Multibody Syst Dyn (2019) 46:127–146

https://doi.org/10.1007/s11044-019-09671-7

ALE beam using reference dynamics

Kilian Grundl1 · Thorsten Schindler1 ·

Heinz Ulbrich1 · Daniel J. Rixen1

Received: 23 July 2017 / Accepted: 19 February 2019 / Published online: 8 March 2019
© Springer Nature B.V. 2019

Abstract We present a small strain beam model based on the Arbitrary Lagrangian Eu-
lerian setting for use in multibody dynamics. The key contribution of the present paper is
to provide a formulation with large flexible reference motion and small overlaid deflec-
tions. We point out that the reference motion is described by actual degrees of freedom of
the model. Therefore, we use a vector of generalized positions and an Eulerian coordinate,
which itself is a degree of freedom and in which the flow of the beam material through an
arbitrary volume is represented. The additional displacements describe small fluctuations
around the reference motion. With this idea it is easy to separate the motion of belt drives,
cable and rope ways or strings. In particular, the overlaid deflections are described for ef-
ficient numeric computation and may be analyzed in an easy way for vibrational behavior.
The guiding reference motion is arbitrary, i.e., the transmission ratios are degrees of free-
dom and may change dynamically affecting also the fluctuations. Contacts with dry friction
are foreseen and represented in the present model. It is validated and proven to be efficient
in comparison with classic co-rotational and absolute nodal coordinate formulations in our
application. The simulation of pushbelt continuously variable transmissions is taken as a
high-dimensional industrial example.

Keywords ALE · Eulerian · Lagrangian · Beam · Contacts · Friction · CVT

B K. Grundl
[email protected]
T. Schindler
[email protected]
H. Ulbrich
[email protected]
D.J. Rixen
[email protected]
1 Institute of Applied Mechanics, Technische Universität München, Garching bei München,
Germany
128 K. Grundl et al.

1 Introduction

Beams are defined as flexible structures with one dimension much larger than the other two
dimensions. The respective theory provides fast models for the dynamic simulation in com-
parison with the direct application of general finite element (FE) models. The decoupling
into linear cross-section evaluations and nonlinear center line kinematics is usually based on
a small strain assumption [4, 15].
The application of beam models is still crucial within multibody systems, which is proven
by recent publications mainly about geometrically exact descriptions [36] dating back to the
pioneering work of Simo [34, 35]. An overview of flexible multibody systems is given in the
paper of Wasfy [40] or the one of Shabana [33]. Both classify beam formulations concerning
specific criteria. Wasfy distinguishes the floating frame of reference formulation [31, 32], co-
rotated formulations [6, 9], and inertial formulations [3] as mathematical descriptions of the
continuum equations. Shabana’s view is approximation based. He considers the incremen-
tal FE method, the finite segment method, i.e., lumped mass superelements, large rotation
vector formulations, and absolute nodal coordinate formulations (ANCF) [10, 16, 31, 32].
Geometrically exact beams belong to the large rotation vector formulations [19, 20, 39].
A comparison between geometrically exact beams and ANCF beams is given in [25].
Nonlinear elasticity in general is summarized in Antman’s book [2]. In particular, geo-
metrically nonlinear problems are characterized by nonlinear strain–displacement relation-
ships related to large displacement gradients found when the body undergoes large rotations
or large deformations. The right mathematical formulation and approximation has to be
picked. For beams, some are mentioned in the previous paragraph.
The equations of motion (EOMs) of multibody systems are nonlinear in general even for
linear elastic material because of the large overall motion. The choice of how to model a
specific example application is crucial and depends not only on strains and displacements
in the system. External loads can cause effects which can directly be included in the model,
cf. centrifugal stiffening phenomena in helicopter blades. The properties of the cross section
highly influence the behavior of the beam especially for laminated composite materials and
are often calculated in a pre-processing step [4]. In the case of contacts, locality of the
deformation has to be preserved [41, 43].
In this paper, we present a description of beams specifically suited for belt drives [26,
29, 30, 42]. We are interested in an efficient way to model the geometrically nonlinear and
transient motion in an overall small strain setting. We describe the reference motion and
add overlaid small deflection, which can be interpreted as a linearization about the changing
reference state. The EOMs for a one-dimensional closed continuum are derived using an
Eulerian view. The tangential movement of a mass particle is projected to a fixed position
in space to focus on the vibrations and the transient and nonlinear behavior of the refer-
ence curve, e.g. due to changing transmission ratios. As we model the frictional interactions
with arbitrary bodies [1, 23], the Lagrangian view following the mass particles is necessary
for the derivation of the overall equations. Using the Arbitrary Lagrangian Eulerian (ALE)
transformation, we meet all requirements. The EOMs are derived with the Lagrange II for-
malism, which may be stated in the general language of Irschik and Holl [18], but they are
derived from scratch in a simple setting also in the present paper. In general, the sliding
beam problem is quite popular. Behdinan [5] describes two different formulations starting
from Hamilton’s principle. Vetyukov [37, 38] gives an overview with belt and plate examples
based on a fixed reference kinematics and prescribed guiding velocity, whereby we also refer
to the literature therein. Also in [12] and [11], the reference kinematics of belt drive is fixed.
A well-known application of the Eulerian view is the Spaghetti problem of VuQuoc [39]. We
ALE beam using reference dynamics 129

refer in particular to the ALE-technique used by Pechstein and Gerstmayr in combination

