Engineering, 2010, 2, 1-11
doi:10.4236/eng.2010.21001 Published Online January 2010 (http://www.scirp.org/journal/eng/).
1
Using Microgripper in Development of Automatic Adhesive
Glue Transferring and Binding Microassembly System
R. J. CHANG1, C. C. CHEN2
1Department of Mechanical Engineering, National Cheng Kung University, Tainan, Taiwan, China
2Micro System Technology Center, Industrial Technology Research Institute, Tainan, Taiwan, China
Received August 13, 2009; revised September12, 2009; accepted September 20, 2009
Abstract
A system using microgripper for gluing and adhesive bonding in automatic microassembly was designed,
implemented, and tested. The development of system is guided by axiomatic design principle. With a com-
pliant PU microgripper, regional-edge-statistics (RES) algorithm, and PD controller, a visual-servoing system
was implemented for gripping micro object, gluing adhesive, and operating adhesive bonding. The RES al-
gorithm estimated and tracked a gripper’s centroid to implement a visual-servoing control in the microas-
sembly operation. The main specifications of the system are: gripping range of 60~80μm, working space of
7mm×5.74mm×15mm, system bandwidth of 15Hz. In the performance test, a copper rod with diameter 60μm
was automatically gripped and transported for transferring glue and bonding. The 60μm copper rod was dipped
into a glue container and moved, pressed and bonding to a copper rod of 380μm. The amount of binding glue
was estimated about 5.7nl.
Keywords: Micro Gripper, Adhesive Bonding, Microassembly, Visual Servo
1. Introduction
In the manufacturing cycle of microsystems, assembly
is a crucial operation in production [1]. The assembly
tasks are varied according to the areas and methods of
microproducts in manufacturing. There are several
methods such as anodic bonding, eutectic bonding,
welding bonding and adhesive bonding which can be
utilized for micro joining [1,2]. Anodic bonding, eutectic
bonding, and welding bonding usually are achieved under
some critical conditions, such as high temperature and
pressure. In comparing with other methods, adhesive
bonding has the advantages of free heating, cleanness,
speediness and no influence on parts.
In the application of adhesive bonding, a traditional
method is to use a glue dispenser [3] for the injection of a
micro drop on the binding surface of a component. Then,
another component is applied and pressed for bonding. By
using a dispenser, it is very difficult to apply small amount
of glue on a specific small surface of a micro component
[3]. The limitation of injection of dispenser and opera-
tional working space makes the technique of micro join-
ing most severe in the production of Microelectrome-
chanical System (MEMS). Instead of using a dispenser,
an effective technique for micro adhesive bonding is to
transfer glue by holding and dipping a slender tool into a
glue container and applying it to the surface for bonding
[2]. However, a direct approach by utilizing a manipulator
to grasp and hold a component instead of using a tool can
be applied for transferring glue in the micro adhesive
bonding. By utilizing a manipulator for the adhesive
bonding operation, it is most effective to hold and dip the
surface of a component into a glue container and to move
the component touching, pressing, and bonding to another
component.
Recently, the automation of microassembly has be-
come an important technology which attracts many in-
vestigators [4–10]. Although there are several different
designs in the microassembly system, a systematic engi-
neering approach from conceptual design to realization is
still not proposed. Actually, in the development of an
automatic microassembly system, the stringent engi-
neering requirements such as small size, high reliability,
clean operation, fast response, and high accuracy usually
make a conventional engineering scheme ineffective in
the process from design to manufacturing. Systematic
tools in realization concerning the relationship between
design and manufacturing are most useful to assure ef-

