Professional Documents
Culture Documents
SUMMARY
The equivalent static load (ESL) method is a powerful approach to solve dynamic response structural
optimization problems. The method transforms the dynamic response optimization into a static response
optimization under multiple load cases. The ESL cases are defined based on the transient analysis response
whereupon all the standard techniques of static response optimization can be used. In the last decade, the
ESL method has been applied to perform the structural optimization of flexible components of mechanical
systems modeled as multibody systems (MBS). The ESL evaluation strongly depends on the adopted formu-
lation to describe the MBS and has been initially derived based on a floating frame of reference formulation.
In this paper, we propose a method to derive the ESL adapted to a nonlinear finite element approach based
on a Lie group formalism for two main reasons. Firstly, the finite element approach is completely general
to analyze complex MBS and is suitable to perform more advanced optimization problems like topology
optimization. Secondly, the selected Lie group formalism leads to a formulation of the equations of motion
in the local frame, which turns out to be a strong practical advantage for the ESL evaluation. Examples are
provided to validate the proposed method. Copyright © 2016 John Wiley & Sons, Ltd.
KEY WORDS: structural optimization; equivalent static load (ESL) method; flexible multibody systems;
nonlinear finite element approach; Lie group formalism
1. INTRODUCTION
Although the fundamentals of static response optimization are well established, structural opti-
mization based on transient structural analysis is still a field of active research. Comprehensive
studies began in 1970s [1] with one of the earliest approaches proposed by Fox and Kapoor [2] to
perform structural optimization under dynamic loads. The bottleneck of the dynamic response opti-
mization lies in the evaluation of the dynamic response and its incorporation in the optimization,
that is, the treatment of time-dependent constraints. Moreover, the sensitivity computation of the
displacements, velocities and accelerations can become costly [3].
The structural optimization of mechanical system component was previously achieved with a
component-based approach, which consists in isolating the component from the system and then,
in enforcing boundary and loading conditions to predict its mechanical responses, that is, deforma-
tion, stress distribution and so on. These conditions usually come from the designer’s experience,
empirical relations or experimental efforts. However, albeit the mechanical system components
are dynamically loaded, the optimization was typically performed under (quasi-)static or fre-
quency domain loadings due to the difficulties of considering the dynamic response optimization as
explained previously.
*Correspondence to: Emmanuel Tromme, Aerospace and Mechanical Engineering Department, Quartier Polytech 1,
University of Liège, Allée de la Découverte 9, 4000 Liège, Belgium.
† E-mail: emmanuel.tromme@ulg.ac.be
With the evolution of virtual prototyping, the component-based approach has been extended
towards a system-level approach that incorporates a multibody system (MBS) simulation to evalu-
ate the response of the whole system. The first contribution came from Bruns and Tortorelli [4] who
proposed an approach combining rigid MBS analysis and optimization techniques to design optimal
components. The optimization procedure was performed with load cases evaluated during the MBS
analysis. They illustrated their method on the design of a slider-crank mechanism loaded with the
maximum tensile force calculated during the simulation.
In the late 1990s, a breakthrough has been realized by Choi and Park [5] who introduced the
equivalent static load (ESL) method to perform the dynamic response optimization of structures.
Basically, the latter method transforms the dynamic response optimization into a static response
optimization under multiple load cases, namely the ESL, that are defined based on the transient
analysis response. Ergo, all the standard techniques of static response optimization can be used.
The main motivations of developing such a method are to avoid the expensive computation of the
state variable derivatives, the possibility of using efficient commercial solvers for static response
structural optimization and to circumvent the difficult treatment of time-dependent constraints. The
ESL method has been developed in a series of articles, and more recently, Kang et al. [1] published
a review article gathering the main developments on this method.
Continuing along the idea of the ESL approach, Kang et al. [6] introduced a weakly coupled
method to optimize mechanical system components. The weakly coupled method reformulates the
problem using a two level approach. Firstly, an MBS simulation computes the dynamic response of
the system. Secondly, each component is optimized independently within its body-attached frame
using a quasi-static approach, wherein a series of ESL derived from the MBS simulation responses
are applied to the respective components. As the ESL is not updated as the design evolves during the
static response optimization, cycles between MBS analysis and optimization process are needed to
account for the effect of mass redistribution over the load cases. Several works were realized using
this weakly coupled method, for example [7–9]. The weakly coupled method contrasts with the
fully coupled method [10–13] wherein the optimization process considers the time response coming
directly from the MBS simulation.
In [6], the ESL derivation is carried out for MBS described with a floating frame of reference
formulation. This formalism is suitable to define ESL because the flexibility of the components
is introduced within a body-attached frame and that the component responses are described with
respect to this frame. Extending the ESL evaluation to a nonlinear finite element formalism is inter-
esting as this formulation is completely general and can analyze complex MBS. Furthermore, this
formulation is also convenient to perform more advanced optimization problems like topology opti-
mization. However, the ESL derivation for a standard nonlinear finite element formulation [14, 15] is
not straightforward for several reasons inherent to the formalism. First, the equations of motion are
developed in an inertial frame, that is, the elastic forces are not expressed in a body-attached frame.
