Steel Skies and Parametric Engineering

Abstract

This paper describes innovative aspects in the parametric structural design process of an automotive flagship store from competition to construction documentation. Novel approaches in design workflow and tight integration of architecture and engineering from the earliest phases on enable a strong correlation of design intent and realized artifact. Parametric structural analysis, cross section sizing, and multi-goal optimization are part of an adaptive procedural model managing a growing body of information throughout all planning phases. Intuitive representation enhances the interface of design engineer to model resulting in amplified adaption of changes in design conditions and requirements. Software interoperability and custom drafting pipelines form a building information system to finally document and deliver complex architectural engineering within a seamless digital chain.

Full text

Proceedings of the International Association for Shell and Spatial Structures (IASS)
Symposium 2015, Amsterdam
Future Visions
17 - 20 August 2015, Amsterdam, The Netherlands
Steel Skies and Parametric Engineering
MSc. Robert VIERLINGER *a, DI Arne HOFMANN *b, DI Martin EPPENSCHWANDTNER *c,
Mag.arch. Moritz HEIMRATH *d
*Bollinger Grohmann Schneider ZT GmbH, Franz Josefs Kai 31 / 1 /4, 1010 Vienna
a rvierlinger@bollinger-grohmann-schneider.at
b ahofmann@bollinger-grohmann-schneider.at
c meppenschwandtner@bollinger-grohmann-schneider.at
d mheimrath@bollinger-grohmann-schneider.at
Abstract
This paper describes innovative aspects in the parametric structural design process of an automotive
flagship store from competition to construction documentation. Novel approaches in design workflow
and tight integration of architecture and engineering from the earliest phases on enable a strong
correlation of design intent and realized artifact. Parametric structural analysis, cross section sizing,
and multi-goal optimization are part of an adaptive procedural model managing a growing body of
information throughout all planning phases. Intuitive representation enhances the interface of design
engineer to model resulting in amplified adaption of changes in design conditions and requirements.
Software interoperability and custom drafting pipelines form a building information system to finally
document and deliver complex architectural engineering within a seamless digital chain.
Keywords: Digital workflow, building information system, continuative procedural model, detail
design, structural optimization, parametric engineering, karamba, multi-objective optimization,
octopus, cross section optimization
1. Introduction
A distinctly ‘shaped sky of steel’ opposes the concrete exhibition landscape in the design for an
automotive flagship store by Delugan Meissl Associated Architects (Fig. 1), re-articulating the ancient
desire for flying structures in architecture. The roof achieves his incisive appearance particularly on
account of lacking perceptible columns (Fig. 2). The horizontal continuity is only partially intercepted
by four concrete cores. All further supports are organized in clusters of five to seven single pillars;
each of these is accordingly small in diameter and therewith extremely slender. Interpretations of a
forest are trying to blur the structural connection of the two parts and integrate in the architectural
context not only visually, but respond to different functions and usages.

Proceedings of the International Association for Shell and Spatial Structures (IASS) Symposium 2015, Amsterdam
Future Visions
Fig. 1: Rendered view of parametric structural model before construction documentation phase
Clustered columns reflect the trees in the vertical green gardens adjacent to each of the four cores.
Like trees the columns are distributed irregularly within each cluster and each cluster is irregularly
placed under the 'shaped sky'. Architecturally spoken the columns should not make an appearance as
columns at first glance but leave focus on the horizontality of the two opposing shapes and the space
formed in between.