R. J. CHANG ET AL. 2
fective development of the systems [11,12].
In realizing an automatic microassembly system, a
visual control through microscope has been a challenging
task. Actually, visual servo has been a viable method in
industry for assembly and material handling jobs. Since
the early contribution of Shirai and Inoue, considerable
efforts have been devoted to develop visual control ma-
nipulators in manufacturing [13]. There are many visual
estimation algorithms developed for tracking control [14].
These algorithms mainly include optical flow method,
sum of square difference (SSD) method, model-based
method, motion energy method, and template matching
method. Although template matching method is widely
employed in automatic microassembly system [6–9],
different algorithms usually have their advantages and
disadvantages in different applications. The issues of
robustness, resolution, and efficiency have been identified
for further improvement of algorithms in the visual
tracking applications [15].
In the present research, a visual-servoing automatic
system with microgripper for adhesive bonding is de-
veloped and tested. By employing the precision design
axioms on system design, a visual-servoing automatic
microassembly system is developed to achieve the re-
quirements of adhesive bonding in micro manipulations.
For achieving the operation of precise and accurate grip-
ping and transportation of micro objects in assembly task,
a compliant microgripper actuated by piezoelectric ac-
tuator is designed and fabricated. In order to provide an
efficient, robust and accurate estimation for the auto-
mation of microassembly operation, an algorithm of
regional edge statistics (RES) is developed and imple-
mented. For implementing a closed-loop control, the
system model will be identified and the controller is
synthesized. The performance of a prototype microas-
sembly system is tested. Finally, the research on the de-
velopment of the visual-servoing microassembly system
is concluded.
2. Preliminary Design Consideration
The major steps in system development consist of
conceptual design, preliminary design, and detail de-
sign [11]. Conceptual design forms the fundamental
step in the whole development process and a physical
realization should capture the essence of the conceptual
design. Because of the lack of detailed information in
the early development process, a prescriptive model to
aid decision making is most useful in system devel-
opment [16]. For the effective development of a micro-
assembly system under the stringent engineering con-
straints such as small size, high reliability, clean op-
eration, fast response, and high accuracy, a precision
design method guided by axiomatic design principle is
employed to aid decision making in the system devel-
opment [12].
2.1. Design Objective and Constraints
In the area of microassembly, there are numerous func-
tional principles which can be applied for the manipula-
tion of micro objects [1,17]. In the present design, the
objective is to develop an automatic microassembly sys-
tem utilizing mechanical micro gripper to handle bonding
of micro parts in the clean room of MEMS industries. For
the microassembly operation, a reliable automatic opera-
tion of picking, gluing, attaching, and binding is required.
No lubrication and no wear are essential constraints for
the clean room operations. The objects or tools to be
picked up, glued, and transported for assembly are slender
hard components with diameter or width around 60~80μm.
The bonding of micro parts is to be operated in room
temperature around 25 ℃. Under the constraints of lim-
ited working space, micro fabrication technology, geo-
metrical configuration of gripper, and the size of micro
object, the planar size of gripper mechanism with width
about 500μm and length about 700μm is required.
2.2. Design Principle
For the development of a microassembly system, the
optimal design of the microassembly system is to achieve
two design axioms in the mappings from Functional Re-
quirements (FRs) to Design Parameters (DPs) and from
DPs to Process Variables (PVs). The two design axioms
are stated as follows [12,18]:
Axiom 1 (Functional Independence). An optimal de-
sign of a micro assembly system must maintain the in-
dependence of functional requirements of subsystems.
Axiom 2 (Information Minimization). The best design
of a micro assembly system is a design of functionally
independent subsystems with minimum information con-
tent. Here, the information content is a measure of un-
certainties in physical realization of the design specifica-
tions of a system and subsystems.
The microassembly system is to be realized under the
guidance of precision design axioms. By employing the
axiom of Functional Independence, a micro assembly
system can be designed and realized with the merits of
independent module design, independent functional test-
ing, and degrees-of-freedom in controller implementation.
With the axiom of Information Minimization in the
processes of design, assembly, and manufacturing, a mi-
cro assembly system can be effectively realized to satisfy
the stringent requirements in assembly operations.
2.3. Conceptual Design
For realizing an adhesive-bonding operation by using a
micro gripper, a conceptual design on the microassembly
system is described. The configuration of the microm
assembly system is designed by following the axiomatic
design principle. The highest FRs of the microassembly

R. J. CHANG ET AL. 3
system is identified as independent functions: FR1=
Gripping and releasing of object, FR2= Carrying gripper
and components in assembly operation, and FR3= Ac-
quiring working states during bonding process. The cor-
responding DPs of the microassembly system is identified
as DP1= Micro Gripper System, DP2= Working Stages,
and DP3= Visual System. The functional mapping be-
tween the FRs and DPs in the first level can be formulated
as (1):
00
FR1 DP1
11
FR20DP2
2122
FR3DP3
313233
a
aa
aaa
=









(1)
where the matrix by [a
ij] is to be characterized in the
physical realization of the microassembly system. The
mapping between FRs and DPs in (1) satisfies a decoup-
led module design. In addition, the micro gripper system
can be uniquely designed without considering the effects
of other modules. By following the axiom of Functional
Independence, the functional mapping between the FRs
and DPs can be realized by utilizing sensor, actuator,
mechanism, and controller. The decomposition of FRs is
to be unique and independent and will be used as the basis
for formulating DPs which are corresponding to FRs,
respectively. In the decomposition and mapping, the
branches of FRs are mapped into physical domain to
develop lower levels of DPs as:
(1) Micro Gripper System
FR1-1: Gripper opening to fit object size
FR1-2: Gripper closing and releasing
FR1-3: Afford gripping force
DP1-1: Gripper Mechanism
DP1-2: Gripper Controller
DP1-3: Gripper Actuator
(2) Working Stages
FR2-1: Carry object or tool for transferring glue
FR2-2: Moving gripper
FR2-3: Carry glue container and assembly part for
bonding
DP2-1: Object/Tool Stage
DP2-2: Gripper Carrier Stage
DP2-3: Glue Container and Assembly Part Stage
DP2-4: Controller
(3) Visual System
FR3-1: Monitor and control gripper working states
FR3-2: Automatic control the motion of gripper stage
FR3-3: Acquire image for tracking
DP3-1: User Interface
DP3-2: Tracking Algorithm
DP3-3: CCD Subsystem
A system structure of the microassembly system to satisfy
the FRs by synthesizing the corresponding DPs is de-
Figure 1. System structure.
picted as Figure 1. From Figure 1, the subsystem consists
of three modules as Micro Gripper System, Working
Stages, and Visual System.
2.4. Design Procedure and Constraints on Design
Parameters
For the development of a microassembly system, the
design procedure in general can be obtained from (1). The
mapping between FRs and DPs in (1) can be decoupled
and realized starting from a11. From the mapping by a11,
the Micro Gripper System will be first realized. With the
Micro Gripper System and a21, the a22 is obtained by
realizing Working Stages to satisfy the functional re-
quirement. With the mapping set up by a11, a21 , a22 , a31,
and a32, the a
33 is obtained finally by realizing Visual
System to satisfy the functional requirement. As a result,
the design procedure is given by
FR1 → DP1 → FR2 → DP2 → FR3 → DP3
By following above design procedure, the design con-
straints in implementing DPs will be analysed. The DP1
as Micro Gripper System is constructed by Gripper
Mechanism, Gripper Controller and Gripper Actuator. In
the design and realization of a Micro Gripper System, the
selections of material, manufacturing processes, system
configuration, and components assembly are crossly re-
lated. From the requirement of design objective, the con-
straints of DPs in the design of Micro Gripper System can
be identified as clean environment, precision dimension,
micron operation, and gripper size. In the process domain,
the PVs are stated as material selection, fabrication
method, configuration, and assembly works. The func-
tional mapping between the constraints and PVs can be
formulated as (2):
Clean environmentMaterial selection
Precision dimensionFabrication method
Micron operationConfiguration
Gripper sizeAssembly works
Bij