Furthermore, no decoupling exists between rigid body motion and elastic deformation. Finally, the
tangent stiffness matrix is not constant and evolves with the system configuration.
In this paper, we continue considering mechanical systems described with a nonlinear finite
element approach, but we adopt a Lie group formalism to derive the equations of motion. The
configuration of the flexible MBS is still described with absolute nodal variables, that is, displace-
ments and orientations with respect to the inertial reference frame. These two kinds of variable
have a different nature: displacements are classical vectors, while rotations are linear maps of vec-
tors, namely rotation matrices. However, both can be handled at the same time if the configuration
space is described as a k-dimensional manifold G that has a matrix Lie group structure. According
to [16–18], the Lie group approach can be exploited not only to describe the configuration but also
to formulate the equations of motion and to integrate them in the time domain. In this research, the
particular Lie group formalism based on the special Euclidean group SE(3) is adopted, for example,
see [17, 19, 20], offering several advantages. Firstly, the equations of motion are derived and solved
directly on the nonlinear manifold, without an explicit parameterization of the rotation variables,
leading to important simplifications in the formulation and algorithms. Secondly, displacements
and rotations are represented as increments with respect to the previous configuration, and those
increments can be expressed in the material frame (local frame of the nodes). Therefore, global
Copyright © 2016 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2016)
DOI: 10.1002/nme
ON THE ESL METHOD FOR FLEXIBLE MBS DESCRIBED WITH A NLFE FORMALISM
geometric nonlinearities are automatically filtered from the relationship between incremental dis-
placements and elastic forces, which strongly reduces the fluctuations of the iteration matrix (and of
the tangent stiffness matrix) during the simulation [21]. Finally, in dynamic simulations, the formu-
lation completely avoids the artificial definition of a corotational frame, and no decomposition of
the motion into a large amplitude rigid motion and a small amplitude elastic deformation is needed.
Taking advantages of the inherent characteristics of this particular Lie group formalism, we pro-
pose an efficient method to evaluate the ESL for flexible MBS described within a nonlinear finite
element approach.
The first part of the paper reminds the analysis of flexible MBS using a Lie group formalism, that
is, the derivation of the equations of motion and the time integration algorithm. Once the equations
of motion are established, the novel method to derive ESL within a nonlinear finite element approach
is developed. The next part introduces the general framework of the structural optimization process.
Classical examples of MBS optimization problems are revisited to validate the proposed method.
where RI 2 SO.3/ is a rotation matrix and xI 2 R3 is a position vector. Such frames are used
to describe the absolute position and orientation of each node of the rigid and flexible bodies and
to describe the relative transformation occurring in kinematic joints, leading to a mixed (absolute
and relative) variable formulation. For a nodal frame, the matrix HI has kI D 6 degrees of free-
dom, namely, three translations and three rotations, while for a relative transformation frame, it is
restricted to belong to a subgroup of SE.3/ and has kI < 6 degrees of freedom. The dimension
k of the configuration space is thus the sum of the degrees of freedom of the nodes plus the degrees
of freedom of the relative transformations. The state of the system is represented as a block diagonal
matrix H composed of the HI ’s.
The composition operation G G ! G that associates two elements of the group to one element
of the group is based on the matrix product
Ht ot D H1 H2 : (2)
The time derivative can be introduced by means of a left invariant vector field as
P D He
H v (3)
where ev is an element of the Lie algebra g, namely, the tangent space at the identity element of
the group. The Lie algebra is a vector space such that e v can be associated with a velocity vec-
tor v of dimension k, which includes 6 (resp. kI ) velocity variables for each node (resp. relative
transformation). For each node, the element e
v is defined as
e u
e
vD (4)
0 0
Copyright © 2016 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2016)
DOI: 10.1002/nme
E. TROMME ET AL.
e is the well-known
where u corresponds to the three linear velocity component vector and
skew-symmetric matrix composed of the three angular velocity components D Œ1 2 3 T ,
that is
2 3
0 3 2
e D 4 3 0 1 5 : (5)
2 1 0
The velocity vector v represents the absolute velocity of a node expressed in a nodal local frame.
Therefore, the interpretation of the velocity vector is the same as with the classical nonlinear finite
element approach.
The exponential map is an operator that maps any element of the Lie algebra to an element of the
Lie group:
exp W g ! G; e q/ :
q 7! q D exp .e (6)
The inverse map is called the logarithmic map:
log W G ! g; q 7! e
q D log.q/: (7)
The nodal variables are a priori considered as independent variables. Nonetheless, six kinematic
constraints between nodal frames and relative transformation frames are afterwards introduced for
each kinematic joint. Formally, m constraints ˆ W G ! Rm are enforced, restraining the dynamics
to the submanifold N of dimension k m, such as
N D ¹H 2 G W ˆ.H/ D 0º : (8)
k
where ıh 2 R is the virtual motion vector in the local frame introduced as
Copyright © 2016 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2016)
DOI: 10.1002/nme
ON THE ESL METHOD FOR FLEXIBLE MBS DESCRIBED WITH A NLFE FORMALISM
e
ı.H/ D H ıh; (13)
g.H/ is the external and internal force vector defined such that
e D ıhT g.H/
D V .H/ ıh (14)
where the left-hand-side represents the directional derivative on the Lie group and B is the m k
matrix of constraint gradients such that
e D B.H/ıh:
Dˆ.H/ ıh (15)
ı.v/ D ı hP C b
vıh (17)
where b is a linear operator of the Lie group which transforms a k 1 vector into a k k matrix.
