Rotation Formulations for Dynamic Relaxation with Application in

3D Framed Structures with Large Displacements and Rotations

Jian-Min Li
PhD student
Institute of Building Structures and
Structural Design (ITKE),
University Stuttgart
Stuttgar, Germany

Jan Knippers
Prof. Dr.-Ing.
Institute of Building Structure and
Structural Design (ITKE),
University Stuttgart
Stuttgart, Germany

[email protected]

[email protected]

Summary
Dynamic relaxation (DR) has been widely used as a powerful method for structural analysis and
form-finding. Compared with the direct stiffness method, one advantage of DR is that there is no
need to compose a global stiffness matrix; for DR, every node of the structure can be seen as a
single particle constrained by forces generated by the interactive mechanism of the structural
members. The equations of motion for particles are much easier to consider than the direct stiffness
approach, especially when structures undergo large displacements and rotations.
However, most applications of DR are restricted in the field of cable, membrane, hanging chain
structures. The common feature behind these structures is that only three degree of freedom (3DOF) nodes are used. And for 3-DOF nodes, no bending moment but translational forces can be
applied.
In this paper, we propose the rotation formulations for DR. The description of motion of one
particle is then separated into two parts, the translational and the rotational. Therefore, additional
rotation degree of freedom is introduced into the computation system and we can use DR to
simulate structure systems composed of 6-DOF nodes.
We further show how we apply the 6-DOF DR, with the beam mechanism, to the simulation of
framed structures. Validity is examined through examples of structures which undergo large
displacements and rotations.
Keywords: six degree of freedom node; Dynamic Relaxation; large rotations; large displacements;
3D beam; framed structures

1.

Related works

Few attempts have been made to extend DR to have rotational degree of freedom. Wakefield uses
rotation displacements to describe 3D rotations in DR [1]. In his derivation, Johnson and Brottons
method [2] was used to get the sway angles.
Williams proposed to solve global translational displacements by solving the nodal displacement
individually by assuming the objective node is not restrained and the adjacent nodes are fixed [3]
and [4]. The velocity terms are not used in his derivation. Instead, he uses displacements to play the
role of velocities.
Besides, there were other attempts to simulate bending within the 3-DOF framework. However, by
the natural limitation of the 3-DOF node, these approaches can not deal with torsion effect and
unsymmetrical section properties [5].

2.

Dynamic relaxation

When a structure undergoes large displacements and rotations, the Newton-Raphson method (NR),
which is commonly used as the iteration solver for nonlinear FEM, might diverge due to the

transient zero tangent stiffness. This is a common situation when using FEM to deal with large
displacements and loose constraints. Being an explicit solver, DR can be more robust to find
equilibrium state, because it solves the nodal increments individually without composing a global
stiffness matrix.
However, NR is proved to be more efficient to deal with structure systems with less elements. But,
it will lose its advantage when it goes to a system with large element number. Because the
computation time needed for DR is proportional to N4/3 (N is the number of elements in the system),
and for NR it is proportional to N7/3 [6].
2.1. Translation formulations for DR
Dynamic relaxation uses the concept of particles and constrained forces to simulate structure
systems. It builds the inter relationship between the displacement, velocity and acceleration of one
particle in a central difference form as in eq. (1) and (2), where n denotes the time point t n t .
These equations can be further arranged as in eq. (3) and (4), in which the unknown displacement
and velocity of the current time step are expressed as functions of known values of the previous
time step. Calculating repeatedly, the motion of the system can be thereby simulated in time history.

qn1 qn1

vn
2 t

v v

an1 n n2
2 t

qn1 qn1 2tvn

vn vn2 2tan1

(1)
(2)
(3)
(4)

Using Newtons third law as in eq. (5), force can be introduced into the system. We substitute the
acceleration term in eq. (4) with the force term to get eq. (6).

(5)
Fn 1 man 1

( 6)
vn vn 2 2tm 1 Fn 1
2.2
Rotation formulations for DR
Rotation matrix is used to describe the orientation of one particle/node. The matrix is composed of
three unit vectors pointing the local x, y and z directions as in eq. (7). Similarly, here we can build
relationship between the current and previous orientations, Rn+1 and Rn-1, as we did for
displacements [7]. Their relationship is built
up through eq. (8) where R is a transformation matrix
which is defined by a rotation vectors e as in eq. (9). The mathematical meaning of the
transformation matrix
is that after rotating the previous local coordinates Rn-1 with degree

around vector e , Rn-1 will meet the current local coordinates Rn+1 [8].

x1
R x2
x3

y1
y2
y3

z1
z 2
z3

(7 )

This rotation vector can be approximated by the product of the mean angular velocity n and the
time interval as in eq. (10), in which we have applied the core meaning of central difference. Eq. (8)
can be further expressed as eq. (11).

