Content uploaded by Niloufar Emami

Author content

All content in this area was uploaded by Niloufar Emami on Sep 26, 2017

Content may be subject to copyright.

Proceedings of the IASS Annual Symposium 2017

“Interfaces: architecture . engineering . science”

September 25 - 28th, 2017, Hamburg, Germany

Annette Bögle, Manfred Grohmann (eds.)

Copyright © 2017 by <Niloufar Emami, Harry Giles, Peter von Buelow>

Published by the International Association for Shell and Spatial Structures (IASS) with permission.

Continuous to discrete: computational performative design and

search of shell structures

Niloufar EMAMI*, Harry GILESa, Peter VON BUELOWb

*, a, b University of Michigan

Taubman College of Architecture and Urban Planning, 2000 Bonisteel Blvd., Ann Arbor, MI 48109, USA

nemami@umich.edu

Abstract

This research assesses structural and daylighting performance of perforated shell structures. By

employing computational design tools and performance assessment methodologies, an array of

generated topologies of perforated shell structures spanning the two extremes is studied. These

generated topologies are coupled with structural and daylighting performance criteria to allow a

performance-oriented exploration of the design space. ParaGen is used to automate the cycle of

generation and evaluation. ParaGen is a method that uses a genetic algorithm (GA) to search for well

performing geometric solutions to architectural engineering problems. By using ParaGen, the

quantitative performance results are stored in a SQL database, accessible through an online website.

The significance of this study is twofold: first, it studies a spectrum of generated forms of well-

established structural typologies with perforations; and second, it couples structural and daylighting

performance with geometry towards computational performance-based design and search of well

performing alternatives. The results of the study contribute to the area of computational design, multi-

objective design exploration as well as shell structures.

Keywords: Computational design, computational search, shell structures, concrete shells, topology, structural performance,

daylighting performance

1. Introduction

Robert le Ricolais’s influential structural engineering work from the 1960’s was guided by his belief

that “the art of structure is where to put holes”(Pennsylvania [14] ). The paradox is that the removal of

material that weakens a structure also makes it lighter, thus makes it more efficient. In the case of

continuous shell structures, having holes does not stop the thin curved surface from being a shell. But

the perforations in a continuous shell do provide an opportunity for introducing daylight into the

space. On the other hand, grid shell structures are formed by intersecting members which provide the

opportunity to design openings which admit daylight into the space. A colander and a sieve are the

commonly used analogies to compare continuous shells and grid shells. Where a colander has a

continuous surface with a relatively small areas removed, a sieve is a surface made from a large

number of initially straight wires which are woven into a flat sheet and then bent into a hemisphere

(Adriaenssens et al.[1]). From a topological standpoint, a continuous shell can be perforated until it

becomes a grid shell. Therefore, the array of shell topologies between a continuous and a grid shell is

certainly worth investigation.

An instance of a designed and built perforated thin shell structure is the roof of the Fronton Recoletos

(1935), created by Eduardo Torroja (1899-1961), the brilliant Spanish engineer. The shell structure

Proceedings of the IASS Annual Symposium 2017

Interfaces: architecture.engineering.science

2

was destroyed during the Spanish Civil War (Mungan and Abel [11]), but once covered a 55 by 32.5

meter space designed for a ball game. The perforations allowed for natural daylight. A thin concrete

shell was designed, and in those areas where skylights were needed, the shell was replaced by a

triangulated structure designed for the insertion of glass panels (Figure1) (Lozano-Galant and Payá-

Zaforteza [10]).

Figure 1: Thin shell with triangulated openings designed for Fronton Recoletos

Looking at the history of shell structures in general and from the beginning of the 20th century,

designers started to see the potentials offered by the plastic qualities of liquid concrete. Kirovski et al.

(2014) points out that the first wave of the era of thin shells from 1922 to 1965 is regarded as the

golden age of thin shell structures (Krivoshapko et al. [9]). Remo Pedreschi (2008) describes that the

majority of the designs used concrete as a ‘tectonic compact’, where its qualities go beyond purely

structural to become shelter and skin. According to Pedreschi, Angerer was the person who proposed

“surface structures” as a new classification of these structures to differentiate them from the historical

classification of the solid or the skeleton and skin (Pedreschi [13]). Giles (2005) describes shells as

more than a general surface that describes three-dimensional space as a continuum. “The skin of a

shell has rigidity, therefore it is more than a simple poly-surface graphic, it is a surface that possesses

intelligence based on material and geometric properties, which could include texture, transparency and

climatic performance” (Giles [7]).