=





(2)

R. J. CHANG ET AL. 4
The Bij in (2) in general is not independent and con-
straints in (2) are to be considered in the physical realiza-
tion of the Micro Gripper System.
2.5. Design Features by Employing Axiomatic
Design
The design axioms are employed first for the design of a
Micro Gripper System. For the micron scale of a Gripper
Mechanism to be operated in a clean room, a one-piece
compliant gripper, without assembly works, is selected
and designed to provide accurate tip motion and gripping
force transmission. The design of the Gripper Mechanism
satisfies the axiom of Information Minimization. In the
realization of assembling Gripper Mechanism and Grip-
per Actuator, physical uncertainties are to be minimized
by the axiom of Information Minimization. A packed
piezoelectric actuator with very rigid steel case is selected
since the case can be employed as a structural frame to
align and install Gripper Mechanism. The design feature
of independent Gripper Mechanism and Gripper Actuator
satisfies the axiom of Functional Independence.
For the design consideration of Working Stages, the
precision of Working Stages for microassembly is re-
quired to be one micron since the working range of grip-
per belongs to micron scale. The selection of three inde-
pendent stages in industrial use satisfies the axioms of
Functional Independence and Information Minimization.
A Visual System is composed of User Interface,
Tracking Algorithm, and CCD Subsystem. The CCD
Subsystem consists of CCD, Illumination Light, Micro-
scopic Lenses, and Image Processing Card. The selection
of Visual System satisfies the design axioms of Func-
tional Independence and Information Minimization.
In the final consideration of control hardware and
software, a friendly software tool is to be selected to im-
plement a User Interface on Display for system monitor-
ing and control. For the consideration of real-time and
precise processing of image signal and piezoelectric con-
trol signal, a functional independent PC with Digital
Signal Processor (DSP), and interface cards are selected.
3. Gluing-Adhesive Microassembly System
Based on the conceptual design, a system structure of
image-based automatic gluing-adhesive system to realize
the DPs is shown in Figure 1. The detail design of micro
gripper system, working stages, and visual system is de-
scribed in the following sections.
3.1. Micro Gripper System
Micro Gripper System consists of Gripper Mechanism
(DP1-1), Gripper Actuator (DP1-3), and Gripper Con-
troller (DP1-2). At first, the design of Micro Gripper
system will be considered to satisfy FR1-1 to FR1-3. The
Micro Gripper System is expected to grip a slender object
with diameter or width around 60~80μm. For the object
with slenderness ratio, characteristic length to width ratio,
above 2 and in dry condition, the micro sticking force will
not be an issue in object releasing operation. Therefore,
the design and fabrication of the Micro Gripper System is
expected to provide the operating functions of precise and
accurate gripping of micro objects.
Compliant polymer micro grippers have been designed
to provide accurate tip motion and sufficient gripping
force in a clean room operation [18–20]. The selection of
a lumped-compliant gripper mechanism is to minimize
the effects of creeping, relaxation, and hysteresis loop due
to polymer material [18]. A lumped-compliant gripper
mechanism is composed of pseudo linkages and compli-
ant joints. By following the axiom of Information Mini-
mization, the topological structure of the Gripper mecha-
nism is designed to be symmetric with minimum number
of compliant joints. The micro compliant gripper mecha-
nism and its associated PLM (Pseudo linkages model) [20]
is designed as shown in Figure 2. In Figure 2, the contour
line shows a geometrical shape of the Gripper Mechanism
and its structural frame. The PLM is modeled as a
six-linkage mechanism for providing one degree of
freedom in input-output motion. When an actuating force
F is applied, the actuator will drive link 4 to produce
displacement. Due to the constraints of structure frame 1,
both link 3 and 5 will rotate and translate to cause the
gripping operation by tips C and C’.
The assumption of small deformation is used in the
following derivation. From Figure 2, the horizontal and
vertical displacement gains can be derived as the ratio
between the tip displacement
x
∆
,
y
∆
and input displace-
ment
i
∆
of slider 4, respectively as
2
1
cos
∆
==
∆
x
Lx
GiL
β
(3)
Figure 2. PLM of kinematic relation of microgripper.