Rn 1 R ( ) Rn 1
(8)
c e x2 (1 c )
e x e y (1 c ) e z s e z e x (1 c ) e y s c cos

R ( ) e x e y (1 c ) e z s
c e y2 (1 c )
e y e z (1 c ) e x s , s sin

e z e x (1 c ) e y s e y e z (1 c ) e x s
c e z2 (1 c) e

(9)

 tn1

dt n 2t

(10)

Rn 1 R(2tn ) Rn 1

(11)

t n1

Similar to the relationship between the velocity and acceleration in eq. (4), we can also express
the current angular velocity in terms of the previous angular velocity and angular acceleration as in
eq. (12). In contrast to eq. (5), which describes translational mechanism, eq. (13) describes
rotational mechanism which can be derived by applying virtual power to a rotating body [9], where
T denotes torque acting on the particle and I denotes the moment of inertia of the particle.
Since, in this research, we care about only the final equilibrium state not the dynamic process, we
can assign the mass and moment of inertia for our own convenience. Therefore, we assign a scalar
mass in eq. (6) and an isotropic rotational inertia here. The later assignment further reduces eq. (13)
to eq. (14). Finally, we replace the angular acceleration in eq. (12) with the torque term to get eq.
(15). Table 1 is presented here as a concise comparison between the translational and rotational
formulations of DR.

n n 2 2t n 1
(12)

Tn 1 I n 1 n 1 ( I n 1 )
(13)

(14)
Tn 1 I n 1

(15)
n n 2 2t I 1Tn 1
Table 1: translation and rotation formulations of DR

3.

Translational part

Rotational part

qn 1 qn 1 2tvn

vn vn 2 2t m 1 Fn 1

Rn1 R(2tn ) Rn1

n n 2 2 t I 1Tn 1

Beam mechanism

Our beam model is visualized as two particles set apart with distance as the beam length and having
concentrated mass and given rotational inertia. The particles will shift and rotate freely if there is
no any constrained force or torque. Once we add the constrained forces to the system, the particles
movements will be coupled together. A simplest example is two balls oscillating on the two ends of
a spring.
In the following sub sections, we introduce how we calculate the internal forces and torques acting
on the particles. These forces and torques are caused by the relative movement between the particles.
First, we calculate the relative displacements and included angles and then use them to calculate the
corresponding forces and torques.
Noteworthily, using particles and constrained forces to simulate the coupled movements might be
not so common in structural analysis but it has been a basic method to study molecule dynamics.
3.1
Orientations of beam ends
In sec.2, we give the description of motion of nodes/particles including its orientation definition.
However, in order to calculate the beam deformation, there is need to define additional orientations
for beam ends, as in eq. (16). Where Ra,n-1 and Ra,n+1 are the previous and current orientations of the
beam-end- a and R is the transformation matrix of the node where beam-end- a is situated, as
defined in eq. (11).
(16)
Ra ,n 1 R Ra ,n 1

Herein we have implied the rigid joint assumption with using the same transformation matrix to
update the nodal and beam ends orientations such that the beam end is able to rotate with the node
simultaneously. The initial beam ends orientations of one beam, Ra,0 and Rb,0 , are equal to the
beams initial orientation. Correspondingly, the initial nodal orientation is an identity matrix.
3.2
Included angles
The included angles, as in eq. (17)-(20), are used to compute bending moments and shears in a
deformed beam. Their corresponding definitions are shown in the fig.1. These formulations are
suitable for small relative rotations between the two ends of one beam. One should bear in mind
that the orientations used here are the orientations of the ends of the beam instead of the nodes
/particles orientations.

i , y u x z i , i a , b
(17)

i , z u x yi , i a, b
(18)


(19)
b , x a , x ( y a z b yb z a ) / 2

(d d )

