Robotics and Computer-Integrated Manufacturing 37 (2016) 292–301

Contents lists available at ScienceDirect

Robotics and Computer-Integrated Manufacturing

journal homepage: www.elsevier.com/locate/rcim

Modeling and simulation for fatigue life analysis of robots with flexible
joints under percussive impact forces
Songliang Nie a, Yuwen Li b, Guo Shuai a,n, Song Tao a, Fengfeng Xi b
a
Shanghai Key Laboratory of Intelligent Manufacturing and Robotics, HC206A, Shanghai University, 99 Shangda Road, BaoShan District, Shanghai, China
b
Department of Aerospace Engineering, Ryerson University, 350 Victoria Street, Toronto, ON M5B 2K3, Canada

art ic l e i nf o a b s t r a c t

Article history: This paper presents a method for modeling and analyzing the fatigue life of robots with flexible joints,
Received 4 August 2014 with a particular focus on applications under percussive impact forces. This development is motivated by
Received in revised form growing interests in robotic automation for operations with percussive impact tools. The most important
19 March 2015
characteristic of percussive operations is the repetitive impacts generated by the tool, such as a per-
Accepted 17 April 2015
cussive rivet gun. After modeling of a flexible joint robot, a forced vibration solution is provided by
Available online 5 May 2015
including the impact forces generated by the percussive gun, projecting them onto the robot joint space
Keywords: and treating them in terms of the Fourier transform. As a result, the joint angular displacements can be
Fatigue life solved using a standard vibration method. Then the joint stresses can be determined through Hooke's
Percussive rivet
law. To consider the stress variations caused by the robot operating at different poses using different
Impact
rivets, a multiple-loading fatigue model is applied from which an equation is derived to determine the
Robot
total number of the rivets that can be riveted before robot's fatigue failure. Based on simulation using our
model, the following observations are received. First, the joint torsional stresses vary with robot's po-
sition and orientation. Second, no joint will always experience the maximum stress and the joint stress
dominancy also varies with robot's position and orientation. Third, at a given riveting point, the rivet gun
direction considerately affects the joint stresses. Fourth, the fatigue life of each joint is different;
therefore robot's fatigue life should be evaluated based on the shortest joint fatigue life.
& 2015 Elsevier Ltd. All rights reserved.

1. Introduction
In the past few decades, robots have been playing a significant
role in industrial automation to provide high productivity, adapt-
ability, quality, and low cost. They have been used in a wide range
of fields, such as machining, assembly, packaging, and material
handling [1]. A robotic automation system usually includes a robot
and a tooling system. A tooling system is composed of tools, such
as grippers and cutting tools, along with a tool mount that is at-
tached to the robot's end-effector. Though most industrial robots
are designed for general applications, tooling systems are usually
customized according to specific tasks. In this paper, we focus on
the automation tasks that require percussive tools. These tools
have been widely used as hammers, drills, chippers, road breakers,
and rivet guns in many industries, such as the civil construction
and aerospace manufacturing [2,3]. For some industrial applica-
tions, manual percussive operations can be tedious, repetitious,
costly, and prone to error, and can cause health and ergonomic
problems related to human joint fatigue [4,5]. Therefore, there
n
Corresponding author.
have been increasing interests in the automation for manual op-
erations using percussive tools, especially in the aerospace man-
ufacturing industry [6–8]. This automation becomes necessary for
new unmanned tasks using percussive tools. For instance, per-
cussive drilling tools have recently been proposed for unmanned
Mars exploration [9].
The most important characteristic of percussive operations is
the repetitive impacts generated by the tool. Due to this highly
impact feature, the success of their robotic automation demands
specific research on these applications. Limited works have been
published in literature on the robotic automation of percussive
operations. Glass et al. [9] designed, tested, and analyzed the
performance of a new percussive drilling tool for unmanned Mars
exploration. Jayaweera and Webb [10] presented a robotic riveting
assembly system for typical aircraft panels and investigated the
positioning accuracy of the robot with a laser metrology device.
Though not explicitly mentioned, their robotic system could be
adopted to percussive riveting. These works are mainly focused on
the design, integration, and test of automatic system for percussive
operations. The performance of the percussive tool and the robot
are demonstrated respectively, but little discussion has been given
on the dynamics mechanism of percussive impacts.

http://dx.doi.org/10.1016/j.rcim.2015.04.001
0736-5845/& 2015 Elsevier Ltd. All rights reserved.
 S. Nie et al. / Robotics and Computer-Integrated Manufacturing 37 (2016) 292–301 293

