Documente Academic
Documente Profesional
Documente Cultură
··································································································································································································································
Abstract
This study shows a computer-oriented design tool conceptually extracting a so-called steel-framed “Diagrid”, which has been well-
known as a specific and effective structural unit cell for mainly free-formed building frames. The goal of this study is to provide a
smart design optimization task to measure the effects of variation of optimally geometrical configurations like shape and topology of
the diagrid to structural engineers. Numerical applications verify how optimal solutions by using the present topology optimization
design tool can evaluate promising possibility to conceptually generate practical diagrid unit cells as a trial of building related issues.
Keywords: steel-framed, diagrid, material layouts, topology, shape, optimization, structural design
··································································································································································································································
*Assistant Professor, Dept. of Architectural Engineering, Sejong University, Seoul 143-747, Korea (Corresponding Author, E-mail: dongkyulee@
sejong.ac.kr)
**Professor, Dept. of Architectural Engineering, Sejong University, Seoul 143-747, Korea (E-mail: jhlee@sejong.ac.kr)
***Leader, Building Structures & Materials Research Dept., Steel Structures Research Division, Research Institute of Industrial Science & Technology,
Incheon 180-1, Korea (E-mail: jin800f13t@rist.re.kr)
****Professor, Structural Analysis and Steel Structures Institute, Hamburg University of Technology, Hamburg D-21073, Germany (E-mail: starossek@
tuhh.de)
− 549 −
Dongkyu Lee, Jaehong Lee, Jinho Kim, and Uwe Srarossek
Fig. 1. New Trend in Building Designs by Utilizing Structural Forms in Nature: (a) Tree in Nature, (b) Buildings with Twisted, Tilted, and
Tapered Shapes (The Left, Middle, and Right Figures are Turning Torso in Sweden, Puerta de Europa in Spain, and Swiss Re in UK.)
Fig. 2. Practical Examples of Construction Plans of Diagrid Systems in Korea: (a) Cyclon Tower, (b) Lotte Super Tower, (c) Future-X
Tower
Fig. 3. Practical Examples of Diagrid Systems in the World: (a) Swiss Re Building, (b) Hearst Tower, (c) CCTV Headquarters
constructing large and tall buildings with mainly steel by using 2. Integration between a Good Design Optimiza-
straight, curved, or horizontal rings and triangulated beams tion Tool and Diagrid Unit Cell for High-Rise
which together make up a complete structural system. In general Buildings
buildings, columns are utilized for providing vertical load carrying
capacity, diagonals or braces offers stability and resistance to A good design optimization task is one in which there are
large horizontal forces such as wind and seismic loads. However design parameters considered together and the engineer needs to
diagrid members take several advantages in addition to eliminating surely understand and measure the influences of their changes on
perimeter columns. At most diagrid system optimizes each performances, to both meet design constraints and improve
structural element. In other words the diagonals and braces try to performances.
join in the vertical load transfer, and the columns participate in the In order to achieve it, the number of design parameters and
lateral load. The two functions are combined into a diagrid and then valances among performances should be considerable enough to
diagonals, braces, and columns become finally all one system. verify the effort involved in creating a new optimization model,
In general diagrid can be made from wood, concrete, and finding or developing an integrated computational tool to utilize
steel or their composite. However steel is the most common an appropriate design optimization task, and carrying out
material of choice due to its high abilities to resist both tensile optimization works.
and compressive forces. Therefore for example, in Hearst Geometry in specific free-formed high-rise buildings as
tower steel diagrid has resulted in a highly efficient structural shown in Figs. 1 and 2 can be present at a lot of different design
system that consumed 20% less steel material in comparison to levels for a good design optimization task and a computational
conventional moment frame structures (Rahimian and Eilon, design tool, as varying from overall geometry to geometry of
2006). In addition to the above-mentioned material selection, the structural details. The two most components on geometry
another component to derive structural efficiency into a diagrid of high-rise buildings are the global structural geometry and the
system is the global structural geometry and the structural structural topology within these buildings. By defining both the
topology within this structure. By defining both shape and components parametrically, associated with a free-formed
topology which are generated by structural optimization based shape, and driven by structural optimization based on an
on a numerical technique an optimized diagrid structure can be artificial intelligence technique an optimized structure can be
designed. designed.
