Toward Mass Customized Architecture. Applying Principles of Mass Customization While Designing Site-Specific, Customer-Inclusive and Bespoke Timber Structures: Innovative Techniques of Representation in Architectural Design

Abstract

Mass customization is established in many industries, but are not yet integrated in architecture and the building industry. This article presents a parametric timber toolkit under development. A flexible toolkit for parametrically designed timber structures, a toolkit that simplifies and substantiates a continuous digital workflow from global shape to digital fabrication and assembly—a toolkit that requires parametric thinking, not only parametric modelling skills. The toolkit proposes solutions to four recurrent workflow-related challenges that limit efficiency and quality while designing timber structures. A series of built case projects are used to exemplify and explain the toolkit. An important finding discussed in the end of the article is that parametric modelling, and partly the toolkit, changes our conception of what is considered a similar structure.

Full text

TOWARD MASS CUSTOMIZED ARCHITECTURE. Applying principles of
mass customization while designing site-specific, customer-inclusive
and bespoke timber structures
John Haddal Mork*, Marcin Luczkowski, Bendik Manum and Anders Rønnquist
* Correspondence: PhD-candiate, john.h.mork@ntnu.no
Abstract: Mass customization is established in many industries, but are not yet integrated in architecture
and the building industry. This article presents a parametric timber toolkit under development. A flexible
toolkit for parametrically designed timber structures, a toolkit that simplifies and substantiates a continuous
digital workflow from global shape to digital fabrication and assembly - A toolkit that requires parametric
thinking, not only parametric modelling skills. The toolkit proposes solutions to four recurrent workflow-
related challenges that limit efficiency and quality while designing timber structures. A series of built case
projects are used to exemplify and explain the toolkit. An important finding discussed in the end of the article
is that parametric modelling, and partly the toolkit, changes our conception of what is considered a similar
structure.
Keywords: “Parametric design and fabrication strategies, CNC and Woodworking Technology, Parametric
timber engineering”
1. Introduction
Is it possible to integrate the benefits of mass customization (MC) into architecture? How can architects
apply the principles of MC while designing site-specific, customer-inclusive and bespoke timber structures?
Mass customization, a term introduced by Stanley Davis in 1987(Davis, 1997), was later redefined by Andreas
M. Kaplan and Michael Haenlein as follows: “Visionary traditional MC in a strategy that creates value by some
form of company-customer interaction at the design stage of the operations level to create customized
products, following a hybrid strategy combining cost leadership and differentiation.”(Kaplan & Haenlein, 2006)
A challenge that distinguishes architecture from other professions is the large design space that has to
be offered. Naval architects always make variations of a ship with a hull, an engine, and a variety of equipment.
Designers in the footwear industry always deal with feet of different sizes and customers with different uses
and styles. In contrast, architectural design projects range from working on a bridge to a high-rise to small
furniture. Johanna Daaboul stated that MC can be offered either via product variability or process
variability(Daaboul, Cunha, Bernard, & Laroche, 2011). Offering both, a wide range of process variability and
product variability, is a core service of being an architect and makes mass customization especially complex.
First, every project site is unique and includes the functional requirements, locally available materials, climate
and building culture of a given location to be unique. Second, the user and the user’s needs and functional
demands are unique. Adapting projects to their contexts is what makes architecture, architecture. Furthermore,
architecture is a highly subjective task, partly a poetic process from a vision to a built structure, and mass
customization must not remove such an important quality.
The use of Computer Numerical Controlled (Rapp & Johnson, 1980) (CNC) Machines has become
widespread in the building industry (BI). CNC-machines are used for a large range of manufacturing
operations, such as cutting, drilling and milling, and are digitally controlled. One of the biggest benefits of using
such tools is that the production process is automated and uses the same amount of time manufacturing a set
of unique components, such as a set of identical components. However, automated manufacturing of unique
building components does not implicate automated, manufacturing ready computer drawings. In contrast,
unique building components require unique drawings and demands instead of an automated strategy to be
economically sustainable.
Computer-Aided Design(Burns, 1986) (CAD) has existed since the 80s, but in reality, it is more or less a
digital, augmented variant of manual drawing. Later, Building Information Modeling(Kymmell, 2007) (BIM) was
introduced. From dumb geometry, 3D geometry suddenly represented a building component—a building
component aware of where it was, what function it had, optional data about cost, manufacturer, etc. According
to Jared Banks, BIM enabled designers to spend more time designing and less time documenting and
coordinating(Banks, 2016).
Digital parametric workflows, which widely appeared a decade ago, represent a dramatic new way of
applying digital power while designing our built environment. Instead of drawing geometry, parametric