As a result, it yields
d T
ıh Mv D ı hP T Mv C ıhT MPv D ı.v/T Mv C ı.h/T MPv b
vT Mv : (18)
dt
Integrating Equation (12) by parts gives
Z tf
T tf
ıh Mv t ıhT MPv vO T Mv C g.H/ C BT .H/ C ıT ˆ.H/ dt D 0 (19)
i
ti
where the first term vanishes due to the boundary conditions. Finally, the stationarity conditions lead
to a set of index-3 differential-algebraic equations
P D He
H v (20)
Notice that the first equation is a matrix equation while the other two are vector equations. These
equations of motion represent the dynamics of a general class of conservative flexible MBS. The
formulation can be easily extended to account for non-conservative forces. We observe that no
parameterization of rotations is needed to formulate those equations and that they are expressed in
the local frame.
Let us develop the linearized form of the equations of motion, which is useful for implicit time
integration as well as later for the ESL evaluation. Considering infinitesimal variations about a given
state .H ; v ; vP ; / lead to
where the tangent damping and stiffness matrices are defined from @r.H; v; vP ; /=@v D Ct and
e D Kt ıh, respectively.
from the directional derivative Dr.H; v; vP ; / ıh
Copyright © 2016 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2016)
DOI: 10.1002/nme
E. TROMME ET AL.
ˆ.HnC1 / D 0 (27)
combined with the time integration formulae
nnC1 D hvn C .0:5 ˇ/h2 an C ˇh2 anC1 (28)
v ! v C v (33)
vP ! vP C Pv (34)
! C (35)
where ./ denotes a finite variation and where the n C 1 subscript is dropped for the sake of
convenience. The correction terms are computed from the linearized form of the residual equation
around the approximate solution .H ; v ; vP ; / such as
rli n .h; v; Pv; / r H ; v ; vP ;
D C
ˆ li n .h/ ˆ .h /
(36)
M0 Pv Ct 0 v Kt BT h
C C
0 0 0 0 B 0
Copyright © 2016 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2016)
DOI: 10.1002/nme
ON THE ESL METHOD FOR FLEXIBLE MBS DESCRIBED WITH A NLFE FORMALISM
where T is the tangent operator associated with the exponential map. Then, the integration for-
mulae (28)–(30) are manipulated by solving Equation (30) for anC1 and inserting that result into
Equations (28) and (29). The resulting two equations are
PvnC1 D .1 ˛m /= ˇh2 .1 ˛f / nnC1 (38)
As a result, using Equations (37)–(38), the linearized form of the update equation (36) reduces to
r n
D St (40)
ˆ
Algorithm 1 solveTimeStep(vn ; vP n ; an )
vP nC1 WD 0
nC1 WD 0
anC1 WD ˛f vP n ˛m an = .1 ˛m /
vnC1 WD vn C h .1 / an C hanC1
nnC1 WD vn C h .0:5 ˇ/ an C ˇhanC1
for i D 1 to imax do
HnC1 D Hn exp.e nnC1 /
r .HnC1 ; vnC1 ; vP nC1 ; nC1 ; tnC1 /
res WD
ˆ .HnC1 /
if jjrjj < tolr and jjˆjj < tolˆ then
break
end if
t WD St .HnC1 ; vnC1 ; vP nC1 ; nC1 ; tnC1 /
S
n
WD S1t res
nnC1 WD nnC1 C n
vnC1 WD vnC1 C =.ˇh/n
vP nC1 WD vP nC1 C .1 ˛m /=.ˇh2 .1 ˛f //n
nC1 WD nC1 C
end for
anC1 WD anC1 C .1 ˛f /=.1 ˛m /PvnC1
return vnC1 ; vP nC1 ; anC1 ; nC1
Copyright © 2016 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2016)
DOI: 10.1002/nme
E. TROMME ET AL.
Definition 1
When a dynamic load is applied to a structure, the ESL is defined as the static load that makes the
same displacement field as that by the dynamic load at an arbitrary time [5, 26].
The main purpose of defining ESL is to mimic the dynamic load by a series of static loads, that
is, the obtained series of ESL gives at each time step the same displacement field as the one given
by the dynamic loading considering a local frame approach. From a pure analysis point of view,
this seems useless. However, in the context of structural optimization, the ESL enables to compute
the displacement during the optimization procedure. As a consequence, the dynamic optimization
problem can be transformed into a static optimization problem subject to multiple static load cases,
each load case being associated with each integration time step.
