Beruflich Dokumente
Kultur Dokumente
net/publication/261596652
CITATIONS READS
11 976
2 authors:
Some of the authors of this publication are also working on these related projects:
Robotically Fabricated Lightweight Timber Shell, Bundesgartenschau Heilbronn, 2019 View project
All content following this page was uploaded by Julian Lienhard on 19 June 2014.
Reprinted from
INTERNATIONAL JOURNAL OF
SPACE STRUCTURES
Volume 28 · Number 3 & 4 · 2013
was predominantly developed through physical models, slender single span beam in its post buckling state.
scale factors for self-weight were derived to correctly After some first considerations by James Bernoulli in
simulate dead load deformation of the scaled model [3]. 1691, Euler describes the shape of these curves in ‘Des
Other expressions of active-bending in building curves Elastics’ in 1744, offering analytical solutions
structures are yet to be analysed for their scaling for post buckled curves of a pin supported beam [4]
behaviour. The research presented in this paper is based (Fig. 1)
on the experiences from various research structures in The residual stress in such an elastically deformed
the range of 2 to 10 m span. In all of these structures the beam can be determined with the Euler-Bernoulli law
scalar jump from a physical structural model to a which states that the bending moment My is
medium scale structure was successfully undertaken. proportional to the change in curvature as shown in
The question analysed in this paper concerns the scalar equation (1) [5]. With the section modulus Wy and the
jump from a medium scale to a large scale structure and consideration that the width b of a cross-section has no
is thereby aiming to fathom the scaling limits of various influence on the maximum bending stress we can write
forms of bending-active structures. the residual bending stress as an expression of the
Scaling in the most general sense is concerned with cross-sectional height h, the Modulus of Elasticity E
power-law relationships between two or more and the Curvature 1/r (2).
variables of a system. Investigating the scaling of
building structures the variables concerned are 1 My
= (1)
deformation and stability on the one side, load and r E ⋅ Iy
mechanical properties on the other. If the relation of
these variables is independent of the system’s
dimensions we consider the system to be self-similar. E ⋅ Iy E ⋅h
σM = = (2)
Some of the more common effects to consider for r ⋅ Wy 2⋅r
the scaling of building structures are:
- Dimension effect: cubic increase of mass with
scale (s ⋅ Li )2 s ⋅ fi L 2 + 4 ⋅ fi 2
r= + = s⋅ i (3)
- Load effect: quadratic increase of surface area s ⋅ 8 ⋅ fi 2 8 ⋅ fi
leads to quadratic increase of surface load
- Size effect of material: probability of material
In (2) we can see that both curvature (1/r) and cross
defects increases with size, whereas the influence
sectional height h have a linear influence on the
of material defects increases for small size
residual stress caused by active bending. The moment
specimens.
of inertia Iy is therefore limited by a given minimal
- Height effect: exponential growth of wind-speed
curvature in the system and the permissible bending
with height combined with quadratic growth of
stress of the chosen material. The radius of curvature
wind-load with speed.
can be expressed as a function of the span Li and the
- Dynamic effects: Wind induced vibration etc.
rise fi. In (3) this relation is given with a scale factor s.
Since the investigations in this paper are aimed to
Simplification of the equation shows that s can be
be of general nature we are only considering the
excluded as a linear factor, thus linear up-scaling of a
change in mass and load. The effects of change in
structure allows for a linear up-scaling of the cross
material properties, wind load and dynamic behaviour
with scale are very individual to each project and
therefore not taken into account in the further Li
investigations. Their influence on scaling however
fi
will play a role on the construction of some large scale EI r h
bending-active structures.
2. DIMENSIONAL ANALYSIS OF
f
ELASTICA
In order to gain a principal understanding of the Wx
scalability of a simple bending-active system some L
initial studies are made on the elastica arch. The
elastica describes the elastic deformation curve of a Figure 1. Development elastica curves.
sectional height while keeping the residual stress procedure it must be noticed that there is no unique set
constant. Assuming constant material properties this of Pi-terms that can be derived for a given problem. Pi-
leads to the overriding question; whether the influence terms may differ in type depending on the choice of a
of the span L on the deflection Uz can be compensated repeating variable that eliminates dimension,
by the moment of inertia Iy, if the scaling of the cross- additionally transformations of Pi-terms are possible.
