Transaction B: Mechanical Engineering

Vol. 16, No. 6, pp. 479{495

c Sharif University of Technology, December 2009

Motion Equations Proper for Forward Dynamics

of Robotic Manipulator with Flexible Links by
Using Recursive Gibbs-Appell Formulation
M.H. Korayem1; and A.M. Shafei1
Abstract. In this article, a new systematic method for deriving the dynamic equations of motion
for exible robotic manipulators is developed by using the Gibbs-Appell assumed modes method. The
proposed method can be applied to the dynamic simulation and control system design of exible robotic
manipulators. In the proposed method, the link de ection is described by a truncated modal expansion.
All the mathematical operations are done by only 3  3 and 3  1 matrices. Also, all dynamic expressions
of a link are expressed in the same link local coordinate system. Based on the developed formulation, an
algorithm is proposed that recursively and systematically derives the equation of motion, then this method
is compared with the recursive Lagrangian method. As shown, this method is computationally simpler
and more ecient and it reduces a large amount of computational complexity. Finally, a computational
simulation for a manipulator with two elastic links is presented to verify the proposed method.
Keywords: Manipulator; Flexible link; Recursive; Gibbs-Appell; Complexity.

INTRODUCTION of exible robotic manipulators are the nite element

method [4] and the assumed modes method [5-9]. The
The derivation of dynamic equations of motion describ- nite element method is a general method and can be
ing the dynamic behavior of robotic manipulators is applied to manipulators with complex shaped links.
necessary for dynamic simulation and control system But, this method requires sophisticated software for
design. Today, many systematic methods can be used performing assembly and the order reduction of the
for deriving the dynamic equations of robotic manip- element equations.
ulators [1-3]. But, these methods are only suitable The assumed mode method of modeling exible
when the individual links of a robotic manipulator are manipulators is mainly presented by Book [5]. He
assumed rigid. represented the link deformation and kinematics of rev-
Based on recent advances in robot utilization olute joints with a 4  4 matrix and used modal analysis
and also the demand for faster robots with great for link deformations. This method of formulation had
quality, a light robot usage idea is represented. Robots acceptable eciency in comparison with other methods
with elastic links are introduced as a solution for the of that time. King applied Walker-Orin's method,
deformation phenomena in light robots with heavy based on Newton-Euler formulation, to improve Book's
loads. In this case, deformation causes accuracy method [6]. But, his method still su ered from great
reduction and system instability. Therefore, there is an computational complexity. Jin and Sankar also have a
obvious requirement for a complete dynamical model systematic approach for elastic links [7]. They obtained
for this kind of robot to control light links at high dynamical equations by using Lagrange formulation
velocity and in heavy load situations, appropriately. and the modes approach assumption. In this method
The two main approaches for the dynamic modeling 3  3 matrixes are used for computations and the
1. Department of Mechanical Engineering, Iran University of
results are simulated for a robot with one link. The
Science and Technology, Tehran, P.O. Box 13114-16846, computations, however, are massive.
Iran. Highly ecient multi- exible-body methods have
*. Corresponding author. E-mail: [email protected] been previously presented by Anderson [10] and Baner-
Received 9 September 2008; received in revised form 14 January jee [11] based on Kane's Method with many other com-
2009; accepted 14 April 2009 parably ecient multi- exible-body routines developed
480 M.H. Korayem and A.M. Shafei

by E. Haug, J. Angeles, R. Singh, R. Schwertassek, this method is computationally simpler and more
A. Jain, R. Wehage, J. Ambrosio and others [12]. ecient and it reduces a large amount of computational
Many of these methods are so-called O(N ) routines, complexity. Finally, for veri cation of this method, a
being able to form equations of motion with an overall computational simulation for a manipulator with two
cost that increases only linearly with the number of elastic links is presented.
system degrees of freedom N , (for rigid body systems).
For exible body systems, this overall cost (equations
formation) is adjusted somewhat, being approximated KINEMATICS OF FLEXIBLE LINK
as O(n2f  m2 ). In this section, the kinematics of a chain of n elastic
Dynamic equations of motion by the Gibbs- links is taken into consideration. The coordinate
Appell formulation begin with a de nition of Gibbs' system of every link is attached according to the rules
function (acceleration energy) [13]. Then, a set of developed by Denavit and Hartenberg. X0 Y0 Z0 is the
independent quasi velocities (linear combination of coordinate system that is attached to the base of the
generalized velocities) should be selected. By taking manipulator and can be considered as the reference
the derivative of the Gibbs' function, with respect to coordinate system. Because of the elastic property of
quasi accelerations (time derivate of quasi velocities), the links, two rotations occurred one of which is in the
and equalizing them with generalized forces, these joints and the other of which is in the links. It is useful
equations will be obtained. But, this method has been to separate the transformations due to the joints from
the least used for resolution of the dynamic problem the transformations which are due to the exible links.
of manipulating robots. In the eld of robotics, So, we allocate two coordinates system to each link.
Popov proposed a method later developed by Vuko- xi yi zi is a coordinate system on link i whose origin is
bratovic and Potkonjak in which the G-A equations located at the beginning of this link, but x^i y^i z^i is the
were used to develop a closed form representation of coordinate system that is attached to the end of this
high computational complexity [14]. This method was link. When link i has no deformation, the axes of x^i y^i z^i
used by Desoyer and Lugner to solve, by means of are parallel to the axes of xi yi zi .
a recursive formulation, O(n2 ), the inverse dynamic In Figure 1, the arbitrary point, Q, is shown.
problem, using the Jacobian matrix of the manipulator The position of this point with respect to the ith
with the purpose of avoiding the explicit development body's local reference system is expressed by i~rQ=Oi .
of partial derivatives [15]. Another approach was sug- To incorporate the de ection of the link, the approach
gested by Vereshcahagin, which proposed manipulator of modal analysis is used. So:
motion equations from Gauss' principle and Gibbs'
function [16]. This approach was used by Rudas Xm i
i~r = ~ +  (t)~rij (); (2)
and Toth to solve the inverse dynamic problem of Q=Oi j =1 ij
robots [17]. Recently, Mata et al. presented a formu-
lation of order O(n) which solves the inverse dynamic where ~ = f 0 0gT and ~rij = fxij yij zij gT . Also,
problem and establishes recursive relations that involve  is the undeformed distance between the origin, Oi ,
a reduced number of algebraic operations [18]. and the point, Q; xij ; yij and zij are the displacement
In this article, a new systematic method for components of j mode of the ith link; ij is the time
dynamic modeling of exible robotic manipulators varying amplitude of mode j of link i; and mi is the
is developed using the Gibbs-Appell assumed modes number of modes used to describe the de ection of
method. In this method, the equation of motion for link i.
exible robotic manipulators is written in the following By using the rotation matrix, j Ri , we can express
form: the arbitrary vector, i~a, in every coordinate system, j ,
! in the following form:
I ()~ = Re; (1)
j~a = j R i~a: (3)
i
where I () is the inertia matrix of the whole system;
~ denotes the vector of the generalized coordinate
 As noted above, it is better to separate the rotations,
!
containing joint and de ection variables; and Re is the due to joints from de ections. So, j Ri can be presented
vector composed of the strain, gravitational, Coriolis, recursively as follows:
centrifugal forces or torques and also the generalized
forces or torques exerted to the joint and link vari- jR = j Ri 1 Ei 1 Ai ; (4)
i
ables. Also, a recursive algorithm is proposed that
systematically derives the equation of motion of elastic where Ai is the rotation matrix of the ith joint that
robotic manipulators. Then, this method is compared shows the orientation of the xi yi zi coordinate system
with the recursive Lagrangian method and, as shown, with respect to x^i 1 y^i 1 z^i 1 . The coecients of this
Gibbs-Appell Formulation for Flexible Link Manipulators 481

