applied

sciences
Article
Modeling and Identification of an Industrial Robot
with a Selective Modal Approach
Matteo Bottin 1 , Silvio Cocuzza 1, *, Nicola Comand 2 and Alberto Doria 1
1 Department of Industrial Engineering, University of Padova, 35131 Padova, Italy;
matteo.bottin@unipd.it (M.B.); alberto.doria@unipd.it (A.D.)
2 Department of Management and Engineering, University of Padova, 36100 Vicenza, Italy;
nicola.comand@phd.unipd.it
* Correspondence: silvio.cocuzza@unipd.it; Tel.: +39-049-827-6793




Received: 29 May 2020; Accepted: 28 June 2020; Published: 3 July 2020


Abstract: The stiffness properties of industrial robots are very important for many industrial
applications, such as automatic robotic assembly and material removal processes (e.g., machining
and deburring). On the one hand, in robotic assembly, joint compliance can be useful for compensating
dimensional errors in the parts to be assembled; on the other hand, in material removal processes,
a high Cartesian stiffness of the end-effector is required. Moreover, low frequency chatter vibrations
can be induced when low-stiffness robots are used, with an impairment in the quality of the machined
surface. In this paper, a compliant joint dynamic model of an industrial robot has been developed,
in which joint stiffness has been experimentally identified using a modal approach. First, a novel
method to select the test configurations has been developed, so that in each configuration the mode
of vibration that chiefly involves only one joint is excited. Then, experimental tests are carried out
in the selected configurations in order to identify joint stiffness. Finally, the developed dynamic
model of the robot is used to predict the variation of the natural frequencies in the workspace.

Keywords: robot; compliance; machining; modes of vibration; impulsive testing

1. Introduction
In robotic processes, the compliance of the robot arm plays a very important role. In some
conditions, for example, in robotic assembly, robot arm compliance can compensate for small position
and orientation errors of the end-effector. In other processes, like machining, robot compliance may
generate chatter vibrations, with an impairment in the quality of the machined surface. Indeed,
still very few robots have been applied in this economically important application area [1], mainly
due to their low stiffness [2]. In industrial robots, the compliance of the end-effector is mainly due
to joint compliances [3–5], even if there are some examples of robots [6] having structural modes
(dominated by link or bearing compliance) in the band of frequency that contains the modes dominated
by joint compliance. The compliance properties of robots are very important in emerging fields
as well, such as flexible assembly systems [7,8] and collaborative robotics [9]. Once robot joint stiffness
has been identified, the kinematic redundancy of the arm with respect to the robotic task can be
exploited to minimize (or maximize) the Cartesian compliance of the end-effector through optimization
methods [10–13], as already proposed for the minimization of base reactions and kinetic energy.
In material removal processes, low stiffness causes imprecise products, due to the robot deflections
during the robotic task. From a dynamic point of view, low frequency chatter vibrations [3,14,15] can
be induced when low-stiffness robots are used, with an impairment in the quality of the machined
surface. Moreover, vibrations cause a reduction of tool life and can lead to breakages in robot joint

Appl. Sci. 2020, 10, 4619; doi:10.3390/app10134619 www.mdpi.com/journal/applsci

Appl. Sci. 2020, 10, 4619 2 of 20

transmission. In particular, a low joint stiffness of the robot is one of the main issues in using robotic
machining instead of computer numerical control (CNC) machining.
In robotic processes, the directions along which the robot arm shows large compliance are
important. The stiffness matrix in the joint space does not directly give this piece of information;
moreover, for a serial robot, the stiffness matrix in the Cartesian space is not diagonal and it is
configuration-dependent. This means that the force and deformation in the Cartesian space are
coupled and this can generate some counterintuitive results. A geometric and intuitive interpretation
of the end-effector displacements in different robot operative conditions has been presented in [16].
Static tests are widely used to obtain the joint stiffness of industrial robots [17]. In the static tests,
a set of forces is applied to the robot end-effector in different robot configurations, while the displacement
sensors (laser sensors, vision systems, coordinate measuring machines) measure the static deformation
of the end-effector. Therefore, the Cartesian stiffness of the robot can be calculated. Then, the stiffness
of the joints is obtained through the analytical relation between the joint and Cartesian stiffness based
on the kinematic model of robot. Various identification methods (least squares or genetic algorithms)
are employed to numerically solve the problem.
Most of the researches (e.g., see [4,18–21]) neglect link flexibility and identify the joints’ rotational
stiffness (modeled with linear rotational springs) using the above-mentioned method. In [22], the robot
arm is modeled considering rigid links and three lumped rotational springs for each joint to take
into account joint compliance, bearings compliance, and link deflections. In [23], an analysis method
in which both joint and link stiffness are considered is presented. In [24,25], two types of robot
dynamic models are presented by considering either only joint compliance or its combination with
link compliance, in order to predict the robot dynamic behavior in different postures. In [6] a hybrid
approach is adopted, in which a critical structural compliance in a direction perpendicular to a joint
axis is taken into account, together with joint compliance.
In [26], experimental modal analysis is used to identify the joint and base stiffness of an industrial
robot represented with a four-degrees-of-freedom (four-DOF) planar model. Experimental modal
analysis was also used in [6,16], to identify the stiffness of the first three joints of an industrial robot.
The static tests for the identification of joint stiffness have some drawbacks. The equipment for
the measurement of end-effector displacements along three directions may be complex and expensive,
especially if high accuracy is required and the workspace of the robot is large. The application of vertical
forces on the end-effector is rather simple, since weights can be used, but the application of horizontal
forces may require specific actuators that have to be connected with the end-effector without introducing
additional constraints. Since Cartesian stiffness is configuration dependent, many test configurations
are needed to map the robot workspace. The identification of joint stiffness from the measured Cartesian
stiffness requires, in most cases, an optimization method. Finally, the measurements are affected by
the structural compliance of the links, and it is difficult to separate the contribution of end-effector
displacement caused by joint compliance from the one due to link compliance.
For the above-mentioned reasons, this research aims to use the methods of impulsive modal
analysis to identify the joint stiffness of industrial robot arms. In future, the same methodology could
be extended to other kinds of robots e.g., walking robots [27]. Some factors foster the development of
testing methods based on impulsive excitation. The basic equipment (which includes an instrumented
hammer, an accelerometer, a data acquisition board, and a computer) is nowadays affordable for most
laboratories. Hammer excitation can be exerted in three directions, even if the robot workspace is wide.
Modal analysis identifies all the modes of vibration, both the ones dominated by joint compliance,
and the ones dominated by link compliance (if present), hence, the modal method allows a better
understanding of the importance of link compliance than the static method.
The modes of vibration of a robot typically involve the rotations of some joints and the natural
frequencies are functions of the stiffness of these joints. Therefore, the main issue of the modal methods
for the identification of joint stiffness is the possibility of finding robot configurations with modes of
vibration dominated by the stiffness of only one joint. In this specific case, modal stiffness coincides with
Appl. Sci. 2020, 10, x FOR PEER REVIEW 3 of 19

The modes of vibration of a robot typically involve the rotations of some joints and the natural
frequencies
Appl. are4619
Sci. 2020, 10, functions of the stiffness of these joints. Therefore, the main issue of the modal 3 of 20
methods for the identification of joint stiffness is the possibility of finding robot configurations with
modes of vibration dominated by the stiffness of only one joint. In this specific case, modal stiffness
joint stiffness,
coincides withand jointthestiffness,
latter can be the
and identified frombethe
latter can measured
identified natural
from frequency natural
the measured and the frequency
calculated
value of moment of inertia. Other important issues are the choice of
and the calculated value of moment of inertia. Other important issues are the choice of proper proper excitation directions for
the selected
excitation configurations
directions for theand the minimization
selected configurations of and
test configurations.
the minimization of test configurations.
This paper is organized as follows. In Section
This paper is organized as follows. In Section 2, a serial 2, a serial six-DOF
six-DOF robot
robot isis considered
considered (Omron(Omron
Adept Viper s650), since it is representative of industrial robots used
Adept Viper s650), since it is representative of industrial robots used for a wide range of for a wide range of operations,
operations,
and aa dynamic
and dynamic modelmodel of of robot
robot vibrations
vibrations is is developed. Section 33 deals
developed. Section deals with
with selective
selective modal
modal testing,
testing,
and some criteria for finding robot configurations with modes of vibration dominated by onlyjoint,
and some criteria for finding robot configurations with modes of vibration dominated by only one one
or by a
joint, orcouple of joints,
by a couple of are presented
joints, and discussed.
are presented Experimental
and discussed. tests carried
Experimental tests out in the
carried outselected
in the
configurations
selected are presented
configurations are in Section 4,inwhich
presented Sectionsignificantly
4, which extends a previous
significantly extendswork of the authors
a previous work [16],
of
in which only the first three joints of the robot were considered. Section 5 deals
the authors [16], in which only the first three joints of the robot were considered. Section 5 deals with with the identification of
jointidentification
the stiffness, andof anjoint
optimization
stiffness, technique is presentedtechnique
and an optimization to cope with the cases in
is presented to which
cope withthe identified
the cases
mode is dominated by a couple of joints. The use of the robot dynamic
in which the identified mode is dominated by a couple of joints. The use of the robot dynamic model model with the identified
stiffness
with thevalues
identifiedis presented
stiffness invalues
Section is 6, and the variation
presented in Sectionof 6,theand natural frequencies
the variation of in
thethe robot
natural
workspace
frequenciesisinanalyzed.
the robot Finally,
workspace in Section 7, conclusions
is analyzed. Finally, in areSection
drawn.7, conclusions are drawn.