The goal of this study is to provide the engineer with a good For the present study, a so-called diagrid unit cell, i.e., a
indication of the feasibility of the diagrid structure, based on widely-known local structural topology system for the present
the geometrical parameters of the diagrid structure. For this free-formed high-rise monumental buildings will be designed by
purpose a typical diagrid structure into a tall building will be a good optimization task. SIMP optimizing both structural
designed by using SIMP method (Bendsoe, 1989; Bendsoe, topology and global structural geometry is one of the most
1995) of topology optimization methods. The original research efficient artificial intelligence techniques and it is termed a
of topology optimization is to generate optimal topologies in material topology optimization design tool. It provides the
structural design using homogenization introduced by Bendsoe engineer with a good indication of the feasibility of the diagrid
and Kikuchi (1988). unit cell as a significant structural element of global free-formed
Topology optimization method of structural optimization high-rise buildings.
methods takes an advantage of obtaining both optimal shape and Therefore successful combinations among a good SIMP-based
topology. Moreover SIMP method is fast and simple for engineering design optimization task, understanding a specific local topology
aspects in comparison to other topology optimization methods such like the diagrid unit cell, and acknowledge of global high-rise
as homogenization, discrete, or geometric methods. In order to building geometry like a free-formed shape require engineers
practically investigate the feasibility of the diagrid system, who are willing to think outside the box in terms of both the
dynamic problem on maximum first or second order eigenfrequency design variants they are willing to consider and embracing new
as well as static problem depending on maximum stiffness is design processes that take advantage of emerging computational
together considered in this study. methods and tools.
The outline of this study is as follows: Possibility of integration
between a design optimization task and a steel-framed diagrid 3. SIMP Material-Based Topology Optimization
unit cell for free-formed high-rise buildings is presented in Problems
Section 2. The SIMP material-based static and dynamic material
topology optimization problems are described in Section 3. In Topology optimization technique can easily be used at the
Section 4 numerical algorithms of the static and dynamic SIMP beginning of the design stage and it provides the optimal material
method are briefly explained. The diagrid extraction by combining distribution into a restricted space called the design space or
optimal solutions of static and dynamic problems is studied in domain. Then connectivity or topology of structures is automatically
numerical applications in Section 5. Section 6 presents conclusions extracted through conceptual shapes formed by density distributions.
of this study. The schematic (Lee et al., 2010) of the solids and voids topology
T T T
subject to: ∫ δε CεdΩX = ∫ δu bdΩX + ∫ δu t dΓt (3) 3.4 Variational and Discrete Sensitivity Analyses for Static
ΩX ΩX Γt and Dynamic Optimization Problems
∫ dΩX – Vref = 0 (4) In this study the analytical variational method is utilized since
ΩX it is numerically more efficient than the discrete method for
where the material tensor C depends on the penalty formulation certain optimization problems like this study. In terms of design
of Eq. (1). The inequality optimization constraint is 0 ≤ Φ ≤ 1, in parameter s, for example here element densities, the total
strictly, not 0 but almost 0 for avoiding numerical singularity is sensitivity formulation (Haug et al., 1986) of the objective function
assumed. Equality constraints indicate a linear elastostatic is composed of an explicit and an implicit partial derivatives and
equilibrium of Eq. (3), and a control of volume of the used is written as.
Table 1. Stiffness and Mass Matrix for Static and Dynamic Topology Optimization
Stiffness matrix Mass matrix
T
Ke = ∫V B CB dV
k1 k2 k3 k4 k5 k6 k7 k8
· k1 k8 k7 k6 k5 k4 k3
e T
· · k1 k6 k5 k4 k3 k2 MC = ∫V ΦN N dV
h h
E i ( Φi ) · · · k1 k8 k3 k2 k5
= ----------------
2 4 0 2 0 1 0 2 0
1–v · · · · k1 k2 k3 k4
· 4 0 2 0 1 0 2
· · · · · k1 k8 k7
· · 4 0 2 0 1 0
· · · · · · k1 k6 1
sym. · · · · · · k1 = --- ΦA · · · 4 0 2 0 1
4 · · · · 4 0 2 0
1 v 1 v · · · · · 4 0 2
k1 = --- – ------ k5 = --- + ------
2 12 4 12 · · · · · · 4 0
1 v 1 v sym. · · · · · · 4
k2 = --- + --- k6 = – --- – ---
8 8 8 8 e 1
ML = --- ΦAI(8 × 8)
1 v v 4
k3 = – --- – ------ k7 = ---
4 12 6
1 3v 1 3v
k4 = – --- – ----- k8 = --- – -----
8 8 8 8
For dynamic topology optimization, the total derivative is 5. Numerical Examples and Discussion
written as a discrete form:
2 e
T
k–1 2 k–1 e The objective of these numerical applications is to present
∂ω u1 [k ( Φe ) K 0 – ω 1 k ( Φe ) M0 ]u 1
---------1 = --------------------------------------------------------------------------------
- (13) numerical stages determining practical diagrid by using com-
∂Φe T
u1 Mu 1 bination strategy between static and dynamic topological optimal
The sensitivities for the eigenvalue problem (Seyranian et al., solutions presented in this study.