R. J. CHANG ET AL. 5
and
2
1
sin
∆
==
∆
y
Ly
GiL
β
(4)
where 1
LAB
=, 2
LAC
= andβis the angle between a
vertical line and
AC
. The ratio between the horizontal
and vertical displacements is
cot
x
y
β
∆=
∆ (5)
From (3) to (5), there are two important parameters
x
G
and β to be designed. For achieving the function of
stable operation, the horizontal displacement is required
to be much larger than the vertical displacement in grip-
ping. Therefore, it is expected to have smaller β from (5).
For the compliant gripper, the lateral stiffness of gripper
arms against pay load needs to be taken into consideration.
It is essential when a gripper is fabricated by using elastic
polymer material. The lateral stiffness depends on the
material and geometry of a gripper. The ratio of L 2 to L1 in
(3) and (4) gives a measure of the lateral stiffness for a
fixed thickness in gripper. The ratio is smaller for higher
lateral stiffness.
The formulation from input force F to output horizontal
displacement can be derived by employing the conserva-
tion of energy. In closing operation, ideally, the input
work is equal to the strain energy stored by the compliant
joints to give
222
11
111
2()
222
sAB
Fikskk×∆=×∆+∆+∆θ
θ
(6)
Since
1
∆
∆=θ
i
L
and 11
sin()
sLi
∆=∆=∆
θ, then
222
1
11
22
11 2
1
111
2(()()())
222
()
sAB
sAB
ii
Fikikk
LL
Lkkk i
L
∆∆
×∆=××∆+×+×
++
=∆
(7)
Equation (7) can be expressed to give
eq
FKi
=∆
(8)
where
eq
K
is the equivalent stiffness as
2
11 2
1
++
=
sAB
eq
Lkkk
KL (9)
The optimal shape and size of the microgripper is syn-
thesized based on the following considerations. The size
of gripping object is of primary consideration. The
maximum size of outlines and the minimum size of the
compliant joints are constrained by the capability of mi-
cro fabrication. The design objective is to trade off the
parallel gripping and maximum horizontal displacement
Figure 3. Mask designed for fabricating microgripper (Unit:
mm).
gain under the geometrical constraints. For gripping a
hard object with size 60μm, the present design is to trade
off the lateral stiffness and horizontal gripping to
give
4
x
G
=
, and
21
β=
o
. The selection of thickness of the
Gripper Mechanism is constrained by the fabrication
material, micro fabrication techniques, and the lateral
stiffness of the gripper. After synthesizing the geometri-
cal size of the gripper, the shape and size of the opening
of gripping surfaces can be further modified. For helping
the gripping object fed into the opening of gripping sur-
faces and maintaining parallel gripping, an opening with
2° in slope and round tips are designed. With the con-
straint of the geometrical size 500μm×700μm, the final
geometrical shape of the Gripper Mechanism is designed
and to be fabricated to give one tenth of a mask as shown
in Figure 3.
The detail design of a Gripper Mechanism based on the
axiomatic design principle satisfies the FRs of the con-
ceptual design. By employing axiomatic design as a tool
of robust design in considering the relationship between
design and manufacturing, the stringent functional re-
quirements of a micro gripper mechanism is to be realized
effectively. The design performance of the Gripper Me-
chanism will be further analysed. The Gripper Mecha-
nism is to be fabricated by using a Polyurethane (PU) film
of thickness 0.2mm. The Young’s modulus of the PU is
E=7.775×107N/m. By employing finite element analysis
through Ansys with planar linear elastic model, the
stress-strain behavior of the gripper is analysed. From the
results by Ansys and through the equivalent model by
PLM, the horizontal gain is obtained as 3.85 and the
equivalent stiffness is 62.2N/m. The error of horizontal
gain and equivalent stiffness between the Ansys results
and PLM design is 3.75% and 5%, respectively. These

R. J. CHANG ET AL. 6
results support the validity of utilizing PLM in design.
The fabrication of the micro Gripper System is con-
sidered by following the axiom of Information Minimi-
zation. The PU micro Gripper Mechanism is fabricated by
employing a mask projection method instead of a direct
write method. By employing an excimer laser, Exitech
2000, the fabricated micro Gripper Mechanism through
mask is reduced by the optical lens of 10x. The fabricated
Gripper Mechanism with Gripper Actuator is assembled
through a metal coupler by utilizing simple hand tools and
with adhesive glue. The Gripper Actuator with Gripper
Controller can provide 10μm linear displacement. With
much higher stiffness of Gripper Actuator to drive the
lower stiffness of Gripper Mechanism, the gripper as-
sembly can provide sufficient gripping force in the grip-
ping range 60μm to 80μm by input 0 to 80 volts from the
Gripper Controller.
3.2. Working Stages
Working Stages include Object/Tool Stage (DP2-1),
Gripper Carrier Stage (DP2-2), and Glue Container and
Assembly Part Stage (DP2-3). By following the guidance
of design axioms, these stages are implemented by se-
lecting industrial products to satisfy FR2-1 to FR2-3. Two
three-axes stages for Object/Tool Stage and Glue Con-
tainer and Assembly Part Stage are obtained from New-
port. The three-axes stages are driven by stepping motors
and can provide 1.3μm resolution with 15mm stroke.
Gripper Carrier Stage obtained from Physik Instrumente
is implemented to give 0.5μm resolution with 15mm
stroke for planar X-Y motion.
3.3. Visual System
Visual System consists of User Interface (DP3-1),
Tracking Algorithm (DP3-2), and CCD Subsystem
(DP3-3). By following the guidance of Information
Minimization in implementing an image acquisition sys-
tem for visual servo, the installation of Visual System is to
adopt Endpoint Closed Loop (ECL) scheme and the
structure of visual servo is to adopt the Dynamic Im-
age-Based Look-and-Move (DIBLM) scheme.
For image acquisition to satisfy FR3-3, hardware is set
up to include JAI CVS3200 CCD with lens, TI vDB im-
age grabber card, and TI DSP320C6711 DSK board. The
imaging model adopts affine projection. In this case, a
point with coordinates as c
P
, expressed with respect to
the Cartesian coordinate frame of a camera, will project
onto an image plane as
[
]
y
T
x by
c
x
y

