See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/385893926

CNC Milling Robotic Arm: Eight torus

Conference Paper · November 2024

CITATIONS READS

0 12

3 authors, including:

Habibe Şenkal Orkan Zeynel Güzelci

Istanbul Technical University Istanbul Technical University
8 PUBLICATIONS 11 CITATIONS 81 PUBLICATIONS 185 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Habibe Şenkal on 17 November 2024.

The user has requested enhancement of the downloaded file.

 CNC Milling Robotic Arm: Eight torus

Habibe Şenkal¹, Sema Alaçam¹, Orkan Zeynel Güzelci¹,²

¹Istanbul Technical University, Istanbul, Türkiye, ²University of Porto, Porto, Portugal

senkal21@itu.edu.tr; alacams@itu.edu.tr; guzelci@itu.edu.tr/oguzelci@arq.up.pt

Abstract. This study focuses on how using a CNC robot arm in producing
mathematical surfaces can expand production boundaries. It investigates the repeated
production of a double-curved eight torus surface defined by mathematical equations
using a CNC robot arm and timber material through two fabrication experiments. The
eight torus is parametrically modeled in Rhino-Grasshopper with the MathForm plugin.
The milling operations for roughing and finishing are simulated using SprutCAM
software. The form is produced from laminated ayous timber using a KUKA KR200
CNC robot arm. The fabrication parameters for both experiments are largely similar.
The fabrication process is investigated based on speed and time, comparing the
simulation and fabrication processes as well as the experiments themselves. The
results highlight the challenges in milling concave curved parts of double-curved forms
and the impact of the assembly method on fabrication time and speed.

Keywords: Robotics, Robot-aided subtractive fabrication, CNC milling, Double-curved

form, Mathematical surface

1 Introduction

Mathematics, a tool used to understand the order of the universe and nature
through numbers (Fuller, 1975; King, 1997), holds significant potential for
designers who strive to comprehend and interpret nature in their form
generation processes. Although pure algebraic expressions are not a familiar
language for designers, computational design and production methods, along
with algorithmic thinking, translate the language of mathematics into a more
comprehensible format for designers, revealing the embedded knowledge of
forms (Kolarevic, 2003; Terzidis, 2006; Báez, 2007; Choma, 2015; Eyce and
Alaçam, 2021). Kotnik (2010) emphasizes the role of the discovery of
computers' ability to simulate natural processes in this transformation, stating
that the incorporation of natural and scientific knowledge into architecture and

995
 the use of digital methods have mathematized architectural discourse. This
relationship can be traced through the development of mathematical
knowledge.
The axioms of Euclidean geometry have been examined from various
perspectives by numerous mathematicians. One of these mathematicians is
Gauss and his proposition that "more than one parallel line can be drawn
through a point outside a given line" (Güner and Çağdaş, 2019) facilitated the
development of non-Euclidean geometries. According to Kotnik (2010), the
use of digital methods in architecture has enabled the integration of these new
geometries such as topology, differential geometry, and algebraic geometry.
This integration allows for the incorporation of properties of mathematical
functions into design processes, such as continuity and smoothness,
differentiation, bifurcation, morphing, and self-organization. Among the non-
Euclidean geometries is algebraic geometry, which is inherently complex due
to its reliance on abstract algebraic expressions and calculations. However,
algebraic geometries can be reproduced through repetition or manipulation
because of their rule-based and orderly structures. In this regard,
computational design and computer-aided design tools are recognized as
effective methods to overcome the complexity constraint and make use of
their advantages of their rule-based properties.
Digital methods not only provide solutions for overcoming design
complexities but also open up new possibilities for addressing challenges in
the manufacturing process. Among these methods, robot-aided fabrication
has been integrated into the field of architecture due to its advantages such
as speed and precision. Through robot-aided fabrication, 3D printing
technology is employed at the architectural scale using materials with various
properties, such as clay, soil, concrete, and steel, based on additive
processes (Yuan et al., 2016; Mense-Korte, 2021; Parkes, 2021; Joosten et
al., 2022:16; Curth et al., 2024). It also enables the production of complex
geometries, including double-curved forms, from materials of varying
hardness, such as metal, stone, wood, plastic, and foam, through subtractive
methods (Burry, 2016; Shaked et al., 2020; Yazar et al., 2023).
In this study, a robot-aided subtractive fabrication workflow was developed
and experimented for the production of double-curved forms, focusing on the
eight torus surface, which is an algebraic surface. This surface was chosen
due to their topological and algebraic complexities, which pose challenges for
conventional design methodologies. Computational design methods were
employed to explore the manufacturing feasibility of this form based on their
mathematical equations.
The research questions to be answered in the study are as follows:
• Is robot-aided subtractive fabrication an effective method for
producing double-curvature forms?
• What are the parameters that affect the fabrication speed and time
of the same form with robot-aided subtractive fabrication?