Figure 1. Manipulator with elastic links.

matrix can be presented by dot products of a pair of SYSTEM'S ACCELERATION ENERGY

unit vectors as follow: (GIBBS FUNCTION)
2 3 In this section, the expression for the system's accelera-
xi :x^i 1 yi :x^i 1 zi :x^i 1 tion energy is developed for use in Gibbs-Appell's equa-
Ai = 4 xi :y^i 1 yi :y^i 1 zi :y^i 1 5 : (5) tions. First, the acceleration energy for a di erential
xi :z^i 1 yi :z^i 1 zi :z^i 1 element is written. Then, integration of this di erential
Also, Ei is the ith link rotation matrix that shows the acceleration energy over the link gives the link's total
orientation of x^i y^i z^i the coordinate system with respect contribution. Summation over all the links provides
to xi yi zi . Like Ai , this matrix is also composed of dot the total acceleration energy. The acceleration energy
products of a pair of unit vectors, but because of the of a point on the ith link is:
small angles between these vectors, Ei is simpli ed in 1
the following form: dsi = dm(i~rQ T :i~rQ ); (9)
2
2 3 where dm is the di erential mass of point Q and i~rQ is
1 zi yi the absolute acceleration of di erential element Q that
Ei = 4 zi 1 xi 5 ; (6) is expressed in the ith body's local reference system:
yi xi 1
i~r =i ~r +i ~r i ~ i ~r_
Q Oi Q=Oi + 2 ! i Q=Oi
where xi ; yi and zi are in nitesimal rotations of
x^i y^i z^i , with respect to xi ; yi and zi axes, respectively. It + i !~_ i i ~rQ=Oi + i!~ i  (i !~ i i ~rQ=Oi ): (10)
should be noted that all the angles in Ei are evaluated
at  = li , where li is the length of the ith link. Now, In the above expression, i~rOi is the absolute acceler-
we de ne i ~i as follows: ation of the origin of the ith body's local reference
system, i !~ i and i !~_ i are angular velocity and angular
i ~
i = fxi yi zi gT : (7) acceleration of the ith link, respectively and i~r_Q=Oi and
i~r
Q=Oi are the velocity and acceleration of di erential
These small angles can be represented by truncated element Q, with respect to the origin of the ith body's
modal expansion as follows: local reference system which will be obtained by taking
the time derivative of Equation 2 as follows:
Xm i Xmi
i ~ =  (t)~ij ; (8) i~r_
i j =1 ij Q=Oi = _ (t)~rij ();
j =1 ij
(11)
Xmi
where ~ij = fxij yij zij gT . By taking the time i~r =  (t)~rij (): (12)
Q=Oi j =1 ij
derivative of i ~i , the angular velocity and acceleration
of x^i y^i z^i the coordinate system, with respect to xi yi zi , By substituting Equation 10 in Equation 9 and inte-
will be obtained. grating over the link, one can obtain the link's total
482 M.H. Korayem and A.M. Shafei

acceleration energy. In this paper, it is assumed that tensors representation of the i~rQ=Oi ; i~r_Q=Oi ; i~rQ=Oi and
the links are slender beams. For slender beams, dm = i!
~ i vectors. For developing an expression for S , these
d where  is mass per unit length. So, one can vector relations, ~a:~b = ~b:~a, ~a  ~b = a~~b and (~a  ~b):~c =
integrate over  from 0 to li . Only the terms in i~rQ=Oi ~a:(~b  ~c) are frequently used. By interchanging the
and its derivatives (i~r_Q=Oi ; i~rQ=Oi ) are functions of  integration and summation in Equations 14 to 22, one
for this link. Thus, the integration can be performed obtains:
without knowledge of i !~ i ; i !~_ i and i~rOi . Summing over
all n links, one nds the system's acceleration energy Xm i
~
iB
1i =  ~" ;
j =1 ij ij
(23)
to be:
Xn Z li Xmi
S= i=1 0
dsi B2i = _ "~ ;
j =1 ij ij
(24)
Xn 1 i T i Xmi
S= M ~r : ~rOi + i~rOiT :i B~ 1i
i=1 2 i Oi
B3i = M
gi r ri +  "~ ; (25)
j =1 ij ij

2i~rOi T B2i i !~ i i~rOi T B3i i !~_ i i~rOi T i !~ i B3i i !~ i Xmi Xmi

B4i = j =1
 
k=1 ij ik
sC
ijk ; (26)
1
+ B 4i 2i !~ i T :i B~ 5i + i !~_ Ti :i B~ 6i Xmi Xmi
2 ~
iB =  _ ~c ; (27)
5i j =1 k=1 ij ik ijk
1 _ T i_
i! ~ i T B 7i i ! ~ i + 2i ! ~_ i T B8i i !
~ i + i!
~ B !~
2 i 9i i Xm i
~
iB
6i =  ~ ;
j =1 ij ij
(28)
+ i !~_ i T i !~ i B9i i !~ i + irrelevant terms; (13)
Xmi
where: B7i = 
j =1 ij ij
; (29)
Z li
~
iB
1i = i~rQ=Oi d; (14) B8i =
Xmi
_ ; (30)
0 j =1 ij ij
Z li
Xm i Xmi
B2i = i r~_Q=Oi d; (15) B9i = ci +  c T +  ; (31)
0 j =1 ij ij k=1 ik ik
Z li
B3i = i r~Q=Oi d; (16) where:
0
Xmi
Z li ~ ij = ~cij +  ~c ;
k=1 ik ikj
(32)
B4i = i~rQ=Oi :i~rQ=Oi d; (17)
0 Xm i
= cij +  mc : (33)
Z li ij k=1 ik ikj
~
iB
5i = i r~Q=Oi i~r_Q=Oi d; (18)
0 By de nition, ~ and r~ij are skew-symmetric tensors
Z li associated with ~ and ~rij vectors. The expressions of
iB
6i = i r~Q=Oi i~rQ=Oi d; (19) ~"ij ; "~ij ; M
g s m
i rri ; cijk ;~cijk ;~cij ; cij ; cijk ; ci that appeared
0 in Equations 23 to 33 can be written in the following
Z li form:
B7i = i r~Q=Oi T i r~Q=Oi d; (20)
0 Z li
Z li ~"ij = ~rij d; (34)
0
B8i = i r~Q=Oi T i r~_Q=Oi d; (21)
0 Z li
Z li "~ij = r~ij d; (35)
B9i = i r~Q=Oi T i r~Q=Oi d: (22) 0
0
Z li
In Equation 13, Mi is the total mass of the ith link. M
gi rr = ~d; (36)
Also i r~Q=Oi , i r~_Q=Oi , i r~Q=Oi and i !~ i are skew-symmetric i
0
Gibbs-Appell Formulation for Flexible Link Manipulators 483