(20)
u x b a
db d a

yb

ya

xa

za

xb

a,z

xa

yb

Fig.1-b: definitions of included angles i, y

zb

xb

zb

xb

xa

za

a, y
za

3.3

ya

zb b, z

Fig.1-a: definitions of included angles i, z

ya

yb

db

da

b, y

Fig.1-c: definitions of u x

Internal forces

3.3.1 Axial forces and bending moments

The formulations for axial force, eq. (21), and bending moments of internal forces, eq. (22)-(24) are
the same as the terms in the stiffness matrix of a finite beam element. They can be derived by
applying virtual work method to a continuous deformable body [10].
(x x )
L
1) b, x a , x ,
fb,x f a,x
L0
L
GJ
(b, x a ,x ) ,
a,x
b, x a, x
L
2 EI y
2 EI y
a, y
(2 a , y b, y ) ,
b, y
( a , y 2 b, y )
L
L
2 EI z
2 EI z
a,z
(2 a , z b, z ) ,
a,z
( a ,z 2 b, z )
L
L

f a , x AE(

(21)
(22)
(23)
(24)

3.3.2 Coordinate transformation

The internal forces and torques derived above are defined in local beam ends coordinates. Before
further computing resultant torques, we have to be transformed them to the global coordinates with
eq. (25).

Ta Ra a
(25)
3.3.3 Shear forces
The missing terms of shear forces in sec. 3.3.1 can be derived by using the equation of zero
resultant force and torque of one beam. Solving eq. (26) and (27), we can get shears as in eq. (28).
The reason for not using the shears defined in the stiffness matrix is we use local beam ends
coordinates to transform bending moments instead of using local beam coordinates, which is the
case for FEM. Therefore, we need to derive shear forces in a way to ensure zero resultant force and
torque of one element. Finally, the applied translational force for one node, from one beam end, is
the sum of the axial force and shear force as in eq. (29).

Va Vb 0
( 26

T a T b ( d a d b ) V a 0

( 27

(T a T b ) ( d b d a )
Va

2
db da
( 28

Fa R a f a V a
( 29

4.

Implementation

4.1
Algorithm
a. Input initial geometry and mechanical properties
b. Compute length extension and included angles as defined in eq. (17)-(20).
c. Compute beam axial force and torque defined in eq. (21)-(24) and transform them to global
coordinates. Use eq. (28) and (29) to compute the beam shear force and nodal translational force.
d. Compute resultant force and torque for each particle.
e. Check if the resultant force and torque are small than given values. If it is true for every particle,
go to step g. Else, go to step f.
f. Use eq. (3), (6), (8) and (15) to update the position, orientation and velocity of each node, use eq.
(16) to update each beam ends orientation, and back to step b.
g. Terminate the calculation and output the result.
4.2
Mass and moment of inertia
Observing eq. (6) and (15), one can find that less the
Intital configuration
mass and moment of inertia, larger the increments
of translational and angular velocities. And
generally, faster the velocities, less time needed to
attain equilibrium. However, too small values of
mass and inertia will lead the calculation divergent.
Final configuration
A simple model as in fig.2 is introduced here to
Fig.2: simple model for determining
determine the minimized mass and rotational inertia
at the same time. The structure, composed of two
mass and moment of inertia
elements, is initially bended and will attain its
straight and unstressed geometry finally. By try and error, we can find the minimised mass and the
minimised rotational inertia, which can then be applied to a larger system with more elements and

)
)
)
)

more complicated geometries while having analogous section properties and element lengths as the
simple model.
4.3

Damping

4.2.1 Velocity damping

Damping is used to reduce systems kinetic energy until a final static equilibrium is attained. The
velocity damping defined here is by multiplying a factor smaller than 1.0, to the translational
velocities or angular velocities in every iteration loop.
4.2.2 Mixture of kinetic damping and velocity damping
Kinetic damping is proved to be a very efficient way to attain the equilibrium. Setting systems
velocities to zero once the system passes through the kinetic energy peak is how it works [4].
Here we mix two damping methods together. For translational velocities, we use kinetic damping
and for angular velocities, we use velocity damping. Thereby, translational and angular velocities
are treated separately. We set the translational velocities to zero once the system passes through the
peak of translational kinetic energy, while we run out the angular kinetic energy gradually by
applying velocity damping.

5.

