Beruflich Dokumente
Kultur Dokumente
Abstract
In this work, a novel six-component force sensor with its force-sensing member in the form of four identical T-shaped bars is presented. The
force-sensing member is subjected to finite element analysis in conjunction with a design optimization for high measurement sensitivities.
Although significant measurement couplings exist in this six-component force sensor, however, they distribute only in a few sparse places in
the calibration matrix, making the calculations for the force components relatively easy and quick. The condition number under the full rated
loading conditions for this sensor is 1.543, which represents a rather good measurement isotropy, as compared to approximately 24 for a
Maltese crossbar sensor under similar conditions. In addition, only 20 strain gauges are required in the design, which is less than that used in a
Maltese crossbar type sensor, in which at least 24 strain gauges are used.
# 2002 Elsevier Science B.V. All rights reserved.
Keywords: Six-component force sensor; Measurement isotropy; Condition number; Finite element analysis; Design optimization
0924-4247/02/$ see front matter # 2002 Elsevier Science B.V. All rights reserved.
PII: S 0 9 2 4 - 4 2 4 7 ( 0 2 ) 0 0 1 3 5 - 8
224 S.A. Liu, H.L. Tzo / Sensors and Actuators A 100 (2002) 223230
are the sensitivity of measurement, the isotropy of measure- compared to those based on a Maltese crossbar. The force-
ment, and the decoupling of measurement. For example, if sensing member is subjected to finite element analysis in
the sensor is to be equipped in a robotic wrist, which is being conjunction with a design optimization to maximize the
under a constant control to perform variable and flexible measurement sensitivities. The final results offer design
tasks, it then requires that the decoupling be one of the sizes, strains, etc., from which the various theoretical mea-
important factors. Furthermore, since there have six force surement characteristics of this six-component force sensor
components to be measured, it is generally hoped that all can be evaluated before a real sensor is made.
these components can be determined with approximately
equal measurement sensitivity. A sensor of best measure-
ment isotropy is one in which the maximum possible 2. The sensing principle of a six-component force sensor
measurement error due to matrix calculation is minimized.
In all cases, the measurement sensitivities should be max- In general, a strain-gauged force sensor utilizes the linear
imized to give high measurement accuracies. elastic behavior of structures generally made of metallic
A Maltese crossbar seems to be the earliest elastic struc- alloys. Consider a structure which is loaded (input) at a
ture developed [1] for a six-component force sensor. It particular point of it by an unknown force vector,
responds to some force components with relatively higher ~
F Fx ; Fy ; Fz ; Mx ; My ; Mz in its linear range. Then, on
stiffness than to others, ensuing a poor measurement iso- the surface of the structure there must produce (output)
tropy. In addition to this particular structural deficiency, a vector of n strain signals, ~ S S1 ; S2 ; S3 ; . . . ; Sn , in the
the measurements based on this bar present a high degree prescribed measurement locations. The superposition prin-
of coupling. For improvements on these deficiencies, it ciple of linear elasticity then provides the following linear
requires that more strain gauges be imposed in the sensing relationship:
bridges. This inevitably introduces error sources. In contrast
to the crossbar, various types of force-sensing structure had ~
S C ~
F (1)
also been proposed [214], some are modified crossbars
where [C] is an n 6 strain compliance matrix, with its
[2,4,9,12] and some are new ones [3,7,8,10,11,13,14].