sectional height is limited to be linear for keeping the Here the results of the system deflection under a linear
residual stress σM constant. load were derived with the step wise procedure using L
as the repeating variable. The resultant dimensionless
2.1. Deriving the dimensions using the Pi-terms are:
Buckingham Pi Theorem:
Dimensional analysis enables a convenient
Uz L4 q
investigation of physical behaviour by combining the π1 = ; π2 = ; π3 = z
variables of a system into dimensionless groups (Pi- L Iy E⋅L
(4)
terms). The Buckingham Pi-Theorem states that the U L4 q
relations in any physical system can be described by a → z = φ , z
group of n–rd Pi-terms, in which n is the number of
L Iy E ⋅ L
variables and rd the number of basic dimensions
therein (rank of the dimensional matrix) [6]. In An analysis of the given Pi-terms now enables to
mechanics the basic dimensions are mass, length and clarify at which power each variable is effected by
time. In the following considerations on static scaling. In the common terminology of dimensional
structural behaviour, force is chosen as a basic analysis one differentiates between the scale of the
dimension without further reduction into its model and that of the prototype. Investigating the
constituent components, for better comparison to scaling of Iy we can compare the Term π2 in a model
known engineering equations. Based on the Pi-terms a state m to an s times larger prototype state p.
functional equation can be derived which shows a
reduced form of the relevant variables, however does L4 (s ⋅ L ) 4
it not give information about the nature of the solution. I = I → I y , p = s 4 ⋅ I y ,m
(5)
y m y p
The exact form of the functional relationship has to be
empirically obtained by a set of experiments in which
the Pi-terms are systematically varied. Analytical If all dimensions of the system are scaled by the
analysis the individual Pi-terms often is sufficient same factor s we can show that the similitude in (5) is
enough to describe the change of system behaviour satisfied by a linear up scaling of the cross sectional
with scale, without knowing the complete solution of dimensions, here shown for a flat section:
the functional equation.
Investigating the deflection for a given elastica bm ⋅ hm 3 s ⋅ bm ⋅ (s ⋅ hm )3
curve of span L stiffness EI and the line load qz and s4 ⋅ Im = s4 =
12 12 (6)
excluding the influence of mass and residual axial force
= I p
→ b p ⋅ h p = s ⋅ bm ⋅ hm
we may derive the following functional equation:
Uz = f (L, E, Iy, qz)
5 Variables: Uz, L, E, Iy, qz With (5) we can show that the deflection of an
5-2 = 3 Pi-Terms elastic arch is self-similar, if the effects of self-weight
2 Dimensions: [mm], [N] and residual force are excluded.
The dimensional Matrix is: Similar to (5) the Term π3 can be used to analyse the
relation of the loading condition in model and
Uz L E Iy qz prototype state, showing that the line load qz scale
[N ] 0 0 1 0 1 linearly for constant material properties between
model and prototype state:
[mm] 1 1 −2 4 −1
qz qz
=
→ qz , p
The dimensionless Pi-terms may be derived using E⋅L m s ⋅ E ⋅ L ) p (7)
various procedures, some of which are explained and
discussed in detail by Barr [7]. Independent of the = s L ⋅ s E ⋅ qz ,m
(a) (b)
s.q s.L 250
Uz Straight
f/L: 0.05
s.f f/L: 0.15
200 f/L: 0.20
f/L: 0.30
Uz (mm) (with qz = s × qz )
f/L: 0.50
f/L: 0.60
150 f/L: 0.75
50
0
0 2 4 6 8 10
Scale s (−)
Figure 2. (a) Study of elastica curves with constant span and varying rise at different scales.
(b) deflection curve for different scales.
Fig. 2a show the various arches in the initial and behaviour at higher loads and final snap through
comparative deformed position, clearly showing buckling. For each scale the system was calculated in
maximum stiffness at ratios of f/L = 0.15 to 0.35. As three different scenarios.
this ratio tends towards f/L = 1.0 the arches show First (indicated by the continuous lines) including
increasingly low stiffness and nonlinear behaviour; dead load and axial force, showing a difference in load
compare dashed line in Fig. 2a. The graphs in Fig. 2b factor ∆λ = 5 between scale s = 1 and s = 8 at the point
show the displacement curves in the linear range of snap through buckling.