Fig. 2: Exterior and interior architectural renderings showing column clusters and vertical greens
The four underground levels are designed to house parking and technical service areas for cars, the
first exhibition level at +0m then forms a transition zone between those regular areas and the
geometrically differentiated architectural articulation above ground. The ‘shaped sky’ of 9,000m²
footprint holds a 5D cinema, a multi-purpose hall of 700 m², administrative rooms, exhibitions spaces
and heavy weight permanent installations as well as a terrace to pick up and test-drive new cars, all
supported on just the four concrete cores and 6 column clusters of 5 to 7 columns each.
2. Structural system
Although with the architectural program the geometry was subject to profound change several times
during the schematic design phases, the design of a hierarchical structural system came about early on.
The four cores are the main elements stabilizing the shaped sky vertically and horizontally, see the
detail for the connection in Figure 4. The vertically supporting column clusters are built just as pinned
rods for the sake of their slenderness, not taking any bending moments or horizontal forces. Spanning
in between and cantilevering out from the cores, the main trusses form the primary load bearing

Proceedings of the International Association for Shell and Spatial Structures (IASS) Symposium 2015, Amsterdam
Future Visions
system in the roof. They use the entire height of the roof, which is up to 9 meters, to bridge the gaps of
up to 40 meters, respectively cantilever up to 26 meters.
Arrays of I-beams form a secondary system with distances of less than 3 meters to easily produce
closed surfaces by application of simple trapezoidal metal sheets. Regularity hereby is an important
objective for economic fabrication (Fig. 3). The column clusters are arranged after several conflicting
criteria, and support both the primary and secondary system.
Fig. 3: Rendered bird’s eye view of structural model before construction documentation;
Rationalization and regular appearance was an important objective in the design.
The final steel structure of the shaped sky weighs around 3000 tons with 6500 modeled elements. The
design happened after Korean Building Code 2009 such that 38 governing load combinations were
modeled in karamba. The maximum displacements in the ultimate limit state were bounded by 25 cm,
under live load it was 5 cm. A beam to hang the entire glass façade from is placed on the underside of
the roof, set in from the perimeter between 5 and 20 meters. The maximum vertical displacements for
the façade beam were bounded to 10 cm (see Figure 4).
3. Digital Design Workflow
Dimensions of 160 x 75 meters in plan, irregularity as ubiquitous condition, an inherently complex
structure, and a dynamic design process from competition to design documentation require advanced

Proceedings of the International Association for Shell and Spatial Structures (IASS) Symposium 2015, Amsterdam
Future Visions
algorithmic parameterization- and optimization technology to conclude with efficiency of workflow
and designed structure in the given time frame.
Fig. 4: Left: Detail developed for three of the four concrete cores to support the shaped sky; right:
façade beam (red) in explosion of concrete landscape and shaped sky.
3.1. Complex modeling
The project is developed in different phases throughout about a year. Though the architectural idea
remains the same, profound changes still happen in late phases. This way, a number of different
approaches in fitting structures into the design had to be developed and tested to facilitate feedback
quick enough to inform the design process again. The establishment of a feedback loop between the
design teams involved is of great importance to the development of a synergetic, well negotiated
solution. On each design iteration, the time needed to incorporate changes and new requirements is
critical to supply quantitative feedback as a basis for decisions.
Procedural modeling is used to set up different building blocks which can be recombined and adjusted
upon each new design adaptation. Generation of the axe geometries in early phases, a search process
for the placement of column positions, dynamic application of load scenarios, assignment of
ambience-aware cross-sections, or many different custom result analysis modules are so being grown
alongside the project.
Collaborative modeling is possible to a certain extent. In the first place the work on subjects which
were to remain divided on the long run is split up to different engineers. The difficulty in apportioned
work on the actual parametric structural model is that dependencies between tasks sometimes are
unforeseeable and dynamic, and therefore are challenging to be done separately.
Visual procedural modeling bears advantages to traditional programming in the development of a
model like this. The visual data flow makes it easier to learn, use, and debug, such that also non-
programmers can become professional users. Speaking of modular collaborative modeling, though it
can become hard to re-use the building blocks after a software update of the platform or the plug-ins
used, which does not happen to code as easy (Park, 2010). Further for a large scale project the bare
handling of big data becomes a challenge, as Grasshopper3d (Rutten, 2015) being the tool used is
geared towards user-friendliness and for this sake produces a lot of overhead computational load.
Software faults and debugging show to be a major factor, especially when exploring new applications
with software packages still in development.