Kang et al. [6] extended this definition to flexible MBS components wherein the MBS dynamics is
described using a floating frame of reference formulation. For the static load case, they consider that
the component is isolated from the system and that the displacement field is evaluated with respect
to the body-attached frame. Because the component is isolated, rigid body modes must be prevented
in the static analysis. In the floating frame of reference formulation, the problem is naturally circum-
vented. Indeed, this formalism requires to enforce boundary conditions when introducing flexibility
in the MBS. For instance, boundary conditions are applied for the modal analysis when employing
the component mode synthesis method to introduce flexibility according to the floating frame of ref-
erence method [27]. In the static load case, the same boundary conditions are considered so that all
rigid body modes are blocked.
The vector g.H/ collects the internal and external forces so that Equation (42) can be expressed as
Copyright © 2016 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2016)
DOI: 10.1002/nme
ON THE ESL METHOD FOR FLEXIBLE MBS DESCRIBED WITH A NLFE FORMALISM
is not unique but any solution is equivalent to Hbn up to a rigid body motion, that is, it is of the form
Hbn D Hbrig Hbst at . Hbn is itself a solution of Equation (45).
To fix this non-uniqueness issue, some boundary conditions must be enforced to prevent rigid
body motions. In this Lie group formalism, boundary conditions are imposed by restricting the
motion to a subgroup G b˘ G b , whose dimension is k b k bc , where k bc equals 6 in 3D. Then,
Equation (45) has a unique solution in G b˘ , which is denoted as Hb˘ st at .
In practice, the boundary conditions may be enforced by fixing some components of the nodal
variables in Hbst at , which are then eliminated from the initial static problem. The corresponding
entries in the vector gi nt;b can be eliminated leading to a vector gi nt;b˘ of dimension k b k bc . After
this elimination, the equivalent static problem is formulated as
gi nt;b˘ Hb˘ b˘
st at D gn;eq : (46)
Hbn D Hb˘ b˘
rig Hlocal : (47)
Accordingly, Hb˘local
is the representation of the current configuration of body b with respect to
the body-attached frame. We then argue that Hb˘ local
D Hb˘ st at . Firstly, considering Equation (44),
it is clear that Hlocal is a solution of Equation (45). Secondly, by construction, Hb˘
b˘
local
satisfies the
boundary conditions used in Equation (46). Thus, Hb˘ local
satisfies Equation (46) and the conclu-
sion follows from the unicity of the solution for this equation. In summary, we have shown that the
displacements resulting from the static analysis is equivalent to the displacements resulting from
the dynamic analysis but represented in the body-attached frame. That is, the relative displace-
ment between two material points of the component is exactly the same, either in the dynamically
deformed configuration Hbn or in the statically deformed configuration Hb˘ st at . The different notations
are depicted in Figure 1.
Figure 1. Graphical interpretation of the notations. (a) The multibody system; and (b) Isolated body b with
imposed boundary conditions.
Copyright © 2016 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2016)
DOI: 10.1002/nme
E. TROMME ET AL.
After enforcing the boundary conditions, the ESL corresponds to the internal force vector gb˘ n;eq
acting on the considered body b at time tn , and they can be stored during the MBS analysis. This
ESL definition is valid because the internal forces are computed in the local frame of each node.
Our definition slightly differs from Kang et al. [6] because the ESL is defined in the local frame of
each node and not in a floating frame.
The selection of boundary conditions has a limited influence on the procedure because, whatever
the choice of this frame, the equivalence between the static and dynamic response is strictly guaran-
teed in the sense explained previously. The methods developed in the context of the floating frame
of reference formulation [27] can be considered. A selection tending to reduce the amplitude of the
local displacements Hb˘st at may offer the advantage of a reduced nonlinearity of the static problem
in Equation (46). However, this argument becomes meaningless in the frequent situation where the
local displacements are very small. Sometimes, a specific choice can lead to a simpler formulation
of the static optimization problem described later. Hence, the choice can be guided by practical
motivations to simplify the set-up of the overall procedure.
During the optimization, the resulting displacement field can be evaluated by solving
Equation (46), at the price of a nonlinear computation. However, because small deformations are
often encountered in MBS optimization, the next section presents a linearization of Equation (46).
Kb˘ b˘ b˘
t nst at D gn;eq (48)
where nb˘ b˘
st at is interpreted as the increment that links Hst at , resulting from the equations of
motion, to Href , that is the reference configuration such that gb .Hb˘
b˘
ref
/ D 0. The tangent stiffness
matrix Kb˘
t does not depend on the configuration because it is defined in the local frame and that
small deformations are assumed.
During the optimization process, the tangent stiffness matrix Kb˘ t of the optimized component
evolves with respect to the design parameters p and the deflection resulting from the ESL at time tn
is recomputed according to
1
nb˘ b˘
st at D Kt .p/ gb˘
n;eq : (50)
Several properties of this particular Lie group formalism render the approach suitable and effi-
cient to evaluate the ESL and then to formulate an equivalent linear static problem. Firstly, the
equations of motion are derived and solved directly on the nonlinear manifold, without an explicit
parameterization of the rotation variables, leading to important simplifications in the formulations
and algorithms. Secondly, displacements and rotations are represented as increments with respect
to the previous configuration, and those increments can be expressed in the local frame. As a result,
geometric nonlinearities are automatically filtered from the relationship between incremental dis-
placements and elastic forces, which strongly reduces the fluctuations of the iteration matrix during
the simulation. Furthermore, because the special Euclidean group SE(3) combined with the left
Copyright © 2016 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2016)
DOI: 10.1002/nme
ON THE ESL METHOD FOR FLEXIBLE MBS DESCRIBED WITH A NLFE FORMALISM
invariant representation of derivatives is adopted, the tangent stiffness matrix Kbt can be considered
as constant under the assumption of small deformations. This contrasts with classical nonlinear finite
element formulation wherein the tangent stiffness matrix Kt evolves with the system configuration.
minimize f0 .p/
p
This formulation provides a general and robust design framework that can be solved by various
types of optimization algorithms.