Z li
sc = ~rij T ~rik d; (37) Here, it should be noted that, in the above expres-
ijk sion, this property of the skew-symmetric matrix, in
0
Z li which aT = a is used. The partial derivative of
~cijk = r~ij ~rik d; (38) Gibbs' function with respect to jf is more complex,
0 because in addition to i~rOi and i !~_ i , the expressions
Z li of i B~ 1i ; B4i ; i B~ 5i ; i B~ 6i and B7i are also functions of
~cij = ~~rij d; (39) de ection variables. So, the expression of @@S jf can be
0 presented as follows:
Z li
cij = ~ T r~ij d; (40) @S Xn @ i~rOi T  i 
0  =  Mi ~rOi + i B~ 1i
@ jf i =j +1 @ jf
Z li
mc r~ij T r~ik d;
ijk = (41) 2B2i i !~ i ~_ i
B 3i i ! i!
~ i B3i
i!
~ i )
0
Z li Xn @i!~_ i T  i 
ci = ~T ~d: (42) + i=j +1 @ jf
B3i ~rOi + i B~ 6i
0
Now, it should be noted that B9i has a unit of inertia + 2B8i i !~ i + B9i i !~_ i i!
~ i! )
matrix. For example, its rst term (ci ) represent rigid- i B9i ~ i
body-inertia terms. It can also be shown that m cijk = Xmj Xmj
m c T . The terms de ned in Equations 23 to 33 are
ikj +  sc
k=1 jk jfk
2j !~ j T _ ~c
k=1 jk jfk
easily simpli ed if one link in the system is considered
rigid (mi = 0). Furthermore, the expression for B9i j!
~ j T jf j !
~j + j ~rOj T ~"jf + j !~_ j T ~ jf : (44)
has a term of order 2 , which is small and a candidate
for later elimination [5]. Finally, the integration of the An additional simpli cation of @S arises, due to the
@ jf
modal shape products in Equations 34 to 42 can be fact that scjfk = scjkf .
done o -line one time for a given link structure.
SYSTEM'S POTENTIAL ENERGY
Derivatives of Acceleration Energy
The potential energy of the system arises from two
G-A equations are obtained by taking the derivative of sources:
Gibbs' function, with respect to generalized accelera-
tions (qj ; jf ): 1. Potential energy due to gravity,
@S @S 2. Potential energy due to elastic deformations.
; :
@ qj @ jf The e ect of gravity on manipulators can be
In Equation 13, there was a term named an irrelevant considered simply by putting 0~rO0 = ~g, where ~g is the
term. In fact, in Gibbs' function, the terms that are acceleration of gravity. Under these circumstances, we
not functions of qj and jf can be eliminated, because can assume that the base of the manipulator has an
they have no role in construction of the derivative of acceleration of 1 g to the top. So, the e ect of gravity
acceleration energy. has been considered without additional computations.
In Gibbs' function, only i~rOi and i !~_ i are functions To express the strain potential energy stored in
of qj . So, the partial derivative of Gibbs' function with the ith link, let us assume that the assumptions of
respect to qj becomes: the classical beam (Euler-Bernoulli) hold. So, the
strain potential energy will be expressed in terms of
@S Xn @ i~r T  de ections and rotations as follows:
= i=j +1 Oi Mi i~rOi + i B~ 1i
@ qj @ qj Z li "    2 2
1 @ui 2 @ wi
 Vei = EA + EIy
2B2i i !~ i ~_ i
B 3i i ! i!
~ i B3i i !~ i 2 0 @ @2
 2 2   #
Xn ~_ i T  i 
@i! @ vi @xi 2
+ B3i ~rOi + i B~ 6i + EIz + GIx d; (45)
i=j @ qj @2 @

+ 2B8i i !~ i + B9i i !~_ i + i !~ i B9i i !~ i : (43) where EIy and EIz are the bending sti ness in the OY
and OZ directions, respectively; EA is the extensional
484 M.H. Korayem and A.M. Shafei

sti ness; GIx is the torsional sti ness; ui ; vi and wi Also, Kxikl ; Kyikl and Kzikl are de ned as follows:
are the de ections in the OX; OY and OZ directions, Z li
respectively; and xi is the rotation in the OX direction @xil () @xik ()
Kxikl = GIx d; (52)
as shown in Figure 2. 0 @ @
It is easy to show that the following relations Z li
between the component of de ections and rotations @yil () @yik ()
exist: Kyikl = EIy d; (53)
0 @ @
@vi Xmi @yij Z li
@zil () @zik ()
zi = = j =1 ; (46) Kzikl = EIz d: (54)
@ @ @ @
0
@wi Xmi @zij It should be noted that Kikl = Kilk . For deriving the
yi = = ; (47)
@ j =1 @ dynamic equation of motion, the partial derivatives of
where yi and zi are the rotations in OY and OZ strain potential energy with respect to the generalized
directions, respectively. coordinate is needed. Upon taking the partial deriva-
By substituting Equations 46 and 47 in Equa- tive with respect to qj , one obtains:
tion 45 and ignoring the strain potential energy due @Ve
to axial deformation, in comparison with the strain = 0: (55)
@qj
potential energy due to bending and torsion [5], the
expression for Vei is simpli ed as follows: But taking partial derivatives with respect to jf
Z li      results in:
1 @yi 2 @zi 2 @Ve Xmj
Vei = EIy + EIz = k=1 jk Kjkf ; (56)
2 0 @ @ @jf
  