However, during the last decades, and according to Jörg Schlaich, concrete shells lost more ground to

cable nets, textile membranes and steel grids (Adriaenssens et al. [1]). Sasaki lists the reasons for

reduced construction of reinforced concrete structures nowadays, as “scarcity of skilled workers,

rising prices for formwork, cost and schedule challenges, the inefficiency of on-site fabrication, large

deformations, deteriorations in the concrete, and a transition to steel grid shell to satisfy new

demands” (Sasaki [17]).

Recently, and in the half-century following the decline of concrete shell construction, there is a

renewed interest in these structures. The reason is threefold. Firstly, there is a rise of a new trend in

architecture towards expressing informal three-dimensional shapes that possess free, complex, and

organic characteristics towards a contemporary shell design. Secondly, material developments lead to

availability of lightweight, thin reinforcement materials for concrete on one hand, and more efficient

concrete mixes on the other hand. And thirdly, CAM (Computer Aided Manufacturing) methods have

evolved as an aid in constructing these complex forms.

This research builds on the increased interest in shell forms and in particular free form shell structures

that may be optimized on functional performance. Performance-based design, in which building

performance becomes a guiding design principle, emerged as a design paradigm due to the emphasis

on sustainability as a defining socio-economic issue, and due to the recent developments in technology

and cultural theory (Koleravic [8]). In performance-based design, the emphasis shifts from the

“product” to the “process” of form generation based on performative strategies of design. Simulation

Proceedings of the IASS Annual Symposium 2017

Interfaces: architecture.engineering.science

3

is one of the influential concepts of performance-based design for analysis, which can also redefine

“performative design” when integrated with generation (Oxman and Oxman [12]). Past researchers

have employed computational design strategies to optimize shell structures regarding their structural

performance (Pugnale and Sassone 15]). Assessing the structural and daylighting performance of

perforated shells simultaneously has also been conducted on a case study (Emami [4]). However, past

research has not demonstrated a comparison between the performances of the two ends of the

continuum of shell structures, nor the topologies generated in between. This research study is focused

on the performative design and exploration of perforated shell structures. It investigates a spectrum of

generated forms of shell typologies, considering their coupled structural and daylighting performance

criteria.

2. Methodology

A parametric model is built to demonstrate the transition of a dome into a vault and then a saddle.

Change in the curvature of the edge curves allows the parametric model to generate a range of

geometries, with positive, zero and negative Gaussian curvature (Figure. 2). According to

Blaauwendraad et al (2014), if point A is defined on a smooth surface, there will be a tangent plane at

that point, where the normal to the surface is defined to be normal to the tangent plane. There are an

infinitely many plane curves at point A, but one of them has a minimum curvature and another has a

maximum curvature. These two plane curves are called the “principal sections”, and their curvatures

are called the “principal curvatures” denoted by k1 and k2. The product of the two principal curvatures,

kg = k1 * k2 , is called the Gaussian curvature of the surface. The Gaussian curvature can be positive,

negative or zero. If it is positive, k1 and k2 are having the same sign and the resultant surface is

synclastic or a dome-shaped surface. If the Gaussian curvature is negative, k1 and k2 have opposite

signs, and the surface is anticlastic or a saddle-shaped surface. If either of k1 or k2 is zero, their product

is zero thus the surface has a single curvature or is vault-shaped (Blaauwendraad and Hoefakker [2]).

Figure 2: synclastic, single curvature, and anticlastic forms with a positive, zero and negative Gaussian curvature

The parametric model also demonstrates the transition of a continuous shell into a grid shell.

Grasshopper, a visual programming environment integrated with Rhino NURBS modeling software

has been employed as the modeling tool. Once the edge curves are generated, a surface is constructed

and RhinoMembrane plugin in Grasshopper is used for membrane form finding. Afterward,

Weaverbird, a plugin for Grasshopper, is used for triangulating the mesh. The mapped triangulation is

the basis for subtracting perforations. A python algorithm systematically removes apertures from the

shell and makes it perforated. The geometry generation workflow is demonstrated in Figure 3.

