Professional Documents
Culture Documents
art ic l e i nf o
a b s t r a c t
Article history:
Received 14 November 2014
Received in revised form
9 March 2015
Accepted 12 March 2015
Available online 26 March 2015
The paper deals with geometric calibration of industrial robots and focuses on reduction of the measurement noise impact by means of proper selection of the manipulator congurations in calibration
experiments. Particular attention is paid to the enhancement of measurement and optimization techniques employed in geometric parameter identication. The developed method implements a complete
and irreducible geometric model for serial manipulator, which takes into account different sources of
errors (link lengths, joint offsets, etc). In contrast to other works, a new industry-oriented performance
measure is proposed for optimal measurement conguration selection that improves the existing
techniques via using the direct measurement data only. This new approach is aimed at nding the calibration congurations that ensure the best robot positioning accuracy after geometric error compensation. Experimental study of heavy industrial robot KUKA KR-270 illustrates the benets of the developed pose strategy technique and the corresponding accuracy improvement.
& 2015 Elsevier Ltd. All rights reserved.
Keywords:
Serial industrial robots
Geometric calibration
Enhanced partial pose measurement
Design of experiments
Industry-oriented performance measure
1. Introduction
In robotic literature, the problem of geometric calibration is
already well studied and has been in the focus of the research
community for many years [18]. As reported by a number of
authors, the manipulator geometric errors are responsible for
about 90% of the total positioning error [9]. Besides of the errors in
link lengths and joint offsets, the end-effector positioning errors
can be also caused by the non-perfect assembling of different links
and arise in shifting and/or rotation of the frames associated with
different elements, which are normally assumed to be matched
and aligned [10]. It is clear that the geometric errors do not vary
with the manipulator conguration, while their inuence on the
positioning accuracy depends on the latter. At present, there exist
various calibration techniques that are able to calibrate the manipulator geometric model using different modeling, measurement and identication methods [1116]. The identied errors can
be efciently compensated either by adjusting the controller input
(the target point) or by direct modication of the model parameters used in the robot controller.
The classical calibration procedure usually includes four steps:
n
Corresponding author at: Ecole des Mines de Nantes, 4 rue Alfred-Kastler,
Nantes 44307, France. Fax. 33 251 85 83 49.
E-mail address: alexandr.klimchik@mines-nantes.fr (A. Klimchik).
http://dx.doi.org/10.1016/j.rcim.2015.03.007
0736-5845/& 2015 Elsevier Ltd. All rights reserved.
152
proposed, they are usually categorized as closed-loop and openloop ones. The closed-loop calibration uses physical constraints on
the manipulator end-link (point, line or plane constraints, for instance). It is claimed to be autonomous and does not require any
external device [13,21,24]. However in this case, the manipulators
must have some redundancy to perform self-motion, and the robot
conguration should be carefully selected to satisfy particular
constraints. Therefore, the open-loop methods have found wide
applications; they are based on the full or partial pose measurements of the end-effector location using external devices. In
practice, the partial pose information is often used and provides
from one to ve dimensional measurements [11,25,26] instead of
the full pose information (6-dimensional location). In general, the
lower dimensional measurement is more attractive due to simplicity of calibration experiment setup. For this so-called partial
pose measurement technique, various external devices can be applied, such as laser tracking system [23], the ball-bar system [27]
and wire potentiometer [22], etc.
The identication step in robot calibration can be treated as the
best tting of the experimental data (given input variables and
measured output variables) by corresponding models. This problem has been addressed by a number of researchers who have
used various modeling methods and identication algorithms,
such as linear least square technique, LevenbergMarquardt algorithm, Kalman ltering technique and maximum likelihood estimator etc. [16,28]. Among them, the least square technique is the
most often applied one, which aims at minimizing the sum of
squared residuals [29]. An important problem here is non-homogeneity of the residual errors (distances and angles, for instance).
To solve this problem, usually a straightforward solution is applied: assigning weights or normalization, but this weight assigning procedure is very non-formal and not rigorous (while
being essential for the nal results). To solve the corresponding
optimization problem, there exist various numerical algorithms
such as gradient search [27,30], heuristic search and the others
[31]. However, these numerical techniques are often difcult to
apply due to large number of parameters to be tuned, that often
lead to low convergence. Nevertheless, for the case of geometric
calibration, the errors in the parameters are relatively small, so the
linearization technique can be successfully applied. In this case,
Table 1
Summary of related works for geometric calibration
Application (Manipulator) Number of model
parameters
Number of measurement
congurations
Measurement device
Identication algorithm
Achieved accuracy,
[mm]
35
80(1)
Two inclinometers(a)
0.40
42
27
36
24
23
42
23
30
42
15(1)
25(1)
800(1)
48(1)
100(3)
800(2)
100(2)
10(4)
18(5)
Single theodolite(a)
Laser tracking system(a)
Ball-bar system(a)
Wire potentiometer(a)
(b)
(b)
Touching probe(b)
(b)
Vision system(c)
LevenbergMarquardt
method
Non-linear LS
Random congurations.
Well distributed congurations.
Noise amplication index.
4
Minimum condition number.
5
Several observability indices.
Measurement technique:
2
3
a
b
c
Open-loop measurement.
Closed-loop measurement.
Simulation.
0.50
0.10
0.08
0.50
0.25
5.70
0.22
3.60
1.30
153
154
deviations. Under this assumption, the actual location of the manipulator end-effector, which incorporates the geometric errors is
expressed as t=g (q, 0 + ). In practice, the geometric errors
are usually relatively small, therefore the following linerized
model can be used
t=t 0 + J
(1)
ti Ji 2
i=1
min
(2)
However, the residual components of this system of identication equations are non-homogeneous (millimeters and radians,
for instance). In some cases, these components are normalized
before computing the squared sum, but it is a non-trivial step that
affects the identication accuracy. To overcome this difculty, it is
proposed to enhance the partial pose measurement method that
uses directly and only the positioning coordinates, but for several
reference points for each manipulator conguration. More details
of this method and its advantages will be presented in Section 4.
Step 4: This step deals with the identication and is aimed at
estimating the geometric parameters by using the corresponding
model and proper identication algorithm. Usually, the identication algorithms are based on the minimization of the leastsquare objectives that are derived assuming that the measurement
tool has a single reference point (see Eq.(2)), while the proposed
measurement technique operates with several of them. For this
reason, it is required to revise the existing identication techniques, taking into account both modication of the objective
function and increasing of the number of parameters (since each
reference point introduces additional parameters).
In addition, this step includes the evaluation of the parameter
identication accuracy. In practice, different sources of error may
affect the identication precision. They include the measurement
errors of the external device providing the end-effector position
coordinates (laser tracker in our case), the errors in the actuator
encoders (internal measurement devices) giving the manipulator
joint coordinates that depend on encoder resolution, etc. Besides,
the assumption concerning the manipulator model (the link rigidity, for instance) may also affect the identication accuracy. It is
clear that, all these sources of error can be hardly taken into account in calibration. For this reason, only the most signicant of
the above mentioned sources of error should be considered in the
accuracy analysis. As follows from our experience, the inaccuracy
of external measurement system has the most signicant impact
on the robot positioning accuracy, comparing to other sources of
error that can be assumed negligible in the frame of geometric
calibration.
Step 5: At the last step (implementation), the geometric errors
are compensated by modication of the geometric parameter values embedded in the robot controller. In the case when some
errors cannot be entered in the controller directly, an off-line error
compensation technique is required. This technique should compensate the manipulator errors via modication of the target
trajectory that becomes slightly different from the desired one
[46].
It is clear that the proposed scheme of robot calibration procedure allows us to improve the calibration accuracy for given
number of experiments (or to minimize the number of experiments for given accuracy). The steps 1, 4 and 5 in the calibration
procedure have been already well studied [9,47], while the steps
2 and 3 still require some revision in terms of the applicability to
particular manufacturing process where the robot is used. Hence,
the goal of this work is the enhancement of calibration technique
for manipulator geometric parameters using enhanced partial
pose measurement and design of experiments. Particular
155
(3)
Fig. 3. The industrial serial robot KUKA KR-270 and its geometric parameters. (a) Industrial robot KUKA KR270 and (b) the manipulator architecture.
156
R y (q2 + q2 )[TxTz R x R z ]L 2
R y (q3 + q3 )[TxTz R x R z ]L 3 R x (q4 + q4 )
0 = { d2 0 0 0 d 3 0 0 0 d 4 d5 0 0 0 0 0 0 0 0 }
(4)
Trobot = R z (q1)TxTyR x
L1
(5)
(6)
1
It should be stressed that 6 parameters related to the base transformation
and 6 parameters describing the tool transformation are not included in this expression (see Remarks 1 and 5), so it is in good agreement with common expression of Zhuang [52] that for robot with 6 rotational joints yields 30 independent
parameters.
(7)
ti = Ji , i = 1, 2, ... m
(8)
where t i = (p xi , p yi , p zi , xi , yi , zi )T is the pose deviation caused by small variation in the model parameters . It is
clear that the corresponding system of linear equations can be
solved with respect to if the number of experiments m is sufciently high and the manipulator congurations {qi , i = 1, m} are
different to ensure non-singularity of relevant observation matrix
used in the identication procedure. For this technique, each
conguration qi produces six scalar equations to be used for the
identication. Corresponding optimization problem (2) whose
solution leads to the desired parameters is often solved without paying attention to the non-homogeneity of the residual
components. In some cases, the weighted least-square technique
is used to resolve this problem, but the weighting coefcients are
usually dened intuitively, which may affect essentially the
identication accuracy.
The main difculty of this conventional technique (full-pose) is
that the orientation components ( xi . yi . zi )T cannot be measured directly. So, these angles are computed using excessive
number of measurements for the same conguration qi , which
produce Cartesian coordinates {(p xij , p yij , p zij )T | j = 1, n ; n 3} for
several reference points of the measurement tool attached to the
manipulator mounting ange (Fig. 2). Hence, instead of using 3mn
scalar equations, that can be theoretically obtained from the
measurement data, this conventional approach uses only 6m scalar
equations for the identication. This may obviously lead to some
loss of the parameter estimation accuracy.
