Linear dynamics of flexible multibody systems : a

system-based approach
Jawhar Chebbi, Vincent Dubanchet, José Alvaro Perez Gonzalez, Daniel
Alazard

To cite this version:

Jawhar Chebbi, Vincent Dubanchet, José Alvaro Perez Gonzalez, Daniel Alazard. Linear dynamics
of flexible multibody systems : a system-based approach. Multibody System Dynamics, 2016, pp.0.
�10.1007/s11044-016-9559-y�. �hal-01405184�

HAL Id: hal-01405184

https://hal.science/hal-01405184
Submitted on 9 Dec 2016

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.
 Open Archive TOULOUSE Archive Ouverte (OATAO)
OATAO is an open access repository that collects the work of Toulouse researchers and
makes it freely available over the web where possible.

This is an author-deposited version published in: http://oatao.univ-toulouse.fr/

Eprints ID: 16559

To link this article: http://dx.doi.org/10.1007/s11044-016-9559-y

To cite this version: Chebbi, Jawhar and Dubanchet, Vincent and Perez Gonzalez, José
Alvaro and Alazard, Daniel Linear dynamics of flexible multibody systems. (2016)
Multibody System Dynamics. ISSN 1384-5640

Any correspondence concerning this service should be sent to the repository

administrator: [email protected]
Linear dynamics of flexible multibody systems
A system-based approach

Jawhar Chebbi1 · Vincent Dubanchet1 ·

José Alvaro Perez Gonzalez1 · Daniel Alazard1

Abstract We present a new methodology to derive a linear model of flexible multibody

system dynamics. This approach is based on the two-port model of each body allowing
the model of the whole system to be built just connecting the inputs/outputs of each body
model. Boundary conditions of each body can be taken into account through inversion of
some input–output channels of its two-port model. This approach is extended here to treat
the case of closed-loop kinematic mechanisms. Lagrange multipliers are commonly used
in an augmented differential-algebraic equation to solve loop-closure constraints. Instead,
they are considered here as a model output that is connected to the adjoining body model
through a feedback. After a summary of main results in the general case, the case of pla-
nar mechanisms with multiple uniform beams is considered, and the two-port model of the
Euler–Bernoulli beam is derived. The choice of the assumed modes is then discussed re-
garding the accuracy of the first natural frequencies for various boundary conditions. The
overall modeling approach is then applied to the well-known four-bar mechanism.

Keywords Flexible structure · Multibody system · Linear system

1 Introduction
Dynamics of flexible multibody systems has been vigorously studied over the last decades
in connection with various application fields such as light weight robotic arm, rotating ma-
chinery, helicopter rotor with flexible blades, large flexible space structures, and so on. We
can find hundreds of papers about the writing, solving, and simulation of the differential
and algebraic equations governing the dynamics of flexible multibody systems [21, 22]. We
can distinguish methods used to build a validation model including nonlinear terms (cen-
trifugal, Coriolis) [4, 12] and methods used to derive a preliminary design model, which
can be also used for control design [10, 14]. The latter often focus on the linear behavior
of flexible multibody systems under a small displacement assumption. Link flexibility is

B D. Alazard
[email protected]
1 ISAE-SUPAERO, Toulouse, France
 J. Chebbi et al.

approximated by discretized models using finite element methods (FEM) or assumed mode
methods (AMM) [28]. The model of the whole multibody system is then derived using
the Euler–Lagrange formalism. For example, in space applications, considering a flexible
robotic arm fitted on a chaser spacecraft, the inertial load seen from the tip of a flexible
link can vary considerably according to the geometrical configuration of the robotic arm.
The AMM is then often used [4, 28] with the first clamped–mass eigenfunctions. The main
drawback is that the mass must be a time-variant parameter and need to be updated along
with the geometrical configuration. Otherwise, an approximation by an average value is still
possible. However, if we consider the capture of a massive space debris, then the boundary
conditions at the tips of each flexible link may be closer to the clamped–clamped condition
than to the clamped–mass condition. During the preliminary design study of such a system,
we can argue that the system dynamics is more sensitive to such a change in the boundary
conditions than in the effects of nonlinear terms. Thus, there is a real need to have a model
for each link or substructure that is valid for arbitrary boundary conditions before doing the
assembly of the whole structure.
Among all substructuring approaches, the transfer matrix method (TMM) has motivated
lots of literature during last decades. Leckie [11] stated the fundamentals of this method: the
transfer matrix is a relation between the left point and right point state vectors of a flexible
body. The state vector at a point is defined as the augmented vector composed of the general-
ized displacements and the corresponding generalized forces. This method was particularly
well suited for modeling structures composed of serially connected bodies or open-chain-
like structures. Later, this approach was merged with the finite-element (FE) method and
presented as the FE-TM method [5]. The FE-TM method was first developed in [13] to
provide reduced models using a condensation procedure, before being extended in [19] for
complex shape bodies to reduce computation efforts in solving eigenvalue problems. Some
extensions were also developed for control design purposes [27], including noncollocated
feedback [10], and for hybrid (rigid and flexible) multibody systems [20]. In [9], the TMM
was applied to a closed-loop kinematic chain of flexible bodies and more particularly to
the four-bar mechanism. Nevertheless, the TMM has some limitations due to the inversion
of submatrices, which are not always square or invertible, depending on the boundary con-
ditions [27]. Furthermore, another limitation comes from the fact that this method cannot
be directly applied to tree-like structures, like a spacecraft composed of a main rigid body
fitted with several flexible appendages (solar panels, antenna, etc.). In such a case, the objec-
tive of the modeling is to find a relation between the generalized forces and displacements
at the root of the tree-like structure, that is, the main body. Such a model is required for
the design of the attitude and orbit control system (AOCS). Some alternatives and efficient
approaches were developed to treat that case. They are based on the effective mass/inertia
model of each flexible appendage (i.e., the relation between forces and accelerations at the
anchorage point of the appendage on the main body) [7] or also on the impedance matrix
[15, 26]. These methods were first introduced in [3] to analyze dynamic coupling between
substructures through the component mode synthesis (CMS) and are commonly used in
space engineering [8]. The limitation of the effective mass/inertia approach is the loss of
appendage information beyond the anchorage point, such as the deflection at the free tip.
Therefore, such a method cannot be applied to model an arbitrary chain-like structure of
flexible bodies.
The main contribution of this paper is to propose a system-based approach to model the
linear dynamic behavior of a flexible link in a multibody system. This approach is inde-
pendent of the boundary conditions applied at the connecting points of this link. The inputs
and outputs of this model, named two-input–two-output port (TITOP) model, are the forces
Linear dynamics of flexible multibody systems

Fig. 1 Multibody systems

and accelerations at the connecting points of the link with its neighbors. This model em-
beds, in the same minimal state-space representation, the direct and the inverse dynamic
models of the link. Its various input–output channels (or ports) are thus invertible. By ap-
plying very simple channel inversion operations this model can be directly plugged into the
whole multibody system model, seen as a block-diagram model. The boundary conditions
at a connecting point of the link are seen as an external feedback loop between forces and
accelerations. The same model of the link can be used for any kind of open-loop or closed-
loop kinematic mechanisms involving this link. This method fills the gap between the TM
method and the effective mass-inertia method. Note that from the terminology point of view,
what is called the state vector of left and right points in the TM approach are in fact the in-
puts and outputs of the transfer matrix. These inputs and outputs are completely revisited in
the TITOP approach. Finally, accelerations at the connecting points are considered instead
of generalized displacements. The state vector is now the vector of variables associated to
the minimal state-space representation of the TITOP model (which is twice the number of
flexible modes of the considered body). First introduced in [1], the TITOP model approach
is also illustrated on space engineering applications in [14, 16–18].
In Sect. 2, the main results of [1] are summarized and reformulated in the context of
general multibody system with a focus on closed-loop mechanisms. In Sect. 3, the TITOP
model is derived for an Euler–Bernoulli beam. It will be shown that the first two natural
frequencies for any kind of tip conditions are accurately modeled from the simple TITOP
approach. In Sect. 4, the methodology is then applied to the well-known four-bar mecha-
nism: the first natural frequencies and modal shapes are analyzed versus the crank angle and
compared with the results already published [9], [29].

