Professional Documents
Culture Documents
Mechatronics
journal homepage: www.elsevier.com/locate/mechatronics
a r t i c l e i n f o a b s t r a c t
Keywords: Manipulation of small components and assembly of microsystems require force measurement. In the
Micro-force measurement microworld (the world of very small components), signal/noise ratio is very low due to the weak ampli-
Measurement noise
tude of the signals. To be used in feedback control or in a micromanipulation system, a force sensor must
Kalman lter
allow static and dynamic measurements. In this paper, we present a micro-force measurement system
Piezoelectric cantilevers
Strain gauges based on the use of strain gauges and a Kalman optimal lter. Using a model of the measurement system
and a statistical description of the noise, the optimal lter allows ltering the noise without loss of
dynamic measurement. The performances of the measurement system are improved and fast force vari-
ations can be measured.
2008 Elsevier Ltd. All rights reserved.
0957-4158/$ - see front matter 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.mechatronics.2008.11.012
458 Y. Haddab et al. / Mechatronics 19 (2009) 457462
The designed system allows us to use both parts as actuator: the First, a calibration of the cantilever is done. Calibrated weights
linear motor for high stroke/high resolution displacements and the are used to apply forces at the tip of the cantilever and the dis-
piezoelectric cantilever to perform fast displacements. However in placement is measured by a high resolution (10 nm) laser sensor
our study, we use the cantilever equipped with the strain gauges from KEYENCE. The results show a good linearity and a compli-
only as a force sensor. Moreover we will not study the control of ance of 1.23 lm/mN.
the linear motor.
3.2. Dynamic characterization
namic behavior of the cantilever, k1 and k2 are the static gains be- 1.2
tween F and d, and V and d, respectively.
From the previous static characterization, we can deduce that
1
k1 1:23 103 m=N: 2
The gain k2 can be easily identied by applying voltages and 0.8
measuring the deection using the laser sensor
Vout(V)
In order to determine G(s), a voltage step is applied and the step re-
sponse is recorded using the laser sensor (see Fig. 4). 0.4
4. Force measurement using strain gauges Fig. 6. Measured voltage Vout in response to a voltage step of 15 V.
In order to measure the force, the two strain gauges glued sym-
metrically on the cantilever are included in a Wheatstone bridge.
higher signal/noise ratio. In order to obtain a good signal, the stan-
This conguration results in better thermal compensation and
dard bonding process (specied by the manufacturer) must be
strictly respected. Two other resistors are used to complete the
bridge, an amplier is used to increase the voltage level and the
signal is ltered using a low-pass lter (see Fig. 5). The 4 V supply
x 10
-5
voltage has been chosen according to the characteristics of the
2.5 gauges. The amplier has a gain of 51.3 and the cutoff frequency
of the lter is 3100 Hz which is more than four times greater than
2
the resonant frequency of the cantilever.
As the dynamic behavior is similar for a force input and a volt-
age input, in order to characterize the measurement system, we
1.5 have chosen to use a voltage step because it is easier to apply.
Fig. 6 shows the measured voltage in response to a voltage step
delta(m)
5. Kalman ltering
delta (m)
0 1 0 1
0 1 0 0 0 0
B 0 0 1 0 C B0C
B C B C
AB C; B B C; -0.5
@ 0 0 0 1 A @0A
0:04258 0:70475 2:50860 2:75595 1 -1
C 0 3:6139 108 6:60332 108 3:47543 108 :
-1.5
6
If we consider the measurement and the process noises, the sys- -2
tem can be represented by the following linear stochastic -0.02 0 0.02 0.04 0.06 0.08 0.1
time (s)
equation:
Fig. 8. Measurement noise.
X k1 A X k B V k wk
7
dk C X k v k
where w and v represent the process and measurement noises, 700
respectively.
Assuming that the noises are independent from each other,
600
white and with normal probability distributions, a discrete Kalman
lter provides an efcient recursive method to estimate the state of
the process (and therefore the output) in a way that minimizes the 500
number of samples
The process noise is mainly due to the input V (noise produced -50
by the voltage generator).
The process and measurement noises are independent from
each other because they are generated by different devices. -100
-150
magnitude (dB)
-200
Vk Process k
-250
Kalman X k
filter C k
-300
Fig. 7. Estimation of the process state and output using a Kalman lter. Fig. 10. Spectrum of the measurement noise.
Y. Haddab et al. / Mechatronics 19 (2009) 457462 461
-6 -6
x 10 x 10
20
1.5
15
1
measured
10 0.5
delta (m)
delta (m)
0
5
-0.5
0 estimated
-1
-5
-0.02 0 0.02 0.04 0.06 0.08 0.1 -11 -10 -9 -8 -7 -6 -5 -4 -3 -2
time (s) time (s) -3
x 10
Fig. 11. Measured deection d in response to a voltage step of 15 V. Fig. 13. Comparison between the measured deection d and the estimated
deection ^
d in static mode.
-6
x 10
20 x 10
-5
1.55
15 1.5
estimated
1.45
10
delta (m)
delta (m)
1.4
5
1.35
1.3
0
1.25 measured
-5
-0.02 0 0.02 0.04 0.06 0.08 0.1
time (s) 0.04 0.042 0.044 0.046 0.048 0.05
time (s)
Fig. 12. Estimated deection ^
d using Kalman ltering in response to a voltage step
of 15 V. Fig. 14. Comparison between the measured deection d and the estimated
deection ^
d in dynamic mode.