To overcome this difculty, the proposed technique is based on
reformulation of the optimization problem (2) using only the data
directly available from the measurement system, i.e. the Cartesian
coordinates of all reference points pij (see Fig. 2). This idea allows
us to obtain homogeneous identication equations where each
residual has the same unit (mm, for instance), and the optimization problem is rewritten as follows
m
pij Jji(p) 2
i=1 j=1
min
(9)
157
Fig. 4. Difference between conventional full-pose approach and enhanced partial pose approache.
Here, the matrix J j i(p) with the superscripts (p) denotes the
position rows of the corresponding identication Jacobian J j i , the
index i denes the manipulator conguration number, and the
index j denotes the reference point number. An obvious advantage of this formulation is its simplicity and clarity of the residual vector norm denition (conventional Euclidian norm can be
applied here reasonably, the normalization is not required). So, the
problem of the weighting coefcient selection does not exist in
this case. In fact, under the assumption that measurement errors
are modeled as a set of independent and identically distributed (i.i.
d.) random values (similar for all directions x, y, z and for all
measurement congurations), the optimal linear estimator should
operate with equal weights for all equations. Besides, the most
important issue is related to the potential benets in the identication accuracy, since the total number of scalar equations incorporated in the least-square objective increases from 6m to 3mn.
To compare the efciency of the presented approach with the
conventional one, a simulation study has been carried out, which
dealt with geometric calibration of a 3-link spatial manipulator
(Fig. 4). Detail description of this example can be found in [53],
where it has been proved that the enhanced technique based on
partial pose information ensures essential improvement of parameter identication accuracy. Using these identied geometric
parameters, it is possible to evaluate the manipulator end-effector
positioning accuracy throughout the workspace. Corresponding
results are shown in Fig. 4, in which the achieved robot accuracy
has been compared for two techniques. As follows from this gure,
using the proposed approach, the maximum positioning error has
not even reached the minimum one by using conventional technique. Moreover, the minimum positioning error has been reduced
by a factor of 4. Fig. 5
Therefore, the partial pose technique is rather promising and
will be further used for calibration experiments in this work. In
contrast to the conventional methods, this technique allows us to
avoid the problem of non-homogeneity of the relevant optimization objective and does not require any normalization (which
arises in the case when full pose residuals are used).
(10)
Fig. 5. Improvement of manipulator positioning accuracy after calibration due to the enhanced partial pose technique: (a) Conventional technique and (b) proposed
technique.
158
where the vectors pij are incorporated in the fourth column of the
homogenous transformation matrix T ij , the matrix Tbase denes the
j
robot base location, the matrices T tool
, j = 1, n describe the locations of the reference points that are observed by the measurement system (see Fig. 2). Here, the matrix function Trobot ( qi , )
describes the manipulator geometry and depends on the current
values of the actuated coordinates qi and the parameters to be
estimated. Taking into account that any homogeneous transformation matrix T ba can be split into the rotational R ba and translational
pba components and presented as
j
pij = pbase + R baseprobot (qi , ) + R baseR robot (qi , )ptool
;
i = 1, m, j = 1, n.
(12)
(16)
pirobot
p base
i
T
i
+ I ~probot R robot base
uj
tool
(17)
and
j
j
utool
= R base ptool
(18)
j
Here the vectors pbase , base and utool
, j = 1, n are treated as
unknowns.
Applying to the linear system (17) the linear least-square
technique, the desired vectors dening the base and tool transformation parameters can be expressed as follows
m
1 m
p ; ; u1 ; ... un = A j T A j A j T p
tool
base base tool
i
i
i
i
i = 1
i = 1
(19)
where
j
pij pbase R baseprobot (qi , ) R baseR robot (qi , )ptool
2
i=1 j=1
j
= pbase + pirobot pirobot [~ base ] + R irobot utool
(11)
(15)
pij
pij
b b
R p
T ba = a a ,
0 1
0
z
y
0 x
[~] = z
0
y x
min
j , }
{p base,R base,ptool
(13)
j
where the vectors/matrices pbase , R base , ptool
, j = 1, n and are
treated as unknowns.
The main difculty with this optimization problem is that some
of the unknowns are included in the objective function in highly
non-linear way. So, to solve this problem, numerical optimization
technique is required. However in practice, the deviations in the
model parameters are relatively small, which allows us to linearize
the manipulator geometric model (12). This leads to a linear leastsquare problem, whose solution can be obtained straightforwardly
with the matrix pseudo-inverse. However, to simplify computations, here it is proposed to apply the linearization technique sequentially and separately with respect to two different subsets of
the model parameters (corresponding to the base/tool transformations and the manipulator geometry). Consequently, the identication procedure is split into two steps. In the frame of this
approach, the rst step deals with the estimation of pbase , R base ,
j
, which are related to the base and tool transformations (asptool
suming that the manipulator parameters are known). The second
step focuses on the estimation of under the assumption that the
base and tool components have been already identied. In order to
ensure that the desired identication accuracy can be achieved,
these two steps are repeated iteratively.
Step 1. For the rst step, taking into account that the errors in
the base orientation are relatively small, the matrix R base is presented in the following form
(14)
where I is a 3 3 identity matrix, vector base includes the deviations in the base orientation angles, and the operator [~] transforms the vector = (x , y , z )T into the skew symmetric matrix
as
j
Ai = I
...
~pi
T i
robot R robot
~pi
T
robot
...
0
...
R irobot
...
~pi
T
robot
...
0
...
...
... R irobot
...
(20)
pij = Jj i(p)
pij
pij
(21)
i
probot
where
is the residual vector corresponding to the
=
m n
1 m n
(22)
reasonable to investigate the rst approach that deals with optimization of the measurement congurations for limited number of
experiments.
In more general case when the measurement errors differ from
direction to direction, the expectation E i Ti can be expressed as
p ; ; u1 ; ... un
tool
base base tool
m
1 m
T
T
= A ij W ij2A ij A ij W ij2pi
i=1
i=1
(23)
and
m n
1 m n
j (p) T
j2 j (p)
= Ji W i Ji Jj i(p) T W ij2pij
i=1 j=1
i=1 j=1
(24)
pij = Jj i(p) + ij ; i = 1, m, j = 1, n
(xij ,
yij ,
ij kjT
) = diag (
xij2,
yij2,
zij2
), if i = k
(28)
(see our previous study on this issue presented in [54]). So, the
covariance matrix dening the calibration accuracy can be rewritten in the following form
ij
159
(25)
m n
1
^) = 2 J j (p) T W j2J j (p)
cov(
i
i=1 j=1
(29)
zij )T
m n
1 m n
i=1 j=1
i=1 j=1
(26)
m n
1
^) = 2 J j (p) T J j (p)
cov(
i
i
i=1 j=1
(27)
160
pd , as shown in Fig. 6.
The dispersion of these points can be evaluated by the variance
E pk T pk which in accordance with the above denition is equal
to the square of the performance measure 0. This yields the following expression
02 = E T J(p0) T J(p0)
(32)
Fig. 6. Dispersion of the manipulator positioning errors after calibration and performance measure for selection of measurement congurations (for given single
target point).
E (pk ) = pd
(30)
0 =
E (pk pd )T (p k p d )
(31)
which is the root-mean-square distance between the target position and the end-effector position after calibration. This indicator
is used below to describe the geometric errors compensation efciency. It is clear that the performance measure 0 is directly
related to the manipulator accuracy in an engineering viewpoint.
It is clear that the positioning error scattering and relevant
performance measure 0 highly depend on the target point position and varies throughout the workspace. In the frame of this
work, it is assumed that the manipulator accuracy can be evaluated for so-called test-pose that is specied in the relevant
technological process. This idea allows us to use the above mentioned performance measure 0 as an objective in the calibration
experiments design.
Using the adopted notations and assuming that the manipulator geometric model is linearized, the distance k can be computed as the Euclidean norm of the vector pk = J(p0) k , where the
subscript0 in the identication Jacobian J(p0) is related to the test
^ k is the difference between the estimated
pose q0 and k =
and true values of the robot geometric parameters respectively.
Further, taking into account expression (26) and the assumptions
concerning the measurement errors that are treated as unbiased
and i.i.d. random variables, it can be easily proved that the expectation E ( pk ) = 0. Therefore, the points pk that the end-effector
attains after compensation are located around the desired position
using
the
identity
equation
(33)
02
2trace J(p0)
i = 1
j=1
1
(p) T
J 0
(34)
02
J j i(p) T J j i(p)
m n
(35)
(36)
where px and py dene the end-effector position, q1, q2 are the joint
coordinates that dene the manipulator conguration. It can be
proved that in this case the parameter covariance matrix does not
depend on the angles q1i and is expressed as
2
cov() =
2
m
2
m ( i = 1 cos q2i )
cos q2i
i=1
(37)
m
m
m
i=1
cos q2i
where the vector = (l1, l2 ) denotes the parameters deviations to be identied, m is the number of experiments and is the
standard deviation. of the measurement noise.
For comparison purposes, the plans of experiments were obtained using three different strategies:
(1) the measurement congurations were generated randomly;
(2) the measurement congurations were obtained using the
conventional approach based on D-optimality principle;
(3) the measurement congurations were obtained using the
proposed test-pose based approach (see Section 6.1).
For the rst approach (i), the measurement congurations were
found in a trivial way, using a uniform random number generator
scaled within the joint limits. For the conventional approach
(ii), where the D-optimality principle was used (that has been
proved to be efcient in many applications), the performance
measure is equal of the covariance matrix determinant (37), which
yields
det(cov()) =
m2
m
i = 1
cos q2i )
(38)
(39)
i=1
It should be mentioned that this optimality condition also satises the A- and G-optimality principles. More details concerning
the calibration experiment planning using the above conditions
can be found in [56].