2. Dynamic
2. Dynamic ModelModel
In the
In the framework
frameworkof ofthis
thisresearch
researchananAdept
AdeptViper s650
Viper robot
s650 manufactured
robot manufactured by Omron
by Omron Adept Inc.
Adept
(Pleasanton, CA, USA) is considered. It is a medium size six-DOF industrial robot, its
Inc. (Pleasanton, CA, USA) is considered. It is a medium size six-DOF industrial robot, its workable workable space
is roughly
space a hemisphere
is roughly with a with
a hemisphere radius of about
a radius of 600 mm.
about 600Figure 1 shows
mm. Figure 1 that
showsthis robot
that thishas a typical
robot has a
anthropomorphic
typical structure.
anthropomorphic There are
structure. threeare
There main joints
three mainthat define
joints thedefine
that position
theof the wrist
position of and three
the wrist
minor joints that define the orientation of the end-effector. All joints are driven by
and three minor joints that define the orientation of the end-effector. All joints are driven by alternate current
(AC) servomotors.
alternate current (AC) servomotors.

Figure
Figure 1.
1. The
The robot
robot Adept
Adept Viper
Viper s650
s650 with the tested
with the tested axes.
axes.

The
The vertical
vertical plane
plane passing
passing through
through the
the axis
axis of
of joint
joint 1,
1, perpendicular
perpendicular to
to the
the axes
axes of
of joint
joint 22 and
and 3,
3,
is named the meridian plane of the robot arm.
is named the meridian plane of the robot arm.
Appl. Sci. 2020, 10, 4619 4 of 20

This paper aims at developing a dynamic model for the study of vibrations around a working
configuration of the robot defined by vector q0 of joint variables, therefore joint variables, velocities,
and accelerations are defined by the following equations:

q = q0 + ∆q
. .
q = ∆q (1)
.. ..
q = ∆q

where vector ∆q contains the small variations in joint variables around the selected configuration.
Joint compliance around the rotation axis is considered, whereas link and bearing compliance
is neglected.
The axis of joint 6 coincides with the approach axis of the end-effector. Since, in most cases,
the external force exerted on the end-effector is aligned with or intersects the approach axis, compliance
about joint 6 is neglected.
If the non-linear Coriolis and centrifugal terms [28] are neglected, according to the assumption of
small oscillations, the equations of free vibrations take the form [6]:
  .. .
M q0 ∆q + Cq ∆q + K∆q = 0 (2)
   
where M q0 is the mass matrix that depends on q0 . The elements on the diagonal of M q0 are
the direct inertia terms, the off-diagonal terms are the inertial cross-coupling terms.
Cq is a diagonal matrix that accounts for joint damping, K is a stiffness matrix that accounts for
joint compliance and configuration-depending gravity torques. Gravity can generate large restoring
or unbalancing torques about joint axes only when the robot arm is close to a vertical configuration
(upwards or downwards).
If configuration-depending gravity torques are negligible the equations of free vibrations become:
  .. .
M q0 ∆q + Cq ∆q + Kq ∆q = 0 (3)

where Kq is a diagonal matrix that accounts for joint compliance. Specific calculations have been be
carried out to estimate the effect of gravity torques in the selected configurations.
The identification of matrices Kq and Cq is the main task of this research.
 
The mass matrix M q0 can be calculated using the inertial data provided by the computer
aided design (CAD) models of the robot. These CAD files have been retrieved from the website
of the manufacturer and the links are assumed to have uniform density and no free spaces
within the external shell. This is due to the uncertainty of the mass distribution within the links of
the robot that could not be eliminated without a complete disassembly of the robot.
Starting from the CAD files, the following inertial parameters have been retrieved:

• Mass of link i (mi ) has been calculated by distributing the total mass mtot based on the volume of
each link;
• Center of mass of link i (Gi ) has been calculated with respect to the corresponding
Denavit-Hartenberg reference frame (see Figure 2);
• Inertia tensor of link i (Ii ) has been calculated at the center of mass aligned with the correspondent
Denavit-Hartenberg reference frame [28].
Appl. Sci. 2020, 10, x FOR PEER REVIEW 5 of 19

Appl. Sci. 2020, 10, 4619 5 of 20

Figure 2. Robot reference frames.

Figure 2. Robot reference frames.  
On the basis of the above-mentioned parameters mass matrix M q0 can be calculated as follows:
3. Selective Modal Testing 5 
  X 
T T
M q0 = aboutmmechanical
Vibration signals contain information Ri Ii RTi JO,i
i Jp,i Jp,i + JO,isystems [29], and this paper aims to use(4)
impulsive modal analysis to identify the stiffness i=1 properties of an industrial robot, and to develop a
compliant robot model.
where Ri is the rotation matrix from link i to the base frame and Jp,i and JO,i are the Jacobians:
Modal analysis is a mature technology for the study of vibrations [30], which has been
successfully used in many fields Jp,i of
=aerospace
[z1 × (Gi − [31],
O )vehicle,
· · · zi × (and
Gi −automation
Oi ) 0 · · · 0] engineering [16,32,33].
h 1
In the field of industrial robotics, it is Jwellz recognized that joint stiffness has a larger effect on(5)
i
O,i 1 · · · z j 0 · · · 0
robot performance than link stiffness, therefore the main task of impulsive modal analysis is the
identification
3. Selective of the stiffness
Modal Testingof the joints.
A robot with assigned joint values (𝒒) is considered. When an impulse force is exerted on a
Vibration
point of the robotsignals contain
by means ofinformation
a hammer about mechanical
for modal testing systems [29], and thiscomponents
and 𝑁 acceleration paper aims to use
are
impulsivebymodal
measured means analysis to identify thea stiffness
of accelerometers, set of 𝑁properties
modes ofofvibrationsan industrial can robot, and to develop
be identified. These a
compliant robot model.
modes of vibration transform the equations of motions (3) into a set of 𝑁 independent equations in
Modal
the modal analysis is𝜂a mature
coordinates 𝑘 = 1, …technology
, 𝑁: for the study of vibrations [30], which has been successfully
used in many fields of aerospace [31], vehicle, and automation engineering [16,32,33].
𝑀∗ of industrial
In the field 𝒒) 0 robotics, it0 is well𝜂recognized 𝐶 ∗ 𝒒) that joint 0 stiffness 0 𝜂
has a larger effect on robot
∗ ∗
0 𝑀 𝒒)
performance than link stiffness, therefore 0 𝜂 + 0 𝐶 𝒒) 0 𝜂
the main task of impulsive modal∗analysis is the identification
0 0 𝑀∗ 𝒒) 𝜂 0 0 𝐶 𝒒) 𝜂
of the stiffness of the joints.
𝐾 ∗ 𝒒) 0 0 𝜂 0 (6)
A robot with assigned joint values (q) is∗ considered. When an impulse force is exerted on a point
+ 0 𝐾 𝒒) 0 𝜂 = 0
of the robot by means of a hammer for modal testing and N acceleration components are measured by
0 0 𝐾∗ 𝒒) 𝜂 0
means of accelerometers, a set of N modes of vibrations can be identified. These modes of vibration
where the mass, damping and stiffness matrices are diagonal, and elements 𝑀∗ 𝒒), 𝐶 ∗ 𝒒) and
𝐾 ∗ 𝒒) represent modal mass, damping and stiffness associated to the 𝑘th mode.
When the robot is tested in a generic configuration the modes of vibration typically involve
rotations about many joints of the robot. Therefore, modal coordinate 𝜂 is a combination of joint
Appl. Sci. 2020, 10, 4619 6 of 20

transform the equations of motions (3) into a set of N independent equations in the modal coordinates
ηk k = 1, . . . , N:
 ..    .
η  η
 
 M∗11 (q) 0 0   C∗11 (q) 0 0
 .. 1   .1
  
  
ηk  ηk
  
0 M∗kk (q) 0 + 0 C∗kk (q) 0
     
  
..  η.
   