Nomenclature F(t) Vector of periodic pulsive forces

G Shear modulus of the joint shaft
Symbol Description J
l Link length J 6 × n Jacobian matrix
mi Mass of the ith link K n × n generalized stiffness matrix
n Unit vector representing the force direction Kd n × n decoupled stiffness matrix
nd Unit vector used in the decoupled system M n × n generalized mass matrix
ni Number of cycles at working stress Md n × n decoupled mass matrix
q0 Vector of joint coordinates at an equilibrium ME Vector of the external moments acting at the end-
configuration effector
Δq Vector of joint deflections Ni Number of cycles to failure
Δq̇ Vector of joint deflection rates S Matrix of the time-independent normal mode shape
Δq0 Vecror of transient response of joint vibration Te Elastic energy
displacements Tk Kinetic energy
Δq ss Vector of steady-state response of joint vibration γ Shear strain
displacements η Vector of the time-dependent generalized coordinates
r Radius of the joint shaft ηi The ith element of vector η
Δs Arc length of the joint shaft θ Joint angle
v Vector of the velocity of the end-effector ς Damping ratio
w Vector of the wrench acting on the end-effector τ Vectror of torsional stresses
D Diagonal matrix of ωi2 ω Vector of the angular velocity of the end-effector
FE Vector of the external forces acting on the end-effector ωi The ith natural frequency of the system