workflows let the user define the design as a series of decisions, systems and relationships(Krymsky, 2015).
Hence, the digital design process has the potential of becoming more flexible than existing, partly standardized,
BIM processes. Robert Aish described it as the computation era and argued that the objective of design
computation is to overcome many of the limitations of BIM (Aish, 2013). We see a computational approach
and parametric thinking as the next, decisive step to reach mass-customized architecture and BI.
Parametricism (P. Schumacher, 2008) is a buzzword that has been around for some years and is often
associated with organic, prestigious buildings made by architects such as Zaha Hadid or Frank Gehry.
However, parametric workflows have greater potential than creating signature architecture and are thoroughly
described in Architectural Design’s special edition, “Parametricism 2.0”. Patrik Schumacher claims following:
“In order to reverse the current marginalisation of Parametricism, it is necessary to relaunch it in a self-critical
redirection as Parametricism 2.0. Parametricism is architecture’s answer to contemporary, computationally
empowered civilisation, and is the only architectural style that can take full advantage of the computational
revolution that now drives all domains of society.”(Patrik Schumacher, 2016)
Great examples of such methodology can be seen in the works of Shigeru Ban and Achim Menges. Shigeru
Ban’s famous Nine Bridges Country Club gridshell is realized with the help of DesignToProduction (consultant
company) parameterization and automated manufacturing(Scheurer, 2013). Another outstanding example is
Achim Menges, who applies parametric tools throughout the entire process while making his pavilions and
structures. Nevertheless, the mentioned projects are more customized than those that are mass customized.
Large budgets and partly research-funded projects make such projects happen.
What is mass customization in architecture? Parametric workflows are extremely efficient when
established, but establishing such workflows in each project may not be economically justifiable. Hence, the
goal must be to create a continuous parametric workflow from sketch to fabrication that is applicable to a wide
range of projects. Through designing and building a series of timber structures according to an established
parametric approach, we identified four recurrent workflow-related challenges that limit efficiency and quality
while designing timber structures. These are also the challenges that often cancel a continuous workflow from
sketch to fabrication and are the motivation to create the parametric timber toolkit:
1) Parametric Detailing: Finding a robust way to sort and identify geometric data that are time-consuming
in each individual project. Among other reasons, data is sorted to be able to design the different details in a
structure.
2) Tectonic Architecture: Processing timber components that are subtractive, meaning that the material
is subtracted from a stock. In contrast, digital design is largely additive, e.g., extrusions, revolves, sweeps, etc.
Thus, a recurrent design challenge focuses on how a component is physically manufactured.
3) Manufacturing Descriptions: Concerning timber building design, architects and structural engineers
tend to output geometry that the manufacturer has to redraw. Geometry that does not include information about
the manufacturing process. The result is redundant work and increases the chances of modeling errors.
4) Structural Analysis: Since the timber is orthotropic material and the most demanding phase in
structural analysis is the connection design, the faster the structural analysis is implemented, the greater the
chance of achieving a feasible project. Similarly, in manufacturing conversion, a challenge is to automatically
convert geometry modeled by an architect to structural analysis.
“First build your tool” is the title of the abovementioned Aish’s article(Aish, 2013). The objective of the
ongoing research project has been to develop a flexible toolkit for parametrically designed timber structures,
a toolkit that simplifies and substantiates a continuous digital workflow from global shape to digital fabrication
and assembly. A toolkit that requires parametric thinking, not only parametric modeling skills.
The toolkit is based upon the process of designing timber structures, but many of the principles are
general and apply to other materials. This article thoroughly describes the system and principles of using
elements and nodes as a basis for most types of parametrically designed architectural structures. It describes
how architects, engineers and manufacturers considers an element, and a shared solution is proposed.
Furthermore, the article describes the algorithm that sorts details into detailing groups. A series of built case
projects are used to exemplify and explain the toolkit. An important finding discussed in the end of the article
is that parametric modeling, and partly the toolkit, changes our conception of what is considered a similar
structure. A structure can be similar in parametric description but can vary much in function, form, scale and
all other attributes. What appears as visually and radically different structures from an architectural point of
view is very similar from an algorithmic point of view.
2. The computational workflow

2.1 Software platform
The toolkit is implemented in Grasshopper 3D(McNeel, 2015), an add on to Rhinoceros 3D (McNeel,
2009). The visual programming software was introduced in late 2007 and accelerated the architectural
software revolution.(Krymsky, 2015). Currently, the toolkit is an in-house beta-version, but is planned to be
released as an open-source plugin early 2019. Using C# and Visual Studio, the toolkit is now developed as a
proper plugin. Earlier development, as described in the case-studies, have been a combination of Grasshopper
and IronPython scripting. The current toolkit is based on C# classes, which corresponds to the detailing groups
described in chapter 4.1. The export from Rhinoceros is based on Building Transfer Language (BTL), and will
be described in chapter 4.4
Figure 1 and Figure 2 shows a simple example of the concept described in this paper. A few bars are
connected in a node. If the bars are shorter than 925 mm, they are trimmed with an angle. The left side of
Figure 1 shows the result in Rhinoceros. The right side of the figure shows the manufacturing readyBTL-export.
Figure 2 shows the script in Grasshopper.
Figure 1: Rhinoceros to the left, BTL-viewer to the right
Figure 2: The grasshopper script required to create the geometry illustrated in figure 1
2.2 Workflow
To use parametric modelling as an efficient method, it is crucial to ensure a continuous flow from overall
geometry, via detailing and structural analysis to manufacturing output. Furthermore, it is crucial to sustain a
workflow that suits both architects, engineers and manufacturers. This chapter explains briefly the chosen
computational workflow. The following chapters explains in-depth how the introduced challenges are solved
and implemented.