1994; Rodrigues et al., 1995) are correct for the case of single A given design structure for the numerical generation is a 2D
valued eigenfrequencies. However, when there is a double wall with fixed supports near four edge parts as shown in Fig. 5.
eigenfrequency, the problem is no longer differentiable, and Note that the solution obtained will absolutely depend on the
when performing the maximization of the lowest frequency this initial fixed supports. The influence of these supports would
usually happens, since the first frequency approaches the second. provide some insight to the problem. A 2.8 m × 3.6 m design
domain is discretized using 28 × 36 square finite elements with
4. Numerical Algorithm for Material Topology four nodes. The material parameters are assumed to be Young’s
Optimization Design modulus of steel E = 1.0 GPa and Poisson’s ratio of steel v = 0.3 .
For only static optimal problem, an applied load of Fv (a vertical
Numerical algorithm of static and dynamic topology optimization load) or Fh (a horizontal load) = 1 kN is concentrated at the
is sequentially composed of a structural analysis, a sensitivity center of the structure. For both static and dynamic problems the
analysis, and then an optimization method. For elastostatic and penalty parameter is k = 3.0 in the SIMP algorithm.
dynamic structural analyses, a finite element method is usually Objective functions are minimal strain energy (kN·m) for static
used. A variational approach with adjoint method or discrete design and maximal fundamental first or second-order eigen-
5.1 The First Design Stage: Optimal Design for Static Dia-
grid Unit Cell Extraction
Figure 6 shows the optimal topologies by using a 0.5 density
isoline contour (Youn and Park, 1997) of all density distribution
Fig. 6. Optimal Topologies for Static Problem: (a) Fv and volume =
contours for static minimal strain energy-based topology 20%, (b) Fh and volume = 20%, (c) Fv and volume=30%,
optimization. Here the 0.5 lsoline is used to extract obvious (d) Fh and volume = 30%
optimal form. Total volume is fixed to 20% in Fig. 6(a) and (b),
and 30% in Fig. 6(c) and (d) during every 50 optimization
procedures.
It can be found that global optimal shapes and topologies
are a X-form in all cases in spite of different X-angles and
their forms are somewhat similar to the typical diagrid node
detail as shown in Fig. 7. When a vertical and a horizontal
load are applied to a given design space, their optimal results
are Fig. 6(a) and (c), and (b) and (d), respectively. As can be
seen in Fig. 6(a) and (c), density distributions make toward
vertical directions in enforcing a vertical load since material
is concentrated and reinforced into positions, in which
stresses are large. Positions with large stresses are relatively Fig. 7. Practical Diagrid Details: (a) Constructed Diagrid Detail, (b)
structural weak and then more material supplement is Schematic Diagrid Detail
needed. In the same manner, when a horizontal load is
applied to a given design space, densities are distributed
toward horizontal directions. Therefore optimal density girder combined to the X-type does not appear contrary to Fig. 7.
assignments conform to Von-Mises stress distributions as From this fact it can be inferred that it is necessary to additionally
shown in Fig. 8. consider dynamic behaviour by repetitive horizontal loadings for
Since X angles yielded on two loading cases considerably complete topology optimization design. The result for dynamic
differ from that of Fig. 7 and each other, appropriate angles of X- problem will be presented in next Section 5.2.
form need to be controlled through a link between both angles in Figure 8 shows Von-Mises stress distributions of optimal
a horzontal and vertical loadings. topologies in Fig. 6. As can be seen, stresses in the center node
In addition, please note that in this static problem a beam or a part of X-structure produced by the loading condition of a
Fig. 8. Von-Mises Stress Distributions of Optimal Topologies for Static Problem: (a) Fv and Volume = 20%, (b) Fh and Volume = 20%,
(c) Fv and Volume = 30%, (d) Fh and Volume = 30%
horizontal load are larger than those by a vertical load regardless As can be seen in Fig. 9(a) and (c), and (b) and (d), global
of volume magnitude of 20% or 30%. It can be found that topologies between eigenmode 1 and 2 are similar but shapes are
material reinforcement for resisting a horizontal loading is more different each other, including effect of volume difference.
