dVs( J. Wdt) =0 (A2-7)
tl
When joint part B of the elastic fin and the piezo bimorph cause
displacement d, the point B moves to point B' and elastic fin tip A
moves to point A'. Assuming that elastic fins do not deform during
relation,
U1,3(ro)=BC-B'C'
=BC-(A'B'2_A'C'2) 1/2
=BC-(A'B'2-(AC-d)2) 1/2
=Llcos a _(Ll2_( Llsin a -d)2) 1/2
And
U2,4(ro)=D(ro)sin rot tan a. (A3-2)
follows :
3.2 Characteristics
We made two robots Land R of the same size, and investigated
their movement characteristics. Figure 26 shows photograph of the
robot. The robots are 30 mm long, and the tube is 18 mm wide.
Figure27 shows velocity vs. frequency of the two robots. Their
characteristics are almost identical. The robots start moving at 200
Hz. The first velocity peak is 1000 Hz and the second peak is 1500 Hz.
The maximum velocity is 180 mm/sec at 1600 Hz. Because the elastic
fins are flexible, the robot can move into a winding tube, Therefore
this robot can be applied to a display device. Figure 28 is a display
200
150
~
!!!.
.s 100
E
.~ 50
0
0
Q;
> 0
-50
0 500 1000 1500 2000 2500 3000
Frequency [Hz)
Fig.26 Photograph of the robot. Fig.27 Velocity vs. frequency of the two robots
Conductive
Electrode elastic fin p' b'
f Mi'mO~h~ IL2 _ I L ,
(a)
(b)
Fig .25 Bending vibration
Fig.24 Robot structure.
A MINIATURE MOBILE ROBOT USING PIEZO ELEMENTS 405
device formed into the "dynamic loop" logo of NTT. The diameter of the
larger loop is 60 cm. The diameter of the smaller loop is 20 cm. The
robot can complete a circuit in 30 seconds. Figure 29 shows the robot
in a circular tube.
~~
Fig .30 Two direction mobile robot structure.
406 S AOSHIMA AND T YABUTA
U 20
3l
E
oS 10
-o6
~ -10
on off
1
3
on
4
5
on
6
408 S. AOSHIMA AND T. YABUTA
Amp.
Reflection
plate
\Nylon wire
Switching Micro
circuit computer
2.5
J----n,o
E
.Eo 2.0 ho+ 60
~ 1.5
50
E =Ol
<I>
g 1.0 .Eo 40
a.
is 0.5 :J:
LL
30
e0
<I>
- - - Cal.
Fig.34 Displacement vs. voltage.
100 200 300 400 500
Voltage IV]
.e.
E
oS X10-1. . - - - - - - - - - - - - ,
...J £=20 (mm)
£=20 (mm)
0
E
8
• Exp.
:g; 4
Q)
6 oS 3
E
Q) :c
u u.
co
a.
III
4
Applied voltage eo
Q) 2
'5 500 (v)
(ij
2 u.
E
0
•
N 00 60 120 180 240 60 120 180 240
"5
:c Arch form coefficient K [%] Arch form coefficient K [%)
Fig.37 Horizontal displacement DL Fig.38 Force FH vs. arch form coefficient K.
vs. arch form coefficient K.
following equation.
W=DL.FH/2 (52)
:'10- 3 ..---------.--=Ex-p-.--, E
E
;;:::
.2l 1.6 L=20 (mm) .2l 4
.c: • •
~ 1.2 MAX ~-,-_ _ _ '0 co
Q)
3
.c: iii
c,.c:
.c: 0.8 c:
Q)
u
...
2
c, 500 (V) - co
c: ~---
... .c:
Q)
Applied voltage 0 eco
~ ~
~ 0
o 60 92 120 180 240 0 60 120 180 240
Arch form coefficient K [%]
Arch form coefficient K [%)
Fig.40 [work/length of arch/arch area]
Fig .39 [work/length of arch]
vs. arch form coefficient K.
vs. arch form coefficient K.
412 S. AOSHIMA AND T. YABUTA
E
oS
...J 2.0 K £=20 (mm)
0
• 100 0100
'E 60 I
OJ 1.5 60 2 layers
E
OJ
'"
0 30
0
CIl 1.0 1 layer t;::"
50
0.. Ol
:a oS 40
I/)
0.5
(ij I
'E u.
0
.t:l
00 100 200 300 400 500
OJ
~ 30
(; 0
I Voltage [V] u.
20
Fig.41 Horizontal displacement)
vs. voltage of the arch bimorph. 10 Cal.
u
Q)
Applied voltage
soo ty)
(J)
E
1 .s
W10 (mm) E
e
0.1~"118
A. C component 2 layers
£Q)
0.010.1
8 component 3 layers
'0
Teflon tube inner diameter ~0.65
Nylon wire diameter 910.40
.2:-
'0
o
Q)
>
1 10
'-:-----L----L-L..J...J...LL1JL-----L--'.-L..LLLllJ
Frequency [Hz]
Fig 44 Trial mobile robot (large model) Fig 45 Velocity of the robot
vs frequency (large model)
414 S AOSHIMAANDT YABUTA
1"-
~ 2.5
f/)
• A direction
o C direction I
f1 I
E I I
1 E.
0:
£J
e15
2.0
Applied
I
I
"
~
I I
\I
~
TI
I
P
t;)
/\ I
\
I
~0.40
10 . voltage j I I \
500 ( v ) I }6
A, C component 1 layer e / j'1
'0 1.0 /. /' \ 'i'
B component 1 layer
f ( ~ \
Teflon tube inner diameter ~0.40 o 0 .5 ,;// \
OJ ,.;e
Nylon wire diameter ~0.29 > /'
o ,.
o 10 20 30 40 50 60 70
Frequency [Hz]
Rod
8 expansion and contraction
component Table2 One cycle of on-off control.
C clamp component
Plate
~O~
/ 1
I I
2 J-, OFF
I~
~
I I
Lead wire
3 ~
I
P I I
4 II----, ON
Switching circuit Micro computer V' ~
I I
5 OFF~
~
I I
I I
6 ~ .:1X I I
Fig.49 Miniature actuator for plate movement '-----J ~ l
and controller.