entries in each column being the strains induced by a unit
Among them, Bayo and Stubbe [7] proposed a frame/truss
force for each corresponding force component in the strain
type and employed its axial deformation behavior to elim-
measurement locations. A particular entry cij then denotes
inate disadvantages of the crossbar, and obtained better
the strain contribution in the ith strain measurement location
measurement isotropy and decoupling. Quinn and Mote
due to the application of a unit magnitude of the jth force
[8] designed an uncoupled six-degree-of-freedom dynam-
component. Provided now that there are six strain measure-
ometer with a new type of force-sensing member made of six
ment locations (i.e., n 6), [C] is a 6 6 diagonal matrix
T-sections with thin webs and flanges called shear panel
and the solution for the force vector is directly obtained by a
elements. Chao and Chen [12] modified the Maltese crossbar
matrix inversion operation on Eq. (1):
by using ball bearings on the outer sides of the four spokes to
allow sliding and rotating at the rims. Although such an idea
F C 1 ~
~ S A ~
S (2)
is quite novel and attractive, it may introduce behavior
deficiencies such as nonlinearity, nonrepeatability, and hys- The matrix [A] here is the calibration matrix of the sensor,
teresis. Huang et al. [11] consulted an original design due to which represents a matrix that directly multiplies the strain
Kassi et al. [2], and proposed a design based on a light and signal vector (output) to obtain the to-be-measured force
simple thin cross beam structure. Kim et al. [13] designed a vector (input). For the case of n > 6, the direct inversion
sensing structure composed of 16 plate-beams, and claimed cannot be performed and the force vector can then be
a maximum interference errors to be below 3%. Recently, a evaluated using a pseudo-inverse technique [15] for matrix
six-axis force transducer based on the Stewart platform was operation, giving the to-be-measured force vector as
also proposed [14], and a maximum measurement accuracy
within 7.5% was claimed to have been obtained. F CT C 1 CT~
~ S (3)
This study proposes a novel six-component force sensor
with its force-sensing member in the form of four identical The constant proportional matrix [C] can be obtained, either
T-shaped bars, which is one that can react to each individual analytically using a method in structural analysis or experi-
force component with the required specific deformations. mentally using actual calibration, by applying a unit mag-
Good isotropy of measurement, large measurement sensi- nitude for each force component individually and then
tivities, and easy calculations for the force components are arranging the strains obtained in the n strain signal measure-
obtained in such a sensor. Although large cross sensitivities ment locations in the corresponding column. Since the
exist, they distribute only in a few sparse places in the components of the force vector ~ F are either forces or
calibration matrix. The condition number [4,9], which is moments, the values in the first three columns and those
an index of indicating the isotropy of measurement with in the last three columns thus are, respectively, the induced
respect to a specific rated loading condition, is quite good, as strains per unit force and per unit moment. This makes [C]
S.A. Liu, H.L. Tzo / Sensors and Actuators A 100 (2002) 223230 225
depend on the units. To eliminate such an undesirable may be evaluated according to the form of the calibration
situation, [C] may be subjected to normalization with matrix (or the strain compliance matrix). If the calibration
respect to the prescribed maximum rated forces and matrix is diagonal, then the sensor is decoupled, which
moments. That is, by applying individually the maximum means that the output of a particular strain signal corre-
rated magnitudes instead of unit magnitudes for the six force sponding to a specific force component is not affected by the
components, a normalized [C], denoting now as [Cn], can be application of any of the other force components. If this is
obtained. Such a normalization scheme compensates for the the case, then
difference in units between force and moment, and makes
the normalized [Cn] being independent of the loading. The Fi aii Si ; i 1; 6; aij 0; i 6 j (5)
ijth elements of this normalized matrix then record the strain Since the sensor body is normally monolithic, a completely
responses of the sensor under the three maximum rated diagonal calibration matrix is never obtainable in practice,
forces and the three maximum rated moments, and the and therefore various cross sensitivities exist. The cross
sum of the elements of a particular row i is the superposed sensitivities may be defined by the cross sensitivities (CS)ij
strain measured in the ith strain measurement location under coefficients [5]:
the simultaneous effect of the six maximum rated force
cij
components. CSij P6 (6)
Since a six-component force sensor is designed particu- j1 jcij j
larly for measuring the specified rated magnitudes for all
where cijs denote the elements of the above-described
components, it is hoped that the sensor itself should provide
normalized strain compliance matrix [Cn].
the same measurement sensitivities for these six components
The elements of the strain signal vector ~S are not directly
in their rated magnitudes in order to give approximately
obtainable, and therefore have to be measured through the
uniform measurement accuracies for all components. The
application of electrical circuits. The commonly used circuit
condition of such isotropy of measurement can be evaluated
is the full Wheatstone bridge, as shown in Fig. 1. It is
by an index called the condition number C0 [4,9], which is
composed of four active strain gauges bonded to the appro-
a measure of relative sensitivity and provides a means of
priate locations of the elastic sensor body. In practice,
comparison of force sensors. Such a number is defined out
gauges 1 and 3 serve to sense the tensile strain pair, whereas
of the normalized strain compliance matrix [Cn] as
gauges 2 and 4 are for the compressive strain pair. Provided
s1 that all four gauges have the same magnitude of strain e, then
C0 (4)
s6 e may be computed via [16]
where s1 and s6 denote, respectively, the maximum and the V0
e (7)
minimum nonzero singular values [4,9] of this normalized KVi
matrix. It has been shown [7] that the condition number is
Here K is the gauge factor normally ranging from 2 to 4 for
also an influencing factor that effects the transmission of the
metal-based gauges. If the strain magnitudes in the two
errors from the measured strains to forces. For the case
tensile and compressive gauge pairs are only approximately
of n 6, the singular values are the same as the characteristic
equal, then the strain evaluated by Eq. (7) can be considered
values. The value of C0 represents the degree of the measure-
as the average value. The measurement sensitivity corre-
ment isotropy for the prescribed maximum rated force set.