Figure 3: The steps for generating the geometry in Rhino

Proceedings of the IASS Annual Symposium 2017

Interfaces: architecture.engineering.science

4

Figure 4: The variables that change the parametric geometry in Grasshopper and with related plugins

There are a total of 8 variables that are implemented in the parametric model that drive the generated

geometry. Five variables change the global geometry by changing the boundary curves, and three

variables change the local geometry by defining the location and percentage of perforations as well as

changing the shell thickness (Figure 4).

The strategy for subtracting the apertures from the mapped triangulation is twofold. Firstly, a margin

is isolated along the outer boundary of the mesh where no meshes are subtracted. This approach is

essential for later structural analysis that requires a solid edge beam. Secondly, the python algorithm

divides the remaining triangulated meshes into nine zones. Later, the algorithm chooses any number of

zones and combines them together (equation 1). In each combination, the order of zones doesn’t

matter and the combinations with same items in different orders are not allowed to repeat. As an

example, if only one zone is to be chosen, stated as “choose 1 from 9”, there are nine possible

combinations including “zone 1 or zone 2 or… zone 9” (equation 2), whereas if nine zones are to be

chosen, there is only one possible combination including “zone 1 and zone 2 and … zone 9” (equation

3). The number of total zone combinations are calculated as 511 (equation 4). In the extreme case

where all apertures are subtracted, an open grid mesh is retrieved.

))!(!( !

),( rnr n

rnC

(1)

9

))!19(!1( !9

)1,9(

C

(2)

1

))!99(!9( !9

)9,9(

C

(3)

511)9,9()8,9()7,9()6,9()5,9()4,9()3,9()2,9()1,9()0,9( CCCCCCCCCC

(4)

Once any number of zones are combined, another python algorithm maps a subtraction pattern onto

the grid in order to control subtractions at a micro level. Not all the apertures of the selected zone are

subtracted, but a 25%, 50%, 75% or 100% perforation ratio facing various orientations can be

selected.

Proceedings of the IASS Annual Symposium 2017

Interfaces: architecture.engineering.science

5

After the geometry is generated, the structural, daylighting and energy performance are assessed.

Karamba, DIVA and Archsim in Grasshopper are used for structural, daylighting and energy

performance assessments respectively. Karamba is a Finite Element Method (FEM) simulation engine,

whereas DIVA and Archsim use Radiance and Energy Plus engines respectively. Ghowl, a plugin for

Grasshoper, is used to read and write the input and output data to an excel file. ParaGen is used to

automate the cycle of generation and evaluation by using AutoHotkey to script the sequence of steps.

ParaGen is a method that uses a genetic algorithm (GA) to search for well performing geometric

solutions to architectural engineering problems (von Buelow [18]). By using ParaGen, the quantitative

performance results as well as the qualitative renderings of the solutions are stored in a SQL database,

accessible through an online website. The workflow is illustrated in Figure 5.

Figure 5: The automated ParaGen workflow

2.1. Structural analysis

Karamba is used for structural analysis. The corners of the shell are chamfered before the supports are

defined. This reduces formation of local stresses at the support and transfers load to the ground on a

wider support. The movement and rotation of the supports is restricted in x, y and z directions.

Concrete is used for the shell as the main material. The material properties are given in Table 1.

Table 1: material properties

Elastic Modulus (E)

[KN/cm2]

Shear Modulus (G)

[KN/cm2]

Specific weight

(Gamma) [KN/m3]

Coefficient of thermal

expansion (alpha T)

[1/Cº]

Yield Strength (fy)

[KN/cm2]

2600

1080

23.52

0.00001

3

A uniform load of 1 KN/m2 is applied to the shell in the negative z direction (equation 5). This load is

adjusted once the shell is perforated. The equation for adjusting the load is described below.

]/[1 2

mKN

Area

Area

Load

Shellperforated

solidShell

(5)

As an instance, a 1 KN/m2 load is applied to a solid shell of any shape, but if the same shell is 50%

perforated, then the load will be doubled (equation 6). The load adjustment is necessary since the

glazing are not loaded, towards comparing different shells under the same loading conditions.