416 S AOSHIMAANDT YABUTA
150
50 50 50 Spring
In
(I)
ScrewM4
! 30 !
...
0
Rod
0 0
In M
19 15
53
Clamp component
Large model 0.35 mm
Small model 0.087 mm
Model L
'0 - 0 - - A direction
Q) 500
--E
(J)
..:!: 400
- _ C direction
Model S
-O-A
~ --0-- Aonly
'(3
300
0
Qi -e-C ~ ~---
>
200
'E
500 (V) Q)
E
o ~---<>-C>---
Q) 100
Applied voltage
> ~- Applied voltage
0 500 (V)
:E
00 20 40 60 80 100 120
Fig.54 Device for measuring displacement
Frequency [Hz]
vs. frequency.
Fig.55 Movement velocity vs. frequency.
miniature mobile robot. The robot moves on a plate when the applied
voltage and frequency are changed. Figure 57 shows a eel type
miniature mobile robot. The robot was moved by progressive wave at a
velocity of 3 J.l.m/sec. This value agreed with the calculated value. A
+.
_
~
;:;:
60
t}
- 60
~. I(S)
_v\
~2
I
5
St~~~ ?~
Its) \
/
3
E 0 U 1
...
Q)
Q.
Model motion
12mm
E
E
~
""
robot was several 10 11m/min. This value was much smaller than the
calculated value because the nail slipped on the surface of the base.
The velocity will be improved to 1 mm/min by increasing the surface
friction. The robot can be used as a precise positioning actuator in a
super high vacuum or in a place where the connection of lead wires is
difficult. At present this technology is at the basic research stage, but
future practical application seems possible by improving materials and
mechanisms. A selection of both mechanisms and materials that suit
the purpose of the system is necessary in order to construct a good
piezo smart system. Possible materials are polymers, gums, ceramics
or compound piezo materials. These materials have unique
characteristics. Possible mechanisms are the pile type, bimorph type,
inch worm type, progressive wave type, push type or displacement
expansion-contraction mechanism type. Many systems can be
constructed by combining these mechanisms and materials. Many
robots have been devised based on animal, insect or fish motion. For
example the inch worm type and the eel type.. These mobile
mechanisms and principles are very helpful. Now we should thoroughly
investigate the mechanisms and motor principles of the leg and wing
movements of insects in order to make a good piezo smart system.
Moreover for a piezo micro system, we should study on motor
mechanisms and principles of microorganisms.
7. Conclusion
In this paper we desribe miniature mobile robots using piezo
elements. First we describe a miniature mobile robot that can move
within a thin tube using piezo vibration. The robot consists of a piezo
A MINIATURE MOBILE ROBOT USING PIEZO ELEMENTS 421
Acknowledgments
The authors would like to express their appreciation to NTT
Transmission Systems Laboratories Dr. Sadakuni Shimada, and Tokyo
Institute of Technology Prof. Teru Hayashi.
References
Micro Electro Mechanical Systems, Napa Valley, CA, Feb. 1990, pp. 1-8.
[31 S. Nakagawa, S. Shoji, M. Esashi, "A Micro Chemical Analyzing System Integrated on
a Silicon Wafer", The 3 rd IEEE Workshop on Micro Electro Mechanical Systems, Napa
[6] S. Aoshima, T. Tsujimura and T. Yabuta, "Design and Analysis of a Midget Mobile
Robot using Piezo Vibration for Mobility in a Thin Tube", Proceedings of The Japan Society
[7] S. Aoshima and T. Yabuta,"Wireless Mobile Robot using Piezo Vibration for Mobility
in a Thin Tube", Proceedings of The 8th the Robotics Society of Japan Annual Meeting 1990
The Japan Society of Precision Engineering 1986 Spring Annual Meeting (in Japanese).
1986, pp.319-320.
[9] S. Aoshima and T. Hayashi, "Piezo Micro Actuator for Plate Movement", Proceedings
of the Japan Society of Precision Engineering 1987 Spring Annual Meeting (in Japanese),
1987, pp.733-734.
[11] M. Komuro, O. Kitamura and S. Okayama, "Ion Beam Fabrication using Piezo-
Actuator and Liquid Metal source", Microcircuit Engineering '88 September, 1989, Wien,
[1 2] T. Okada, and T. Sanemori, "MOGRER: A Vehicle Study and Realization for In-Pipe
Inspection Tasks," IEEE J. Robotics Automation, Vol. RA-3, No.6, December 1987,
pp.573-582.
[13] T. Fukuda, and H. Hosokai, "A Study on an Inspection Mobile Robot for the Inside of a
Pipe-line," Trans. Jpn. Soc. Mech. Eng.,(in Japanese), Vo1.52, No.4??, C, 1986, pp.1584.
A MINIATURE MOBILE ROBOT USING PIEZO ELEMENTS 423
[1 4] T. Morimitsu, and H. Sakata, "A Study of a Vibrating Bristled Vehicle for Small
[1 5] S. Yamamoto, K. Sato, H. Fukushima, and H. Saito, "A Study of a Pipe Inside Mover
with Elastic Bristles," Trans. Jpn. Soc. Mech. Eng.,(in Japanese), Vo1.54, NO.506, C,
1988, pp.2471.
[16] S. Aoshima, T. Morimitsu, and T. Tsujimura, "A Study on a Midget Mobile Robot in a
Thin Tube using Piezo Vibration," Trans. Jpn. Soc. Mech. Eng.,(in Japanese), Vo1.55,
deflection for a midget mobile robot in a thin tube," Proceedings of the 6th Annual the
[1 8] H. Nukiyama, and T. Suzuki, "On the Mode of Vibration and Electroacoustic Constants
June1942, pp.367.
H. S. Tzou
Department of Mechanical Engineering
Center for Robotics and Manufacturing Systems
University of Kentucky
Lexington, Kentucky 40506--0046
ABSTRACT
Hamilton's principle, and linear piezoelectric theory. The derived equation can be
simplified to account for many other common geometries (e.g., plates, cylinders, beams,
etc.) which are the basic components of robotic and machine structures/components. A
simple reduction procedure is proposed and examples are demonstrated in case studies.
