Professional Documents
Culture Documents
2, 2013
Reference to this paper should be made as follows: Avci, E., Nguyen, C-N., Ohara, K., Mae, Y.
and Arai, T. (2013) ‘Analysis and suppression of residual vibration in microhand for high-speed
single-cell manipulation’, Int. J. Mechatronics and Automation, Vol. 3, No. 2, pp.110–117.
Biographical notes: Ebubekir Avci is currently a PhD student at the Department of Systems
Innovation, Graduate School of Engineering Science, Osaka University. He received his BS in
Mechatronics from Faculty of Engineering and Natural Science, Sabanci University, Turkey, in
2008, and MS in Robotics from Graduate School of Engineering Science, Osaka University,
Japan, in 2010. His current research interests include parallel mechanisms, micro-robotics and
high-speed micro-manipulation.
Kenichi Ohara received his MS in Electrical Engineering from Shibaura Institute of Technology,
Tokyo, Japan, in 2004, and his PhD from the University of Tsukuba, Tsukuba, Japan, in 2008.
Since 2007, he has been a member of Ubiquitous Function Research Group, National Institute of
Advanced Industrial Science and Technology, Tsukuba. In 2007, he joined the Graduate School
of Interdisciplinary Information Studies, University of Tokyo, as an Assistant Professor. From
2008 to 2012, he was an Assistant Professor at Osaka University. He is currently an Associate
Professor at the Department of Systems Innovation, Graduate School of Engineering Science,
Osaka University. His research interests include robot vision, wireless sensor networks, and
middleware technology for robots.
Yasushi Mae received his BS, MS, and PhD degrees in Engineering from Osaka University,
Osaka, Japan in 1993, 1995, and 1998, respectively. From 1998 to 2004, he was a Research
Associate at the Graduate School of Engineering Science, Osaka University. From 2004 to 2007,
he was an Associate Professor at the Faculty of Engineering, University of Fukui. He is currently
an Associate Professor at the Department of Systems Innovation, Graduate School of
Engineering Science, Osaka University. His research interests include robot vision, working
mobile robots and intelligent environments.
Tatsuo Arai received his BS, MS and PhD degrees from the University of Tokyo in 1975, 1977,
and 1986, respectively. He joined the Mechanical Engineering Laboratory, AIST, MITI
(nowMETI) in 1977, and was engaged in research and development of new arm design and
control, mobile robots, teleoperation, and microrobotics. He stayed at MIT as a Visiting Scientist
in 1986–1987. He moved to Osaka University, Osaka, Japan, in 1997 and since then he has been
a Full Professor at the Department of Systems Innovation, Graduate School of Engineering
Science. His current research topics are mechanism design including parallel mechanisms, legged
working robots, microrobotics for bio applications, humanoid robots and haptic interfaces.
The two end effectors of the microhand and the target object
are observed under an IX81 motorised inverted optical
microscope using an Olympus LUCPlanFLN 20x/0.45na
Ph1 objective lens. The images are captured with a
highspeed camera (Photron FASTCAM MC2) and
displayed on a Windows PC (Intel Core i7 CPU, 2.93GHz
with 4 GB RAM) monitor. The end effectors – the right and
left fingers – of this micromanipulator are glass needles
which have a 1 mm diameter and sharpened ends with a
Table 1 Specifications of motorised stages curvature of less than 1 μm. While the left end effector is
connected to the lower module, the right end effector is
Fine stage Coarse stage mounted onto upper module. In total, the microhand can
Travel (mm) 0.1 (X, Y, Z) 25 (X, Y), 10 (Z) therefore perform movements in a large workspace (rough
Resolution (μm) 0.01 1 stage) with high precision (fine stage), and grab, rotate and
Accuracy (μm) 0.1 17 release objects by the relative movement of the right end
Repeatability (μm) 0.15 2 effector (parallel link).
Max. speed (mm/s) 2.3 0.1 The maximum speed of the microhand is 2.3 mm/s. As
the feasible maximum speed of the actuators depends on the
The upper module is designed for local motion, i.e., the applied voltage and the features of the actuator itself, higher
movement of the right end effector. With local motion, the speed can be realised with other actuators.
grasping and releasing of different-size target objects can be When the stage runs at normal speed (0.1 mm/s), there is
achieved. The upper module consists of the parallel link no effective vibration for most of the microhand systems.
mechanism used for manipulating different-size objects and To grasp a cell quickly, however, the end effector of the
a manual stage on which the parallel mechanism is attached. microhand needs to move towards the cell at high speed.