]/[2]/[1*

5.01

%50 22 mKNmKNShellPerforatedForLoad

(6)

Dead load is also calculated and applied to the shell. The weight of the shell is calculated based on the

solid parts only whereas holes are simply zero weight. In other words, the weight of the glazing on the

shell is not included in the dead load, which gives an approximation of the self-weight of the shells.

Proceedings of the IASS Annual Symposium 2017

Interfaces: architecture.engineering.science

6

Once the material, supports and load are applied to the shell, the maximum displacement and the von

Mises stress levels are analyzed in Karamba. The aforementioned steps are illustrated in Figure 6.

Figure 6: boundary conditions and structural analysis

It should be noted that although maximum displacement is calculated for serviceability conditions,

maximum von Mises is not considered for the strength criteria. Instead, a proportional statistical

distribution analysis is carried out using the von Mises stress levels of all the nodes, where only von

Mises stress at the third quartile (Q3) is considered. In statistics, the quartiles of a ranked set of data

are described as three points that divide the data set into four equal groups. The third quartile is the

middle value between the median and the highest value of the data, meaning that 75% of the data lie

below it. This approach acknowledges the fact that the nodes that are subject to extreme stresses

significantly higher compared to the majority of the remaining nodes are not the base line for

designing the shell. It averages out the maximum stresses, assuming that a degree of redistribution will

take place in reality, whereas the analysis software tends to predict extremely high stresses at points of

discontinuity, which is also dependent on the mesh configuration used by the analysis algorithm.

These highly localized extreme stresses are also considered local effects that do not impact on the

sizing criteria for the shell as a whole.

2.2. Daylighting and energy simulations

DIVA for Rhino and Archsim are used for daylighting and energy analysis respectively. A 50 m by 60

m floor plan is consistent among all shell typologies and topologies. A 1 m by 1 m horizontal grid is

defined at the height of 0.9 meters from the ground to measure daylight illumination. The surrounding

walls, floor, ceiling and glazing are defined as a 35% reflectance Outside Facade, 20% reflectance

Generic Floor, 80% reflectance Generic Ceiling and 20% reflectance Generic Translucent Panel. It

should be noted that there are no external nor internal shading devices modeled, therefore, the glazing

is set to translucent to mitigate the effect of glare while providing natural daylight for the space. The

Radiance parameters are set as default which are -aa .15 -ab 2 -ad 512 -ar 256 -as 128 -dr 2 -ds .2 -lr 6

-lw .004 -dj 0 -lr 6 -sj 1 -st 0.15. The location is set to Boston (42.36º N, 71.05º W) and the weather

data for Boston Logan Airport is downloaded from the Energy Plus website

i

.

Once the daylighting analysis is completed, a daylighting schedule is retrieved which feeds into the

energy analysis setup. In other words, the times of the year where the required light levels are being

met by daylight alone will be excluded from the energy analysis schedule which ultimately reduce the

electricity energy needed for lighting. Custom concrete materials are added to the Archsim library for

various concrete thicknesses which result in different U-values and thus differing heat transfer rates. In

the energy analysis set up, the floor and walls are considered adiabatic, thus the shell roof is the only

surface that transfers heat. This strategy will eliminate the difference between the surface areas of

different walls due to various alternatively formed shell curvatures (Figure 7).

Figure 7: different shell curvatures affect the surface area of the surrounding walls, thus the walls are modeled adiabatic

Proceedings of the IASS Annual Symposium 2017

Interfaces: architecture.engineering.science

7

According to the three performance categories expressed as daylight availability, visual comfort, and

thermal loads (Reinhart and Wienold [16]) for environmental analysis, the metrics presented in Table

2 are selected for daylighting availability, visual comfort performance and energy assessment. It

should be noted that the target required light level of 300 lux has been selected. In addition, since the

annual Daylight Glare Probability (DGP) simulation is computationally expensive, a point-in-time

DGP is calculated for noon at June 21st.