without dead load or residual axial force. Their In a second scenario (dashed line) the residual stress
linearity prove the similitude derived above for all and thereby axial force N was disabled. This leads to a
variations of the elastica curve, in which the rise to shift of both graphs by load factors ∆λ = 5 for s = 1 and
span ratio is a constant in correspondence with the ∆λ = 6 for s = 8 higher at the point of failure. The
incline of the graph. difference between scales is reduced by one factor
In Fig.3a an elastica curve with f/L = 0.15 is from ∆λ = 5 to ∆λ = 4.
investigated further by plotting the load deformation Finally the dotted line shows the elastica in a
diagram with corresponding compression stress for a calculation without dead load and residual stress. Here
step wise increasing scaled line load. In order to the two load deflection curves almost perfectly match
exclude findings that are limited to symmetric and snap through buckling occurs at the same load
systems, the line load was applied asymmetrically. The factor, ∆λ = 0. This clearly supports the hypothesis
investigations were compared at two different scales made in section 2 that the bending-active elastica arch
with s = 1 and s = 8 in Fig. 3b showing the nonlinear is self-similar if dead load and axial force are omitted.
(a) (b)
25
0.90
−2.00
1.00
0.80
∆λ = 0
20
0.00
∆λ = 5
Snap through
10
−1.78
−1.79
−18.2
−18.7
−35.8
−35.0
−50.3
−49.6
−63.2
−60.0
−59.6
Nonlinear
−10.00 −8.00 −6.00 −4.00 −2.00 0.00
5
s.L
Beam elements, compression stress Mpa,
nonlinear loadcase, load factor 5 λ .s.qz 0.8. λ .s.qz
Linear
Uz
0
0 20 40 60 80 100
−6.37
−18.5
−18.9
Displacement Uz /s (mm)
−4.01
−10.5
−42.5
−44.9
−67.4
−65.1
−79.2
−75.2
−6.55
−16.4
−27.0
−35.1
−39.8
−45.8
−47.6
−51.2
−53.0
−51.8
2.00
Figure 3. (a) Asymmetrical line load on elastica curve showing compression stress and deformation to the point of snap through
buckling. (b) Load deflection curve of elastica curve with 15% f/L ratio at two scales; showing linear, nonlinear and snap
through failure range.
In general the graph shows how the curves are the elastic bending behaviour of 6.5 mm birch
very close in the linear range and the influence of plywood strips. The strips were robotically
dead load grows with size (∆λ = 1 for s = 1 and ∆λ manufactured as planar elements, and subsequently
= 5 for s = 8). connected to coupled arch systems. Radial
arrangement and interconnection of the self-
4. SCALING OF THREE CASE STUDY
equilibrating arch system lead to the final torus
STRUCTURES
shaped design of the pavilion with a total diameter of
With the scaling investigations on three successfully
10.5 m, maximum free span of 4 m and a height of
built case study structures, the above drawn
3.5 m. The form-finding and structural behaviour of
conclusions are verified. A jump in scale from a
this structure was explained in [8].
prototypical structure, in the size of an exhibition
The FE form finding routine developed for this
pavilion, to a large building structure is investigated.
structure was used for the scaling investigations
The choice of material for these bending-active
presented here (Fig 4). The system was studied on
structures is limited by availability of materials
three different scales in which each model was
offering high strength with low bending stiffness such
separately form-found. Thus all effects from residual
as plywood and GFRP. In addition the scalar
stress are included in the investigations.
investigation are made in a relatively small range of s
In Fig. 5a the comparison of the deformation under
= 0.5 to 5. Therefore material properties are assumed
vertical load shows significant stiffening effects in the
constant. For all systems several wind and combined
system where the residual stresses from the erecting
snow load cases were investigated; the graphs in this
process are included. This may be explained due to the
paper are a selection that highlight best the scale
fact that the curved arch sections are inducing tensile
dependent behaviour discussed here. In contrast to the
stress into the straight sections (compare normal forces
investigations on the elastica, loads are areal and
in the upper graph). This tension pre-stress increases
therefore constant across scale. Height effects of wind
the stiffness the coupled arch structure hence the
are not considered because of the relatively small
geometrical stiffness of the entire system, from
scalar range investigated.
nonlinear structural analysis generally known as
The case study structures have been discussed in-
stress-stiffening effect.
depth in previous publications, therefore only short
Fig. 5b shows the load deflection curves of the
introductions are given with the references for further
system on three different scales. The applied load qw is
information.