Using the manual stage, the 3D position of the right end This motion involves accelerating to required operational
effector can be adjusted with respect to the left one. The speed and decelerating to a full stop. The abrupt changes in
manual stage can move 6 mm in the X, Y and Z directions, acceleration or deceleration often result in residual
with a resolution of 3 μm. The parallel link mechanism vibration. When the grasping motion is carried out at high
Analysis and suppression of residual vibration in microhand for high-speed single-cell manipulation 113
speed (≥ 1 mm/s), the residual vibration is greater which To analyse the vibration of the system, the fine stage moves
may cause dropping and even damaging a cell during with the speed of 2.3 mm/second for 80 μm. An exemplary
transportation. In addition, the settling time is longer, which plotted result of the oscillation is given in Figure 5. The
is not a desirable feature in the high-speed grasping of the vibration during the motion is small. The residual vibration
microobjects. Therefore, to realise an appropriate vibration after the motion is great however. Looking at the
control method, we carried out the vibration analysis in the magnitudes of the oscillations, we can say that the right end
microhand (Avci et al., 2012). effector oscillates with higher magnitudes (up to 29.1 μm)
than the left one (5.76 μm). Furthermore, the oscillation in
the y-direction is smaller than the one in the x-direction.
3 Vibration analysis in the microhand
Since objects are grabbed and held by the relative
3.1 Oscillation observation movement and force in the x-direction, vibration control
with the aim of removing the oscillations in the x-direction
The oscillations of the end effectors during and after the was prioritised.
motion were recorded with a high-speed camera. The
camera was set to capture images at 2,000 frames per
second with a magnification of 20× (252 × 252 μm2 image 3.2 Frequency analysis of oscillation
size and 1 pixel = 0.494 μm). With this configuration, the
oscillation could be easily observed. The magnitude and the As discrete Fourier transform (DFT) takes a discrete signal
relation to the moving direction (perpendicular or parallel to in the time domain and transforms it into its discrete
the end effectors) could be identified roughly and directly frequency domain representation, the DFT is an appropriate
from the captured images. method for frequency (spectrum) analysis. Let χ0, …, χN–1 be
For more accurate and faster handling of big image complex numbers. The DFT is defined by the formula:
batches, the end effector position is saved to be used in N −1
n
image processing. With image processing, the contours of Xk = ∑χ e n
− i 2 πk
N
k = 0, ..., N − 1. (1)
the end effectors are detected based on the difference in n=0
brightness between the background and the end effector. A fast Fourier transformation (FFT) computes the DFT and
The coordinates of the end effector tip are determined as produces exactly the same result as when evaluating the
follows. The topmost pixel of the contour corresponds to the DFT definition directly; the only difference is that an FFT is
y coordinate. Calculating the centroid of the contour and much faster.
assuming a constant angular orientation of the end effector, The datasets (obtained from image processing) were
the x coordinate can be derived, as seen in Figure 4. This analysed using the FFT. The objective of this analysis was
algorithm can be applied to the image batches saved to determine the frequencies occurring within the end
beforehand. effector oscillations. This analysis provides information
about the nature and cause of the vibration. In addition, it
Figure 4 Position detection that is showing the contour, centroid
and end effector tip (see online version for colours) shows the highest present frequency.
The FFT analysis shows that the frequency spectra of
the left and right end effectors are quite different, as shown
in Figure 6. The left end effector oscillates with but a
single – its natural – frequency (170 Hz). The right end
effector sways with the superposed frequencies. The highest
occurring frequency is the right end effector’s natural
frequency, at about 260 Hz.
4 Vibration suppression of the end effectors threshold (set to ± 1 μm in our experiments). The natural
frequency of the right end effector is 60 Hz as shown in
4.1 Structural design to increase damping Figure 9. The parallel link mechanism, moving the right end
To increase the damping of the system, the thickness and effector, has very thin and long spherical joints, which are
mass of the fingers were considered. Equation (2) shows the main cause of the vibration as shown in Figure 10.
that a heavier finger has a lower natural frequency (Fn).