sponding to a bridge for a particular force component may be
The condition number depends not only on the structural
defined as the ratio of the output voltage to the input voltage
configuration of the sensor body, the locations of the mea-
under the application of the maximum rated force value.
suring strain gauges, and the total number of strain gauge
This measurement sensitivity may then be calculated as
measurements, but also on the ratio of maximum moment
V0 =Vi Ke. For a bridge composed of metal-based strain
to maximum force rather than the absolute values of the
gauges, a ratio that is greater than 1.0 mV/V is considered
maximum rated force components. When the condition
good enough, and a ratio greater than 1.5 mV/V is usually
number has a value of unity, the measurement isotropy is
difficult to achieve. A full bridge circuit has the property of
considered to be perfect. It is possible to choose a rated set
of forces to be measured that will minimize the condition
number, but unless this set corresponds to the specified set of
a particular moment to force ratio, a comparison of perfor-
mance as based on the condition numbers of various designs
is not appropriate or has no meaning. One of the objectives
in the design of a particular sensor is to pursue a value of C0
as close as possible to a value of unity under a prescribed
rated set of forces.
In addition to the measurement isotropy described above, Fig. 1. A full bridge composed of identical gauges with equal resistance
another major characteristic of a six-component force sensor changes.
226 S.A. Liu, H.L. Tzo / Sensors and Actuators A 100 (2002) 223230
particular symmetry, so it can to a great extent eliminate the implies that maximum strains are being pursued. However,
effect due to temperature change. For a six-component force the maximum normal strains produced in the member at
sensor, at least six bridge circuits are required. the locations of bonded strain gauges ought not to exceed
a maximum value as appropriate to the capability of the
strain gauges intended for use. In general, for strain gauges
3. Finite element analysis and shape optimal design commonly used, the maximum strain allowed is 500
1500 mm. In addition to such a strain constraint, the elastic
A number of design factors should be considered in the member must also not be overstressed to avoid permanent
design of a six-component force sensor. These include the damage to this member. As have been noted, one of the
specific geometry of the force-sensing member, material, major design goals is to maximize the strains in the member,
design of sensing bridges, stress and strain limits, size so that large measurement sensitivity may be obtained to
limitation, etc. In general, the design of a force sensor is ensue high measurement accuracies. Such a design goal
expected to come out with high measurement sensitivities. may best be achieved by finding the minimum volume of
Since the sensing member of a six-component force sensor is the sensing member, since lower volume generally implies
generally monolithic and geometrically complicated, its smaller sectional sizes and thus larger strains. The optimiza-
analysis as based on a pure theory of solid mechanics is tion problems thus involved may be defined as
generally difficult. To overcome this problem, the numerical Minimize Vd1 ; d2 ; d3 ; . . . ; dp (8a)
analysis as based on the commercial finite element package
can then be implemented to obtain quite accurate results and subject to
serve as one of the necessary steps before a real sensor is e eallow (8b)
made. Based on the numerical analysis, the structural beha-
vior of the sensing member can be totally realized and the sY
se (8c)
locations of appropriate strain gauge bonding can be deter- N
mined. In particular, the elastic force-sensing member under
dil di diu ; i 1; . . . ; p (8d)
the simultaneous effect of the full rated force set must be
capable of being strained sufficiently to give large strains, where V; e; eallow ; se ; sY ; N represent, respectively, the
while at the same time not subject to overstressing to cause volume designated as the objective function, the maximum
any permanent damage to this member. One of the design normal strain, the maximum strain that the strain gauge can
goals is then to maximize the strains in the member, so that tolerate before failure, the effective stress, the yield stress,
large measurement sensitivity may be obtained to ensue high and the factor of safety in stress, whereas di ; dil ; diu are the
measurement accuracies. Such a design goal may be accom- independent design variables (sensor body sizes) and their
plished by resorting to the finite element analysis technique in appropriate geometric limitations.