2 Two-port model approach

An N degrees-of-freedom (d.o.f.) multibody system can be divided into bodies (or links)
connected to each other by (prismatic, revolute, cardan, spherical) joints. We can distin-
guish chain-like multibody systems (Fig. 1a), tree-like systems (Fig. 1b), and closed-loop
kinematic systems (Fig. 1c). In all the cases, the objective of this paper is to derive the lin-
ear model of the system dynamics around a given equilibrium configuration of N d.o.f.,
 J. Chebbi et al.

Fig. 2 The ith flexible link in

the structure

θ = [θ1 , θ2 , . . . , θN ]T and θ̇ = 0, assuming small variations of the d.o.f. around this con-
figuration and small deformations of flexible links. Therefore, the Coriolis and centrifugal
forces are neglected throughout this paper.

Frame definition Under this small displacement assumption, it is possible to define an

inertial frame Ri = (Pi0 , xi , yi , zi ) attached to the equilibrium configuration of each link i.
In the chain-like and tree-like cases, Pi0 is the connection point of link i with the parent link
at the equilibrium. Pi/j is the direction cosine matrix (DCM) between frames Ri and Rj
(i.e., the matrix of components of unitary vectors xj , yj , zj in Ri ). Pi/j depends only on the
given configuration θ for all i and j .

2.1 Two-port model of a link

The link Li connected to the parent substructure Li−1 at the point Pi and to the child sub-
structure Li+1 at point Ci is depicted in Fig. 2. The double-port or TITOP (two-input–two-
L
output port) model GPii,Ci (s), proposed in [1], is a linear dynamic model between 12 inputs:
– the six components in Ri of the wrench WCi = [FTCi TTCi ]T applied by the substructure
Li+1 to the link Li at point Ci : FCi stands for the three-component force vector applied at
point Ci , and TCi stands for the three-component torque vector applied at point Ci ,
– the six components in Ri of the acceleration twist ẍPi = [aTPi ω̇TPi ]T (time-derivative of
the twist) of point Pi : aPi stands for the three-component linear acceleration vector at
point Pi , and ω̇Pi stands for the three-component angular acceleration vector at point Pi ,
and 12 outputs:
– the six components in Ri of the acceleration twist ẍCi = [aTCi ω̇TCi ]T of point Ci ,
– the six components in Ri of the wrench WPi = [FTPi TTPi ]T applied by the link Li to the
substructure Li−1 at point Pi ,
and can be represented by the block-diagram depicted in Fig. 3.
L
The way to obtain such a TITOP model GPii,Ci (s) will be detailed in Sect. 3 for Euler–
Bernoulli beams and is also detailed in [1] and [18] in the general case. The general proce-
dure is based on the modal analysis of the link Li with the clamped at Pi -free at Ci boundary
Linear dynamics of flexible multibody systems

Fig. 3 Block-diagram of the

L
TITOP model GP i,C (s) of
i i
link Li

conditions. Indeed, the Lagrange formulation of dynamics using the assumed mode method
(AMM) or finite element method (FEM) leads to a generalized second-order differential
equation [22, 28]:
 Li        
DPi LTPi ẍPi 0 0  0 0 
+ +
LPi 1ni η̈i 0 diag(2ξj ωj ) η̇i 0 diag(ωj2 ) ηi
  
−I6 τ TCi Pi WPi
= T , (1)
0 Φ Ci WCi

where:
– ni is the number of flexible clamped–free modes of link Li characterized by the modal
coordinates vector ηi , the frequencies ωj , j = 1, . . . , ni , and the damping ratio ξj , j =
1, . . . , ni ,
– 1ni is the identity matrix of size ni ,
– LPi is the ni × 6 modal participation factor matrix of the link at point Pi and projected in
the frame Ri ,
– Φ Ci is the 6 × ni projection matrix of the ni clamped–free modal shapes on the six d.o.f.
at point Ci and projected in the frame Ri ,
– τ Ci Pi is the
 “rigid” kinematic
 model between point Ci and Pi projected in the frame Ri :
∗ −−→ −−→
τ Ci Pi = 013 ( C1i Pi ) , where (∗ Ci Pi ) is the skew-symmetric matrix associated with the
3×3 3
vector from Ci to Pi ,
L
– DPii is the 6 × 6 rigid mass model of the link at point Pi and projected in the frame Ri .
The right-hand term of Eq. (1) describes the contribution of the wrenches WPi and WCi
to the generalized force vector. The acceleration twist ẍCi at point Ci can be easily expressed
from the generalized acceleration vector, the modal shapes, and the kinematic model:
 
ẍ
and: ẍCi = [τ Ci Pi Φ Ci ] Pi (2)
η̈i

From Eqs. (1) and (2), the general state-space realization (see also the definition of a
L
state-space realization of a given transfer in Appendix 1) of the TITOP model GPii,Ci (s)
reads:
⎡ ⎤ ⎡ 0ni ×ni 1ni 0ni ×6 0ni ×6
⎤
η̇i
⎢ η̈i ⎥ ⎢ − diag(ωj2 ) − diag(2ξj ωj ) Φ TCi −LPi ⎥
⎢ ⎥=⎢ ⎢
⎥
⎥
⎣ ẍ ⎦ ⎣ −Φ diag(ω2 ) −Φ diag(2ξ ω ) Φ ΦT (τ − Φ L )⎦
Ci Ci j Ci j j Ci Ci Ci Pi Ci Pi
L
WPi LTPi diag(ωj2 ) LTPi diag(2ξi ωi ) (τ Ci Pi − Φ Ci LPi )T −DPii + LTPi LPi
⎡ ⎤
ηi
⎢ η̇i ⎥
×⎢
⎣W ⎦.
⎥ (3)
Ci
ẍPi
 J. Chebbi et al.

This model embeds, in the same minimal state-space realization, the direct dynamic
model (transfer from acceleration to force) of link i at the point Pi and the inverse dy-
namic model (transfer from force to acceleration) of link i at the point Ci . All data re-
quired in Eq. (3) are commonly provided by FEM software. In [18], the link between
the TITOP model and substructuring methods like the Graig–Bampton decomposition and
the component modes synthesis (CMS) is detailed in the general case. In [14], a custom
NASTRAN/S IMULINK interface was developed to declare and manipulate the TITOP model
L
GPii,Ci (s) as a linear dynamic system object under the M ATLAB/S IMULINK environment.
Let us note that the modal analysis of the left-hand term of Eq. (1) corresponds to the free
(at Pi )–free (at Ci ) case and provides six rigid modes in addition to the free–free flexible
L
modes. These rigid modes are not taken into account in the model GPii,Ci (s) defined by
Eq. (3). This is one the main advantages of the TITOP model: the boundary conditions
and the rigid modes can be taken into account “outside” the model. This is detailed in the
following section.

2.2 Interconnection and channel inversion of TITOP models

L
The open-loop dynamics of the TITOP model GPii,Ci (s), that is, when the inputs are null,
corresponds to the link dynamics when it is clamped at Pi (ẍPi = 0) and free at Ci (WCi = 0).
L
It must be noticed that all the 12 input–output channels of the model GPii,Ci (s) are invertible
assuming that a residual mass (inertia) of link i is attached to the points Pi and Ci along
(around) the three axes. Such an assumption is valid by using the finite-element method or
the assumed mode method to model the flexibility of the link.
L
[GPii,Ci ]−1I (s) denotes the model where the channels numbered in the index vector I are
inverted according to the procedure described in Appendix 1. This channel inversion opera-
tion is very useful to analyze the dynamics of the link Li for various boundary conditions:
L
– clamped at Pi and clamped at Ci , using [GPii,Ci ]−1[1:6] (s) whose inputs are ẍCi and ẍPi ,
L
– free at Pi and free at Ci , using [GPii,Ci ]−1[7:12] (s) whose inputs are WCi and WPi ,
L
– free at Pi and clamped at Ci , using [GPii,Ci ]−1 (s) whose inputs are ẍCi and WPi ,
– or to take into account a revolute (resp. prismatic) joint at the point Pi around (resp. along)
L L
the zi -axis using [GPii,Ci ]−112 (s) (resp. [GPii,Ci ]−19 (s)), that is, no torque (resp. force) can
be applied by the link Li to the parent substructure Li−1 around (resp. along) the zi -axis.
L
A spherical joint can be taken into account using [GPii,Ci ]−1[10:12] (s).
More generally, once the TITOP model of each link is defined and the DCM Pi/j for
a given nominal configuration θ are computed, the linear model of the whole multibody
system can be built by simply connecting the inputs/outputs of the TITOP models. Hereafter,
some examples of interconnection are presented.