Figure 9 Vibration frequency of the right end effector is 60 Hz
EI (see online version for colours)
Fn ∝ . (2)
mL4
E stands for elastic modules, while I stands for the moment
of inertia; m represents the mass of weight, and L is the
length of the end effector. Moreover, a thicker finger
oscillates with a considerably smaller magnitude (3) where
δmax represents the maximum deflection and w stands for
applied force.
wL4
δmax ∝ . (3)
EI
(a) (b)
4.2.1 Real-time image processing 1,000 Hz. The delay between giving a command on the
Linux PC and the parallel link executing the command is
To realise the vibration control in real-time, the image
about 3 ms. In total, however, the entire delay time should
processing can no longer be done on a batch of images
add up to 4.0-4.5 ms.
saved beforehand. The finger position has to be determined
directly after capturing an image. Therefore, the processing
time has to be decreased as much as possible. 4.2.3 Set-point determination
The image width was constrained by defining a region To control the vibration, the actual end effector position has
of interest (ROI). By observing the oscillation (3.1), we to be compared to a setpoint value. This is the value the
learn that the finger moves less than the maximum distance control task aims to reach. The set-point for the control task
between the two following frames. This distance depends on once the global motion has ended is identical to the targeted
the frame rate and the lens used. A 20X magnification lens position of the end effector. By converting the moving
(1 pixel = 0.494 μm) was chosen to cover a big workspace distance d into an amount of pixels and adding it to the
of 252 × 252 μm2. The ROI to be processed in the following initial position (pos0), the coordinates of the destination
frame is defined by adding this maximum distance to both (posdest) can be calculated (4).
sides of the current finger position. As a result, an image
capturing and processing rate of 2,000 Hz was realised. posdest = pos0 + d / λ
(4)
Sometimes, the finger oscillates too strongly and will where : [ λ] = μm / pixel
not be within the processed region (occurrence < 1%). In
this case, the ROI is reset to the centre of the image in order The ratio λ depends on the lens in use. In the case of the 20×
to increase the chance of re-detecting the finger in the next magnification, it is λ = 0.494.
frame.
The controller output has to be sent to the computer 4.2.4 Experimental results and discussion
commanding the parallel link mechanism. The connection
The procedure of the high-speed grasping motion is as
between the two computers used was tested for possible
follows: First, the fine stage runs with 2.3 mm/second for 80
sending rate and delay time. A sending rate of 2,000 Hz
μm to make the end effectors approach to the target object
with a transmission delay time of 0.5 ms was ascertained.
quickly. When the end effectors reach the target position,
the system informs the parallel mechanism in real-time by
4.2.2 Control loop design visual feedback, in order to suppress the residual vibration.
Considering the control aims, the deviation between the When the fine stage stops, the parallel mechanism moves
actual and the target position is processed and fed back to the end effectors to the opposite direction of the residual
the system. The difference is calculated on the Windows vibration for a decided number of cycles using the
computer doing the image processing. The result is then sent feedforward control, in order to settle the oscillation. During
to a second computer (Linux), responsible for moving the the feedforward control, there is no sensory information
parallel link mechanism, as shown in Figure 11. The two used. With the feedforward control, the residual vibration of
computers are connected via a regular LAN. The system has the right end effector is decreased. The settling time is 49
to be setup on two different computers, because the high- ms, with a maximum deflection of 15.36 μm, as seen in
speed camera only provides a Windows driver, while some Figure 12.
control tasks for the microhand require an ART-Linux To achieve stable grasping at high speed, we analyse the
computing system. vibration of the end effectors. Analysis results show that
There are two levels of vibration control, as seen in both end effectors show great oscillation at the end of the
Figure 11. The first level is the feedback control (PI) of the motion. To increase the damping of the system, we modify
parallel mechanism, meant to compensate hysteresis of the the structure of the end effectors. As a result, the amplitude
piezo actuators. However, we need to choose P and I gains and settling time of the vibration for the left end effector
very carefully so as not to produce any vibration of the decrease to the imperceptible level, at which high-speed
piezo actuators. Results show that, while compensating stable grasping is feasible. For the right end effector, the
hysteresis to avoid producing any vibration, P and I gains residual vibration is still great, however. Therefore, we
should be very small (P:0.3, I:0.1) which takes about 3 ms apply the feedforward control to the right end effector and
to complete parallel mechanism motion. Second level of achieve a further decrease in the vibration, as illustrated in
vibration control is achieved with the feedforward method. Table 2.