with ANCF elements for a planar belt model with a fixed reference shape [22] as it motivated
our work. However, in contrast, we propose a model for belt drives with a non-fixed refer-
ence shape that dynamically undergoes large changes such as the variation of pulley radius
in a continuously variable transmission (CVT). These changes are modeled with an extra
set of (reference) degrees of freedom (DOFs), in particular the transformation parameter be-
tween Eulerian and Lagrangian coordinate space. Our model uses planar reference motion
but includes spatial, overlaid deformation. We verify the general behavior of the model and
show inherent properties using numerical tests.
The model is difficult to place in the classification of Wasfy. The reference motion is
an inertial description, whereas the overlaid deformation shows characteristics of a floating
frame or co-rotational approach. In this sense, it is a specific model to deal with the high
demands in belt drive systems. In particular, the new development is motivated by reducing
the computation time for a large pushbelt CVT simulation model [26, 28, 30]. The belt speed
and the transmission ratio define the DOFs of the reference motion. We compare simulation
results and computation time for several beam elements used in the pushbelt CVT so far
(classic absolute nodal coordinate beams [10] and redundant coordinate methods [26, 30,
41, 43], i.e., co-rotational descriptions). We highlight the performance of the new model
concerning the mentioned criteria in the designated application. This paper extends the ideas
that are presented in [17, Chap. 2]. Essentially we introduce a new theory that models a
system with a reference curve, which allows us to approximate the behavior of a complex
dynamic system much more efficiently.
The outline of the paper is as follows. First, we derive the EOMs for a one-dimensional
closed continuum in the Eulerian view. Next, we deduce the position and velocity formulas
for the center line of the beam model using the idea of a reference motion. We show the
generalized forces in the case of a planar reference motion and apply the equations to an in-
dustrial example. We derive the specific reference motion for the system ‘pushbelt CVT’and
test the model numerically in academic settings. Finally, we compare the outputs in an over-
all model with other beam models to prove the applicability and efficiency of the new model.
After the summary, we list possible considerations for improvements and applications.

2 Lagrange equations for an Eulerian description of a one-dimensional

closed continuum

In the following, the transformation between the Eulerian description and the Lagrangian
description is outlined and the related Lagrange equations are written for a one-dimensional
case. Irschik and Holl deduce the equations [18] for the general case. Our derivation holds
for a closed one-dimensional continuum.

2.1 Preliminaries

Figure 1 describes the relation between the material Lagrangian coordinate x̄ and the re-
spective spatially fixed Eulerian observer location ξ along a curve with fixed length, i.e.,
assuming no stretching along the one-dimensional curve, i.e., only a homogeneous speed ṡ.
In other words, the displacement along the curve s(t) (a scalar) is assumed to depend only
on time t .
The Eulerian coordinate ξ along the continuum defines a point fixed in space. The La-
grangian coordinate x̄ along the continuum follows the particle over time and can also be
130 K. Grundl et al.

Fig. 1 Spatially fixed Eulerian

observer coordinate ξ and
material Lagrangian coordinate x̄

understood as the spatial position of the material point if the displacement would be null.
Hence, the observer position ξ is reached by x̄ with the displacement s(t), assuming a uni-
form curvilinear velocity:

x̄ = ξ − s(t) (1)

and we can write

∂ x̄ ∂ x̄ d x̄ ds
= 1, = −1, = − = −ṡ, (2)
∂ξ ∂s dt dt

where we assume that the observation location ξ does not move.

If we define a function f as an integral of an arbitrary function g(x̄, t) over time depend-
ing limits l = l(t), u = u(t) as
 u
f := g(x̄, t)d x̄ ,
l

the absolute time derivative of f is given by the Leibniz integral rule:

 u
df ∂f dl ∂f du dg
= + + d x̄
dt ∂l dt ∂u dt dt l
 u  u
dl du ∂g dx̄ ∂g
= −g(l, t) + g(u, t) + d x̄ + d x̄, (3)
dt dt l ∂ x̄ dt l ∂t

where d symbolizes the absolute derivative and ∂ the partial derivative.

2.2 Closed curve

For a closed one-dimensional structure, the integral limits as well as the inner function
evaluation change equally over time:

dl du
= and g(l, t) = g(u, t).
dt dt
Hence, the first two terms in (3) vanish. Further, the first integral of (3) is zero as well since
 u
∂g  
− ṡ d x̄ = −ṡ g(u, t) − g(l, t) = 0.
l ∂ x̄
ALE beam using reference dynamics 131

A transformation of variables to integrate along ξ with u = L − s and l = 0 − s and the

length L of the continuum yields
 L
df ∂g
= dξ. (4)
dt 0 ∂t

2.3 Energy expressions

The EOMs can be derived considering the Lagrangian L = T − V and the standard La-
grange II equation (e.g. [24])
d ∂L ∂L
− =0
dt ∂ q̇k ∂qk
where qk index the generalized DOFs of the model and q̇k their time derivatives and where
we assume all forces deriving from a potential V . The kinetic energy is defined as
 
1 1
T = v vρdΩ = ρ
T
v T vdΩ (5)
2 Ω 2 Ω
where Ω is the reference (undeformed) volume for the integration, ρ is the constant den-
sity of the undeformed material and v the absolute velocity field. Considering the one-
dimensional continuum to be a beam-like structure, we can further write
 u  L
1 1
T = ρA v vd x̄ = ρA
T
v T vdξ (6)
2 l 2 0

where we neglect the energy related to the rotation of the cross section and where A denotes
the area of the cross section. The expression for the Lagrange II equation results in
d ∂T ∂T ∂V
− = Qk and Qk =: − . (7)
dt ∂ q̇k ∂qk ∂qk
Using (4) and (6), it follows

d ∂T ρA L
∂ ∂(v T v)
= dξ. (8)
dt ∂ q̇k 2 0 ∂t ∂ q̇k
Observe that the limits of the integral do not depend on the generalized velocities q̇k . The
derivative of the kinetic energy with respect to the generalized positions follows analo-
gously:
  
∂T ρA L ∂ v T v
= dξ. (9)
∂qk 2 0 ∂qk
As the potential energy can be generally represented as an integral over a function gV , yet
to be defined, it follows
 u  L
∂V ∂gV
V := gV d x̄ with = dξ ,
l ∂q k 0 ∂qk
which gives the last term in (7).
All expressions are written using Eulerian description. For the special case of a closed
beam structure, the ALE-transformation does not yield additional terms compared to the
Lagrange setting.
132 K. Grundl et al.

3 Kinematics

The general equations of the previous section are applied to a belt drive, i.e., a closed beam in
the following. In order to clarify how the deformations in the belt will be later approximated,
we start this section by introducing the shape functions that will determine the kinematics of
the centerline of the belt. The displacement of the belt will first be represented by an overall
shape that is parametrized by DOFs governing the reference shape of the unstretched belt.
Around that reference configuration, additional displacements – assumed to remain small
– are allowed by introducing deformation shape functions. These allow the belt to have
a general displacement, to also stretch and have additional perturbation in the binormal
directions to the reference shape.