Chain-like multibody system Let us consider a flexible spacecraft S whose reference

body frame is (O, x0 , y0 , z0 ), fitted with a flexible boom B at point P1 with a revolute joint
along the (P1 , z1 )-axis. The boom holds a flexible antenna A at point P2 . A sketch of such
a multibody system is depicted in Fig. 4 (left). The block-diagram associated to this system
is presented in Fig. 4 (right). It involves the three TITOP models of the three bodies with
the channel inversions required by the different boundary conditions. Pai/j is the augmented
6 × 6 DCM:
 
Pi/j 03×3
Pi/j =
a
.
03×3 Pi/j
Linear dynamics of flexible multibody systems

Fig. 4 A flexible boom B linking a flexible antenna A to a flexible spacecraft S through a revolute joint
(left) and its block diagram model (right)

In this block-diagram, seven double integrators are added on the outputs to represent
the seven rigid modes associated to the seven d.o.f. of this system: the six d.o.f. of the
spacecraft and one d.o.f. of the revolute joint. The seven inputs of this model are the torque
u1 applied inside the revolute joint by the driven mechanism and the six components of the
wrench W0 applied by the attitude and orbit control system at the reference point O. All
dynamic couplings between the flexible modes of the antenna, the boom, and the spacecraft
are directly taken into account through the feedback connections between the three TITOP
models. Furthermore, the inertia j1 , the stiffness k1 , and the damping factor f1 of the driven
mechanism located in the revolute joint can be added by a feedback loop u1 = −j1 δθ ¨1−
˙ 1 − k1 δθ1 , directly plugged into the block-diagram. This can be used to handle any
f1 δθ
arbitrary boundary condition in the revolute joint, from the free condition (j1 = k1 = f1 = 0)
to the clamped condition (k1 → ∞).

Tree-like multibody system In this case, one link B may have two (or more) child bodies
connected to B at points C1 and C2 . This case is quite similar to the previous one and
requires a model of the link augmented with additional ports. This case is not detailed in
this paper. The reader is advised to refer to [1] where the three-port model of a link (denoted
GBP ,C1 ,C2 (s)) and an example of a tree-like system are detailed (see Fig. 18 in [1]).

Closed-loop multibody system Such mechanisms are commonly modeled through the
Lagrange approach using Lagrange multipliers [22, 24]. They are used to be denoted λ and
correspond to the generalized constraint forces. Considering the model of the link Li given
in Eqs. (1) and (2) with a kinematic-chain closed at point Ci , the Lagrange multipliers are
nothing else than the wrench at this point applied by the link Li : λ = −WCi . The classical
Lagrange approach leads to an augmented system of differential-algebraic equations:
⎡ Li ⎤⎡ ⎤ ⎡ ⎤⎡ ⎤
DPi LTPi τ TCi Pi ẍPi 0 0 0 
⎣ LP 1ni Φ TCi ⎦ ⎣ η̈i ⎦ + ⎣ 0 diag(2ξj ωj ) 0 ⎦ ⎣ η̇i ⎦
i
τ Ci ⎡ Φ Ci 0 λ 0 0 0 
Pi ⎤⎡ ⎤ ⎡ ⎤
0 0 0  −WPi
+ ⎣ 0 diag(ωj2 ) 0 ⎦ ⎣ ηi ⎦ = ⎣ 0 ⎦ . (4)
0 0 0  ẍCi
 J. Chebbi et al.

Fig. 5 A flexible closed-loop structure (left) and its block diagram model (right)

Fig. 6 A uniform beam B in the plane (P 0 , x, y)

Considering the TITOP formulation of the model in Eq. (3), λ (up to the sign) is an input of
L L
GPii,Ci (s) or an output of the model [GPii,Ci ]−1[1:6] (s) where the upper port is inverted. Thus,
L
the upper port of [GPii,Ci ]−1[1:6] (s) can be connected in a feedback loop to the upper port of
the adjoining link in order to close the kinematic chain. Such an interconnection is presented
in Fig. 5 (right) in the case of a very simple closed-loop structure composed of two links L1
and L2 clamped to the ground and one to each other at point C1 ≡ C2 . In that case, the model
outputs are the two wrenches WP1 and WP2 applied by the system to the ground at points
P1 and P2 . Similarly to the case of the chain-like multibody system example, a revolute
L
joint around the (C2 , z1 )-axis can be taken into account using the model [GP22,C2 ]−1[1:5] (s) to
cancel the torque around this axis in the joint between L1 and L2 , that is, λ(6) = WCi (6) = 0
(see also the illustration on the four-bar mechanism in Sect. 4).

3 Two-port model of a beam in the planar case

Let us consider a uniform beam B between points P and C, characterized by the following
parameters (see Fig. 6):
– mass density ρ (kg/m3 ),
– section S (m2 ),
– length l (m),
– Young modulus E (N/m2 ),
– second moment of area w.r.t. the z-axis: Iz (m4 ).
The objective is to compute the TITOP model GB P ,C (s). In the (3 d.o.f.) planar case, this
model will be denoted G(s) and is a transfer between six inputs:
y
– three components FCx (N), FC (N), and TCz (Nm) in the frame R = (P 0 , x, y, z) of the
planar wrench applied to the beam at point C,
Linear dynamics of flexible multibody systems

Fig. 7 Parameterization of the beam deflection at a given time t

– three components üP (m/s2 ), v̈P (m/s2 ), and α̈P (rad/s2 ) in the frame R of the planar
acceleration twist of the beam at point P ,
and six outputs:
– three components üC (m/s2 ), v̈C (m/s2 ), and α̈C (rad/s2 ) in the frame R of the planar
acceleration twist of the beam at point C,
y
– three components FPx (N), FP (N), and TPz (Nm) in the frame R of the planar wrench
applied by the beam at point P .
This model G(s) is derived in following sections considering a decoupling between the
pure flexion in the plane (P 0 , x, y) and the pure traction–compression along the (P 0 , x)-
axis.

3.1 Pure flexion in the plane (P 0 , x, y)

A well-known approach to derive the dynamic model of a flexible beam is based on the
Lagrange technique combined with the assumed modes method (AMM) [4]. This approach
involves a moving body frame R = (P , x̃, ỹ, z) attached to the beam at point P (see Fig. 7).
Under the small deflection assumption, the displacement v(x, t) along the y-axis at any point
of abscissa x and at any time t reads as follows:

v(x, t) = vP (t) + xαP (t) + ṽ(x, t) ∀x ∈ [0, l]. (5)

The deflection ṽ(x, t) in the moving frame R is split up into n shape functions φi (x)
(φ(x) = [φ1 (x) · · · φn (x)]T being the shape function vector), associated with n time-domain
functions qi (t), n = 1, . . . , n (q(t) = [q1 (t) · · · qn (t)]T ):
n
ṽ(x, t) = φi (x)qi (t) = φ(x)T q(t). (6)
i=0

Then, the kinetic energy reads

⎡  ⎤T ⎡ ⎤
 l v̇P   v̇P
1 2 1 Mrr MTrf
T = ρS v̇P + x α̇P + Σi=0
n
φi (x)q̇i dx = ⎣ α̇P ⎦ ⎣ α̇P ⎦ ,
2 0 2 Mrf Mff
q̇ q̇
 J. Chebbi et al.

where:
 2 
ρSl ρSl
– Mrr = is the 2 × 2 rigid mass matrix,
2
ρSl 2 ρSl 3
l 2 3
l
– Mrf = [ρS 0 φ(x) dx ρS 0 xφ(x) dx] is the n × 2 rigid-flexible coupling mass matrix,
l
– Mff = ρS 0 φ(x)φ(x)T dx is the n × n flexible mass matrix.