@xi 2 where Kjkf can analytically or numerically be deter-
+ GIx d: (48)
@ mined.
As noted previously, angles xi ; yi and zi can be
presented with a truncated modal approximation. For DERIVATION OF DYNAMIC EQUATIONS
example the rotation about the OX axis is presented OF MOTION USING G-A'S FORMULATION
as follows: The components of the complete equations of motion
Xm i in G-A's formulation, except for the external forcing
xi () =  (t)xik ();
k=1 ik
(49) terms, have been evaluated in Equations 43 and 55
for the joint equations and in Equations 44 and 56
where xik is the angle corresponding to the kth mode for de ection equations. The generalized force in joint
of link i at point . By substituting the achieved equations is the torque, j , that applies to joints.
expressions of xi ; yi and zi in Equation 48, the strain But, in de ection equations, the corresponding gener-
potential energy for the whole system will be obtained alized force will be zero, if the corresponding modal
as follows: de ections or rotations have no displacement at those
1 Xn Xmi Xmi locations where external forces are applied [5]. So, with
Ve =   K ; (50) this assumption, the dynamic equation of motion in G-
2 i=1 k=1 l=1 ik il ikl
A's formulation will be completed as follows:
where:
1. The joint equations of motion:
Kikl = Kxikl + Kyikl + Kzikl : (51)
@S
= j : (57)
@ qj
2. The de ection equations of motion:
@S @Ve
+ = 0: (58)
@ jf @jf
The above equations are in the form of inverse dynamic.
In this type of dynamic, the forces exerted by the
actuators are obtained algebraically for certain con-
Figure 2. De ections and rotations of a link. gurations of the manipulator (position, velocity and
Gibbs-Appell Formulation for Flexible Link Manipulators 485

acceleration). On the other hand, the forward dynamic and i~zi = f0 0 1gT . By taking the time derivative of
problem computes the acceleration of the joints of the Equation 63, the expression of angular acceleration will
manipulator, once the forces exerted by the actuators be obtained:
are given. This problem is part of the process that must  
be followed to perform the simulation of the dynamic ~_ = i R
i!
i ~_ + i 1 ~ + i 1 !
i 1!
i 1 i 1~  i 1 ~_ i 1 i 1 i 1
behavior of the manipulator. This process is completed
 
after it calculates the velocity and position of the + i Ri i 1! i 1 ~_  i~zi q_i + i~zi qi : (65)
joints by means of a process of numerical integration 1 ~ i 1+ i 1
in which the acceleration of the joints and the initial
con guration are data input to the problem [15]. In the above expression, i 1 ~i 1 (li 1 ) is the angular
acceleration that is produced because of the elasticity
FORWARD DYNAMIC EQUATIONS OF of the i 1th link:
MOTION i ~ (l ) =
Xmi
 (t) ~ (l ): (66)
i i j =1 ij ij i
In this section, the rst step will extend the equations
of i~rOi and also i !~_ i . These equations are used to Now, by having i~roi and i !~_ i in recursive form, we can
separate the second derivatives of joint variables and convert them in summation form as follows:
de ection variables from the dynamic equations of Xi 1
motion. i~r = i R k ~r
Oi k=1 k Ok+1 =Ok
The absolute acceleration of the origin of the ith
body's local reference system in recursive form can be Xi 1  
presented as follows: + k=1
~_
iR k !
k k  k~rOk+1 =Ok + i~rOv;i ; (67)
 Xi
i~r = i Ri i 1~r + i 1~rOi =Oi 1 + 2i 1 !~ i
Oi 1 Oi 1 1 ~_ i
i! = k=1
i R k ~z q
k k k

 i 1~r_Oi =Oi 1 + i 1 !~_ i 1  i 1~r

Oi =Oi 1 +
Xi 1
i R k ~
k k (lk ) + ~_ v;i ;
i! (68)
k=1



+ i 1 !~ i 1 i
i 1!
~ 1~r ; (59)
i 1 Oi =Oi 1 where:
Xi 1  
where: i~r
Ov;i =2 iR k !
~ k k  k~r_Ok+1 =Ok
Xmi k=1
i~r
Oi+1 =Oi = ~li +  (t)~rij (li );
j =1 ij
(60) Xi 1


+ iR
k=1 k
k!
~k  k!
~ k  k~rOk+1 =Ok ; (69)
Xmi
i~r_ = _ (t)~rij (li ); (61)
Oi+1 =Oi j =1 ij
Xi 1
i~r
Xmi ~_ v;i
i! = iR k !
~ k k  i Rk+1 k+1~zk+1 q_k+1
Oi+1 =Oi =  (t)~rij (li );
j =1 ij
(62) k=1
Xi 1
+ i R k ~_  i Rk+1 k+1~zk+1 q_k+1 : (70)
k k
and also ~li = fli 0 0gT . Before developing an k=1
expression for angular acceleration, we should present In fact, i~rov;i and i !~_ v;i are those constructive terms of
angular velocity, because by taking its time derivative, i~r and i !
angular acceleration will be obtained. The angular oi ~_ i that do not contain the second derivatives
velocity of the ith link is the same as the i 1th link of joint variables and de ection variables. By having
i~r and i ! ~_ i in summation form, the calculation of par-
plus two new components, one of which (i~zi q_i ), comes oi
from the angular velocity of the ith link and the other tial derivatives that appeared in the dynamic equations
of motion can be done as follows:
(i 1 ~_i 1 ) is produced due to the elasticity of the i 1th
link. So, the expression of angular velocity can be @i!~_ i i j
presented as follows: = Rj ~zj ; (71)
@ qj
 
i 1 ~_ (l )
i!
~i = i Ri 1
i 1!
~ i 1+ i 1 i 1 + i~zi q_i ; (63) @i!~_ i i ~
 = Rj jf (lj ); (72)
@ jf
where:
i ~_ (l ) =
Xmi @ i~rOi i j
i i _ (t)~ij (li ); (64) = Rj ~zj  i~rOi =Oj ; (73)
j =1 ij @ qj
486 M.H. Korayem and A.M. Shafei

@ i~rOi i n
X
= Rj ~rjf (lj ) + i Rj ~jf (lj )  i~rOi =Oj+1 ; (74) j
k=
j r~ j i
Oi =Oj Ri B3i Rk ; (77)
@ jf i=max(k;j +1)
where i~roi =oj is a position vector drawn from the j th nX1
body's local reference system to the ith body's local jU
k = (j t + j t+ )t r~Ot+1 =Ot t Rk : (78)
reference system (j < i). t=k