For the proposed approach (iii), it is assumed that the calibration quality is evaluated in the predened manipulator test conguration (q10 , q20 ). In this case, the performance measure 02 (34)
can be computed as
02 = 2 2
(40)
02 min =
cos2 q20
2
m 1 sin q20
(41)
i=1
cos q2i = m
1 sin q20
cos q20
(42)
161
Table 2
Accuracy comparison of the proposed and conventional approaches
Test-pose (q20 ), [deg.]
30
60
90
120
150
180
02 /c2
0.5
0.75
0.83
0.83
0.75
0.5
41
15
10
10
15
41
162
Fig. 7. Dispersion of manipulator positioning errors after calibration for different plans of experiments: (a) Random plan, (b) conventional plan and (c) proposed plan.
Table 3
Comparison of the optimal and quasi-optimal solutions for measurement congurations in calibration experiments, evaluated via 0min , [mm] (case of a 6-dof
m=3
km = 3
km = 12
Computational time
m=6
m = 12
0.0637
(3 1)
0.0521
(4 1)
km = 4
km = 6
m=4
0.0450
(3 2)
0.0319
(3 4)
38 min
0.0301
(4 3)
45 min
0.0426
(6 1)
0.0301
(6 2)
56 min
0.0301
(12 1)
1.6 h
163
Fig. 8. The experimental work-cell environment: (a) general view; (b) typical machining conguration (test-pose).
164
Table 4
The joint limits of robot KUKA KR-270
Table 6
Comparison of calibration plans with different diversity of measurement
congurations.
q1
q2
q3
q4
q5
q6
qmin , [deg.]
180
145
110
180
125
180
qmax ,
180
155
180
125
180
Calibration plans
[deg.]
Table 5
The work-cell space boundaries with respect to the robot base frame
px
py
pz
pmin , [mm]
1400
3000
300
pmax , [mm]
1800
2200
3500
(i)
{Sol. #1 3}
{Sol. #2 3}
{Sol. #3 3}
(ii) {Sol. #1 2, Sol. #2}
{Sol. #1 2, Sol. #3}
{Sol. #1, Sol. #2 2}
{Sol. #2 2, Sol. #3}
{Sol. #1, Sol. #3 2}
{Sol. #2, Sol. #3 2}
(iii) {Sol. #1, Sol. #2, Sol. #3}
Random congurations (for
comparison)
7.85
7.84
7.83
7.84
7.84
7.83
7.83
7.83
7.83
7.83
17.33
56 min
56 min
56 min
1.9 h
1.9 h
1.9 h
1.9 h
1.9 h
1.9 h
2.8 h
0.05 s
Table 7
Identication results for manipulator tool transformations
Reference point #1 (P1) Reference point #2 (P2) Reference point #3 (P3)
Value, [mm] CI
Value, [mm] CI
Value,
[mm]
CI
px
277.23
70.05
276.49
7 0.05
278.44
70.05
py
46.53
70.04
48.25
7 0.04
103.73
70.05
pz
93.87
70.04
94.05
7 0.05
2.17
70.05
Table 8
Identication results for manipulator geometric parameters
Parameter Unit
Value
Condence interval
Estimated using covariance matrix
px1 d2
[mm]
0.353
7 0.086
7 0.102
py1
[mm]
0.426
7 0.272
7 0.421
x1
[deg.]
0.015
7 0.005
7 0.005
q2
[deg.]
0.007
7 0.005
7 0.004
7 0.060
px2 d3
[mm]
0.458
7 0.082
x2
[deg.]
0.022
7 0.014
7 0.022
z2
[deg.]
0.023
7 0.005
7 0.005
q3
[deg.]
0.023
7 0.019
7 0.013
px3 d4
[mm]
0.214
7 0.089
7 0.093
pz3 d5
[mm]
0.508
7 0.363
7 0.259
z3
[deg.]
0.011
7 0.017
7 0.022
q4
[deg.]
0.001
7 0.008
7 0.009
py4
[mm]
0.167
7 0.113
7 0.044
pz4
[mm]
0.018
7 0.073
7 0.044
z4
[deg.]
0.025
7 0.015
7 0.010
q5
[deg.]
0.011
7 0.027
7 0.009
pz5
[mm]
0.016
7 0.104
7 0.041
z5
[deg.]
0.008
7 0.018
7 0.007
Table 9
Evaluation of the manipulator accuracy improvement based on residual analysis
Criterion
Coordinate-based
residuals, [mm]
Distance-based residuals, [mm]
max
RMS
max
RMS
Before
calibration
After
calibration
Improvement
factor
1.25
0.54
1.31
0.94
0.32
0.10
0.39
0.17
4.0
5.3
3.5
5.5
165
solutions have been obtained that ensure almost the same value of
the considered performance measure 0 (13.6mm). Corresponding
solutions (measurement congurations) are presented in [45]. For
comparison purposes, these solutions have been evaluated both
separately and in different combinations, assuming that the
measurements are performed 18 times in the following way: (i)
repeating three times the measurements in congurations from a
single set; (ii) using twice congurations from one set and only
once from the second set; (iii) using all congurations from three
sets simultaneously (but only once). Corresponding values of 0
are presented in Table 6. As follows from this table, the diversity of
manipulator congurations has almost negligible contribution to
the improvement of robot accuracy (it is about 7.85 mm, the difference is less than 0.2%). This conrms the results from Subsection 6.3, which claims that using simple repetition of the optimal
plan with lower number of measurement congurations essentially reduces the experimental complexity while the same calibration accuracy can be achieved.
7.3. Identication of geometric parameters
The obtained measurement congurations have been used for
the calibration experiments for KUKA KR-270 industrial robot. It is
worth mentioning that each manipulator conguration provides
27 values of the position coordinates. These coordinates have been
obtained using two different locations of the laser tracker (see more
details in [45]). However, at certain congurations, some of the
reference points were not visible for both laser tracker locations.
This problem can be solved by increasing the number of laser
tracker locations, but in practice such solution is limited by the
experimental time as well as the work-cell constraints. On the
other hand, since the calibration experiment employs two laser
tracker placements, 6 additional parameters describing the second
laser tracker location should be also identied. In total, the system
of identication equations contains 432 expressions that can be
used to identify the whole set of 39 geometric parameters. To
achieve the highest identication accuracy, here it is proposed to
use all measurements corresponding to 18 manipulator congurations simultaneously for calibration of the geometric
parameters.
Using the obtained measurement data, the two-step identication procedure has been applied (see Section 5). On the rst
step, the base and tool transformations have been computed,
corresponding results are presented in Table 7. On the second step,
these transformations have been used for the identication of the
manipulator geometric parameters, which are presented in Table 8. It should be mentioned that in order to increase the identication accuracy, this two-step procedure has been repeated
iteratively (280 iterations, computing time was less than two
Fig. 10. Histograms of residual distribution along X-, Y-, and Z-directions after geometric calibration: (a) X-direction, (b) Y-direction and Z-direction.
166
Fig. 11. Residual distribution after geometric calibration for different measurement congurations.
Parameters
Comparing to the machining accuracy required for the considered milling process (0.050.25 mm), the above listed positioning error impacts are not negligible for the most of the geometric parameters. So, their deviations should be compensated
either in the geometric model embedded in the robot controller or
at the step of generation of the machining trajectory.
For comparison purposes, the manipulator accuracy improvement due to calibration has been studied based on the residual
analysis before and after calibration (computed using the nominal
and identied values of geometric parameters respectively). Here,
two types of residuals have been examined, the coordinate-based
and distance-based ones. Corresponding results are presented in
Table 9, which includes the maximum and root mean square
(RMS) values of the relevant residuals. As follows from the results,
both types of the residuals have been essentially reduced after
calibration. In particular, the maximum values have been reduced
by a factor of 4 and 3.5, while the RMS values have been decreased
by a factor of 5.3 and 5.5, respectively.
Hence, the obtained results allow us to improve essentially the
manipulator accuracy for the measurement congurations that
were used in the identication. So, it is reasonable to expect that
using the geometric model, which integrates the identied parameters, the desired positioning accuracy for the given test conguration can be also achieved. A more detailed analysis concerning the parameter identication accuracy and its impact on
8. Conclusions
This paper presents a new approach for calibration experiments design for serial industrial robots. This approach employs a
new industry-oriented performance measure, which evaluates the
quality of calibration plan via the manipulator positioning accuracy after geometric error compensation, and considers the industrial requirements associated with the prescribed manufacturing task. It is proved that the proposed performance measure
can be presented as the weighted trace of the relevant covariance
matrix, where the weighting coefcients are dened by the corresponding test-pose. Such an approach allows us to nd the
Acknowledgments
The authors would like to acknowledge the nancial support of the
French Agence Nationale de la Recherche (Project ANR-2010-SEGI003-02-COROUSSO), France and FEDER ROBOTEX project, France.
References
[1] S. Aguado, D. Samper, J. Santolaria, J.J. Aguilar, Identication strategy of error
parameter in volumetric error compensation of machine tool based on laser
tracker measurements, Int. J. Mach. Tools Manuf. 53 (2012) 160169.
[2] G. Du, P. Zhang, D. Li, Online robot calibration based on hybrid sensors using
Kalman lters, Robot. Comput.-Integr. Manuf. 31 (2015) 91100.
[3] A. Joubair, M. Slamani, I.A. Bonev, A novel XY-Theta precision table and a
geometric procedure for its kinematic calibration, Robot. Comput.-Integr.
Manuf. 28 (2012) 5765.
[4] S. Marie, E. Courteille, P. Maurine, Elasto-geometrical modeling and calibration
of robot manipulators: application to machining and forming applications,
Mech. Mach. Theory 69 (2013) 1343.
[5] A. Nubiola, I.A. Bonev, Absolute robot calibration with a single telescoping
ballbar, Precis. Eng. 38 (2014) 472480.
[6] Z.S. Roth, B. Mooring, B. Ravani, An overview of robot calibration, IEEE J. Robot.
Autom. 3 (1987) 377385.
[7] J. Santolaria, F.-J. Brosed, J. Velzquez, R. Jimnez, Self-alignment of on-board
measurement sensors for robot kinematic calibration, Precis. Eng. 37 (2013)
699710 7//.