The elastic potential energy reads

⎡ ⎤T ⎡ ⎤
 l 2 2 vP   vP
1 ∂ ṽ(x, t) 1 02×2 02×n
V= EIz dx = ⎣ αP ⎦ ⎣ αP ⎦ ,
2 0 ∂x 2 2 0n×2 Kn×n
q q

l 2 d 2 φ(x) T
 l  T
where K = EIz 0 [ d dxφ(x)
2 ][ dx 2 ] dx = EIz 0 φ (x)φ (x) dx is the n × n stiffness ma-
trix.
Applying the action–reaction principle at point P , the work of external loads is

Wext = −vP FP − αP TPz + v(l)FC + v  (l)TCz

y y
(7)
⎡ y ⎤
FP
  ⎢ TPz ⎥
= −vP − αP vP + lαP + qT φ(l) αP + qT φ  (l) ⎢⎣ FCy
⎥
⎦ (8)
TCz
⎡ ⎤
⎡
⎤T ⎡ ⎤ FPy
vP −1 0 1 0 ⎢Tz ⎥
= ⎣ αP ⎦ ⎣ 0 −1 l 1 ⎦⎢ P ⎥
⎣ FCy ⎦ , (9)

q 0n×1 0n×1 φ(l) φ (l)
TCz

and the Lagrange derivation leads to the following second-order equation:

y ⎤ ⎡
⎡ ⎤ ⎡ ⎤ FP
  v̈  v   
Mrr MTrf ⎣ P ⎦ 0 02×n ⎣ P ⎦ −I2 τT ⎢ Tz ⎥
α̈P + 2×2 αP = ⎢  Py  ⎥
Mrf Mff 0n×2 Kn×n 0n×2 Φ (l) ⎣ FC ⎦
T
q̈ q
TCz
  (10)
with τ = 10 1l , the rigid kinematic model between point C and P of the beam in the planar
case, and Φ(l) = [φ(l) φ  (l)]T . The relation between accelerations at points C and P , under
the low-speed motion assumption reads

   
v̈C v̈
= τ P + Φ(l)q̈. (11)
α̈C α̈P

From Eqs. (10) and (11) we can easily derive the state-space representation, associated with
the state vector xf = [qT q̇T ]T , of the TITOP model Gf (s) relative to the pure flexion
Linear dynamics of flexible multibody systems

Fig. 8 TITOP model Gf (s)

block diagram of a flexible beam
in bending

dynamics in the plane (P 0 , x, y) (see Fig. 8):

⎡ ⎤
q̇ ⎡ ⎤
⎢ q̈ ⎥ 0n 1n 0n×2 0n×2
⎢ ⎥
⎢ ⎥ ⎢ −M−1 K 0 M−1 T
Φ (l) −M−1
ff Mrf
⎥
⎢ v̈C ⎥ ⎢ ff n ff ⎥
⎢ ⎥=⎢ ⎥
⎢ α̈C ⎥ ⎣ −Φ(l)M−1 ff K 0n Φ(l)M−1 T
ff Φ (l) τ − Φ(l)M−1 ff Mrf
⎦
⎢ y ⎥
⎣ FP ⎦ −1
Mrf Mff K 0n (τ − Φ(l)Mff Mrf )T
T −1
MTrf M−1
ff Mrf − Mrr
z
TP
⎡ ⎤
q
⎢ q̇ ⎥
⎢ ⎥
⎢ Fy ⎥
×⎢ C ⎥ (12)
⎢ T z ⎥.
⎢ C ⎥
⎣ v̈ ⎦
P
α̈P

3.1.1 Choice of the modal decomposition

In this section, we discuss the choice of the decomposition defined in Eq. (6) regarding
the capability of the TITOP model Gf (s) (and its inverses) to take into account various
boundary conditions. Among all possible boundary conditions, a very simple test consists
in computing and comparing the natural frequencies ωcf,i and ωf c,i of models Gf (s) and
G−1
f (s), respectively. Indeed, Gf (s) is the model of the beam under the clamped–free condi-
tion and G−1
f (s) is the model of the beam under the free–clamped condition. Since the beam
is uniform, the natural frequencies of the two models must be equal: ωcf,i = ωf c,i ∀i.
Two decompositions are considered to build the TITOP model Gf (s):
– the well-known clamped–free assumed mode decomposition [12]. The model obtained
from this decomposition is denoted Gf,n (s), where n is the number of assumed modes.
– the decomposition based on fifth-order polynomial shape functions [14]. The model ob-
tained from this decomposition is denoted Gf,pol (s).

Model Gf,n (s) In this case, the decomposition is based on the n first modes of the Euler–
Bernoulli beam in the clamped–free condition, and the shape functions in (6) are [2, 6]

cos βi l + cosh βi l
φi (x) = cos βi x − cosh βi x − (sin βi x − sinh βi x), (13)
sin βi l + sinh βi l

where βi , i = 1, . . . , n, are the n first solutions of the characteristic

 equation cos βl cosh βl +
1 = 0. The natural frequencies are given by ωi,ref = βi2 EI
ρS
z
and are the clamped–free refer-
ence values to evaluate approximate solutions. Such a decomposition basis is orthonormal
and leads to Mff = In and K = diag([ωi,ref
2
]). The values of Mrf and Φ(l) are not detailed
for brevity.
 J. Chebbi et al.

Model Gf,pol (s) In this approach, given in more details in [14], the deflection ṽ(x, t) in
the moving frame R is expressed as

ṽ(x, t) = a2 (t)x 2 + a3 (t)x 3 + a4 (t)x 4 + a5 (t)x 5 .

This fifth-order polynomial ensures that ṽ(0, t) = ṽ  (0, t) = 0 ∀t . The coefficients ai (t) are
then expressed as a function of four time-domain functions qi (t) defined by
 T
q(t) = ṽ  (0, t) ṽ(l, t) ṽ  (l, t) ṽ  (l, t) , (14)

and the shape functions vector reads

⎡1 ⎤
x 2 − 2l3 x 3 + 2l32 x 4 − 2l13 x 5
2
⎢ 10 3
x − 15 x 4 + l65 x 5 ⎥
φ(x) = ⎢
⎣
l3
−4
l4 ⎥.
⎦ (15)
l2
x 3
+ 7
l3
x 4
− 3 5
l4
x
1 3
2l
x − 1 4
l2
x + 1 5
2l 3
x

Data required to build the TITOP model Gf,pol (s) defined in Eq. (12) are obtained using
following decompositions:
⎡ ⎤
55440 27720l 462l 2 27720 −5544l 462l 2
⎢ 27720l 18480l 2 198l 3 19800l −3432l 2 264l 3 ⎥
  ⎢ ⎥
Mrr MTrf ρSl ⎢⎢ 462l 2 198l 3 6l 4 181l 2 −52l 3 5l 4 ⎥
⎥
= ⎢ ⎥,
Mrf Mff 55440 ⎢ 27720 19800l 181l 2 21720 −3732l 281l 2 ⎥
⎢ ⎥
⎣ −5544l −3432l 2 −52l 3 −3732l 832l 2 −69l 3 ⎦
462l 2 264l 3 5l 4 281l 2 −69l 3 6l 4
(16)
⎡ ⎤
1 0
⎡ ⎤ ⎢l
6l 4 −30l 2 8l 3 l4  T  ⎢ 0⎥⎥
EIz ⎢ ⎢ ⎥
⎢ −30l
2
1200 −600l 30l 2 ⎥
⎥,
τ ⎢0 0⎥
K= = ⎢ ⎥. (17)
70l 3 ⎣ 8l 3
−600l 384l 2 −22l 3 ⎦ T
Φ (l) ⎢1 0⎥
⎢ ⎥
l4 30l 2 −22l 3 6l 4 ⎣0 1⎦
0 0

The clamped–free and free–clamped models obtained from the TITOP models Gf,pol (s) and
Gf,n (s) (for various numbers n of assumed modes) are now evaluated in Table 1 regarding
the accuracy of the first two natural frequencies. Of course, the model Gf,n (s) provides
the exact clamped–free frequencies for any values of n since the clamped–free modes are
exactly taken into account in this approach. But its inverse, which must represents the free–
clamped model, is not accurate at all, even for high values of n (for n = 16, the order of the
model Gf,16 (s) is 32!!). In comparison, the model Gf,pol (s) is an eighth-order model with
only four flexible modes, which is quite accurate regarding the first two natural frequencies
from the clamped–free to free–clamped conditions.
In the following section, the TITOP model Gf,pol (s) is used to model an Euler–Bernoulli
beam under the various well-known boundary conditions: free, clamped, support, or sliding
on each of the two tips of the beam. In each case, the first two natural frequencies are
compared with the reference values provided by the beam theory.
Linear dynamics of flexible multibody systems

Table 1 Comparison of the first two frequencies of TITOP models and their inverses. The direct model
corresponds to the clamped–free model of an Euler–Bernoulli beam. The
inverse model corresponds to the
EIz
free–clamped model. Frequencies are expressed using normalized units ( )
ρSl 4

i ωi,ref Gf,pol (s) G−1

f,pol (s) Gf,n (s) G−1
f,4 (s) G−1
f,8 (s) G−1
f,16 (s)

1 3.5160 3.5160 3.5160 3.5160 ∀n 4.4432 3.9084 3.7508

2 22.034 22.158 22.158 22.034 ∀n 27.486 24.579 23.585

3.1.2 Other boundary conditions

The various classical (clamped, free, pinned, sliding) boundary conditions are taken into ac-
count by setting to zero the corresponding inputs of Gf,pol (s), or of its inversion around one
or more channels. The procedure for such a channel inversion is described in Appendix 1
under state-space formalism and can be directly applied to the TITOP model Gf,pol (s) de-
fined in Eq. (12). Note that the eight-component state vector xf of Gf,pol (s) or of its various
inverses has a physical meaning and is defined by
 
q
xf = ,
q̇

where more particularly (from (5) and (14))

xf (2) = q(2) = ṽ(l) = vC − vP − lαP ,

(18)
xf (3) = q(3) = ṽ  (l) = αc − αP .