The components and workflow are described in Figure 3, and the procedure is as follows:
Overall geometry
1) Centerline-geometry is generated with the help of Rhinoceros or conventional Grasshopper-
components
2) The centerline-geometry is fed into one or multiple element components. Here, the centerline is
materialized.
3) In the element components, the cross-section, cross-section orientation and material are defined.
4) The property description tools describe each detailing group. Here, the node-properties are also
assigned.
5) The loads are defined.
6) The loads, the elements and the descriptions of the detailing groups are attached to the assembler.
The assembler generates relations between the nodes and elements, generates the details and
assigns the details to its detailing groups. This step relates to challenge 1 and are thoroughly described
in chapter 4.
Structural analysis
7) Finite element analysis is performed using Karamba (Preisinger & Heimrath, 2014). However, the pre-
processing and post-processing of the analysis is included in the toolkit. To integrate the structural
analysis and the architectural design, the structural FE objects are being made simultaneously with
definition of the geometrical elements and its components (material, cross section). More over the
results from FE analysis are automatically send for post-processing. The components are checking
the elements and joints according to the EC5 criteria and are informing user (designer) about utilization
ratio and which combination of failure state (compression, tension, bending, combinations) is crucial.
Detailing
8) The respective detailing groups are extracted by a component. Here, the user can extract each
member of the detail.
9) Properties from the members of the details are further extracted.
10) Subtractive tools detail the timber structure. This step relates to challenge 2 and 3. The Subtractive
tools are described in chapter 5.
Output
11) The outputs from the subtractive operations can be used for visual inspection, FEA or BTL-export.

Figure 3: the computational workflow
3. Establishment of a shared parametric approach for architects, structural engineers and manufacturers
An established method of modeling digital structures is to start with the center-line geometry, modeling
the theoretical center-line of any building component. Columns, beams, chords and bars are described by two
points —the start and end points that construct a line. If two lines share the same points, the lines are
connected neighbors. The shared point will then be a connection node. With only points, lines and geometric
properties such as cross section and materials, a majority of the different kinds of structure can be described
in architecture or engineering.
As in many other applications, the parametric timber toolkit is based on a system of curves and points.
Curves can represent any building component with one dominant geometrical direction. In this article, these
components will be described as elements. In buildings, elements are mostly attached by a physical connector,
but digitally, they can be described as single points where elements start, end or intersect (hereby called
nodes). See Figure 4.
Figure 4: The article and toolkit’s definition of elements and nodes
As a general description of a topology, the overall shape of the structure, the description of using curves
and points, works very well. However, when architects, engineers and manufacturers are going into detailed
design, analysis and manufacture of the structure, the described system is not optimal. The following is a
generalized description of how geometry generation methods are preferred according to each of the building-
process participant:
Figure 5: How architects, structural engineers and manufactures define an element
3.1 Architect’s conception of an element
What you see, is what you get. For an average architect, it is enough to consider an element as a building
component. The physical component ordered from the manufacturer is what is being considered as one
element. The example in Figure 5 shows two columns and an arched beam. Depending on the detail-level,
further detailing of the node and even the refinement of the elements might be required. Regardless, the
architect can consider the illustrated structure as three elements.

3.2 Structural engineer’s conception of an element
The structural engineers view the objects from two perspectives, physical and numerical. A numerical
model always aims to simulate the physical behavior of the structure or at least it part (here called element).
Currently, the numerical models are mostly built using the finite element method (FEM).
In general, (the physical model) one element is understood by the structural engineer as an object with
material continuity that allows continuous stress distribution. The engineering simplification of the objects to
the beam element is very intuitive and allows operating on force and deformation for estimating the element
capacity. Two rules are essential:
1) Commonly, the beams are represented by the linear finite element; for representing curves, we
divide it into a sufficient number of smaller finite elements.
2) The connection/continuity between elements is described by the nodes, and the finite elements
have to begin or end in the nodes.
With the rules in mind, we see that the architectural model on the left side (Figure 5) does not fulfil the
requirements for a structural analysis. First, the arch must be segmented into a polyline and fragmented into
linear elements. Second, the beam must be divided where the columns are attached. That is, for a structural
engineer, the structure consists of seven elements.
3.3 Manufacturer’s conceptions of an element
The architectural conception of an element is also often sufficient for a manufacturer. One element is one
building component. However, sometimes, an element consists of multiple sub-elements joined together as
one element. There are many reasons contributing to this phenomenon. For glulam-manufacturers, the
reasons to split the element can be due to size limitations or being able to mill a complex detail.
For such purposes, the element must be divided into multiple elements, but not likely in the same manner
as for the structural engineer.
If the structure in the example were detailed and slotted in plates and dowels as nodes, the manufacturer
would likely prefer to construct each element from at least three sub-elements. In this way, space for the plates
could be easily milled, thereby being glued together as one element. The result is seen in Figure 5
3.4 A solution that integrates the three conceptions
How is this problem solved? How does one make a model that satisfy these three approaches? Owing to
object-oriented programming(Goldberg & Robson, 1983), a class-structure is developed to contain the three
ways of describing an element. The master element is similar to the physical component delivered to the
building site. However, a subclass is also storing one or more structural elements within the main element.
Similarly, one or more manufacturing sub element is stored in the main element. The system is illustrated in
Figure 6. Note that the structural sub element is subdivided transversal while the manufacturing sub element
is subdivided longitudinal.