996
2 Methodology

This chapter details the experiment setup with its assumptions and variables,
form generation process in Rhino-Grasshopper environment using the
MathForm plugin, the fabrication planning and simulation in SprutCAM
software.

2.1 Experiment Setup

Assumptions and variables of experiment setup are listed below:

• Material: The laminated and glued timber blocks are made from
ayous, a tropical tree native to West Africa.
• Table: The table where the material that the robot arm will mill is
assembled.
• Support: A support block is used to elevate the timber mass above
the machining table to prevent the robotic arm from colliding with
the table during milling.
• Strut: Cross-cut timber element used to increase the stabilization
of the support block.
• Form: A three-dimensional and volumetric double-curved surface
created from mathematical equations.
• Size: The boundaries of the bounding box surrounding the form,
as well as the size of the laminated timber used in fabrication.
• Tool: The milling tool used to subtract timber material layer by
layer, depending on its diameter and type.
• Axis: It represents the direction in which the robot arm mills the
material in the three-dimensional Cartesian system, that is the
machining axis.

2.2 Form Generation

The eight torus surface is formed by bending a rectangle along its two short
edges to create an S shape, then joining the short edges at the center of the
S to form an 8. This 8 is then twisted and stretched along a circle at its center.
The resulting toroidal surface, characterized by an eight shaped cross-
section, is a closed loop, single edged, twisted, and non-oriented algebraic
surface (Peterson, 2001). The closed loop signifies that the surface's starting
and ending points are joined, and this loop also serves as the surface's single
edge. The non-oriented nature of the surface means that it is impossible to
distinguish between the interior and exterior sides of the surface.
Traditional methods of manufacturing and manipulating this surface
present challenges. Therefore, algebraic representation was utilized for form
generation using Rhino-Grasshopper, a Computer-Aided Design (CAD)
software. The MathForm plugin, compatible with this software and introduced

997
 by Şenkal et al. (2024), was employed. This plugin accepts input parameters
such as mathematical equations and variable ranges, processes them in the
background, and outputs the equation result as a point cloud. The process of
transforming equations into forms using the MathForm plugin is illustrated in
equation 1, 2 and 3. Finally, to create a surface from the point cloud, the
Surface from Points component of Grasshopper is used, and the generated
point cloud and surface are simultaneously reflected on the Rhino (Figure 1).
The eight torus equation contains a variable coefficient as well as the
unknowns u and v, as presented in equations. Although the equations do not
contain diameter information, the range of values for the u and v allows the
surface to form a closed loop, similar to a circle. Therefore, the value range
for the u and v is defined as u(Ux, Uy) = v(Vx, Vy) = [-π, π]. The a is set to 2
units, which affects the radius of the closed loop. Changes in the coefficient
affect the size of the surface, causing it to grow or shrink without altering the
sectional dimensions of the eight torus.
Fx(u,v) = cos(u) ( a + sin(v)cos(u) - ( (sin(2v)sin(u) ) / 2) (1)
Fy(u,v) = sin(u) sin(v) + ( cos(u)(sin(2v) ) / 2) (2)
Fz(u,v) = sin(u) ( a + sin(v)cos(u) - ( (sin(2v)sin(u) ) / 2) (3)

Figure 1. Generation of eight torus form using MathForm. Black line is a closed
loop, and red dots are points of the surface. Source: Authors, 2024.

2.3 Fabrication Planning and Simulation

Fabrication planning of the form was executed using SprutCAM X Robot,

which is a Computer-Aided Manufacturing (CAM) software specifically
designed for robot-aided subtractive manufacturing. This software provides
predefined operation options with different machining strategies for CNC robot
arms, along with specialized parameters for each operation (SprutCAM Tech
Limited, 2022). The operations involve directing the milling tool along a
specified path to process the material based on defined parameter values.
The production process was planned to proceed with two-stage milling,
including roughing and finishing operations. For the milling process, the
software's "roughing plane" and "5D surfacing" operations were utilized. Some
of the parameters used in these operations are presented in Table 1.