[8] J. Santolaria, M. Gins, Uncertainty estimation in robot kinematic calibration,
Robot. Comput.-Integr. Manuf. 29 (2013) 370384.
[9] A. Elatta, L.P. Gen, F.L. Zhi, Y. Daoyuan, L. Fei, An overview of robot calibration,
Inform. Technol. J. 3 (2004) 7478.
[10] A. Klimchik, A. Pashkevich, D. Chablat, G. Hovland, Compliance error compensation technique for parallel robots composed of non-perfect serial chains,
Robot. Comput.-Integr. Manuf. 29 (2013) 385393.
[11] D. Daney and I. Z. Emiris, Robust parallel robot calibration with partial information, in: Proceedings of the IEEE International Conference on, Robotics
and Automation, ICRA, 2001, pp. 32623267.
[12] M.R. Driels, Full-pose calibration of a robot manipulator using a coordinatemeasuring machine, Int. J. Adv. Manuf. Technol. 8 (1993) 3441.
[13] M. Ikits and J. M. Hollerbach, Kinematic calibration using a plane constraint,
in: Proceedings of the IEEE International Conference on Robotics and Automation, 1997, pp. 31913196.
[14] W. Veitschegger and C.-h. Wu, A method for calibrating and compensating
robot kinematic errors, in: Proceedings of the IEEE International Conference
on Robotics and Automation, 1987, pp. 3944.
[15] H. Hage, P. Bidaud, and N. Jardin, Practical consideration on the identication
of the kinematic parameters of the Stubli TX90 robot, in: Proceedings of the
13th World Congress in Mechanism and Machine Science, Guanajuato, Mexico,
2011, p. 43.
[16] J.-M. Renders, E. Rossignol, M. Becquet, R. Hanus, Kinematic calibration and
geometrical parameter identication for robots, Robot. Autom. IEEE Trans. 7
(1991) 721732.
[17] B. Mooring The effect of joint axis misalignment on robot positioning accuracy,
in: Proceedings of the Engineering Conference on ASME International Computers. 1983, pp. 151156.
[18] W. Khalil, E. Dombre, Modeling, Identication and Control of Robots, Butterworth-Heinemann, Oxford, 2004.
167
[19] S. A. Hayati, Robot arm geometric link parameter estimation, in: Proceedings
of the 22nd IEEE Conference on, Decision and Control 1983, pp. 14771483.
[20] H. Zhuang, Z.S. Roth, F. Hamano, A complete and parametrically continuous
kinematic model for robot manipulators, Robot. Autom. IEEE Trans. 8 (1992)
451463.
[21] M. A. Meggiolaro and S. Dubowsky, An analytical method to eliminate the
redundant parameters in robot calibration, in: Proceedings of the IEEE International Conference on, Robotics and Automation ICRA'00 2000, pp. 3609
3615.
[22] M.R. Driels, W. Swayze, Automated partial pose measurement system for
manipulator calibration experiments, Robot. Autom. IEEE Trans. 10 (1994)
430440.
[23] Y. Bai, H. Zhuang, and Z. S. Roth, Experiment study of PUMA robot calibration
using a laser tracking system, in: Proceedings of the IEEE International
Workshop on Soft Computing in Industrial Applications SMCia/03. 2003, pp.
139144.
[24] A. Rauf and J. Ryu, Fully autonomous calibration of parallel manipulators by
imposing position constraint, in: Proceedings of the IEEE International Conference on, Robotics and Automation ICRA. 2001, pp. 23892394.
[25] S. Besnard and W. Khalil, Calibration of parallel robots using two inclinometers, in: Proceedings of the IEEE International Conference on, Robotics
and Automation 1999, pp. 17581763.
[26] A. Rauf, A. Pervez, J. Ryu, Experimental results on kinematic calibration of
parallel manipulators using a partial pose measurement device, Robot. IEEE
Trans. 22 (2006) 379384.
[27] A. Goswami, A. Quaid, and M. Peshkin, Complete parameter identication of a
robot from partial pose information, in: Proceedings of the IEEE International
Conference on, Robotics and Automation 1993, pp. 168173.
[28] B.W. Mooring, Z.S. Roth, M.R. Driels, Fundamentals of Manipulator Calibration,
Wiley, New York, 1991.
[29] C. Rao, H. Toutenburg, Linear Models and Generalizations: Least Squares and
Alternatives, Springer, New York, 1999.
[30] H. Zhuang, J. Yan, O. Masory, Calibration of Stewart platforms and other parallel manipulators by minimizing inverse kinematic residuals, J. Robot. Syst. 15
(1998) 395405.
[31] R. Fletcher, Practical Methods of optimization, John Wiley & Sons, Hoboken,
NJ, Wiley, 2013 452 p.
[32] E. Moore, H., On the reciprocal of the general algebraic matrix, Bull. Am. Math.
Soc. 26 (1920) 394395.
[33] R. Penrose, A generalized inverse for matrices, Proc. Cambridge Philos. Soc
(1955) 406413.
[34] J.-H. Borm, C.-H. Meng, Determination of optimal measurement congurations
for robot calibration based on observability measure, Int. J. Robot. Res. 10
(1991) 5163.
[35] M.R. Driels, U.S. Pathre, Signicance of observation strategy on the design of
robot calibration experiments, J. Robot. Syst. 7 (1990) 197223.
[36] H. Zhuang, O. Masory, and J. Yan, Kinematic calibration of a Stewart platform
using pose measurements obtained by a single theodolite, in: Proceedings of
the IEEE/RSJ International Conference on Intelligent Robots and Systems 95.
Human Robot Interaction and Cooperative Robots 1995, pp. 329334.
[37] M. A. Meggiolaro, G. Scrifgnano, and S. Dubowsky, Manipulator calibration
using a single endpoint contact constraint, in: Proceedings of ASME Design
Engineering Technical Conference, Baltimore, USA, 2000.
[38] H. Zhuang, J. Wu, and W. Huang, Optimal planning of robot calibration experiments by genetic algorithms, in: Proceedings of the IEEE International
Conference on, Robotics and Automation 1996, pp. 981986.
[39] D. Daney, Y. Papegay, B. Madeline, Choosing measurement poses for robot
calibration with the local convergence method and Tabu search, Int. J. Robot.
Res. v24 (2005) 501518.
[40] Y. Sun and J. M. Hollerbach, Observability index selection for robot calibration,
in: Proceedings of the IEEE International Conference on, Robotics and Automation ICRA 2008, pp. 831836.
[41] A. Nahvi and J. M. Hollerbach, The noise amplication index for optimal pose
selection in robot calibration, in: Proceedings. of the IEEE International Conference on Robotics and Automation 1996, pp. 647654.
[42] A. Nahvi, J. M. Hollerbach, and V. Hayward, Calibration of a parallel robot using
multiple kinematic closed loops, in: Proceedings of the IEEE International
Conference on Robotics and Automation 1994, pp. 407412.
[43] D. Daney, Optimal measurement congurations for Gough platform calibration, in: Proceedings of the IEEE International Conference on, Robotics and
Automation ICRA'02. 2002, pp. 147152.
[44] H. Zhuang, K. Wang, and Z. S. Roth, Optimal selection of measurement congurations for robot calibration using simulated annealing, in: Proceedings of
the IEEE International Conference on Robotics and Automation 1994, pp. 393
398.
[45] Y. Wu, Optimal pose selection for the identication of geometric and elastostatic parameters of machining robots, Ecole des Mines de Nantes, PhD Thesis
(2014).
[46] Y. Chen, J. Gao, H. Deng, D. Zheng, X. Chen, R. Kelly, Spatial statistical analysis
and compensation of machining errors for complex surfaces, Precis. Eng. 37
(2013) 203212 1//.
[47] Z. Roth, B. Mooring, B. Ravani, An overview of robot calibration, Robot. Autom.
IEEE J. 3 (1987) 377385.
[48] H. Stone, A. Sanderson, and C. Neuman, Arm signature identication, in:
Proceedings of the IEEE International Conference on, Robotics and Automation
1986, pp. 4148.
168
DOI 10.1007/s12206-013-1167-7
KDC, 205 GNATC, 676 Bangji-ri, Sanam, Sachen, Gyungnam, 664-942, Korea
Dept. of Bio-Nano System Engineering, Chonbuk National University, 664-14 1-Ga, Deokjin-Dong, Deokjin-Gu, Jeonju, Jeonbuk, 561-756, Korea
3
Vehicle Advanced Team, Corporate R&D Division, Hyundai-Kia Motors, 772-1 Jangduk-Dong, Hwaseong, Gyeonggi, 445-706, Korea
4,*
Div. of Mechanical System Engineering, Chonbuk National University, 664-14 1-Ga, Deokjin-Dong, Deokjin-Gu, Jeonju, Jeonbuk, 561-756, Korea
2
(Manuscript Received July 27, 2012; Revised August 9, 2013; Accepted September 16, 2013)
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
Abstract
This paper optimizes the combination of design parameters for improving the kinematic characteristics of a midsize truck using both
design of experiment and computer simulation. A computational model of the front suspension and steering system of a midsize truck is
developed for analyzing kinematic and compliance characteristics. A taper leaf spring is modeled as a flexible body using finite elements.
A bump mode test is performed to validate the reliability of the developed computational model. Mean absolute values of the toe angle
and wheel base change are used as objective functions. Modifiable hard points are selected as design parameters. An optimal combination of design parameters for improving kinematic characteristics is suggested based on analyses of variance and factor effects using a
table of orthogonal arrays.
Keywords: Design of experiment; Kinematics analysis; Toe angle; Wheel base; Suspension parameter measuring device (SPMD); Table of orthogonal
array; Analysis of variance (ANOVA); Analysis of factor effect
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
1. Introduction
Since the online purchase of goods has become more common, the demands on midsize trucks have increased rapidly.
In order to safely and quickly deliver goods, midsize trucks
must perform reliably. Part of the process of improving the
performance of midsize trucks is improving straight-ahead
driving stability.