Particularizing the general formulation (51) to MBS optimization leads to the following dynamic
response optimization:
minimize f0 .p; H; v; vP ; /
p
subject to P D He
H v;
MPv vO T Mv C g.H/ C BT .H/ D 0;
(52)
ˆ.H/ D 0; 8t 2 ti ; tf ;
fj .p; H; v; vP ; / 6 f j ; j D 1; : : : ; nc ;
p i 6 pi 6 p i ; i D 1; : : : ; nv ;
Cost and constraint functions are, for example, mass, displacement or stress measures at particular
points and time steps. The design variables nv may be either sizing, shape or topological variables.
subject to Kb˘ b˘ b˘
t .p/nn D gn;eq ; b D 1; : : : nb ;
(53)
fj p; nb˘
n 6 f j; j D 1; : : : ; nc ;
p i 6 pi 6 p i ; i D 1; : : : ; nv ;
for n D 1; : : : ; ne nd where nb and ne nd number the components and the time steps, respectively.
Copyright © 2016 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2016)
DOI: 10.1002/nme
E. TROMME ET AL.
In the optimization problem formulation (53), each body is loaded by as many load cases as the
number of time steps. The ESL is fixed during the static response optimization process whereupon
cycles between MBS analysis and static optimization are needed to account for the effects of design
modifications over the ESL. Held [30] demonstrates the equivalence between keeping the ESL
fixed and neglecting the time dependency in the sensitivity analysis of the dynamic optimization
problem (52).
To solve the dynamic response optimization problem using the ESL method, the algorithm
proposed in [31], apart from the stopping criteria, is as follows:
(1) Initialize the design variables and set it D 0.
(2) Perform a dynamic MBS analysis.
(3) Extract the ESL for the optimization.
(4) If it D 0, go to step 5. If it > 0 and if
teP
nd
kgb˘ b˘
n;eq;i t gn;eq;i t 1 k
nD1
teP
< "; (54)
nd
kgb˘
n;eq;i t 1 k
nD1
Copyright © 2016 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2016)
DOI: 10.1002/nme
ON THE ESL METHOD FOR FLEXIBLE MBS DESCRIBED WITH A NLFE FORMALISM
can probably be referred to as one of the oldest optimization approaches. These methods have been
employed to solve large scale structural and multidisciplinary optimization problems with great suc-
cess [36, 37]. The major advantage of these methods is their good convergence rate, which limits
the number of function evaluations to obtain an optimal design. As encountered for most methods,
these are also sensitive to local optima.
Amongst the well-known applications of mathematical programming, we can mention the
sequential quadratic programming [38], the CONvex LINearization [39], the method of moving
asymptotes [40] and its extensions, for example, the globally convergent method of moving asymp-
totes [41] and the globally convergent method [42]. In this study, method of moving asymptotes
algorithm is adopted. This algorithm is based on the sequential convex programming approach,
which relies on two concepts. First, a local approximation of the original optimization problem is
built. Indeed, the original optimization is generally nonlinear and implicit with respect to the design
variables, whereas the number of function evaluations to solve the problem can be relatively impor-
tant. The local approximations are established using the sensitivities as well as a variant of the Taylor
series expansion of the design functions. Second, each local convex sub-problem is solved resort-
ing to efficient mathematical programming algorithms such as a Lagrangian maximization (dual
method) or interior point methods. Figure 3 illustrates the sequential convex programming approach.
The first example concerns the mass minimization of a two-DOF planar robot inspired from [6, 10]
and is illustrated in Figure 4. Each robot arm has a length of 600 mm and is modeled by two equal
length beam elements with a hollow cross section. The adopted beam element model is described
in [20]. The robot is made of aluminium with a Young’s modulus of E D 72 GPa, Poisson’s ratio
of D 0:3 and mass density of D 2700 kg/m3 . Each revolute joint is driven by an ideal motor
which imposes a smooth joint trajectory 1 .t / and 2 .t / such that the robot tip deploys along a
straight line. The initial and terminal joint angles are 1 .ti / D 120°, 2 .ti / D 150°and 1 .te nd / D
60°, 2 .te nd / D 30°, respectively. The ideal tip displacement corresponds to the following
trajectory equation:
0:5 T 2 t
xt ip .t / D yt ip .t / D t sin (55)
T 2 T
Copyright © 2016 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2016)
DOI: 10.1002/nme
E. TROMME ET AL.
hollow beam elements whose wall thickness is set to 0:1 pi . Initial values of the design variables
are set to 50 mm. During the static response optimization, each arm is considered as a fixed-free
beam to fix the singularity of the tangent stiffness matrix.