Figure 6: Object-oriented programming gathers architects, structural engineers and manufacturers’ conceptions of an
element in one master element.
In addition, geometric representations of the element and a series of other data are generated and stored
in the element object. First, geometric data such as length, height, width and local planes and vectors are
stored. Second, cross-sectional data, material properties, manufacturing data and other metadata are stored
in the element object, making it trivial to extract relevant data when detailing or analyzing a structure.
4. Rethinking parametric detailing: Introducing property-based detailing groups
The beauty of parametric modeling is the ability to create a limited amount of parametric details that
become valid for numerous instances in a given structure. The challenge is to create a robust sorting algorithm
that updates any geometry correctly. The first level of complexity is to make the sorting work for a given
topology; a more complex sorting is to make it work for any kind of topology, allowing the topology of a structure
to change fundamentally, but the details are updated and distributed correctly. Through the research project,
a general and flexible sorting algorithm for parametric modeling has been developed. The method is
surprisingly simple.
However, to describe the concept properly, let us start from the beginning. What is a detail? When
detailing a node, one is also detailing the ends of the connected elements. Hence, a detail of the node
influences the node, and due to, for example, holes in the element, the detail changes the element size.
Mutually, when detailing an element, the connected nodes may be affected. Thus, the system defines a detail
as a coupled system of elements and nodes, that is, either a node and its elements or an element and its
nodes. The principle is shown in Figure 7.

Figure 7: Node-focused details and element-focused details. A node has elements and an element has nodes.
Every type of structure, both parametrically and conventionally designed, contains a set of details. How
the column meets the ground, how the members in a gridshell are connected, and how the top-chord is
connected to the two bars are examples of details. However, there is a big difference between parametric and
conventional detailing:
Parametric detailing principle
One parametric recipe for a detail generates specific geometry for all instances of the details in the
structure. A good parametric recipe of a detail is able to handle a wide range of variations. The more general
the recipe is, the bigger algorithm has to be.
One model – several possible detail instances
Conventional detailing principle
If there are no variations, then one detail applies to all. However, if there are variations within the same
detailing principle, one either has to draw several variations or describe only one in detail. The rest depends
on the builders to replicate. Since each detail needs a new model in this approach, the time of design and the
probability of making a mistake by the designer increase.
One model – one detail instance..
4.1 Detailing Groups
How a detail is crafted is decided by a series of considerations. Architectural, structural and manufacturing
considerations are taken into account, but all of the other disciplines included in a building project influences
a design. Are the details of the structure load-bearing? Is the detail geometrically planar? Is the connection
hinged or fixed? Will the connectors be visible? Do the details need fire-protection? What class of materials
will be applied? How will the components be assembled on the construction site? The possible considerations
are almost endless.
Figure 9 shows two structures. An experienced human can intuitively point out what is similarly detailed.
The challenging part is to precisely describe the logics in a language that a computer understands. The method
used in proposed algorithm is surprisingly similar to the children’s game called “Guess who?”. The purpose of
the game is to use as few possible descriptive questions to ask of whom the competitor is thinking. It is not
allowed to ask topology-based questions, such as “is it a person in the bottom-left corner?” See Figure 8.

Figure 8: “Guess who?” Property descriptions filter the different groups
Figure 9: Similar to “Guess who”, the detailing groups can be described by property descriptions.

A similar approach is suitable to identify details in a structure. Figure 9 shows two structures. By using
property descriptions, the same descriptions can be used on two different structures. In figure 9, all details that
contains two elements becomes a foundation node.
The key advantage of the described sorting principle is not to use descriptions relative to the topology
and space. Rather, descriptions of local parameters, only influencing each individual detail, are used to identify
the details. Hence, a change in the topology does not disturb the sorting system.
The sorting principle is implemented in the toolkit by a series of components regarding property
description. Detailing groups are defined by telling whether a property description is true or whether a property
description is false. When the detailing groups are defined, each detail is analyzed to determine of which
detailing groups they belong. A diagram of the system is described in Figure 10. Table 1 presents a list of
property descriptions that have been or will be implemented in the toolkit:
Name
Description
Option
Element length
Defining Min/Max Length
Valid for one or all
Element angle
Max/Min angle between elements
Valid for one or all
Element
parameter
Defining Min/Max parameter where the node
is connected to the element
Valid for one or all
Element
amount
Min/Max amount of elements
Node position
Checks if node is inside defined bounding box
Element
Names
Checks if Detailing group contains defined
names
Including at least or exactly defined
elements
Element
direction
Defines min/max angle deviation from defined
vector
Valid for one or all
Normal Force
Defines min/max Normal Force value
Valid for one or all
Moment Force
Defines min/max Moment Force value
Table 1: Property descriptions that have been or will be implemented in the toolkit