=+


APc
(10)
where A and c are calibrated 2x3 and 2x1 matrix, respec-
tively. The model is purely linear. The A and c are easily
computed by using linear regression techniques. Since
affine projection does not correspond to any specific
imaging situation, the issue of camera calibration is
greatly simplified. The affine projection can be simpli-
fied to a scaled orthographic projection if A is reduced to
a scalar and c is zero.
The DIBLM scheme provides several advantages. First,
the system with DIBLM structure doesn ’t need coordinate
transformation and consequently, the control performance
is independent of the calibration. Second, it is more effi-
cient in computation. Thus, the computational load on
control structure is not stringent in implementation. Third,
it is suitable for the present realization of a planar vis-
ual-servoing microassembly operation.
The design axioms provide an efficient and effective
tool to help logical reasoning and decision making in
obtaining an optimal design of the microassembly system.
The system framework of hardware installation is shown
in Figure 4. The image-based automatic operation of
gluing-adhesive bonding is to implement control software
on a PC and through a motion control card, NI-3744 and
communication interface, RS-232. LabView is selected to
implement control windows for DP3-1. The selection of
LabView for implementing control windows to satisfy
FR3-1 is based on the guidance of Minimum Information.
For the microassembly system, all the assembly works are
operated under the field of view of a visual system by
DP3-1 to DP3-3. The detail designs of DP3-1 and DP3-3
were described in this section. A tracking algorithm to
realize DP3-2 will be developed in next section.
4. Tracking Algorithm
The performance of an algorithm which tracks an object
from the features of its image plays a key role to imple-
ment a visual-servoing system. In general, edges are most
essential features of an object employed in image tracking.
If the shape of an object has clear edge features in the
image, the geometric features such as length, width, area,
and centroid, depending on the representative of these
edge features, can be fully or partially determined to
construct a feature space of the object. An object with
symmetrical geometry in image is a typical object that the
edge is representative of its centroid. For estimating the
geometric centroid of an object from the edge features of
the image, it is required that the image is under conditions:
1) The edge of an object is representative of its cen-
troid.
2) There is no high frequency spatial noise in back-
ground.
By employing the concepts of interesting region and
edge feature, a dynamic image tracking algorithm, Re-
gional Edge Statistics (RES), is developed. In the initial
time instant k, the initial center of a tracking region was
defined as the centroid of the micro gripper and repre-

R. J. CHANG ET AL. 7
Figure 4. System framework of hardware installation.
sented as (Xc(k),Yc(k)) in RES algorithm. The height and
the width of the tracking region are determined by the
height Lo and width Wo of the microgripper, the maximum
velocities (Vv, Vh) of the microgripper, and the sampling
period T in moving one step. The height L and the width
W of the tracking region are given as (11) and (12), re-
spectively,
ov
LLVT
=+ (11)
oh
WWVT
=+ (12)
The limits of boundary of the tracking region in height
are max
()
Xk
and min
()
Xk
as shown in (13) and (14),
respectively,
max
()()(/2)
c
XkXkL=+ (13)
min
()()(/2)
c
XkXkL=− (14)
The limits of boundary of the tracking region in width
are max
()
Yk
and min
()
Yk
as shown in (15) and (16), re-
spectively,
max
()()(/2)
c
YkYkW=+ (15)
min
()()(/2)
c
YkYkW=− (16)
The present algorithm utilizes a 2-D mask to extract
the edge features of the target on an image. For a 2-D
mask with height
2
a
and width
2
b
, the image response
(,,)
Gxyk
with a mask
(,)
Wst
on an original image
(,,)
Fxyk
can be expressed as
(,,)(,)(,,)
ab
satb
GxykWstFxsytk
=−=−
=++
∑∑ (17)
The horizontal and vertical Sobel masks
((,),
v
Sobel
Wst
(,))
h
Sobel
Wst
with given appropriate threshold values
()
vh
SobelSobel
VV are utilized, respectively, to extract the
horizontal and vertical edge features of the microgripper.
The vertical edge features extracted is given by (18) or
(19),
minmaxmin
max
()|(,,),
()()(),
()()
v
vSobel
v
ekGxykV
EkXkxXkY
kyYk
>


=<<


<<

(18)
minmaxminmax
()()|(,)(,,),
()(),()()
vx
ab
vvSobelSobel
satb
EkekWstFxsytkV
XkxXkYkyYk
=−=−

=++>



<<<<


∑∑
(19)
The horizontal edge features extracted is given by (20)
as
minmaxminmax
()()|(,)(,,),
()(),()()
hy
ab
hhSobelSobel
satb
EkekWstFxsytkV
XkxXkYkyYk
=−=−

