Professional Documents
Culture Documents
Assembly-aware design of
masonry shell structures: a
computational approach
Conference Paper
Author(s):
Kao, Gene Ting-Chun ; Körner, Axel; Sonntag, Daniel; Nguyen, Long; Menges, Achim; Knippers, Jan
Publication date:
2017
Permanent link:
https://doi.org/10.3929/ethz-b-000554318
Rights / license:
Creative Commons Attribution 4.0 International
This page was generated automatically upon download from the ETH Zurich Research Collection.
For more information, please consult the Terms of use.
September 25th - 28th, 2017
Proceedings of the IASS Annual Symposium 2017
“Interfaces: architecture.engineering.science”
25 - 28th September, 2017, Hamburg, Germany
architecture . engineering . science IASS 2017 hamburg Annette Bögle, Manfred Grohmann (eds.)
Abstract
This paper proposes a workflow for Assembly-Aware Design (AAD) of masonry shell structures and
introduces an interactive tool in a CAD environment to assist the design process while simulating the
step-by-step assembly of masonry blocks. Thus designers can explore the design space of masonry
shell structures and be aware of structural performance before the assembly phase, at the early design
stage. Masonry shell structures are an old construction technique, which has recently received a lot of
attention due to new computational methods. Even though the form of such a structure is optimised for
structural performance, its incomplete form during construction often requires the support of
falseworks, which can be extensive, costly and time-consuming. To tackle this unsolved problem, we
developed an assembly strategy that significantly reduces the falsework usage while still maintaining
the equilibrium of the incomplete shell at each assembly step. The key idea is to compute a
disassembly strategy inspired by the Jenga game and then reverse it to obtain the actual assembly
sequence of the masonry blocks. Rather than using discrete element methods to predict the structural
behaviour of the masonry blocks, we employed the GPU-based rigid-body dynamic solver from the
engine NVIDIA PhysX, this allows very fast computation speeds while still offering sufficient
accuracy for our purposes. Finally, we verified our method using small-scale 3D printed models.
Keywords: Masonry structure, shell structure, assembly sequence, disassembly sequence, discrete structures, construction
with less falsework, discrete element modelling, game engine, rigid body dynamics, physical simulation.
1. Introduction
Shell structures play an important role both in architecture and engineering due to their aesthetic
qualities and their efficient load bearing behaviour, the latter being the result of a double curved
geometry.
Although only a few masonry shells exist nowadays, there has been a revived interest in masonry shell
structures over the last few years. This is largely due to the introduction of new computational
methods for design and analysis coupled with significant advancement in digital fabrication methods.
Ongoing research such as Panozzo et al. [8] continues to extend the design space of masonry shell
structures. The load bearing behaviour of masonry structures severely constrains the potential design
space, a solution to which is the appearance of new computational design tools. Specifically, the issue
of fabricating double curved masonry shells with individually shaped blocks can be tackled with
modern computational fabrication tools. However little attention has been given so far to the issue of
assembly, and extensive falsework is usually necessary during the construction stage to temporarily
support the unfinished structure. This necessity of falsework is still a major drawback of this
construction technique. Deuss et al. [5] proposed a different approach by using a sparse set of chains
to maintain the stones in equilibrium during construction, while another method mentioned in Fitchen
[6] uses tensioned ropes instead of dense falsework to hold blocks in place. These methods, however,
can still be time consuming and result in extra structures that are required to hold the supporting
elements in place.
This research aims to investigate possible solutions to saving falsework during the design process, by
trying to find assembly sequences, where intermediate construction stages are already in equilibrium
and therefore require only minimal falsework. This requires the development of a digital design
environment, where the structural behaviour of various construction stages can be digitally predicted
and fed back in order to modify the assembly sequence or potentially have influence on overall form
and tessellation. Furthermore, the necessity to investigate large amounts of possible configurations
requires the finding of time-efficient analysis methods for those intermediate construction stages, all
of which are discussed in the following sections.
2. Related works
2
Proceedings of the IASS Annual Symposium 2017
Interfaces: architecture.engineering.science
(a) (b)
In the first experiment, one block A is laid on top of another block B. As shown in figure 1a, d is the
distance between the vertical projection of Block A’s centre of gravity and the edges of Block B, with
d being positive when the vertical projection of the centre of gravity is inside the projected boundaries.
The goal of the experiment is to test if tiling failure is correctly represented. Both methods provide
similar results in the first experiment, and the boxes fall down once d becomes negative, see figure 2a.
In the second experiment the angle between both blocks is progressively increased in order to simulate
sliding failure, see figure 1b. It can also be seen, that both methods provide similar results. In both
experiments the computation using PhysX is approximately 10 times faster compared to 3DEC.
Close to the instability situation the DEM simulation provides higher accuracy by taking deformation
of the building blocks into account. Nevertheless, the results of the RBD method are close enough to
make less time-consuming approximations to filter out a high number of impossible solutions. Once
an assembly sequence is found, DEM can be used for verification.
3
Proceedings of the IASS Annual Symposium 2017
Interfaces: architecture.engineering.science
(a)
(b)
Figure 2: 3DEC (left) and Rhino (right) viewport of two experiments. (a) See figure 1a. (b) See figure1b.