a wind pressure surface load and therefore remains
4.1. Research pavilion ICD/ITKE 2010 constant for all scalar investigations. The general
At the end of July 2010 the Institute of Computer behaviour is similar to that shown of the elastica curve
based Design (ICD) and the Institute of building in Fig 3; showing snap through buckling at decreasing
structures and structural design (ITKE) at the load factors with increasing scale. With increasing
University Stuttgart realised a temporary research scale the system also shows distinct nonlinear
pavilion made of plywood. The design of the behaviour. At scale s = 4 snap through already occurs
pavilion was the result of a student workshop which under dead load which leads to a completely nonlinear
focused on the integration of physical experiments behaviour with increasing load. In contrast to the
and computational design tools to develop bending- elastica curve in Fig. 3b were the residual compression
active structures. The structure is entirely based on stress is destabilising to the system, the coupled arches
10.5 m
S1:50 S1:1
(a)
qz
−34.4
−34.4
−34.2
−33.6
−31.9
−32.3
−33 .8
−34.5
−3 34.4
−3 1.5
.7
−3
2
−
9.
.4
4
16.1
−29 9
33.7
−3 .4
33.7
34.2
−3 1.4
34.0
34.7
2.1
−30
35.3
34.4
−3
−3 9
.5
34.8
36.0
0.
2.
35.3
36.7.5
1 −3
37
−31 .6 36 .2
−31
.5
35 .3 4 .9 −32
.4 3 .0
3 .9 35 .1 −30.5
30.2 3 5.6
−28 5 35 .3 −28.1
3 .4 35
−27 .5 5.2 41.2
−26 .8 44.7 −29
.6
.6 37.6
.8
−24.8 40.8 48.2 −28
51.8 −26.9
−22.7 44.2
55.5 −24.9
−20.4 47.6
59.2 − 21.7
.8
−17 .0 51.1 62.5 −18
1 5 .1
− 1. 4 54.8
− 1 57.9 Residual normal force (kN/m)
108.4
104.2
105.4
96.2
94.1
82.3
89.1
65.7
59.0
49.6
43.4
35.6
31.0
6.62 19.8
3.87 17.3
2.77
5.80
6.46
5.84 Displacement Uz (mm)
syst. excl.residual stress
90.5 89.1
87.5
82.5
80.0
69.0
72.1
36.5
59.8
13.9
45.9
33.6
36.1
27.0
11.0 22.0
5.32 12.2
4.77
16.1
5.25
5.52
1.01
1.72
Displacement Uz (mm)
5.69
5.27
(b)
s = 1 N: yes s = 1 N: no s.L
λ .qw
s = 2 N: yes s = 2 N: no
Uz = λ.0.1 kN/m2
s = 4 N: yes
6
Loadfactor λ (−)
∆λ = 4
∆λ = 5
Figure 5. (a) Displacement Uz of a coupled arch section from the pavilion, showing 17% increase in stiffness for the system
including residual stress induced by active bending. (b) Load deflection curve of the ICD/ITKE 2010 research pavilion at 3
different scales for a given wind pressure load case.
in the pavilion store residual tension stress in their s = 1 and s = 2 where the system was calculated
straight sections. The resultant increase in geometrical without its residual stress from the form-finding of the
stiffness can be seen in the dashed line graphs of scale erection process (N: no). Here snap through buckling
occurs at an earlier point if the geometry only is taken and pre-stressed membrane was similar at all
into consideration (∆λ = 0.5 for s = 1 and ∆λ = 1 for investigated scales. This resulted in under-proportional
s = 8). However, the advantageous influence of scaling of membrane pre-stress.
residual tension stress to the system is reduced over In this system the bending-active beams are
time due to enhanced relaxation of timber. shortcutting tension forces between the edge cables
These investigations show that the reduced size and the low point. Independent of the load direction
structure is already close to the limit size of this the bending-active beam elements are therefore always
system. In this case increasing the scale would under compression. These compression forces
necessitate a change in material and topology. increase over proportionally with scale and generally
destabilise the system. In this combination of bending,
4.2. Marrakech umbrella shear and axial load the linear beam elements cannot
The project is based on a student workshop that compensate the quadratic growth in membrane surface
developed shading solutions for an outdoor plaza as seen in Fig. 7. Here the arm-end deformations Uz
space at an architecture school in Marrakech Morocco. are shown for both wind suction (left) and wind
The design proposal of a funnel shaped membrane pressure (right), showing highly nonlinear load
roof was further developed by the authors with the aim deflection curves and significant differences in scale.