Numerical examples

5.1
Straight beam subjected to end bending
This example was seen in Ibrahimbegovics work [11]. A straight cantilevered beam composed of
10 elements is shown in fig.3 with material properties: E (elastic modulus) = 79577, (Poissons
ratio) = 0. The beam is subjected to a pure end torque Tz = 2.5. The exact solution of this
problem is Tz l / EI , where l denotes total length which is 10. That is when Tz =2.5, is equal
to /4. The result from the presented method is compared with Ibrahimbegovics, the analytic and
the result from Sofistik, a FEM commercial software, as in table 2.
( 9 . 0050 , 3 . 7311 , 0 )

y
0.4

Final state

0.4
Tz

(10 , 0 , 0 )

Intital state
z

Fig.3: definition of the straight beam and its deformation

Table 2: displacement of straight beam under end moment

present
Ibrahimbegovic
FEM commercial
Analytic

5.2

x
-0.9950
-0.9962
-0.9945
-0.9968

y
3.7311
3.7290
3.7302
3.7292

0
0
0
0

Curved beam subjected to end force

The curved beam proposed by Bathe and Bolourchi as in fig. 4 has a 45-degree bent geometry with
radius 100 in. The material properties are: E = 107 psi, = 0. The beam composed of 8 elements
with equal length lies in the x-y plane and is subjected to a concentrated load in z-direction.
Comparison is made between Bathes [12], Crisfields [13] and Sofistiks results, as in table 3.

(15.604, 46.873, 53.703)

1"

1"

Final state

P 600 lb

Intitalf state

450

(29.289, 70.710, 0)

R 100"

Fig.4: definition of the curved beam and its deformation

Table 3: displacement of Bathes curved beam
present
Bathe and Bolourchi
Crisfield
FEM commercial

6.

x
13.685
13.4
13.63
13.559

y
-23.838
-23.5
-23.87
-23.552

z
53.703
53.4
53.71
53.294

Conclusion and further work

We have proposed the rotation formulations for dynamic relaxation, with which we are able to
consider structure systems with 6-DOF per node. That is the bending and torsion effects can be
considered in the DR framework. We further expressed how we integrate the rotation and
translation formulations of DR with the beam mechanism. Validity has been examined through
examples considering structures undergo large displacements and rotations.
To build a more robust algorithm, we will improve the way we optimise the mass and rotational
inertia such that fine tuning is no more necessary.

References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]

Wakefield, D.S., Dynamic Relaxation Analysis of Pretensioned Networks Supported by

Compression Arches, PhD thesis,1980
Johnson D., and Brotton, D.M., A Finite Deflection Analzsis for Space Structuraes, Proc.
Int. Conf. on Space Structures, Univ. of Surrez, 1966.
William, C.J.K., Private Report in the Context of Partial Supervision and Work on Grid
Shells, Bath University,1999
Adriaenssens, S.M.L., Stressed Spline Sturctures, PhD thesis, 2000
Adriaenssens, S.M.L., Barnes, M.R. and Williams, C.J.K., A new Analytic and Numerical
basis for the Form-finding and Analysis of Spline and Grid-shell, Computing Developments
in Civil and Structural Engineering, 1999, pp. 83-90
Sauve, R.G.., Advances in Dynamic Relaxation Techniques for Nonlinear Finite Element
Analysis, Journal of Pressure Vessel Technology, Vol. 117, 1995, pp. 170-176
Krzsl, P., Explicit Newmark/Verlet algorithm for Time Integration of the Rotational
Dynamics of Rigid Bodies, International Journal for Numerical Methods in Engineering,
vol. 62, 2005, pp. 2154-2177
"Rotation matrix", http://en.wikipedia.org/wiki/Rotation_matrix

[9]
[10]
[11]
[12]
[13]

Wittenburg, J., Dynamics of Multibody Systems, Springer, 2008

McGuire, W., Matrix structural analysis, Wiley, 2000
Ibrahimbegovi, A., Computational Aspects of Vector-like Parameterization of Threedimensional Finite Rotations, International Journal for Numerical Methods in Engineering,
vol. 38, 1995, pp. 3653-3673
Bathe, K. J., "Large displacement analysis of three-dimensional beam structures",
International Journal for Numerical Methods in Engineering, vol. 14, 1979, pp. 961-986
Crisfield, M.A., "A consistent co-rotational formulation for non-linear three-dimensional
beam-elements",Computer Methods in Applied Mechanics and Engineering, Vol 81,1990,
pp.131150