Parametric Design

Process of a Complex
Building In Practice
Using Programmed Code
As Master Model
Kat Park and Nicholas Holt

international journal of architectural computing

issue 03, volume 08

359

Parametric Design Process of a Complex Building

In Practice Using Programmed Code As Master
Model
Kat Park and Nicholas Holt

Abstract
Parameter based design explorations inevitably require a unified master
model that represents the current design state, where each parameter
being explored is essentially a critical sub-case of this master model.
Throughout the constantly changing design state, it is beneficial to
maintain a master model that is flexible and adaptive.This paper
describes the design process of a complex building whose master
model documented the design logic through implementation of
software code.This process is illustrated by the case study of Lotte
Super Tower (Seoul, Korea) from the beginning of schematic design to
end of construction document phase. By maintaining the master model
as a platform-free software code, in contrast to platform-dependent
methods, the case study illuminates the advantages of documenting the
generative logic behind design variations in a way that allows greater
flexibility and a higher level of alignment with design intent.

360

1. Introduction
Integrating parametric models into the process of design requires a
formalization of the generative logic and a systematic way of evolving this logic
in concert with design changes.A successful master model must, therefore,
keep track of the various parameters being explored.The critical sub-cases
that these parameters represent might be as simple as studying the effects of a
numerical value on the buildings form or a much more complex set of values
that explore the buildings response to factors such as environmental
conditions. In the more complex scenarios it might be more efficient for the
explorations to take place outside of the master model; nevertheless, whatever
design decisions might have been made need to be fed back into the master
model in order to propagate its influence on the overall design, as well as
manage the relationships among the various sub-cases. Parametric
representation of a design is a constantly evolving task, since most complex
design problems in practice are continually reacting to addition and deletion
of inputs at multiple scales and levels of complexity. Setting up a parametric
model requires defining the major parameters in the initial stage, but this
initial set of parameters continues to grow and change as the project
develops into different stages. Parametric modeling platforms existing today
offer assistance in defining the parameters and their associative
relationships; however, it is inevitable that using their predefined set of tools
will make the representation dependent on the specific platform.The notion
that availability of a tool invites use is well accepted in design.[1] Different
modeling platforms demand the end-user to have differing levels of
understanding and rigor in the process of defining the parameters and their
relationships. Rigor obtained through the constraints opposed by a software
platform is not only limited and often biased, but it is also dependent on the
form defined by the platform. As design process is inherently parametric, it
is important that designers use an open platform that is most appropriate
for building the relationships as they arise.
The main roles that should be fulfilled by the unified master model are:

offer a platform for the iterative-evaluative loop for design options

capture and document knowledge of other related disciplines
affecting the design
provide a platform for negotiating optimization of competing
disciplines.

This paper demonstrates a new method of managing a complex

parametric design system where programmed code captures the
understanding of the inherent generative logic as well as how the system
might be constructed or fabricated in an efficient and flexible way, while
fulfilling the main roles of a master model mentioned above.

Parametric Design Process of a Complex Building In Practice

Using Programmed Code As Master Model

361

2. Existing Approaches to Master Data Models

It is commonly agreed that all design tools have particular affordances and
biases, and they implicitly embody particular assumptions and values. In
general, a designers toolkit represents a provisional equilibrium of capability
and demand.[1] In this section, we provide an overview of design tools
commonly used in practice to aid in the design of projects similar in scale
and complexity to Lotte Super Tower, seek to illuminate the assumptions
and values embodied in each, and assess how well each meets the criteria of
a successful unified master model as outlined above.