conjunction with the shape optimal design algorithm, which
can offer the best one among many design candidates and
has nowadays become a usual practice in mechanical design. 4. Design of a six-component force sensor
The power of optimization methods to determine the best
case without actually testing all possible cases comes 4.1. A novel elastic force-sensing member
through the use of a modest level of mathematics and at
the cost of performing iterative numerical calculations using A novel force-sensing member for a six-component force
clearly defined logical procedures implemented by the use of sensor is designed and shown in Fig. 2. Four square-sec-
computer. In order to apply the numerical techniques of tioned T-shaped bars compose the main part of the member.
optimization theory to concrete engineering problems it is A force transmitting thick plate sits on the four bars as a
necessary to clearly delineate the boundaries of the problem monolithic whole. This particular design utilizes the defor-
to be optimized, to define the quantitative criterion on which mation of the vertical and horizontal beams of the T-shaped
the best is assumed to base, to select the system variables bars. The design with undetermined sizes designated as
that characterize the problem, and to define a model where B1 ; B2 ; L1 ; L2 is shown in Fig. 3.
the variables are related. In general, three clearly differenti-
able portions are involved in a typical optimization problem. 4.2. Conditions of design
These are defining the objective function to be optimized,
identifying the independent design variables that explicitly The full rated force set for this sensor are specified to be
or implicitly form the objective function, and giving the Fx Fy 20 kgf; Fz 40 kgf;
various constraints associated with both the design variables Mx My Mz 100 kgf cm 1 kgf m 1000 kgf mm
and the structural behavior.
The measurement sensitivity of a strain-gauged force which correspond to a medium-size six-component force
sensor under full rated force set depends mainly on the sensor. The sensor body is assumed to be made of aluminum
amount of strains produced in the elastic member. This alloy 7075T6. The factor of safety in stress design based on
S.A. Liu, H.L. Tzo / Sensors and Actuators A 100 (2002) 223230 227
the yield stress is taken to be N 2, which is quite appro- The finite element model is one that uses solid elements
priate. The modulus of elasticity, the Poissons ratio, and the of eight nodes, and is bisected and trisected, respectively, in
yield stress required in the analysis are, respectively, the height and width of the sections of the T-shaped bars. To
analyze this force-sensing member, the full rated capacity of
E 7:342 kgf=mm2 ; n 0:33;
each individual force component is separately applied to the
2
sY 51:5 kgf=mm model. Based on the final results of the design optimization
as stated above, the deformed shapes can be plotted to
4.3. Design optimization provide demonstration of the specific deformations corre-
sponding to each individual force component, and also to
Considering the capability of the strain gauges, the max- provide a comprehension of the interaction of different force
imum allowable normal strain is assumed to be 800 mm components.
(0.0008). Also, taking into account the size requirement for Fig. 5 shows the exaggerated deformed shape of the finite
element model subjected to the force component Fx or Fy.
From the figure, it is clear that for the vertical beams, in
addition to the usual bending, tension and compression also
occur. As for the horizontal beams, two of them are in bending
and the other two are in torsion. The exaggerated deformed
shape for the force component Fx is shown in Fig. 6, where it is
obvious that the vertical beams are in tension, whereas the
horizontal ones are in bending. Fig. 7 is the exaggerated
deformed shape under Mx or My. It appears that two T-shaped
bars are severely strained, whereas the other two are not.
Finally, the exaggerated deformed shape for Mz is shown in
Fig. 8. For this case, both the vertical beams and the horizontal
beams of the T-shaped bars are in a condition of bending.
The normal strains from the results of the finite element
analysis associated with the last iteration of the design
optimization are listed in Table 1. Note here that the values
listed have been taken as the averages of the four nodes on
the surfaces of the second elements from the ends, and the
elements are about 1.5 mm in length. From the values in the
table, the measurement characteristics can be evaluated.