1. INTRODUCTION
"hardware" way, i.e., the machines and robots structures are integrated with sensors and
actuators.) This report presents a new integrated piezoelectric sensor/actuator design for
1.1. Background
In general, machine and robot structures are "distributed" in nature, i.e., structural
dynamics are functions of spatial and time variables, which are classified as
regarded as a general DPS from which many other structures, e.g., plates, cylinders,
spheres, etc., can be derived [2,3]. Moreover, conventional sensors and actuators usually
measure and control spatially discrete locations. Serious problems can occur when these
devices are installed at modal nodes or lines. One solution to this problem is to use
distributed sensors and actuators for structural identification and control of DPS's [4].
Thus, new development on active distributed vibration control and structural identification
much interest in recent years. Tzou and Gadre designed a dual-purpose piezoelectric
exciter and vibration isolator [5,6]. Test data showed a close comparison with the
theoretical solutions. This technique was also applied to rotordynamic vibration controls
[7]. Use of distributed piezoelectric actuator in control of flexible beam oscillation were
studied [8-12]. Tzou and Gadre derived a generic layered shell actuator theory for
distributed vibration control of flexible shells [13]. Tzou derived an integrated distributed
sensing and control theory for thin shells [1,4]. Tzou and Tseng also formulated a new
piezoelectric finite element for distributed sensing and control of shells and plates [14]. A
generic theory on structural identification and vibration control of DPS's using electroded
1.2. Objectives
In this study, based on a generic DPS - a generic shell element, theories on the
sensor and actuator are derived. Simplification of the theories to other geometries will also
be discussed. Modal expansion method is incorporated with the theories to express the
general identification and control equations in modal coordinates. Active distributed
2. PIEZOELECTRICITY THEORY
Piezoelectric phenomena were first observed by the Curies brothers, Jacques and
referred to as the direct piezoelectric effect. Conversely, when an electric field is applied to
the material, a mechanical stress or strain is induced; this is called the converse
piezoelectric effect. In this study, the direct effect is used for distributed structural
identification and the converse effect for the active distributed vibration control of DPS's.
There are two constitutive equations representing the direct and converse piezoelectric
and
where is} is the strain vector (i.e., is} = { S11 S22 S33 S23 S31 S12 }t); {T} is the stress
vector (i.e., {T} = { T11 T22 Taa T23 T31 T12 }t); [sE] is the elastic compliance matrix
measured at constant electric field; [fT] is the dielectric matrix evaluated at constant
strain; [d] is the piezoelectric constant matrix; [cD] is the elasticity matrix evaluated at
constant dielectric displacement; [h] is the piezoelectric constant matrix; {D} is the electric
DISTRIBUTED PIEZOELECTRIC TRANSDUCERS 429
displacement vector; [.J t indicates the matrix transpose; {E} is the electric field vector;
and [tf] is the dielectric impermeability matrix evaluated at constant strain. The
mechanically and electrically balanced and the two effects can be decoupled.
direction (1l3), Le., isotropic in transverse direction (1l3) and anisotropic in in-plane
directions (Ill and 1l2). Polymeric polyvinylidene fluoride (PVDF) and some piezoceramics
are in this category. The piezoelectric matrix [d] of this category can be expressed as
[16,17]'
000
[d..J = 0 0 0 (5)
IJ
In this section, a distributed piezoelectric sensing theory for a generic DPS - shell-
is developed. The output signal equation is extended to incorporate with the modal
structural identification theories. (This shell structure has been proved that its system
equations can be directly simplified to account for many other commonly occurred
geometries, such as plates, spheres, cylinders, beams, arches, cylindrical panels, etc [2,3].)
The generic shell structure is made of a passive elastic material sandwiched between two
thin piezoelectric layers, Figure 1. Note that the top layer serves as a distributed actuator
for distributed vibration control and the bottom a distributed sensor for structural
identification.
It is assumed that the piezoelectric layers are relatively thin and flexible compared
with the elastic shell. Thus, the structural dynamics of the shell can be approximated by
theory for thin shells is developed using Maxwell theory [1,4]. Note that in the distributed
sensing application, only the direct piezoelectric effect is considered. Also note that only
the transverse electric field E3 is considered in the shell configuration. Thus, the strains
Using Maxwell's equation, one can derive the electric field E. related to the electric
I
potential ¢* as
{E.}
I
=- V ¢* (6)
where V = { .J-,
U(}:l
.J-, .J- }t.
U(}:2 U(}:3
The voltage across the electrodes can be obtained by
integrating the electric field over the thickness of the piezoelectric sensor layer, i.e.,
(7)
where the superscript IpS" denotes the distributed piezoelectric sensor layer and h Ps is the
Appendix 1.) Since the piezoelectric layer is relatively thin and flexible, it is assumed that
the strains in the layer are constant and equal to the outer surface strains of the shell.
432 H.S.1Z0U
Considering the induced in-plane strains (S~s and S~s) and the piezoelectric
coefficients (hal and h a2 ), one can derive an electric voltage equation as,
(8)
where S~s and S~s are the in-plane strains in al and a2 direction, respectively. The
in-plane strains in the piezoelectric sensor layer, the bending strains of the outer shell
+ 1 aA l [ u2 1 aUa (9)
A!A2 ---oa; IIf2 - ---x; 8a2
where d~s and d~s are the distances measured from the shell neutral surface; Al and A2 are
Lame'S parameters; Ul and U2 are the in-plane displacements; 1R1 and 1R2 are the radii of
The electric displacement D~s can be expressed in terms of the voltage <p
(11)
DISTRIBUTED PIEZOELECTRIC TRANSDUCERS 433
The distributed piezoelectric sensor outputs can be calculated in two ways: 1) a discrete
voltage ~s and 2) an averaged ¢lis. For discrete voltage calculation, it is assumed that
there are infinite electrodes on the surface of the piezoelectric layer. Thus, knowing the
discrete voltage amplitude of every discrete location, one can construct a distributed
voltage contour, voltage map, corresponding to any time instant or state. In the second
method, there are limited number of electrodes made on the surface of the layer so that an
Integrating Eq.(ll) over the electrode surface APs yields a charge. Setting the
resulting charge expression equal to zero gives an averaged open-circuit voltage ¢lis of the
sensor,
(12-a)
(12-b)
Note that (ll and (l2 constitute a tw~imensional curvilinear plane. For discrete point
voltage ¢l~s, the above equation can be modified as,
434 H.S.1Z0U
(13-a)
+ A1A2
1 aA I [u 2 1 aU3 ] 1
Ga2 IIf2 - -x; 8Ci2
Dynamic response of the shell can be expressed in modal coordinates Ak using the
modal expansion method [3]. That is the dynamic response ui(Cil'~,t) is a summation of
all participating modes Uik with individual modal participation factor \ '
!D
where i = 1, 2, 3 indicating three principal axes. Since for a distributed system, the
number of modes are infinite, k =1 -!D. Substituting the modal expansion equation into
f
ID ID
a a "-'). U
pS
¢a =
h ps
APs APs
{ h dPs [ 1
31 1 At 7m!