2.1. Master Data Model Using Non-parametric Platforms

Non-parametric methods of representing a design are the most widely used
digital design tool in todays practice, and thus provide a baseline against
which to asses the value of more advanced techniques. Setting up the logic
for a parametric model requires initial investment of time and effort, which
always need to be evaluated against the projected lifespan of the model. A
quick conceptual study might sometimes call for a model without the need
to document the logic or associative relationships, or the initial resources
might simply not be available.
A 3D Rhinoceros model of a form (without the use of Grasshopper
plug-in) is an example of non-parametric modeling platform. In this platform,
one method of exploring or documenting design intent is to use
geometrical elements and geometrical operation tools offered by Rhino.This
method is perhaps most susceptible to the design being influenced by the
tool.While a complete account of modeling by direct manipulation of the
geometry suggests a design process that is beyond the scope of this paper,
we may speculate that a systematic approach to form-generation is possible
in this media by using points, construction lines, etc, to leave a trace of the
logic applied in creating the form. In other words, the subassembly of parts
used to create the desired form is deliberately saved by the operator.

Figure 1. Construction figures are
saved in order to document the design
generation explicitly

362

Kat Park and Nicholas Holt

This approach, the manual parametric exploration, leaves a history of the

generative logic, but the set of data representing this logic is rarely
recoverable and difficult to decipher. More often than not, different design
options created in this manner require similar amount of data and workhours for each additional option. If this method persists into the later stages
of design, the representation needs to be updated to the new design state
by repeating the modeling effort each time.While this representation might
capture the knowledge of a partnering discipline (for instance, duct size
coordination with mechanical consultants requires the floor-to-floor height
of the model to change), the knowledge is indirectly embedded in the
model, and thus hidden.

2.2. Master Data Model Using BIM Platforms

Parametric modeling have matured in recent years well beyond the state of
art when they first appeared on the professional architectural design stage,
both in terms of computational power and well-resolved user interface. BIM
systems like Autodesk Revit and Dessaus Digital Project provide a platform
to represent and manage 3D building components and systems, capable of
integrating the various members in design and construction phase such as
structural engineers, MEP engineers, and fabricators. By applying objectbased parametric modeling, controlled manipulation of properties and
inheritance, these platforms organize the building data using the general
object-oriented computation concepts.These platforms run behind an
interface that is slightly closer to how the software themselves operate than
their predecessors, and require that the user learn the structure of their
language. In exchange for this greater investment on the part of the user,
the database becomes more computer-interpretable, which carries with it a
wide range of advantages: exporting quantities of materials and assorted
other metadata about the components and systems, carrying two-way links
to building specifications, automatically translating the design information
into analysis applications, integrating energy, structural and user simulations
to inform the design, and many more.[2]
One of the key benefits of such BIM system is to create a single central
model that can be explored from the view of multiple disciplines of the
design. Integrated coordination with structure or MEP arm of design is
possible when the architectural model is compared to structural model
through clash-detection; but much of this expectation-realignment can
happen at a conceptual level if there were an efficient way to communicate
the design logic or the structural logic.With Revit or DP master models,
the architectural model can be coordinated with the structural model only
when the structural engineers have put together a structural model. In initial
form-finding stages, the current set of interfaces in Revit imposes more
constraints than invite use. More often than not, formal explorations are
executed elsewhere (typically a more flexible environment such as Rhino)

Parametric Design Process of a Complex Building In Practice

Using Programmed Code As Master Model

363

and imported into Revit in order to document or append more

construction-relevant data. For this reason, it is difficult to discuss Revits
current state as a parametric design platform, as it is used more successfully
as a project delivery platform.

Figure 2. Digital Projects
Specification Tree organizes the
parameter set and relationships into a
tree structure

Digital Projects engineering-caliber geometry engine provides the

designer a toolkit for constructing complex 3D surfaces and solids.They
provide abstractions and interfaces to help define the design problem,
interfaces to organize and manage the parameters and their relationships
with each other.The specification tree lists all the geometry and features
but this organization method is rigid which binds the designer to think of
the parameters in a particular, Digital Project-centric way. Moreover, unless
the associations and parametric relationships have been developed properly
to anticipate certain changes, not only is it difficult to alter them, it is
difficult to maintain any parts that depend on a relationship that might need
to change. Especially at earlier stages of design, such constraints imposed by
the platform are not desirable. Scripting or creating macros within DP can
relieve some of the rigidity but the user is still tied to the platform-specific
geometrical primitives. For example, a designer might speed up the
generation of surfaces through a script, but the script must refer to the DP
definition of a surface, not a mathematical definition of a surface.The
parametric modeling toolset inherent in these BIM systems provide a
snapshot of the artifact being designed; but they cannot capture the intent
or the progression of the design logic through various iterations and
changes, either due to immaturity of the toolset or its rigidity.

