Content uploaded by Luisa Caldas

Author content

All content in this area was uploaded by Luisa Caldas on Feb 06, 2018

Content may be subject to copyright.

Automation of CAD models to BEM models for performance based goal-oriented

design methods

Luis Santosa†

Simon Schleicher

Luisa Caldas

UC Berkeley

College of Environmental Design

Center for the Built Environment

luis_sds82@berkeley.edu

UC Berkeley

College of Environmental Design

Department of Architecture

simon_s@berkeley.edu

UC Berkeley

College of Environmental Design

Department of Architecture

lcaldas@berkeley.edu

aCorresponding author.†Current address: Department of Architecture. University of California,

Berkeley, 232 Wurster Hall #1800, Berkeley, CA 94720-1800, USA.

Abstract

This work presents a new methodology to automate the derivation of Building Energy Models

(BEMs) from complex 3D Computer-Aided Design (CAD) geometry. The goal is to combine current

parametric modeling, digital fabrication, and computer graphics techniques to automatically generate

the geometric input of an energy model from any digital 3D model of a building. Such automation

facilitates the use and implementation of goal-oriented design methods that integrate energy

performance with other types of building performance models. In this work, mesh planarization

algorithms, which are currently used in computer graphics and in digital fabrication methods, are used

and adapted to automate and optimize the parsing of non-planar surfaces to EnergyPlus (E+), a

popular BEM engine. The proposed methodology facilitates the modeling of thermal zones with

double-curved envelopes, which is a time-consuming task that typically requires a high level of

expertise from the energy modeler. The proposed single, streamlined workflow generates digital

models that are suitable for both energy optimization and digital fabrication, thereby facilitating the

integration of two parallel design procedures at the core of an architectural design process. Through

this workflow, a single CAD model generates solutions that are energy efficient and feasibly

fabricated using digital techniques. This goal-oriented design workflow is applied in the study of

fritting pattern densities for three complex double-curved building geometries.

Keywords: generative design systems, goal-oriented design, energy optimization, digital interfaces

and fabrication

1. Introduction

The recent rise of parametric, procedural, and algorithmic design processes may be partially attributed

to the development of new domain-specific programming languages for preexisting 3D modeling

software, such as Grasshopper for Rhinoceros [1, 2] and Dynamo for Revit [3]. These processes

empower designers to more easily iterate and more deeply explore the resultant solution space for a

given design; however, studies show that human beings have limited abilities to assess large

alternative sets [4]. Thus, it is common to complement parametric and/or procedural architectural

design with performance-based design using simulation software, which provides information about

the current design’s performance at any given point. Examples of such digital simulation tools

include: Radiance [5], a ray tracing-based software used for rendering and lighting, EnergyPlus [6], a

whole building energy simulation program, and SAP2000 [7], a simulation program for structural

analysis. However, the parallel use of both parametric and performance-based methods can lead to a

tiresome iterative cycle of model-simulate-evaluate-remodel. The inefficiencies of this cycle may

limit the depth of exploration within the design solution space.

Different types of Performance-based Generative Design Systems (PGDSs) have been proposed to

overcome these limitations. These systems directly apply the goal-oriented design method, in which

the designer establishes performance goals for a given design and allows the system to automatically

search that design’s solution space for the solution that best meets the desired objectives. PGDSs for

the built environment integrate three different modules: (i) a generative or parametric model that is

able to generate a multitude of design alternatives from a set of procedural rules or algorithmic

processes; (ii) a building performance simulation engine to assess the fitness of each design

alternative generated by the generative model; and (iii) a search mechanism or solver that

automatically steers the iteration process until it finds a solution, or set of design alternatives, whose

performance is closer to the pre-established goals. There are several PGDSs dedicated to the built

environment. These include: Audioptimization, a goal-oriented design system for acoustic design

proposed by Monks et al. [8]; EifForm, for structural design for structural design [9]; GENEARCH

[10,11,12], and GenOpt [13,14] for building energy performance, and more recent methods and

systems based on goal-oriented approaches such as the ones presented by Turrin et al. [15], Wright et

al. [16], Xu et al. [17], and Futrell et al. [18].

Recently, several modular programming tools with commercial optimization packages have emerged,

which facilitate the implementation of custom PGDSs for architectural design. For example, it is

possible to build the framework for an energy-based PGDS by coupling DIVA for Rhino [19] or

Ladybug/Honeybee [20] with Grasshopper functions and methods for parametric design, and any

Grasshopper-based evolutionary solver, such as Galapagos or Octopus [21]. If Karamba [22] was used

instead of DIVA or Ladybug/Honeybee in this scenario, a framework for a structure-specific PGDS

would instead be developed.

Despite remarkable progress in the integration of several tools and modules used in goal-oriented

design methods, severe limitations persist in energy performance-based generative processes. These

limitations are primarily attributed to the geometric limitations of the energy simulation module used.