1
[ ---urt ..1~01
"-'). U
k 1k -
1
At ---oat ..1L k 3k]
k=l k=l
+ aA 1 1 "-'). ID U a "-').
ID u ]
~ ---oa; [ IR2 ..1~01 k 2k
1
-
1
A2 0"ii2 ..1L k 3k]
k=l k=l
ID ID
+h
32
d~s [ 1 a 1 "-'). U
A2 002 [ IR2 ..1~01 k 2k -
a "-'). U
1
A2 0"ii2 ..1L k 3k]
k=l k=l
+ ID U
"-'). a "-').
ID u ]
~ ---oa; [ IR2 ..1L ..1~01
1 aA I 1 1
k 2k - A2 0"ii2 k 3k]
k=l k=l
ID ID
+h a a "-').
32
d~s [ 1
A2 002 [ IR2
1
..1~01
"-'). U
k 2k -
1
A2 0"ii2 ..1~01 U
k 3k]
k=l k=l
+ aA ID u _ 1 a "-'). u ]J
~
1
~ 1 "-'). 2 ID
[ ---urt..1L k 1k At ----aiil..1L k 3k] . (16)
k=l k=l
It is observed that the distributed sensor output signal (either averaged or discrete
formulation) is contributed by all shell vibration modes. That is the distributed sensor can
measure or identify all vibration modes of the shell structure. Also note that the modal
participation factor ).k could be zero for any kth mode not participating in the shell
oscillation.
436 H. S. TZOV
However, for a fully electroded shell, anti-symmetrical modes can induce zero
averaged voltage because the positive output is canceled out with the negative output in
the DPS. In this case, segmenting or shaping surface electrodes is required to produce
non-zero outputs.
As discussed previously, a discrete point voltage output ¢~s( O!I *,0!2 *) can also be
DPS. Connecting all discrete local voltages forms a voltage distribution contour
representing the current vibration state of the DPS. A kth distributed modal voltage
¢~s k( 0!1 '0!2,t) function, which describes modal voltage distribution, can be calculated in
this way,
(17)
in which the modal participation factor Ak is unity. (Note that any Uik could be zero if it
does not contribute the mode.) Note that the modal voltage distribution contour is
In the tri-Iayer shell structure, the top piezoelectric layer serves as a distributed
actuator for active vibration suppression and control. For a bi-axially polarized
piezoelectric actuator, a voltage applied to the distributed actuator layer results in two
in-plane strains (SiC and S~c) due to the converse piezoelectric effect, where the
superscript "pc" denotes the distributed piezoelectric controller/actuator layer. Since these
strains are located a distance, a moment arm, away from the shell neutral surface, two
distributed counteracting control moments are then introduced. Figure 2 illustrates the
Piezoelectric
Strains
It is assumed that the piezoelectric actuator layer is not constrained and free from
external in-plane forces, the stress effects are neglected in the analysis. Besides, the
applied control voltage ¢pc is much larger than the induced voltage ¢ due to the direct
effect in the distributed actuator. Thus, this induced ¢ is neglected in the derivation.
In this section, control moments of the open-loop and closed-loop feedback controls
of the DPS are derived. In the closed-loop control systems, two control algorithms,
negative velocity feedback control, are proposed [4,14]. Systems dynamic equations and
the state equation of the open and closed loop controls are also formulated.
In the open-loop control, a reference signal ¢pc can be injected into the distributed
piezoelectric actuator. (N ote this ¢pc is not taken from the sensor layer.) Due to the
converse piezoelectric effect, two induced strains, S~c and S~c, in the distributed
where hPc is the piezoelectric actuator thickness. In order to keep force equilibrium in the
in-plane directions (defined by at and a2 axes), the induced strains in the piezoelectric
layer cause effective in-plane strains. These strains result in resultant forces and
moments. The moments are produced through the moment arm from the mid-plane of
DISTRIBUTED PIEZOELECTRIC TRANSDUCERS 439
each layer to the neutral surface of the shell. Since a biaxially oriented piezoelectric
actuator is used for the distributed sensor/actuator and its thickness is much thinner than
that of the shell, the produced line moments in (}:1 direction M~~ can be calculated by,
(19)
where dl c is the moment arm (distance measured from the neutral surface to the
modulus of the piezoelectric actuator. Similarly, the line moment in (}:2 direction can be
calculated by,
(20)
Note that the injected voltage needs to be controlled in a way that the produce moments
can counteract and suppress the shell oscillation. In a feedback control system, this
reference signal is directly taken from the distributed sensor measurements. Thus, a
In this case, it is assumed that the distributed sensor output ¢ps is directly used in
a feedback loop. In the later derivation, two feedback algorithms are considered: 1)
negative velocity, proportional feedback control and 2) constant amplitude, negative velocity
feedback control [4,14]. Note that the piezoelectric sensor output ¢ps used in the later
In this case, the sensor output voltage is directly differentiated, amplified and fed
back into the distributed actuator, i.e., a negative velocity proportional feedback, which
feedback control voltage ¢pc and the sensor output voltage ¢ps can be related by
(21)
where ( is the control gain (voltage amplified ratio). Substituting distributed (averaged)
sensing expression into the above equation, one can write a feedback voltage in terms of
velocity.