The feedforward control starts according to image We can predict the exact model of the vibration based
processing results. The parallel mechanism moves the right on the analysis results. Thus, we can say that the
end effector to the opposite direction of the oscillation to feedforward control is appropriate for the residual vibration
control the residual vibration. problem. If there is any unexpected disturbance, e.g.
In summary, it can be stated that an image capturing, vibrational motion of the microhand actuators on the
processing and error calculation rate of 2,000 Hz is platform, however, the resulted vibration can not be
achieved. The delay time for sending the data to the Linux suppressed with the feedforward method. Consequently, to
PC is 0.5 ms. Thus, the entire control loop can be run at realise a more robust control of the vibration, the feedback
116 E. Avci et al.
method will be applied as a next task after achieving a (29.1 μm and 5.76 μm max. deflections with ≈200 ms
sampling time of 10 times the vibration frequency. settling time) which is not appropriate for stable
manipulation at high speed. To decrease the residual
Figure 11 Block diagram of the vibration control system vibration of both end effectors, stiffness of the finger
(see online version for colours) holders is increased. Consequently, the oscillation of the left
end effector decreases to imperceptible level (1 μm with
20 ms settling time). However, the residual vibration of the
right end effector is still great (25.88 μm with 173 ms
settling time). To suppress the oscillation further, we
implement the feedforward control. As a result, the
amplitude of the vibration decreases by 47 % and the
settling time decreases by 74 %.
For this system, as we can predict the behaviour of the
vibration, the feedforward control method is suitable. To
control the vibration due to any unexpected disturbance
(vibrational motion of the microhand actuators on the
Figure 12 Vibration after compensation (see online version platform due to imperfect design), feedback method can be
for colours) applied. Thus, the combination of the feedforward and
feedback methods will be our future work.
Acknowledgements
This work was supported by Grant-in-Aid for Scientific
Research on Innovative Areas ‘Bio Assembler’ (23106005)
from the Ministry of Education, Culture, Sports, Science
and Technology of Japan.
References
Abe, A. and Komura, K. (2012) ‘Minimum energy trajectory
planning for vibration control of a flexible manipulator using
a multi-objective optimisation approach’, International
Table 2 Vibration and settling time after vibration
Journal of Mechatronics and Automation, Vol. 2, No. 4,
suppression
pp.286–294.
Left end effector Right end effector Avci, E., Nguyen, C., Ohara, K., Mae, Y., Arai, T. and Gobel, C.
(2012) ‘Vibration analysis of microhand for high speed single
Deflection Settling Deflection Settling cell manipulation’, 2012 International Conference on
(μm) (ms) (μm) (ms) Mechatronics and Automation (ICMA), pp.75–80.
Original 5.76 226 29.1 192 Chang, T. and Sun, X. (2001) ‘Analysis and control of monolithic
values piezoelectric nano-actuator’, IEEE Transactions on Control
Structural 1 20 25.88 173 Systems Technology, Vol. 9, No. 1, pp.69–75.
design Inoue, K., Nishi, D., Takubo, T., Tanikawa, T. and Arai, T. (2007)
Average 83 91 12 10 ‘Performance evaluation of teleoperation for manipulating
improvement micro objects using two-fingered micro hand’, Journal of
(%) Robotics and Mechatronics, Vol. 19, No. 5 pp.577–584.
Feedforward - - 15.36 49 Kim, N. and Hong, M. (2004) ‘Diagnosis and reduction of robot
control arm vibration for 12-inch wafer spin scrubber’, Key
Engineering Materials, Vols. 270–273, No. 1, pp.884–889.
Average - - 47 74
Lambrechts, P., Boerlage, M. and Steinbuch, M. (2004)
improvement
‘Trajectory planning and feedforward design for high
(%)
performance motion systems’, Proc. of American Control
Conference, pp.4637–4642.
Lu, Z., Moraes, C., Zhao, Y., You, L., Simmons, C.A. and Sun, Y.