  
M∗NN (q)  ηN 
  
0 0 0 0 C∗NN (q)
 
N (6)
η
    
 K∗11 (q) 0 0   1
  0 
   
ηk
  
 
+ 0 K∗kk (q) 0  = 0
  

 
 
 

K∗NN (q)  ηN
   
 0 
0 0  

where the mass, damping and stiffness matrices are diagonal, and elements M∗kk (q), C∗kk (q) and K∗kk (q)
represent modal mass, damping and stiffness associated to the kth mode.
When the robot is tested in a generic configuration the modes of vibration typically involve
rotations about many joints of the robot. Therefore, modal coordinate ηk is a combination of joint
rotations, which is defined by the corresponding mode of vibration. Modal mass, damping and stiffness
do not have a direct physical meaning, and are not very useful to identify the stiffness and damping
characteristics of the robot joints.
Conversely, when the same robot is modally tested in a specific configuration in which the mass
matrix shows that a joint variable i has negligible inertial cross coupling terms, one of the modes of
vibration is dominated by the rotation about joint i, and the corresponding modal equation becomes:
.. .
Iizz (q)qi + ci qi + ki qi = Fi (t) (7)

The modal coordinate ηi coincides with joint variable qi , modal mass coincides with inertia
about joint i (Iizz (q)), modal stiffness and damping coincide with joint stiffness (ki ) and damping (ci ),
respectively. Mode i is very useful to identifying the stiffness of joint i.
Inertia about joint i (Iizz (q)) depends on the inertias of the following links and can be calculated by
using the Huygens–Steiner theorem:

N 
X 
Iizz (q) = R ji Ii RTji + m j P ji (8)
j=i

 2
 y ji + z2ji

−x ji y ji −x ji z ji 
 
P ji =  −y ji x ji
 x2ji + z2ji −y ji z ji  (9)
 
 −z x
ji ji −z ji y ji x2 + y2 
ji ji

where R ji is the rotation matrix that transforms the reference frame of link j to the reference frame of
link i, and xl ji , yl ji and zl ji are the coordinates of the center of mass of link j defined in the reference
frame of link i.
It is worth noticing that usually mode i (dominated by joint i) includes small contributions
of the other joints, and sometimes of structural compliance of the links, therefore assuming
M∗ii (q) = Iizz (q), K∗ii (q) = ki and C∗ii (q) = ci is only an approximation.
For the above-mentioned reasons, a specific modal testing method is proposed, which is based on
robot configurations and excitation directions that make possible the excitation of modes of vibrations
dominated by only one joint.
In order to have an insight into the properties of the mass matrix of the tested robot, the elements of
the mass matrix were evaluated in random configurations. In particular 1,000,000 random combinations
of q1 . . . q5 with values within the actual joint ranges were considered.
Since the inertial cross-coupling terms may be positive or negative, depending on the configuration,
the mean values and standard deviations of the moduli of the elements of the mass matrix were
calculated to avoid cancellations.
Appl. Sci. 2020, 10, 4619 7 of 20

Appl. Sci. 2020, 10, x FOR PEER REVIEW 7 of 19

The results of this analysis are summarized in Figure 3, and show the following properties.

Figure 3. Mean values and standard deviations of mass matrix terms (106 random configurations),
Figure 3. Mean values and standard deviations of mass matrix terms (10 random configurations),
direct inertia terms (red bars) and inertial cross-coupling terms (blue bars).
direct inertia terms (red bars) and inertial cross-coupling terms (blue bars).
(a) The direct inertia terms of the first three joints are by far larger than the ones of the wrist
In a specific that
joints—notice configuration, the scale
a logarithmic inertial cross-coupling
is adopted terms
in Figure of joint
3. Joint 𝑖 (𝑀
2 (J2) has ,the
𝑗 ≠ largest
𝑖) are small
meanor
negligible if the velocity of
value followed by Joint 1 (J1).joint 𝑖 generates a distribution of robot arm velocities that cannot be
(b)generated
The directby inertia
the velocities of the
term of joint i (Mother joints. Even if the previous condition is satisfied,
ii ) depends on the joint variables of the following links,
cross-coupling terms may be present, if the
hence M11 depends on q2 . . . qN , whereas MNN axes of the joints areTherefore,
is constant. not principal
theaxes.
standard deviation
According to the results
decreases from J1 to J5. of Figure 3, the configurations for selective modal testing summarized
in Figure 4 and Table 1 were chosen.
(c) The inertial cross-coupling terms between the first three joints are much smaller than the direct
inertia terms with the exception of M23 , which is comparable with M33 .
(d) M23 has a large standard deviation, and this means that there are robot configurations with a
minimum coupling between J2 and J3.
(e) The mean values of the inertial cross-coupling terms of the first three joints with the wrist joints
are very small in comparison with the direct inertia terms of the first three joints. They are more
important if compared with the direct inertia terms of the wrist joints
(f) There is a negligible coupling between wrist joints (M45 ).

In a specific configuration, the inertial cross-coupling terms of joint i (Mij , j , i) are small or
negligible if the velocity of joint i generates a distribution of robot arm velocities that cannot be
generated by the velocities of the other joints. Even if the previous condition is satisfied, cross-coupling
terms may be present, if the axes of the joints are not principal axes.
According to the results of Figure 3, the configurations for selective modal testing summarized
in Figure 4 and Table 1 were chosen.

Figure 4. Robot configurations.

 In a specific configuration, the inertial cross-coupling terms of joint 𝑖 (𝑀 , 𝑗 ≠ 𝑖) are small or
negligible if the velocity of joint 𝑖 generates a distribution of robot arm velocities that cannot be
generated by the velocities of the other joints. Even if the previous condition is satisfied,
cross-coupling terms may be present, if the axes of the joints are not principal axes.
According to the results of Figure 3, the configurations for selective modal testing summarized
Appl. Sci. 2020, 10, 4619 8 of 20
in Figure 4 and Table 1 were chosen.

Figure 4. Robot configurations.

Figure 4. Robot configurations.
Table 1. Robot testing configurations.

Configuration Joint 1 [◦ ] Joint 2 [◦ ] Joint 3 [◦ ] Joint 4 [◦ ] Joint 5 [◦ ] Joint 6 [◦ ]

Test 1 0 0 90 0 0 0
Test 2 0 0 17 0 −17 0
Test 3 0 −189 243 0 −54 0
Test 4 0 0 90 0 90 0
Test 5 0 −90 180 0 90 0

Joint 1 (J1) is always perpendicular to joints 2 and 3 (J2 and J3). If joint variables q2 , q3 are set to 0◦
and 90◦ , respectively, the robot arm is horizontal, and if q4 q5 are set to 0◦ , the approach axis is horizontal
as well. In this configuration (Test 1) only J1 is able to generate relevant velocities in the horizontal
plane. This configuration minimizes the inertial coupling between J1 and the other joints (which is
usually small) and enhances the excitation of J1, when a lateral force (in y direction) is exerted on
the end-effector.
Joints J2 and J3 are always parallel, and generate velocity components in the same plane,
Figure 3 shows that the inertial cross coupling may be large. The mass matrix of a robot with two
parallel joints [28] shows an inertial cross-coupling term depending on joint variable q3 (the angle
between the links). A numerical calculation of the ratios between the inertial cross-coupling term M23
and the direct inertias of J2 and J3 (M22 and M33 respectively) is depicted in Figure 5. They show
that the inertial cross-coupling term tends to decrease when the two links tend to overlap, notice
that the two links are aligned when q3 = 90◦ . Minimum coupling configurations can be found for
q3 = −38◦ and q3 = 293◦ . These theoretical values have to be compared with the actual range of
the joint (−29◦ ≤ q3 ≤ 256◦ ), and, for this reason, a configuration with q3 = 243◦ (Test 3) is suited to
identifying the stiffness both of J2 and J3. It is worth noticing that, in this configuration, a vertical
force generates a large moment about J3, thus, it is able to excite the mode dominated by this joint.
Conversely, in this configuration a vertical force produces a small moment about J2, thus, the excitation
of the mode dominated by this joint is poor. Figures 3 and 5 show that the inertial cross-coupling
term is less important for J2 than for J3, since the direct inertia M22 term is much larger than M23 .
For this reason, configuration Test 2 of Table 1, in which links 2 and 3 are almost perpendicular,
gives a good compromise between the reduction of inertial cross coupling and the generation of a large
moment about J2.
 large moment about J3, thus, it is able to excite the mode dominated by this joint. Conversely, in this
configuration a vertical force produces a small moment about J2, thus, the excitation of the mode
dominated by this joint is poor. Figures 3 and 5 show that the inertial cross-coupling term is less
important for J2 than for J3, since the direct inertia 𝑀22 term is much larger than 𝑀23 . For this
reason, configuration Test 2 of Table 1, in which links 2 and 3 are almost perpendicular, gives a good
Appl. Sci. 2020, 10, 4619
compromise between the reduction of inertial cross coupling and the generation of a large moment9 of 20
about J2.