Table 2: Metrics for assessing the daylighting, glare and energy loads

Criteria

Metric

sDA300/50%>50%

Spatial Daylight Autonomy of 300 lux for at least 50% of the year should be

met by 50% of the surface area

DAoversupply <= 5% of the year

The maximum amount of DA is 10 times more than the target daylight:

DAmax= 10 DA = 10 300 lux = 3000 lux.3000 lux is considered as

oversupply and should not occur more than 5% of the year.

DGP < 0.04

Daylight Glare Probability should not be intolerable or disturbing

[Kw/hr]

Lighting, cooling and heating energy loads

3. Multi-objective design space exploration: coupled analysis

The design space can be explored interactively through a webpage that shows the renderings of the

alternatives (Figure 8). Each solution can be scrutinized more by clicking on it to read its detailed

design parameters and performance criteria in both structural and daylighting discipline. Design space

exploration provides the opportunity to compare multiple solutions together based on the daylighting

and structural performance (Figure 9). Therefore, a design option which is good at both daylighting

and structural performance can be recognized as opposed to a solution which performs well in only

one discipline. In order to compare a greater number of alternatives, plotting the Pareto front graphs is

more promising as it shows a better landscape of the alternatives’ performance.

Figure 8: ParaGen’s interactive website that assists the designer to explore the design space

Pareto fronts can be plotted in only one discipline: as an example, weight versus displacement is

plotted to find the well-performing alternatives that have a low weight as well as meeting the

serviceability criteria. Design space exploration can be further filtered, in order to focus on a specific

shell typology. For example, one might want to look at the weight versus maximum displacemnet

limited specifically to a dome-shaped, vault shaped or a saddle shaped shell (Figure 10). Exploration

Proceedings of the IASS Annual Symposium 2017

Interfaces: architecture.engineering.science

8

can be further narrowed, by looking at a specific thickness of a specific shell typology. As an example,

one might want to compare the 10 cm thick saddle-shaped shells to the remaining saddle shaped shells

(Figure 11).

Figure 9: Coupled analysis regarding daylighting and structural performance

Figure 10: Interactive plots of weight versus displacement for specific shell typologies (from left to right: a dome, vault and saddle)

Figure 11: Exploring saddle-shaped shells based on their weight-displacemnet performance. The (left) 10 cm thick shells and

(right) 40 cm thick shells are presented as red dots

Pareto fronts can also be plotted based on both structural and daylighting disicpline. Since the metrics

for measuring different disciplinary performance criteria are different, a set of selected variables have

Proceedings of the IASS Annual Symposium 2017

Interfaces: architecture.engineering.science

9

been transferred to cost: material cost tied to the weight of the structure, and operational cost tied to

the lighting, heating and cooling energy load when an alternative is used to cover a courtyard. This

provides a common ground for conducting multi-disciplinary multi-objective analysis (Figure 12). The

parallel bands of points shown in this plot represent different shell thicknesses ranging from 10 to 40

cm. It should also be noted that operational cost is calculated over one year, and if it is calculated over

a 10, 20 or 30 year period, the plot scale will change.

Figure 12: Material cost versus operating cost

4. Discussion

Computer scientists have been seeking ways to progress from single objective problems to multi-

objective problems. Based on the work of French-Italian economist and sociologist, Vifredo Pareto,

currently Pareto-optimization is the most widely used multi-objective search technique for problems

with conflicting parameters (Diwekar [3]). An algorithm needs to be used to determine a

representation of the Pareto optimal set. Genetic algorithms (GAs), a posteriori algorithm, is used in

ParaGen in order to generate solutions from an initially randomly-generated pool of alternatives. A

complete and more detailed description of ParaGen using a Non-Destructive Dynamic Population

Genetic Algorithm (NDDP GA), along with a comparison between two selection methods namely “a

multi objective selection using SQL queries” with “selection through a Pareto set” is presented in

another paper [19]. Interactive design space exploration provides the opportunity of exploring design

alternatives regarding various performance criteria. There can be multiple objectives within one

discipline to be defined as the acceptable performance, or multiple objectives within multiple

disciplines can be set. This is research in progress that aims to study the relation between the structural

and daylighting performance of perforated shell structures spanning a range of geometries and

topologies. The results will contribute to the area of computational search, multi-objective, multi-

disciplinary performance assessment, as well as shell structures.