Inertia Coecients Also j t and j t+ are de ned as follows:

n
For construction of the inertia coecients that multiply j
X
j r~ j
the second derivatives, we substitute Equations 71 to t= Oi =Oj Mi Rt ; (79)
74 and also the summation form of i~roi and i !~_ i (Equa- i=max(t+1;j +1)
tions 67 to 68) into the relevant parts of Equations 43 n
X
and 44. By collecting the terms that contain qj and j + = j R B iR :
i 3i t (80)
t
jf and by arranging them, we obtain expressions that i=max(t+1;j )
should be written in matrix form. By assembling these
matrixes, the inertia matrix of the whole system will In the next section, Expression 75 will be written in
be obtained. In continuation the details of the above matrix form that makes the inertia matrix of the joint
steps are brought. variable in the joint equations. As will be shown, this
matrix is symmetric and this fact reduces the necessary
Inertia Coecients of Joint Variable in Joint computations. Also, the expressions appeared in sum-
Equations mation form (j t+ ; j t ; j Uk ; j k ; j k ) can be calculated
All occurrences of qj in Equation 43 are in the ex- recursively. This is an important issue that causes
pressions of i !~_ i and i~roi ; by isolating these terms and the reduction of necessary computations and will be
interchanging the order of summations as follows: considered in detail in the next section.
n Xi
X nX1 Xn Inertia Coecients of De ection Variables in
k=1
= i=max(k+1;j )
; Joint Equations
i= j k=1 In consideration of Equation 43, we observe that the
n Xi n Xn
de ection variables jf appear not only in i~rOi and i !~_ i ,
X
=
X
; but also in i B~ 1i and i B~ 6i . By isolating these terms, the
i=j +1
k=1
k=1
i=max(k;j +1) expression for the terms that contain jf is obtained as
below:
n Xi 1
X nX1 Xn 
Xn 1 Xmk

k=1
= i=max(k+1;j )
; j ~z T j  +
j k
j
k+ ~kt
i= j k=1 k=1 t=1
Xn 1 Xmk

n Xi 1
X nX1 Xn + j ~z T j  + +j ~rkt
k=1 j k
t=1 k
k=1
= i=max(k+1;j +1)
;
i=j +1 k=1 Xn 2 Xmk
j ~z T j U + ~
k=1 j k kt
t=1
nX1 Xk nX1 Xn 1
= ; Xn Xm k
+ j ~z T j r~ j "
t=1
t=1
k=t
k=j +1 t=1 j Ok =Oj Rk ~kt
k=1
Xn Xm k 
the below expression for the terms that contain qj is + j ~z T j R ~ kt ; (81)
obtained: k =j t=1 j k kt

  where:
Xn Xn 1
j ~z T (j  j )k ~z j ~z T j U k ~z qk ;
k=1 j k k k k=1 j k k
n
(75) j +
X
j R B iR ;
k = i 9i k (82)
where: i=max(k+1;j )

n
X n
X
j = jR i (76) j = j r~ j i (83)
k i B9i Rk ; k+ Oi =Oj Ri B3i Rk ;
i=max(k;j ) i=max(k+1;j +1)
Gibbs-Appell Formulation for Flexible Link Manipulators 487

nX1 n
X
jU + = j j
t t j+  + = j R B iR ; (87)
k t + t+ r~Ot+1 =Ot Rk : (84) k i 3i k
t=k+1 i=max(k+1;j +1)

By writing Expression 81 in matrix form, the inertia nX1

jV = j  t r~ t (88)
coecients of de ection variables in the joint equations k t ot+1 =ot Rk ;
will be obtained. t=k+1
It can be shown that the inertia coecients for
joint variables in the de ection equations are the same n
X
j+ = j r~ j i (89)
as the coecients of de ection variables in the joint k+ oi =oj +1 Ri B3i Rk ;
equations. This issue implies the symmetry of the i=max(k+1;j +2)
inertia matrix of the whole system and can be used n
for reduction of necessary computations. X
j+ j r~ j
k= oi =oj +1 Mi Rk ; (90)
Inertia Coecients of De ection Variables in i=max(k+1;j +2)
De ection Equations j 1
X
In a manner much the same as the previous two steps, jW j R t r~ t
k= t ot+1 =ot Rk ; (91)
the below expression is obtained by isolating the terms t=k+1
that contain jf in de ection equations:
nX1  
 j+ U + = j+ j +  + t r~ t
Xn 1 Xmk  +
j+

k t+ t ot+1 =ot Rk ; (92)
k=1 t=1
~jf T j k+ k+ ~kt t=k+1

Xn 2 Xm k n
X
~jf T j + Uk+ ~kt
 j = Mi j Rk : (93)
k=1 t=1 k
i=max(k+1;j +1)
Xn 2 Xm k
k ~kt
~r T jV
k=1 t=1 jf Like the previous two steps, the above expression is
Xn 1 Xm k written in matrix form. The symmetry of this matrix
~r T j +  + ~ can be shown by expanding its coecients. On the
k=1 t=1 jf k kt
other hand, all the expressions in summation form can
Xj 2 Xmk
T j W ~ be calculated recursively.
k=1
~"
t=1 jf k kt

Xj 1 Xmk Xm j Final Form of Forward Dynamic Equations

+ k=1
~ T j Rk ~kt +
t=1 jf t=1
sc
jft
The complete simulation equations have now been
Xn 1 Xm k  
+ ~ T j+ j+ derived. It remains to assemble them in nal form
k=1 t=1 jf k + k+ ~rkt and point out some remaining recursions that can be
Xn 1 Xm k used to reduce the number of calculations. The second
+ ~r T j  ~r derivatives of the joint and de ection are desired on
k=1 t=1 jf k kt
the \left hand side" of the equation as unknowns,
Xj 1 Xmk
T j R ~r and the remaining dynamic e ects and the inputs are
+ k=1
~"
t=1 jf k kt desired on the \right hand side". To carry out this
Xn Xm k process completely, one would take the inverse of the
+ ~r T jR ~ inertia matrix, I () , and premultiply the vector of
k=j +1 t=1 jf k "kt !
Xn Xm k
other dynamic e ects, Re. Because of its complexity,
+ ~ T j r~ j "
Ok =Oj +1 Rk ~
this inverse can only be evaluated numerically. Thus,
k=j +2 t=1 jf kt
for the purpose of this paper, the equations will be
Xn Xmk  considered in the following form:
+ k=j +1
~ T j Rk ~ kt kt ;
t=1 jf
(85)
!
~ = Re;
I () (94)
where:
n
where:
X
j+  + = jR i (86)
k i B9i Rk ; I () The inertia matrix consisting of coecients
i=max(k+1;j +1) will be obtained in the next section;
488 M.H. Korayem and A.M. Shafei

~
 The vector of generalized coordinate; where:
~
 fq1 11 12


1m1 q2 21


Xm j
2m2


qk k1


kmk


nmngT ; Qjf = 2j !j T : _ ~c + j!
~ j T jf j !
~j
qk The joint variable for the kth joint; k=1 jk jfk
kt The de ection variable (amplitude) of j ~r T :~ ~_
j! T
~ jf : (102)
the tth mode of link k; ov;j "jf v;j
!
Re Vectors of remaining dynamics and external Like the previous step, the following recursive equation
forcing terms, f Re1 Re11