Figure
Figure 5. Effect
5. Effect of joint
of joint variableq3𝑞3ononinertial
variable inertialcross
cross coupling
coupling between
betweenJ2J2
and J3.J3.
and

Joint J4 is able to generate velocity components perpendicular to the vertical plane that contains
the robot arm (meridian plane). J2, J3 generate velocity components in meridian plane, and the inertial
cross coupling with J4 is small. Conversely, J1 is able to generate velocity components in the same
direction as J4, and there is a significant cross coupling with J4 (see Figure 3). The inertial cross coupling
between J4 and J5 is always negligible (see Figure 3). In order to enhance the moment that a lateral hit
can generate about J4, q5 was set to ±90◦ and configuration Test 4 was defined.
Finally, when q4 is set to zero, Joint 5 (J5) is able to generate velocity components of the last links
in the meridian plane of the robot arm, but these components can be generated by J2 and J3 as well.
When q4 is set to 90◦ , J5 is able to generate velocity components out of the meridian plane of the robot
arm, but these components can be generated by J1 as well. Therefore, Figure 3 shows that there is
always a significant coupling between J5 and the arm joints. The test configuration can be chosen only
with the aim of increasing the moment about J5 generated by the hammer hit. Test configuration Test 5
satisfies this condition for a hammer hit in the longitudinal direction.
The relevance of gravity restoring (or unbalancing) torques was evaluated in the selected
configurations. The calculated values resulted much smaller than reference values of joint torques
caused by joint stiffness [16].

4. Experimental Tests
The equipment needed to carry out the modal analysis of a robot with the impulsive method
is rather simple and cheap. It includes a hammer for modal testing, an accelerometer, and a data
acquisition board. In the framework of this research, a PCB 086C01 hammer (with load cell sensitivity
0.2549 mV/N), a PCB 356A17 tri-axial accelerometer (sensitivity 50.5 mV/(m/s2 )), and a NI9234 data
acquisition board were used. The hammer and the accelerometer were made by PCB Piezotronics
Inc., Depew, NY, USA, whereas the board was made by National Instruments Inc., Austin, TX, USA.
Measured signals were analyzed in the frequency domain by means of ModalVIEW R2 (2017), which
is a specific software for modal analysis developed by ABSignal-National Instruments Inc., Austin,
TX, USA.
In order to characterize the vibrations of the robot 10 testing points were defined. Figure 6 shows
that testing point 1 is located on the flange of the end-effector, testing point 10 is located on the base,
before joint 1, and the other testing points are evenly distributed on the links. The number of testing
points is a trade-off between the need for a large number of measurement locations, to accurately
describe robot vibrations, and the need for a quick testing procedure. Modal analysis was carried
out with the rowing response approach [34] Excitation was always exerted on the end-effector,
 In order to characterize the vibrations of the robot 10 testing points were defined. Figure 6
shows that testing point 1 is located on the flange of the end-effector, testing point 10 is located on
the base, before joint 1, and the other testing points are evenly distributed on the links. The number
of testing points is a trade-off between the need for a large number of measurement locations, to
accurately describe robot vibrations, and the need for a quick testing procedure. Modal analysis was
Appl. Sci.carried
2020, 10,out
4619with the rowing response approach [34] Excitation was always exerted on the 10 of 20
end-effector, whereas the tri-axial accelerometer was sequentially moved to the 10 testing points.
The three axes of the accelerometer were always parallel or perpendicular to the robot surface.
whereas the tri-axial accelerometer was sequentially moved to the 10 testing points. The three axes of
the accelerometer were always parallel or perpendicular to the robot surface.

Figure 6. Measurement points for modal testing.

Figure 6. Measurement points for modal testing.
Since motions of the base of the robot may have a negative effect on the quality of measurements,
the robot was rigidly fastened to a large steel base.
Appl. Sci. 2020, 10, x FOR PEER REVIEW 10 of 19
After some preliminary tests, data acquisition was carried out with a sampling frequency of
1024 Hz and 2048Since samples.
motions of the base of the robot may have a negative effect on the quality of
measurements,
Figure 7 deals with the robot was rigidly fastened
the experimental to a large
results steel base.
obtained in configuration Test 1, with hammer
After some preliminary tests, data acquisition was carried out with a sampling frequency of
excitation in the lateral direction (y). Plots (a) and (b) show the modulus and phase of the direct
1024 Hz and 2048 samples.
point frequency Figureresponse
7 deals function (FRF), which
with the experimental is the
results FRF in
obtained measured in the
configuration Testexcitation point between
1, with hammer
the acceleration
excitation in in
thethedirection of the hammer
lateral direction force
(y). Plots (a) and and the hammer
(b) show the modulus force.
and Plot
phase(c)
of shows
the directthe complex
point frequency response function (FRF), which is the FRF measured
mode indicator function (CMIF) calculated over the 30 FRFs [30]. Both the modulus of the FRFin the excitation point between
the acceleration in the direction of the hammer force and the hammer force. Plot (c) shows the
and the CMIF show a high and isolated peak at about 13 Hz. The phase of the direct point FRF shows a
complex mode indicator function (CMIF) calculated over the 30 FRFs [30]. Both the modulus of the
large phase
FRFchange
and theat the show
CMIF samea frequency. These
high and isolated properties
peak at about 13are
Hz. hints of the
The phase presence
of the of aFRF
direct point well-defined
mode of vibration
shows a large at phase
aboutchange
13 Hz. ModalVIEW
at the made
same frequency. it possible
These to identify
properties are a presence
hints of the mode ofofvibration
a at
well-defined
12.8 Hz with a damping moderatioof vibration
of 2.3% at [30].
about This
13 Hz.mode,
ModalVIEWwhich made it possible to identify
is represented a mode ofof Figure 7,
at the bottom
vibration at 12.8 Hz with a damping ratio of 2.3% [30]. This mode, which is represented at the
is dominated by the rotation of the whole robot arm about J1.
bottom of Figure 7, is dominated by the rotation of the whole robot arm about J1.

Figure 7. Test 1 configuration, (a) direct point frequency response function (FRF) modulus, (b) phase,
Figure 7. Test 1 configuration, (a) direct point frequency response function (FRF) modulus, (b) phase,
(c) complex mode indicator function (CMIF).
(c) complex mode indicator function (CMIF).
Figure 8 shows the experimental results obtained in configuration Test 2 with hammer
excitation in the vertical direction (z). In this case, the direct point FRF and the CMIF show a high
and isolated peak at about 17 Hz. ModalVIEW made it possible to identify at 17.5 Hz a mode of
vibration dominated by the rotation of the whole robot arm about J2, which is represented in Figure
8. The corresponding damping ratio is 1.9%, this value is congruent with the damping ratio of J1.
Appl. Sci. 2020, 10, 4619 11 of 20

Figure 8 shows the experimental results obtained in configuration Test 2 with hammer excitation
in the vertical direction (z). In this case, the direct point FRF and the CMIF show a high and isolated
peak at about 17 Hz. ModalVIEW made it possible to identify at 17.5 Hz a mode of vibration dominated
by the rotation of the whole robot arm about J2, which is represented in Figure 8. The corresponding
damping ratio is 1.9%, this value is congruent with the damping ratio of J1.
Appl. Sci. 2020, 10, x FOR PEER REVIEW 11 of 19