Figure 10: Diagram of the sorting algorithm. The algorithm checks each detail and determines if it fits one or several
detailing groups.
The result from the described sorting system is that one may end up with details present in either zero or
multiple detailing groups. If that is not the intention of the specific project, then the detailing groups must be
further described. However, there are cases in which details connected to zero or multiple detailing groups are
relevant. First, being connected to zero detailing groups implies that no detailing is required, the stock does
not need any refinement. Second, there might be sub-details that overlap in some details. An example is the
chord connections of the Follo Bridge shown in Figure 11. The upper chord and lower chord are primarily
similar, but the bottom chord node detail includes a suspended connection to the secondary structure. Hence,
the connections in the top chord only belong to the ChordNodeDetail, while the connection in bottom chord
also belongs to the SuspensionDetail
Figure 11: The bottom chord detail belongs to the two detailing groups

4.2 Node Properties
The properties of the elements can be created simultaneously by defining the elements. Assigning
properties to the nodes is a slightly more complicated process because the node geometry is a consequence
of the intersection of the elements. The features of the node can be changed, added or limited according to
changes in the global geometry. The system assumes that all nodes within one node-based detailing group
have shared properties. Hence, the desired properties are connected to the property description component.
Generating a sufficent node plane is crucial for efficent parametric detailing. As previously explained, the
node is generated from the end points or the element intersections and is by default just a point with no normal
direction. However, when detailing a spatial object around this point, the orientation of a local coordinate-
system, or a local node plane, must be clearly identified relative to the global coordinate-system of the overall
structure. How a node plane is oriented is highly dependent on the type of structure and how the node is
detailed, but this orientation is valid for all parametric detailing procedures; a consistent and logical node plane
generation is crucial for efficient detailing.
What is a consistent and logic generation of a node plane? The concept can be well-explained in a truss
bridge example. By default, a node plane is an XY-plane (Figure 12 A) but is most likely not suitable for the
nodes in a bridge. If the bridge is straight and parallel to a global axis, an XZ-plane or a YZ-plane is usable,
but it is not flexible. A more consistent rule would be to say that the plane is planar to the truss plane (Figure
12 B). Then, the structure can be oriented in any direction. However, the logics of the planes are still not
consistent. The normal direction may be flipped, and the X-alignment may not be defined. By stating that the
normal direction of the plane always faces from or to the secondary structure and the X-axis is parallel to the
bottom chord, a fully consistent and logical plane-generation is established (Figure 12 C).
Figure 12: Different strategies of node-plane generation. Figure 9 C shows a consistent and logical plane-generation
These rules apply well for a truss bridge, but the logic does not necessarily apply for all other structures.
For example, a gridshell structure is easier to detail if the normal axis of the node plane is parallel to the normal
surface. Thus, the toolkit should supply various node plane alignment components.

5. Digital Subtractive tools
There are many purposes of creating a 3D CAD-model. The way a model is made influences the model
output possibilities. A sketched volume model is relevant to understand the space to be created but most likely
is not adequate for fabrication purposes.
Most of the big manufacturers in the timber industry use CNC tools, such as Hundegger Speedcutter
(Hundegger, 2001). These tools have two shared characteristics: they are controlled by numerical data, and
they subtract material. These features must be taken into account when designing timber structures digitally.
Since manufacturing timber is highly specialized and tool-dependent, a design must often be redrawn by the
manufacturer. Redundant work is done, but it also increases the chances of human modeling errors. To solve
this challenge, the modeling purposes and geometrical outputs of the architects, structural engineers and
manufacturers were investigated. The timber toolkit aims to create an effective solution that fits all three
professions. The following is a brief description of these three professions’ objective for 3D-detailing timber
structures. The description focuses on the part that influences all three professions, namely, the load-bearing
system.
5.1 Architect:
While some architects leave the load-bearing system to structural engineers, other architects base the
architectural expression on how a structure is made. They use architecture to communicate how the forces in
a building work and how a building has been crafted. To design such architectural expressions, it is suitable
to make a detailed 3D-model, including the bolts and brackets. Indeed, the architect does not dictate the
dimensioning but can hold a great influence on how the structural load-bearing system and detailing are
composed. Thus, the architectural geometrical output is primarily visual. Secondary output such as cost,
manufacturer, and volume may be generated.
4.2 Structural engineer:
The timber structures due to natural imperfections in the material (orthotropic, not linear grain angle,
rods..) are very sensitive to imprecise detailing. Even the best calculations and perfectly chosen static schemes
can be ruined because of bad explanations of the joint fabrication/production (the detail).
The issue that is characteristic of timber engineering is finding the connection stiffness. Most of the
connections are designed to be rotation-free hinges. Finding the real characteristic of the joint behavior is left
either to experimental studies or sophisticated experiments.
Two computational approaches are described—one which is simple and fast, and one which is advanced
and more CPU-demanding. The simple detail analysis is made by applying analytical equations (e.g., from
Eurocode 5). These equations can check a whole structure in real time. The advanced analysis is made based
on sub-modeling. Critical details from global analysis are sent to local analysis. Finite element analysis is
performed with volume-objects and more precise material descriptions, and it includes eventual steel
connectors such as dowels, bolts and screws.
At first glance, the geometry needed for advanced analysis is very similar to an architectural model.
However, the major difference is that how the mesh of the analysis model is built up is critical to make the
analysis work.
4.3 Manufacturer:
In regard to digital fabrication, the manufacturer’s output is preferably an assembly-ready building
component, but the input is a primitive stock. Thus, one of the roles of a high-tech manufacturer is to create
instructions for a CNC machine. Two-dimensional cuts performed by lasers or mills are possible to import from
a standard CAD-model, but more complex operations have to be defined as virtual timber processing
operations.
4.4 Finding a shared detailing solution
The three described methods and geometric outputs look surprisingly similar, but the difference is how
the geometry is being made. There are some existing tools that are bridging the gap between the disciplines,
but assuming a user-controlled parametric workflow, the options are limited. CadWork and HSBCAD are tools
that are tailored for timber structures and connect architecture and manufacturing. Further, Woodpecker, an
innovative tool developed by Lignocam and Designtoproduction, enables the export from Grasshoper 3D to
any CAM-software. This export occurs through an open-source format called Building Transfer Language.