Re1m1 for Rejf is obtained:
Re2 Re21


Re2m2


Rej Rej 1



Rejmj


Renmn gT ; Rejf =
Xm j
 K + Qjf
Rej Dynamics from the joint equations j k=1 jk jkf
(Equation 43), excluding second derivatives
of the generalized coordinate; ~rjf T :j ~ j ~jf T j Rj +1 j +1 ~ j +1 : (103)
Rejf Dynamics from the de ection equations jf
(Equation 44), excluding second Equations 98 and 103 are used to construct the right
derivatives of the generalized coordinate. hand side equations of motion.
At rst, consider Rej . In joint equations, by collecting
the terms that do not contain qj and jf , the PROPOSED ALGORITHM
expression is obtained as below:
Now, we shall present an algorithm that results from
Xn @ i~rOi T i ~ Xn ~_ i T i ~
@i! the expressions developed in previous sections. In
Rej = j i=j +1 @ qj
: Si i=j @ qj
: Ti ; this algorithm, all cross products are done in tensor
(95) notation. And, also, each speci c algorithmic ex-
pression is accompanied by information that indicates
where: the number of algebraic operations that are involved,
showing separately products M and A sums. The
~
iS
i = Mi i~rOv;i 2B2i i !~ i ~_ v;i
B 3i i ! i!
~ i B3i
i!
~ i; calculations are done in a step by step process, as
(96) follows:
i T~
i = B3i i~rOv;i + 2B8i i !~ i + B9i i !~_ v;i + i !~ i B9i i !~ i : Step 1: The rotation matrix will be calculated by this
(97) algorithm.
By substituting Equations 71 and 73 in Equation 95 for i = 2 : 1 : n
and changing it to a recursive expression, a new
equation for Rej is obtained: i 1R
i = E i 1 Ai & iR
i 1 = i 1 Ri T ; 15M 6A
Rej = j j ~z T j  (98)
j ~j;
Step 2: The vectors of i !~ i , i !~_ v;i and i~rOv;i can be
where: calculated recursively, as follows.
j
~j = j T~j + j r~Oj+1 =Oj j ~ j + j Rj +1 j +1 ~ j +1 ; (99) Initialize:
and:
 
1!
~1 = 1~z1 q_1 ; & ~_ v;1
1! = f0 0 0gT ; &
~
j
j = j Rj +1 ~
j +1 S
j +1 + ~ j +1
j +1  : (100)
1~r
Ov;1 = A1 T fgx gy gz gT ;
Now, consider Rejf . If in the defection equation the
terms that do not contain qj and jf are collected, the for i = 2 : 1 : n
following expression will be obtained:
Equation 63 9M 10A
Xm i Xn @ i~rOi T i ~
Rejf =  K
k=1 jk jkf i=j +1 @ jf
: Si  
~_ v;i
i! = i Ri i 1!
~ i 1 ~_ i 1 R i ~z q_
1 i 1+ i 1 i i i
Xn @i!~_ i T i ~ 
i=j +1 @ jf
: Ti + Qjf ; (101) + i 1 !~_ v;i 1 ; 18M 18A
Gibbs-Appell Formulation for Flexible Link Manipulators 489


i~r = i Ri 2i 1 !~ i i 1~r_ and, also, the inertia matrix of the whole system.
ov;i 1 1 oi =oi 1

+ i 1 !~_ v;i 1
i 1~r
oi =oi 1
Step 7: Calculation of the compound rotation matrix,

for j = 1 : 1 : n
+ i 1 !~ i 1
i 1!
~ i 1
i 1~r
oi =oi 1
jR = I33 ;
j
+i 1~rov;i 1 ) ; 33M 27A
for j = 1 : 1 : n 2; k = j + 2 : 1 : n
Step 3: In this step, the vectors of i S~i and i T~i are
calculated. jR = j Rk k 1R
k 1 k; &
for i = 2 : 1 : n kR
j = j Rk T ; 27M 18A
Equation 96; 27M 21A
Step 8: The following algorithm evaluates the vector
for i = 1 : 1 : n of i~rOj =Oi :

Equation 97; 39M 33A for j = 2 : 1 : n 1; j = i 1 : 1 : 1

Step 4: The vectors of i ~i and i ~ i can be calculated j ~r

Oi+1 =Oi = j Rj +1 j +1~rOi+1 =Oi ; 9M 6A
by the following algorithm:
for i = 1 : 1 : n 2; j = i + 2 : 1 : n
Initialize:
i~r = i~rOj + i~rOj =Oj 1 ; 0M 3A
Oj =Oi 1 =Oi
~n
n = f0 0 0g T; & n
~n = n T~n
Step 9: In this step, the variables that have been
for j = n 1 : 1 : 1 appeared in summation form in the inertia matrix are
evaluated.
Equation 100; 9M 9A
 Calculation of j k :
Equation 99; 15M 15A for k = n : 1 : 1; j = k : 1 : 1
Step 5: Calculation of Qjf ,
if (k = j )
for j = 1 : 1 : n; f = 1 : 1 : mj n
if (k = n) n = B 9n ;
Equation 102; 21M 17A
else
Step 6: In this step, Equations 98 and 103 are used
to calculate Rej and Rejf . k
k = B9k + k k+1 k+1 Rk ; 27M 27A
for j = 1 : 1 : n else
Equation 98; 0M 1A j  = j R j +1  ;
k j +1 k & k = j  T ;
j k 27M 18A
for f = 1 : 1 : mn A recursive algorithm, like the one mentioned
Xmn above, for calculation of j k , can be used. However,
Renf =  K
k=1 nk nkf
+ Qnf ; 0M 1A it should be noticed that, instead of B9i , we have
B3i and, also, at the last line, we have:
for j = 1 : 1 : n 1; f = 1 : 1 : mj k j T ;
j = k
Equation 103; 15M 13A
 Calculation of j k :
At the end of this step, the right hand side of
the equations of motion is completely evaluated. In for j = n 1 : 1 : 1
continuation, a recursive algorithm is presented that j
evaluates the left hand side of the equations of motion n = j r~On =Oj j n ; 18M 9A
490 M.H. Korayem and A.M. Shafei

for j = n 1 : 1 : 1; k = n 1 : 1 : 1  Calculation of j Vk :
if (k > j ) for j = 1 : 1 : n 1
j j k+1 R + j r~ j
k= k+1 k Ok =Oj Rk B3k ; 63M 45A jV
n 2 = j n 1
n 1 r~ n 1R
n 2; 18M 9A
On =On 1

else j =j k+1 R
k k+1 k; 27M 18A for j = 1 : 1 : n 1; k = n 3 : 1 : 1
 Calculation of j k : jV
k = j k+1 k+1 r~Ok+2 =Ok+1 k+1 Rk
for k = n 1 : 1 : 1; j = k : 1 : 1
+ j Vk+1 k+1 Rk ; 45M 36A
if (k = j )
 Calculation of j k+ :
if (k = n 1) n 1 n 1 = Mn I33 ; 3M 0A
for k = 1 : 1 : n 1; j = 1 : 1 : n
else
j +
k = j k+1 k+1 Rk ; 27M 18A
k
k = Mk+1 I33 + k+1 k+1 ; 3M 3A
 Calculation of j+ k+ :
else
j  = j R j +1  ; k = j  T ;
for j = 1 : 1 : n 1; k = 1 : 1 : n 1
k j +1 k & j k 27M 18A
j+  + = j Rj +1 j +1 k+ ; 27M 18A
 Calculation of j k : k