of minimising anchoring forces to the surrounding Scaling of a membrane structure with bending-
buildings. The introduction of a bending-active active beam elements therefore necessitates a change
supporting structure for the free edges of the in topology i.e. increasing the number of beam
membrane proved to be a very efficient solution. After elements when scaling up in overall size. However,
the successful test setup in Stuttgart June 2011, the beyond spans of approx. 30 m other effects may
structure was mounted by students from Stuttgart and become decisive, as the membrane loses its stabilising
Morocco in spring 2012 [9]. The structure features 6 effect on the beam elements due to flexibility
elastically bent glass fibre rods with a length of increasing with size. Finally, even the availability of
approx. 7.5 m. The rods are pushing out three the necessary large size GFRP cross sections may
additional corner points on both free edges of the become limiting factors to the scale of membrane
structure. The funnel shaped membrane has a span of structures with integrated bending-active support
approx. 11 m ∞ 11 m and an eaves height of 5.5 m systems. Similar conclusions were drawn by
resulting in a membrane surface of approx. 110 m2. Adriaenssens [10].
This structure highlights the possibility of using active
bending in hybrid form-and bending-active structures 4.3. Flectofin®, an elastic kinetic shell
where the system is stabilised by the elastic beams system
which, in turn, are restrained by the membrane surface. The focus of the Flectofin® project is the optimisation
For the scaling of this system, special attention had of deployable systems in architecture using bio-
to be paid to the change of membrane pre-stress. Since inspired solutions. The Flectofin® is based on a
the pre-stress of the membrane has a direct influence sophisticated pollination mechanism, which performs
on the curvature of the bending-active beam elements, a reversible deformation when an external mechanical
it was altered in such a way that the geometry resulting force is applied. The system is comprised of a stiff
from combined form-finding of bending-active beams beam element with attached flexible shell elements.
11 m
s=1 s=1
s = 2.5 s = 2.5
s=5 8 s=5
s.L s.L
λ .qw = λ λ .qw = λ.0.1 kN/m2
λ .0.1 kN/m2
Uz Uz
−1000 −800 −600 −400 −200 0 200 400 600 800 1000
Displacement Uz / s (mm) Displacement Uz / s (mm)
Uniaxial bending of the beam causes the shell influence to scaling. Additionally, the bending-active
elements to deviate into out-of-plane bending due to surface element makes up the entire load bearing
their low lateral stiffness, see Fig. 8 and for further surface, the quadratic growth in surface is therefore
information [11]. The prototype lamellas are 2.2 m compensated by the quadratic growth of the load
long and 0.25 m deep, they are made out of GFRP and bearing shell element itself.
produced in a vacuum bagging process. This system suggests that the above made
In Fig. 9 the load deflection curves of a Flectofin® assumptions of dead load and stability only, being the
lamella are plotted in three different scales for both limiting factors for scalability of bending active
wind suction and wind pressure. Besides minor structures may the generalized. This is additionally
nonlinearities the system shows almost perfect supported by a recently built large scale adaptive
similitude. This behaviour may be explained by a façade: For the theme pavilion of EXPO Yeosu, Korea
combination of reasons: 2012, an elastic kinetic façade shading system inspired
In its deformed state the Flectofin® lamellas by the Flectofin® was built. Some early feasibility
predominantly hold residual tensile stress, leading to studies using the Flectofin® already suggested that
considerable stress stiffening effects. Destabilising scalability to a size range of 14 m high lamellas was
effects due to residual compression forces are possible. Covering a total length of 140 m and varying
therefore eluded. Positioned vertically on a façade the in height between 3 and 14 m it was designed to
dead load direction acts in its upright position which withstand the very high wind loads at the Korean coast
offers maximal elastic stiffness, dimension effects [12], proving the applicability of this system for large
considering dead load may therefore be of minor scale structures.
0° 90°
2.2
m
S1:50 S1:1
S0.5:1 S1:1 S5:1
Physical model FE-simulation Real construction FE-simulation
s.L
s.L
Load factor λ (−)
6 λ.qw
λ.qw dl dl
Figure 9. Load deflection curve of the Flectofin® Lamella at 3 different scales, for a given wind- suction and pressure load case.