The optimization problem concerns the weight minimization of the robot W .p/ subject to two
types of constraints. The first restriction enforces a limit on the maximal deflection of the robot
defined as
q
2
fj D yA .tj / C yt2ip .tj / (56)
where yA and yt ip are the transversal deflection in the respective local frame, respectively, for
the first and second arm. The second constraint limits the axial stresses at each time step tj for
each element i at the extremity of both sides of the neutral fibre. Therefore, the stress constraint
vector reads
N M N M T
j;i D
j;i C
j;i
j;i
j;i (57)
where the superscripts ./N and ./M stand for the normal and bending stresses, respectively. It
follows that the optimization problem is stated as
minimize W .p/
p
q
2
subject to yA .tj / C yt2ip .tj / 6 0:001 mm; (58)
Œ75 75T 6 j;i 6 Œ75 75T MPa; j D 1; : : : ; nc ;
0:02 m 6 pi 6 0:06 m; i D 1; : : : ; 4:
Copyright © 2016 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2016)
DOI: 10.1002/nme
ON THE ESL METHOD FOR FLEXIBLE MBS DESCRIBED WITH A NLFE FORMALISM
Figure 5. Optimization results of the two-DOF robot example. The elements are numbered from the ground
to the tip. (a) Cost function history; (b) Deflection constraint for the optimized design; (c) Stress constraint
1 for the optimized design; and (d) Stress constraint
2 for the optimized design.
element mesh is coarse so that differences also arise from this discretization because the convergence
of the finite element mesh is not achieved.
The optimization problem concerns the mass minimization of a 4-bar mechanism inspired from [6,
11, 43]. The 4-bar mechanism consists of three flexible links connected to each other and to
the ground by revolute joints (Figure 6). The three links have a constant solid circular cross
Copyright © 2016 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2016)
DOI: 10.1002/nme
E. TROMME ET AL.
section, Young’s modulus of E D 68:95 GPa, Poisson’s ratio of D 0:3 and mass density of
D 2757 kg/m3 . Each link is modeled by six beam finite elements [20]. The lengths of the links
are l1 D 0:3048 m, l2 D l4 D 0:9144 m and l3 D 0:762 m. The gravity field is considered
in the opposite y-direction. The input crank is driven by an ideal motor, which imposes a con-
stant angular velocity ! D 10 s1 . The simulation is performed using the generalized-˛ scheme
with a time step h D 1 103 s and a spectral radius of 1 D 0. This spectral radius value
enables to filter high frequency loadings in order to incorporate smoothly stress constraints in the
optimization problem.
The design variables pi are the link diameters. Initial values of design variables are set to
356.8 mm. Fixed-free beam conditions are adopted to fix the singularity of the stiffness matrix
during the static response optimization.
The optimization problem is to find the mobile link diameters which minimize the total mass of
the system m .p/ subjected to bending stress constraints
j;e at each time step tj for each beam
element e. The MBS analysis computes the bending moment Mj;e for each element as a function of
time whereupon the bending stresses are calculated from
p
4
j;e D 3=2 Mj;e (59)
Ae
where Ae is the cross sectional area of element e. Mathematically, the optimization problem is
stated as
minimize m .p/
p
subject to
j;e 6 27:58 MPa; j D 301; : : : 500; (60)
0:015 m 6 pi 6 0:5 m; i D 1; : : : ; 3;
where e numbers the beam finite element and goes from 1 to 18. The optimization considers the
time interval between 0.3 and 0.5 s which exactly covers one period of steady state motion after the
transient has died away.
The initial design has a mass m D 546 kg, and all the bending stress constraints are satisfied.
The optimization process is deemed converged when the relative variations of the design variable
values are less than 0.01%. The stopping criteria are different from Equation (54) because small
modifications of the design variable values strongly affect the ESL, which hinders the convergence
with the previous stopping criteria.
The cost function history is illustrated in Figure 7(a) where it is observed that the convergence is
attained after 15 iterations. Figure 7(b)–(d) shows the bending stresses for the optimal design. The
constraints are active for each link at one time step. Table II gathers the numerical results which are
compared with the results given in [6, 11]. According to Etman et al. [11] who performed a dynamic
response optimization process using approximation concepts, this example has multiple local
optima. Hence, the compared results may correspond to different optimal solutions. Furthermore,
the MBS analysis and the optimization process are different for each study. However, the overall
trend of the three methods is in good agreement, and the convergence is achieved within a similar
number of iterations.
Copyright © 2016 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2016)
DOI: 10.1002/nme
ON THE ESL METHOD FOR FLEXIBLE MBS DESCRIBED WITH A NLFE FORMALISM
Figure 7. Optimization results of the four-bar mechanism example. The elements are numbered from the
ground and going clockwise with a counter reset at each new component. (a) Cost function history; (b) Link
1 at the optimized design; (c) Link 2 at the optimized design; and (d) Link 3 at the optimized design.