2.3. Master Data Model Using Parametric Design Platforms

McNeels Grasshopper is much more flexible as a parametric design tool.
The designer can explore multiple ways of generating one form or element,

364

Kat Park and Nicholas Holt

which can be easily saved as a cluster of modules, often within the same
Grasshopper file.This allows the designer to save the generative logic in an
efficient, organized manner. Nevertheless, the designer is still tied to a very
particular user interface, and consequently, only explore within its
boundaries under its indirect design influences. For instance, one quickly
runs into the way data structures are inherently handled within all the
modules and its large influence in many of the tasks. Because the data
structures are required to be set up in a particular way, users often resort
to creating workaround solutions rather than build a parametric
relationship or association that is intuitive to them. Creating ones own
modules have recently become fairly easy, which have added greater
flexibility. More importantly, Grasshopper has not yet been able to handle
the amount of data or complexity of a project described in this paper.

3. Case Study of Lotte Super Tower

3.1. Project Background
The geometry of 555m tall Lotte Super Tower begins with a constant
transformation from a square base to a circle top.This idea is conceptually
structural as well as architectural: transformation in order to shed wind
vortices which occur in unchanging form, taper for efficient mass
distribution of mixed-use program which require varying lease spans.The
geometrical challenge of transforming a square into a circle was resolved by
creating triangular facets on the building.
 Figure 3. Lotte Super Tower

Parametric Design Process of a Complex Building In Practice

Using Programmed Code As Master Model

365

3.2. Design Variation Exploration

It is important to note that the master model described here did not grow
out of a desire to extend the capabilities of AUTOCAD platform as a design
tool, but rather the need for a platform for the mathematical development of
the geometrical design moves evolving in the tower.The operations in the
programmed code did not use AUTOCAD specific functions, but AUTOLISP
was used for its ease of handling lists of data.The formulas and generative
logic were described in the code with AUTOLISP; the parameter values were
kept in a separate configuration file that the software read in as part of its
initialization process; and different forms of outputs were generated for
different needs.The series of configuration files, each associated with a
specific version of the logic file, made it possible to keep an organized record
of design status as constant changes were being applied.
Figure 4 shows a simple example of the descriptive logic that was used
to provide iterative-evaluative loop for design exploration in the initial
phase of the project: the overall dimension or the number of facet modules
for proportional evaluation, as well as functional evaluation such as floor
plate dimension requirements, enclosed area, etc.

Figure 4. Initial parameter
set for overall tower
dimensions, subdivision
modules, and their
relationships.

Figure 5 shows the simple parameter set and their relationships from
Figure 4 translated into a function named BuildData in AUTOLISP code. As
mentioned previously, the specific values of the parameters are stored in a
separate configurations file in order to keep an organized history of the
values at different design stages, often accompanied by the reasons for their
change from the previous version.The BuildData function abstracts the idea
of a line at the top and bottom of the tower, not a circle or a rectangle.This
lower-level of abstraction allowed the flexibility of changing the top figure

366

Kat Park and Nicholas Holt

 Figure 5. Function BuildData builds

the list of intermediate points for
subdivision modules at various
elevation

into a square (which was one of the numerous design explorations the
project went through) or a circle.This kind of flexibility can often be difficult
to achieve when the platform restricts the operator to use its specific set of
user interface tools to create parametric relationships.