The dynamics of percussive operations involves the impacts fatigue cycle. Miclosina and Campian [22] investigated the fatigue
generated by the tool and the interaction between the tool and the of a parallel robot, where the stress of the flexible linkages was
part. Kadam [11] and Bloxsom [12] investigated the modeling of calculated by Finite Element Analysis (FEA) in SolidWorks. In both
pneumatic percussive hammers. They demonstrated how the re- works, the robot was considered to performance simple point-to-
petitive impacts were produced through simulation. Quan et al. point motion. Thus, these works cannot represent the percussive
[13] presented and implemented the dynamic simulation of the operations with highly impact dynamics.
percussive driving mechanism for a rotary-percussive drilling tool. To estimate the cost, efficiency, and life limit of a robotic system
The impact energy produced by the percussive tool was modeled for percussive operations, the fatigue life of the robot under re-
by a spring-mass model. Johnson et al. [14] proposed a three de- petitive impact forces must be predicted. This paper aims to
grees-of-freedom (DOF) analytical dynamics model to simulate overcome this problem.
percussive riveting and to investigate the dynamic interaction
between the operator, the gun, and the part. Li et al. [15] studied
how the inertia of the percussive rivet gun could affect the ac- 2. System description and problem statement
celeration, natural frequencies, and energy consumption of the
robot. The above researches have provided some insight into the A robotic riveting system has been developed at Shanghai
complex dynamics of percussive operations, indicating the im- University [23]. As shown in Fig. 1, this system includes four ro-
portance of studying the dynamics of percussive operations for bots. A Fanuc M-20iA robot is used to hold and drive a percussive
their robotic automation. rivet gun. Two Kawasaki JS010G-A robots are used to form a
A most significant issue for percussive operations is the struc- flexible jig to hold a piece of sheet metal, and an ABB IRB 2600
tural fatigue due to the repetitive impacts, which has been ad- robot is used to hold a bucking bar for support during riveting. As
dressed for manual operations [4,5]. The human health risks can shown in Fig. 2, in riveting a hole has to be drilled first, then a rivet
be eliminated by robotic automation, but robots are flexible is inserted into the hole and deformed by force. For complete sheet
structures and have limited life expectancy. The life limit of a robot metal riveting, the Fanuc robot will move point-to-point along a
relates to the performance of its mechanical and electrical com- pre-planned path and the ABB robot will follow the same path in
ponents. The life expectancy of an industrial robot can largely vary
from five to twenty years, depending on the operating conditions
and care of service [16]. For percussive operations, because of the
repetitive impacts, the torsional stress in robot joints can vary
greatly in a relatively short time. It has been demonstrated that
varying torsional stress can lead to stress fatigue to machine shafts
[17]. Extensive research on the dynamics and control of flexible
robots can be found in literature. These works cover various topics
including stiffness mapping, dynamic modeling, inverse dynamics,
vibration control, and parameter estimation. The readers are re-
ferred to three review articles [18–20] for more detail on the
works on the dynamics and control of flexible robots. However, to
the knowledge of the authors, only a few studies have been pub-
lished on the fatigue analysis of flexible robots. Du, Yu, and Su [21]
analyzed the dynamic stress of linkages of a 3-RRR parallel robot Fig. 1. A robotic percussive riveting system.
under bending and then used the stress results to calculate the
294 S. Nie et al. / Robotics and Computer-Integrated Manufacturing 37 (2016) 292–301
Fig. 2. Riveting procedure.
synchronization with the Fanuc robot to provide the support for
riveting.
In principle there are two riveting methods, the first called
squeezing (or one-shot) riveting, where a large upsetting force is
applied to deform a rivet instantly. This method requires a large
riveter operating under high pressure beyond the yield strength of
aluminum rivets. This type of riveter is made of either a hydraulic
cylinder or an electromagnetic piston, very heavy, bulky and
usually requires a lifting assisted device if used for manual op-
eration. The automated riveting machines and existing riveting
robots all employ this type of riveter, and hence they are gigantic
and only limited to riveting large and simple components.
The second method is called percussive (or hammering) rivet-
ing, widely used for manual riveting, where a series of impact
forces is applied to deform a rivet accumulatively by a series of
hits. This method uses a rivet gun of regular hand-held power tool
size, very compact and light, operating under much lower pressure
in a range less than 100 psi; very safe and energy efficient. Our
robotic riveting is based on this principle, and as such a light
weight Fanuc robot of 20 Kg payload can be used to develop for
riveting.
While the main problem of squeezing robotic riveting is large
static deformation, percussive robotic riveting experiences ex-
cessive vibration due to striking from the percussive riveting gun.
Since the gun is mounted on robot's end-effector, when a series of
impact forces is employed, forced vibrations will be induced inside
the robot. There are hundreds and thousands of rivets in an air-
craft, if a percussive robotic riveting is used, it is necessary to in-
vestigate the fatigue life of the robot. To do so, a dynamic model
must be developed to analyze robot vibration and influence on its
fatigue life. This motivates the research reported in this paper. The
remaining paper is organized as follows. A theoretical model is
presented in Section 3 and simulation is carried out in Section 4.
3. Theoretical modeling
3.1. Equations of motion
Fig. 3 illustrates an n-DOF robot with n rigid links and n flexible
joints. The position vector of the ith joint in the inertial frame
{O _xyz} is denoted by Pi and the position of the end-effector E on
the last link is denoted by Pn + 1. For convenience, local body-fixed
frames {Oi _x iyi zi} (i = 1, 2, … , n) are established on the links at
their corresponding joints. For percussive operations, the robot
moves to a required position, stops in this position, and then sends
a signal to the percussive tool to complete the operation. There-
fore, to investigate the robot fatigue life, the vibration model of the
robot in a stationary position must be developed.
For industrial robots, the main source of deformation is joint
flexibility [1]. Thus, in our discussion, it is assumed that all the
links are rigid and all the joints are flexible revolute joints which
are modeled as linearly elastic torsional springs, as illustrated in
Fig. 3. An n-DOF flexible joint robot with a percussive impact tool acting on the
end-effector.
Fig. 3. If we choose the input joint angles as the generalized co-
ordinates, i.e. q = [θ1, θ2, ... , θn]T , the kinetic energy Tk and elastic
energy Te of a stationary flexible-joint robot can be respectively
written in terms of the mass and stiffness matrices as
1 T 1
Tk = Δq̇ M(q)Δq̇ and Te = ΔqTK(q)Δq
2 2 (1)
where Δq ∈ Rn is the perturbations in joint angles from an equi-
librium configuration q0 , representing the joint deflections, Δq is
the deflection rate, M(q) is the n × n symmetric generalized mass
matrix, and K is the n × n diagonal stiffness matrix in the joint
space, with its entries equal to the torsional stiffnesses of the
corresponding joints. Then, the dynamic model when the robot is
the stationary and under forced vibration can be represented as [1]
M (q)Δq + K Δq = J T w (2)
The 6 × n Jacobian matrix J represents the mapping from the
joint velocities to the velocities of tool center E such that
⎡ Jv ⎤
J=⎢ ⎥
⎣ Jω ⎦ (3)
where v = Jv q, ω = Jω q, Jv ∈ R3 × n , Jω ∈ R3 × n , v is the linear velocity
of point E , and ω is the angular velocity of the last link. The wrench
T
vector w can be written as W = ⎡⎣FTE , MTE ⎤⎦ , in which FE and ME are
the external forces and moments acting at the end-effector E . If
point E is defined at the tip of the percussive tool at which the
external repeated impact forces are applied to the robot, the ex-
ternal moments ME can be ignored. Furthermore, since the change
in the direction of percussive impact force during the operation
can be ignored, the external force can be written as
FE = F (t)n (4)
where n = [n1, n2, n3]T is a unit vector representing the force di-
rection and F (t) changes with the time. For percussive operations,
F (t) can be modeled as a periodic pulse train, as shown in Fig. 4.
Then, Eq. (2) is simplified as
 S. Nie et al. / Robotics and Computer-Integrated Manufacturing 37 (2016) 292–301 295

where ηi is the ith element of vector η , ndi is the element on the ith
element of vector nd . Also, damping ratio ζ is incorporated in Eq.
(10) to consider the influence of material viscous damping to the
system vibrations.
The periodic pulse train input force F (t), as shown in Fig. 4, is
approximated as an expansion of N Fourier series as follows [24]
N N
F (t) = a 0 + ∑ aj cos(jωt) + ∑ bj sin(jωt)
j=1 j=1 (12)