998
 Table 1. Roughing and finishing simulation parameters. Source: Authors, 2024.

Parameters Roughing Finishing

Tool 16 mm Cylindrical mill 16 mm Cylindrical and

Spherical mil

Milling step %50 Ø (8 mm) or %20 Ø (3,2 mm)

%25 Ø (4 mm)

Depth of cut %50 Ø (8 mm) %50 Ø (8 mm)

Tool orientation - Fixed

Tolerance 0.2 mm 0.02 mm

Spindle, Rapid and 200 rev/min(rpm), 10000 200 rev/min(rpm),

Work speed mm/min and 8500 mm/min 8500 mm/min and
3750 mm/min

The production of form, a cylindrical type tool with a diameter of 16mm was
used in the roughing operation, and a spherical type tool with the same
diameter was used in the finishing operation. In Experiment 1, the machining
steps used half the diameter of the roughing tool, while in Experiment 2, this
parameter was used 25% of the tool diameter. The depth of cut, which is the
amount of material the tool removes from the timber in a single pass, is 50%
in both operations. There is no tool orientation option in roughing. This
parameter was fixed in both experiments during finishing, because there are
many collision errors in other modes. The tolerance value is 0.2 mm for
roughing operations and 0.02mm for finishing operations in both experiments.
For roughing operations, the tool was operated at 200 rpm, with feed rates
of 8500 mm/min for cutting and 1000 mm/min for movement. In finishing
operations, these values were 200 rpm for rotation, 3750 mm/min for cutting,
and 8500 mm/min for movement.
After the definitions, production simulations were conducted for both
experiments to be carried out with the robotic arm, following the optimization
of the tool path and the resolution of collision detections. The time-dependent
variation of the simulation results for Experiment 1 is presented in Figure 2,
while only the durations have changed for Experiment 2.
The different planned times for the two experiments are related to the
varying milling step values used in each experiment. As the milling step value
increases, the surface area milled by the tool in a single pass increases,
thereby reducing the milling time. Therefore, in Experiment 1, which used a
larger milling step value, the material was milled with wider intervals and
fewer steps. It was planned to complete the roughing of one surface in 90
minutes, totaling 180 minutes for two surfaces, and the finishing of one
surface in 9 minutes, totaling 15 minutes. In Experiment 2, the planned times
were 300 minutes for roughing and 20 minutes for finishing.

999
 Figure 2. Time-dependent simulation result of Experiment 1 for both operations.
Source: Authors, 2024.

3 Experiments

The fabrication of eight torus was investigated through two experiments.

These experiments utilize robotic arm for fabrication, same size of laminated
ayous block for milling, and same milling axis. However, the method used to
assembly the material is different (Figure 3). The impact of this difference on
the speed and duration of material in robot-aided subtractive manufacturing
processes has been explored. This effect was discussed by comparing
production data with simulation data from the production planning process.

Figure 3. Mathematical equations, computational design and robotic fabrication of

form. Red one is the strut used in Experiment 2. Source: Authors, 2024.

In the experiments, 60x60x20 cm laminated ayous material was milled

using a rail-supported KUKA KR200 robotic arm. In these experiments, a
wooden support block, 50 cm in height, was placed on the machining table,
and the ayous block for forming was fixed onto this support. Thus, the CNC
robotic arm milled the material bidirectionally in the XZ axis without colliding

1000
with the machining table. To safely conduct the milling process and prevent
the milled material from falling, the initial 5 cm height of the ayous block after
mounting was defined as a restricted zone in simulation process where no
milling was performed.
In the roughing operation, after milling one surface of the material in the XZ
axis to half the thickness of the material, the operation was completed, and
the remaining portion of the material was milled from the opposite surface to
outline the form. Following roughing, a finishing operation was applied. In the
finishing operation, in addition to the XZ axis, the form was milled in the XY
axis for more detailed shaping. For the finishing operation, the form was
divided into lower sections and milled gradually. Although the fabrication
planning and fabrication process are basically based on similar principles, the
experiments proceeded differently in execution.

3.1 Roughing

Experiment 1 began the roughing operation at 100% speed as planned.