(22)
M~~ = - ( d~C d 31 YP
. (It
h PS
d [ APs f APS {h 31 d'!' [ 1\1
1 Gal a [ ----nf1
u, 1
- 1\1
0u3
8Q:l ]
+ 1 f)A l [U 2 1 f)U3 ]]
1\lA2 0Ci2 IIf2 - 1\2 8Q:2
and
M~~ = - ( d~c d 32 y P
. (It
h Ps
d [[ A ps f Aps {h 31 d~s [ 1\1 a [u,
1 Gal ----nf1 - 1
1\1
0u3
8Q:l ]
+ 1 f)A l [u2
1\lA2 0Ci2 IIf2 - 1\2 8Q:2
1 f)U3
J]
+ h32 d~s [ 1\2
0u3
f) [u 2
1 7m2 1
IIf2 - 1\2 8Q:2 ]
+ 1 f)A 2 [u l 1 f)U3
] ] } A1A2 dQ:\ dQ:2]' (24)
1\lA2 0Cil ----nf1 - 1\1 8Q: 1
Note that the feedback gain ( is constant for all vibration modes. This concept can be
extended to a more general case in which a modal control gain (k is considered so that the
control effort for each mode can be different. This concept will be demonstrated in the
next control algorithm. (However, determination of modal control gains is not the topic in
this report. Therefore, it is not discussed.) The counteracting control moments can also be
rewritten in modal coordinates as discussed previously.
442 H. S. TZOU
Note that if only a single point velocity is considered in the feedback control, the
In this case, the amplitude of feedback voltage is constant and the sign is opposite
where «: is the feedback gain and "SGN" is a sin gum function, i.e.,
where (.*
1
= C d~c
II
d 3· Y.
p
Note this velocity could be a single point transverse velocity
* *
U3(t,G!1,G!2) in the feedback * * denotes a specific location on the DPS.
control, where (G!bG!2)
Thus,
(28)
DISTRIBUTED PIEZOELECTRIC TRANSDUCERS 443
And,
Note that the feedback gains are constant for all vibration modes in the above
equations. To extend the theory, an individual modal control gain Ck can be introduced
into the equations to represent different feedback effects for various vibration modes.
Thus, substituting the modal expression into the above equations yields,
(3D-a)
(3D-b)
(31-a)
(31-b)
(31--c)
QI3* = AIA2
1
[ {!1M8alIIA* 2) + 8(M I *2AI)
8a2
+ M12* BAI
002 -
M
22
* BA2 ]
001 ' (32-a)
*
~MI2A2) + * + M * BA2 M * BAI] .
Q23* = AIA2 [ fj(M 22AI)
1
8 a2 12 001 - 11002 ' (32-b)
8al
moments as discussed previously. In a special case in which the in-plane twisting effect is
neglected, i.e., MI2* = MI2 and N 12* = N 12, the transverse system equation becomes,
(33)
DISTRIBUTED PIEZOELECTRIC TRANSDUCERS 445
Note that the superscript * terms include the feedback control effects induced by the
converse piezoelectric effect. Detailed definition of Nij and Mij can be found in References
[2,3,13).
Note that in all control algorithms, the moment arm, the actuator stiffness, and
moments. In addition, the feedback gains «: can also be extended to be modal dependant -
In this section, the system dynamic equation is re-written in a state equation form.
Separating the control moments and induced piezoelectric resultant forces, moving them to
H (at,a2,t)
0[11\2 o(M~~A!)_M~~
+ 0a2 aa2 1 OA!]}+ 1 [N~~+N~~].
1\2 oa; pn lR\ lIf;- (34)
The original elastic terms associated with the DPS are defined as,
446 H. S.1Z0U
+AA [Nll+N22]}
1 2 lIfl lIf2 ' (35)
where L is a differential operator. Thus, the original equation can be simplified to:
(36)
Using state variable transformation, one can rewrite the above equation in a state equation
form,
(37-a)
or
ax
----ar-=AX+Bm, (37-b)
The system dynamic equations and the state equation can be further simplified to
account for many other geometries if four geometric parameters (A\, A 2, IRI, and 1R 2) are
given, e.g., 1) Al =1 (dx), A2 = 1 (dy), IRI = w, and 1R2 = w for a rectangular plate; 2)
Al = 1 (radial direction dr), A2 = r (angular direction dB), IRI = w, and 1R2 = w for a circular
and 1R2 = a (radius) for a cylinder (or a cylindrical panel); etc [2,3].
5. CASE STUDIES
There are three cases presented in this section. The first case demonstrates a simple
reduction procedure illustrating how to apply the generic distributed vibration sensing and
control theories to a flexible manipulator case [1]. In the second case, a prototype flexible
beam with distributed piezoelectric sensor/actuator is design and tested. The third case
parameters, one can easily reduce the original DPS equations to a variety of mechanical
systems. In the flexible robot application, it is assumed that the effective piezoelectric
Figure 3. (Note that the transverse direction is a3 or z.) Since this is a uniaxial
448 H.S.1Z0U
application of the distributed sensing theory, h32 is neglected from the equation. Besides,
the Lame's parameters for a beam structure are Al =1 and A2 = 1. The radii of
I // y
~(""______~I~I~x
Motor Piezoelectric Actuator
Flexible Manipulator
Piezoelectric Senso r
Thus, the distributed (averaged) sensor output from the distributed piezoelectric
(39)
sense the oscillation and generate output signals. In the active distributed vibration
control, this output voltage can be further processed and amplified. Then, the amplified
high voltage is injected into the distributed piezoelectric actuator layer generating a
feedback force to control the oscillation of distributed systems. The counteracting control
=-{dPcd
1
Y hPSf
31 P pS
Ps
pS (h 31 d 1 ~ )dAPS., rrU3 (40)
A A Ox
(41)
cylinders, rings, etc., one can follow the same procedure to derive the sensing/control
equations.
theories, a physical model was designed and tested in Dynamics and Systems Laboratory.
One of the simplest DPS's - a cantilever beam - was used as an example. A plexiglas
cantilever beam (0.15m x O.01m x 0.0018m) was sandwiched between two polymeric
piezoelectric layers (40 Jilll). The top piezoelectric layer serves as a distributed actuator
and the bottom a distributed sensor, Figure 3.