=++>



<<<< 

∑∑
(20)
From (19) and (20), the set of edge features of the mi-
crogripper is extracted to give
()
Ek
as
{
}
()()|()()()
vh
EkekekEkEk
=∈
U
(21)
The RES algorithm finally utilizes the set of edge fea-
tures to estimate the centroid of an object
((),())
oo
XkYk
as given by the following two equations,
1
1
01
()()
j
i
N
Me
eh
vj
ivh
vh
x
x
Xkcc
ccMN
=
=
=+
+
∑
∑ (22)

R. J. CHANG ET AL. 8
1
1
1
()()
j
i
N
Me
eh
vj
i
ovh
vh
y
y
Ykcc
ccMN
=
=
=+
+
∑
∑ (23)
In (22) and (23), the
(,)
ii
ee
vv
xy
, and
(,)
jj
ee
hh
xy
are the
ith and jth image coordinates of elements in the vertical
and horizontal edge features, respectively. The
v
C
,
h
C
which are between 0 and 1 are given as the statistical
weighting coefficients of vertical and horizontal features,
respectively. The numbers of elements of vertical and
horizontal edge features, respectively, as M and N are
counted in the procedure of edge extraction. When the
object is moving one step, the center of tracking region
in next time instant k+1 is replaced by the centroid which
was estimated in time instant k as (24) and (25),
(1)()
+=
co
XkXk
(24)
(1)()
+=
co
YkYk
(25)
The application of RES algorithm for tracking the mo-
tion of a microgripper is illustrated. For the present mi-
croassembly system, the microgripper for RES algorithm
satisfies two conditions described above. The images of a
microgripper under RES are shown in Figure 5. Figure
5(a) is an original image of the microgripper. In Figure
5(b), the gray region is shown as a tracking region. The
horizontal edge features are extracted with
0
v
C
=
as
shown in white. In Figure 5(c), the bright white spot is
the centroid extracted. Figure 5 reveals that the centroid
of the microgripper as a tracking target can be estimated
accurately by employing the RES algorithm.
The RES method is compared with other methods.
Since a template matching method has good performance
in robustness and resolution, it is selected for the com-
parisons. The simulated results of RES, Gray-Scale Cor-
relation (GSC), Normalized Gray-Scale Correlation
(NGSC), and Normalized Gray-Scale Correlation with
Image Pyramid (NGSC-IP) methods are shown in Table.
1. From Table 1, it is observed that the RES method can
provide high resolution as those of GSC and NGSC.
However, the computational efficiency is much im-
proved compared with GSC related methods.
A template matching method utilized GSC related
methods to find out a position which gives maximum cor-
relation between a template and image. The position is then
(a) Original image (b) Edge features (c) Centroid extracted
Figure 5. Images of microgripper under RES.
Figure 6. Block diagram with ideal Smith predictor.
Table 1. Comparison of RES and template matching related
methods.
Template
size
(pixel)
Resolution
(pixel)
Computation
load
(second)
Tracking
area
(pixel)
RES 51x45 1 0.01 11x11
GSC 51x45 1 0.03 11x11
NGSC 51x45 1 0.06 11x11
NGSC-IP 51x45 2 0.11 31x31
used to track any target through the known template. If
the selected template has no apparent feature or template
has noisy feature on the image, the GSC related methods
may fail to find out the correct position of the selected
template. However, the RES algorithm can statistically
collocate weighting coefficients to improve the robust-
ness and computational load. The positioning accuracy is
improved by the high resolution in image tracking. Actu-
ally, if a tracking region always covers the target on an
image, the feature-based RES algorithm is expected to
track the centroid of the target fast and precisely.
5. System Modeling and Controller Design
A visual-based automatic microassembly system utiliz-
ing the architecture of DIBLM is implemented. The pre-
sent system utilizes the RES algorithm to estimate and
track the microgripper for assembly task. The size of the
tracking region by the RES algorithm is 51x45 pixels.
The modeling and controller design for the system will
be described.
5.1. System Modeling
In the DIBLM architecture, the stage controller, PI
C-843.20, and the sub-micron stage, M1-111.DG, are
considered as a plant. Two axes of the sub-micron stage
are orthogonal and can be modeled independently. The
image controller is denoted as
()
Cs
. In the preliminary
step-response tests of both axes with
()3
Cs
=
, the re-
sponses showed dead-time delays and overshoots. By
employing an ideal model of Smith predictor [21], the
system architecture is simplified as shown in Figure 6.
For the X-axis, the step-response test with
()3
x
Cs
=