In the various cases presented in Appendix 2, the channel inversion to be used on the TITOP
model Gf,pol (s) in order to take into account the corresponding boundary conditions is de-
tailed. Some boundary conditions can introduce also some internal state conditions, which
can be used to reduce the order of the model. These conditions are also detailed in the
various cases presented in Appendix 2. The first two natural frequencies ωi , i = 1, 2, are
then compared with the reference values ωi,ref provided by beam theory [2] under the same
|ωi −ωiref |
boundary conditions, and the relative errors ωi = 100 ωiref
, i = 1, 2, are computed and
summarized in Tables 3 to 11.
Finally, we have to keep in mind that the model Gf,pol (s) and its various inverses describe
the relationship between forces and accelerations. These models provide the frequencies of
the flexible modes for various boundary conditions. To take into account the rigid modes that
can appear in some cases (free–free, pinned–free, sliding–free, sliding–sliding), we have to
consider the positions vP , αP , vC , and αC as outputs of an augmented model. Of course,
these positions can be obtained from accelerations using twice integrations, but it is rec-
ommended to use the internal state xf and Eqs. (18) to describe this augmented model by
a minimal state-space representation. The block-diagram representation of this augmented
model and the number of rigid modes are also detailed for the various cases presented in
Appendix 2.
In every case, the TITOP model Gf,pol (s) and its inverses provide a quite good approxi-
mation of the first two natural frequencies.
In conclusion, the polynomial approximation done in the TITOP model Gf,pol (s) leads
to a simple analytical model (order 8), which can be used during preliminary design phase.
 J. Chebbi et al.

It allows varying conditions on a very large range, from clamped to free conditions, for
each translational or rotational d.o.f. The fifth-order polynomial shape functions chosen to
build the TITOP model allow the first two natural frequencies to be representative enough
for arbitrary boundary conditions. The only task needed to vary these boundary conditions
consists of the inversion of input/output channels of the TITOP model or the feedback of
some dynamic model between its input and output ports. That will be illustrated in Sect. 4.

3.1.3 Modal shapes of TITOP models

I −1
Let us consider the eighth-order TITOP model of the beam Gf,pol (s) or its inverses Gf,pol (s).
For all these models, the state vector is still xf (t) = [q (t) q̇ (t)] .
T T T

Let A, B, C, and D be the four state-space matrices of one of these models. Then
the eigenvalue decomposition of A characterizes the free response of the model to initial
conditions xf (0) (i.e., when the inputs are set to 0). The eigenvalues λi of A are auto-
conjugate pairs associated with each natural frequency (λi = ±j ωi ) of the ith flexible mode
(i = 1, . . . , 4). The associated eigenvector vi represents the magnitude of the ith mode on
the components of the state vector xf :

  4
q(t)
xf (t) = = gi vi ej (ωi t+ϕi ) , (19)
q̇(t)
i=1

where the scalars gi and ϕi depend on the initial conditions xf (0).

From Eqs. (5) and (6), the deformation of the beam is
⎡⎤
 vP (t)

v(x, t) = 1 x φ (x) ⎣ αP (t) ⎦ ,
T

q(t)

where φ(x) is defined by (15).

Modal shapes can be characterized by their contributions to the second time-derivative
of the beam deflection v̈(x, t) (which is proportional to a factor −ωi2 to the beam deflec-
tion v(x, t)). Indeed, let ψi (x) be the modal shape of the ith mode; then v(x, t) can be
decomposed as
4
v(x, t) = gi ψi (x)ej (ωi t+ϕi )
i=1

and
⎡ ⎤
4
 v̈P (t)
v̈(x, t) = − gi ωi2 ψi (x)ej (ωi t+ϕi ) = 1 x φ T (x) ⎣ α̈P (t) ⎦ . (20)
i=1 q̈(t)
q̈(t) can be decomposed using (19) and the state matrix A (see Appendix 1 for the submatrix
notation):

4
  4
 
q̈(t) = gi A [5 : 8], : vi ej (ωi t+ϕi ) = − gi ωi2 vi [1 : 4] ej (ωi t+ϕi ) . (21)
i=1 i=1
Linear dynamics of flexible multibody systems

Fig. 9 Modal shapes of the first (top) and second (bottom) free–clamped modes: ψi (x) (solid bold), reference
φi (x) (dashed), and error φi (x) − ψi (x) (solid)

If the beam is clamped at P , then v̈P (t) = α̈P (t) = 0, and the identification of Eq. (20) with
(21) gives
   
ψi (x) = −φ T (x)A [5 : 8], : vi /ωi2 = φ T (x)vi [1 : 4] .

In the other cases, according to the channels (defined in the vector of indexes I) which are
−1I
inverted in the model Gf,pol (s), v̈P (t) and α̈P (t) can be outputs in the model, and their
contribution to the modal shape ψi (x) can be characterized using the matrix C of the model
state-space representation. In a general way, we can define:

CvP = C(3, :) if 3 ∈ I, CvP = 01×8 otherwise,

CαP = C(4, :) if 4 ∈ I, CαP = 01×8 otherwise.

Then,
⎡ ⎤
  CvP
ψi (x) = − 1 x φ T (x) ⎣ CαP ⎦ vi /ωi2 . (22)
A([5 : 8], :)

Thus, the modal shapes for arbitrary boundary conditions can be easily characterized
from the eigenvectors/eigenvalues decomposition of the TITOP model or its inverses. As
an example, the modal shapes ψ1 (x) and ψ2 (x) of the first two modes of the free–clamped
−1[1:4]
beam modeled by Gf,pol (s) are depicted in Fig. 9 and compared with the reference modal
shapes φ1 (x) and φ2 (x) provided by Eq. (13) by changing x by l − x. We can appreciate the
accuracy of the first two modal shapes provided by the TITOP model.
 J. Chebbi et al.

3.2 Pure traction along (P 0 , x)-axis

Only one flexible mode is considered to represent the traction–compression along the
(P 0 , x)-axis. Let u(x, t) be the axial deformation at any point of abscissa x (x ∈ [0; l])
due to axial forces −FPx and FCx applied to the beam at points P and C, respectively. Then,
we assume that
  
x  x x uP (t)
u(x, t) = uP (t) + uC (t) − uP (t) = uP (t) + δu (t) = 1 ,
l l l δu (t)

that is, the deformation at x is proportional to the total deformation δu (t) of the beam.
Then, the kinetic energy reads
  2  
1 l ∂u 1 u̇P
Tx =
ρS (x, t) dx = [u̇P δ̇u ]Mx
2 0 ∂t 2 δ̇u
 
1 1/2
with Mx = ρSl 1/2 1/3 , and the potential energy reads

 l  2  
1 ∂u 1 up
Vx = ES (x, t) dx = [up δu ]Kx
2 0 ∂x 2 δu
 
with Kx = ES
l
00
01
.
The TITOP model Gt (s) relative to the traction/compression dynamics along the (P 0 , x)-
axis is
⎡ ⎤ ⎡ 0 1 0 0
⎤⎡ ⎤
δ̇u δu
⎢ δ̈u ⎥ ⎢
3 ⎥⎢ ⎥
− 3E
0 3
− ⎥ ⎢ δ̇u ⎥
⎢ ⎥=⎢ ρl 2 ρSl 2
⎢
⎣ ü ⎦ ⎣ − 3E 0 3 ⎥ ⎢ ⎥. (23)
C ρl 2 ρSl
− 12 ⎦ ⎣ FCx ⎦
FPx 3ES
0 − 12 − ρSl üP
2l 4

3.3 Two-port model G(s) of a beam in the planar case

From Eqs. (12) and (23), the 6 × 6 tenth-order TITOP model G(s) of the beam in the plane
(P 0 , x, y) reads
 
Gf (s) 04×2
G(s) = T TT , (24)
02×4 Gt (s)
where the permutation matrix T is
⎡ ⎤
0 0 0 1 0 0
⎢1 0 0 0 0 0⎥
⎢ ⎥
⎢0 1 0 0 0 0⎥
T=⎢
⎢0
⎥.
⎢ 0 0 0 0 1⎥⎥
⎣0 0 1 0 0 0⎦
0 0 0 1 0 0

The block-diagram of G(s) is depicted in Fig. 10.