450 H.S. TZOV
The physical model was clamped at one end on a fixture which was directly
mounted on a shaker. A random noise signal was input into the shaker which provided a
direct excitation to the beam model. The distributed piezoelectric sensor responded to the
excitation and generated an output signal (which is a function of all distributed strains);
and this signal was phase-shifted, amplified, and then fed back into the distributed
piezoelectric actuator. The level of excitation was kept constant. Thus, the counteracting
control moments were solely controlled by a gain control on the high-voltage amplifier.
Figure 4 illustrates the laboratory setup.
I
IB~
o ~I
PC-AT + WAYEPAK
TRANSOUCER
-HI---SEHSITIVI TY
.-11--- RANGE
AMPlI FI ER
SENSOR SIGNAL
-
It"
TRA SF RMER
FEEDBACK SIGNAL
MINI SHAKER
random excitations was recorded first. Then, various feedback gains were applied to the
feedback control system and the attenuations after feedback were also recorded and
compared with the original (uncontrolled) response. Figure 5 shows a test sample of two
frequency responses of the first three modes, in which the higher represents the
uncontrolled response and the lower controlled response with a feedback gain of 1700. Note
that the frequencies were also slightly reduced due to the enhanced system damping in
feedback controls.
-
FRECI RES .. 1 121 A ..... IIIXO v 1 w-"" o vI C.Z
2111.121
I 1\
I';.
It-..
~ J,\ ............ ~
/ f"-.. I--- I /"
V\(
-elll.1II
V
.--
F""dXY " w.
FRECI RESP 1111" v IIIXD v 1 w ov lC.Z
1 Bill
~h_ • •
f'... ~
I'..
~ 1\
~
.... \.
Figure 5 shows sixty hertz line noise and its harmonics which were induced by the
feedback control electronics. This noise could be filtered out if appropriate filter was used.
Note that the distributed actuator controlled all three modes simultaneously. Table 1
summarized the distributed control effectiveness of the first three modes, five tested cases,
Table 1. Vibration control for the first three beam vibration modes (db).
o
100 -Q.265 -Q.165 -Q.361
650 -3.444 -11.164 -11.689
870 -3.766 -11.311 -12.326
1300 -4.915 -11.874 -12.647
1700 -6.694 -12.027 -13.410
It is observed that at lower feedback gains the control effectiveness was not as
significant as those at higher gains. This is because that there was a combined
electromechanical effect of the direct and converse piezoelectric effects at lower feedback
voltages. The control voltages were "neutralized" and the feedback control effects were
reduced. At high feedback voltages, however, the feedback was much larger than the
self-generated voltage; and the direct effect was suppressed. Thus, the control
that the controlled attenuation to higher modes is more effective than the lower modes.
DISTRIBUTED PIEZOELECTRIC TRANSDUCERS 453
This can be explained that the distributed actuator can effectively constrain the modeshape
variation at higher modes. It should also be noted that in the above measurements no
technique using a piezoelectric actuator [5,6]. A general theory for the piezoelectric
actuator subjected to mechanical excitations and feedback voltages is first developed.
Effectiveness of the piezoelectric micro-position attenuation is evaluated analytically and
experimentally.
piezoelectric slab to counteract the induced base excitation and to improve operation
accuracy. A prototype model was designed and tested in laboratory to validate the theory.
Figure 6 shows the prototype model with a layer of piezoelectric PVDF actuator. The
model has a 0.25"-thick steel base with a standard 10-32 stud which can be mounted on a
shaker. A 1mm thick PVDF polymer with an effective surface area of 4x10-4m 2 is
sandwiched between two 0.25" plexiglas layers which provide the same boundary conditions
to the piezoelectric actuator. The bottom plexiglas is epoxied to the steel base, and an
interchangeable metal plate is screwed onto the top plexiglas layer. A mini-accelerometer
is attached above this metal plate. Thus, the seismic mass consists of all the items above
454 H. S. TZOV
the piezoelectric actuator - the plexiglas and metal plates and the mini-accelerometer.
The vibration of this seismic mass was monitored by the mini accelerometer. The
acceleration signal was phase shifted, amplified, stepupped using a transformer and then
injected into the piezoelectric polymer to achieve active feedback position control.
11 I Acce 1 erometer
Seismic Mass
I I
Plexiglas
L
~ ~'
1
I-- Case
1 :1 I-- Piezoelectric
I-- Plexiglas
I - - S tee 1
~, Mount
Using general energy and force equilibrium concept, one can derive a general
(42)
where p is the mass density and K is a elasticity constant. It is assumed that the electric
(43)
where A is the velocity of wave propagation in the piezoelectric actuator, A = /1/ pK.
Consider the general mechanical and electric boundary conditions at fr3= 0 and h (h is the
1) fr3 = 0
T33 = T33 sinwt (44-a)
{ 0
E = Eosinwt (44-b)
2) fr3 = h
T33 = T33 sinwt (45-a)
{ 0
E = Eosinwt (45-b)
where w is the excitation frequency and E3 = Vfb/h and Vfb o is the feedback voltage.
The steady-state solution of u3 can be derived as,
(46)
where if> = (wh/A). The feedback acceleration Gfb(t) is produced by the piezoelectric
polymer resulting from the converse piezoelectric effect and the g-level can be expressed as
456 H. S.1Z0U
(47)
where g is the gravity. It is assumed that the piezoelectric actuator is used to isolate the
piezoelectricity induced force Ffb(t) resulting from the feedback voltage Vfb o (controlled
(48)
Similarly, the equivalent force Fb(t) introduced by the base excitation Gb/t) is given by
(49)
where m T is the total mass (including the piezoelectric actuator mass). The resultant
acceleration G due to the combining effects of excitations Gb and feedback Vfb can be
r o o
obtained by balancing the forces,
(50)
Substituting Eqs.(48) and (49) into Eq.(50) yields a general equation for the piezoelectric
(51)
This is a generic equation for a piezoelectric actuator with a seismic mass and it can also be
used for other piezoelectric actuators. Integrating the acceleration twice yields a general
stroke equation. In our case, we use a polymeric piezoelectric PVDF as the actuator.