For example, although EnergyPlus (E+) is a state-of-the-art Building Energy Modeler (BEM), its

modeling features and geometry protocols are very limited relative to those of a typical 3D CAD

software. Shape convexity and planarity are the most significant geometric limitations in E+. All

surfaces must be planar and convex, triangular or quadrilateral. Mesh triangulation can avoid these

limitations, but it can lead to others. For example, mesh triangulation may result in the exponential

increase of simulation time, especially if the triangulation result is a dense mesh. The simulation time

increment is related to view factor calculations necessary for simulating radiant heat transfer;

therefore, it is difficult to simulate a thermal zone with any type of curved and/or double-curved

surfaces in E+. Such geometric limitations force the designer to manually oversimplify the design's

geometry whenever an energy simulation is needed. Although energy modeling experts have

streamlined the manual simplification of model geometry, there are inevitable consequences of this

approach: (i) the geometric gap between the energy model and any other model, which simultaneously

confuses the relationship between geometric features and energy performance and hinders the

integration of building energy performance considerations in the multi-objective optimization process

[23, 24]; (ii) the loss of accuracy, which is more relevant if the building uses passive solar design

strategies and has an envelope with a significant amount of self-shaded surface area; (iii) the loss of

complex parametric modeling as a tool in goal-oriented design, as manual adjustment of an inherently

automated process compromises the value of the automation.

2. Goals and envisioned solution

This work proposes an automatic and robust method of deriving BEM from fully parametric building

models, thus providing a method to efficiently and accurately generate parametric energy models

during a goal-oriented design process. Automatically deriving BEMs from their

parametric/architectural counterparts allows designers to directly use energy-based generative design

processes without the use of intermediary simplified models. This will facilitate the integration of

building energy performance in multi-objective PGDS, which are generative design systems that have

more than one performance objective [25]. For example, a multi-objective PGDS might optimize a

design for both structural and energy performance. The proposed automation will also reduce the

current gap between the different model representations of the same design, thereby strengthening the

relationship between them. For example, if the relationship between the BEM and the parametric

model is bidirectional, visualizing the impact of design changes for energy performance optimization

during the goal-oriented design workflow becomes a trivial task.

To achieve the goal of deriving a thermal model from a 3D model of a design, the proposed method

combines current techniques used in digital fabrication and remeshing with energy modeling ones.

Thus, mesh planarization algorithms will be used to generate discrete meshes from double curved

Non-Uniform Ratio Basis Spline (NURBS) surfaces. The resultant meshes may be easily exported to

E+. Applying such algorithms minimizes the differences among the geometries of the parametric

model, the digital fabrication model, and the energy simulation model. In this way, one model can

directly inform the next.

Because mesh planarization algorithms are also used in digital fabrication, the envisioned goal-

oriented design approach focuses both on energy performance and on construction feasibility. In this

particular study, the approach is applied to the study of glass fritting patterns for the envelope of all-

glass thermal zones with curved and complex envelopes. The theme of a complex curved all-glass

building was selected because it presents at the same time construction obstacles, where planar glass

panels are desirable due to cost constrains, and a difficult thermal problem. In this test case, the

algorithms search for appropriate shading frit patterns, and adapt the emergent forms within pre-

defined boundaries and rules until buildable forms are identified (e.g., maximum glass panel size and

deviation from the original surface). The proposed integration of mesh planarization methods with

energy performance-based design is illustrated in Figure 1.

Figure 1 – Workflow diagram of the proposed energy based generative design system. From an initial surface, a

mesh based on planar quad meshes is produced both for fabrication and energy simulation. The Genetic

Algorithm (GA) search process will optimize the fritting density of the fabrication model glass panels.

In the proposed workflow, planarization algorithms, available in Kangaroo Grasshopper’s add-on

package, process an initial freeform geometry. The result is a mesh composed of planar quads with

constraints applied to their maximum dimensions. Each planar quad corresponds to a planar glass

panel with maximum size of a standard double glass panel: 3x3 m. This model corresponds both to a

model, where all the glass panels are ready for digital fabrication, and a geometry model that can be

directly fed to E+ and Radiance by assigning specific E+ and Radiance materials to each panel.

Radiance is used to map solar radiation on the building envelope and, in turn, clusters facade panels

by the amount of radiation they receive. The proposed workflow uses Honeybee as the E+ interface

and DIVA for Grasshopper as the interface for Radiance. A Genetic Algorithm (GA) controls the frit

density assigned to each panel cluster. In this way, the GA searches for the best fritting densities for

each solar exposure. Finally, when the GA converges, the resultant frit pattern is remapped on the

digital fabrication model. Through this loop, a single architectural geometric representation is

informed by two parameters: construction and energy consumption.

3. Related Work

Recent development in energy modeling has improved the automation of thermal zoning in BEMs

[26, 27, 28, 29, 30, 31]; however, such improvements have been limited to simple geometry. Already

faceted geometries are the most complex cases found in the literature. A review of existing modeling

approaches finds that the parsing of curved surfaces from a CAD or BIM model to a BEM is limited

to standard triangulation algorithms, which allow only limited control and do not follow energy

modeling best practices. Such approaches are found in Ladybug/Honeybee and Green Building Studio

(GBS).

In fact, although possible, perfectly curved forms are relatively rare in architectural design. In most

cases, the curved surfaces that are built are faceted approximations. The infrequency of perfectly

curved structures is typically attributed to their high construction costs, particularly for steel and glass

construction. The only exceptions to this are certain types of concrete building technologies (e.g.;

concrete elements that use metal or polymer based frameworks). Thus, it is reasonable to think that

the modeling problem in goal-oriented design for energy performance could be addressed as a

construction problem where the planarity of the elements is desirable. This is a common research

topic in architectural geometry and digital fabrication [32, 33, 34, 35].