3.1 Position

Consider a belt drive with a fixed belt length L (Fig. 2). A reference curve rRef (ξ, qR ) is
described along ξ ∈ [0, L[ in an inertial frame with axes x, y and z. The current reference
curve of the elastic centerline may depend on the m DOFs qR . On each reference position,
a frame is attached holding the tangential tRef , normal nRef and binormal bRef . A tangential
drift s defines the position of a particle along the curve. The reference motion is described
with the degrees of freedom s and qR . Additional n DOFs qf are introduced which enable
a small displacement rf of the centerline around the reference position rRef (Fig. 3). The
vector qf holds the DOFs qf i which describe the local deviation at certain nodes in the three
possible spatial directions. Local trial function vectors Si (ξ ) ∈ R3 span the local deformation
field rf in-between the nodes. The position of a particle is given by


n
r = rRef + rf = rRef + Ai Si qf i = rRef + Bqf (10)
i=1

where Ai (ξ, qR ) is introduced as a general transformation matrix for the displacements Si qf i

from a local frame for DOFs qf i to the global one. Two possibilities to choose Ai are dis-
cussed in this paper:
1. The most simple option is to choose a constant Ai , e.g. as the identity matrix. It does
neither depend on ξ nor on qR . The trial functions interpolate the qf i in an inertial frame.

Fig. 2 Reference curve

ALE beam using reference dynamics 133

Fig. 3 Reference curve with

deformations

2. The most general option is to choose Ai to depend on qR and ξ . It could be defined as the
orientation matrix Ai = [tRef , nRef , bRef ] which changes over the arc with ξ . The local
DOFs can be interpreted in the direction of the local tangential, normal and binormal
direction along the whole curve, respectively.

n derive the equations for this general, second case, i.e., B(ξ, qR )qf :=
We
i=1 Ai (ξ, qR )Si qf i .

3.2 Velocity

To get the absolute velocity, the material derivative1 has to be applied2 with
dr drRef dB dqf
v= = + qf + B
dt dt dt dt
 
∂rRef ∂rRef ∂B ∂B
= ṡ + q̇R + ṡqf + q̇R qf + Bq̇f
∂ξ ∂qR ∂ξ ∂qR
which can be rearranged as
∂r ∂r
v= ṡ + q̇R + Bq̇f .
∂ξ ∂qR
Using the vectors of generalized coordinates and generalized velocities

qT = s qR qf , q̇ T = ṡ q˙R q̇f (11)

the velocity can be written as

v = Pq̇ where P(q) := r  ∂r

∂qR
B (12)

1 The absolute change in the position of a mass particle is derived based on r(x̄) with x̄ being constant (cf.
(2)).
2 The derivative of a matrix G w.r.t. to a vector d applied to a vector e is defined as

dG  dG
k
e= ei
dd ddi
i=1
where d and e have the dimension k.
134 K. Grundl et al.

is the interpolation matrix between the generalized velocities q̇ and the velocity of the mass
particles. It depends solely on the generalized positions q. Prime  defines the derivative with
respect to ξ . We want to emphasize that besides qR and qf also s is a degree of freedom of
the model.

4 Equations of motion and generalized forces

In the following, we set-up the generalized forces. The terms resulting from the kinetic en-
ergy are derived directly from the kinematical description which holds for the general spatial
case. We choose a reference motion in the xy-plane. Thus, the reference position rRef is pla-
nar and the overlaid small deformation rf introduces an arbitrary spatial motion. For the
elastic energy, a standard homogeneous material is used which is suitable for many appli-
cations. Advanced constitutive laws may be applied in a straightforward manner depending
on the application.

4.1 Kinetic energy

The gradient of the kinetic energy with respect to the generalized coordinates is given by

∂T L
∂v
= ρA vT dξ = ρAq̇ T IPTdPdq q̇ (13)
∂q 0 ∂q

where the integral

 L
∂P
IPTdPdq := PT dξ
0 ∂q

has to be evaluated in each time step.

The time derivative of the gradient of the kinetic energy with respect to the generalized
velocities is
     L
d ∂T ρA L ∂ ∂  T  ρA L ∂ ∂v ∂ T T 
= v v dξ = 2 vT dξ = ρA q̇ P P dξ
dt ∂ q̇ 2 0 ∂t ∂ q̇ 2 0 ∂t ∂ q̇ 0 ∂t

= 2ρAq̇ T IPTdPdt + q̈ T ρAIPTP (14)

where the first integral is defined as

 L
∂P
IPTdPdt := PT dξ (15)
0 ∂t

and the second term yields the symmetric mass matrix

 L
M := ρAIPTP := ρA PT Pdξ. (16)
0

The time derivative is written as a partial time derivative according to (4): this includes the
indirect dependency hidden in the generalized coordinates.
ALE beam using reference dynamics 135

4.2 Elastic energy

In this work, we assume the belt to be slender for the bending around both normal directions
so that the Kirchhoff–Bernoulli assumptions can be applied: cross sections remain plane
and orthogonal to the elastic centerline. The shear deformation is thus assumed to be null so
that the deformation energy originates only from tangential stretching and bi-axial bending
described by the local curvatures. The stretching originates only from the deformation shape
function of the centerline, whereas the curvature and associated bending are generated both
by the reference shape and the additional deformation.
The elastic energy V is defined following [19] with
 L  L
1 1
V := γ Cγ γ dξ +
T
κ T Cκ κdξ = Vγ + Vκ (17)
2 0 2 0

where γ and κ are measures for the strains related to deformations originating from transla-
tion and rotational deformation, respectively, and the matrices Cγ and Cκ contain the related
beam stiffnesses. The partial derivatives are given by
 
∂ Vγ L
∂γ ∂ Vκ L
∂κ
= γ T Cγ dξ and = κ T Cκ dξ (18)
∂q 0 ∂q ∂q 0 ∂q

with the normal strain energy Vγ and the bending strain energy Vκ . We assume the material
to be linear elastic and isotropic and that the deformations can be measured in the principal
axis. With the Young modulus E and the cross section A as well as the moments of inertia
for bending In and Ib in normal and binormal direction, respectively, the constitutive laws
follow as