The kinematic and compliance characteristics of the front
suspension and steering systems of midsize trucks influence
the wheel alignment and wheel center locus. Changes in wheel
alignment and wheel center locus influence straight-ahead
driving stability. The joints and links that comprise the front
suspension and steering system influence kinematic characteristics significantly. The springs and bushings used for connecting links significantly influence compliance characteristics. In addition, the left and right asymmetry of the steering
system influences the kinematic and compliance characteristics.
In the early stage of the development of a midsize truck,
many vehicle and part tests are performed to find an optimal
combination of design parameters to influence straight-ahead
*
Corresponding author. Tel.: +82 63 270 2377, Fax.: +82 63 270 2388
E-mail address: ohcy@jbnu.ac.kr
2. Computational modeling
The computational model of the front suspension and steer-
964
B. M. Kim et al. / Journal of Mechanical Science and Technology 28 (3) (2014) 963~969
Fig. 2. Computational model of the front suspension and steering system of a mid-sized truck on a test rig.
ment leaf spring model. The dummy part was joined to the
front axle with a bushing element. The upper end of a shock
absorber was connected to the frame with a universal joint.
The lower end of the shock absorber was connected to the
front axle with a spherical joint.
The steering system of the midsize truck used in this paper
is composed of a steering wheel, steering column, steering
column housing, steering shaft, steering rack, drop arm, drag
link, knuckle, and tie rod. The steering wheel was connected
to the steering column with a revolute joint. The steering column was connected to the steering column housing with a
cylindrical joint. Also, it was connected to the steering shaft
with a universal joint. The steering column housing was connected to the ground part with a fixed joint. The steering shaft
was connected to the steering rack with a revolute joint. The
steering rack was connected to a drop arm with a revolute
joint. The drop arm was connected to a drag link with a spherical joint. The drag link was connected to the left knuckle with
a universal joint. The left end of a tie rod was connected to the
left knuckle with a spherical joint, and the right end of a tie
rod was connected to the right knuckle with a universal joint.
Fig. 2 shows the computational model of the front suspension
and steering system of a midsize truck on a test rig.
B. M. Kim et al. / Journal of Mechanical Science and Technology 28 (3) (2014) 963~969
changes in the left and right wheel rates. The simulation results predict the trends of the test results over the whole range
of the stroke. Fig. 4 shows a comparison of the changes in the
left and right toe angles. The trends seen in the simulation
show a little difference from the results of the test at strokes
between -55 mm ~ -30 mm for both the left and right toe angles. However, the trends seen in the simulation are in good
agreement with those from the test for the remainder of the
stroke. Fig. 5 shows a comparison of the changes in the left
and right caster angles. For the left caster angle, the trend seen
in the simulation show a little difference with those from the
test stroke between -55 mm ~ -30 mm, but the simulation
result provides a good estimate of the trend seen in the test
result for the remainder of the stroke. For the right caster angle,
the simulation result provides a good estimate of the trend
seen in the test result over the whole range of stroke. Fig. 6
shows a comparison of the changes in the left and right wheel
base. The simulation results predict the test results over the
whole range of the stroke. From these comparisons, we can
conclude that the wheel rate change, toe angle change, caster
angle change, and wheel base change considered as major
factors representing the kinematic characteristic acquired from
the simulation provide a good estimate of those acquired from
the test in general.
965
Fig. 4. Comparison of the changes in toe angle from the simulation vs.
the SPMD test.
toe angle and wheel base change are considered to be the two
most important factors influencing straight-ahead driving stability. Therefore, we sought to optimize the design parameters
in order to minimize toe angle and wheel base changes.
To design an experiment, we need to define an objective
function, select proper design parameters, and select the num-
966
B. M. Kim et al. / Journal of Mechanical Science and Technology 28 (3) (2014) 963~969
ber of levels. This paper defines objective functions to minimize the mean absolute values of toe angle and wheel base
change as shown in Eqs. (1) and (2).
n
T (j )
i
Minimize
F (j ) =
i =1
(1)
(2)
D (d )
i
Minimize
D (d ) =
i =1
(3)
Level 0
Level 1
Level 2
Current
-20
Current
Current
+20
Current
-20
Current
Current
+20
Current
-20
Current
Current
+20
Current
-20
Current
Current
+20
Current
-20
Current
Current
+20
Current
-20
Current
Current
+20
967
B. M. Kim et al. / Journal of Mechanical Science and Technology 28 (3) (2014) 963~969
Design parameter
Trial
1 2 3 4 5 6 7 8 9 10 11 12
13 Toe angle
Wheel
base
0 0 0 0 0 0 0 0 0 0
0.4899
5.2583
0 0 0 0 1 1 1 1 1 1
0.4648
5.2204
0 0 0 0 2 2 2 2 2 2
0.4377
5.1794
0 1 1 1 0 0 0 1 1 1
0.1578
4.5886
0 1 1 1 1 1 1 2 2 2
0.5917
5.4122
0 1 1 1 2 2 2 0 0 0
0.6319
5.4729
0 2 2 2 0 0 0 2 2 2
0.1926
4.8014
0 2 2 2 1 1 1 0 0 0
0.2354
4.8744
0 2 2 2 2 2 2 1 1 1
0.7356
5.6297
10 1 0 1 2 0 1 2 0 1 2
0.2834
4.9468
11 1 0 1 2 1 2 0 1 2 0
0.2541
4.9016
12 1 0 1 2 2 0 1 2 0 1
0.9164
5.9023
13 1 1 2 0 0 1 2 0 1 2
0.2578
4.9081
14 1 1 2 0 1 2 0 1 2 0
0.2305
4.8663
15 1 1 2 0 2 0 1 2 0 1
0.8295
5.7714
16 1 2 0 1 0 1 2 0 1 2
0.2363
4.8758
17 1 2 0 1 1 2 0 1 2 0
0.2109
4.8370
18 1 2 0 1 2 0 1 2 0 1
0.7577
5.6632
19 2 0 2 1 0 2 1 0 2 2
0.1708
4.6439
20 2 0 2 1 1 0 2 1 0 2
0.7177
5.6024
21 2 0 2 1 2 1 0 2 1 0
0.7150
5.5981
22 2 1 0 2 0 2 1 0 2 1
0.1556
4.6332
23 2 1 0 2 1 0 2 1 0 2
0.6529
5.5049
24 2 1 0 2 2 1 0 2 1 0
0.6478
5.4984
25 2 2 1 0 0 2 1 0 2 1
0.1429
4.6243
26 2 2 1 0 1 0 2 1 0 2
0.5989
5.4235
27 2 2 1 0 2 1 0 2 1 0
0.5937
5.4155
AG e e F e e e B C
e T = 12.310 140.05
Factor
Sum of
square
Degree of
freedom
Mean
square
(%)
Contribution
0.0143
0.0072
0.260
0.0684
0.0342
1.230
2.7376
1.3688
49.13
1.7260
0.8630
30.98
0.0003
0.0001
0.001
1.0256
0.5128
18.41
Optimal
combination
Toe
angle
A1G1F1B1C1D1
A0G2F0B2C2D2
0.44
0.15
Wheel
base
A1G1F1B1C1D1
A0G2F0B2C2D2
5.19 mm
4.54 mm
65.9% Down
12.5% Down
CT = 5.613 726.49
Table 3. Result of analysis of variance of toe angle.
Factor
Sum of
square
Degree of
freedom
Mean
square
(%)
Contribution
0.0144
0.0072
0.53
0.0314
0.0157
1.16
0.9738
0.4869
35.90
0.6587
0.3294
24.28
0.0087
0.0044
0.32
1.0256
0.5128
37.81
968
B. M. Kim et al. / Journal of Mechanical Science and Technology 28 (3) (2014) 963~969
References
Fig. 9. Comparison of the changes in toe angle and wheel base from
the current combination of design parameters vs. the suggested optimal
combination of design parameters.
5. Conclusions
This paper described the development of a computational
model of the front suspension and steering system of a midsize truck for the purpose of analyzing the kinematic characteristics. A taper leaf spring was modeled as a flexible body
using finite elements. A bump mode test was performed with
a SPMD and a simulation was performed under the same conditions as the test. The trends in both the left and right toe
angles and left caster angle acquired from the simulation
showed a little difference from those acquired from the test at
strokes between -55 mm ~ -30 mm. However, the trends in the
wheel rate, toe angle, caster angle, and wheel base change
acquired from the simulation were good estimates of those
acquired from the test in general. Design of experiment and
computational simulation were both used to seek an optimal
combination of design parameters for minimizing the change
in toe angle and wheel base. Mean absolute values of toe angle and wheel base change were used as objective functions.
Changes in the x, y, and z directions of hard points connecting
the drag link to the knuckle, and connecting the drop arm to
the drag link were selected as design parameters for optimization. A table of orthogonal array having 13 columns and 27
rows was generated. The result of analysis of variance showed
that x and z coordinate of a hard point joining drop arm and
drag link, and z coordinate of a hard point joining drag link
and knuckle were primary factors influencing toe angle and
wheel base change. The result of factor effect analysis suggested that design parameters A and F need to be decreased
B. M. Kim et al. / Journal of Mechanical Science and Technology 28 (3) (2014) 963~969
Bo Myung Kim received his B.S. degree in Aerospace Engineering and M.S.
degree in Precision Mechanical Engineering from Chonbuk National University, Korea, in 2003 and 2009, respectively. He is currently a researcher at
KDC Co. in Sacheon, Korea. Mr. KIMs
research interests are in the area of vehicle dynamics.