In its search of construction approaches for double-curved geometry based on planar elements, recent

research in digital fabrication applies developments in mathematics around quadmesh quasi-

planarization [36, 37]. To obtain a planar quadrilateral subdivision of any given surface, quasi-

isothermic remeshing algorithms are used to transform a given mesh into a S-isothermic quadrilateral

mesh. A quadrilateral mesh is S-isothermic if: (i) all the quadrilaterals are planar, (ii) all faces have

incircles, and (iii) the incircles of adjacent quadrilaterals touch. Figure 3 illustrates the concept of a

quadrilateral mesh and how the quasi-isothermic algorithm remeshes a given mesh into a planar quad

mesh. Sechelmann et al., describes in more depth the algorithmic approach to create S-isothermic

quadrilateral meshes in [36].

Figure 2. Left: incircles of a non S-isothermic quad mesh. In order to be S-isothermic the incircles of adjacent

meshes need to touch. This produces the same ratio on both sides of the edge ij; cot(βi/2)/cot(βj/2) [36]. Right:

examples of S-isothermic quad meshes generated by the algorithm proposed by [36].

Figure 3 shows a pavilion that used a similar planarization approach; however, this example departs

from the concept of S-isothermic in-circles through periodic conformal maps [37, 38]. Conformal

maps can extend the panelization from quad planar panels to hexagonal ones. Because these

algorithms can remesh an original surface into a set of planar and convex panels, while

simultaneously controlling how the remeshing is done, the resultant geometry can be simulated in E+

if assigned a proper E+ construction assembly to it.

Figure 3. The Landesgartenschau Exhibition Hall was designed and constructed resorting to surface discrete

planarization algorithms. [38]

4. Methodology

The proposed goal-oriented design system was applied to a case study composed of three glass

pavilions in order to test and validate the approach. The tested pavilions were variations of a semi-

ellipsoid with a 30 m primary axis, a 15 m secondary axis, and a 7.5m height. In each case, the

primary axis was oriented due North-South. The poles of this base geometry were cut to provide two

entrances on the North and South ends. This constitutes solution A. Solution B is a deformation of

solution A, in which the section along the primary axis was rescaled to create a “peanut” shape to

promote self-shading and to create a different solar radiation pattern. As discussed in section 1, a

different irradiance distribution not only affects the overall energy consumption of the solutions but

also the optimized frit pattern/density on the envelope. Solution C is the result of the deformation of

the main meridians of the ellipsoid, which were varied for the same reasons as solution B. Figure 4

shows the initial NURBS geometry of all the solutions of the case study. Table 1 reports the floor

area, surface area, volume, and form factor (surface to volume ratio) of each pavilion.

Figure 4. The base semi-ellipsoid and the three pavilions derived from it.

Floor Area (m2)

Volume (m3)

Surface area (m2)

Form factor

Solution A

339.5

1646.8

583.2

0.35

Solution B

302.5

1182.6

508.3

0.43

Solution C

321.9

1211.5

497.8

0.41

Table 1. Floor area, volume, surface area, and form factor of each solution.

Although the workflow proposed in section 2 could be applied to any complex building shape, it was

only tested in these three scenarios due to the prototypal nature of this work. The goal was to perform

a set of initial tests that verifies the efficacy of the proposed method. Thus, the three t scenarios

represent different degrees within a range of potential geometric complexity for the same type of

building. Although they share the same archetypal form, the surface-to-volume ratios (form factors),

solar radiation distributions, and self-shading patterns are unique for each solution. Solution A

presents a surface of revolution of a simple arc, with a homogeneous, gradient-based solar radiation

distribution and with the smallest form factor of the three examples. The double curvature of solution

B’s revolution profile increases its form factor while introducing more variation in solar radiation

distribution and self-shading patterns. Finally, solution C presents a higher level of geometric

complexity while maintaining the same form factor. Its meridional section varies along the revolution

path, which generates a double curved surface that is difficult to discretize in planar quad meshes.

Finally, the workflow was implemented as an integrated dynamic model (IDM) [39]. An integrated

dynamic model can be defined as a combination of a design tool, a visual programming language

(VPL), and a building performance simulation software (BPS). The IDM paradigm was selected

because it provides valid feedback, it is extremely flexible to accommodate rapid design changes, and

it allows the user to have full control on every step of the process [39]. Thus, the Rhino CAD platform

environment was selected because it offers a wide range of tools that fully support the implementation

of IMDs, namely: (1) dedicated modeling tools, (2) a well developed and known VPL, Grasshopper,

which can be extended by other high-level languages such as Python, and (3) several interfaces to

EnergyPlus and Radiance, such as DIVA and Honeybee. This CAD platform was also important in

the implementation of the proposed workflow because it provides several add-ons for mesh

manipulation, such as Kangaroo [40], and several evolutionary solvers such as Galapagos and

Octopus.

4.1 Modeling

Each solution was panelized and planarized using Kangaroo Physics 2.0 [40], a dynamic physics

simulator for Grasshopper that uses the planarization algorithms discussed in section 3. The number

and size of the glass panels (quadmeshes) was controlled by a small grasshopper script that divides

the original surface along its width (u) and length (v), directions and informs the user about the

dimensions of the largest and smallest glass panel. After this step, the NURBS surface was

transformed into a mesh and planarized using Kangaroo 2.0. Finally, the planarity of each panel was

tested by a grasshopper procedure made specifically for such a test. If the planarity test returns “true”

for all quad mesh panels it validates Kangaroo's remeshing settings, and the energy simulation can be

run. Figure 5 illustrates both the original surfaces and the resulting planar quad meshes computed by

Kangaroo, which constitute both the construction/fabrication model for the glass panels and the

geometric model for E+. Figure 6 shows the output from both the panel sizing algorithm and the

planarity test of a non-optimized quad mesh composed of 120 panels.