However, during the milling process, undesirable vibration of the material was
observed, threatening the fabrication safety. As a result, the milling speed had
to be manually reduced using the robot arm's smartpad tool. The vibration
decreased as the speed was reduced. In this experiment, the robot arm
operated at speeds between 100% and 75% for the first surface, and then the
same process continued for the second surface. As the weight of the material
further decreased on the second surface, vibration issues reappeared at 75%
speed. Therefore, the milling speed was reduced to 50%, and towards the
end of the operation, it was lowered to 30% to complete the roughing.

Figure 4. Roughing process for Experiment 1 (first row) and Experiment 2 (second
row). Source: Authors, 2024.

1001
 In Experiment 2, no vibration issues were observed during the roughing
operation. Consequently, the robot arm was operated at 100% speed as
planned. The fabrication stages of the roughing operation for both
experiments are presented in Figure 4.
During the roughing operation, it was observed that the tool holder
approached the form significantly in both experiments, posing a risk to
fabrication safety. This feedback was taken into account during the finishing
process.

3.2 Finishing

To prevent the tool limit issues encountered during rough milling, the finishing
operations were performed on different axes and in stages.
After the roughing operation in Experiment 1, the physical model was
disassembled due to the high workload in the fabrication facility and later
reassembled for the finishing. The fabrication began by milling the upper part
of the form on the XY axis at 50% speed. However, as the operation
progressed, it was determined that the form was not being shaped as planned
due to discrepancies between the coordinates in the simulation and the
physical environment. This error caused the upper part of the form to be
milled further back than intended. The operation was halted, the error was
solved in the CAM environment, and the next milling phase commenced.
During this phase, the lower parts of the form were sequentially milled along
the XZ axis. In one of these phases, despite correcting the surface
coordinates, the tool path was not updated, leading to incorrect programming
of the robot arm. The robot arm began to mill this region much deeper than
required, causing deformation of both the material and the tool. After
identifying this error, the robot arm was manually halted and moved away
from the form. Subsequent milling phases were completed in the finishing
operation.
In Experiment 2, the finishing operation began immediately after the
roughing. The convex curved surface of the form was milled bidirectionally
along the XZ axis. During the operation, when the tool holder moved very
close to the material or posed a risk of deformation due to collision, the robot
arm was manually controlled and moved away from the form, similar to
Experiment 1. After completing the milling on this axis, the upper part of the
form was milled along the XY axis to the extent allowed by the tool length,
completing the finishing.
As with the roughing, no vibration issues were observed during the
finishing in Experiment 2, and the robot arm operated at 100% speed.
However, due to safety risks encountered during the milling, the operation
was halted as needed, and the robot arm was moved away from the surface.
This affected the fabrication time. The physical models obtained at the end of
the fabrications for Experiment 1 and Experiment 2 are presented in Figure 5.

1002
 Figure 5. Finishing process for Experiment 1 (first row) and Experiment 2 (second
row). Source: Authors, 2024.

4 Results

The speed and duration based results of the experiments are presented in
Table 2.
The results of the milling operations are as follows:
• The roughing operation time in Experiment 1 was planned as 180
minutes. However, manually reducing the speed of the robot arm
during the roughing doubled the planned duration (Figure 6).
• In Experiment 2, the operation was carried out at the planned
speed, and the production time was 300 minutes as planned in the
simulation. This longer duration was due to the milling step value
(Figure 6).
• The finishing operation time in Experiment 1 was planned as a
total of 15 minutes in the simulation. However, the vibration issue,
operating the robot arm at 50% speed instead of 100%, manual
control of the robot in case of errors, and skipping some milling
steps affected the operation time. Therefore, the finishing took 20
minutes (Figure 6).
• In Experiment 2, the finishing operation time was also performed
at the planned speed. However, manual intervention to move the
robot arm away from the form when the tool holder approached
undesirably affected the time, extending the time from 18 minutes
to 24 minutes (Figure 6).

1003
 Table 2. Fabrication process results. Source: Authors, 2024.

Simulation Fabrication
Roughing Finishing Roughing Finishing
Axis
Experiment Axis Operation Operation
and Speed Time Speed Time
and and Time and Time
Angl (%) (min) (%) (min)
Angle (min) (min)
e
First side
Experiment Roughing 5D 100-75 330
(100
1 Plane XY Surfacing Second 50 20
and
XZ 180 and 15 side
230)
90 XZ 50-30
Experiment Roughing 90 5D
2 Plane Surfacing 100 300 100 24
300 18

Figure 6. Comparison of experiments with simulation and fabrication data of

roughing and finishing operations based on speed and time. Source: Authors, 2024.