Proceedings of the International Association for Shell and Spatial Structures (IASS) Symposium 2015, Amsterdam
Future Visions
Nevertheless the very same procedural model is refined over the course of over 6 months, where the
focus firstly lies in the fast adaption of changed architectural design, and later becomes the
incorporation of additional detail information. Feedback from a .dstv file based interface to a
secondary engineering program used to perform the official checks and non-linear dynamic analysis is
broken down to simple conditions and idealizations to be fed into the generative pipeline again.
The result is a process distinguished by its flexibility in design changes, its efficiency in producing
detailed output, but also by its dependency on software compatibility and a few persons operating the
master models. Still the approach proved to be very successful as the lever achieved by the emergent
complexity and its optimization capabilities by far outbalanced the efforts needed to produce and
maintain the model.
3.2. Design Methods
Karamba3d (Preisinger, 2013) is used as a structural modeling tool which provides high-performance
feedback and seamless integration with flexible geometric modeling. The parametric model in
Grasshopper3d is acting complementary to the model drawn in Rhino, where the axes of beams are
drawn and fed into the scripting environment (see Figure 5, top row). They are converted to beams
with automatically assigned parameters differentiated on a per-element basis. Those properties (see
Figure 5) include a possible allocation to a smoothing set (middle left), an application of different
joint conditions (middle right), maximum profile dimensions defined by offset surfaces (bottom left),
and loads to be applied for various scenarios and combinations.
Fig. 5: Information for the generative model is supplied geometrically; top left to bottom right: axe
surfaces, groups of beam axes, beam set definition surfaces for size-smoothing, points and lines to

Proceedings of the International Association for Shell and Spatial Structures (IASS) Symposium 2015, Amsterdam
Future Visions
define different joint details, zoned build-up limits to cross section dimensions, surfaces to
dynamically apply loads.
Most of this information is based on the architectural 3D-model plus simple data which is
collaboratively acquired during the planning process. The data is fed into an integrative parametric
pipeline which produces an accurately optimized structure, the basis for all necessary plans, detailed
custom result analysis, a structural model within a secondary engineering program for documentary
calculation protocols, and a deliverable BIM-model in a single process.
Different algorithmic optimization methods are used to improve performance quantitatively, but also
to identify mechanisms and bottlenecks and to identify modeling mistakes. Conflicting intentions are
negotiated with multi-objective genetic algorithms (Vierlinger, 2013) to place the column clusters
(Fig.4), while structural member sizes immediately follow each design change with a Eurocode-based
algorithm until design documentation phase (Fig.5).
The role of the model shifts from a step-by-step approach towards a single growing unit of immediate
feedback on all specifically relevant issues to obtain a consistent buildable structure as early as
possible. Interoperability with industry-standard formats of BIM (Holzer, 2007) and structural models
is maintained along with the flexibility and openness of contemporary parametric design
environments. A semi-automated drafting pipeline at the end of the generative model enables the
rapid production of all necessary drawings directly out of the parametric structural model with
minimal designer interaction (Fig. 6 and Fig. 7). Consistent part-labeling and numbering throughout
the platforms ensures a coherent package of final documents.
Figure 12 shows images from the construction in progress, where the design of the majority of parts
stems from the parametric structural model in karamba and Grasshopper. It can be seen as an example
of successfully applied research and practical use of often experimental and academic platforms which
turn into powerful tools when applied in structured and proficient ways.
Fig. 6: Information in the 3D parametric structural model is used to generate the basis for all
drawings automatically