7. CONCLUSIONS
The ESL method is a powerful tool to perform the optimization of mechanical system components
within a flexible MBS approach. This method transforms the dynamic response optimization prob-
lem into a static response optimization problem subject to multiple load cases. These static load
cases correspond to the ESL that are defined based on the transient analysis response. The ESL
evaluation for MBS optimization strongly depends on the adopted formalism to describe the MBS
Copyright © 2016 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2016)
DOI: 10.1002/nme
E. TROMME ET AL.
dynamics. Initially, the method has been developed for a floating frame of reference formulation
that is suitable to derive ESL because this formalism relies on body-attached frames.
The ESL evaluation for a nonlinear finite element formulation is appealing because this formalism
is completely general to analyze complex MBS and is convenient to perform advanced optimization
processes like topology optimization. However, the extension of the ESL method to a classical
nonlinear finite element formulation is cumbersome mainly because of the tangent stiffness matrix
that evolves with the system configuration.
The present paper proposes an efficient method to derive ESL adapted to a nonlinear finite ele-
ment approach while taking advantages of the Lie group formalism which emerges in the MBS
community. A fundamental property of the particular Lie group formalism concerns the derivation
of the equations of motion with respect to local frames. A definition of ESL based on this for-
malism has been proposed and is valid even in case of large local displacements. The evaluation
of the ESL is a byproduct of the dynamic analysis (no additional processing or frame transforma-
tion). Here, linear displacements within the component have been assumed, which results in a linear
static computation. Classical examples have been treated to demonstrate that the method converges
towards a similar optimized solution as the initial method based on the floating frame of reference
formulation [6].
In comparison with the initial ESL method [6], the use of the particular Lie group formalism
enables to extend the ESL method to a nonlinear finite element formulation without any extra-cost.
This mainly follows from the tangent stiffness matrix of each component that can be considered as
constant in the nodal local frame. Moreover, no additional computation is needed to evaluate the
ESL because they are a byproduct of the MBS analysis. Finally, the use of the Lie group formalism
is promising for advanced optimization because the proposed approach can also account for large
deformation in the static response optimization process at the cost of solving a nonlinear equation.
In the future, the ESL method could be extended to formulate the optimization problem with
respect to the whole system, that is, including kinematic joints and thusly, removing the implicit
hypothesis of formulating the optimization problem with respect to isolated components. Further-
more, the problem of defining boundary conditions on the component during the static optimization
would disappear. Because the proposed method has been validated on classical examples, topology
optimization of more realistic problems could be investigated.
ACKNOWLEDGEMENTS
Part of this research has been supported by the CIMEDE 2 Project sponsored by the pole of competitiveness
“GreenWin" and the Walloon Region of Belgium (Contract RW-7179).
REFERENCES
1. Kang B, Park G, Arora J. A review of optimization of structures subjected to transient loads. Structural and
Multidisciplinary Optimization 2006; 31(2):81–95.
2. Fox R, Kapoor M. Structural optimization in the dynamic regime: a computational approach. AIAA Journal 1970;
8:1798 –1804.
3. Park GJ. Analytic Methods for Design Practice. Springer-Verlag: London, 2007.
4. Bruns T, Tortorelli D. Computer-aided optimal design of flexible mechanisms. Proceedings of the Twelfth Conference
of the Irish Manufacturing Commitee, IMC12, Competitive Manufacturing, University College Cork, Ireland, 1995;
29–36.
5. Choi W, Park G. Transformation of dynamic loads into equivalent static loads based on modal analysis. International
Journal for Numerical Methods in Engineering 1999; 46(1):29–43.
6. Kang B, Park G, Arora J. Optimization of flexible multibody dynamic systems using the equivalent static load
method. AIAA Journal 2005; 43(4):846–852.
7. Häussler P, Minx J, Emmrich D. Topology optimization of dynamically loaded parts in mechanical systems: coupling
of MBS, FEM and structural optimization. Proceedings of NAFEMS Seminar Analysis of Multi-Body Systems Using
FEM and MBS, Wiesbaden, Germany, 2004; 1–11.
8. Hong E, You B, Kim C, Park G. Optimization of flexible components of multibody systems via equivalent static
loads. Structural Multidisciplinary Optimization 2010; 40:549–562.
9. Sherif K, Irschik H. Efficient topology optimization of large dynamic finite element systems using fatigue. AIAA
Journal 2010; 48(7):1339–1347.
Copyright © 2016 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2016)
DOI: 10.1002/nme
ON THE ESL METHOD FOR FLEXIBLE MBS DESCRIBED WITH A NLFE FORMALISM
10. Oral S, Kemal Ider S. Optimum design of high-speed flexible robotic arms with dynamic behavior constraints.
Computers & Structures 1997; 65(2):255–259.
11. Etman L, Van Campen D, Schoofs A. Design optimization of multibody systems by sequential approximation.
Multibody System Dynamics 1998; 2(4):393–415.
12. Brüls O, Lemaire E, Duysinx P, Eberhard P. Optimization of multibody systems and their structural components.
In Multibody dynamics: Computational Methods and Applications, Vol. 23. Springer: Springer Netherlands, 2011;
49–68.
13. Tromme E, Tortorelli D, Brüls O, Duysinx P. Structural optimization of multibody system components described
using level set techniques. Structural Multidisciplinary Optimization 2015; 52:959–971.