R. J. CHANG ET AL. 9
Figure 7. Normalized step response of X-axis servo.
Figure 8. Normalized step response of X-axis servo without
employing PD and then with PD controller after 40 seconds.
is shown in Figure 7 by a solid line. The overshoot is
about 7.68% and the delay time is 0.695 seconds. The
peak time is about 3.125 seconds. By assuming the
transfer function of a plant as ()
x
sT
x
Gse
− and utilizing a
second-order model to approach the step response, one
derives damping ratio
0.633
ς
=
and natural fre-
quency
1.299
n
ω
=
. The closed-loop transfer function of
X-axis is presented as ()
x
sT
x
Tse
− by (26),
0.692
2
1.687
() 1.6441.687
x
sT
s
x
Tsee
ss
− −
=
++ (26)
The plant is derived to give ()
x
sT
x
Gse
− as (27),
0.692
0.562
() (1.644)
x
sT
s
x
Gsee
ss
−−
=+ (27)
A closed-loop step response of the model is compared
with that of experimental test as shown in Figure 7. From
Figure 7, the simulated result does not fit very well with
the experimental one. A higher order model may be util-
ized to improve the modelling error. However, the iden-
tified model fits experimental result accurately in delay
time, peak time, and overshoot. These specifications are
actually concerned in the present gluing and assembly
operations.
By following the same procedure as the X-axis for
modeling and testing of the Y-axis in the sub-micron
stage, the closed-loop transfer function is obtained to
give
0.81
2
4.071
() 2.3814.071
y
sT
s
y
Tsee
ss
−−
=
++ (28)
The plant is derived to give
0.81
1.018
() (2.381)
y
sT
s
y
Gsee
ss
−−
=+ (29)
5.2. Controller Design
In the present microassembly operation, the task is to use
a microgripper gripping an object, gluing adhesive, and
bonding with another object automatically. Organic glue
which can be hardened in room temperature around 25℃
is utilized for the assembly task. For the assembly opera-
tion, the desired control performance is described in the
following:
1) The overshoot in response is minimized since it will
cause collision to make the assembly fail.
2) Response must be fast enough to avoid the glue
hardening.
In order to reduce the overshoot, a PD controller is se-
lected. For the servo in the X-axis, the closed-loop trans-
fer function with PD controller becomes
0.692
2
0.562()
() (1.6440.562)0.562
x
sT
s
xx
x
xx
DsP
Tsee
sDsP
− −
+
=
+++
(30)
The PD controller is finally designed as
()
x
Cs
=
41.21
s
+ to minimize the overshoot and increase the
response speed as shown in Figure 8. For the Y-axis
servo, the PD controller is designed as
()1.90.367
y
Css
=+
to satisfy the control objective.
The stability robustness due to modelling error
G
∆
needs to be investigated. Since the plants of both axes are
type 1, the gain margins should be infinite if the time
delays are zero. The phase margins of the servo system
due to delay time in both axes are analyzed. If both axes
have no delay time, these X-axis and Y-axis servos can
provide phase margins of 78.5 and 80 degrees, respec-
tively. Therefore, for the two axes of a servo system op-
erated with natural frequency of 1.5rad/second, the sys-
tem is stable with sufficient margin.
By the present servo design on both axes, there is no
overshoot in response and the settling time is less than 5
seconds. The PD controller satisfies the control objective
and can be implemented for the microassembly task.

R. J. CHANG ET AL. 10
6. Gluing and Adhesive Bonding Tests
Automatic gluing and adhesive bonding test by the micro
assembly system is to implement an automatic operation
for a microgripper gripping a copper rod with diameter
60μm, gluing adhesive, and bonding to another copper rod
with diameter 380 μm. The copper rod grasped by micro-
gripper can be considered as a glue transfer tool or a cy-
lindrical component. The whole system except PC and
controllers are enclosed in an acrylic case to maintain a
constant temperature about 25 ℃. In the following per-
formance test, an automatic operation takes seven steps to
complete a visual-servoing microassembly task.
First, the system will estimate both centroids of the
microgripper and the copper rod with diameter 60μm.
Also, the system will define the centroid of the micro-
gripper as the origin of system. Second, the system will
drive the microgripper to grip the copper rod and go back
to the origin as defined in the first step. Third, the system
estimates the centroid of glue in a container and the tip
position of another copper rod with diameter 380 μm.
Fourth, the system estimates the pose and the tip position
of the copper rod which has been gripped by the micro-
gripper. The pose of the copper rod, which was held by
the microgripper, provides information for the system to
avoid collision. A copper rod gripped by the microgrip-
per is shown in the left of Figure 9. Fifth, the system
initiates RES algorithm to estimate the centroid of the
microgripper with gripping copper rod. The microgripper
gripping the copper rod is to glue adhesive for three
times. The copper rod gluing adhesive on the tip is
shown in the middle of Figure 9. Sixth, the microgripper
gripping the 60μm copper rod is bonding to a 380μm
copper rod. The microgripper stays and waits for ten
minutes for glue hardening. Finally, the microgripper
releases the 60μm copper rod as shown in the right of
Figure 9 and goes back to the origin as defined before.
In the present system, the steady-state error in visual
tracking is less than 1 pixel. The bandwidths of system in
X-axis and Y-axis are about 16Hz, and 19Hz, respectively.
The microassembly system takes about 3 minutes to
finish one operation and 10 minutes to wait for glue
hardening in room temperature. The volume of glue in
assembly can be estimated from the image of Figure 9.
Figure 9. Microassembly procedure from gripping (left),
gluing (middle), to bonding and releasing (right).
Since the image resolution is 22μm/ pixel and the thick-
ness of glue in system is about 1 pixel, the thickness of
glue is estimated as 22μm. Therefore, the volume of
binding glue is about 5.7nl in the microassembly opera-
tion.
7. Conclusions
A PC-based visual-servoing automatic microassembly
system was developed and tested. By employing the pre-
cision design axioms, a visual-servoing automatic mi-
croassembly system is realized efficiently and effectively
to achieve the requirements of adhesive bonding in micro
manipulations. The micro adhesive bonding by a PU
microgripper is highly reliable without failure in the
compliant joints during more than a hundred of opera-
tional tests. The system utilized the RES algorithm to
track the microgripper and achieve the task of adhesive
bonding. The present RES algorithm can accurately track
and estimate the microgripper in a real-time operation.
The steady-state error in visual tracking is less than 1
pixel. The system bandwidth is about 15 Hz. The per-
formance was tested for gripping a copper rod with 60μm
in diameter, gluing adhesive, and bonding to another
copper rod with 380μm in diameter. The volume of
binding glue in the microassembly operation is about
5.7nl.The system took about 3 minutes to finish one as-
sembly operation if the waiting time for glue hardening
was ignored.
8. Acknowledgments
The authors would like to thank the NSC for the support
under contract No. (95)-2221-E-006-157.
9. References
[1] H. V. Brussel, J. Peirs, D. Reynaerts, A. Delchambre, G.
Reinhart, N. Roth, M. Weck, and E. Zussman, “Assembly
of microsystems,” Annals of the CIRP, Vol. 49, pp.
451–472, 2000.
[2] T. Tanikawa, Y. Hashimoto, and T. Arai, “Micro drops for
adhesive bonding of micro assemblies and making a 3-D
structure `Micro scarecrow',” IEEE International Confer-
ence on Intelligent Robots and Systems, Vol. 2, pp.
776–781, 1998.
[3] J. Bergkvist, T. Lilliehorn, J. Nilsson, S. Johansson,
and T. Laurell, “Miniaturized flowthrough microdis-
penser with piezoceramic tripod actuation,” Journal of
Microelectromechanical Systems, Vol. 14, No. 1, pp.
134–140, 2005.
[4] S. Fatikow, J. Seyfried, A. Buerkle, and F. Schmoeckel,
“A Flexible microrobot-based microassembly station,”
Journal of Intelligent and Robotic Systems, Vol. 27, pp.
135–169, 2000.
[5] P. Korhonen, Q. Zhou, J. Laitinen, and S. Sjovall, “Auto-