where ω = 2π /T , a0 = F0(Td/T ), aj = (F0/jπ)sin(2jπ (Td/T )), and

bj = (F0/jπ) − (F0/jπ)cos(2jπ (Td/T )).
Fig. 4. Periodic pulse train.
With Eqs. (11) and (12), the analytical solution of the steady-
state responses of ηi can be written as
N
M (q)Δq + K Δq = F (t)J T n (5) ηi(t) = ndi A 0 + ndi ∑ Aj cos(jωt − φj)
j=1 (13)
Eq. (5) represents the vibration model of the robot under per-
1
cussive impact forces. Matrices M and Jv depend on the config- where A0 = a 0 / k , Ai = k
ai2 + bi2 /(1 − r 2)2 + (2ζr)2 , r = iω/ωn ,
uration of the robot, force F (t) and its direction n depend on the φi = tan−1(2ζr /1 − r 2) + tan (bi /ai). Once vector η is obtained, the
−1
specific percussion operation at that configuration. responses of the joint displacements can be calculated from Eq.
(7).
3.2. Steady-state responses of joint displacements
3.3. Joint displacement and stress envelopes
Since the robot stays in a stationary configuration during the
percussive operations, the model described in Eq. (5) is a time- As shown in Fig. 5, the joint angular displacement Δq can be
invariant forced vibration system. Its solution can be written as related to the shaft shear strain, where r is the radius of the joint
Δq = Δq ss(t) + Δq 0(t) (6) shaft, l is the length, Δs is the arc length produced by Δq and γ is
the shear strain. First, Δs can be written in terms of Δq as
where Δq ss is the steady-state responses of the joint displacements
Δs = r Δq (14)
and Δq0 is related to the initial conditions. Since Δq0 can be
damped out with time, we only focus on the steady-state re- Since the arc displacement Δs is small, it can also be expressed
sponses. In particular, we present a method to find out the ap- in terms of the shear strain as
proximate analytical solution of the steady-state responses. First,
Eq. (5) is decoupled into n single-DOF vibration systems, and the Δs = γl (15)
external force acting on each single-DOF system is formulated.
From Eqs. (14) and (15), the shaft shear strain can be derived as
After solving the decoupled vibration systems, the responses of
the joint displacements can be easily computed. Δqr
γ=
For this purpose, we write the joint displacement vector Δq as l (16)

Δq = Sη (7) Further, based on Hooke's law, the torsional stress can be ob-
tained that
where S is the time-independent normal mode shape matrix and
can be obtained by solving the eigenvalue problem KS = MSD , in τ = Gγ (17)
which D is a diagonal matrix as
where G is the joint shaft shear modulus.
⎡ω 2 ⎤
⎢ 1 ⎥
⎢ ⎥ 3.4. Fatigue life estimate
D=⎢ ω22
⎥
⎢ ⋱ ⎥ The fatigue life of the robot joints under torsional vibrations
⎢⎣ ωn ⎥⎦
2
(8) can be estimated from the shear stresses of the joint shafts. In
practice, the robot will perform riveting at different positions in
where ωi (i = 1, 2, ... , n) is the ith natural frequency of the system.
Substituting Eq. (7) into Eq. (5), we have
Mdη + Kη = F (t)nd (9)

Where Md = ST MS , K d = ST KSη , nd = STJTv n , and the subscript d re-

presents the decoupled system. Matrices Md and Kd are both n × n
diagonal matrices and they can expressed as

⎡m1 ⎤ ⎡k ⎤
⎢ ⎥ ⎢ 1 ⎥
m2
Md = ⎢ ⎥, K d = ⎢ k2 ⎥
⎢ ⋱ ⎥ ⎢ ⋱ ⎥
⎢⎣ mn ⎥⎦ ⎢ ⎥
⎣ k n⎦ (10)

Then, Eq. (9) can be written as

̇ ̇ Fig. 5. Joint angular displacement and shear strain.
mi ηi + 2ς mi ki ηi + kiηi = F (t)ndi , i = 1, 2, …, n (11)
296 S. Nie et al. / Robotics and Computer-Integrated Manufacturing 37 (2016) 292–301

Table 1
Inertial properties of links.