3.3. Knowledge Capture of Partnering Disciplines

Documenting the generative logic into software code also allowed
coordination of design intent with structural engineers to begin at a
conceptual level, as well as capture their knowledge into the master model.
The optimal angle for the diagrid structure (diagonal columns than span
the subdivided arcs across the discretized elevation points) was determined
mathematically as a function of the load each member was required to
carry at their individual locations. Since the angles cannot change midmember, the results from the theoretical solution was reinterpreted with
discrete values pertaining to the dimension of the diagrid, which is shown in
dark blue lines in Figure 6.Translating this type of information into software
code allowed the communication between the design team and the
structural engineers to remain at a level that provided accuracy and
efficiency unachievable by 3D models.The values of zn elevations specified
at right in Figure 6 indicates where these structural diagrid discretizations
occur in the tower.These values were primarily determined by
programmatic and architectural requirements, which also fluctuated
throughout the design phases.The exact nature of the lines which connect
the top arc and bottom line, as well as the logic of exactly how these

Parametric Design Process of a Complex Building In Practice

Using Programmed Code As Master Model

367

discretizations occur along the lines were all design explorations that
affected architecture as well as structure.These ideas were translated into
code and passed back and forth with the structural engineers sometimes as
code, sometimes as pseudo-code.When parameter values changed,
coordination was as easy as communicating a new set of values, since both
disciplines had already agreed on the parameter logic and their relationships
through code.This type of platform-independent communication between
the disciplines facilitated a generative - as opposed to a strictly analytical structural design consultancy, and provided significant flexibility in contrast
to the platforms typically used by architects and structural engineers.

3.4. Documentation of Critical Sub-Cases

The master model served as a central repository of various iterations and
results obtained from specific analysis.This section describes two such
cases: installation feasibility and fabrication analysis of exterior wall detail
and exterior wall panelization strategies.

368

Kat Park and Nicholas Holt

 Figure 6. Left: Engaging Structural

Engineers theoretical solution to the
optimal diagrid angles. Right:The final
values of zn elevations documented at
the end of Construction Documents
phase.

Exterior Wall Fabrication Analysis

Rule-based logic documentation was critical in exploring the buildings
exterior wall design, since the geometry of the towers exterior wall was
determined by offsetting the structures centroidal wireframe in a systematic
manner. After exploring numerous offset methods, we found that the size of
the diagrid nodes was the governing factor in determining the specific offset
dimensions. In other words, the final offset dimension values specified in
Figure 7 represents the minimum distance necessary for the exterior wall
to clear the structural diagrid nodes at each elevation.The new set of
horizontals derived from this method defined the external boundary of the
exterior wall.
 Figure 7. An example of structure
to exterior wall relationship
description

After the horizontal offsets were determined, the diagonal centroid

which locates the line of the diagrid structure was offset as well. Defining
this relationship between structure and curtain wall was a complex task,
since it spanned from a problem of geometrical concept all the way to
understanding fabrication and installation requirements.
The following diagrams show one example of the offset strategy which
was explored in order to determine a structure-to-exterior-wall
relationship that satisfied multiple conditions.

Parametric Design Process of a Complex Building In Practice

Using Programmed Code As Master Model

369


Figure 8. An example of
generative logic for defining the
relationship between structural
centroid and exterior wall

Figure 9. An example of
generative logic (the exploded
option) for exterior wall
expression panel geometry.
Expression panel zone is defined
by the dotted lines enclosing the
gap between surfaces

The variations of offset strategies were designed with rule-based logic in

the master model, after which the results were exported to be visualized in
renderings and drawings to ensure design intent. At the same time, the
relationship was tested for fabrication feasibility through a more
comprehensive torque analysis in Digital Project.This required the master
model to output the exterior wall geometry and the structural geometry
into a set of coordinate data that could be read into Digital Project.

370

Kat Park and Nicholas Holt

 Figure 10. Design model output as

coordinate data for importing into
other analysis platforms.