969
404
Nomenclature
F - steering error; L - steering angle of the wheel;
S - steering wheel angle; V - toe angle; CCD - central
composite design; DSA - design sensitivity analysis;
eL - unit vector of the steering axis; eYR - unit vector of the
wheel rotation axis; FFD - full factorial design;
LF - wheelbase, mm; MSE - maximum steering error, ;
O - centre of the bend; SA - centre of gravity of the vehicle
body; sR - track width, mm; SS - sweep study; zA8 - wheel
displacement, mm
1. Introduction
Through their design simplicity and ease of manufacturing, Ackerman trapezoidal linkage has a broad application area as the steering system of heavy commercial
vehicles equipped with solid axles. On the other hand, as a
result of the comfort and control requirements, one of the
main targets to be reached in the design of vehicle suspensions is to keep the unsprung mass as small as possible. In
order to satisfy these demands, independent front suspensions (IFS) are applied increasingly on busses and trucks
by the heavy commercial vehicle manufacturers [1]. In this
case, more sophisticated systems are demanded to meet the
sufficient steering and independent wheel travel functions
simultaneously. Because of its design advantages, multilink steering linkage (or opposed four-bar linkage [2, 3])
is used in the majority of the passenger busses equipped
with IFS. This mechanism basically consists of two relay
levers, one track rod, two tie rods and two steering arms as
seen in Fig. 1.
Kinematic model of a typical bus IFS including
405
(1)
F Li La Li LaA Li .
(2)
Sweep study
Threshold value
MSE
Standardised effect
Iteration number
DOE-RSM (CCD)
DOE-RSM (CCD)
Response
surface
Factor 1
j
LF
Factor 3
Factor 2
Factor 3
Factor 1
Initial
Optimised
Toe angle
Factor 2
Factor 2
cot Li
Term
Wheel travel
LaA Li tan 1
Factor 1
Base
Optimised
Steering angle
406
2. Multibody model of the double wishbone suspension
3. Methodology
Direction
14
10
15
7
6
12
IIx
1 : Steering wheel
6 : Drag link
11: Upright
Fig.
5 Multibody model
of the bus IFS and
steering system
R
S
Lower wishbone
Upper wishbone
AirIIzspring
Shock absorber
IIy
U
U: Universal
T: Translational
R: Revolute
S: Spherical
C: Homokinetic
IIy
Direction R
y ,
12:
13:
14:
IIx
15:
IIx
(4)
(5)
(6)
where
y y1 ,y2 ,...,yM ;
T
1 x11
1 x12
... ...
1 x1M
x21
x22
...
x2 M
(7)
x51
... x52
;
... ...
... x5 M
...
0 , 1 , 2 , 3 , 4 , 5 ;
T
1 , 2 ,..., M ,
T
(3)
IIz
7 : Relay lever
8 : Track rod
9 : Tie rodR R
10 : Steering arm
i j
y 0 1 x1 2 x2 3 x3 4 x4 5 x5 .
IIy
2 : Steering column
3 : Intermediate shaft
Direction
4 : Steering box
5 : Pitman arm R
i 1
y 0 1 x1 2 x2 3 x12 4 x22 5 x1 x2 .
11
13
y 0 i xi ij xi x j .
(8)
(9)
(10)
T T y .
1
(11)
407
for this purpose. The CCD involves the use of a two-level
factorial or fraction combined with 2k axial or star points.
Hence, the design includes factorial points, 2k axial points,
and total nc centre runs, yielding a total number of
2k+2k+nc runs are carried out to achieve experimental
data. A comparison of the two level FFD and the CCD for
three factors is seen in Fig. 7.
objective (V or F) is defined. Except for some differences, optimisation procedures for V and F are similar.
Absolute value of the maximum deviation of the objective
obtained from the primary simulation is imported to Adams/Insight. The factors and design targets are also defined. In the light of the design constraints, the variation
ranges of the factors are chosen. Effective and noneffective parameters are identified by screening experiments. Results obtained from these experiments are also
used for the SS of F optimisation. For the optimisation
processes, investigation strategy is chosen as DOE-RSM.
Number of the runs is determined according to the design
type. The design space and workspace which contain the
full set of the design trials and the results of their analyses
are generated. Optimum set of the factors which gives the
target value of the design objective is obtained by fitting
the results to a polynomial or a response surface. In order
to control the estimation results of the regression analysis,
a multibody model which contains the optimum values of
the factors is also carried out. Results obtained from this
model are compared with the target value. Some steps of
this flow chart have similar characteristics with the methodology given by [22] for the optimisation of the suspension parameters to improve impact harshness (IH) of road
vehicles.
4. Toe optimisation
Fig. 9 shows the six total translational DOF of the
hardpoints A9 and A10 which determine the position of the
tie rod and directly affect the V in case of wheel travel.
Since tie rod is connected to the relay lever by the spherical joint A10, initial position of the hardpoints A13 and A14
does not have any remarkable effect on V for a given value of L. Hence, only the hardpoints of A9 and A10 are chosen as factors in stage 2. A summary of the design limitations are also given in Fig. 10. Here OV, the midpoint of
the front axle was chosen as the reference point for this
work. Appropriate positions of the hardpoints A 9 and A10
were searched in the design volumes Cube 1 and Cube 2.
Initial positions of the cubes and their edge lengths were
chosen according to the design limitations which are summarised below. Position of A9 in x axis should render possible enough space for wheel-end and brake system (Volume B) to eliminate any penetration of the mechanical elements. Initial value of A10x co-ordinate is chosen such that
the mechanism should not be blocked in the L range of the
front wheels. As a design rule, the angles and between
steering arm and tie rod should not be lower than 15 [2] in
408
course of the maximum steering position of the front
wheels as seen in Fig. 11.
Steel wheel limits the -y co-ordinate of A9 because of the installation issues. A gap e is necessary
which limits the -y co-ordinate of A10 because of the installation issues. In this design, gap e is assumed as 60 mm by
considering the physical diameter of the spherical joint
A10.
IIy
A14
A14
IIx
Li
A7
A13
A13
A10
15
A9
La
A7
A10
j
409
Table 1
Initial and optimised hardpoint co-ordinates for tie rod
Factor, mm
A9x
A9y
A9z
A10x
A10y
A10z
Initial
-268
-795
-80
200
-100
0
shows the contour plots obtained from the software for the
interactions of the design parameters where, the first term
indicates the ordinate and the second is for abscissa.
100
zA8, (mm)
Initial
Optimised
-100
-2
-1
0
V, ()
Optimised CCD
-308
-755
-55.25
240
-140
-42.4
c
Fig. 15 Parallel wheel travel simulation of the IFS: zA8=:
a - +100 mm; b - 0 mm; c - 100 mm
Deviation characteristics of initial and optimised
models are given in Fig. 16. Maximum V values were obtained for + 100 mm bump and -100 mm rebound of the
wheel as -0.26 and -0.32 respectively.
In order to evaluate the interaction effects of the
tie rod factors on V deviation, the design matrix obtained
from CCD was also imported to MINITAB. Fig. 17
410
ing iS ratio which can be defined as:
iS
2.5
V 1.5
-75
-50
-50
-25
-25
0
35
A9z
25
iS, ()
A10z
Table 2
Basic dimensions of the passenger bus (mm)
LV
3957
sRV
2096
15
L1 : 440 mm
LF
6050
sRH
1825
j
1844
(13)
where, S is the steering wheel angle. In this study, possible range of iS was chosen in the range of 18-23 (-) due to
the manufacturers demand. Effect of relay lever length on
iS of the base mechanism can be seen as a function of S in
Fig. 20. L1 was chosen as 540 mm.
0.5
-75
S
,
L
-750
-500
-250
L2 : 540 mm
0
S, ()
250
L3 : 640 mm
500
750
(12)
411
6
MSE20, ()
4
2
15
30
45
60
Iteration number
75
90
Factor (mm)
A9x
A9y
A10x
A10y
A13x
A13y
A14x
A14y
Base
-308
-755
240
-140
-28.5
-140
-300
-140
Optimised CCD
-270.7
-746.49
-240.83
-140.3
-1.9
-140.41
-296.32
-182.77
CCD
0.2
Final model
Design target
0.34
0.5
412
MSE20 (%)
19.9
19.9
15.2
7.6
Base
SS1
SS2
CCD
Base Optimal
F, ()
Base
Optimal
10
15
20
25
Li, ()
30
35
40
45
Fig. 26 Comparison of the F (Li) curves for base and optimised models
6. Conclusions
In this work, a DOE-RSM based design application to obtain a multi-link steering mechanism which gives
optimum deviation of V and steering error was developed
and applied on an MSC.Adams multibody model of a bus
IFS. In order to carry out the optimisation of the V, the
most effective parameters among the tie rod co-ordinates
on V deviation were first identified via DSA by using Adams/Insight multi-objective optimisation tool. The FFD
was used to determine the rank of importance of the coordinates of the tie rod hardpoints on V angle. Results
were evaluated by using MINITAB a practical statistical
software package. Since the FFD merely uses the high and
low values of the factors, it is not adequate to determine
the possible curvature of the response. In order to find out
the intermediate values of the parameters which give the
optimal tie rod position, CCD was also applied. In the final
stage of the study, geometry of the steering trapezoid
which gives the optimum MSE was determined via SS and
CCD. In order to do that the co-ordinate A10z was assumed
as the design constraint which determines the vertical position of the multi-link mechanism plane. Results obtained
from this study are summarised as follows:
1. Results of the DSA showed that for a multi-link
steering mechanism, the most effective factors among the
tie rod co-ordinates on V are the vertical components A9z
1. Timoney, E.; Timoney S. 2003. A review of the development of independent suspension for heavy vehicles, 2003 SAE International Truck and Bus Meeting
and Exhibition, SAE Technical Paper 2003-01-3433.
http://dx.doi.org/10.4271/2003-01-3433.
2. Reimpell, J. 1974. Fahrwerktechnik, Bd. 3, Wrzburg:
Vogel-Verlag, 177p.
3. Reimpell, J.; Stoll, H.; Betzler, J.W. 2002. The Automotive Chassis: Engineering Principles, Warrendale,
PA.: Society of Automotive Engineers, Inc.
4. Simionescu, P.A.; Beale, D. 2002. Optimum synthesis
of the four-bar function generator in its symmetric embodiment: the Ackermann steering linkage, Mechanism
and Machine Theory 37(12): 1487-1504.
http://dx.doi.org/10.1016/S0094-114X(02)00071-X.
5. Zhou, B.; Li, D.; Yang, F. 2009. Optimization design
of steering linkage in independent suspension based on
genetic algorithm, IEEE 10th International Conference
on Computer-Aided Industrial Design & Conceptual
Design: CAID&CD. Wenzhou, 45-48p.
http://dx.doi.org/10.1109/CAIDCD.2009.5374895.