5. Conclusion

From a topological standpoint, continuous and grid shells are two ends of one continuum: as more

holes are added to the surface of the shell, force flow is constantly changed and redirected, until it is

Proceedings of the IASS Annual Symposium 2017

Interfaces: architecture.engineering.science

10

channeled through the discrete members of the grid shell. In this topological transition, structural

performance, including weigh, maximum deflection and maximum von Mises Stress, are in

continuous transition. From another point of view, the amount and distribution of incoming daylight is

in fluctuation. This paper studied the complex interwoven dynamics of form, structural and

daylighting performance. The ParaGen framework is used to couple the two disciplines, by linking

together any commercial or public domain simulation software to accomplish the performance

analysis. In this way, both structural and daylighting performance values are available to assess the

performance of each solution. Since the metrics for measuring different disciplinary performance

criteria are different, a set of selected variables have been transferred to cost: material cost tied to the

weight of the structure, and operational cost tied to the lighting, heating and cooling energy load when

an alternative is used to cover a courtyard. This provides a common ground for conducting multi-

disciplinary multi-objective analysis. This is a research in progress, and the next step is to conduct

sensitivity analysis to understand the effect of design parameters on the performance criteria.

References

[1] Adriaenssens S., et al. Shell Structures for Architecture. Routledge, 2014.

[2] Blaauwendraad J., Hoefakker J., Structural Shell Analysis. Vol. 200. N.p., 2014.

[3] Diwekar U., Introduction to Applied Optimization. 2nd ed. Springer, 2008.

[4] Emami N., Structure and Daylighting Performance Comparisons of Heinz Isler’s Roof Shell Based on

Variations in Parametrically Derived Multi Opening Topologies, in IASS 2015. Future Visions Amsterdam,

2015

[5] Evins R., et al., Multi-Objective Design Optimisation: Getting More for Less, in Proceedings of the ICE -

Civil Engineering 2012,5-10

[6] Flager F., et al., Multidisciplinary Process Integration and Design Optimization of a Classroom Building.

Electronic Journal of Information Technology in Construction, 2009: 595–612

[7] Giles H., On Shells and Skins. ACS, 2005: 668–676

[8] Koleravic B., Computing the Performance in Theories of the Digital in Architecture. Ed. Rivka Oxman and

Robert Oxman. Routledge, 2014: 103–111

[9] Krivoshapko S., et al., Chronology Of Erection Of The Earliest Reinforced Concrete Shells. IJRRAS, 2014:

95–108.

[10] Lozano-Galant J., Payá-Zaforteza I., Structural Analysis of Eduardo Torroja’s Frontón de Recoletos’ Roof.

Engineering Structures 33.3 (2011): 843–854.

[11] Mungan I., Abel J., Fifty Years of Progress for Shell and Spatial Structures. Madrid, Spain: IASS

(International Association for Shell and Spatial Structures), 2011

[12] Oxman R., Oxman R., “Performance/ Generation: From Analysis to Informed Synthesis.” Theories of the

Digital in Architecture. Ed. Rivka Oxman and Robert Oxman. Routledge, 2014. 97–102.

[13] Pedreschi R., “Form, Force and Structure.” Architectural Design 78.2 (2008): 12–19.

[14] Pennsylvania, University of. Visions and Paradox: An Exhibition of the Work of Robert Le Ricolais. N.p.,

1996

[15] Pugnale A., Sassone M., “Morphogenesis and Structural Optimization of Shell Structures With the Aid of

a Genetic Algorithm.” IASS 48.3 (2007): 161–166

[16] Reinhart C., Wienold J., “The Daylighting Dashboard - A Simulation-Based Design Analysis for Daylit

Spaces.” Building and Environment 46.2 (2011): 386–396

[17] Sasaki M., “Structural Design of Free-Curved RC Shells.” Shell Structures for Architecture. N.p., 2014.

259–270

[18] von Buelow P., “ParaGen: Performative Exploration of Generative Systems.” Journal of the International

Association for Shell and Spatial Structures 53.4 (2012): 271–284

[19] von Buelow P, “Choosing parents to produce better preforming children: a comparison of selection

methods used for evolutionary search”, in IASS 2017. Interfaces: architecture. engineering. science, Bögle

A, Grohmann M (eds.), Hamburg, 2017