Figure 5. Original surfaces and the resulting S-isothermic planar quad meshes. The processed meshes are ready

to be fed to E+.

Figure 6. Left: the design system informs the user about the location and the dimensions of the biggest (orange)

and smallest (purple) glass panel. In this way, the user can assess if the surface rationalization is feasible from a

construction point of view. Right: planarity test of an intermediary step of the planarization process. The red

color flags non-planar quadmeshes, while the green color indicates planar ones.

The fritting density of the glass panels was modeled as an E+ shading object; thus, every panel had a

scalable centered opaque shading object. By scaling every shading object, it was possible to infer the

fritting percentage applied to each glass panel. This was the only simplification used to reduce

simulation time without compromising accuracy. The system then reprocessed the simplified shades

back to the digital fabrication model as fritting patterns through a dedicated set of scripted functions

that allows the user to control the size and shape of the frit pattern. Figure 7 shows the energy model

with shading objects equivalent to a fritting density of 60% and their translation to a possible fritting

pattern within the digital fabrication model.

Figure 7. E+ shading objects (60% shading ratio) and their remapping to a possible glass fritting pattern.

4.2 Simulation and optimization process

In the reported test case, the energy and solar radiation simulations were performed with E+ and

Radiance using the climate file of Logan International Airport, Boston, USA. Although a cold

climate, Boston experiences hot and humid summers; thus, it provides a useful climate to test both

winter and summer conditions. Summer thermal performance was of particular importance because

the proposed system aims for the optimization of shading, a passive cooling strategy.

The material used for the envelope was triple low-e glazing with argon. This type of glazing was

chosen for its high thermal performance, specifically its low thermal conductivity (U-factor). The

frame and structure of the glass was modeled as aluminum. Finally, the ground floor was modeled as

a heavyweight concrete slab. Before running a whole building energy simulation with E+, a solar

radiation analysis with Radiance was conducted for each panel. A sensor node was placed in the area

centroid of each panel. The value registered by the sensor node was remapped into a color that was

applied to the corresponding panel quad mesh. The panels were then clustered according to their

irradiance values. The grouping of the panels is based on solar radiation similarity. Table 2, shows the

five groups of panels that were generated for each design alternative according with their irradiance

range. This clustering process avoids panel-by-panel optimization, minimizing the number of design

variables controlled by the GA. Although this strategy constrains the design solution space it

facilitates the search process without losing too much granularity. Finally, Grasshopper’s Galapagos

standard GA controlled the shading fraction of each cluster aiming towards the minimization of the

overall energy consumption. Figure 8 shows the annual solar radiation on the envelope of each

pavilion and illustrates how the panels were clustered in groups of similar solar irradiance (SR).

SR: min.

(kWh/m2)

SR: max.

(kWh/m2)

SR: 0-20%

(kWh/m2)

SR: 20-40%

(kWh/m2)

SR: 40-60%

(kWh/m2)

SR: 60-80%

(kWh/m2)

SR: 80-100%

(kWh/m2)

Solution A

417

1636

417 - 661

661 - 905

905 - 1148

1148 - 1392

1392 - 1636

Solution B

593

1647

593 - 804

804 - 1015

1015 - 1225

1225 - 1436

1436 - 1647

Solution C

830

1613

830 - 987

987 - 1143

1143 - 1230

1230 - 1456

1456 - 1613

Table 2. Clusters of glass panels per solution and their irradiance range (kWh/m2).

After clustering the glass panels by solar radiation, a whole building energy assessment was

performed with E+. The main simulation assumptions and settings used were: (i) simplified variable

air volume (VAV) HVAC system to calculate sensible cooling and heating loads and energy

consumption for space conditioning; (ii) heating setpoint set to 20 Cº and cooling setpoint set to 26

Cº; (iii) infiltration rate of 0.3 air changes per hour (ACH), which represents an airtight building; (iv)

a power density for light fixtures of 11.4 W/m2 (a typical value for fluorescent bulbs); and finally (vi)

an illuminance setpoint of 300 lux to control the dimmable lighting system used in the model.

Figure 8. Up: annual solar radiation mapped in the envelope of the three pavilions. Down: the five glass panel

clusters based on the annual irradiance of Boston, MA.

Three main energy uses were considered and reported: heating, cooling, and lighting. Equipment

loads were excluded because they do not depend on environmental factors or on building envelope

characteristics but on the type of equipment used and occupancy schedules. The total energy

consumption reported is the sum of hourly heating, cooling, and lighting energy consumption. Due to

the differences in area and in volume between the different alternatives, Energy Use Intensity (EUI)

per area unit (EUI - kWh/m2), and per volume unit, (EUI(v) - kWh/m3), were the energy metrics used

to assess results. Finally, the performance-based design exploration included two main tasks:

1) An annual parametric study of different shading conditions – Before setting an automatic

optimization procedure for shading/fritting density, three different shading ratios were annually

simulated in order to better understand the baseline energy profile of the three pavilions. The different

shading alternatives were: (i) no shading (base case); (ii) 40% evenly distributed shading; and (iii) a

shading gradient that is distributed according with solar radiation (SR) measurements (Table 3):

0 to 20% of

total SR

20 to 40% of

total SR

40 to 60% of

total SR

60 to 80% of

total SR

80 to 100% of

total SR

% of shading

25

40

55

70

85

Table 3. Gradient shading alternative - percentage of shading/glass fritting according to solar radiation.