EIn 0
Cγ = EA and Cκ = . (19)
0 EIb

Torsion is not incorporated in this paper. In our application the out-of-plane deformation
in the direction of the binormal b is small and therefore torsion is negligible. Furthermore,
we observed numerically unfavorable behavior due to torsion in our system [26]. The small
out-of-plane motion in binormal direction is incorporated to show the extensibility to the
general spatial case, which is successfully applied in [17].
For the normal strain energy the material strain measure is used with

γ = tT r − 1 (20)

where the local tangential t direction of the center line follows the Frenet formulas, which
also define the local3 directions of the normal and binormal.
r t r × t
t= = n × b, n= =b×t and b= = t × n. (21)
r   t   r  t  

With this definition of the tangential direction t , the Kirchhoff assumption is fulfilled. The
cross section is not able to rotate relatively to the neutral phase [20, Remark 3.4]. Using
Kirchhoff beam theory is advantageous compared for instance to a Timoschenko model

3 [t n b] refers to the local deformations and does not coincide with [t

Ref nRef bRef ].
136 K. Grundl et al.

since, when discretized by finite elements, the latter can exhibit locking unless proper re-
duced integration is used. The normal strain energy follows as
 L
∂ Vγ  ∂  T 
= EA rT t − 1 t r dξ
∂q 0 ∂qk
 L √  ∂ √
= EA rT r − 1 r  T r  dξ
0 ∂qk
 L √  1 ∂ 
= EA rT r − 1 √ rT r dξ
0 rT r ∂qk

and it is finally given by

⎡⎡ ⎤ ⎤
 L  0
∂ Vγ 1 ⎣⎣0⎦ dr 
= EA 1− √ r T
dqR
B1 ... Bn ⎦ dξ =: EAIV γ .
∂q 0 rT r 0

The bending measures are defined as

 
κn
κ := (22)
κb

where κn and κb measure the curvatures in the two planes that are spanned between the
tangential vector tRef and the normal nRef or the binormal bRef , respectively. The energy
derivatives follow as
 L  L
∂ Vκ ∂ ∂
= EIn κn κn dξ + EIb κb κb dξ =: EIn IV n + EIb IV b .
∂q 0 ∂q 0 ∂q

The single measures are discussed in the following.

For the curvature in the tRef /nRef -plane, it is crucial to couple the reference deformation
and the overlaid effects. Only due to this, the full physics of the beam can be represented.
Otherwise, the elastic forces of the reference DOFs would only affect the reference motion,
i.e., the forces would not be projected into the generalized direction of the overlaid DOFs.
With the illustrative example of a circle-like deformation-free state, a damped dynamic sim-
ulation starting from an arbitrary initial state would not result in the circular shape in the
case of no external forces if the coupling of reference and overlaid motion has not been
covered in the right way (see Sect. 5.3). To approximate the curvature in the tRef /nRef -plane,
the second derivative of the position is projected along the reference normal direction:

κn := nTRef r  with r  = rRef


+ B qf .

This approximation yields correct measures for the reference part and linearized overlaid
measures for the additional deformations. The derivatives with respect to the generalized
coordinates are
∂
κn = 0, (23a)
∂s
∂ ∂nTRef  ∂r 
κn = r + nTRef , (23b)
∂qR ∂qR ∂qR
ALE beam using reference dynamics 137

∂ ∂rf  
κn = nTRef = nTRef Ai Si + 2Ai Si + Ai Si . (23c)
∂qf i ∂qf i

Analogously, the binormal curvature measure in the tRef /bRef -plane is approximated as

κb := bRef
T
r 

where derivatives with respect to the generalized coordinates follow with

∂
κb = 0, (24a)
∂s
T
∂bRef 
∂ T ∂r
κb = r  + bRef , (24b)
∂qR ∂q ∂qR
 R
=0

∂ ∂rf   
κb = bRef
T
= bRef
T
Ai Si + 2A Si + ASi = 0 0 1 Si . (24c)
∂qf i ∂qf i

4.3 External forces

External forces from joints or contacts have to be projected into the generalized directions.
The Jacobian of translation has to be found which relates the global free directions with the
generalized directions. According to (12), it is the transformation matrix P:

∂v
JT := = P.
∂ q̇

Interaction torques have to be modeled as force pairs.

4.4 Equations of motion

Summarizing the previous findings, we gain the EOMs

Mq̈ = h (25)

where the mass matrix is defined by (16) and the right hand side vector h follows with two
parts of the kinetic energy and three parts of the potential energy.

h = hT 1 + h T 2 + h V 1 + h V 2 + h V 3 , (26a)
hT 1 = −2ρAq̇ T IPTdPdt , (26b)
hT 2 = ρAq̇ T IPTdPdq q̇ , (26c)
hV 1 = −EAIV γ , (26d)
hV 2 = −EIn IV n , (26e)
hV 3 = −EIb IV b . (26f)
138 K. Grundl et al.

Fig. 4 Pushbelt CVT. ©Bosch

Transmission Technology

Fig. 5 Kinematics of a simple

reference curve

5 Example: continuously variable transmission

We prove the applicability of the derived model within the framework of multibody simu-
lation in this section. The beam model is applied within a pushbelt CVT system (Fig. 4).
The CVT transmits torque and power between two pulleys. Its conical sheaves clamp the
pushbelt on the steel element flanks. The ratio between the clamping forces on the primary
and secondary pulley-arcs determine the transmission ratio. In-between the elements a push
force acts due to the applied torque on the pulleys. The elements guided by two rings. These
are modeled with our presented approach.
First, we define the specific reference kinematics, then we show the inherent properties
of the ring model with numerical tests and finally we compare the results in the overall CVT
system to the ones gained with different beam models.