In [14], the TITOP model of a uniform beam in detailed in the three-dimensional case.
This parametric model is then used to model and design a boom linking a flexible deployable
antenna to a flexible spacecraft.
Linear dynamics of flexible multibody systems

Fig. 10 TITOP model G(s)

block diagram

Fig. 11 Four-bar mechanism

Table 2 Four-bar mechanism links properties

i 0 1 2 3

Name ground crank coupler follower

Length li (m) 0.254 0.108 0.2794 0.2705
Cross section Si (m2 ) – 1.0774 × 10−4 4.0645 × 10−5 4.0645 × 10−5
Flexural rigidity (EI )i (Nm2 ) – 11.472 0.616 0.616

Each link is a uniform beam with mass density ρ = 2714 Kg/m3 and modulus of elasticity E = 7.11010 N/m2 .
The lumped masses of bearing assemblies in revolute joints between link 2 and link 3 and between link 3 and
link 4 are equal: ml = 0.042 kg

4 Illustration: planar flexible four-bar mechanism

In this section, the two-input–two-output model approach presented in the previous section
is applied to the well-known four-bar mechanism depicted in Fig. 11, taking into account
the flexibility in the crank, in the coupler, and in the follower. This mechanism is a closed-
loop kinematic chain with one rigid degree-of-freedom and was largely studied in the past
decades [9, 23, 29] and more recently [25] to design an efficient vibrating actuator. The ob-
jective is to compute the linear model G4 bars (s, θ1 ) between the torque TPz1 applied on the
crank at point P1 and the angular acceleration of the crank α̈P1 = δ θ̈1 for a given angular
configuration θ1 . First natural frequencies and modal shapes of the mechanism are then ana-
lyzed with respect to the crank angle θ1 . The numerical values are taken from [9] (translated
in the International System of Units) for result comparison and summarized in Table 2. The
corresponding natural frequencies are then called free frequencies (and denoted ωf (i)) since
the crank is free to rotate around the (P1 , z1 ) axis. The modeling approach provides also the
cantilevered frequencies ωc (i), thats is, when the crank is clamped on the ground, which are
easily computed by considering the inverse model G−1 4 bars (s, θ1 ).
From data of Table 2, TITOP models of the crank G1 (s), the coupler G2 (s), and the
follower G3 (s) can be built from Eqs. (12), (16), (17), (23), and (24). The ground (i = 0
 J. Chebbi et al.

or 4) is assumed to be rigid. The planar wrenches and acceleration twists on the inputs
and outputs of each model Gi (s) are expressed in the body frame Ri . Considering the linear
model around a given angular configuration (small variations), all the frames Ri are assumed
to be inertial frames. For a given crank angle θ1 , two kinematic constraints hold:

l1 cos θ1 + l2 cos (θ1 + θ2 ) + l3 cos (θ1 + θ2 + θ3 ) = l0 ,

l1 sin θ1 + l2 sin (θ1 + θ2 ) + l3 sin (θ1 + θ2 + θ3 ) = 0.

They are solved in θ2 and θ3 using the equations given by Shigley and Uicker [23]. The
values of θi , i = 1, 2, 3, allow the (planar) direction cosine matrices between the various
frames Ri to be computed:
 
cos θi − sin θi
Pi/i−1 = .
sin θi cos θi

The revolute joints between the links enforce geometrical constraints between inputs and
outputs of models Gi (s):
   
üPi üCi−1
Pi ≡ Ci−1 ⇒ = PTi/i−1 .
v̈Pi v̈Ci−1

Some dynamic boundary conditions also hold on the inputs of the three models using chan-
nel(s) inversion if required:
– for the crank (i = 1) and the coupler (i = 2), the inputs of the model i must take into
y
account that: (i) FCxi and FCi are constrained by the contact force applied by link i + 1
and the inertial force of the lumped mass:
     
FCxi FPxi+1 üCi
y = Pi+1/i y − m l , (25)
FCi FPi+1 v̈Ci

(ii) TCzi = 0 due to the revolute joint (no torque can be applied though the passive revolute
joint), (iii) −TPzi is equal to the torque u applied by the driven system (crank case) or
TPzi = 0 if the revolute joint is free (coupler case). Thus, TPzi must be on the input, and the
−1 −1
models to be used for the crank and the coupler are G1 6 (s) and G2 6 (s), respectively,
– for the follower (i = 3), the same condition on TPz3 appears, but it is necessary to invert
the first two channels of G3 (s) in order to close the kinematic loop, and to impose that
üC3 and v̈C3 are zero due to the revolute joint on ground. Then, the model to be used for
−1
the follower is G3 [1,2,6] (s) and the first two outputs are the forces FCx3 and FC3 applied by
y

the follower on the ground.

All these geometrical and dynamical boundary conditions can be easily expressed using the
block diagram model depicted in Fig. 12. From this block diagram representation, the model
G4 bars (s, θ1 ) between the torque u applied to the crank and the crank acceleration y = α̈P1
can then be easily derived using M ATLAB/S IMULINK and the function linmod. The modal
analysis of this model provides the free pulsations ωf (i) of the mechanism. Note that the
order of G4 bars (s, θ1 ) is 30: this model includes 15 flexible modes (five modes per link with
four modes for bending in the plane (Pi , yi , zi ) and one mode for traction–compression
along the (Pi , xi )-axis). But this model does not include the rigid mode of the mechanism.
Linear dynamics of flexible multibody systems

Fig. 12 Block-diagram of model G4 bars (s, θ1 ) between the crank torque u and the crank acceleration
y = α̈P1

For control design purpose, this rigid mode can be included augmenting the model by a
double integration:
αp1 1
(s, θ1 ) = G4 bars (s, θ1 ) 2 .
u s
To evaluate the accuracy of this model, the cantilevered pulsations ωc (i) are computed
from the inverse model G−1 4 bars (s, θ1 ) and compared with the results already published on the
same mechanism [9], [29]. The first, second, and third cantilevered frequencies are plotted
in Fig. 13 versus crank angle. The modal shapes of the mechanism computed from the
accelerations (see Sect. 3.1.3) of the first three modes for two crank angle configurations
(0 and 180◦ ) are plotted in Figs. 14 and 15. We can see that these results fit exactly the ones
presented in [9].

5 Conclusions

The objective of this paper is to present the interest of the two-input–two-output port ap-
proach to model any kind of flexible multibody systems. The novelty lies in the manipulation
of transfer between accelerations and forces and to embed, in the same minimal state-space
representation, the direct dynamic model of a body seen from one point and the inverse
dynamic model seen from another point. All channels of this transfer are invertible allow-
ing us to consider arbitrary boundary conditions at these points. By comparison with the
Newton–Euler method, commonly used to model multibody system dynamics, the forward
and backward recursions are directly solved by the feedback connections between the inputs
and the outputs of the TITOP models of each body. Such a dynamic model is today restricted
to the linear behavior of the multibody system and to loads lumped at some particular points
 J. Chebbi et al.

Fig. 13 ωc (i), i = 1, 2, 3, versus crank angle θ1

Fig. 14 First three modal shapes for crank angle θ1 = 0

Fig. 15 First three modal shapes for crank angle θ1 = 180 deg

of the structure. The methodological utility of this approach has been highlighted on the
flexible four-bar mechanism.
Further works need to be performed to extend this approach to aeronautical engineering
and, more particularly, to take into account distributed loads along each body of the system
and nonlinear effects, which could appear when the rate of variation of the configuration
Linear dynamics of flexible multibody systems

Fig. 16 Inversion of the ith

channel of G

parameters is determinant. In this current state, this approach is fully applicable to space
applications (large flexible orbital structures), but the proposed extensions are still required
to apply it, for instance, to the helicopter rotor dynamic model.