However, Woodpecker is not a design tool; rather, it is more of a geometry conversion tool (STEHLING,
SCHEURER, & ROULIER, 2017).
The solution developed in this toolkit aims to bridge the gap between all mentioned disciplines while
modelling parametrically. The concept is to allow the user to refine timber components exclusively based on
subtracting material from a digital stock. Thus, the designer is forced into thinking about physical processes
while designing virtually. When the architect applies a subtractive operation to a stock, three outputs are
automatically generated. A 3D preview geometry, an FEA-ready mesh and a BTL-description. (In the current
version of the toolkit, the FEA mesh is not yet implemented). The concept is illustrated in Figure 13.
Figure 13: Different outputs from one subtractive operation
5. Case Studies
A series of case projects have been built to test the capability and flexibility of the toolkit. The following
sub-chapters shows how the toolkit was used to parametrically design two timber bridges, a freestyle water
ramp, a log house and a shelf. The structures are visually and fundamentally different but surprisingly similar
when applying the ideas of elements and nodes. The toolkit have been developed while designing the case-
studies. Mock-ups of the toolkit have been developed in both Grasshopper-scripts, Iron-python and C#. Table
2 shows what elements of the toolkit have been used designing the structures.
Project
Script Language
Detailing
groups
Subtractive tools
FEA
BTL-Export
Orkla Bridges
Iron Python
X
X
X
X
Water Ramp
Iron Python
X
X
X
Log House
GH-components
X
Book-shelf
GH-components
X
Table 2: The table shows which elements of the toolkit have been used to design and build the structures.
5.1 Orkla Bridges
The bridges are designed for a pathway in a park and cross two small rivers. The distance between the
bridges are approximately 1 km; hence, it was natural to create two bridges that had a similar architectural
expression. However, the boundary conditions at the sites were different. The first bridge, Follo Bridge (shown
in Figure 17) had a span of 10 meters and on height-differences but had to be arched to allow the potential ice

drift through. The second bridge, Evjen Bridge (shown in Figure 16) had a span of 15 meters and had a
relatively large height difference. Furthermore, the bridge needed curved platforms on both sides to connect
to the path and fulfill the requirements of a maximum 1:20 rise.
The different boundary conditions gave a nice opportunity to build a parametric model with a large enough
design-space to be adaptable for both bridges. The architectural expression is based on a well-established
system of dowels and slotted-in-plates as a connection system (Mork & Luczkowski, 2017). Due to relatively
heavy loads, a secondary steel structure had to be used, but the rest of the structure is glulam-based. While
the smallest bridge has a classic arched shape, the larger bridge has a more organic appearance with a
doubled curved railing.
Using the principle of detailing groups as outlined, only four different detailing groups had to be developed
in the small bridge: the bottom chord, the top chord, the foundation points and the zero-force node. Figure 14
shows the rules applied
Figure 14: Detailing groups in Follo Bridge. The same rules were applicable while designing Evjen Bridge
The bottom chord node is the most complex detailing group. The structural bars are connected to the
bottom chord, and the secondary structure is suspended from the metal-plates. Due to the scale of the bridge,
the suspension connections were integrated inside the chord, making a more compact detail. As previously
shown, the basis for the detailing is stocks that are not processed. Subtractive tools are used to design the
timber elements. In addition, steel components were designed using conventional parametric tools in the
grasshopper environment. In the bottom chord detailing group, these processes were performed (Figure 15):
• Cutting the bars parallel and with an offset half the height + 30 mm from the tangent direction of
the bottom chord center curve.
• A grid of holes for the dowels. The amount of dowels was calculated based on the force in the
bar/chord, and the size of the grid was calculated based on the required edge distances.
• A pocket milling to make space for the metal-plates. Both the plate that connected the bars to the
chord and the plate for the suspended secondary structure.