5.1 Kinematics of the reference curve

The derived EOMs (25) are valid for any belt system with a planar reference curve. For the
application within the CVT system, we have to choose the corresponding reference curve.
We omit the index ref in this section for the sake of clearness. All kinematical values refer
to the center line of the beam model. The reference curve consists of four parts which are
depicted in Fig. 5:
– primary arc with radius rP and length bP = 2rP (π − ϕ),
– upper strand with length b = dA cos α,
– secondary arc with radius rS and with length bS = 2rS ϕ,
– lower strand with length b.
ALE beam using reference dynamics 139

The relations α := arcsin( rSd−r

A
P
) and ϕ := arccos( rPd−r
A
S
) are used. Denoting the geometric
ratio (of the reference curve radii) by ig := rP , we define the scalar ratio parameter
rS

1 − ig 1−Θ
Θ := ⇔ ig = .
1 + ig 1+Θ

This parameter is sufficient to describe the reference motion and will be used as degree of
freedom for the global shape (qR = Θ).4
For a given geometric ratio Θ, the primary and secondary radii can be deduced using the
fact that

0 = L − 2b − bP − bS . (27)

Considering all the definitions outlined earlier, (27) provides a nonlinear relationship where
dA —the constant distance between the pulley axles—and L—the total length of the ring—
are given geometrical properties of the CVT system. The changing transmission ratio as
explicit DOF allows for the analysis of the dynamic behavior, e.g. in run-up phases [17,
Chap. 5.2.2].
In a pre-processing step, the arc-radii for different Θ are computed solving the nonlin-
ear system (27). The primary radius rP is interpolated over the ratio parameter Θ with at
least C 2 -continuous splines. An explicit dependency rp (Θ) results. The further kinematic
description over the arc follows also explicitly and is C 1 -continuous in ξ . For a given inter-
polated function of the primary radius rP (Θ), the positional description for the upper part
over the arc length ξ is given by
⎧⎛ ⎞
⎪
⎪ −rS cos(ϕS )
⎪
⎪ ⎜ ⎟
⎪
⎪ ⎝ rS sin(ϕS ) ⎠ if 0 ≤ ξ ≤ ϕrS
⎪
⎪
⎛ u⎞ ⎪ ⎪
rx ⎪
⎨ 0
r u = ⎝ryu ⎠ = rS,end + ξT tT if ϕrS < ξ ≤ rS ϕ + b (28)
⎪
⎪ ⎛ ⎞
0 ⎪
⎪ dA − rP cos(ϕP )
⎪
⎪
⎪
⎪ ⎜ ⎟
⎪
⎪ ⎝ rP sin(ϕP ) ⎠ if ϕrS + b ≤ ξ ≤ L2
⎪
⎩ 0

with the following definitions:

ξ ξ
ϕS := , ξT := ξ − rS ϕ, ξP := ξ − rS ϕ + b, ϕP := ϕ + ,
rS rP
⎛ ⎞
−rS cos ϕ
rP ,start − rS,end
rS,end := ⎝ rS sin ϕ ⎠ and tT := .
0 |r P ,start − rS,end |

The lower part follows due to symmetry:

⎛ ⎞
rxu (ξ − L)
r l (ξ ) = ⎝−ryu (ξ − L)⎠ . (29)
0

4 The parameter Θ is defined in such a way that symmetric properties can be used in the numerical evaluation.
140 K. Grundl et al.

5.2 Trial functions

Finite elements, used to discretize the deformation around the reference configuration, are
distributed along the entire length of the belt. Hermite interpolation trial functions are chosen
to respect the prerequisite of C 1 -continuity. In this example we focus on planar overlaid
deformation and write
le le le 2 le 3
N1 := 12 − 34 x + 14 x 3 , N2 := − x− x + x ,
8 8 8 8
le le le 2 le 3
N3 := 12 + 34 x − 14 x 3 , N4 := + x − x − x ,
8 8 8 8
where le is the element length and x ∈ [−1; 1] the local coordinate in the FE coordinate
system. For each direction we introduce one DOF for the position 5 and one additional DOF
for its derivative.6 Four DOFs per node result. The local deformation within one finite ele-
ment is therefore affected by in total eight DOFs which are interpolated by the interpolation
matrix Se
⎡ ⎤
N1 N2 0 0 N3 N4 0 0
S =⎣ 0
e
0 N1 N2 0 0 N3 N4 ⎦ .
0 0 0 0 0 0 0 0

Si is obtained after assembly of the element shape function Se .

5.3 Numerical tests

The equations are implemented in MBSim [21, 27] and tested qualitatively in the following.
The first virtual experiment tests the dynamics of the reference kinematics only, i.e., qf =
0. The parameters follow a real ring of a pushbelt CVT such that steel material and the
geometrical properties are incorporated. The initial condition is for the DOFs Θ = 0.4, s = 0
and Θ̇ = ṡ = 0, i.e., an overdrive ratio of ig ≈ 0.4286 is set. Results are depicted in Fig. 6
for the deformation of the ring. Four different time points are given where t0 is at the initial
condition. At t1 an intermediate state is given. One can see that the ratio changes over time
and oscillates around the neutral state Θ = 0. The parameter s stays zero up to numerical
errors. Keeping in mind that the local deformations are blocked, this is the expected result.
For a general model with all DOFs, one would expect a circular shape as neutral state.
Therefore, another experiment is performed with the same belt, but adding the local defor-
mations, i.e., qf = 0. As the local deformations are able to represent rigid body motion,
the overall position is redundantly described taking into account also the reference DOFs.

Fig. 6 Snapshots of reference oscillation (qf = 0)

5 Interpolated by N and N .
1 3
6 Interpolated by N and N .
2 4
ALE beam using reference dynamics 141

Fig. 7 Snapshots of a deformable ring with 8 FEs

As a consequence, the mass matrix becomes singular. Therefore, local DOFs qf need to be
locked. We fix one node to its reference position such that it moves only due to the refer-
ence coordinates within the xy-plane. We choose the first node for all simulations, which
is at ξ = 0. The same settings as before are used. Eight FEs are used along the belt. All
generalized positions and velocities satisfy q = q̇ = 0 at t0 = 0.
Snapshots of the belt are depicted in Fig. 7 for four different times in chronological order.
One can see that at t1 the ring shows a reasonable behavior. Similar to a (rubber) band, it
starts to form a circular shape. The strands deform outwards. At t2 however, the belt exhibits
a rather unexpected shape. The first node7 can only move with the reference DOFs, i.e., Θ in
this case. To form a circular shape, the left node has to move to the center, i.e., Θ increases.
The typical shape of the CVT ring for Θ > 0 can be seen, i.e. overdrive, as Θ influences the
deformation over the whole arc. Yet, the deformation develops further yielding an oval-like
shape again at t3 , which is the expected behavior.
With this second example we want to emphasize the following: The presented modeling
strategy relies on the description of the reference curve. The model assumes an external
force distribution that yields a shape similar to the one of the reference configuration. Only
small deviations from the reference are incorporated correctly. This explains why the results
obtained between t0 and t1 in the previous examples (Fig. 7) are as physically expected and
why the shape computed at t2 is an artifact of the chosen discretization. Nevertheless the
small deformation behavior is quantitatively correct as can be seen from the shape computed
at t3 .