Proceedings of the International Association for Shell and Spatial Structures (IASS) Symposium 2015, Amsterdam
Future Visions
Fig. 7: Drawing of a main truss produced out of Grasshopper and karamba3d
3. Optimization Process
3.1. Cross section sizing
The parametric finite element software karamba3d (Preisinger, 2013) offers a powerful component to
choose appropriate profiles for each beam in a structure, where it uses the procedures prescribed in
Eurocode 1993-1-1, respectively Appendix B, to check a cross section for its viability (Hofmann,
2011). Given a fixed topology and geometry, it is a tool to economize material distribution within a
system of beams and optimize its load bearing capacity. The procedure is deterministic, which means
there is no randomness involved as opposed to the optimization approach described below, where
meta-heuristic algorithms are used to place the column clusters. Still, the process is iterative and non-
linear as the outputs depend on the parameters and conditions supplied.
Different shapes and families of cross sections are assigned to different groups of beams in the
karamba-model, for each family the sizing algorithm then needs a list of cross sections to choose
from. One profile of a list is assigned to each beam forming an initial model which is analyzed as a
starting condition. For each beam the resulting sectional forces then are checked against the sections
in its assigned list, starting with the first one and continuing until a maximum material utilization of
below 100% is reached, and additionally the checks against local buckling and lateral torsional
buckling are satisfied. In statically indeterminate structures the sectional forces depend on the stiffness
of the members, therefore the process of analyzing and choosing is repeated until the checks are
positive for all elements and no changes need to happen anymore, or the maximum number of
iterations is reached.
The lists of cross sections supplied have significant influence on the outcome of the process. Figure 9
shows a regular grid of beams optimized with differently conditioned lists. An inherent property of
such systems is that bigger and therefore stiffer members attract more forces and again get sized up in
the next step. If the number of iterations is set to a high value typically around 20, out of an initially

Proceedings of the International Association for Shell and Spatial Structures (IASS) Symposium 2015, Amsterdam
Future Visions
smooth distribution following the hogging and sagging moments around the support points, principal
bearing axes emerge which can be recognized as venation-like patterns. The distribution of member
sizes within the list then influences the patterns which are formed, see Fig. 9.
The families of profiles mentioned above in general are shapes of profiles like Is, boxes, tubes,
trapezoids, shells, etc., but in the case of this project are differentiated further. As the structural axes
are drawn in a way such that their ends are always connected (Fig. 5), requirements on different
geometric limits imposed by the architectural surfaces and the different build-up heights have to be
met by applying eccentricities to the cross sections. Therefore, different areas of the building need
different lists of cross sections which follow the local conditions of maximal height and the
eccentricities needed to align them.
The pure structural efficiency is not sufficient for an economic design. The resulting profiles vary a lot
in neighboring and adjacent segments of beams. For this reason the sizing algorithm, in each iteration,
assigns the biggest cross section occurring within a group of beams to all of the members of the group
(Fig. 8). The definition of the groups therefore is a sensitive step, as a single big element would blow
up the entire group. Surfaces drawn manually again serve as an intuitive and quick aid to define these
sets, where the allocation of a line to a group happens when the mid-point of the axis lies on the
surface (Fig. 5). The effect of the smoothing is a slightly higher steel mass but in the first place
enables a manufacturing of the structure within the economic limits.
Further, most of the beam-groups are assigned lists of domestic steel series, welded sections are only
applied where the standards are not sufficient (Fig. 10). The final number of resulting different cross
sections was further reduced by manually selecting a subset of the available profiles.
The algorithm is embedded during all stages of the growing procedural model, giving intuitive
feedback on any problematic spots coming up during the development and refinement of the structural
geometry. The final solution is a carefully adjusted trade-off between structural efficiency and rational
fabrication.
Fig. 8: Effect of group-wise smoothing during cross section sizing; left without, right with smoothing

Proceedings of the International Association for Shell and Spatial Structures (IASS) Symposium 2015, Amsterdam
Future Visions



2iterations20iterationsprogressive20iterationslinear20iterationsregressive
Fig. 9: Different patterns emerging from differently conditioned input lists in cross section sizing;
The graphs show the weighting of cross sections’ height- and width within the same domain.
Fig. 10: Samples of domestic (left) and welded (middle) cross sections; blue lines represent the axes
in the drawings, green lines represent the main cross-sectional axes; eccentricities define the final
position of the profile; Right: beam groups for size-smoothing shown in different colors