By simulating different forms with similar areas and volumes, we can assess how form affects solar

irradiation and total energy consumption, as well as the impact of different shading schemes in the

overall energy profile of each case.

2) Optimization of shading/fritting percentage for cooling energy consumption – Because shading is a

cooling strategy, an optimization cycle was performed, which was informed by the lessons learned in

the parametric study. The summer period considered for the optimization was May 20th to September

21st, including thus the end of the mid-season and the whole summer.

5. Results

Results are grouped in the two sections: (i) Parametric study of different shading conditions; and (ii)

Optimization of shading/fritting percentage for cooling energy consumption.

5.1 Parametric study of different shading conditions

Figures 8, 9, and 10 illustrate the annual solar radiation analysis for Boston, and the three different

shading strategies applied for each solution in this analysis stage. The solar radiation analysis

presented in Figure 8 shows how different forms present different distributions. These differences

directly result from the geometric differences in tilt, angle, and self-shading patterns among the three

tested forms. Thus, although these tested forms derive from the same archetypical geometry, their

annual irradiance distributions are sufficiently diverse to result in mutually unique shading patterns.

Figure 9 illustrates the resultant shading pattern given a desired homogeneous shading factor of 40%

per panel. In contrast, Figure 10 shows the shading pattern given a gradient shading strategy based on

solar radiation values, where incident radiation acts as an attractor/repeller that controls the amount of

shading for each cluster of panels.

Figure 9. 40% homogeneous shading distribution across the three design alternatives modeled with E+ shading

objects.

Table 4 reports both the EUI and EUI(v) of each solution. Figures 11 and 12 shows a bar chart that

compares the energy performance of each of the tested design alternatives in terms of EUI (kWh/m2)

and EUI(v) (kWh/m3). As shown in the graphs, solution A requires the most heating energy, which

can be attributed to this form’s small solar exposure on its north facade. Solutions A and B have

similar irradiance distributions on their north facades; however, solution B compensates for the

associated heat losses via two south-facing areas that receive high solar gains. Solution C has the

lowest heating energy consumption of the three test cases. This can be attributed to this form’s more

uniform solar radiation distribution, which reduces the contrast between heat losses and gains. In all

cases, shading is effective in reducing cooling loads, but it also increases the heating demands. The

gradient shading solution is the most effective in reducing cooling energy consumption, which

supports the intuition that the panels that need more shade are the ones that receive more solar

radiation. Lighting energy consumption varies slightly among the three scenarios; however, these

differences are insignificant relative to those of the space conditioning end-uses. This is explained by

the consistently high amount of glazing across all solutions and each solution’s success in blocking

direct sun while still allowing diffuse light, which minimizes lighting energy needs with relative

consistency.

Figure 10. The gradient shading solution modeled with E+ shading objects.

Solution A

Solution B

Solution C

No

shading

40%

shading

Gradient

shading

No

shading

40%

shading

Gradient

shading

No

shading

40%

shading

Gradient

shading

EUI - kWh/m2

Heating

105.1

110.5

128.9

53.6

54.5

64.7

52.8

57.0

62.2

Cooling

29.6

22.2

11.4

49.2

40.5

25.5

45.1

35.2

26.5

Lighting

10.5

10.5

10.7

11.4

11.5

11.6

11.0

12.0

11.1

EUI [total]

145.2

143.3

151.0

114.2

106.5

101.8

108.9

104.3

99.9

EUI(v) - kWh/m3

!

!

Heating

21.7

22.8

26.6

13.7

13.9

16.6

14.0

15.2

16.5

Cooling

6.1

4.6

2.4

12.6

10.4

6.5

11.97

9.4

7.0

Lighting

2.2

2.2

2.2

2.9

2.9

3.0

2.9

3.2

3.0

EUI(v) [total]

29.9

29.5

31.1

29.2

27.3

26.1

28.9

27.7

26.5

Table 4. EUI and EUI(v) of each solution.

Figure 11. Annual EUI (kWh/m2) of the different design alternatives.

Figure 12. Annual EUI(v) (kWh/m3) of the different design alternatives.

Figures 11 and 12 show that the significant reduction of cooling energy consumption provided by the

gradient solution does not compensate in terms of heating in solution A. From this more holistic

perspective, the 40% homogeneous shading solution is slightly more efficient, as indicated by its

lower EUI and EUI(v).

Regarding solution B energy profile, the gradient shading scheme is the most energy efficient solution

overall. It reports an improvement of 11% when compared to the no shading scheme. This is because

the difference between heating and cooling loads in the unshaded case is smaller than in solution A.

Like solution A, the heating demand is higher for the shaded conditions than for the base case, but

unlike solution A, the reduction in cooling demand is sufficiently compensates for the higher heating

loads. Finally, solution C shows a similar profile to solution B but is less efficient.

This parametric study shows that providing a fixed shading system to a specific all-glass building

geometry can be effective year-round, even in a climate like Boston. Even in solution A, the

experiment shows that with a homogeneous 40% shading ratio we can improve the overall energy

consumption. This supports the choice to run the optimization cycle only for the summer period in

order to find the best shading strategy.

5.2 Optimization of shading/fritting percentage for cooling energy consumption

Two fully automated search procedures were conducted using Galapagos’ standard GA. The first,

Opt#1, was conducted without any restrictions to the parameters that control the amount of shading

for each group of panels receiving the same irradiance range. The second, Opt#2, was conducted with

some constraints assigned to those parameters, which aimed to promote a gradient shading pattern or

at least to minimize shading to panels that receive less solar radiation. In order to optimize the shading

strategy the optimization was constrained to end of spring, May 20th, to the end of summer,

September 21st.