Link m (kg) CG position (m)

(
I Kg m2 )
1 80 0,0,0.2 diag(0.9,1,1)
2 30 0.4,0,0 diag(0.6,6.5,6.5)
3 45 0.4,0,0 diag(0.8,9.5,9.5)

To account for different operating conditions, a relation be-

Fig. 6. Multiple loading fatigue S-N curve.
different directions over its workspace. It is well known that the
robot joint forces are robot's configuration dependent. In other
words, even under the same robot impact forces, the joint vibra-
tions will differ and the joint shear stresses will vary. Therefore, a
multiple-loading fatigue model must be applied. As shown in
Fig. 6, different shear stresses will produce different number of
cycles to failure. To account for this, the Palmgren-Miner cycle-
ratio summation rule, also called Miner's rule, is adopted here [25]
k ni
∑ i= 1 =c
Ni
where ni is the number of cycles at stress level τi , Ni is the number
of cycles to reach failure at stress level τi , and k is the number of
different riveting operations. The parameter c is determined by
experiment and found in a range 0.7 o co 2.2. Usually, c¼ 1 is
used.
The torsional stress τi induced by riveting is fluctuating under a
series of impact forces. There are a number of methods to quantify
the fatigue stress that corresponds to Ni. These methods include
Gerber, Goodman and Soderberg [25], considering peak stress,
valley stress, stress amplitude, mean stress, etc. For simplicity, the
peak stress is used here as the fatigue stress to calculate Ni.
tween the numbers of cycles ni needs to be established. In our
case, the number of cycle is equal to the number of hits generated
by the percussive rivet gun. Our riveting process research [26]
indicates that the number of hits for percussive riveting is a
function of rivet size and supply pressure. Therefore, if defining h1
as the number of hits per rivet for the smallest rivet under the
lowest workable pressure, the number of hits per rivet for differ-
ent sizes and/or different supply pressures can be expressed as
hi ¼aih1, where ai is the ratio between hi and h1. Now Eq. (18) can
be re-written as
hi
∑ p =1
Ni i (19a)
where pi is the total number of rivets that can be riveted under the
ith riveting operation. Note that ni ¼hipi. Furthermore, if defining
p1 as the number of rivets riveted under h1 operation, then mi can
be related to m1 as pi¼bip1, where bi is the ratio between mi and
m1. Now Eq. (19a) can be re-written as
(18) aibi
∑ h1p1 = 1
Ni (19b)
Since the number of hits per rivet, hi, can be determined ex-
perimentally or numerically [26] and Ni determined by a S–N
curve under the fatigue stress τi , m1 can be obtained from Eq. (19b)
as
⎛ aib ⎞
p1 = 1/⎜∑ i h1⎟
⎝ Ni ⎠ (20)
Finally, the total number of rivets, n, that can be riveted by a
percussive riveting robot before robot's fatigue failure can be

Fig. 7. (a) Schematic of the simulation robot and (b) CAD of the simulation robot with rivet gun.
 S. Nie et al. / Robotics and Computer-Integrated Manufacturing 37 (2016) 292–301
determined as
n= (∑ b )p
k
i=1 i 1 (21a)
For percussive riveting of the same rivet size under the same
gun supply pressure, Eq. (21a) can be further simplified as
n = kp1
(21b)
4. Numerical simulation
4.1. System parameters
Typically, an industrial robot is designed to have the first three
joints with the links for position control and the last three joints as
a wrist for orientation control. Therefore, from the fatigue point of
view, only the first three joints are considered. Fig. 7(a) shows a
3-DOF spatial robot with revolute joints for this simulation, which
mimics the Fanuc robot used for our percussive riveting as shown
in Fig. 7(b). The first joint rotates about a vertical axis and the
other two joints rotate about two parallel horizontal axes. The
orientation of the percussive rivet gun can be adjusted by rotating
the wrist. During riveting, gun's orientation is not changed.
The inertial and link-fixed local frames are established as
shown in Fig. 7(a). The inertial frame {O _xyz} is located at the base
of the robot, with the positive z axis vertically upward and the y
axis perpendicular to the plane containing the second and third
links. The zi axes of the local frames {Oi _x iyi zi} are along the rota-
tion directions of the corresponding joints, the x1 axis is parallel to
the x axis when the robot is stationary, and the x2 and x3 axes are
along the link length directions. The directions of the other re-
ference axes are determined from the right hand rule. The link
length vectors of O1O2, O2O3, and O3E can be written in local frames
as
l1b = [0, 0, l1]T , l2b = [l2, 0, 0]T , l3b = [l3, 0, 0]T
where the link lengths are l1 = 0.4 m and l2 = l3 = 0.8m. The masses
Fig. 8. Stress envelop of the flexible joints during percussive riveting.
297
mi , position vectors of the CG, and inertia tensors of the links Ii are
listed in Table 1. Note that these vectors and tensors are written in
the corresponding local frames. The maximum quantities of the
actuator torques are: m1 = m2 = 200 N-m, and m3 = 120 N-m. The
torsional stiffnesses of the three joints are: k1 = 2.4 × 105 N-m/rad
and k2 = k3 = 105 N-m/rad. Finally, the range of the input joint
angles is: θ1 ∈ [−π , π]rad, θ2 ∈ [−π /2, π /2]rad, and
θ3 ∈ [−7π /18, 7π /18] rad.
4.2. Joint displacement and stress envelopes
Based on the parameters given in Section 4.1, the vibration
equations given in Eq. (11) are established for each of the three
joints of the robot under simulation. The impact force generated
by the rivet gun is obtained through vibration measurement at
300 N. Fig. 8 shows a typical joint shear stress under percussive
riveting. It can be seen that the joint stress fluctuates.
The first set of simulation is to examine the effect of gun's or-
ientation. As shown in Fig. 9, four configurations are considered,
mimicking a vertical path of riveting four rivets. In this case, while
the x and y coordinates of the gun tip remain the same, the z
coordinate changes from 1.5 to 1.2, 0.9 and 0.6. At each given rivet
point the gun orientation is changed to investigate the effect of the
force direction. Since the impact force is generated along the gun
axis, the gun direction is the force direction. Several important
observations can be obtained from this simulation. First, all the
joint displacement envelops appear to be a double-ball like shape
with a symmetric plane and the maximum force direction is per-
pendicular to this plane. The said envelop is formed for each joint
by linking all the maximum joint displacements simulated under
different force directions. Second, the size of each joint displace-
ment envelop changes with configuration. As shown in Fig. 8, the
largest joint displacement envelop is at joint 2 for z¼ 1.5, then it
changes to joint 3 for z ¼1.2 and z¼0.9, and to joint 1 for z¼ 0.6.
Third, the force directions causing the maximum joint displace-
ment change with configuration and they are different for differ-
ent joints. As shown in Fig. 9, though the force direction for joint
1 do not change due to the vertical path in the x–z plane, the force
298 S. Nie et al. / Robotics and Computer-Integrated Manufacturing 37 (2016) 292–301