4
Proceedings of the IASS Annual Symposium 2017
Interfaces: architecture.engineering.science
(a) (b)
Figure 4: (a) Algorithm of removing blocks. (b) Comparison of tessellation and stabilities.
5
Proceedings of the IASS Annual Symposium 2017
Interfaces: architecture.engineering.science
The basic block removal algorithm is described in the pseudocode in figure 4a. Whenever
all blocks are stable (V(Bn) = 0)), the block with the biggest displacement from the original
position (the loosest) is removed.
The results of the PhysX simulations of the three tessellations at various stages are shown in
figure 5. The triangular tessellation was the most vulnerable and after several blocks were
removed, the structure collapsed, showing that the triangular pattern is not stable in general.
When the connection is observed in detail, see figure 6a, it can be seen that in the triangular
tessellation, the two blocks support each other by only a point-like connection. In addition,
it can be seen that in general the structure is very sensitive to imperfections at all stages.
While the quadrilateral tessellation is relatively stable, see chart 4b, the removal of certain
blocks leads to instability of entire rows since the integrity of arches within the shell has
been compromised.
Figure 5: Screenshots of three tessellation collapse behaviour. (a) Triangular. (b) Quadrilateral. (c) Polygonal.
On the other hand, in polygonal tessellation, almost all the blocks can be removed one by
one without causing its neighbours to collapse (figure 4c, figure 5c). The polygonal
tessellation allows more interfaces to effectively transfer loads and form arches in more than
one direction, see figure 6c. It can be seen that the polygonal tessellation in most situations
allow for the formation of arches in secondary directions which help stabilize the overall
structure (figure 6c).
6
Proceedings of the IASS Annual Symposium 2017
Interfaces: architecture.engineering.science
disassemble with falsework, i.e. temporarily support the other blocks (figure 7b), disassemble different
items first such as block B (figure 7c), or disassemble several items as subassembly (figure 7d).
Figure 7: Disassemble options. (a) Removing A cause B collapse. (b) Using falsework before removing A. (c)
Alternative removal of B. (d) Remove A and B together.
During the disassembly, the selection of the loose items can also be based on different priority
settings. In figure 8b, all blocks with red colour indicate they are the first priority set of disassemble
items because they have more edges exposed to the open area, which is easier to be removed than grey
areas. The algorithm, which was implemented for the disassembly of items with priority settings, is
shown in figure 8a.
(a) (b)
By following this idea, an assembly sequence can be generated by reversing the disassembly sequence
which results in similar options: assemble with another block, assemble a group of blocks all at once
or assemble blocks using falsework on neighbours, then remove the falsework after the blocks are
stable together.
In the figure 9a, a masonry shell structure with 10 blocks is labelled from A to J. MJBAP disassembly
graph reduces a lot of possibility branches and only considers removing either the loosest block or the
block around the open area. The removal of branches considered unsuitable significantly reduces the
complexity of searching for a feasible assembly sequence among the branches, so the assembly graph
of MJBAP is closer to a linear path instead of a complicated graph with numerous branches.
In the disassembly graph figure 9b, the colour red indicates the subassembly with the loosest item that
can be safely removed. The colour orange indicates the collapse of some blocks after the removal of
an item or multiple items. Square shapes mean the subassembly has more than one item while circle
shapes mean only one item remains in the subassembly.
After the feasible disassembly graph is found, an algorithm simply reverses the graph to generate an
assembly sequence.
7
Proceedings of the IASS Annual Symposium 2017
Interfaces: architecture.engineering.science
(a) (b)
Figure 10: Simple dome disassembly process, simulation and physical model.
In the second result, a simple vault with three open edges is successfully constructed using limited
amount of falsework, which is only needed for the open edges naaC = 40/40 = 100%, and nearly 50%
of falsework is saved compared to the construction methods using common scaffolding elements.
Only 18 blocks used falsework among 58 dynamic blocks. naC = 40/58 = 68.9%, nabC= 32/58 = 55.1%
(only four of them need to be assembled as two sets).
17 ground blocks are fixed among 75 blocks. PhysX parameter settings are similar to the first result:
static friction 0.65, kinematic friction 0.65, density 2000 kg/m3, block thickness 0.295m, structure
span 10m.
The strategy of disassembling an open edges vault is that all blocks on the open edges must be
supported first. Figure 11 shows the numerical and physical model at different steps from disassembly
to assembly.
For the model making some registration notches are used in order to increase the alignment precision.
In our research, we used the same notch approach system used by Deuss, et al. [5]. In the real
construction, even the notches do not transfer the load, the possibilities of some form of joints would
be necessary.
8
Proceedings of the IASS Annual Symposium 2017
Interfaces: architecture.engineering.science
Figure 11: Open edges vault disassembly process, simulation and physical model.
In the last result, a more sophisticated vault geometry as described in Rippmann [10] is shown in
figure 12 to demonstrate the potential usage of AAD workflow and MJBAP with a larger amount of
discrete blocks. PhysX parameters are described here: static friction 1.0, kinematic friction 1.0, density
2000 kg/m3, block thickness 0.295m.