Figure 8. Test 2, (a) direct point frequency response function (FRF) modulus, (b) phase, (c) complex
Figure 8. Test 2, (a) direct point frequency response function (FRF) modulus, (b) phase, (c) complex
mode indicator function (CMIF).
mode indicator function (CMIF).
The experimental results obtained in configuration Test 3 are summarized in Figure 9. The
The experimental results obtained in configuration Test 3 are summarized in Figure 9. The direct
direct point FRF shows a high peak at about 30 Hz with a large phase change, but in this case the
pointmain
FRF shows
peak isanot high peaksince
alone, at about
there30isHz with apeak
another large
at phase
about change,
35 Hz. The butCMIF
in thisplot
caseconfirms
the mainthispeak is
not alone, since there is another peak at about 35 Hz. The CMIF plot confirms
scenario. ModalVIEW made it possible to identify two modes. The first mode is at 30.8 Hz with a this scenario. ModalVIEW
madedamping
it possible
ratiotoofidentify
2.1%. This two modes.
mode, whichThe first modeinisFigure
is represented at 30.8 Hz
9, is with a damping
dominated ratio of 2.1%.
by the compliance
This mode, which
of J3, and showsissome
represented in Figure
contribution 9, is dominated
of J2, owing by the compliance
to the inertial-cross coupling betweenof J3,the
andtwoshows
joints.some
The second
contribution mode
of J2, owingat 34.7 Hz inertial-cross
to the is influenced by the structural
coupling between compliance. In particular,
the two joints. The secondstructural
mode at
compliance in the direction perpendicular to the axes J2 and J3 leads to an apparent
34.7 Hz is influenced by the structural compliance. In particular, structural compliance in the direction torsion of the
robot arm. The large damping ratio (4.4%) is in agreement with the contribution
perpendicular to the axes J2 and J3 leads to an apparent torsion of the robot arm. The large damping of structural
deformation. The presence of structural modes in the range of frequencies of the modes dominated
ratio (4.4%) is in agreement with the contribution of structural deformation. The presence of structural
by joint compliance has been detected by other researchers [6].
modes in the range of frequencies of the modes dominated by joint compliance has been detected by
Figure 10 deals with experimental results obtained in configuration Test 4 with lateral excitation
other(yresearchers
direction). [6].
Both the direct point FRF and the CMIF highlight the presence of an important
Figure
resonance 10 deals with experimental
peak beyond the 10÷40 Hz results obtained
frequency band,inwhich
configuration Testpreviously
contains the 4 with lateral excitation
identified
(y direction).
modes. Modal Bothanalysis
the direct
madepoint FRF to
it possible and the CMIF
identify a mode highlight the at
of vibration presence
56.8 Hz ofwithanaimportant
3.0%
resonance
damping peak beyond
ratio, which the 10 ÷ 40
is similar Hz ones
to the frequency band,joints.
of the other whichThiscontains the previously
mode, which is represented identified
in
Figure 10, shows a large contribution of the rotation about J4, which leads to
modes. Modal analysis made it possible to identify a mode of vibration at 56.8 Hz with a 3.0% damping lateral displacements of
the points of the following links.
ratio, which is similar to the ones of the other joints. This mode, which is represented in Figure 10,
Finally, Figure 11 shows experimental results obtained in configuration Test 5 with longitudinal
shows a large contribution of the rotation about J4, which leads to lateral displacements of the points
excitation (x direction). At high frequency, there is a very important peak at about 170 Hz. Modal
of the following links.
analysis showed that at 171.4 Hz there is a mode of vibration with large displacements localized in
the last link and caused by the compliance of J5 (see Figure 11). The damping ratio of this mode is
2.8%.
Appl. Sci. 2020, 10, 4619 12 of 20
Appl. Sci. 2020, 10, x FOR PEER REVIEW 12 of 19
Appl. Sci. 2020, 10, x FOR PEER REVIEW 12 of 19

Figure 9. Test 3, (a) direct point frequency response function (FRF) modulus, (b) phase, (c) complex
9. Test9.3,Test
Figure Figure (a) 3,
direct point
(a) direct frequency
point response
frequency responsefunction
function (FRF) modulus,
(FRF) modulus, (b)(b) phase,
phase, (c) complex
(c) complex
mode indicator function (CMIF).
mode indicator function (CMIF). mode indicator function (CMIF).

Figure 10. Test 4, (a) direct point frequency response function (FRF) modulus, (b) phase, (c) complex
Figure 10. Test 4, (a) direct point frequency response function (FRF) modulus, (b) phase, (c) complex
Figure 10. Test 4, (a) direct point frequency response
mode indicator function
function (FRF) modulus, (b) phase, (c) complex
(CMIF).
mode indicator function (CMIF).
mode indicator function (CMIF).
Appl. Sci. 2020, 10, 4619 13 of 20

Finally, Figure 11 shows experimental results obtained in configuration Test 5 with longitudinal
excitation (x direction). At high frequency, there is a very important peak at about 170 Hz. Modal analysis
showed that at 171.4 Hz there is a mode of vibration with large displacements localized in the last link
and caused by the compliance of J5 (see Figure 11). The damping ratio of this mode is 2.8%. 13 of 19
Appl. Sci. 2020, 10, x FOR PEER REVIEW

Figure Test
11.11.
Figure 5,5,(a)(a)direct
Test directpoint
pointfrequency
frequency response function(FRF)
response function (FRF)modulus,
modulus,(b)
(b) phase,
phase, (c)(c) complex
complex
mode indicator function (CMIF). mode indicator function (CMIF).

5. 5.
Identification
Identification
Testing
Testing configurations
configurationsTest Test11andandTestTest22mademade it it possible
possible to toidentify
identifymodes
modesofofvibration
vibration clearly
clearly
dominated
dominatedby bythe
thestiffness propertiesofofJ1J1and
stiffness properties and J2 J2 respectively,
respectively, andand
thesethese stiffness
stiffness properties
properties are theare
themost
mostimportant
importantfrom fromthethepoint
pointofofview
viewofofrobotrobotoperations
operations(e.g., (e.g., machining
machining oror assembly).
assembly). Testing
Testing
configurations Test 4 and Test 5 made it possible to identify modes of vibration
configurations Test 4 and Test 5 made it possible to identify modes of vibration with a large with a large contribution
of contribution
the stiffnessofofthe J4 stiffness
and J5. ofOwing
J4 and to J5. the
Owingsmall moments
to the of end-effector
small moments forces forces
of end-effector about about
J4 and J4 J5,
theand
stiffness
J5, theof wrist joints
stiffness is joints
of wrist less critical than the
is less critical thanstiffness of armofjoints.
the stiffness Testing
arm joints. configuration
Testing configuration Test 3
Test 3 minimized
minimized the couplingthe coupling
between between
J2 and J3,J2nevertheless
and J3, nevertheless the identified
the identified modal shapemodal a atshape a atshowed
30.8 Hz 30.8
Hz showed some influence of the rotation about J2 and of structural
some influence of the rotation about J2 and of structural deformation, since in this configuration deformation, since in this
theconfiguration the robotmode
robot has a structural has aad structural
34.7 Hz. modeTherefore,ad 34.7 theHz. Therefore,
measured the measured
frequency is chieflyfrequency
influenced is by
chiefly influenced by the stiffness of J3, but it includes the contribution
the stiffness of J3, but it includes the contribution of other stiffness properties. of other stiffness properties.
For For
thethe aforementionedreasons,
aforementioned reasons,the the stiffness
stiffness about
about J1,J1, J2,
J2,J4
J4and
andJ5J5was
wasdirectly
directly calculated
calculated fromfrom
equation 7, assuming modal coordinate 𝜂 coincident with 𝑞
Equation (7), assuming modal coordinate ηi coincident with qi and using a 1 DOF model: and using a 1 DOF model:

1 s 𝑘
𝑓 = 1 k (10)
fni = 2𝜋 𝐼 𝒒i𝒊 ) (10)
2π Ii q
zz i
𝑘 = 2𝜋𝑓 ) 𝐼 𝒒𝒊 ) (11)
fni )2 Iizz
 
where 𝑓 is the natural frequency measured (2π
ki = in testing qi
configuration 𝒒𝒊 . (11)
The damping coefficient of joint 𝑖 (𝑐 ) was calculated with the following equation:
where fni is the natural frequency measured in testing configuration qi .
2𝜁 𝑘
𝑐 = (12)
2𝜋𝑓
where 𝜁 is the identified damping ratio of mode 𝑖.
The results of these calculations are summarized in Table 2.
Appl. Sci. 2020, 10, 4619 14 of 20

The damping coefficient of joint i (ci ) was calculated with the following equation:

2ζi ki
ci = (12)
2π fni

where ζi is the identified damping ratio of mode i.

The results of these calculations are summarized in Table 2.

Table 2. Robot stiffness and damping properties obtained with the 1-degree-of-freedom (DOF) model.

Joint 1 Joint 2 Joint 3 Joint 4 Joint 5

Stiffness
12,602 9257 7490 643 898
[Nm/rad]
Damping
7.21 3.20 1.63 0.11 0.05
[Nms/rad]

For joint J3, the value given by Equation (11) was considered only a first estimate of the actual
joint stiffness (k3 ). The identification of this stiffness was then improved carrying out an optimization.
The following penalty function was defined:

5 X
X 5  2
Fp (k3 ) = fnjk − fpjk (k3 ) (13)
j = 1k = 1

This is the sum of the squared differences between the measured natural frequencies and the natural
frequencies fpjk predicted by the model with assigned k3 . Five modes of vibration (j = 1, . . . , 5)
in the five testing configurations are considered (k = 1, . . . , 5). Minimization was carried out with
the function fmincon of MATLAB. The optimized stiffness k3 = 6485 Nm/rad was rather different
from the first attempt value. Finally, damping coefficient was calculated according to Equation (12)
and resulted c3 = 1.41 Nms/rad.
The identified stiffness and damping properties were the implemented in the mathematical
model of the robot (Equation (3)), and the natural frequencies and damping ratios were calculated
in some validation configurations that were experimentally tested as well. Joint variables of validation
configurations are summarized in Table 3.