for j = n 1 : 1 : 1 For calculation of j k+ ; j k+ and j Uk+ we use the

algorithm like the+ one +that was
+
used for
+
calculation
j of j k+ . Also, j k+ j k+ j k and j Uk+ can be
n 1 = j r~On =Oj j n 1 ; 18M 9A calculated by the algorithm like the one that was
+
used for calculation of j k+ .
for j = n 1 : 1 : 1; k = n 2 : 1 : 1
Step 10: Finally, calculation of the inertia matrix for
if(k < j ) j =j k+1 R ; 27M 18A the whole system is considered.
k k+1 k

else  Calculation of the inertia matrix for joint variables

in joint equations:
j j k+1 R j j
k= k+1 k + Mk+1 r~Ok+1 =Oj Rk ; 51M 36A
for j = 1 : 1 : n 1; k = j : 1 : n 1
 Calculation of jU :
k
Ijk = j ~zj T j k j j U
k ~z ; &
k k k
for j = 1 : 1 : n
jU j j
n 1 Ikj = Ijk ; 0M 18A
n 1 = n 1+ n 1+ r~ On =On 1 ;
for j = 1 : 1 : n
18M 18A

n
for j = 1 : 1 : n; k = n 2 : 1 : 1 if (j 6= n) Ijn = j ~zj T j n j
n n; &
~z

jU = j j  +
k r~ Inj = Ijn ; 0M 9A
k k+ k Ok+1 =Ok

+ j Uk+1 k+1 Rk ; 45M 45A else Inn = n~zn T n n n~zn ; 0M 0A

Gibbs-Appell Formulation for Flexible Link Manipulators 491

 + 
 Calculation of the inertia matrix for de ection vari- Ijkft = ~jf T j
k+
j+  +
k ~rkt
ables in joint equations:
 + 
+ j k+ j+ j+ U + ~kt
k+ k
for j = 2 : 1 : n; k = 1 : 1 : j 1; t = 1 : 1 : mk
 +j Rk ~ kt + j Rj +1 j +1 r~Ok =Oj+1 j +1 Rk ~"kt )


Ijkt = j ~zj T j +
k
j
k+
jU +
k ~kt  
j+
~rjf T k + k+ ~kt
jV


+ j k+ + j k ~rkt ) ; 18M 42A 
j  ~r jR ~ ; 84M 107A
k kt k "kt
for j = 1 : 1 : n 1; k = j ; t = 1 : 1 : mk
for j =1 : 1 : n 1; k = n; t =1 : 1 : mk ; f =1 : 1; mj


Ijkt = j ~zj T j +
k
j
k+
jU +
k ~kt Ijkft = ~jf T j Rk ~ kt + j Rj +1 j +1 r~Ok =Oj+1 j +1 Rk ~"kt

+ j k + + j k ~rkt + ~ kt ) ; 18M 45A +~rjf T j Rk ~"kt ; 48M 35A

for j =1 : 1 : n 2; k = j +1 : 1 : n 1; t =1:1: mk for j = n; k = n; t = 1 : 1 : mk ; f = 1 : 1 : mj


j U +
~
Ijfkt = s cjft ; 0M 0A
Ijkt = j ~zj T j +
k
j
k+ k kt
The required mathematical operations for calculating
j +
the above steps are listed in Table 1, where n is the
+ k +j j
k ~rkt + Rk ~ kt
total number of links; nf is the number of exible links
+ j r~Ok =Oj j Rk ~"kt ) ; 42M 63A and m is the number of modes describing each exible
link, the same for all exible links.
In Table 2, the computational complexity of this
for j =1 : 1 : n 1; k = n; t =1 : 1 : mk method compared with the ones of [5], are shown. Also,


Table 2 shows the number of operations for two typical
Ijnt = j ~zj T j Rk ~ kt + j r~Ok =Oj j Rk ~"kt ; 24M 18A cases.
As a general comparison, the number of mathe-
matical operations of the method proposed in this ar-
for j = k = n; t = 1 : 1 : m k ticle for the dynamic modeling of exible manipulators
is less than the recursive Lagrangian method in [5].
Ijkt = j ~zk T ~ kt ; 0M 0A
COMPUTATIONAL SIMULATION
 Calculation of de ection variables in de ection equa-
tions: In this section, we verify the proposed method for
the dynamic modeling of exible robotic manipulators
in the preceding sections by means of computational
for j =1 : 1 : n 1; k = j ; t =1 : 1: mk ; f =1 : 1: mj simulation for a manipulator with two elastic links.
 +  The rst mode shape of clamped-free beams is used to
Ijkft = ~jf T j j+  + model the elastic deformation of each link. All neces-
k+ k ~rkt
sary parameters of exible links for this computational
 +
j+ j+ U +
 simulation are shown in Table 3. These parameters are
+ j k+ k+ k ~kt the same as in [4].
   To clearly explain computational procedures for
~rjf T jV j+  + ~kt j  ~r the simulation, we rewrite Equation 94 in state form.
k+ k k kt
~_ 1 = 
 ~ 2;
+ s cjft ; 42M 72A
!
~_ 2 = I 1 (1 )Re:

for j =1 : 1: n 2; k = j +1 : 1 : n 1; t =1 : 1: mk ;
The initial conditions are also the same as in [4] as
f =1 : 1 : mj shown in Figure 3.
492 M.H. Korayem and A.M. Shafei

Table 1. The required mathematical operation for deriving the equation of motion in G-A's formulation.
Sums Products Step
6n 6 15n 15 1
55n 55 60n 60 2
54n 21 66n 27 3
24n 24 24n 24 4
17nf m 21nf m 5
n 12m + 13mnf 15mnf 15m 6
9n2 27n + 18 13:5n2 40:5n + 27 7
4:5n2 13:5n + 9 4:5n2 13:5n + 9 8
276 652:5n + 328:5n2 390 901:5n + 457:5n2 9
9n 9 + 52:5mn2f 79:5mnf + 18m
2 30mn2f 30mnf 60m2 nf + 10
89:5m2 nf + 18m2 + 53:5m2 n2f 18m2 + 42m2 n2f

Table 2. The comparison of computational complexity.