(52)
the accelerometer output can be processed and fed back into the piezoelectric actuator
counteracting the oscillation and eliminating the disturbance. The active vibration
isolation due to the feedback induced converse effect can then be defined as the difference
between the resultant acceleration Gr(t) and the base excitation Gb(t). Note that it is
assumed the residual stress in the piezoelectric actuator is negligible in the active vibration
(53-a)
The gradient of the isolation surface with respect to the excitation frequency and
the feedback voltage is evaluated when the base excitation is constant (G bo = G) and
458 H. S. TZOV
feedback gain C varies (Vfbo = C V0 where V0 is the transducer output). When the base
It is found that the second term is small comparing with the first term after substituting
81R(%)
OY fb
~ [ ? d~3 :s] .w2
T
(55-a)
o
= (cons tant). w2 (55-b)
(55-c)
Cl[ w2
8IR(%)
8w - 200
-----u- d33 m S 1. Vfb
o•W
N [
= -g- ill
T
= (constant).Vfbo·w (56-a)
Eq.(55) shows that the isolation is a quadratic function of the frequency. As the
frequency increases, the isolation will increase as frequency squared. Eq.(56) shows that
The prototype model (Figure 6) was mounted on a shaker that could be excited at
various frequencies using a function generator. The seismic mass acceleration was sensed
by the miniaccelerometer. The acceleration signal was phase shifted, amplified and applied
across the piezoelectric layer in such a way that the piezoelectric vibration was 180' out of
phase with that of the base. The shaker was excited at various frequencies and amplitudes
UTC 5-46
lRAN5fOR~ER HP - 465A
CHARGE SHAKER A~PL I F I ER
UIPl lFI ER
(NOT TO SCALE I
A sample spectra of the resultant accelerations before and after feedback at 500 Hz
are given in Figures 8 and 9. (A detailed performance of the actuator will be discussed
later.) Figure 8 shows the spectrum of the undamped signal while Figure (9-a) shows the
spectrum of the attenuated signal and (9-b) for the feedback (before step-up transformer
at 100 gain) at 500 Hz. The magnitude is 0.09241 V for uncontrolled response, Figure (9),
and 0.09105 V for controlled response, Figure (9-a). It should be noted that the feedback
voltage was very high so that the signal processing equipment could not handle. The
I.
....1.......r....sa
2:7-..JU1..
,"'.,..,..
I StDDD8.
F'Rt G:
"" ;
I
'00.00
O.ONOI
.
1.1t
•a·
1.12
0.01
1.114
r,~~'- 11fm
. ' .'1 •••l re".. •
Slapo"
; '00 . 00
....
l.lZ ~~ 0 . 0110'
6
.....
.., I ,00 aoo 100
Jnlt ..... -
. .0 _~o -
nmu"" poe ... - •• ,to
700 000
"'EO: '00 . 00
MI.' : 1.'a7J7
•
i.. •
6 •
•
I
• '00
- '":...~ ,,,:'0
- 700 ...
Figure 9: Spectra of (a) a controlled response and (b) the feedback signal.
The base excitation was kept constant in this set of experiments. Four excitation
amplitudes to the shaker were chosen for each frequency . Since the base excitation is a
constant, the feedback gain is varied so that the feedback voltage injected into the
piezoelectric polymer can be controlled. The spectrum of the undamped seismic mass
acceleration signal was observed and the frequency and amplitude recorded. Then the
feedback was applied and the spectra of the feedback and attenuated seismic mass
acceleration signals were observed and the peaks recorded . In this way data for a
experiments and isolation percentage 1R(%)(exp) and 1R(%)(thy) can be calculated and then
..
.n.
Experimentally, the active vibration isolation is found to vary between 0.3% at 500
for a given frequency and a quadratic function of frequency for a given excitation voltage.
Both the experimental and analytical data plots show this tendency. Due to an equipment
limitation and the system stability, it was only tested up to 25k Hz. The performance
predictions and experimental results at a given frequency and a feedback voltage is given
by
The absolute errors between the analytical and experimental data of the resultant seismic
mass acceleration can be plotted in Figure 11. The maximum was about 7% .
•
.
....
It is observed that the theoretical data are higher than those obtained
experimentally at high frequency due to the energy dissipation at high feedback gains. The
other possible sources of the errors could be i) an interaction between the direct effect and
the converse effect in the piezoelectric PVDF actuator; ii) the feedback signal not being
exactly 180· out of phase with the seismic mass acceleration; and iii) the nonlinearity
associated with the experimental model (e.g., epoxy). Note that the theory developed and
verified is for transverse direction only. Besides, the operation should be kept within the
linear range of the mechanical system with the piezoelectric actuator. Otherwise, nonlinear
and machines, has driven for innovating new technologies and materials to design
new-generation "smart" machines and robots. This report presents a new integrated
structural identification theory based on a generic DPS - a shell element coupled with
geometrical parameters, two Lame'S parameters Ais and two radii of curvatures IRiS are
defined, the developed theories can be directly simplified to many commonly occurred
distributed structural identification theory was first derived for a generic DPS - a shell.
DISTRIBUTED PIEZOELECTRIC TRANSDUCERS 465
The derived theory shows that the DPS responses can be measured by the distributed
modal voltage equation was also derived, which shows that the modal voltage distribution
(contour) is a function of space and time. The detailed modal voltage distribution pattern
an (area) averaged output was also derived. Note that for "distributed" structural
A theory on active distributed vibration control of the DPS's was also derived using
the same DPS - a shell. The control effect was introduced and contributed by the voltage
induced piezoelectric strains (due to the converse piezoelectric effect) which result in
counteracting control moments to suppress the DPS vibrations. The control moments for
the DPS were formulated and a new set of system dynamic equations as well as state
equation were also derived. Two feedback controls, namely, 1) negative velocity and
derived.
manipulator was proposed and demonstrated. Following the same procedure, one can
apply the theories to other mechanical structures and components, e.g., plates, beams,
cylinders, etc.