Figure 15: Subtractive operations to achieve wanted detail
Figure 16: Evjen Bridge. Photo: Arnfinn Sæthre

Figure 17: Follo Bridge. Photo: Arnfinn Sæthre
Figure 18: The sauna (mostly L-nodes) on the left side and the water ramp with secondary connection and primary
extension nodes.
5.2 Freestyle water ramp
A freestyle water ramp is a ski jump that ends in a lake and is used for summer training. This structure
was a simpler design than those of the two bridges. Due to low loads, a beam-column structure was feasible.
However, the principle of using dowels and slotted-in metal plates applied in this structure was used as well.
Surprisingly, the water ramp had more details to develop than those of a bridge. The detailing groups can be
seen in Figure 19, and Figure 18 shows the ramp under construction

Figure 19: Relatively high amount of detailing groups. Note how the yellow detailing group can be used to trim building
components.
5.3 Log house sauna
Until now, the examples have shown rules based on element names. However, there are many cases in
which element names cannot be used. Often, the elements are nameless or a set of similar elements. In
addition, previous examples have described spacious structures. Is it possible for the principle to apply a more
homogeneous structure, containing only one type of element?
An algorithm was developed to be able to model the wall’s surfaces. The algorithm then splits the structure
into layers with a center-line geometry. What is challenging is to generate the actual length of each individual
element. Some elements are extended, and some elements are trimmed to become an overlapping system.
Looking at the system, there are four kinds of connection types, i.e., four types of detailing groups—the
X-node, the T-node, the L-node and the I-node. These are easy for a human to determine, but what rules can
be applied in the toolkit to distinguish the nodes? An X-detail contains only two laths, but that is the case for
the L and T-detail, as well. In this case, information regarding where the node is relative to the lath was used
to distinguish the details. Descriptions of the detailing groups are shown in Figure 20 and the final result is
shown in Figure 18.

Figure 20 Detailing groups applying the positioning of the node relative to the element to distinguish the details.
With such rules, all the different types are separated, but there is one more condition needed to generate
a log house, namely, the shift in which element is extended. Thus, one more condition to all the detailing
groups is if the group is on an even or odd layer. If the T-node is on an odd layer, the element connected to
the end will be extended, and the opposite will happen if the node is on an even layer. This principle is shown
in the bottom-right of Figure 20.
5.4 Plywood shelf
Finally, the principle has been tested on furniture: a plywood shelf prefabricated using a 3-axis CNC-mill.
Both visually and in function, a shelf is something extremely different from a truss, a ski jump and a log-house;
there are different scales, different appearances, and different functions. However, algorithmically, a shelf is
identical to one layer of the log-house principle. The same type of detailing group applies to the X, T, L and I-
node. Figure 21 shows the nodes and Figure 22 shows the detailing of the X-node.

Figure 21: Same property descriptions as the log-house, but different design of the details.
Figure 22: Detailing the X-node. The parts are slotted to simplify assembly
6. Discussion and conclusion
This article presented a flexible approach designing structures parametrically. The aim was to investigate
how to apply the benefits of MC while designing site-specific, customer-inclusive and bespoke timber
structures. The main finding is that when abstracting any structure to elements and nodes, most of the
structures can be designed through a similar set of rules.
Architects are trained to design unique projects. Reusing elements from a completed project almost feels
like cheating. Such mindset stands in contrast to the idea of mass customization. However, learning from the
described study and case-projects, one conclusion is particularly prominent: To be able to achieve Mass
Customization in architecture, we must standardize projects. Thanks to parametric modelling, we do not have
to standardize the physical result. Rather, we can standardize the workflow, the process and the interfaces
between the project stakeholders. This paper has exemplified a standardized, still flexible, building system in
the age of parametric architecture.
A structure can be similar in a parametric description, but can vary much in function, form, scale and all
other attributes. What appears as visually and radically different structures from an architectural point of view
is very similar from an algorithmic point of view. The log-house and the shelf exemplify this issue very well. In
contrast, two structures that are functionally similar are fundamentally different from an algorithmic perspective.
With such a view, a truss bridge is more similar to a free-form gridshell than a beam bridge. A tendency in
described case-projects, is that one parametric workflow outputs a much larger design space than in ordinary,
manually drawn CAD-projects. By parametric thinking, we are able to shift from standardized products to
standardized, still scalable, parametric building systems.