Acknowledgements This research is supported, in part, by ONERA (The French Aerospace Lab), CNES
(The French Space Agency), ESA (European Space Agency), Thales Alenia Space and Polytechnique Mon-
tréal. We would like to thank more particularly Christelle Cumer, Thomas Loquen, Christelle Pittet, Finn
Ankersen, Luca Massotti, Catherine le Peuvédic, Chiara Toglia, David-Alexandre Saussié, and Hari Murali.

Appendix 1: Matrix and linear system operations

In this section, we present basic operations on matrices and linear dynamic systems and
finally the inversion of one (or more) input–output channel(s) in a linear system.
Let us consider a square (the same number of inputs and outputs) linear system G(s)
(s stands for the Laplace variable) with order n and m channels (i.e., m inputs and m out-
puts). A state-space realization of the system G(s) is a set of four matrices An×n , Bn×m ,
Cm×n , and Dm×m such that
 
−1
AB
G(s) = D + C(sIn − A) B, also noted: G(s) ≡ . (26)
CD

Submatrices and subsystems M ATLAB nomenclature is used to define the submatrix

M(I, J) (resp. subsystem GI,J (s)) restricted to the rows (resp. outputs) and the columns
(resp. inputs) ordered in vectors of indices I and J.

Single-channel inversion The system corresponding to the inversion of the ith channel of
the system G(s) (i ∈ [1, m]) is denoted G−1i (s) and can be characterized by the following
state-space realization (Eq. (27)).
Let J be the vector of indices form 1 to m without i: J = [1, . . . , i − 1, i + 1, . . . , m].
Assume that D(i, i) = 0. Then let us denote fi = D(i,i)
1
and define
⎡ ⎤
A − fi B(:, i)C(i, :) [B(:, J) − fi B(:, i)D(i, J) fi B(:, i)]
G−1i (s) ≡ ⎣ C(J, :) − fi D(J, i)C(i, :)

D(J, J) − fi D(J, i)D(i, J) fi D(J, i)
⎦
.
−fi C(i, :) −fi D(i, J) fi

In G−1i (s), the ith inverted channel appears on the last channel; it is thus required to reorder
the channels using the vector of indices K = [1, . . . , i − 1, m, i + 1, . . . , m − 1], and then
−1
G−1i (s) = GK,Ki (s). (27)

Let u and y be the input and output vectors of G. This inversion can be represented by the
block diagram depicted in Fig. 16.
 J. Chebbi et al.

Multichannel inversion Let I be the vector (with q components) of indices corresponding

to the channels to be inverted. The successive inversion of the q channels in G(s) is denoted
 −1 −1I(2) ··· −1I(q)
G−1I (s) = G I(1) (s). (28)

Appendix 2: TITOP model of the E ULER –B ERNOULLI beam under

various boundary conditions
[3,4] −1
Case 1: free–free beam (Fig. 17) The model Gf,pol (s) is used to set to zero its inputs
y z y z
Fc , Tc , FP , and TP . The augmented model with the two rigid modes is depicted in Fig. 17
(right).

Fig. 17 The free–free beam (left) and its block-diagram model (right)

Table 3 The first two natural  

EIz EIz
frequencies ωi of the free–free i ωi,ref ( ) ωi ( ) ωi (%)
−1[3,4] ρSl 4 ρSl 4
beam model Gf,pol (s) and the
relative errors with the reference 1 22.373 22.564 0.85
values ωi,ref
2 61.673 63.537 3.0

[1,2] −1
Case 2: clamped–clamped beam (Fig. 18) The model Gf,pol (s) is used to set to zero
its inputs v̈C , α̈C , v̈P , and α̈P . There are no rigid modes in this case. The internal state
constraints

ṽ(l) = xf (2) = 0, ṽ  (l) = xf (3) = 0 ∀t ⇒ ẋf (2) = ẋf (3) = ẋf (6) = ẋf (7) = 0

[1,2] −1
can be used to truncate the state-space representation of Gf,pol (s) to a fourth-order model.
The frequencies of the two flexible modes are displayed in Table 4.
Fig. 18 The clamped–clamped
beam (left) and its block-diagram
model (right)

Table 4 The first two natural  

EIz EIz
frequencies ωi of the i ωi,ref ( ) ωi ( ) ωi (%)
ρSl 4 ρSl 4
clamped–clamped beam model
−1 [1,2]
Gf,pol (s) and the relative 1 22.373 22.45 0.34
errors with the reference values
2 61.673 62.929 2.0
ωi,ref
Linear dynamics of flexible multibody systems

4 −1
Case 3: pinned–free beam (Fig. 19) The model Gf,pol (s) is used to set to zero its inputs
y z z
FC , TC , v̈P , and TP . The augmented model with the rigid mode is depicted in Fig. 19 (right).

Fig. 19 The pinned–free beam (left) and its block-diagram model (right)

Table 5 The first two natural  

EIz EIz
frequencies ωi of the pinned–free i ωi,ref ( ) ωi ( ) ωi (%)
−14 ρSl 4 ρSl 4
beam model Gf,pol (s) and the
relative errors with the reference 1 15.418 15.445 0.17
values ωi,ref
2 49.965 51.206 2.5

−1
[1,2,4]
Case 4: pinned–clamped beam (see Fig. 20) The model Gf,pol (s) is used to set to
z
zero its inputs v̈C , α̈C , v̈P , and TP . There are no rigid modes in this case. The internal state
constraints

ṽ(l) − l ṽ  (l) = xf (2) − lxf (3) = 0 ∀t ⇒ ẋf (2) − l ẋf (3) = ẋf (6) − l ẋf (7) = 0

[1,2,4] −1
can be used to reduce the state-space representation of Gf,pol (s) to a sixth-order model.

Fig. 20 The pinned–clamped beam (right) and its block-diagram model (left)

Table 6 The first two natural  

EIz EIz
frequencies ωi of the i ωi,ref ( ) ωi ( ) ωi (%)
ρSl 4 ρSl 4
pinned–clamped beam model
−1 [1,2,4]
Gf,pol (s) and the relative 1 15.418 15.433 0.1
errors with the reference values
2 49.965 50.724 1.5
ωi,ref

[1,4] −1
Case 5: pinned–pinned beam (Fig. 21) The model Gf,pol (s) is used to set to 0 its inputs:
z z
v̈C , Tc , v̈P , and TP . There are no rigid modes in this case.
 J. Chebbi et al.

Fig. 21 The pinned–pinned beam (left) and its block-diagram model (right)

Table 7 The first two natural  

EIz EIz
frequencies ωi of the i ωi,ref ( ) ωi ( ) ωi (%)
ρSl 4 ρSl 4
pinned–pinned beam model
−1 [1,4]
Gf,pol (s) and the relative 1 9.8696 9.8725 0.03
errors with the reference values
2 39.478 39.646 0.4
ωi,ref

3 −1
Case 6: sliding–free beam (Fig. 22) The model Gf,pol (s) is used to set to zero its inputs
y z y
Fc , Tc , FP , and α̈P . The augmented model with the rigid mode is depicted in Fig. 22 (right).

Fig. 22 The sliding–free beam (left) and its block-diagram model (right)

Table 8 The first two natural  

EIz EIz
frequencies ωi of the sliding–free i ωi,ref ( ) ωi ( ) ωi (%)
−13 ρSl 4 ρSl 4
beam model Gf,pol (s) and the
relative errors with the reference 1 5.5932 5.5934 0.003
values ωi,ref
2 30.226 30.734 1.7

[1,2,3] −1
Case 7: sliding–clamped beam (see Fig. 23) The model Gf,pol (s) is used to set to zero
y
its inputs v̈C , α̈C , FP , and α̈P . There are no rigid modes in this case. The internal state
constraints
ṽ  (l) = xf (3) = 0 ∀t ⇒ ẋf (3) = ẋf (7) = 0
[1,2,3] −1
can be used to reduce the state-space representation of Gf,pol (s) to a sixth-order model.