important than one for a vertical loading in the node part of X-
form. The largest stresses act to the node part in all the cases, and 5.3 The Third Design Stage: Final Diagrid Unit Cell by
therefore a node is the most significant element with respect to Combining the First and Second Stages
structural safety of diagrid systems. In this Section, the third design stage may be performed, when
both static and dynamic behaviors of diagrid are inspired to the
5.2 The Second Design Stage: Optimal Design for Dynamic present structural detail. This denotes the so-called rule of thumb
Diagrid Unit Cell Extraction (Pope, 1997), which is a principle with broad application that is
Figure 9 shows the optimal 0.5 density isoline contour for not intended to be strictly accurate or reliable for every situation.
dynamic maximal first and second-order eigenfrequency- It is an easily learned and easily applied procedure for
based topology optimization. Total volumes are fixed to 20% approximately calculating or recalling some value, or for making
in Fig. 9(a) and (b), and 30% in Fig. 9(c) and (d) during every some determination.
100 optimization step. Contrary to Section 5.1, there is no Computational diagrid extractions in Section 5.1 and 5.2
loading condition due to the consideration of only free vibration considerably are different from practical diagrid details in Fig. 7.
problem. In order to reduce the problems of the static and dynamic
It can be found that global optimal shapes and topologies are a optimum results, structural design strategy which links between
X-form similar to typical diagrid form as shown in Fig. 7. static results in Section 5.1 and dynamic results in Section 5.2 is
However it consists of a combination of four diagonal presented in this study. It can be found that the combining
members branched in a main horizontal member. Strictly strategy is structurally proper since surely final diagrid forms can
speaking it is different from the perfect X-shape like Fig. 7, resist to both static and dynamic behaviors.
although a horizontal member acts as a beam together linked According to the final diagrid extraction strategy presented in
to diagonal members. Therefore these results in dynamic this study, we could understand practical diagrid mechanism as
design have to be combined to perfect X-shape in static follows.
design and then it produces more similar X-shape toward 1. Diagrids are redundant and load path following.
practical diagrid details of Fig. 7. 2. Angled setting of column elements allows for a natural flow
Acknowledgements
References
<www.dsg.fgg.uni-lj.si/dubaj2009/.../Diagrid%20tehnologija.pdf>. Seyranian, A. P., Lund, E., and Olhoff, N. (1994). “Multiple eigenvalues
Nordenson G.., Riley T. (2003). Tall buildings, The Museum of Modern in structural optimization problems.” Structural Optimization, Vol.
Art. 8, No. 4, pp. 207-227.
Pedersen, N. L. (2000). “Maximization of eigenvalues using topology Sigmund, O. (2001). “A 99 topology optimization code written in
optimization.” Structural Multidisciplinary Optimization, Vol. 20, Matlab.” Structural and Multidisciplinary Optimization, Vol. 21,
No. 1, pp. 2-11. No. 2, pp. 120-127.
Pope, J. E. (1997). Rule of thumb for mechanical engineers - A manual Svanberg, K. (1987). “The method of moving asymptotes - A new method
of quick, accurate solutions to everyday mechanical engineering for structural optimization.” International Journal for Numerical
problems, Gulf Publishing Company, Houston, Texas. Methods in Engineering, Vol. 24, No. 2, pp. 359-373.
Rahimian, A. and Eilon, Y. (2006). “New York’s Hearst Tower - A Wells, M. (2005). Skyscrapers: Structure and design, Laurence King
restoration, an adaptive reuse and a modern steel tower rolled into Publishing.
one.” Structure Magazine, pp. 25-29. Youn, S. K. and Park, S. H. (1997). “A Study on the shape extraction
Rodrigues, H. C., Guedes, J. M., and Bendsoe, M. P. (1995). “Necessary process in the structural topology optimization using homogenization
conditions for optimal-design of structures with a nonsmooth material.” Computers and Structures, Vol. 62, No. 3, pp. 527-538.
eigenvalue based criterion.” Structural Optimization, Vol. 9, No. 1, Zimmer, A. S. and Bell, G. R. (2006). John hancock center scaffold
pp. 52-56. collapse, ASCE Fourth Forensic Engineering Congress.