1004
5 Conclusions

This study investigated robot-aided subtractive fabrication processes for

timber through experiments. Experiments show that using strut to stabilize the
material on a support block, which offers the advantage of bidirectionally
milling, is important for overcoming the issues of power and vibration created
by the robot arm during milling. This approach positively affects both
fabrication speed and time. These results also point to the limitations of the
study.
The properties of the milling tools used, such as diameter and length, and
the absence of additional tools like positioner (rotary table) that facilitate
fabrication by allowing material movement, were identified as constraints that
limited detailed milling of the form's concave areas. Future studies could
improve upon this work by examining the effects of different milling tools and
additional tools on the fabrication process. Additionally, investigating the
material responses with different types of timber.
By offering a repetitive fabrication process and examining the parameters
that influence fabrication, this study contributes to identifying the factors
affecting the production of double-curved forms from timber using robot-aided
subtractive methods.

References
Báez, R. Z. (2007). A Proposal for the Classification of Mathematical Sculpture. In
Bridges Donostia: Mathematics, Music, Art, Architecture, Culture, 67-74.
Burry, M. (2016). Robots at the Sagrada Família Basilica: a brief history of robotised
stone-cutting. Robotic Fabrication in Architecture, Art and Design 2016, 2-15.
Choma, J. (2015). Morphing: A Guide to Mathematical Transformations for Architects,
Designers. Hachette UK.
Curth, A., Pearl, N., Castro-Salazar, A., Mueller, C., and Sass, L. (2024). 3D printing
earth: Local, circular material processing, fabrication methods, and Life Cycle
Assessment. Construction and Building Materials, 421, 135714.
Eyce, N. and Alaçam, S. (2021). Mathematical Analysis of Architectural Drawing
through Fourier Series. Proceedings of XV. Mimarlıkta Sayısal Tasarım Ulusal
Sempozyumu, 2-13. Istanbul, Türkiye: Istanbul Technical University.
Fuller, R. B. (1975). Synergetics: The Geometry of Thinking. Macmillan Publishing
Company, New York.
Güner, Y. and Çağdaş, G. (2019). A Form Finding Approach With Triply Periodic
Minimal Surfaces. Journal of Computational Design, 1(1), 35-54.
Joosten, S., Ren, S., van Horn, P. and Vola, M. (2022). Printing the built environment.
The Arup Journal, 1, 14-22.
King, J. P. (1997). Matematik sanatı (N. Arık Çev.). Tübitak.

1005
 Kolarevic, B. (2003). Architecture in the digital age. Design and Manufacturing. Nueva
York-Londres: Spon Press-Taylor & Francis Group.
Kotnik, T. (2010). Digital architectural design as exploration of computable functions.
International journal of architectural computing, 8(1), 1-16.
Mense-Korte. (2021). First 3D Residential House in Germany, https://mense-
korte.de/en/3d-printed-house/
Parkes, J. (2021). Tecla house 3D-printed from locally sourced clay,
https://www.dezeen.com/202.1/04/23/mario-cucinella-architectswasp-3d-printed-
housing/
Shaked, T., Bar-Sinai, K.L. and Sprecher, A. (2020). Craft to site. Construction
Robotics, 4, 141–150.
SprutCAM Tech Limited, (2022). SprutCAM X user manual (version 16.2).
https://sprutcam.com/pdf/2022/SprutCAM%20X%20User%20Manual-v16.pdf
Şenkal, H., Alaçam S., and Güzelci, O. Z. (2024). Inspired by Mathematics: MathForm
Plugin. Proceedings of XVIII. Mimarlıkta Sayısal Tasarım Ulusal Sempozyumu,
481-501. Balıkesir, Türkiye: Balıkesir University.
Terzidis, K. (2006). Algorithmic architecture. Routledge.
Yazar, T., Oral-Karakoç, H., Gündüz, G., and Yabanigül, M. N. (2023). Integrating
Robotic Fabrication into the Basic Design Studio. Nexus Network Journal, 25(4),
999-1013.
Yuan, P. F., Meng, H., Yu, L., and Zhang, L. (2016). Robotic multi-dimensional printing
based on structural performance. Robotic fabrication in architecture, art and
design, 92-105.

1006

View publication stats