The toolkit made in fully parametric design environment allows the designer (architect or engineer) to
observe and control the design in a holistic way. From digital conceptual draft to the production codes, every
important decision can be programmed, followed in real time and changed according to the need of hour. The
toolkit does not introduce such methodology, but simplifies and substantiates a continuous digital workflow
from global shape to digital fabrication and assembly. This by offering a series of tools that reduces technical
complexity of parametrically designing timber structures. Especially the detailing groups based on property
descriptions and the digital subtractive tools have drastically reduced the time needed to set up a parametric
model of a project.
According to R.Aish [9] in computation design: “The designer is no longer directly modelling the building:
instead he develops a graph or script whose execution generates the model”. The timber toolkit presented
here allows the designer to go out from strict BIM design, where everything have to be described “a priori”, to
described by R. Aish definition called computation design. The variety of presented cases done with one
approach and one toolkit is a conclusive evidence of applicability of it in building design and industry.
One can definitely state that the Orkla Bridge project fulfils the idea of mass customization. The first
bridge, Follo Bridge were extremely time-demanding, but designing the Evjen Bridge was much more efficient.
Updating the shape of the bridge required only 60-90 seconds, including a simple FEA, regenerating all details
and not the least outputting BTL-instruction. However, these bridges were relatively similar. How about the
other structures? The already-designed bridges contributed to a more efficient detailing of the water ramp. The
slotted-in metal plates for the detailing algorithm were reused when modeling the foundations for the columns.
They are different shapes but based on the same knowledge-based engineering (KBE)-rules.
The flexibility of a building system is often dependent of the hours spent developing the system. A flexible
and highly reusable parametric building system is lightly not to pay off if thinking one project a head. However,
if developing the building system is considered as an investment in a series of future projects, we strongly
believe that such mindset is economically sustainable. Further, to achieve Mass customization in architecture,
we must continue shifting our mindset to better understand the possibilities of applying parametric tools and
computation.
At last, does the timber toolkit ensure site-specific, customer-inclusive and bespoke timber structures?
The author’s answer is, luckily, that the toolkit does not make great architecture. At best, the toolkit reduces
the amount of work spent on redundant work. However, the hours earned can be spent making better
architecture. Architecture is still made by architects; tools are just tools simplifying the creative process.
References
Aish, R. (2013). First build your tools. Inside Smartgeometry: Expanding the Architectural Possibilities of
Computational Design, 36–49.
Banks, J. (2016). Why BIM is still bankrupting your firm. Shoegnome 9.12.
Burns, W. E. (1986). Computer-Aided Design (CAD). Industrial Education, 75(6), 10–11.
Daaboul, J., Cunha, C. D., Bernard, A., & Laroche, F. (2011). Design for mass customization: Product
variety vs. process variety. CIRP Annals, 60(1), 169–174.
https://doi.org/https://doi.org/10.1016/j.cirp.2011.03.093
Davis, S. M. (1997). Future perfect. Basic Books.
Goldberg, A., & Robson, D. (1983). Smalltalk-80: the language and its implementation. Addison-Wesley
Longman Publishing Co., Inc.
Hundegger, H. (2001). Woodworking machine.
Kaplan, A. M., & Haenlein, M. (2006). Toward a parsimonious definition of traditional and electronic mass
customization. Journal of Product Innovation Management, 23(2), 168–182.
Krymsky, Y. (2015, December 11). The Architecture Software Revolution: From One Size Fits All to DIY.
Retrieved January 18, 2017, from http://www.archdaily.com/778619/the-architecture-software-revolution-from-
one-size-fits-all-to-diy
Kymmell, W. (2007). Building Information Modeling: Planning and Managing Construction Projects with
4D CAD and Simulations (McGraw-Hill Construction Series): Planning and Managing Construction Projects
with 4D CAD and Simulations. McGraw Hill Professional.
McNeel, R. (2009). Rhinoceros 3D. Retrieved Jan, 15.

McNeel, R. (2015). Grasshopper 3D.
Mork, J. H., & Luczkowski, M. (2017). One algorithm, two timber bridges built in Orkdal, Norway. In Forum
Wood Building/Nordic Trondheim 17. Faculty of Architecture and Design Trondheim.
Preisinger, C., & Heimrath, M. (2014). Karamba—A Toolkit for Parametric Structural Design. Structural
Engineering International, 24(2), 217–221. https://doi.org/10.2749/101686614X13830790993483
Rapp, E. J., & Johnson, G. A. (1980). Computer numerical control machine tool. Google Patents.
Retrieved from https://www.google.com/patents/US4199814
Scheurer, F. (2013). Digital craftsmanship: from thinking to modeling to building. Digital Workflows in
Architecture. Birkhäuser, Basel, 110–129.
Schumacher, P. (2008). Parametricism as Style-Parametricist Manifesto,.
Schumacher, Patrik. (2016). Parametricism 2.0: Gearing Up to Impact the Global Built Environment.
Architectural Design, 86(2), 10. https://doi.org/10.1002/ad.2018
STEHLING, H., SCHEURER, F., & ROULIER, J. (2017). BRIDGING THE GAP FROM CAD TO CAM:
CONCEPTS, CAVEATS AND A NEW GRASSHOPPER PLUG-IN. In Fabricate 2014: Negotiating Design &
Making (DGO-Digital original, pp. 52–59). UCL Press. Retrieved from
http://www.jstor.org/stable/j.ctt1tp3c5w.10