Table 3. Robot validation configurations.

Configuration Joint 1 [◦ ] Joint 2 [◦ ] Joint 3 [◦ ] Joint 4 [◦ ] Joint 5 [◦ ] Joint 6 [◦ ]

Validation 1 0 30 30 0 90 0
Validation 2 30 −50 160 0 90 0

Table 4 shows the comparison between numerical values and experimental values
in the validation configurations.

Table 4. Validation results.

Configuration Mode 1 Mode 2 Mode 3 Mode 4 Mode 5

Calc. frequency [Hz] 15.4 16.1 33.0 57.2 171.9
Exp. frequency [Hz] 16.3 17.8 36.2 58.4 170.5
Validation 1
Calc. damping % 2.6 1.8 2.4 3.0 2.8
Exp. damping % 3.9 2.1 3.1 3.4 3.2
Calc. frequency [Hz] 12.9 15.7 46.1 57.2 172.3
Exp. frequency [Hz] 13.3 14.5 40.2 59.1 171.8
Validation 2
Calc. damping % 1.33 2.8 3.4 3.0 2.8
Exp. damping % 2.7 2.4 2.5 3.7 2.7
Appl. Sci. 2020, 10, 4619 15 of 20

Appl. Sci. 2020, 10, x FOR PEER REVIEW 15 of 19

In validation configuration 1, the predicted natural frequencies of the first four modes are a bit
lowerLooking
than theatexperimental
modal damping,naturalthere is a generalThe
frequencies. agreement between
largest error numerical
is less and experimental
than −10%, and it takes
values
place in in
theboth configurations.
second mode, whichThe is amodes
mode on thatthe
show a large
meridian experimental
plane dominated damping typically
by the rotation show
about J2.a
large
The numerical
predicted damping.
natural Onlyof
frequency inMode
the first
5 ismode the calculated
a bit larger than the value is significantly
experimental value. lower than the
experimental
In validation value. The presence
configuration of predicted
2, the configuration dependent
natural friction
frequencies forces,
of Mode dueMode
1 and to large gravity
4 are a bit
loads,than
lower maythe beexperimental
the cause of this phenomenon.
values (about −3%), whereas the other natural frequencies are higher than
the experimental values. The largest error (+15%) takes place in the frequency of Mode 3. This happens
6. Numerical
because modalsimulations
analysis (Figure 9) showed that in this range of frequencies structural deformability
affectsRecent
the modes of vibration,
studies and
[2,6,35,36] thishighlighted
have effect is not taken into
that, in account
some robot byoperations
the numerical model,
(e.g., millingwhich is
tasks),
based on joint stiffness.
the dynamic properties of the robot arm are important, and that a simple optimization of robot
Looking at modal
configurations based ondamping, there is adoes
static stiffness general
not agreement
lead to thebetween numerical and
best performance experimental
[2]. The modes of
values in both configurations. The modes that show a large experimental damping
vibration of a robot are configuration-dependent, and the variation of natural frequency and modal typically show
adamping
large numerical damping. Only in the first mode the calculated value is significantly
in the workspace is an important feature of robot dynamics. It gives some hints useful lower thanto
the experimental
improve value. The presence
robot performance selectingof configuration dependent
robot configurations friction
with highlyforces, due tomodes
damped large gravity
having
loads, may be the cause of this phenomenon.
natural frequencies far from the excitation frequencies.
Experimental modal analysis is time consuming, thus, the mathematical model based on modal
6. Numerical Simulations
analysis in a small number of selected configurations is a very useful tool to predict the variation in
Recent studies
the dynamic [2,6,35,36]
properties of the have
robothighlighted
over the whole that,workspace.
in some robot operations (e.g., milling tasks),
the dynamic
The first numerical analysis aimed to understand the and
properties of the robot arm are important, effectthat a simple
of robot optimization
configuration on theofnatural
robot
configurations
frequency andbased dampingon static
of thestiffness
modes of does not lead
vibration. A to theofbest
total performance
10,000 [2]. The modes
random configurations of
of the
vibration of a robot aretoconfiguration-dependent,
robot corresponding 10,000 random combinations and the variation
of joint of natural
variables frequency
J2, J3, J4, and modal
and J5 (within joint
damping in the defined,
ranges) were workspace andis an
theimportant feature of robot
modal properties were dynamics.
calculated.It Since
gives some hints useful
the robot modeltois
improve robot performance
axial-symmetric about J1, theselecting robot
variation of configurations
J1 was not taken with highly
into damped
account, and modes having
this joint natural
variable was
frequencies
set to zero. far
Thefrom the of
results excitation frequencies.
these calculations are collected in Figure 12 and 13, which depict the mean
Experimental
values and standard modal analysisofisnatural
deviations time consuming,
frequenciesthus, and the
modalmathematical
dampings,model based on modal
respectively.
analysisTheinnatural
a smallfrequencies
number ofofselected configurations
the second and third mode, is a very
which useful tool to characterized
are usually predict the variation
by large
indisplacements
the dynamic properties of the robot over the whole workspace.
in the meridian plane, show the largest standard deviations. Therefore, these modes
The first numerical
are influenced analysis
by the robot aimed to understand
configuration. the effect
The first natural of robotwhich
frequency, configuration
usually on the naturalto
corresponds
frequency
a mode of andvibration
dampingwith of thelarge
modes of vibration. Aintotal
displacements theofhorizontal
10,000 randomplane,configurations
shows a mild of the robot
standard
corresponding to 10,000 random combinations of joint variables J2, J3, J4,
deviation, since the robot configuration affects the moment of inertia about J1. The natural and J5 (within joint ranges)
were defined, of
frequencies andthe
thelast
modal
twoproperties were calculated.
modes, which are chiefly Since the robotby
dominated model is axial-symmetric
J4 and J5, show veryabout small
J1, the variation
standard of J1 was
deviations. Thisnot taken into
happens account,
because and this
the robot joint variable
configuration haswas set toeffect
a small zero.on The
theresults
direct
ofinertia
these calculations
term of J4 and are collected
no effect in onFigures 12 and
the direct 13, which
inertia term ofdepict the itmean
J5, but onlyvalues andthe
changes standard
inertial
deviations of natural
cross-coupling termsfrequencies
(see Figureand 3). modal dampings, respectively.

Figure 12. Mean values and standard deviations of natural frequencies in the workspace, (a) random
Figure 12. Mean values and standard deviations of natural frequencies in the workspace, (a) random
variations of J2, J3, J4, J5, (b) random variations of J2 and J3.
variations of J2, J3, J4, J5, (b) random variations of J2 and J3.
Appl. Sci. 2020, 10, 4619 16 of 20
Appl. Sci. 2020, 10, x FOR PEER REVIEW 16 of 19