Table 1
Normal strains in the locations of strain measurement gauges
Gauge Fx Fy Fz Mx My Mz
4 5 5 4 4
2 4.843 10 4.431 10 9.221 10 1.335 10 1.26 10 5.617 10 4
1 4.843 10 4 4.431 10 5 9.221 10 5 1.335 10 4 1.26 10 4 5.617 10 4
6 4.843 10 4 4.431 10 5 9.221 10 5 1.335 10 4 1.26 10 4 5.617 10 4
5 4.843 10 4 4.431 10 5 9.221 10 5 1.335 10 4 1.26 10 4 5.617 10 4
4 4.431 10 5 4.843 10 4 9.221 10 5 1.26 10 4 1.335 10 4 5.617 10 4
3 4.431 10 5 4.843 10 4 9.221 10 5 1.26 10 4 1.335 10 4 5.617 10 4
8 4.431 10 5 4.843 10 4 9.221 10 5 1.26 10 4 1.335 10 4 5.617 10 4
7 4.431 10 5 4.843 10 4 9.221 10 5 1.26 10 4 1.335 10 4 5.617 10 4
11 8.944 10 5 2.279 10 4 4.609 10 4 6.599 10 4 3.619 10 5 7.981 10 5
12 2.489 10 4 1.577 10 4 5.026 10 4 3.963 10 5 7.194 10 4 1.52 10 4
18 2.489 10 4 1.577 10 4 5.026 10 4 3.963 10 5 7.194 10 4 1.52 10 4
17 8.944 10 5 2.279 10 4 4.609 10 4 6.599 10 4 3.619 10 5 7.981 10 5
9 3.191 10 4 2.402 10 4 4.848 10 4 6.957 10 4 8.384 10 5 3.087 10 4
10 3.191 10 4 2.402 10 4 4.848 10 4 6.957 10 4 8.384 10 5 3.087 10 4
15 3.568 10 4 2.402 10 4 4.848 10 4 6.957 10 4 8.384 10 5 3.087 10 4
16 3.191 10 4 2.402 10 4 4.848 10 4 6.957 10 4 8.384 10 5 3.087 10 4
13 2402. 10 4 3.191 10 4 4.848 10 4 8.384 10 5 6.957 10 4 3.087 10 4
19 2.402 10 4 3.191 10 4 4.848 10 4 8.384 10 5 6.957 10 4 3.087 10 4
14 2402. 10 4 3.191 10 4 4.848 10 4 8.384 10 5 6.957 10 4 3.087 10 4
20 2.402 10 4 3.191 10 4 4.848 10 4 8.384 10 5 6.957 10 4 3.087 10 4
locations of the gauges and are numbered, respectively, from 1 obtained from the strain results of applying each individual
to 20 (the numbers in parentheses represent those that are loca- force component in its full rated magnitude separately. The
ted in the other sides). The six bridges for the six force com- entries of [Cn] in each column can be evaluated as the
ponents are each composed of four active gauges. These are averages of the strains in the measurement locations of
Fx : 2; 5; 1; 6; Fy : 4; 7; 3; 8; the strain gauge bonding:
Fz : 12; 18; 11; 17; Mx : 9; 10; 15; 16; c1j 14 e2 e5 e1 e6 j ; j 1; 6 (9a)
My : 13; 14; 19; 20; Mz : 1; 5; 4; 8 c2j 14 e4 e7 e3 e8 j ; j 1; 6 (9b)
The compositions of these bridges are based on the specific
deformations that occur under each individual force com-
ponent, as may be seen in Figs. 58.
Fig. 4. Bonding locations and numbering for the strain gauges. Fig. 6. Exaggerated deformed shape under the force Fz.
S.A. Liu, H.L. Tzo / Sensors and Actuators A 100 (2002) 223230 229
Table 2 References
Measurement sensitivities of the novel force sensor
Gauge factor K 2:1 V0/Vi (mV/V) [1] V.D. Scheiman, A preliminary work on implementing a manipulator
force sensing, Wrist, AI Laboratory Report, Stanford University,
Measurement sensitivities 1971.
Fx, Fy 1.12 [2] M. Kassi, K. Takeyasu, M. Uno, K. Murakaoka, Trainable assembly
Fz 1.15 system with an active sensory table possessing six axes, in:
Mx, My 1.57 Proceedings of the 11th International Industrial Robots, Tokyo,
Mz 1.36 Japan, 1981, pp. 393404.
[3] H. Van Brussel, H. Berlien, H. Thielemans, Force sensing for
advanced robot control, Robotics 2 (1986) 139148.
of the output voltages to the input voltages, can be calculated
[4] M. Uchiyama, Y. Nakamura, K. Hakomori, Evaluation of robot force
accordingly. These are listed in Table 2. sensor structure using singular value decomposition, J. Robot Soc.