5 Conclusions (2010) ‘A micromanipulation system for single cell
deposition’, in 2010 IEEE International Conference on
In this paper, we studied the residual vibration suppression Robotics and Automation, Anchorage, pp.494–499.
in a microhand to achieve high-speed grasping of McEver, M.A. and Leo, D.J. (2001) ‘Autonomous vibration
microobjects. First, we analyse the vibration behaviour of suppression using online pole-zero identification’, ASME
both end effectors. Results show that for the speed of Journal of Vibration and Acoustics, Vol. 123, No. 4,
2.3 mm/second, both end effectors’ vibrations are great pp.487–495.
Analysis and suppression of residual vibration in microhand for high-speed single-cell manipulation 117
Meckl, P.H. and Arestides, P.B. (1998) ‘Optimized s-curve motion Roy, J. and Whitcomb, L.L. (1999) ‘Comparative structrual
profiles for minimum residual vibration’, Proceedings of the analysis of 2-DOF semi-direct-drive linkages for robot arms’,
American Control Conference, pp.2627–2631. IEEE/ASME Transactions on Mechatronics, Vol. 4, No. 1,
Nakamura, M., Iwanaga, S., Arai, K., Toda, H., Capi, G. and pp.82–86.
Nikaido, T. (2011) ‘Computer-assisted biofabrication: the Saito, S., Miyazaki, H.T. and Sato, T. (2002) ‘Micro-object pick
challenges on manufacturing 3-d biological tissues for tissue and place operation under SEM based on micro-physics’,
and organ engineering’, in 2011 Symposium on VLSI Journal of Robotics and Mechatronics, Vol. 14, No. 3,
Technology, Toyama, pp.2–5. pp.227–237.
Norotte, C., Marga, F.S., Niklason, L.E. and Forgacs, G. (2009) Singh, T. and Singhose, W. (2002) ‘Input shaping/time delay
‘Scaffold-free vascular tissue engineering using bioprinting’, control of maneuvering flexible structures’, Proceedings of
Biomaterials, Vol. 30, No. 3, pp.5910–5917. the American Control Conference, pp.1717–1731.
Piazzi, A. and Visioli, A. (2000) ‘Minimum-time system inversion Singhose, W. and Singer, N. (1996) ‘Effects of input shaping on
based motion planning for residual vibration reduction’, two-dimensional trajectory following’, IEEE Transactions on
IEEE/ASME Transactions on Mechatronics, Vol. 5, No. 1, Robotics and Automation, Vol. 12, No. 6, pp.881–887.
pp.12–22.
Takeuchi, M., Nakajima, M., Kojima, M. and Fukuda, T. (2010)
Preumont, A., Francois, A., Bossens, F. and Hanieh, A.A. (2002) ‘Nanoliters discharge/suction by thermoresponsive polymer
‘Force feedback versus acceleration feedback in active actuated probe and applied for single cell manipulation’,
vibration isolation’, Journal of Sound and Vibration, Journal of Robotics and Mechatronics, Vol. 22, No. 5,
Vol. 257, No. 4, pp.605–613. pp.644–650.
Rakotondrabe, M., Clevy, C. and Lutz, P. (2008) ‘Hysteresis
Tao, W.M., Zhang, M.J., Liu, M. and Yun, X.P. (2006) ‘Residual
and vibration compensation in a nonlinear unimorph
vibration analysis and suppression for SCARA robot arm in
piezocantilever’, in 2008 IEEE/RSJ International Conference
semiconductor manufacturing’, in 2006 IEEE/RSJ
on Intelligent Robots and Systems, pp.558–563.
International Conference on Intelligent Robots and Systems,
Rakotondrabe, M., Clevy, C. and Lutz, P. (2010) ‘Complete open pp.5153–5158.
loop control of hysteretic, creeped, and oscillating
piezoelectric cantilevers’, IEEE Transactions on Automation Yang, J. et al. (2005) ‘Cell sheet engineering: Recreating tissues
Science and Engineering, Vol. 7, No. 3, pp.440–450. without biodegradable scaffolds’, Biomaterials, Vol. 26,
No. 33, pp.6415–6422.
Rakotondrabe, M., Rabenorosoa, K., Agnus, J. and Chaillet, N.
(2011) ‘Robust feedforward-feedback control of a nonlinear
and oscillating 2-DOF piezocantilever’, IEEE Transactions
on Automation Science and Engineering, Vol. 8, No. 3,
pp.506–519.