Figure 13. Mean values and standard deviations of modal damping in the workspace, (a) random
Figure 13. Mean values and standard deviations of modal damping in the workspace, (a) random
variations in J2, J3, J4, J5, (b) random variations in J2 and J3.
variations in J2, J3, J4, J5, (b) random variations in J2 and J3.
The natural frequencies of the second and third mode, which are usually characterized by large
The statistical analysis of modal damping variation with robot configurations is depicted in
displacements in the meridian plane, show the largest standard deviations. Therefore, these modes
Figure 13. This figure shows that robot configuration affects the modal damping of the first three
are influenced by the robot configuration. The first natural frequency, which usually corresponds
modes, since standard deviations are comparable with the mean values. Conversely, robot
to a mode of vibration with large displacements in the horizontal plane, shows a mild standard
configuration has a very small effect on the damping of the last modes, which chiefly involve wrist
deviation, since the robot configuration affects the moment of inertia about J1. The natural frequencies
joints, because standard deviations are very small.
of the last two modes, which are chiefly dominated by J4 and J5, show very small standard deviations.
Figure 12b and Figure 13b show the result of a similar calculation, in which only random
This happens because the robot configuration has a small effect on the direct inertia term of J4 and no
variations in J2 and J3 were performed. The standard deviations of the natural frequencies and
effect on the direct inertia term of J5, but it only changes the inertial cross-coupling terms (see Figure 3).
modal dampings are very similar to the ones of Figure 12a and Figure 13a, therefore, this result
The statistical analysis of modal damping variation with robot configurations is depicted
points out that the variations in J2 and J3 have the largest effect on the modal properties of the robot.
in Figure 13. This figure shows that robot configuration affects the modal damping of the first
Figures 12 and 13 highlight another important feature of the robot. Variations in joint variables
three modes, since standard deviations are comparable with the mean values. Conversely, robot
are able to modify the natural frequencies, but they do not change the typical frequency bands of the
configuration has a very small effect on the damping of the last modes, which chiefly involve wrist
modes. In other words, the frequencies of the modes associated to the wrist joints are always higher
joints, because standard deviations are very small.
than the frequencies associated to the arm joints. The first two modes belong to the same frequency
Figures 12b and 13b show the result of a similar calculation, in which only random variations
band (15 ÷ 30 Hz), and this band has a negligible overlap with the bands containing the other
in J2 and J3 were performed. The standard deviations of the natural frequencies and modal dampings
modes.
are very similar to the ones of Figures 12a and 13a, therefore, this result points out that the variations
The statistical analysis showed mild variations in the properties of the first modes of vibrations
in J2 and J3 have the largest effect on the modal properties of the robot.
of the robot, due to variations in J2 and J3. These joint variables determine the configuration of the
Figures 12 and 13 highlight another important feature of the robot. Variations in joint variables
robot in the meridian plane and in particular the location of the wrist (the origin of coordinate
are able to modify the natural frequencies, but they do not change the typical frequency bands of
system 4 in Figure 2). Therefore, a further numerical analysis was carried out to study the
the modes. In other words, the frequencies of the modes associated to the wrist joints are always higher
dependence of the first three natural frequencies on the position of the wrist in the meridian plane.
than the frequencies associated to the arm joints. The first two modes belong to the same frequency
The results are presented in terms of contour plots, in which the darker colors represent the lower
band (15 ÷ 30 Hz), and this band has a negligible overlap with the bands containing the other modes.
frequencies and the lighter colors represent the higher frequencies. Since some locations can be
The statistical analysis showed mild variations in the properties of the first modes of vibrations
reached with two robot configurations (elbow-up and elbow-down), both configurations are
of the robot, due to variations in J2 and J3. These joint variables determine the configuration of
considered in separate plots.
the robot in the meridian plane and in particular the location of the wrist (the origin of coordinate
Figure 14 shows that the first natural frequency decreases in a regular way when the distance
system 4 in Figure 2). Therefore, a further numerical analysis was carried out to study the dependence
between the wrist and the robot base increases, and this effect takes place both in the elbow-up and
of the first three natural frequencies on the position of the wrist in the meridian plane. The results
in the elbow-down configuration.
are presented in terms of contour plots, in which the darker colors represent the lower frequencies
The contour plot of the second natural frequency is more complex (Figure 15). The second
and the lighter colors represent the higher frequencies. Since some locations can be reached with two
natural frequency reaches the lowest values when the arm is extended in order to reach locations in
robot configurations (elbow-up and elbow-down), both configurations are considered in separate plots.
front of the base or on the back of the base. In these configurations, links 2 and 3 tend to align. The
Figure 14 shows that the first natural frequency decreases in a regular way when the distance
maximum value of the second natural frequency takes place when the arm has to reach locations
between the wrist and the robot base increases, and this effect takes place both in the elbow-up
above the base. This effect is present in both the considered configurations.
and in the elbow-down configuration.
Figure 16 shows a rather regular trend of the third natural frequency. In both configurations,
this natural frequency tends to increase when the wrist location is far from the base and links 2 and 3
tend to align (joint variable J3 is close to 90°).
Therefore, the angle (𝑞 ) between link 2 and 3 has opposite effects on the second and third
natural frequencies that usually correspond to modes that take place in the meridian plane. The
second natural frequency decreases when the two links tend to align, because the direct inertia term
𝑀 increases. The third natural frequency increases when the two links tend to align, because the
inertial cross coupling term (𝑀 ) between J2 and J3 increases (see Figure 5). This effect takes place
even if a simple 2 DOF model of the robot in the meridian plane is adopted [28].
Finally, the analysis of Figure 15 and 16 shows that it is rather difficult to maximize both the
second
Appl. and the 10,
Sci. 2020, third
4619natural frequencies. Only in a limited region of the workspace above the robot
17 of 20
it is possible to achieve large values
Appl. Sci. 2020, 10, x FOR PEER REVIEW of both frequencies. 17 of 19

𝑀 increases. The third natural frequency increases when the two links tend to align, because the
inertial cross coupling term (𝑀 ) between J2 and J3 increases (see Figure 5). This effect takes place
even if a simple 2 DOF model of the robot in the meridian plane is adopted [28].
Appl. Sci. 2020, 10, x FOR PEER REVIEW 17 of 19
Finally, the analysis of Figure 15 and 16 shows that it is rather difficult to maximize both the
second and the The
𝑀 increases. thirdthird
natural frequencies.
natural frequencyOnly in a limited
increases whenregion of links
the two the workspace above
tend to align, the robot
because the
itinertial
is possible to achieve large values of both frequencies.
cross coupling term (𝑀 ) between J2 and J3 increases (see Figure 5). This effect takes place
even if a simple 2 DOF model of the robot in the meridian plane is adopted [28].
Finally, the analysis of Figure 15 and 16 shows that it is rather difficult to maximize both the
second and the third natural frequencies. Only in a limited region of the workspace above the robot
it is possible to achieve large values of both frequencies.

Figure 14. Contour

Figure plot of
14. Contour theofnatural
plot frequency
the natural of mode
frequency 1 in the
of mode meridian
1 in plane, (a)
the meridian elbow-up,
plane, (b)
(a) elbow-up,
elbow-down.
(b) elbow-down.

The contour plot of the second natural frequency is more complex (Figure 15). The second natural
frequency reaches the lowest values when the arm is extended in order to reach locations in front of
the base or on the back of the base. In these configurations, links 2 and 3 tend to align. The maximum
value of the second
Figure natural
14. Contour frequency
plot of takes
the natural placeofwhen
frequency modethe
1 inarm has to reach
the meridian locations
plane, above
(a) elbow-up, (b)the base.
This effect is present
elbow-down. in both the considered configurations.

Figure 14. Contour plot of the natural frequency of mode 1 in the meridian plane, (a) elbow-up, (b)
elbow-down.

Figure 15. Contour plot of the natural frequency of mode 2 in the meridian plane, (a) elbow-up, (b)
elbow-down.

Figure
Figure 15.15.Contour
Contour plot
plot of
ofthe
thenatural
naturalfrequency of mode
frequency 2 in 2
of mode theinmeridian plane, plane,
the meridian (a) elbow-up, (b)
(a) elbow-up,
(b) elbow-down. elbow-down.

Figure 16 shows a rather regular trend of the third natural frequency. In both configurations,
this natural frequency
Figure tends
15. Contour tothe
plot of increase
natural when the of
frequency wrist location
mode is far from
2 in the meridian the(a)
plane, base and links
elbow-up, (b) 2 and 3
tend to align (joint variable J3 is close to 90◦ ).elbow-down.

Figure 16. Contour plot of the natural frequency of mode 3 in the meridian plane, (a) elbow-up, (b)
elbow-down.

Figure 16. Contour plot of the natural frequency of mode 3 in the meridian plane, (a) elbow-up, (b)
elbow-down.

Figure16.
Figure 16.Contour
Contour plot
plot of
ofthe
thenatural
naturalfrequency of mode
frequency 3 in3the
of mode in meridian plane,plane,
the meridian (a) elbow-up, (b)
(a) elbow-up,
elbow-down.
(b) elbow-down.
Appl. Sci. 2020, 10, 4619 18 of 20

Therefore, the angle (q3 ) between link 2 and 3 has opposite effects on the second and third natural
frequencies that usually correspond to modes that take place in the meridian plane. The second natural
frequency decreases when the two links tend to align, because the direct inertia term M22 increases.
The third natural frequency increases when the two links tend to align, because the inertial cross
coupling term (M23 ) between J2 and J3 increases (see Figure 5). This effect takes place even if a simple
2 DOF model of the robot in the meridian plane is adopted [28].
Finally, the analysis of Figures 15 and 16 shows that it is rather difficult to maximize both the second
and the third natural frequencies. Only in a limited region of the workspace above the robot it is
possible to achieve large values of both frequencies.

7. Conclusions
The proposed testing method based on selective modal testing requires simple equipment, does
not take a long testing time, can be implemented in every robot configuration, and makes it possible to
detect the presence of modes of vibration affected by structural and bearing compliance.
The combination of the results of selective modal testing with a mathematical model of mass
distribution of the robot makes it possible a straightforward identification of joint stiffness and damping.
A simple optimization procedure on a small set of parameters is carried out, to identify joint stiffness
in the presence of a strong inertial cross-coupling between the joints.
The mathematical model implemented with the identified stiffness and damping properties
gives information about the variation of the dynamic characteristics of the robot in the workspace.
This information is useful to optimize the performance of the robot in machining tasks.
In future, the method will be improved, in order to cope with the presence of the structural mode
in the band of frequency of interest.