Fig. 9. Joint displacement envelopes in four configurations for percussive force (red lines: force directions for maximum displacements). (For interpretation of the references
to color in this figure legend, the reader is referred to the web version of this article).

Fig. 10. (a) Joint 1 distribution of maximum displacements throughout the workspace (red color indicates the high displacement and blue the low displacement, other colors
in between). (b) Joint 2 distribution of maximum displacements throughout the workspace (red color indicates the high displacement and blue the low displacement, other
colors in between). (c) Joint 3 distribution of maximum displacements throughout the workspace (red color indicates the high displacement and blue the low displacement,
other colors in between). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article).
 S. Nie et al. / Robotics and Computer-Integrated Manufacturing 37 (2016) 292–301
Fig. 11. Torsional S-N curve of steel.
Fig. 12. Rivet pattern and corresponding path.
Fig. 13. (a) Riveting forward. (b) Riveting forward.
299
directions for joint 2 and 3 both change from leaning the left to
leaning to the right.
The second set of simulation is to examine the change of the
maximum joint displacement over the robot workspace. Fig. 10
shows a snapshot of this simulation by focusing on the x–z plane,
with figure (a), (b) and (c) for joint 1, 2 and 3, respectively. The
workspace is identical for the three figures as it is in the x–z plane
of the robot's task space. But, the locations of the maximum joint
displacement vary. In Fig. 10(a–c), red color indicates the high
displacement and blue the low displacement, other colors in be-
tween. First, it can be seen that joint 1 experiences the largest
displacement when the robot reaches the left and right end in the
x direction to perform percussive riveting. At these configurations
the robot is fully extended in the horizontal direction and the
implusive riveting forces acting on the tool tip in the x direction
will be fully transmitted to produce the largest torque to joint 1,
thereby causing the largest displacement of joint 1. Second, it can
be seen that joint 2 experiences the largest displacement when the
robot reaches the up and low end in the z direction to perform
percussive riveting. At these configurations the robot is fully ex-
tended in the vertical direction and the implusive riveting forces
acting on the tool tip in the z direction will be fully transmitted to
produce the largest torque to joint 2, thereby causing the largest
displacement of joint 2. Third, it can be seen that joint 3 experi-
ences the largest displacement when the robot performs percus-
sive riveting in the middle area. At these configurations the robot
is not fully extended, and the implusive riveting forces acting on
the tool tip will be mainly absorbed by the nearest joint 3, therey
causing the largest displacement of joint 3.
4.3. Robot fatigue life
As explained in Section 3.3, the joint shear strains are de-
termined from the joint displacements and the joint fatigue
stresses are determined from the joint shear strains. To determine
the fatigue life, an S–N curve of the used material is needed. For
this study, the S–N curve is obtained through fitting the data
presented in [17] for steel. This S–N data was generated from the
experiment for the material similar to those used for common
robot joint shafts. The equation for the fatigue curve is (N) = a⁎N b .
By fitting it is found that a ¼18930, and b¼  0.3548. The fitted
curve is given in Fig. 11.
300 S. Nie et al. / Robotics and Computer-Integrated Manufacturing 37 (2016) 292–301

Fig. 14. (a) Riveting left. (b) Riveting left.