Proceedings of the International Association for Shell and Spatial Structures (IASS) Symposium 2015, Amsterdam
Future Visions
3.2. Column clusters
The general concept of a dissolving and irregular arrangement of columns plays an important role in
the architectural design and is well met by a search process using meta-heuristic evolutionary
algorithms. Further embraced by an irregular plan layout and the inhomogeneous structure of the
shaped sky, a non-deterministic process is developed and refined throughout the design phases and
applied to each iteration of the global design. The positioning of cores is defined by functional needs
like escape routes and people flow. Their arrangement is fixed and not part of the optimization
process. The cores carry all horizontal forces. Accordingly the column clusters provide vertical
support only, just the vertical load cases need to be applied for the optimization.
The optimization uses multi-objective Pareto classification to produce a range of solutions negotiating
aspects of program, forces, and economy as differently weighted factors. In the last step the architects
choose the final design from a set of optimized trade-off solutions.
The results always serve as a check for the integrity of the structural modeling, as the computer would
exploit or react to any mistakes made in the digital representation. Different levels of structural
detailing change the load bearing behavior, which is observed and verified by those runs of search and
optimization alongside the schematic design development.
For each stage the evolutionary process is carried out several times with changed boundary conditions
such as:
So called no-go areas are defined by the architects according to the flow of people and the usage in the
main level beneath the roof, see the left part of Figure 11. In these areas columns cannot be placed.
Three different goals were negotiated in the optimization process: Maximum displacement of the roof
structure, number of clusters and the total mass. As for each configuration a precise sizing of all
elements was done (see 3.1 Cross section sizing), the stress is indirectly considered in the total mass.
High stress levels result in a heavier profile and therefore higher mass.
Fig. 11: Left: Assessed (grey) and final (red) solutions of search for column placement, red and white
areas are No-Go areas, columns are only possible in dotted regions; Right: Pareto front
approximation of the search process in the three goal dimensions

Proceedings of the International Association for Shell and Spatial Structures (IASS) Symposium 2015, Amsterdam
Future Visions
Fig. 12: Images of steel construction in progress; top left: main truss of rectangular hollow sections
cantilevering from concrete core, other structure made from I-profiles; top right and middle left:
column clusters introduce a dense grid of beams to distribute the normal forces in adjacent columns;
bottom: overview of site.

Proceedings of the International Association for Shell and Spatial Structures (IASS) Symposium 2015, Amsterdam
Future Visions
5. Conclusion
Although the project shown here on one hand is structurally driven, at the same time it directly reflects
the architectural design intent. The structural logic merges into, builds a synthesis with and cannot
separate from the architecture. In this way of thinking the design process simultaneously combines
architecture and engineering.
Modern design and planning tools allow a reunion of the disciplines. They give designers a fast and
precise communication platform; the fast isochronic exchange and feedback endows a higher quality
of understanding for each party.
In addition the profound flexibility, which is inherent in parametric design processes, allows the
engineer a reaction to changes in the latter design phases that are unavoidable in complex projects
such as the Automotive Flagship Store.
References
[1] Park K. and Holt N., Parametric Design Process of a Complex Building in Practice Using
Programmed Code as Master Model, International Journal of Architectural Computing 8, no.3,
p.359-376, 2010
[2] Holzer Dominic., Are You Talking to Me? Why BIM Alone Is Not the Answer, Association of
Architecture Schools in Australasia, Sydney, Australia, 2007
[3] Preisinger C., Linking Structure and Parametric Geometry, Architectural Design Special Issue:
Computation Works: The Building of Algorithmic Thought, 83(2): pp. 132-135; Jon Wiley &
Sons, London, 2013
[4] Rutten D., Grasshopper3d, software, http://www.grasshopper3d.com, accessed 2015
[5] Vierlinger R. and Hofmann A., A Framework for Flexible Search and Optimization in
Parametric Design, Conference Proceedings Design Modeling Symposium Berlin, Berlin, 2013
[6] Hofmann A. and Preisinger C., Atlas Moderner Stahlbau: Computer in der Tragwerksplanung,
Edition Detail, p. 54-61, 2011