5.4 Application

The numerical tests show particularly that our system relies on the application of external
forces on it, which keep it close to its reference shape. In fact, the overlaid deformations are
assumed to be small and can be interpreted as a linearization about the changing reference
state. The model does not behave quantitatively correctly, when it is “far off” this state.
Thus, the main application in mind is the simulation of a pushbelt CVT [17], whereby this
section presents the results using the newly derived model and a comparison with other beam
models which are implemented in MBSim [21, 27]. In total four beam models are used:
RCM It follows the co-rotational approach and is a planar implementation of the model of
[13, 14, 41, 43]. In the simulation 32 FEs are used, which yields 160 DOFs.
ANCF It is the planar implementation of the ANCF-theory of [32]. Forty FEs yield 160
DOFs.

7 Here on the left side of the ring.

142 K. Grundl et al.

Table 1 Nine load cases defined

by the transmission ratio and the 1 2 3 4 5 6 7 8 9
primary torque
ig [−] 0.47 1.00 2.42 0.47 1.00 2.42 0.47 1.00 1.30
TP [Nm] 30 30 30 50 50 50 100 100 100

ALE0 It is the planar implementation of the present model without overlaid deformation. It
only considers the two reference DOFs s and Θ.
ALE Compared to the ALE0 model it also considers overlaid deformations with 157 DOFs
and 40 FEs.
In Sect. 3.1, we introduced two possibilities to choose the direction matrix Ai . For the appli-
cation within the CVT the simple option has proven to be most suitable, i.e., Ai is chosen to
be the constant identity matrix. This results in less evaluated derivatives, reducing thereby
the CPU time. In the general second possibility Ai would change at every time step as Θ
does not stay constant but oscillates. The interpretation of the DOFs would change in every
time step and the local DOFs would “see” the oscillations in Θ, resulting in a numerically
unfavorable behavior. As introduced before, the small overlaid deformations are planar.
We draw different general conclusions comparing the results of nine different load cases
(see Table 1). The loads are taken from measurements to ensure realistic settings. The belt
speed has minor influence and is hold constant for the primary pulley with ωP = 2000 RPM.
The transmission ratio is varied in three way, i.e. ig < 1, ig ≈ 1 and ig > 1. In the experi-
ments the torques of TP1 = 30 Nm, TP2 = 50 Nm and TP3 = 100 Nm where applied on the
primary pulley. At the primary pulley we ensure kinematically a constant distance to fix
the transmission ratio. On the secondary pulley we apply torques and forces TS∗ and FS∗ ,
which are taken from measurements of a reference belt and which ensure safe operation.
The measurements were conducted by ©Bosch Transmission Technology.
Thereby, the output with the RCM model is used as reference as it was successfully tested
in prior work [7, 8, 13, 14]. The differences to the other models are evaluated by considering
the thrust ratio iF (Fig. 8a), which is the ratio between the two clamping forces at the pulley
sheaves, as well as by analyzing a measure of efficiency η (Fig. 8b). It can be seen that all
models converge globally to the same level, i.e., a similar thrust ratio and efficiency. Yet, the
ALE0 model yields in overdrive8 very high values for the thrust ratio whereas the ANCF
model yields low values in medium9 and underdrive10 configurations. Furthermore, higher
losses compared to the other models result when using the ANCF model.
Figure 8c reveals the reason for these discrepancies. Given are the push forces acting on
one push element in longitudinal direction along one revolution for case 3. First these build
up in the primary pulley arc. Then they are rather constant in the push-strand between the
pulleys. In the secondary pulley arc they decrease again.
The ANCF model reacts very stiff. High oscillations result in the complete system as
seen in the figure. For the efficiency computation, this means that many oscillations around
a zero level (e.g. after t = 0.08 s) arise. It leads to errors within the numerical integration
of the efficiency values. The ALE0 model deviates from the other three for what concerns
spiral running11 (Fig. 8d). As the model does not support local deformations along the belt

8 Cases 1, 4 and 7.
9 Cases 2, 5 and 8.
10 Cases 3, 6 and 9.
11 The pulley sheaves deform due to the clamping forces. Thus, elements change the running radius within a
pulley arc, which is called spiral running.
ALE beam using reference dynamics 143

Fig. 8 Comparing different beam models. The results gained with the RCM model are used as reference

Fig. 9 Average CPU costs for

beam models where RCM is the
reference. The error bars show
the maximum deviations from the
average value

only a local stiffness in the contact enables small deviations from the circular arc. The pulley
sheaves deform differently, which influences the axial force distribution along the arcs and
therefore the thrust ratio.
For the nine cases, the computational times are compared in Fig. 9. Again the simulation
times of the RCM model are used as reference. The error bars indicate the maximal deviation
from the mean value. The ANCF and the ALE0 model take up only about the half of the
simulation time. The computational time for the full simulation with the ALE approach takes
only about 80% of the computation time with the reference RCM approach.
Altogether, the ALE model proves to be a good reference model. The simulation time is
reduced ensuring a good correlation with an already validated RCM model. Depending on
144 K. Grundl et al.

the application, it is easy to adjust the level of detail12 reducing the simulation times even
further.