R. J. CHANG ET AL. 11
matic dextrous handling of micro components using a 6
DOF microgripper,” Proceedings of 2005 IEEE Interna-
tional Symposium on Computational Intelligence in Ro-
botics and Automation, pp. 125–131, 2005.
[6] A. Ferreira, C. Cassier, and S. Hirai, “Automatic mi-
croassembly system assisted by vision servoing and virtual
reality,” IEEE/ASME Transactions on Mechatronics,
Vol. 9, No. 2, pp. 321–333, 2004.
[7] K. B. Yesin and B. J. Nelson, “A CAD model based
tracking system for visually guided microassembly,”
Robotica, Vol. 23, pp. 409–418, 2005.
[8] Y. H. Anis, J. K. Mills, and W. L. Cleghorn, “Automated
microassembly task execution using vision-based feed-
back control,” IEEE Proceedings of the 2007 International
Conference on Information Acquisition, Vol. 9, No. 2, pp.
321–333, 2007.
[9] L. Ren, L. Wang, J. K. Mills, and D. Sun, “Vision-based
2-D automatic micrograsping using coarse-to-fine grasp-
ing strategy,” IEEE Transactions on Industrial Electronics,
Vol. 55, No. 9, pp. 3324–3331, 2008.
[10] G. Yang, J. A. Gaines, and B. J. Nelson, “Optomecha-
tronic design of microassembly systems for manufacturing
hybrid microsystems,” IEEE Transactions on Industrial
Electronics, Vol. 52, No. 4, pp. 1013–1023, 2005.
[11] B. S. Blanchard, “System Engineering Management,”
New York, Wiley, 1991.
[12] N. P. Suh, “The Principles of Design,” New York, Oxford
University Press, 1990.
[13] S. Hutchinson, G. D. Hager, and P. I. Corke, “A tutorial on
visual servo control,” IEEE Transactions on Robotics and
Automation, Vol. 12, No. 5, pp. 651–670, 1996.
[14] E. Trucco and K. Plakas, “Video tracking: a concise sur-
vey,” IEEE Journal of Oceanic Engineering, Vol. 31, No.
2, pp. 520–529, 2006.
[15] G. Zhu, Q. Zeng, and C. Wang, “Efficient edge-based
object tracking,” Pattern Recognition, Vol. 39, pp. 2223 –
2226, 2006.
[16] D. Braha and O. Maimon, “The design process: properties,
paradigms, and structure, ” IEEE Transactions on Systems,
Man, and Cybernetics, Vol. 27, No. 2, pp. 146–166, 1997.
[17] S. Fatikow and U. Rembold, “Microsystem technology
and microrobotics,” Berlin, Springer, 1997.
[18] R. J. Chang, H. S. Wang, and Y. L. Wang, “Development
of mesoscopic polymer gripper system guided by preci-
sion design axioms,” Precision Engineering, Vol. 27, pp.
362–369, 2003.
[19] R. J. Chang, P. W. Shih, R. Z. Huang, and K. L. Lin,
“Application of piezo-driven polymer microgripper in
automatic transportation of micro object, ” IEEE Interna-
tional Conference of Mechatronics, pp. 140–144, 2005.
[20] R. J. Chang and Y. L. Wang, “Integration method for
input-output modeling and error analysis of four-bar
polymer compliant micromachines, ” ASME Journal of
Mechanical Design, Vol. 121, pp. 220–228, 1999.
[21] O. J. Smith, “A controller to overcome dead time, ” ISA
Journal, Vol. 6, pp. 28– 33, 1959.