Author Contributions: Conceptualization, M.B., S.C., N.C., A.D.; methodology, M.B., S.C., A.D.; software, M.B.,
N.C.; investigation, S.C., N.C., A.D.; writing—original draft preparation, M.B., S.C.; writing—review and editing,
N.C., A.D.; supervision, A.D. All authors have read and agreed to the published version of the manuscript.
Funding: This research was funded by Fondazione Aldo Gini (Call 2018), Ermenegildo Zegna Founder’s
Scholarship 2019/2020, and by University of Padova DII project BIRD187930.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Chen, H.; Dong, F. Robot machining: Recent development and future research issues. Int. J. Adv. Manuf.
Technol. 2013, 66, 1489–1497. [CrossRef]
2. Leonesio, M.; Villagrossi, E.; Beschi, M.; Marini, A.; Bianchi, G.; Pedrocchi, N.; Tosatti, L.M.; Grechishnikov, V.;
Ilyukhin, Y.; Isaev, A. Vibration Analysis of Robotic Milling Tasks. Procedia CIRP 2018, 67, 262–267. [CrossRef]
3. Pan, Z.; Zhang, H.; Zhu, Z.; Wang, J. Analysis of robotic machining process. J. Mater. Process. Technol. 2006,
173, 301–309. [CrossRef]
4. Zhang, H.; Wang, J.; Zhang, G.; Gan, Z.; Pan, Z.; Cui, H.; Zhu, Z. Machining with flexible manipulator: Toward
improving robotic machining performance. In Proceedings of the IEEE/ASME International Conference on
Advanced Intelligent Mechatronics, Monterey, CA, USA, 24–28 July 2005.
5. Carbone, G. Stiffness analysis and experimental validation of robotic systems. Front. Mech. Eng. Chin. 2011,
6, 182–196. [CrossRef]
6. Huynh, H.N.; Assadi, H.; Rivière-Lorphèvre, E.; Verlinden, O.; Ahmadi, K. Modelling the dynamics of
industrial robots for milling operations. Robot. Comput. Integr. Manuf. 2020, 61, 101852. [CrossRef]
7. Rosati, G.; Faccio, M.; Carli, A.; Rossi, A. Fully flexible assembly systems (F-FAS): A new concept in flexible
automation. Assembly Autom. 2013, 33, 8–21. [CrossRef]
8. Rosati, G.; Faccio, M.; Finetto, C.; Carli, A. Modelling and optimization of fully flexible assembly systems
(F-FAS). Assembly Autom. 2013, 33, 165–174. [CrossRef]
9. Faccio, M.; Bottin, M.; Rosati, G. Collaborative and traditional robotic assembly: A comparison model. Int. J.
Adv. Manuf. Technol. 2019, 102, 1355–1372. [CrossRef]
Appl. Sci. 2020, 10, 4619 19 of 20

10. Cocuzza, S.; Pretto, I.; Debei, S. Least-squares-based reaction control of space manipulators. J. Guid. Control
Dynam. 2012, 35, 976–986. [CrossRef]
11. Tringali, A.; Cocuzza, S. Predictive control of a space manipulator through error and kinetic energy expectation.
In Proceedings of the 69th International Astronautical Congress (IAC 2018), Bremen, Germany, 1–5 October
2018.
12. Cocuzza, S.; Pretto, I.; Debei, S. Novel reaction control techniques for redundant space manipulators: Theory
and simulated microgravity tests. Acta Astronaut. 2011, 68, 1712–1721. [CrossRef]
13. Cocuzza, S.; Tringali, A.; Yan, X. Energy-efficient motion of a space manipulator. In Proceedings of the 67th
International Astronautical Congress (IAC 2016), Guadalajara, Mexico, 26–30 September 2016; pp. 8788–8804.
14. Gasparetto, A. Eigenvalue analysis of mode-coupling chatter for machine-tool stabilization. J. Vib. Control
2001, 7, 181–197. [CrossRef]
15. Gasparetto, A. A system theory approach to mode coupling chatter in machining. J. Dyn. Syst. Meas. Control
1998, 120, 545–547. [CrossRef]
16. Doria, A.; Cocuzza, S.; Comand, N.; Bottin, M.; Rossi, A. Analysis of the compliance properties of an
industrial robot with the Mozzi axis approach. Robotics 2019, 8, 80. [CrossRef]
17. Ni, H.; Zhang, C.; Hu, T.; Wang, T.; Chen, Q.; Chen, C. A dynamic parameter identification method of
industrial robots considering joint elasticity. Int. J. Adv. Rob. Syst. 2019, 16, 1729881418825217. [CrossRef]
18. Dumas, C.; Caro, S.; Garnier, S.; Furet, B. Joint stiffness identification of six-revolute industrial serial robots.
Rob. Comput. Integr. Manuf. 2011, 27, 881–888. [CrossRef]
19. Abele, E.; Weigold, M.; Rothenbücher, S. Modeling and identification of an industrial robot for machining
applications. CIRP Ann. 2007, 56, 387–390. [CrossRef]
20. Rafieian, F.; Liu, Z.; Hazel, B. Dynamic model and modal testing for vibration analysis of robotic grinding
process with a 6DOF flexible-joint manipulator. In Proceedings of the International Conference on
Mechatronics and Automation, Changchun, China, 9–12 August 2009.
21. Alici, G.; Shirinzadeh, B. Enhanced stiffness modeling, identification and characterization for robot
manipulators. IEEE Trans. Rob. 2005, 21, 554–564. [CrossRef]
22. Abele, E.; Rothenbücher, S.; Weigold, M. Cartesian compliance model for industrial robots using virtual
joints. Prod. Eng. 2008, 2, 339. [CrossRef]
23. Carbone, G.; Ceccarelli, M. A stiffness analysis for a hybrid parallel-serial manipulator. Robotica 2004, 22,
567–576. [CrossRef]
24. Mousavi, S.; Gagnol, V.; Bouzgarou, B.C.; Ray, P. Dynamic behaviour model of a machining robot.
In Proceedings of the ECCOMAS Multibody Dynamics, Zagreb, Croatia, 1–4 July 2013; pp. 771–779.
25. Mousavi, S.; Gagnol, V.; Bouzgarrou, B.C.; Ray, P. Stability optimization in robotic milling through the control
of functional redundancies. Robot. Comput. Integr. Manuf. 2018, 50, 181–192. [CrossRef]
26. Behi, F.; Tesar, D. Parametric identification for industrial manipulators using experimental modal analysis.
IEEE Trans. Rob. Autom. 1991, 7, 642–652. [CrossRef]
27. Corral, E.; García, M.J.; Castejon, C.; Meneses, J.; Gismeros, R. Dynamic Modeling of the Dissipative Contact
and Friction Forces of a Passive Biped-Walking Robot. Appl. Sci. 2020, 10, 2342. [CrossRef]
28. Craig, J.J. Introduction to Robotics, Mechanics & Control; Addison-Wesley Publishing Company: Reading, MA,
USA, 1986.
29. Gómez, M.J.; Corral, E.; Castejon, C.; García-Prada, J.C. Effective crack detection in railway axles using
vibration signals and WPT energy. Sensors 2018, 18, 1603. [CrossRef] [PubMed]
30. Ewins, D.J. Modal testing: Theory and Practice; Research Studies Press: Hertfordshir, UK, 1984.
31. Verbeke, J.; Debruyne, S. Vibration analysis of a UAV multirotor frame. In Proceedings of the ISMA 2016
International Conference on Noise and Vibration Engineering, Leuven, Belgium, 19–21 September 2016.
32. Cossalter, V.; Doria, A.; Basso, R.; Fabris, D. Experimental analysis of out-of-plane structural vibrations of
two-wheeled vehicles. Shock Vib. 2004, 11, 433–443. [CrossRef]
33. Belotti, R.; Caneva, G.; Palomba, I.; Richiedei, D.; Trevisani, A. Model updating in flexible-link multibody
systems. J. Phys. Conf. Ser. 2016, 744, 012073. [CrossRef]
34. Cossalter, V.; Doria, A.; Mitolo, L. Inertial and modal properties of racing motorcycles. SAE Tech. Pap. 2002.
[CrossRef]
Appl. Sci. 2020, 10, 4619 20 of 20

35. Soriano, E.; Rubio, H.; Castejón, C.; García-Prada, J.C. Design of a low-cost manipulator arm for industrial
fields. In New Trends in Mechanism and Machine Science; Springer: Cham, Switzerland, 2015; pp. 839–847.
36. Rahman, N.; Carbonari, L.; Caldwell, D.; Cannella, F. Kinematic Analysis, Prototypation and Control of a
Novel Gripper for Dexterous Applications. J. Intell. Robot. Syst. 2018, 91, 193–206. [CrossRef]

© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).