Jpn. 5 (1) (1987) 410, (in Japanese).
4.9. Remarks [5] M. Uchiyama, K. Hakomori, A few considerations on structure
design of force sensors, in: Proceedings of the Third Annual
From the above expressions, it is seen that, strain signals Conference of Robotics Society of Japan, 1985, pp. 1718.
[6] Y. Nakamura, et al., Design and signal processing of six-axis force
designed for sensing Fx and My interact each other, i.e., Fx sensors, in: Proceedings of the Fourth International Symposium on
sensor influences My sensor, and vice versa. Similarly, strain Robotics Research, MIT Press, Cambridge, MA, 1988, pp. 7581.
signals designed for sensing Fy and Mx also interact each [7] E. Bayo, J. Stubbe, Six axis force sensor evaluation and a new type of
other in the same manner. It is quite particular that Mz sensor optimal frame/truss design for robotics applications, J. Robotic Syst.
influences Fz sensor, but Fz sensor does not influence Mz 6 (2) (1989) 191208.
[8] T.P. Quinn, C.D. Mote Jr., Optimal design of an uncoupled six degree
sensor. For this six-component force sensor, only Fz sensor is of freedom dynamometer, Exp. Mech. 30 (1990) 4048.
completely independent of the other sensors. [9] M. Uchiyama, E. Bayo, E. Palma-Villalon, A systematic design
procedure to minimize a performance index for robot force sensors, J.
Dyn. Syst. Measur. Contr. 113 (1991) 388394.
5. Conclusion [10] A. Bicchi, A criterion for optimal design of multi-axis force sensors,
Robotics Auton. Syst. 10 (1993) 169286.
[11] W. Huang, H. Jiang, H. Zhou, Mechanical analysis of a novel six-
This study has endeavored in designing a novel six- degree-of-freedom wrist force sensor, Sens. Actuat. 35 (1993) 203
component force sensor, and has obtained good results in 208.
terms of the measurement isotropy and measurement sensi- [12] L.P. Chao, K.T. Chen, Shape optimal design and force sensitivity
tivity. By performing the design optimization in conjunction evaluation of six-axis force sensors, Sens. Actuat. A 63 (1997) 105
112.
with the finite element analysis, the optimal sizes of the
[13] G.S. Kim, D.-I. Kang, S.-H. Rhee, Design and fabrication of a
force-sensing member were determined. six-component force/moment sensor, Sens. Actuat. 77 (1999) 209
Although high degrees of coupling exist in such a six- 220.
component force sensor, however, they distribute sparsely [14] C.G. Kang, Closed-form force sensing of a 6-axis force transducer
only in a few places in the calibration matrix, making the based on the Stewart platform, Sens. Actuat. A 90 (2001) 3137.
[15] R.D. Cook, W.C. Young, Advanced Mechanics of Materials, Wiley,
calculations for the force components reasonably easy and
New York, 1989.
quick. The condition number for this sensor is 1.543, which [16] D.K. Alexander, Transducers and their Elements, Prentice-Hall,
represents a rather good measurement isotropy, as compared Englewood Cliffs, NJ, 1994.
to approximately 24 for a Maltese crossbar type sensor [17] COSMOS/M 2.0, Users Manuals, 1997.
under a similar loading condition. In addition, only 20 strain
gauges are used, less than that used in a Maltese crossbar
type sensor, in which at least 24 strain gauges must be used. Biographies
The results presented in this study are based purely on a
Sheng A. Liu received his PhD degree in mechanical and aerospace
theoretical study. For a real sensor that is made according to
engineering of State University of New York at Buffalo in 1990. Since
the design, the measurement characteristics should be close then, he has been working as an Associate Professor in the Department of
to what have been predicted. However, due to various error Mechanical Engineering at Dayeh University. His research interest
sources such as strain gauge variation, bonding inaccuracy, includes design of force sensors and pressure transducers, computer-aided
material in-homogeneity, etc., the sensor must be subjected engineering analysis, and design optimization.
to characteristic test using suitable calibration apparatus
Hung L. Tzo is a former graduate student in the Department of Mechanical
of known uncertainty in order to obtain the actual calibra- Engineering at Dayeh University, and received his MEng in 2001. He has
tion matrix and also to evaluate the overall measurement been working as an Associate Engineer in the engineering division of
characteristics. Taiwan Railway Administration.