Fig. 15. (a) Riveting upward. (b) Riveting upward.

directions. There are in total 15 r ivets and the numbers in Fig. 12

Table 2
Fatigue life analysis of the simulation robot indicate the sequence of riveting, i.e. the rivet path. For all riveting
the same size of rivets is used. To predict robot fatigue life, the
Direction Forward fatigue stress of each direction must be determined to find the
corresponding cycle numbers to failure. The number of hits per
Joint Joint 1 Joint 2 Joint 3
rivet is found through experiment to be 12. Finally, Eq. (22) is used
Stress Level(MPa) 32.81 37.46 9.43 to determine the maximum number of rivets that can be riveted
Ni 3.08e þ06 2.38e þ 06 3.44e þ 07 before the robot's fatigue failure.
Direction Left
Fig. 13(a) shows the riveting in the forward direction. The im-
Joint Joint 1 Joint 2 Joint 3
Stress Level(MPa) 35.56 37.91 11.13
pact force direction in this case is n ¼[0, 1]. Fig. 13(b) shows the
Ni 2.64e þ 06 2.33eþ06 1.10e þ 07 maximum stresses of the three joints over riveting 15 rivets at 15
Direction Upward rivet spots. The mean value of the fatigue stresses over 15 rivets
Joint Joint 1 Joint 2 Joint 3 are found as 32.81, 37.46 and 9.43 MPa, which are used to find Ni
Stress Level(MPa) 0 54.86 56.23
for the three joints, respectively. Fig. 14(a) depicts the riveting in
Ni +∞ 1.14e þ 06 1.09e þ 06
the left direction, with the impact force direction n¼[0 1 0]. Fig. 14
(b) shows the maximum stresses of the three joints over 5 rivets.
In this simulation, a riveting task including three directions, The mean value of the fatigue stresses over 15 rivets are found as
forward, left and upward, is considered for fatigue life prediction. 35.56, 37.91 and 11.13 Mpa for the three joints, respectively. Fig. 15
The same rivet pattern as shown in Fig. 12 is applied to all three (a) displays the riveting in the left upward direction, and the im-
pact force direction n ¼[0 0 1]. Fig. 15(b) shows the maximum
 S. Nie et al. / Robotics and Computer-Integrated Manufacturing 37 (2016) 292–301
stresses of the three joints over 5 rivets. The mean value of the
fatigue stresses over 15 rivets are found as 0, 54.86 and 56.23 Mpa
for the three joints, respectively. Table 2 lists all the fatigue
stresses for the three directions and corresponding cycle number
to failure. Now, the total number of rivets that can be riveted over
the robot's fatigue life is determined as
3 [2] A. Barber, Pneumatic Handbook, 8th edition,. Elsevier, Oxford (1997) 299–330.
n=
h1⁎((1/N1) + (1/N2) + (1/N3)) (22)
Here, h1 is 12, as mentioned before. For joint 1, n is 355384; for
joint 2 and 3, n are 144797 and 253420, respectively. Finally, the
smallest value among the three joints is used to predict the robot's
fatigue life, i.e. there are 144797 rivets that can be riveted before
the robot's fatigue failure.
5. Conclusions
A complete method for modeling and simulation for fatigue life
analysis of robots with flexible joints under percussive impact
forces is presented in this paper. Though a conventional modeling
method is adopted for modeling of flexible joint robots, a forced
vibration solution is provided to this problem by including the
impact forces generated by the percussive gun, projecting them
onto the joint space and treating them in terms of the Fourier
transform. As a result, the joint angular displacements can be
solved using a standard vibration method. Then the joint stresses
can be determined through Hooke's law. To consider the stress
variations caused by the robot operating at different poses as well
as different rivets, a multiple-loading fatigue model is applied
from which an equation is derived to determine the total number
of the rivets that can be riveted before robot's fatigue failure. Based
on simulation using our model, the following observations are
received. First, the joint torsional stresses vary with robot's posi-
tion and orientation. Second, no joint will always experience the
maximum stress and the joint stress dominancy also varies with
robot's position and orientation. Third, for a given point, the rivet
gun direction considerately affects the joint stresses. Fourth, the
fatigue life of each joint is different; therefore robot's fatigue life
should be evaluated based on the shortest joint fatigue life.