14. Bauchau O. Flexible Multibody Dynamics. Springer: Netherlands, 2011.
15. Géradin M, Cardona A. Flexible Multibody Dynamics: A Finite Element Approach. John Wiley & Sons: New York,
2001.
16. Brüls O, Cardona A. On the use of Lie group time integrators in multibody dynamics. Journal of Computational and
Nonlinear Dynamics 2010; 5(3):031002-1–031002-13.
17. Brüls O, Cardona A, Arnold M. Lie group generalized-˛ time integration of constrained flexible multibody systems.
Mechanism and Machine Theory 2012; 48:121–137.
18. Paraskevopoulos E, Natsiavas S. A new look into the kinematics and dynamics of finite rigid body rotations using
Lie group theory. International Journal of Solids and Structures 2013; 50(1):57 –72.
19. Sonneville V, Brüls O. A formulation on the special Euclidean group for dynamic analysis of multibody systems.
Journal of Computational and Nonlinear Dynamics 2014; 9(4):041002-1–041002-8.
20. Sonneville V, Cardona A, Brüls O. Geometrically exact beam finite element formulated on the special Euclidean
group SE(3). Computer Methods in Applied Mechanics and Engineering 2014; 268:451–474.
21. Brüls O, Arnold M, Cardona A. Two Lie group formulations for dynamic multibody systems with large rotations.
Proceedings of the IDETC/MSNDC Conference, Washington D.C., USA, 2011; 85–94.
22. Holm D, Schmah T, Stoica C, Ellis D. Geometric Mechanics and Symmetry: From Finite to Infinite Dimensions.
Oxford University Press: London, 2009.
23. Boothby W. An Introduction to Differentiable Manifolds and Riemannian Geometry (Second edn.) Academic Press:
San Diego, 2009.
24. Wen H, Reddy T, Reid S. Deformation and failure of clamped beams under low speed impact loading. International
Journal of Impact Engineering 1995; 16(3):435–454.
25. Zhang Q, Vrouwenvelder A, Wardenier J. Dynamic amplification factors and {EUDL} of bridges under random
traffic flows. Engineering Structures 2001; 23(6):663 –672.
26. Choe U, Gang S, Sin M, Park G. Transformation of a dynamic load into an equivalent static load and shape opti-
mization of the road arm in self-propelled howitzer. The Korean Society of Mechanical Engineers 1996; 20(12):
3767–3781.
27. Shabana A. Dynamics of Multibody Systems (Fourth edn.) Cambridge University Press: England, 2013.
28. Haftka R, Gürdal Z. Elements of Structural Optimization (Third edn.) Springer: Netherlands, 1992.
29. Boyd S, Vandenberghe L. Convex Optimization. Cambridge University Press: Cambridge, United-Kingdom, 2004.
30. Held A. On structural optimization of flexible multibody systems. Ph.D. Thesis, Institut für Technische und
Numerische Mechanik, Universität Stuttgart, 2014.
31. Choi W, Park G. Structural optimization using equivalent static loads at all time intervals. Computer Methods in
Applied Mechanics and Engineering 2002; 191(19-20):2105–2122.
32. Jung U, Park G. A new method for simultaneous optimum design of structural and control systems. Computers &
Structures 2015; 160:90–99.
33. Lee H, Park G. Nonlinear dynamic response topology optimization using the equivalent static loads method.
Computer Methods in Applied Mechanics and Engineering 2015; 283:956–970.
34. Stolpe M. On the equivalent static loads approach for dynamic response structural optimization. Structural and
Multidisciplinary Optimization 2014; 50:921–926.
35. Cauchy A. Méthodes générales pour la résolution des systèmes d’équations simultanées. Comptes rendus de
l’Académie des Sciences de Paris 1847; 25:536–538.
36. Sigmund O, Maute K. Topology optimization approaches: a comparative review. Structural and Multidisciplinary
Optimization 2013; 48(6):1031–1055.
37. Deaton J, Grandhi R. A survey of structural and multidisciplinary continuum topology optimization: post 2000.
Structural and Multidisciplinary Optimization 2014; 49(1):1–38.
38. Schittkowski K. NLPQL: A fortran subroutine solving constrained nonlinear programming problems. Annals of
Operation Research 1986; 5(2):485–500.
39. Fleury C, Braibant V. Structural optimization: a new dual method using mixed variables. International Journal for
Numerical Methods in Engineering 1986; 23:409–428.
40. Svanberg K. The method of moving asymptotes – a new method for structural optimization. International Journal
for Numerical Methods in Engineering 1987; 24:359–373.
41. Svanberg K. A class of globally convergent optimization methods based on conservative convex separable
approximations. SIAM Journal on Optimization 2002; 12(2):555–573.
42. Bruyneel M, Duysinx P, Fleury C. A family of MMA approximations for structural optimization. Structural and
Multidisciplinary Optimization 2002; 24:263–276.
43. Sohoni V, Haug E. A state space technique for optimal design of mechanisms. Journal of Mechanical Design 1982;
104(4):792–798.
Copyright © 2016 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng (2016)
DOI: 10.1002/nme