6. Hanzaki, A.R.; Rao, P.V.M.; Saha, S.K. 2009. Kinematic and sensitivity analysis and optimization of planar rack-and-pinion steering linkages, Mechanism and
Machine Theory 44(1) 42-56.
http://dx.doi.org/10.1016/j.mechmachtheory.2008.02.0
14.
7. Oz, Y.; Ozan, B.; Uyanik, E. 2012. Steering system
optimization of a Ford heavy-commercial vehicle using
kinematic & compliance analysis, SAE Technical Paper 2012-01-1937.
http://dx.doi.org/10.4271/2012-01-1937.
413
8. Liang J.; Xin, L. 2012. Simulation analysis and optimization design of front suspension based on ADAMS,
Mechanika 18(3): 337-340.
http://dx.doi.org/10.5755/j01.mech.18.3.1873.
9. Kim, B.; M., Kim, J.W.; Moon, I.D.; Oh, C.Y. 2014.
Optimal combination of design parameters for improving the kinematics characteristics of a midsize truck
through design of experiment, Journal of Mechanical
Science and Technology 28(3): 963-969.
http://dx.doi.org/10.1007/s12206-013-1167-7.
10. Bian, X.L.; Song, B.A.; Becker, W. 2003. The optimisation design of the McPherson strut and steering
mechanism for automobiles, Forschung im Ingenieurwesen 68(1): 60-65.
http://dx.doi.org/10.1007/s10010-003-0107-6.
11. Bian, X.L.; Song, B.A.; Walter, R. 2004. Optimization of steering linkage and double wishbone suspension via R-W multibody dynamic analysis, Forschung
im Ingenieurwesen 69(1): 38-43.
http://dx.doi.org/10.1007/s10010-004-0136-9.
12. Minitab User's Guide 1: Data Graphics and Macros
(Release 13 for Windows), 1999. Pennsylvania State
University, USA.
13. MSC.Adams. 2002. Product Catalog. MSC. Software
Corporation.
14. Blundell, M.; Harty, D. 2006. The Multibody Systems
Approach to Vehicle Dynamics, London: Elsevier Butterworth Heinemann, 172p.
15. Matschinsky, W. 2007. Radfhrungen der Straenfahrzeuge. 3. aktualisierte und erweiterte Auflage, Berlin Heidelberg: Springer-Verlag, 96, 97p.
16. Kuiper, E.; Van Ooster, J.J.M. 2007. The PAC2002
advanced handling tire model, Vehicle System Dynamics: International Journal of Vehicle Mechanics and
Mobility, 45(1): 153-167.
http://dx.doi.org/ 10.1080/00423110701773893.
17. Montgomery, D.C. 2000. Design and analysis of experiments. Fifth Edition, Hoboken, New Jersey: John
Wiley & Sons, Inc. 275p.
18. Park, K.; Heo, S.J.; Kang, D.O.; Jeong, J.I.; Yi,
J.H.; Lee, J.H.; Kim, K.W. 2013. Robust design optimization of suspension system considering steering
pull reduction, International Journal of Automotive
Technology 14(6): 927-933.
http://dx.doi.org/10.1007/s12239-013-0102-3.
19. Han, H.; Park, T. 2004. Robust optimal design of
multi-body systems, Multibody system dynamics
11(2): 167-183.
http://dx.doi.org/10.1023/B:MUBO.0000025414.28789
.34.
20. Myers, R.H.; Montgomery, D.C.; Anderson-Cook,
C.M. 2009. Response Surface Methodology: Process
and Product Optimization Using Design of Experiments, Third Edition. Hoboken, New Jersey: John
ARTICLE IN PRESS
Abstract
This work illustrates simulation approach for optimizing the parametric design and performance of a 2-DOF RR planar
manipulator. Using dynamic and kinematic models of a manipulator different performance measures for the manipulator are obtained
for different combination of parameters with effect of noise incorporated to imitate the real time performance of the manipulator.
A novel approach has been proposed to model, the otherwise difcult to model, noise effects. The data generated during simulation for
various parameter combinations are utilized to analyze the statistical signicance of kinematic and dynamic parameters on performance
of manipulator using ANOVA technique. The parameter combinations, which give optimum performance measures obtained for
different points in workspace, are compared and reported.
r 2006 Elsevier Ltd. All rights reserved.
Keywords: Parametric design; Control factors; Noise factors; Positional error; Reliability
1. Introduction
Optimal design of manipulators, whose importance in
manufacturing is increasing day by day, has been a
challenge. Little success has been obtained in this area
because of difculties associated with geometrical constraints and complex models of the manipulator. Mostly
industrial manipulators are required to perform tasks with
a higher precision and speed than human beings. To
perform a task, a robot is commanded to move its endeffector to a specied position but the actual position
reached may be quite different from the desired one. This
difference in the actual and desired position for the endeffector is termed as positional error of a manipulator and
the average precision with which the manipulator moves its
arm to the commanded position is termed as its positional
accuracy [1]. The positional error for an industrial
manipulator may be 0.1 mm and repeatability as high as
10 mm [2]. Some studies relevant to the stochastic analysis
Corresponding author. Tel.: +91 1596 245073 225.
ARTICLE IN PRESS
240
B.K. Rout, R.K. Mittal / Robotics and Computer-Integrated Manufacturing 24 (2008) 239248
ARTICLE IN PRESS
B.K. Rout, R.K. Mittal / Robotics and Computer-Integrated Manufacturing 24 (2008) 239248
(1)
y l 1 S 1 l 2 S 12 ,
(2)
1
x2 y2 l 21 l 22
.
2l 1 l 2
(4)
(5)
and
x_ x y_ y
y_ 2
,
l 1 l 2 S2
8
2
2
Eqs. (7) and (8) are solved for y 1 and y 2 and are written in a
compact form as
be cd
y 1
ad b2
and
bc ae
y 2
,
ad b2
where a; b; c; d and e; are given as
m
1
1
m2 l 21 m2 l 22 m2 l 1 l 2 C 2 ,
a
3
3
2
l
1
b m2 2 l 1 l 2 C 2 ,
3 2
m
m2
1
m2 gl 1 C 1
gl 2 C 12 m2 l 1 l 2 S2 y_ 1 y_ 2
c
2
2
m2
2
l 1 l 2 S 2 y_ 2 t1 ,
2
(6)
d
_ y
_ represent end-effector velocity ~
where x;
ve with
x_ ve cos a and y_ ve sin a, and a is angle made by ~
ve
with positive x-axis of base frame.
vey
ve
vex
241
m2 2
l ,
3 2
(9)
(10)
(11)
(12)
13
(14)
m2
m2
2
l 1 l 2 S2 y_ 1
gl 2 C 12 t2 .
(15)
2
2
Eqs. (1)(15) are used to identify signicant factors
and compute the performance measures for the robotic
manipulator.
P(x,y)
m2
l2
Y
2
m1
1+2
l1
Various parameters inuencing the working of manipulator are identied with the help of a parameter diagram
(P-Diagram) for manipulator as shown in Fig. 2. The
parameters other than input and output are classied as
control factor (CF) and noise factor (NF).
1
X
ARTICLE IN PRESS
242
B.K. Rout, R.K. Mittal / Robotics and Computer-Integrated Manufacturing 24 (2008) 239248
Noise Factors
Environmental Conditions,
Eletrical Noise, Joint Friction,
Manufacturing and Assembly Errors
Input
Command for
work to be done
Response
Positional Error
Robot
Control Factors
Link Length,
Link Mass, Joint Torque
n
1X
i ,
n i1
(18)
ARTICLE IN PRESS
B.K. Rout, R.K. Mittal / Robotics and Computer-Integrated Manufacturing 24 (2008) 239248
(20)
243
ARTICLE IN PRESS
244
B.K. Rout, R.K. Mittal / Robotics and Computer-Integrated Manufacturing 24 (2008) 239248
4. Simulation
The numerical values used to simulate the performance
are given below:
(a) Number of levels for each control factor 2.
(b) Nominal values of six control factors at two levels and
standard deviations are given in Table 1.
(c) Number of combinations in factorial design 26 64.
(d) Design matrix containing 64 CF combinations is
provided in Appendix A.
(e) Chosen number of replications for each combination 8
(for factorial design) 100 (for reliability).
(f) Coordinates of target point in workspace: case I: x
0:40 m; y 0:30 m and case II: x 0:50 m, y 0:40 m.
(g) The step size of increment for search, range of search
and permissible error value e for the search algorithm
were chosen as: y1incr 0:01y1 , y2incr 0:01y2 0:5 y1
py1 p1:5 y1 and 0:5 y2 py2 p1:5 y2 , e 0:05.
(h) Tolerance selected for target point for computation of
reliability Dx 0:0005 m; Dy 0:0005 m:
Using the above numerical values simulation were
carried out. To simulate the performance for each factor
combination is run for eight replications for cases I and II.
The simulated performances are analyzed for statistical
analysis and its results are provided in Tables 2 and 3,
respectively. The performance measures, i.e. positional
error and SN ratio are computed for each parameter
combination and its trend are displayed in Figs. 3(a) and
3(b) for case I and Figs. 4(a) and 4(b) for case II. To
validate the results of factorial design, performance
measure reliability is computed after running the simulation for 100 cycles. The results for each combination are
displayed in Figs. 3(c) and 4(c) for the two cases,
respectively. After comparing the values of performance
measure, optimal combinations having optimal performance are presented in Tables 4 and 5.