5.2.2. Process noise In this section we present the results of the implementation of
The process noise is assumed to be mainly due to the input the Kalman Filter. In the algorithm, the process noise covariance
noise. (covariance of wk) and the measurement noise covariance (covari-
ance of vk) are calculated from noise characteristics.
wk B v nk : 8
Figs. 11 and 12 show the measured deection and the estimated
In the previous equation, vnk is the input noise. deection using Kalman ltering (respectively) in response to a
vnk is measured at the input when a zero input is applied. voltage step of 15 V. These results show that Kalman ltering
(zero input means that the generator is connected to the system allow obtaining a signicant decrease of measurement noise (see
and therefore the noise produced is applied at the input of the sys- Fig. 13) without loss of dynamic behaviour measurement (see
tem). The same analysis was performed to characterize the input Fig. 14).
462 Y. Haddab et al. / Mechatronics 19 (2009) 457462
From Fig. 13, we can see that the measurement noise amplitude References
is decreased 40 times and then force measurement can be affected
by a maximum error of 50 lN. [1] Arai F, Nakajima M, Dong L, Fukuda T. The pico-Newton order force
measurement with a calibrated carbon nanotube probe. In: Proceedings of
In the current results, we have considered the voltage input and the 2003 IEEE/ASME international conference on advanced intelligent
the noise produced by the voltage generator. Considering the force mechatronics, Kobe, Japan; 2003. p. 6916.
input will lead to similar approach as the behaviour of the cantile- [2] Dao DV, Toriyama T, Wells J, Sugiyama Susuma. Silicon piezoresistive six-
degree of freedom forcemoment microsensor. Sens Mater
ver is similar. However, exact performance of the Kalman ltering 2003;15(N3):11335.
will depend on the force noise level. This noise depends on the ob- [3] Fung CKM, Elhajj I, Li WJ, Xi N. A 2-D PVDF force sensing system for micro-
ject in contact with the cantilever tip. In a practical case, the force manipulation and micro-assembly. In: Proceedings of the 2002 IEEE
international conference on robotics and automation, Washington, DC, USA;
noise should be evaluated prior to real-time Kalman ltering. Mea-
2002. p. 148994.
surement and process noises characterisation can also be periodi- [4] Haddab Y, Chaillet N, Bourjault A. A microgripper using smart piezoelectric
cally updated for best performances. These noises can vary if the actuators. In: Proceedings of the 2000 IEEE/RSJ international conference on
working conditions are different or if the gripper is used in differ- intelligent robots and systems. Takamatsu, Japan; 2000.
[5] Kalman RE. A new approach to linear ltering and prediction problems.
ent environments. transactions of the ASME. J Basic Eng 1960:3545.
[6] Katupitiya J, Dutr S, Demay S De Geeter J. Estimating contact and grasping
uncertainties using kalman lters in force controlled assembly. In: Proceedings
6. Conclusion and future work of the 1996 IEEE/RSJ international conference on intelligent robots and
systems, Osaka, Japan; 1996. p. 696703.
[7] Kemper M. Development of a tactile low-cost microgripper with integrated
Force measurement is very important for the success of manip- force sensor. In: Proceedings of the IEEE international conference on control
ulation and assembly of small components and devices. In this pa- applications, Taipei, Taiwan; 2004. p. 14616.
per, a measurement system based on the use of strain gauges has [8] Kim D, Kim B, Park J. Implementation of a piezoresistive MEMS cantilever for
nanoscale force measurement in micro/nano robotic applications. KSME Int J
been implemented. Strain gauges are low-cost, easy to integrate 2004;18(N5):78997.
sensors for force measurement. [9] Liu JJ, Ma CK, Kung IC, Lin DC. Input force estimation of a cantilever plate by
Because of the weak amplitude of the forces in the microworld, using a system identication technique computer methods in applied
mechanics and engineering, vol. 190. Elsevier; 2000. p. 130922.
signal/noise ratio is low. Moreover, closed loop control of microma- [10] Lu Z, Chen PCY, Lin W. Force sensing and control in micromanipulation. IEEE
nipulators requires dynamic force measurement. A Kalman optimal Trans Syst Man Cybern Part C 2006;36(6):71324.
lter has been implemented in order to lter the measurement [11] Ma CK, Chang JM, Lin DC. Input forces estimation of beam structures by an
inverse method. J Sound Vib 2003;259(2):387407.
noise and allow measurement of dynamic forces. The use of a [12] Maybeck PS. Stochastic models, estimation and control, vol. 1. Academic Press,
Kalman lter allows signicant performance improvement. Inc.; 1979.
In the future, the cantilever will be used in a microgripper and [13] Nasir M, Dickinson M, Liepmann D. Measurement of insect ight forces using a
MEMS based physical sensor. In: Proceedings of Hilton Head 2006, Solid State
the strain gauges and the Kalman lter will be used to estimate
Sensors, Actuators and Microsystems Workshop. Hilton Head Island; 2006.
the force applied on the manipulated micro-objects. Various manip- [14] Nilsson B, Nygards J, Larsson U, Wernersson A. Control of exible mobile
ulation and transportation tasks will be performed in different envi- manipulators: positioning and vibration reduction using an eye-in-hand range
ronmental conditions and the results compared. Moreover, the camera. Control Eng Pract 1999;7:74151.
[15] Woods RJ, Fearing RS. Flight force measurements for a micromechanical ying
ability to use the piezoelectric cantilever as actuator and as force insect. In: Proceedings of the IEEE/RSJ international conference on intelligent
sensor opens new perspectives for integrated force/position control. robots and systems, vol. 1, Hawaii, USA; 2001. p. 35562.