Figures 13 and 14 show the results of Opt#1. Figure 14 shows the remapping of shading ratio,

presented in Figure 13, to a frit pattern. The Opt#1 solution is counterintuitive because it provides

more shading to the panels that receive less solar radiation. To reduce cooling loads, the GA blocks

the panels with the largest area, except the ones that are more exposed to the sun. Nevertheless, this

simulation indicates that the GA probably found a local minimum and that the process needed more

guidance.

Figures 15 and 16 show the results of Opt#2. Figure 16 presents the frit pattern of the shading

percentage shown in Figure 15. With the extra guidance provided by the constraints, the GA found a

more expectable solution that places the higher shading ratio in the areas that report higher irradiance

values.

Table 5, figures 17, and 18, compare the energy performance of each design solution for no shading

(baseline), Opt#1, and Opt#2 for the time period under study. The charts show that Opt#2 is extremely

effective in solution A, slightly effective in solution B and not as effective as Opt#1 in solution C.

However, the results of solution A and B show that steering the optimization process through

constraints on the parametric model based on initial analysis can lead to better results that are more

energy efficient and closer to the designer’s intents. In summary, the GA was able to find more

efficient shading strategies that could improve the energy performance of the all glazing pavilions in

the summer period.

Figure 13. Opt#1 shading solution with E+ shading objects.

Figure 14. Opt#1 shading ratio remapped as a fritting pattern. Shading ratio label indicates the average.

Figure 15. Opt#2 shading solution with E+ shading objects.

Figure 16. Left: Opt#2 shading ratio remapped as a fritting pattern. Shading ratio label indicates the average.

Right: Examples of fritting patterns generated with the remapping algorithm. All of the three options has the

same shading ratio, 70%.

kWh/m2

Solution A

Solution B

Solution C

No

shading

Opt#1

Opt#2

No

shading

Opt#1

Opt#2

No

shading

Opt#1

Opt#2

Heating

0.37

0.46

0.54

0.08

0.13

0.11

0.07

0.11

0.09

Cooling

24.24

6.59

4.34

34.65

12.05

11.26

32.85

6.47

12.47

Lighting

4.28

4.29

4.41

4.55

4.49

4.58

4.34

4.42

4.46

EUI [total]

28.89

11.33

9.29

39.28

16.66

15.95

37.26

11.00

17.01

kWh/m3

!

!

Heating

0.08

0.09

0.11

0.02

0.03

0.03

0.02

0.03

0.02

Cooling

5.00

1.36

0.90

8.87

3.08

2.88

8.73

1.72

3.31

Lighting

0.88

0.88

0.91

1.16

1.15

1.17

1.15

1.18

1.18

EUI(v) [total]

5.96

2.34

1.92

10.05

4.26

4.08

9.90

2.92

4.52

Table 5. EUI and EUI(v) of each solution for no shading, Opt#1, and Opt#2.

Figure 17. EUI (kWh/m2) of the different design alternatives of the optimization experiment.

Figure 18. EUI (kWh/m2) of Solution A for no shading, Opt#1, and Opt#2 alternatives.

6. Discussion

The parametric study tested three shading scenarios for each of the three solutions: no shading, 40%

homogeneous shading for the glass panels, and a gradient shading approach based on solar radiation

incident upon the building envelope. The goal was to test the efficacy of a shading strategy based on a

fritting pattern in Boston, Massachusetts over one year. Boston experiences all seasons. The summers

are warm to hot and humid, with periods that can exceed 32 ºC, while winters are cold, with freezing

temperatures from November through the end of March. Typically, Boston’s climate is heating-

dominated, but because it also experiences hot and humid summers, cooling loads are relevant in this

period. That is why shading is an important passive strategy for a whole glass pavilion in this climate.

The annual parametric simulations show two different patterns. Table 4, Figure 11, and Figure 12,

show that heating loads clearly dominate solution A, while in solutions B and C, the difference

between heating and cooling loads is much smaller. There are several reasons for this, but it seems

that the relationship between surface area and volume plays a relevant role. Because it has the largest

volume of the three, solution A has the largest volume of air to heat; simultaneously, it has the largest

surface area of the three solutions, which means that it has the most envelope heat loss. The result is

significantly higher heating than cooling loads.

Because solution A does not have self-shading and has more exposed surface area, it also requires a

higher percentage fritting in the glass panels in the gradient shading scenario. This eventually blocks

desired solar heat gains in the winter, which also contributes to the relatively high heating loads. That

is why the gradient shading scenario is the only one where shading is unfavorable. The volumes and

surface areas of solutions B and C are smaller; therefore, they require a smaller heating load and

experience less heat dissipation than does solution A. Self-shading and/or a more even distribution of

solar radiation on the envelopes of solutions B and C help distribute a more balanced frit pattern in the

gradient shading scenario. This experiment also demonstrates that the optimization of individualized

fritting patterns in glass panels requires a geometrically accurate energy model.

Table 6 shows the percentage of improvement of EUI by adding shading to the glass pavilions. It is

clear that shading is a beneficial a passive strategy year round for an all-glass pavilion in Boston.