Sums Products Principle Authors
329n + 115:5mn2f + 19m2 nf 279n + 118mn2f + 17:5m2 nf
+123mnf + 85n2 + 68nnf m +137:5mnf + 84n2 + 74nnf m L-E Book
+6:5m2 n2f 91 + 80nnf + 111nf +6m2 n2f 57 + 86nnf + 126nf
6m 553n 49:5mnf 15m 790:5n + 6mnf
+351n2 + 18m2 + 53:5m2 n2f +475:5n2 + 18m2 + 42m2 n2f G-A This work
+52:5mn2f 89:5m2 nf + 188 +30mn2f 60m2 nf + 300
n = 6 ; nf = 6 ; m = 3 n = 3; n f = 2; m = 2
24569A 25251M 4851A 4922M Book
18332A 19557M 2134M 2706M This work

1 = 90 deg; 2 = 5 deg; ures 4 to 15 represent the exural responses of the

system. The response of these variables portrays
_1 (0)= _2 (0)= 11 (0)= _12 (0)= 21 (0)= _21 (0)=0: the vibration modes of the system response and their
in uence on the quality of the system response. Sim-
Then, by using a numerical method, such as Runge- ulations results show that the response of the exible
Kutta, a set of di erential equations will be solved. manipulator is highly undesirable and, in order to get
By solving this set of di erential equations, the
time response of the system will be obtained,
(q1 ; q_1 ; 11 ; _11 ; q2 ; q_2 ; 21 ; _21 ). In [4], which uses FEM
for simulation, the results of exural displacement and
angular displacement in the middle and at the end of
each link are shown. So, for comparison, we present
the same results in Figures 4 to 15.
Variables u3 ; u4 ; u5 ; u6 ; w3 ; w4 ; w5 and w6 in Fig-

Table 3. The necessary parameters for simulation.

Parameters Value Unit
The length of the links L1 = L2 = 1 m
Module of elasticity E1 = E2 = 2:0  10 N/m2
11
Moment of inertia I1z = I2z = 5:0  10 9 m4
Mass per unit length 1 = 2 = 5 kg/m Figure 3. Initial condition for simulation.
Gibbs-Appell Formulation for Flexible Link Manipulators 493

Figure 4. Angular displacement of the rst joint. Figure 7. Y position of end e ector.

Figure 8. Flexural displacement in the middle of the rst

Figure 5. Angular displacement of the second joint. link.

Figure 9. Flexural displacement at the end of the rst

Figure 6. X position of end e ector. link.
494 M.H. Korayem and A.M. Shafei

Figure 10. Angular displacement in the middle of the Figure 13. Flexural displacement at the end of the
rst link. second link.

Figure 11. Angular displacement at the end of the rst Figure 14. Angular displacement in the middle of the
link. second link.

Figure 12. Flexural displacement in the middle of the Figure 15. Angular displacement at the end of the
second link. second link.
Gibbs-Appell Formulation for Flexible Link Manipulators 495

the dynamics of the system to be acceptable for most IEEE Trans. System Man Cybernet, SMC 17, pp.
practical purposes, very e ective controls are needed 1059-1068 (1987).
to control the vibration modes. On the other hand, 7. Jin, C. and Sankar, T.S. \A systematic Method of
as seen, the results are in good concordance with ones dynamics for exible robot manipulators", J. Dynamic
in [4]. It should be noted that the simulation is done System., 9(7), pp. 861-891 (1992).
by using only one mode shape. More accurate results
will be obtained by using more mode shapes. 8. Korayem, M.H., Yau, Y. and Basu, A. \Application
of symbolic manipulation to inverse dynamics and
kinematics of elastic robot", International Journal of
CONCLUSION Advanced Manufacturing Technology, 9(4), pp. 343-350
(1994).
This article has presented an ecient and systematic
method for the dynamic modeling of exible robotic 9. Korayem, M.H. and Basu, A. \Automated fast sym-
manipulators. The proposed method can be applied bolic modeling of robotic manipulators with compliant
to the design of control systems and the dynamic link", International Journal of Scienti c Computing &
Modeling, 22(9), pp. 41-55 (1995).
simulation of exible manipulators. The advantages of
this method in comparison with others are as follows: 10. Anderson, K.S. \An ecient formulation for the mod-
eling of general multi- exible-body constrained sys-
1. A reduction in computations by using only 3  3 tems", International Journal of Solids and Structures,
and 3  1 matrices. 30(7), pp. 921-945 (1993).
2. Increase in the speed of generating the equations 11. Banerjee, A.K. \Block-diagonal equations for multi-
of motion by reducing the number of additions and body elastodynamics with generic sti ness and con-
multiplications, as shown in Table 2. straints", AIAA Journal of Guidance, Control and
3. Ease of understanding, as it uses primitive dynamic Dynamics, 16(6), pp. 1092-1100 (1993).
concepts. 12. Dwivedy, S.K. and Eberhard, P. \Dynamic analysis of
exible manipulators, a literature review", J. Mecha-
REFERENCES nism and Machine Theory, 41(7), pp. 749-777 (2006).
1. Walker, M.W. and Orin, D.E. \Ecient dynamic 13. Baruh, H., Analytical Dynamics, New York, McGraw-
computer simulation of robotic mechanisms", ASME Hill (1998).
J. Dynamic systems Meas. Control., 104, pp. 205-211 14. Vukobratovic, M. and Potkonjak, V., Applied Dynam-
(1982). ics and CAD of Manipulation Robots, Springer-Verlag,
2. Megahed, S. and Renaud, M. \Minimization of the Berlin (1985).
computation time necessary for the dynamic control 15. Desoyer, K. and Lugner, P. \Recursive formulation for
of robot manipulators", Proc. of the 12th ISIR, pp. the analytical or numerical application of the Gibbs-
469-478 (1982). Appell method to the dynamics of robots", Robotica.,
3. Vukobratovic, M., Li, S.G. and Kircanski, N. \An 7, pp. 343-347 (1989).
ecient procedure for generating dynamic manipulator
models", Robotica., 3, pp. 147-152 (1985). 16. Vereshchagin, A.F. \Computer simulation of the dy-
namics of complicated mechanisms of robotic manipu-
4. Usoro, P.B., Nadira, R. and Mahil, S.S. \A nite lators", Eng. Cyber., 6, pp. 65-70 (1974).
element / Lagrange approach to modeling lightweight
exible manipulators", ASME J. Dynamic Systems 17. Rudas, I. and Toth, A. \Ecient recursive algorithm
Meas. Control., 108, pp. 198-205 (1986). for inverse dynamics", Mechatronics., 3(2), pp. 205-
5. Book, W.J. \Recursive Lagrangian dynamics of exible 214 (1993).
manipulator arms", Int. J. Robotics. Res., 3, pp. 87- 18. Mata, V., Provenzano, S., Cuadrado, J.I. and Valero,
101 (1984). F. \Serial-robot dynamics algorithms for moderately
6. King, J.O., Gourishankar, V.G. and Rink, R.E. \La- large number of joints", J. Mechanism and Machine
grangian dynamics of exible manipulators using an- Theory, 37, pp. 739-755 (2002).
gular velocities instead of transformation matrices",