Only 36 open edge blocks use falsework among 175 blocks, this helps the construction of masonry
shell save most of the falsework. NaC = 139/175 = 79.4%, nabC = 109/175 = 62.2%, naaC = 139/139 =
100%.
6. Conclusion:
This research demonstrates the basic knowledge of Assembly-Aware Design’s application for
masonry shell structure. Even though this approach is still a work in progress and in the very early
stages of development, it can potentially assist the designers in reducing extensive falsework during
the construction phase of masonry shell construction.
In future research the following factors should be addressed. First, imperfections due to manufacturing
of the components are currently not considered in the simulations. Although, registration marks such
as notches or spheres can reduce tolerance issues, during the construction phase fabrication and
assembly inaccuracies accumulate to larger deviations. Therefore, scanning and adjusting the
fabrication of each stone after each step could be one possibility to minimize the tolerances of
masonry shells.
Second, safety value and stabilities should be applied in order to construct the structure safely. For
example, after a block has been added to the assembly, we can quickly test its stability by applying a
“disturbance” force, pointing downward with magnitude proportional to the block’s weight, at an
unsupported corner point, and verify if the block still manages to maintain equilibrium
Third, the current approach assumes planar interfaces between the building blocks. By adjusting the
interface geometry in direction or adding interlocking shapes such as wedges while considering the
assembly sequence and direction, the stability can be increased. Fourth, some more accurate
comparisons between simulation and physical model, even with different methods such as Limit
Analysis, can be further explored. Last but not least, a complete AAD workflow can include the TNA
methods with feedback and evaluation methods between all steps to act on all stages of the assembly,
from the global underlying shell geometry, to mesh topology and tessellation and adjustment of
interface geometry.
9
Proceedings of the IASS Annual Symposium 2017
Interfaces: architecture.engineering.science
Acknowledgements
The research was realised as a master thesis as part of the Integrative Technologies and Architectural
Design Research M.Sc. programme (ITECH) at the University of Stuttgart, led by the Institute of
Computational Design (ICD) and the Institute of Building Structures and Structural Design (ITKE).
Special thanks to Dr. Rippmann and Prof. Dr. Block from BRG at ETH Zurich for providing digital
models which were used as a case study.
References
[1] Beyeler L, Bazin JC, Whiting E. A graph-based approach for discovery of stable deconstruction
sequences. In Advances in Architectural Geometry 2014 2015 (pp. 145-157). Springer
International Publishing.
[2] Block PP. Thrust network analysis: exploring three-dimensional equilibrium. PhD [dissertation],
Massachusetts Institute of Technology. 2009.
[3] Cundall PA. Formulation of a three-dimensional distinct element model—Part I. A scheme to
detect and represent contacts in a system composed of many polyhedral blocks. In International
Journal of Rock Mechanics and Mining Sciences & Geomechanics Abstracts 1988 Jun 1 (Vol.
25, No. 3, pp. 107-116). Pergamon.
[4] Deuss M, Deleuran AH, Bouaziz S, Deng B, Piker D, Pauly M. ShapeOp—A Robust and
Extensible Geometric Modelling Paradigm. In Modelling Behaviour 2015 (pp. 505-515).
Springer International Publishing.
[5] Deuss M, Panozzo D, Whiting E, Liu Y, Block P, Sorkine-Hornung O, Pauly M. Assembling
self-supporting structures. ACM Transactions on Graphics (TOG). 2014 Nov 1;33(6):214.
[6] Fitchen J. The construction of Gothic cathedrals: a study of medieval vault erection. University
of Chicago Press; 1981.
[7] Lee S. Backward assembly planning. In Tools for Artificial Intelligence, 1991. TAI'91., Third
International Conference on 1991 Nov 10 (pp. 408-415). IEEE.
[8] Panozzo D, Block P, Sorkine-Hornung O. Designing unreinforced masonry models. ACM
Transactions on Graphics (TOG). 2013 Jul 21;32(4):91.
[9] Rippmann M, Mele TV, Popescu M, Augustynowicz E, Echenagucia TM, Barentin CC, Frick U,
Block P. The Armadillo Vault: Computational design and digital fabrication of a freeform stone
shell. Advances in Architectural Geometry 2016. 2016.
[10] Rippmann M. Funicular Shell Design: Geometric approaches to form finding and fabrication of
discrete funicular structures. PhD [dissertation]. Zurich: ETH Zurich, Department of
Architecture, Feb. 2016.
[11] South M. A Real-Time Physics Simulator for Jenga™. Master's thesis, University of Sheffield,
South Yorkshire, England, May. 2003 May 7.
[12] Tai AS. Design for assembly: a computational approach to construct interlocking wooden frames.
[master’s thesis], Massachusetts Institute of Technology. 2012
[13] Van Mele T, McInerney J, DeJong M, Block P. Physical and computational discrete modeling of
masonry vault collapse. In Proceedings of the 8th international conference on structural analysis
of historical constructions, Wroclaw 2012 Oct.
[14] Wester T. Nature teaching structures. International Journal of Space Structures. 2002 Jun 1;17(2-
3):135-47.
10