Fig. 23 The sliding–clamped beam (right) and its block-diagram model (left)
Linear dynamics of flexible multibody systems

Table 9 The first two natural  

EIz EIz
frequencies ωi of the i ωi,ref ( ) ωi ( ) ωi (%)
ρSl 4 ρSl 4
sliding–clamped beam model
−1 [1,2,3]
Gf,pol (s) and the relative 1 5.5932 5.5933 0.002
errors with the reference values
2 30.226 30.561 1.1
ωi,ref

[1,3] −1
Case 8: sliding–pinned beam (see Fig. 24) The model Gf,pol (s) is used to set to zero
z y
its inputs v̈C , TC , FP , and α̈P . There are no rigid modes in this case.

Fig. 24 The sliding–pinned beam (right) and its block-diagram model (left)

Table 10 The first two natural  

EIz EIz
frequencies ωi of the i ωi,ref ( ) ωi ( ) ωi (%)
ρSl 4 ρSl 4
sliding–pinned beam model
−1 [1,3]
Gf,pol (s) and the relative 1 2.4674 2.4674 0.0
errors with the reference values
2 22.207 22.272 0.3
ωi,ref

[2,3] −1
Case 9: sliding–sliding beam (Fig. 25) The model Gf,pol (s) is used to set to zero its
y y
inputs Fc , α̈c , FP , and α̈P . The augmented model with the rigid mode is depicted in Fig. 25
(right).

Fig. 25 The sliding–sliding beam (left) and its block-diagram model (right)

Table 11 The first two natural  

EIz EIz
frequencies ωi of the i ωi,ref ( ) ωi ( ) ωi (%)
ρSl 4 ρSl 4
sliding–sliding beam model
−1 [2,3]
Gf,pol (s) and the relative 1 9.8696 9.8697 0.0008
errors with the reference values
2 39.478 40.988 3.8
ωi,ref

References
1. Alazard, D., Perez, J.A., Cumer, C., Loquen, T.: Two-input two-output port model for mechanical
systems. In: AIAA SciTech, American Institute of Aeronautics and Astronautics (2015). doi:10.2514/
6.2015-1778
2. Bishop, R., Johnson, D.: The Mechanics of Vibration. Cambridge University Press, Cambridge (1979)
 J. Chebbi et al.

3. Craig, R.R., Chang, C.J.: Substructure coupling for dynamic analysis and testing. NASA contractor
report 2781. National Aeronautics and Space Administration (1977)
4. De Luca, A., Siciliano, B.: Closed-form dynamic model of planar multilink lightweight robots. IEEE
Trans. Syst. Man Cybern. 21(4), 826–839 (1991). doi:10.1109/21.108300
5. Dokainish, M.A.: A new approach for plate vibrations: combination of transfer matrix and finite-element
technique. J. Eng. Ind. 94(2), 526–530 (1972). doi:10.1115/1.3428185
6. Euler–Bernoulli beam theory. https://en.wikipedia.org/wiki/Euler-Bernoulli_beam_theory. Accessed:
2015-09-30
7. Girard, A., Roy, N.: Dynamics of Structure in Industry. Lavoisier, Paris (2003)
8. Guy, N., Alazard, D., Cumer, C., Charbonnel, C.: Dynamic modeling and analysis of spacecraft with
variable tilt of flexible appendages. Journal of Dynamic Systems Measurement and Control 136(2)
(2014). http://oatao.univ-toulouse.fr/11016/
9. Kitis, L., Lindenberg, R.: Natural frequencies and mode shapes of flexible mechanisms by a transfer
matrix method. Finite Elem. Anal. Des. 6(4), 267–285 (1990). doi:10.1016/0168-874X(90)90020-F
10. Krauss, R.W., Book, W.J.: Transfer matrix modeling of systems with noncollocated feedback. J. Dyn.
Syst. Meas. Control 132(6), 061301 (2010). doi:10.1115/1.4002476
11. Leckie, F., Pestel, E.: Transfer-matrix fundamentals. Int. J. Mech. Sci. 2, 137–167 (1960)
12. Mohan, A., Saha, S.: A recursive, numerically stable, and efficient simulation algorithm for serial robots
with flexible links. Multibody Syst. Dyn. 21(1), 1–35 (2009). doi:10.1007/s11044-008-9122-6
13. Mucino, V.H., Pavelic, V.: An exact condensation procedure for chain-like structures using a finite
element-transfer matrix approach. J. Mech. Des. 103(2), 295–303 (1981). doi:10.1115/1.3254907
14. Murali, H., Alazard, D., Massotti, L., Ankersen, F., Toglia, C.: Mechanical-attitude controller co-design
of large flexible space structures. In: Bordeneuve-Guibé, J., Drouin, A., Roos, C. (eds.) Advances
in Aerospace Guidance, Navigation and Control, pp. 659–678. Springer, Berlin (2015). doi:10.1007/
978-3-319-17518-8_38
15. Pascal, M.: Dynamics analysis of a system of hinge-connected flexible bodies. Celest. Mech. 41, 253–
274 (1988)
16. Perez, J.A., Alazard, D., Loquen, T., Cumer, C., Pittet, C.: Linear dynamic modeling of spacecraft with
open-chain assembly of flexible bodies for ACS/structure co-design. In: Bordeneuve-Guibé, J., Drouin,
A., Roos, C. (eds.) Advances in Aerospace Guidance, Navigation and Control, pp. 639–658. Springer,
Berlin (2015). doi:10.1007/978-3-319-17518-8_37
17. Perez, J.A., Pittet, C., ALazard, D., Loquen, T., Cumer, C.: A flexible appendage model for
use in integrated control/structure spacecraft design. IFAC-PapersOnLine 48(9), 275–280 (2015).
doi:10.1016/j.ifacol.2015.08.096
18. Perez, J.A., Alazard, D., Loquen, T., Pittet, C., Cumer, C.: Flexible multibody system linear modeling
for control using component modes synthesis and double-port approach. J. Dyn. Syst. Meas. Control
(2016). doi:10.1115/1.4034149
19. Rong, B., Rui, X., Wang, G.: Modified finite element transfer matrix method for eigenvalue problem of
flexible structures. J. Appl. Mech. 78(2), 021016 (2010). doi:10.1115/1.4002578
20. Rui, X., Wang, G., Lu, Y., Yun, L.: Transfer matrix method for linear multibody system. Multibody Syst.
Dyn. 19(3), 179–207 (2008). doi:10.1007/s11044-007-9092-0
21. Schiehlen, W.: Multibody system dynamics: roots and perspectives. Multibody Syst. Dyn. 1(2), 149–188
(1997). doi:10.1023/A:1009745432698
22. Shabana, A.: Flexible multibody dynamics: review of past and recent developments. Multibody Syst.
Dyn. 1(2), 189–222 (1997). doi:10.1023/A:1009773505418
23. Shigley, J., Uicker, J.: Theory of Machines and Mechanisms. McGraw-Hill Series in Mechanical Engi-
neering. McGraw-Hill, New York (1980). https://books.google.com.et/books?id=hkhSAAAAMAAJ
24. Simeon, B.: On Lagrange multipliers in flexible multibody dynamics. Comput. Methods Appl. Mech.
Eng. 195(50–51), 6993–7005 (2006). Multibody Dynamics Analysis. doi:10.1016/j.cma.2005.04.015
25. Sitti, M.: Piezoelectrically actuated four-bar mechanism with two flexible links for micromechani-
cal flying insect thorax. IEEE/ASME Trans. Mechatron. 8(1), 26–36 (2003). doi:10.1109/TMECH.
2003.809126
26. Sylla, M., Asseke, B.: Dynamics of a rotating flexible and symmetric spacecraft using impedance ma-
trix in terms of the flexible appendages cantilever modes. Multibody Syst. Dyn. 19, 345–364 (2008).
doi:10.1007/s11044-007-9102-2
27. Tan, T., Yousuff, A., Bahar, L., Konstantinidis, M.: A modified finite element-transfer matrix for control
design of space structures. Comput. Struct. 36(1), 47–55 (1990). doi:10.1016/0045-7949(90)90173-Y
28. Theodore, R.J., Ghosal, A.: Comparison of the assumed modes and finite element models for flexi-
ble multilink manipulators. Int. J. Robot. Res. 14(2), 91–111 (1995). http://dblp.uni-trier.de/db/journals/
ijrr/ijrr14.html#TheodoreG95
29. Turcic, D.A., Midha, A.: Dynamic analysis of elastic mechanism systems. Part I: applications. J. Dyn.
Syst. Meas. Control 106(4), 249–254 (1984). doi:10.1115/1.3140681