More than ten different geometrical relationships between structure and

exterior wall were explored during this process, and since all the
relationships were documented as software code, generating the result of a
particular offset method as coordinate data (delimited text file) was natural.
Maintaining the logic as programmed code allowed flexibility to produce
various outputs each catered for different needs, eliminating the problem of
competing file formats prevalent in traditional BIM approach. Figure 10
depicts one particular example of the coordinate data generated for the
triangular facets which describe the surface of the exterior wall.This
coordinate data was imported into a Digital Project model to measure the
level of torque or clearance in order to ensure ease of fabrication and
installation. Modeling the multiple variations of the relationship between the
structure and exterior wall with a parametric platform like Digital Project
would not only have taken much longer due to its heaviness, but also would
have required a specification tree size similar to that of multiple (and
separate) instances of the model.This is a clear advantage of maintaining a
master model as a central repository of design logic, both documenting
design intention and capturing the recommendations of fabrication and
construction consultants.
Exterior Wall Panelization
The exterior wall of the tower is divided into three distinct sections, each

Parametric Design Process of a Complex Building In Practice

Using Programmed Code As Master Model

371

with different set of conditions and characteristics that required a different

panelization logic:
1.

2.

3.

At podium, the exterior wall is composed of approximately 45m

tall cable net wall.Therefore, panelization strategy prioritizes
continuation of structure cables across adjacent triangular facets
(tri-surfaces).
Most of the office and hotel zone of the tower is curtain wall.
Anticipating interior partitions, perpendicular mullion fabrication
takes priority over mullion alignment across adjacent tri-surfaces. In
a few cases, the anchor point of the vertical grid shifts to straddle
the midpoint of the triangle, rather than coincide with the
midpoint.This was a simple solution to maximize mullion alignment
across adjacent tri-surfaces.
The exterior wall at the observation deck and restaurant zone is
free from any interior wall module requirement; therefore, visual
transparency is prioritized.To minimize view obstruction, cable net
structure is defined by a diagonal grid, producing diamond or
triangle panels.

Due to the taper and transformation in the tower form, each triangular
facet (tri-surface) of the exterior surface resulted in a unique geometry,
aside from the reflective symmetry across the center of the overall surface.
Therefore, it was necessary to handle panelization of each surface
individually.The logic of how each surface should be panelized was explored
and documented in the software code; and depending on the location of the
surface, one of final three strategies mentioned above were applied to the

372

Kat Park and Nicholas Holt

 Figure 11. Left: Panelization strategy

for cable net wall at podium. Middle:
Panelization strategy for cable net wall
at observation deck.

tri-surface. Once a panelization strategy was chosen, a more quantitative

analysis of the panelization was performed. For this purpose, the software
produced a delimited text file of coordinates, dimensions, and angles for
every panel in the curtain wall.This initial panel data was quantified to
assess how many panels were regular (rectangle) and how many were
irregular.

 Figure 12. Left: Initial panel types of

curtain wall. Right: Excel-based
algorithm documenting combination
logic of curtain wall engineers,
fabricators, and glass manufacturers.

As shown at left in Figure 12, it was necessary to quantify the small

irregular panels adjacent to the expression zone that could not be
fabricated as individual panels. After consulting curtain wall engineers and
other various parties involved in manufacturing and installing a curtain wall,
a procedural algorithm that described a strategy for combining the small
irregular panels into its larger neighboring panel was developed.The rulebased logic in the algorithm implemented the level of certainty offered by
the consultants.Variables such as the maximum acute angle or the minimum
width of a glass piece depended on the manufacturer, which became
parameterized inputs in the algorithm.
Running the algorithm to create panel combinations according to
manufacturers criteria produced a data set for each panel in the curtain
wall as well as quantify the types of fabrication necessary for the entire wall,
allowing preliminary cost estimation, etc. An optimization routine as
comprehensive as this may only be performed algorithmically and only
implemented within a procedural environment.The methodology of keeping
a code-based master model offers the designer a more seamless

Parametric Design Process of a Complex Building In Practice

Using Programmed Code As Master Model

373

optimization environment in comparison to other methods which might

allow script routines.The hierarchical nature of BIM or 3D parametric
modeling tools can be an obstruction when trying to integrate script
routines into specific locations in the model hierarchy. In the case of
procedures that need to return to particular places to reevaluate and
essentially disrupt the hierarchy, a software code can achieve this with little
effort.