Table 1
Values for control factors at two different levels
Control factors
Low level
High level
Standard deviation
l1 (m)
l2 (m)
m1 (kg)
m2 (kg)
t1 (N m)
t2 (N m)
0.40
0.25
5.5
4.0
500
100
0.50
0.35
6.5
5.0
800
105
0.0001
0.0001
0.01
0.01
0.1
0.1
ARTICLE IN PRESS
B.K. Rout, R.K. Mittal / Robotics and Computer-Integrated Manufacturing 24 (2008) 239248
Table 2
Analysis of variance of factorial design for case I
Source
Sum of
squares
DF
11.52
1
l1
47.33
1
l2
m1
3.154 105 1
m2
30.19
1
1.02
1
t1
t2
2.56
1
l1l2
1.13
1
m1 t1
3.53
1
4.03
1
m2 t1
l 1 l 2 m1 t1 t2
3.56
1
Residual
385.27
478
Corrected 506.35
511
sum total
Mean
square
F-value
Remark
11.52
3.67
Signicant
47.33
58.72
Signicant
3.154 105 3.913 105
30.19
37.46
Signicant
1.02
1.26
2.56
3.18
1.13
1.040
3.53
4.38
Signicant
4.03
5.00
Signicant
3.56
4.41
Signicant
0.81
Table 3
Analysis of variance of factorial design for case II
Source
Sum of
squares
DF
Mean
square
F-value
Remark
l1
l2
m2
t1
l1l2
l 1 t1
l 2 t1
l 1 l 2 t1
Residual
Corrected
sum total
193.54
415.76
44.59
39.04
47.08
147.30
11.03
80.47
2106.19
3085.01
1
1
1
1
1
1
1
1
503
511
193.54
415.76
44.59
39.04
47.08
147.30
11.03
80.47
4.19
46.22
99.29
10.65
9.32
11.24
35.18
2.63
19.22
Signicant
Signicant
Signicant
Signicant
Signicant
Signicant
Signicant
245
ARTICLE IN PRESS
B.K. Rout, R.K. Mittal / Robotics and Computer-Integrated Manufacturing 24 (2008) 239248
246
2.5
2
1.5
1
0.5
0
1
7 10 13 16 19 22 25 28 31 34 37 40 43 46 49 52 55 58 61 64
Combination number
2
64
61
58
55
52
49
46
43
40
37
34
31
28
25
22
19
16
13
10
S/N ratio
0
-2
-4
-6
-8
-10
Combination number
0.08
Reliability
0.06
0.04
0.02
61
57
53
49
45
41
37
33
29
25
21
17
13
Combination number
Fig. 3. Performance measures for case I: (a) mean positional error; (b) SN ratio; (c) reliability.
8
6
4
2
Combination number
58
55
46
43
40
37
34
31
28
25
22
19
16
13
10
0
1
-10
-15
0.08
Reliability
-20
0.06
Combination number
0.04
0.02
Combination number
64
61
58
55
52
49
46
43
40
37
34
31
28
22
19
16
13
10
0
25
S/N ratio
-5
64
61
55
52
58
52
49
49
46
43
40
37
34
31
28
25
22
19
16
13
10
0
1
ARTICLE IN PRESS
B.K. Rout, R.K. Mittal / Robotics and Computer-Integrated Manufacturing 24 (2008) 239248
Table 4
Optimum parameters for different performance measure for case I
Control factor
Value
(combination no.)
l1 (m)
l2 (m)
m1 (kg)
m2 (kg)
t1 (N m)
t2 (N m)
Reliability
Mean positional
error (m)
0.446933 (29)
0.071 (49)
0.00739 (29)
0.40
0.35
6.5
5
500
105
0.50
0.35
5.5
4
500
100
0.40
0.35
6.5
5
500
105
Table 5
Optimum parameters for different performance measure for case II
Control factor
Reliability
Mean positional
error (m)
Value
(combination no.)
1.66023 (48)
0.072 (22)
0.00889 (55)
l1 (m)
l2 (m)
m1 (kg)
m2 (kg)
t1 (N m)
t2 (N m)
0.50
0.25
6.5
5
800
105
0.40
0.35
5.5
5
500
105
0.50
0.35
5.5
5
800
100
247
l1 (m)
l2 (m)
m1 (kg)
m2 (kg)
t1
(N m)
t2
(N m)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.40
0.50
0.50
0.50
0.50
0.50
0.50
0.50
0.50
0.50
0.50
0.50
0.50
0.50
0.50
0.50
0.50
0.50
0.50
0.50
0.50
0.50
0.50
0.50
0.50
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.35
0.35
0.35
0.35
0.35
0.35
0.35
0.35
0.35
0.35
0.35
0.35
0.35
0.35
0.35
0.35
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.25
0.35
0.35
0.35
0.35
0.35
0.35
0.35
0.35
5.5
5.5
5.5
5.5
5.5
5.5
5.5
5.5
6.5
6.5
6.5
6.5
6.5
6.5
6.5
6.5
5.5
5.5
5.5
5.5
5.5
5.5
5.5
5.5
6.5
6.5
6.5
6.5
6.5
6.5
6.5
6.5
5.5
5.5
5.5
5.5
5.5
5.5
5.5
5.5
6.5
6.5
6.5
6.5
6.5
6.5
6.5
6.5
5.5
5.5
5.5
5.5
5.5
5.5
5.5
5.5
4
4
4
4
5
5
5
5
4
4
4
4
5
5
5
5
4
4
4
4
5
5
5
5
4
4
4
4
5
5
5
5
4
4
4
4
5
5
5
5
4
4
4
4
5
5
5
5
4
4
4
4
5
5
5
5
500
500
800
800
500
500
800
800
500
500
800
800
500
500
800
800
500
500
800
800
500
500
800
800
500
500
800
800
500
500
800
800
500
500
800
800
500
500
800
800
500
500
800
800
500
500
800
800
500
500
800
800
500
500
800
800
100
105
100
105
100
105
100
105
100
105
100
105
100
105
100
105
100
105
100
105
100
105
100
105
100
105
100
105
100
105
100
105
100
105
100
105
100
105
100
105
100
105
100
105
100
105
100
105
100
105
100
105
100
105
100
105
ARTICLE IN PRESS
B.K. Rout, R.K. Mittal / Robotics and Computer-Integrated Manufacturing 24 (2008) 239248
248
Table A1 (continued )
Combination
number
l1 (m)
l2 (m)
m1 (kg)
m2 (kg)
t1
(N m)
t2
(N m)
57
58
59
60
61
62
63
64
0.50
0.50
0.50
0.50
0.50
0.50
0.50
0.50
0.35
0.35
0.35
0.35
0.35
0.35
0.35
0.35
6.5
6.5
6.5
6.5
6.5
6.5
6.5
6.5
4
4
4
4
5
5
5
5
500
500
800
800
500
500
800
800
100
105
100
105
100
105
100
105
[13] Chen W, Allen JK, Tsui KL, Mistree F. A procedure for robust
design: minimizing variation caused by noise factors and control
factors. Trans ASME J Mech Des 1996;118:47885.
[14] Welch WJ, Buck RJ, Sacks J, Wynn HP, Mitchell TJ, Morris MD.
Screening, predicting and computer experiments. Technometrics
1992;34(1):1525.
[15] Chen W, Allen JK, Mavris D, Mistree F. A concept exploration
method for determining robust top level specications. Eng Optim
1996;26:13758.
[16] Chen W, Simpson TW, Allen JK, Mistree F. Use of design capability
indices to satisfy the range set of design requirements. ASME Design
Automation Conference, Irvine, CA, ASME paper 96-DETC/DAC1090.
[17] Simpson TW, Chen W, Allen JK, Mistree F. Conceptual design of a
family of products through the use of the robust concept exploration
method. 6th AIAA/USAF/NASA/ISSMO symposium on multidisciplinary analysis and optimization Bellevue, WA, 1996, AIAA,
2. p. 153545.
[18] Peplinski JD, Allen JK, Mistree F. Integrating product design with
manufacturing process design using robust concept exploration
method. Design theory and methodology conference, Irvine, CA,
ASME, Paper No.96-DETC/DTM-1502, 1996.
[19] Koch PN, Allen JK, Barlow A, Mistree F. Conguring turbine
propulsion systems using robust concept exploration. In: ASME
design automation conference. Irvine CA: ASME; 1996; Paper No.
96-DETC/DAC-1285.
[20] Koch PN, Allen JK, Mistree F, Mavris D. The problem of size in
robust design. ASME design automation conference, Sacramont, CA,
ASME, paper No. DETC97DAC3983, 1997.
[21] Guinta AA, Balabanov V, Haim D, Grossman B, Mason WH,
Watson LT. Wing design for a high speed civil transport using a
design of experiments methodology. 6th AIA/USAF/NASA/ISSMO
symposium on multidisciplinary analysis and optimization, Bellevue,
WA, AIAA vol. 1, 1996. p. 16883.
[22] Rout BK, Mittal RK. A review on minimizing variability in product
performance with the help of computational tools by controlling
active and noise factor for robust design. Proceedings of national
conference on current applications of computers in design engineering, J.N.V. University, Jodhpur, India, 2001. p. 1323.
[23] Rout BK, Mittal RK. Parametric robot manipulator design
optimization using taguchi method. Proceedings of 48th congress,
Indian Society for Theoretical and Applied Mechanics, IIT Guahati,
India, 2003. p. 1527.
[24] Rout BK, Mittal RK. Manipulator kinematic parameter selection
design of experiment approach. Conference proceeding of emerging
trends of concurrent engineering, N.I.T Rourkela, Orissa, India,
2005. p. 1209.
[25] Rout BK, Mittal RK. Parametric tolerance design of manipulator.
Proceedings of conference on emerging trends in mechanical
engineering, N.I.T. Kurukshetra, Haryana, India, 2005. p. 5561.
[26] Rout BK, Mittal RK. Modeling and optimization of A 2 DOF RR
planar manipulator using ANN. 14th ISME international conference,
Delhi College of Engineering, New Delhi, India, 2005. p. 4017.
[27] Mittal RK, Nagrath IJ. Robotics and control. New Delhi: Tata
McGraw Hill Publication; 2003.
[28] Park SH. Robust design and analysis for quality engineering.
London: Chapman & Hall Publication; 1998.
[29] Free down load software. Design Expert (DX) Version 6: http://
www.statease.com.
[30] Montgomery DC. Design and analysis of experiments, 5th Ed. New
York: Wiley; 2001.