Even solution A sees a small benefit from 40% homogeneous shading in its glass panels, which

indicates that shading can be an effective passive design strategy to reduce energy consumption in

cases with a large exposed surface area. The table also shows that deriving a shading pattern based on

solar radiation distributions can be more efficient than using homogeneous shading factors. These

results reinforced the need of the second set of tests that aimed to optimize the glass fritting in each

glass panel.

% of improvement

(Base case: No shading)

% of improvement

(Base case: 40% homogeneous

shading)

40% homogeneous

shading

Gradient

Gradient

Solution A

1%

-4%

-5%

Solution B

7%

11%

4%

Solution C

5%

8%

3%

Table 6. Percentage of improvement in energy consumption of the parametric study of different shading

conditions. The percentage of improvement was calculated for: (1) 40% homogeneous shading and the gradient

shading against no shading; (2) gradient shading against 40% homogenous shading.

Because shading is a passive cooling strategy, the goal of the second set of tests was to determine the

efficacy of glass fritting as a cooling strategy. Thus, the optimization was constrained to the warm

months in Boston. As previously described in section 5.2, two optimization procedures were

performed in the second set of tests. The first, Opt#1, was conducted without constraining the amount

of shading for each group of panels receiving the same range of solar radiation. Figure 13 shows that

the shading optimization prioritizes the area size of the panel clusters instead of the amount of solar

radiation incident upon the building envelope. To test if the strategy adopted by the GA in Opt#1 was

optimal, a second optimization procedure was conducted that imposed a more direct relationship

between incident solar radiation and the amount of fritting.

% of improvement

(Base case: No shading)

% of improvement

(Base case: Opt#1)

Opt#1

Opt#2

Opt#2

Solution A

61%

68%

18%

Solution B

58%

59%

4%

Solution C

71%

54%

-55%

Table 7. Percentage of improvement of energy consumption in the optimization study. The percentage of

improvement was calculated for: (1) Opt#1 and Opt#2 against no shading; (2) Opt#2 against Opt#1.

Table 7 compares both Opt#1 and Opt#2 to a non-shaded base case, as well as to each other. The

automated search procedure provided by the GA was able to considerably reduce the overall energy

consumption of the non-shaded base cases. As expected, the optimization produced a high

improvement in solution A in both optimization procedures. This is related to the fact that solution A

has the largest surface area and, consequently, the most exposed envelope of the three scenarios.

Comparing the two optimization strategies, constraining the search procedure to promote a gradient

shading pattern based on the solar radiation distribution yields better results in solutions A and B. In

solution A, Opt#2 was able to find a solution that is 18% better than the Opt#1. In the case of solution

C, Opt#2 found a worst solution than Opt#1. This indicates that the success of steering the search by

imposing Opt#2 constraints is closely related to the amount of exposed surface area. Thus, the poor

performance of Opt#2 in solution C could be related to that scenario’s lower solar radiation variance,

which makes this option a less suitable candidate for a gradient-based shading approach.

7. Conclusions

This work proposed a new, goal-oriented design approach, which integrates multiple 3D

representations of a building into a single parametric model that may be optimized for multiple

parameters, including energy efficiency. The proposed approach uses planarization methods, which

have been developed and applied in digital fabrication, to overcome the limitations of energy

modeling of curved and double curved geometries. This allows to automatically generate the

geometry of an energy model from a complex 3D CAD model, thus, avoiding time consuming

manual modeling tasks that are incompatible with fully automated optimization workflows.

Because the planarization algorithms are accessible, the user can directly specify how the

discretization of the original surfaces is performed, therefore controlling the degree of detail of the

energy model. The proposed approach also minimizes the gap between construction model and

thermal model by using the geometry of the first as an input to the second. Only the shades are

modeled differently for energy simulation. To solve this discrepancy, a dedicated algorithm remaps

the most thermally effective shade into a fritting pattern. Synchronized or shared models in

performance-based generative systems improve and streamline the feedback between energy

simulation and architectural design. In this way, energy simulation can easily inform the construction

model and vice-versa.

The use of planarization algorithms to optimize an original free form also indicates a high potential

for model integration in multi-objective goal-oriented design methods, where the automated search

process aims to simultaneously optimize different aspects of a single design.

The results showed the usefulness of goal-oriented methods to fully understand the complex trade-offs

associated with energy-related problems in buildings. Unexpected results from these methods could

inform the design process with new perspectives and different solutions to solve specific design

problems. The two optimization procedures showed how an accurate geometry can be important in

energy optimization workflows. Optimizing the fritting density of different glass panel clusters of the

same building envelope requires a detailed energy model that can simulate self shading and variable

solar radiation distribution patterns. This is hard to incorporate in the typical oversimplified box-based

energy models. The optimization experiments also showed that in some cases steering the

optimization process, by constraining certain design parameters, can improve the quality of the search

procedure. Finally, the experiments also showed how the energy performance approach can actually

inform the physical and aesthetic properties of the building envelope with different densities of

fritting patterns.

In sum, the generative-design system prototype proved to be:

• Robust – it was able to automatically translate a diversity of double-curved building

geometries to E+.

• Adaptable – by allowing the user to control the generation of the BEM model and steer the

optimization process through the use of constraints.

• Reliable – it found energy efficient passive solutions that resulted in percentages of

improvement of almost 70%.

The limitations of the proposed methodology are mostly related with simulation computational time.