6 Summary

In this paper, we present a beam model based on planar geometrically nonlinear reference
dynamics which is described by actual degrees of freedom of the model and we add over-
laid spatial deflections. The derivation is based on the small strain assumption and makes
use of the Arbitrary Lagrangian Eulerian transformation. Hence, we deal with both peri-
odic motion, which can be reduced due to the Eulerian view, and contact conditions, which
are represented by the Lagrangian view. The planar large reference motion may change the
shape dynamically, which is a significant extension to models from literature. The origi-
nal application of the derivation is a large spatial model for pushbelt continuously variable
transmissions. The idea—known from e.g. the floating frame of reference method—is to de-
scribe the nonlinear reference motion and additional small deviations separately. We show
that the new model represents the phenomena even better than the previous models or in a
comparable way, and that it saves computational time. Hence, the contribution of the paper
is twofold: theoretically as there is no model with changing reference shape in the literature;
practically as it allows for efficient and reliable spatial simulation of pushbelt CVTs for the
first time.

7 Further considerations

Based on the summary, we propose to answer the following questions for future work.
– Application to other belt systems
The presented approach is not restricted to the pushbelt CVT. It can be used for any belt
system with a closed one-dimensional curve and a reference kinematics.
– Separated integration
The separation of the reference curve and the overlaid deformation enables the possibil-
ity to separate these DOFs for the numerical integration. This makes sense, if e.g. the
reference DOFs change only slowly compared to the overlaid deformations and so they
need not to be updated in every time step. This reduces the simulation costs and increases
the robustness of the overall simulation model as the dynamical coupling of the DOFs is
weakened.
– Model reduction
A nonlinear model order reduction technique, e.g., the proper orthogonal decomposition
(POD), is necessary to reduce the computational effort for the CVT application in the
Lagrangian view. The coupling of the overall motion with local small deformation is the
reason why both motions have to be identified at the same time. The Eulerian view in
combination with the reference curve description offers the possibility to apply linear
model order reduction techniques only to the overlaid deformations.
– Quasi-static overlaid deformation field
The evaluation of the integrals with respect to the trial functions is time consuming and
yields the most computational cost. For an overall good approximation, the reference

12 For instance ALE0.

ALE beam using reference dynamics 145

DOFs suffice to give good results (Fig. 8). However, the spiral running in the arcs is not
represented which results in small deviations. From the overall state of the variator, it is
possible to approximate the kinematics of the ring. The small deviations could be defined
as quasi-static. Besides the spiral running, this idea can be adapted for the misalignment.
The axial deviations of the rings can be deduced from the sheave positions as is done for
the initialization in [26]. In both cases, the stiff behavior needs not be treated dynami-
cally, which saves computational costs by adding spiral running and misalignment to the
reference DOFs.

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps
and institutional affiliations.

References
1. Acary, V., Brogliato, B.: Numerical Methods for Nonsmooth Dynamical Systems: Applications in Me-
chanics and Electronics, 1st edn. Lecture Notes in Applied and Computational Mechanics, vol. 35.
Springer, Berlin (2008)
2. Antman, S.: Nonlinear Problems of Elasticity. Springer, Berlin (2005)
3. Bathe, K.J., Bolourchi, S.: Large displacement analysis of three-dimensional beam structures. Int. J.
Numer. Methods Eng. 14, 961–986 (1979)
4. Bauchau, O.: Flexible Multibody Dynamics. Springer, Berlin (2010)
5. Behdinan, K., Tabarrok, B.: Dynamics of flexible sliding beams – non-linear analysis part II: tran-
sient response. J. Sound Vib. 208(4), 541–565 (1997). https://doi.org/10.1006/jsvi.1997.1168. http://
www.sciencedirect.com/science/article/pii/S0022460X97911688
6. Belytschko, T., Schwer, L.: Large displacement, transient analysis of space frames. Int. J. Numer. Meth-
ods Eng. 11, 65–84 (1977)
7. Cebulla, T.: Spatial dynamics of pushbelt cvts: model enhancements to a non-smooth flexible multibody
system. Ph.D. thesis, Technische Universität München (2014)
8. Cebulla, T., Grundl, K., Schindler, T., Ulbrich, H., van der Velde, A., Pijpers, H.: Spatial dynamics of
pushbelt CVTs: model enhancements. In: SAE Technical Paper of SAE World Congress, Detroit, 24th–
26th April 2012 (2012)
9. Crisfield, M.A.: A consistent co-rotational formulation for non-linear, three-dimensional, beam elements.
Comput. Methods Appl. Mech. Eng. 81, 131–150 (1990)
10. von Dombrowski, S., Schwertassek, R.: Analysis of large flexible body deformation in multibody sys-
tems using absolute coordinates. In: Advances in Computational Multibody Dynamics, Lisbon, 20th–
23rd September 1999, pp. 359–378. Instituto Superior Tecnico, Lisbon (1999).
11. Dufva, K., Kerkkänen, K., Maqueda, L.G., Shabana, A.A.: Nonlinear dynamics of three-dimensional belt
drives using the finite-element method. Nonlinear Dyn. 48(4), 449–466 (2007). https://doi.org/10.1007/
s11071-006-9098-9. https://link.springer.com/article/10.1007/s11071-006-9098-9
12. Funk, K.: Simulation eindimensionaler Kontinua mit Unstetigkeiten. In: Fortschrittberichte VDI: Reihe
18, Mechanik, Bruchmechanik, vol. 294. VDI Verlag, Düsseldorf (2004)
13. Geier, T.: Dynamics of Push Belt CVTs, Fortschrittberichte VDI: Reihe 12, Verkehrstechnik,
Fahrzeugtechnik, vol. 654. VDI Verlag, Düsseldorf (2007)
14. Geier, T., Förg, M., Zander, R., Ulbrich, H., Pfeiffer, F., Brandsma, A., van der Velde, A.: Simulation of
a push belt CVT considering uni- and bilateral constraints. J. Appl. Math. Mech. 86, 795–806 (2006)
15. Geradin, M., Cardona, A.: Flexible Multibody Dynamics: A Finite Element Approach. Wiley, New York
(2001)
16. Gerstmayr, J., Sugiyama, H., Mikkola, A.: An overview on the developments of the absolute nodal
coordinate formulation. In: Proceedings of 2nd Joint International Conference on Multibody System
Dynamics, Stuttgart, 29th May–1st June 2012 (2012)
17. Grundl, K.: Validation of a pushbelt variator. Ph.D. thesis, Technische Universität München (2015)
18. Irschik, H., Holl, H.: The equations of Lagrange written for a non-material volume. Acta Mech. 153,
231–248 (2002)
19. Jelenic, G., Saje, M.: A kinematically exact space finite strain beam model – finite element formulation
by generalized virtual work principle. Comput. Methods Appl. Mech. Eng. 120(1), 131–161 (1995).
http://www.sciencedirect.com/science/article/pii/004578259400056S
146 K. Grundl et al.