3.5. Facilitating Negotiation of Competing Optimizations

After initial panelization of the exterior wall, the pure geometrical results
were evaluated against construction conditions.Various manufacturers and
other disciplines related to exterior wall, as well as architects with years of
construction experience, combined their expertise and wisdom regarding
the available construction tolerances and the accuracy that is practically
required by a panel dimension.Taking these factors into account, the original
panelization procedure ran an additional layer of optimization that grouped
certain panel dimensions into an average dimension in order to standardize
panel types for fabrication purposes.This type of optimization was
conducted for both the curtain wall and cable net wall at the observation
deck. In the case of optimizing the cable net wall panel dimensions, it was
necessary for the changes in the dimension to be tested for structural
integrity of the cable net concurrently. Multiple disciplines competing to
optimize for its own best solution is natural in a design process. Due to the
inherent ability of a programmed code to document such generative logic,
the master model was able to facilitate the negotiation of trade-off
optimizations of multiple disciplines, as well as document the compromises,
abandoned variations and final results of the optimization.

374

Kat Park and Nicholas Holt

 Figure 13. Left:Visual representation

of the exterior wall panelization.
Middle:The rule-based logic for
combination documented in code.
Right:Visual model of this logic applied
to the original panelization method

 Figure 14. Example of the panel

dimension optimization at cable net
wall

4. Conclusion
Parametric thinking in design may be conceived of as the process of
abstracting a specific task into its operation and inputs. For many designers,
scripting is the first step to explicitly defining and documenting the design
logic. As the Lotte Super Tower project demonstrates, when approached as
a loosely organized network of scripts, a master data model in the form of
programmed code can provide significant flexibility, efficiency, and
coordination capability, especially in the context of complex and prolonged
design problems in practice. In large complex projects, the malleable code
can respond to constantly growing and evolving changes in the project much
faster and more easily than other comparable software platforms.

Acknowledgements
The author would like to thank all members of Lotte Super Tower project
team.

Parametric Design Process of a Complex Building In Practice

Using Programmed Code As Master Model

375

References
1. Mitchell,W.,Thinking in BIM, Architecture and Urbanism, 2009, 09:08 Special Issue,
10-15.
2. Eastman, C.,What is BIM?, Architecture and Urbanism, 2009, 09:08 Special Issue,
16-17.
3. Turkle, S., Simulation and Its Discontents,The MIT Press, Cambridge, MA, 2009.
4. Holzer, Dominic. Are You Talking to Me? Why BIM Alone Is Not the Answer. In
Association of Architecture Schools in Australasia. Sydney, Australia, 2007.
5. Holzer, Dominik, Richard Hough, and Mark Burry. Parametric Design and
Structural Optimisation for Early Design Exploration. International Journal of
Architectural Computing 5, no. 4 (2007): 625-643.
6. Kilian, Axel. Design Evolution through Bidirectional Modeling of Constraints.
PhD, Cambridge, MA: Massachusetts Institute of Technology, 2006.
7. Haghparast, Farzin, and Andrew Marsh. The Application of Computer-Optimised
Solutions to Tightly Defined Design Problems. Eindhoven, Netherlands, 2004.
8. Cross, Nigel. Design as a Discipline. De Montfort University, 2002.
9. Silver, Mike. Towards a programming culture in the design arts. Architectural
Design 76, no. 4 (2006): 5-11.
10. Silver, Mike. Building without drawings: Automason Ver 1.0. Architectural Design
76, no. 4 (2006): 46-51.
11. Visser,Willemien. Designers Activities Examined at Three Levels: Organization,
Strategies and Problem-Solving Processes. Knowledge-Based Systems 5, no. 1
(March 1992): 92-104.
12. Kalay,Yehuda. Redefining the Role of Computers in Architecture: from draftingmodeling tools to knowledge-based design assistants. Computer-Aided Design
17, no. 7 (September 1985): 319-328.
13. Shaviv, Edna,Yehuda Kalay, and U.J. Peleg. An Integrated Knowledge-Based and
Procedural System for the Design of Energy Conscious Buildings. Automation in
Construction 1, no. 2 (September 1992): 123-141.

Kat Park and Nicholas Holt

Skidmore, Owings & Merrill, LLP
United States
Email: [email protected]; [email protected]

376

Kat Park and Nicholas Holt