However, there exist limitations for the proposed model. The fa-
tigue prediction model presented is based on the torsional fatigue
of the shaft made of steel. To give a more realistic prediction, the
exact material should be identified to run a torsional fatigue test
on it. Moreover, a more complete modeling and analysis method
should also include joint tolerance and clearance.
Acknowledgment
This work is supported by the Shanghai Municipal Science and
Technology Commission (Granted nos.12111101004,13DZ1101700)
301
and the Shanghai Key Laboratory of Intelligent Manufacturing and
Robotics (Granted no.ZK1304).
References
[1] B. Siciliano, O. Khatib, Springer Handbook of Robotics, Springer-Verlag, Berlin
Heidelberg (2008) 229–244.
[3] F.C. Campbell, Manufacturing Technology for Aerospace Structural Materials,
Elsevier, New York (2006) 495–537.
[4] S.-L. Peng, Characterization and ergonomic design modifications for pneu-
matic percussive rivet tools,, Int. J. Ind. Ergonom. 13 (3) (1994) 171–187.
[5] J.G. Cherng, M. Eksioglu, K. Kizilaslan, Vibration reduction of pneumatic per-
cussive rivet tools: mechanical and ergonomic re-design approaches,, App.
Ergon. 40 (2) (2009) 256–266.
[6] B. Rooks, Assembly in aerospace features at IEE seminar, Assem. Autom. 25 (2)
(2005) 108–111.
[7] B. Morey, Robotics seeks its role in aerospace, Manuf. Eng. 139 (4) (2007)
AAC1–AAC6.
[8] P. Webb, S. Eastwood, N. Jayaweera, Y. Chen, Automated aerostructure as-
sembly, Ind. Robot: Int. J. 32 (5) (2005) 383–387.
[9] B.J. Glass, A. Dave, C.P. McKay, G. Paulsen, Robotics and Automation for ‘Ice-
breaker’, J. Field Robot. 31 (1) (2014) 192–205.
[10] N. Jayaweera, P. Webb, Adaptive robotic assembly of compliant aero-structure
components, Robot. Comput. Integr. Manuf. 23 (2) (2007) 180–194.
[11] R.S. Kadam, Vibration Characterization and Numerical Modeling of a Pneu-
matic Impact Hammer, Virginia Polytechnic Institute and State University,
Blacksburg, VA, 2006, MS thesis.
[12] W.A. Bloxsom, Modeling of the Reciprocating, Pneumatic Impact Hammer,,
University of Nevada, Reno, NV, 2003, Ph.D. thesis.
[13] Q. Quan, P. Li, S. Jiang, Development of a Rotary-percussive Drilling Mechanism
(RPDM), Proc. IEEE Int. Conf. Robot. Biomim. (2012) 950–955.
[14] T.J. Johnson, R. Manning, D.E. Adams, R. Sterkenburg, K. Jata, Diagnostics of
tool-part interactions during riveting on an aluminum aircraft fuselage, J.
Aircraft 43 (3) (2006) 779–786.
[15] Y. Li, F. Xi, R.P. Mohamed, K. Behdinan, Dynamic analysis for robotic integration
of tooling systems, ASME J. Dyn. Syst. Measur. Control 133 (4) (2011) 041022.
[16] C.-D. Garron, Prolonging a robot's life expectancy, Ind. Robot: Int. J. 22 (2)
(1995) 16–17.
[17] Masami Wakita and Takanori Kuno and Ayamitsu Amano, Akihiko Nemoto and
Katsushi Saruki and Keisuke Tanaka,“Study on Predicting Formula for Tor-
sional Fatigue Strength of Spring Steel – Effect of Environment, Notch and
Hardness”.
[18] S. Dwivedy, K. Eberhard P., Dynamic analysis of flexible manipulators, a lit-
erature review, Mech. Mach. Theory 41 (7) (2006) 749–777.
[19] M. Benosman, G. Le Vey, Control of flexible manipulators: a survey, Robotica
22 (5) (2004) 533–545.
[20] N.H. Rahimi, M. Nazemizadeh, Dynamic analysis and intelligent control
techniques for flexible manipulators: a review, Adv. Robot. 28 (2) (2014)
63–76.
[21] Z.-C. Du, Y.-Q. Yu, L.-Y. Su, Analysis of dynamic stress and fatigue property of
flexible robots, Proc. IEEE Int. Conf. Robot. Biomim. (2006) 1351–1355.
[22] C.-O. Miclosina, C.V. Campian, Fatigue analysis of low level links of a parallel
topology robot guiding device mechanism, Appl. Mechan. Mater. 162 (2012)
98–105.
[23] T. Song, et al., A comparison Study of Algorithms for Surface Normal De-
termination Based on Point Cloud, Precision Enginnering, 2014.
[24] R. Bracewell, The Fourier Transform & Its Applications, 2nd Edition, McGraw-
Hill, 1978.
[25] R.G. Budynas, J.K. Nisbett, Chapter 6 Fatigue Failure Resulting from Variable
Loading, in Shigley's Mechanical Engineering Design, McGraw-Hill Science/
Engineering/Math, 2010.
[26] Li, Y., Xi, F., and Behdinan, K., Modeling and Simulation of Percussive Riveting
for Robotic Automation, ASME Journal of Computational and Nonlinear Dy-
namics, Vol. 5, No. 2, 021011.