controlled multi-modes. However, the modal control effectiveness was different for each
466 H. S. 120U
mode, i.e., more effective for higher modes in this case. At low feedback voltages, it
showed that the vibration control was not significant due to a combined electromechanical
effect of the direct and converse piezoelectricity. When the feedback voltage overcame the
was also studied analytically and experimentally. A general mathematical model of the
piezoelectric actuator was first formulated and the theory associated with the model was
also proposed. Theoretical solutions for the actuation and isolation control at variable
feedback gains were also derived. A prototype piezoelectric actuator made of a ,8-phase
piezoelectric polyvinylidene fluoride (PVDF) polymer was designed and evaluated. A mini
accelerometer was used to monitor the system responses; and the output signals were
processed, phase shifted, fed back, and then injected into the piezoelectric actuator to
counteract the base excitations. Effectiveness of the actuation and isolation was proved
significant. Note that this technique is primarily for micro-position feedback controls of
high-precision operations. For relatively large stroke, it would require much thicker
piezoelectric actuator. Some other factors, such as breakdown voltage, temperature effects,
noise, phase shift, time delay, etc. could also influence the control effectiveness. The
formulation can introduce zero voltage when involving symmetrical modes of DPS's.
Should this problem occur, a point reference voltage or a local averaged voltage of an
interested bounded area can be used in feedback control systems. Note that if a bounded
Determination of various modal control gains to achieve optimal controls of DPS's is also
an interesting and challenging problem which needs to be further studied and explored.
ACKNOWLEDGEMENT
This research was supported by a grant from the National Science Foundation (No.
RlI-8610671) and the Kentucky EPSCoR program, and a grant on Flexible Assembly and
Intelligent Machines (1988-1989) from the Center for Robotics and Manufacturing Systems
(CRMS) at the University of Kentucky. All my past graduate students (C.1. Tseng, S.
Pandita, and M. Gadre) who contributed to this research are also gratefully acknowledged.
REFERENCES
piezoelectric slab with constant feedback gains," Journal of Sound and Vibration,
7. A. B. Palazzolo, R.R. Lin, R.R. Kascak, and R.M., Alexander, "Active control of
structure with piezoelectric sensors and actuators," 1988 AlA A paper No.88-2418,
11. A. Baz and S. Poh, "Performance of an active control system with piezoelectric
15. J. Curie and P. Curie, Acad. Science (Paris), 91, 1880,294 and 383.
16. H.F. Tiesten, Linear Piezoelectric Plate Vibrations, Plenum Press, New York, 1969.
17. H.S. Tzou and S. Pandita, "A mUltipurpose dynamic and tactile sensor for robot
APPENDIX A: NOMENCLATURE
vector
matrix
matrix transpose
area
{E.}
1
electric field vector
shell thickness
sign function
strain vector
470 H.S. TZOU
t time
{T.}
1
stress vector
U.
1
velocity in (}i direction
ii.
1
acceleration in (}i direction
[If··]
lJ
dielectric impermeability matrix evaluated at constant strain
v differential operator
electric potential
sensor output
feedback to actuator
density
(C /PzTutrl-IEEE9UEEE91)
Mechanics
SOLID MECHANICS AND ITS APPLICATIONS
Series Editor: G.M.L. Gladwell
Aims and Scope of the Series
The fundamental questions arising in mechanics are: Why?, How?, and How much? The aim of this
series is to provide lucid accounts written by authoritative researchers giving vision and insight in
answering these questions on the subject of mechanics as it relates to solids. The scope of the series
covers the entire spectrum of solid mechanics. Thus it includes the foundation of mechanics;
variational formulations; computational mechanics; statics, kinematics and dynamics of rigid and
elastic bodies; vibrations of solids and structures; dynamical systems and chaos; the theories of
elasticity, plasticity and viscoelasticity; composite materials; rods, beams, shells and membranes;
structural control and stability; soils, rocks and geomechanics; fracture; tribology; experimental
mechanics; biomechanics and machine design.
1. R.T. Haftka, Z. Giirdal and M.P. Kamat: Elements of Structural Optimization. 2nd rev.ed.,
1990 ISBN 0-7923-0608-2
2. J.J. Kalker: Three-Dimensional Elastic Bodies in Rolling Contact. 1990
ISBN 0-7923-0712-7
3. P. Karasudhi: Foundations of Solid Mechanics. 1991 ISBN 0-7923-0772-0
4. N. Kikuchi: Computational Methods in Contact Mechanics. (forthcoming)
ISBN 0-7923-0773-9
5. Y.K. Cheung and A.Y.T. Leung: Finite Element Methods in Dynamics. 1991
ISBN 0-7923-1313-5
6. J.F. Doyle: Static and Dynamic Analysis of Structures. With an Emphasis on Mechanics and
Computer Matrix Methods. 1991 ISBN 0-7923-1124-8; Pb 0-7923-1208-2
7. 0.0. Ochoa and J.N. Reddy: Finite Element Analysis of Composite Laminates.
ISBN 0-7923-1125-6
8. M.H. Aliabadi and D.P. Rooke: Numerical Fracture Mechanics. ISBN 0-7923-1175-2
9. J. Angeles and C.S. Lopez-Cajun: Optimization of Cam Mechanisms. 1991
ISBN 0-7923-1355-0
10. D.E. Grierson, A. Franchi and P. Riva: Progress in Structural Engineering. 1991
ISBN 0-7923-1396-8
11. R.T. Haftka and Z. Giirdal: Elements of Structural Optimization. 3rd rev. and expo ed. 1992
ISBN 0-7923-1504-9; Pb 0-7923-1505-7
12. J.R. Barber: Elasticity. 1992 ISBN 0-7923-1609-6; Pb 0-7923-161O-X
13. H.S. Tzou and G.L. Anderson (eds.): Intelligent Structural Systems. 1992
ISBN 0-7923-1920-6
14. E.E. Gdoutos: Fracture Mechanics. An Introduction ISBN 0-7923-1932-X
15. J.P. Ward: Solid Mechanics. An Introduction. 1992 ISBN 0-7923-1949-4
16. M. Farshad: Design and Analysis of Shell Structures. 1992 ISBN 0-7923-1950-8
17. H. S. Tzou and T. Fukuda (eds.): Precision Sensors, Actuators and Systems. 1992
ISBN 0-7923-2015-8