20. Lang, H., Linn, J., Arnold, M.: Multi-body dynamics simulation of geometrically exact Cosserat rods.
Multibody Syst. Dyn. 25(3), 285–312 (2011)
21. MBSim – multi-body simulation software. GNU Lesser General Public License.
https://github.com/mbsim-env/
22. Pechstein, A., Gerstmayr, J.: A Lagrange–Eulerian formulation of an axially moving beam based
on the absolute nodal coordinate formulation. Multibody Syst. Dyn. 30(3), 343–358 (2013). http://
link.springer.com/article/10.1007/s11044-013-9350-2
23. Pfeiffer, F.: Mechanical System Dynamics, corr. 2nd printing edn. Lecture Notes in Applied and Com-
putational Mechanics, vol. 40. Springer, Berlin (2008)
24. Pfeiffer, F., Schindler, T.: Introduction to Dynamics. Springer, Berlin (2015)
25. Romero, I.: A comparison of finite elements for nonlinear beams: the absolute nodal coordinate and
geometrically exact formulations. Multibody Syst. Dyn. 20, 51–68 (2008)
26. Schindler, T.: Spatial Dynamics of Pushbelt CVTs. In: Fortschritt-Berichte VDI: Reihe 12,
Verkehrstechnik, Fahrzeugtechnik, vol. 730. VDI Verlag, Düsseldorf (2010). http://mediatum.ub.tum.de/
node?id=981870
27. Schindler, T., Förg, M., Friedrich, M., Schneider, M., Esefeld, B., Huber, R., Zander, R., Ulbrich, H.:
Analysing dynamical phenomenons: introduction to MBSim. In: Proceedings of 1st Joint International
Conference on Multibody System Dynamics, Lappeenranta, 25th–27th May 2010 (2010)
28. Schindler, T., Friedrich, M., Ulbrich, H.: Computing time reduction possibilities in multibody dynam-
ics. In: Blajer, W., Arczewski, K., Fraczek, J., Wojtyra, M. (eds.) Multibody Dynamics: Computational
Methods and Applications, Computational Methods in Applied Sciences, vol. 23, pp. 239–259. Springer,
Dordrecht (2011)
29. Schindler, T., Geier, T., Ulbrich, H., Pfeiffer, F., van der Velde, A., Brandsma, A.: Dynamics of pushbelt
CVTs. In: Umschlingungsgetriebe: Ketten und Riemen – Konstruktion, Simulation und Anwendung
von Komponenten und Systemen, Tagung Berlin, 21. und 22. Juni 2007, VDI-Berichte. VDI Verlag,
Düsseldorf (2007).
30. Schindler, T., Ulbrich, H., Pfeiffer, F., van der Velde, A., Brandsma, A.: Spatial simulation of pushbelt
CVTs with timestepping schemes. Appl. Numer. Math. 62(10), 1515–1530 (2012)
31. Schwertassek, R., Wallrapp, O.: Dynamik Flexibler Mehrkörpersysteme. Vieweg, Wiesbaden (1999)
32. Shabana, A.: Dynamics of Multibody Systems, 3rd edn. Cambridge University Press, New York (2005)
33. Shabana, A.A.: Flexible multibody dynamics: Review of past and recent developments. Multibody Syst.
Dyn. 1(2), 189–222 (1997). https://doi.org/10.1023/A:1009773505418
34. Simo, J.: A finite strain beam formulation. The three-dimensional dynamic problem. Part I. Comput.
Methods Appl. Mech. Eng. 49(1), 55–70 (1985). https://doi.org/10.1016/0045-7825(85)90050-7
35. Simo, J., Vu-Quoc, L.: A three-dimensional finite-strain rod model. Part II: computational aspects. Com-
put. Methods Appl. Mech. Eng. 58(1), 79–116 (1986). https://doi.org/10.1016/0045-7825(86)90079-4
36. Sonneville, V., Cardona, A., Brüls, O.: Geometrically exact beam finite element formulated
on the special Euclidean group. Comput. Methods Appl. Mech. Eng. 268, 451–474 (2014).
https://doi.org/10.1016/j.cma.2013.10.008
37. Vetyukov, Y.: Mechanics of axially moving structures at mixed Eulerian-Lagrangian description. In:
Analysis and Modelling of Advanced Structures and Smart Systems, Advanced Structured Materials.
Springer, Singapore (2018). https://doi.org/10.1007/978-981-10-6895-9_13. https://link.springer.com/
chapter/10.1007/978-981-10-6895-9_13
38. Vetyukov, Y.: Non-material finite element modelling of large vibrations of axially moving strings
and beams. J. Sound Vib. 414, 299–317 (2018). https://doi.org/10.1016/j.jsv.2017.11.010. http://
linkinghub.elsevier.com/retrieve/pii/S0022460X17307824
39. Vu-Quoc, L., Li, S.: Dynamics of sliding geometrically-exact beams: large angle maneuver and
parametric resonance. Comput. Methods Appl. Mech. Eng. 120(1–2), 65–118 (1995). https://doi.org/
10.1016/0045-7825(94)00051-N
40. Wasfy, T., Noor, A.: Computational strategies for flexible multibody systems. Appl. Mech. Rev. 56,
553–613 (2003)
41. Zander, R.: Flexible multi-body systems with set-valued force laws. Fortschritt-Berichte VDI: Reihe
20, Rechnerunterstützte Verfahren, vol. 420. VDI Verlag, Düsseldorf (2009). http://mediatum2.
ub.tum.de/node?id=654788
42. Zander, R., Rettig, F., Schindler, T.: Concepts for the simulation of belt drives – industrial and aca-
demic approaches. In: Proceedings of 1st Joint International Conference on Multibody System Dynam-
ics, Lappeenranta, 25th–27th May 2010 (2010)
43. Zander, R., Ulbrich, H.: Reference-free mixed FE-MBS approach for beam structures with constraints.
Nonlinear Dyn. 46, 349–361 (2006)