There is direct correlation between the number of panels and simulation time, meaning that the higher

the number of panels, the higher the simulation computational cost due mostly to view factor

calculations for E+ radiant heat transfer computation. Another limitation is related with steering the

GA search procedure. The Opt#1 experiment shows that the GA probably got stuck in a local

minimum. This problem can be related with the type of GA used, Galapagos standard GA. In future

work, it will be worthy to test more sophisticated and robust GAs, such as SPEA2 [41] commonly

used for multi-objective problems.

In order to further develop the proposed methodology, future work should focus on:

1) Test the GA results against a problem with a known optimal value - this step would

measure and validate the search mechanism accuracy.

2) Comparing different evolutionary solvers - this would inform the methodology about the

trade-offs between speed and quality of several search mechanisms, thus helping in the

selection and the setting of the parameters of the most suitable evolutionary solver.

3) Extend the proposed methodology to multi-zone models. Due to the prototypal nature of

this work only single zone models of a specific building type were tested. Extending the work

to more complex multi-zone models of different building types would improve the

methodology regarding its applicability and generalization.

4) Modeling the fritting panels as Bidirectional Scattering Distribution Function (BSDF)

based materials – this would improve accuracy and simulation time.

5) Adding more details to the fabrication model, such as structure and variable thicknesses –

to test how flexible is the method to contemplate more building design disciplines and their

respective representational models.

6) Extending the methodology to multi-objective optimization generative design systems

where, for example, structural optimization and construction cost are contemplated.

8. Acknowledgments

Work presented in this paper was partially financed by the Portuguese Science and Technology

Foundation (FCT) through the PhD scholarship SFRH/BD/98658/2013. The authors would

like to thank Sara Tepfer for her precious collaboration and help.

9. References

[1] McNeel, Robert. "Rhinoceros 3D." Retrieved June 15 (2016).

[2] McNeel, Robert. "Grasshopper-Generative Modeling with Rhino." Computer software (2011b),

http://www.grasshopper3d.com (2010).

[3] Keough, I. “Dynamo: Designing a Visual Scripting Interface for the Revit API (notes).” Also see

https://github. com/ikeough/Dynamo/wiki for more information about coding in Dynamo (2011):

n.pag. Print.

[4] Daru, Roel, and H. P. S. Snijder. "GACAAD or AVOCAAD? CAAD and Genetic Algorithms for

an Evolutionary Design Paradigm." (1997): n. pag. Web. 30 June 2016.

[5] Ward, Gregory J., and Francis M. Rubinstein. “A New Technique for Computer Simulation of

Illuminated Spaces.” Journal of the Illuminating Engineering Society 17.1 (1988): 80–91.

[6] Crawley, Drury B., et al. "EnergyPlus: creating a new-generation building energy simulation

program." Energy and buildings 33.4 (2001): 319-331.

[7] Wilson, E. L., and A. Habibullah. "SAP2000 integrated finite element analysis and design of

structures." Analysis reference. Computers and Structures (1997).

[8] Monks, M., Oh, B. M., and Dorsey, J. "Audioptimization: Goal-based acoustic design." Computer

Graphics and Applications, IEEE 20.3 (2000): 76-90.

[9] Shea, K., Aish, R., and Gourtovaia, M. "Towards integrated performance-driven generative design

tools." Automation in Construction 14.2 (2005): 253-264.

[10] Caldas, L.G., and Norford L. K. "A design optimization tool based on a genetic algorithm."

Automation in construction 11.2 (2002): 173-184

[11] Caldas, L. “Generation of Energy-Efficient Architecture Solutions Applying GENE_ARCH: An

Evolution-Based Generative Design System", Advanced Engineering Informatics, 22.1 (2008): 59-70.

[12] Caldas, Luisa G., and Luis Santos. "Generation of Energy-Efficient Patio Houses with

GENE_ARCH: Combining an Evolutionary Generative Design System with a Shape Grammar."

Proceedings of the 30th Conference on Education in ComputerAided Architectural Design in Europe,

eCAADe. 2012.

[13] Wetter, M. "GenOpt - A generic optimization program." Seventh International IBPSA

Conference, Rio de Janeiro. 2001.

[14] Rahadian, K., and M. I. Alhamid. "Simulation and Optimization of Solar Thermal Cooling

System in Manufacturing Research Center Building to Reduce Operational Cost Using Software

EnergyPlus and GenOpt." Applied Mechanics & Materials 780 (2015).

[15] Turrin, M., Peter von Buelow, and Rudi Stouffs. “Design Explorations of Performance Driven

Geometry in Architectural Design Using Parametric Modeling and Genetic Algorithms.” Advanced

Engineering Informatics 25.4 (2011): 656–675.

[16] Wright, Jonathan A., Alexander Brownlee, Monjur M. Mourshed, and Mengchao Wang. “Multi-

Objective Optimization of Cellular Fenestration by an Evolutionary Algorithm.” Journal of Building

Performance Simulation 7.1 (2014): 33–51.

[17] Xu, Jun, Jin-Ho Kim, Hiki Hong, and Junemo Koo. “A Systematic Approach for Energy

Efficient Building Design Factors Optimization.” Energy and Buildings 89 (2015): 87–96.

[18] Futrell, Benjamin J., Ertunga C. Ozelkan, and Dale Brentrup. “Bi-Objective Optimization of

Building Enclosure Design for Thermal and Lighting Performance.” Building and Environment 92

(2015): 591–602.

[19] Jakubiec, A., and C. Reinhart. “DIVA-FOR-RHINO 2.0: Environmental Parametric Modeling in

Rhinoceros/grasshopper Using RADIANCE, Daysim and EnergyPlus.” Conference Proceedings of

Building Simulation. N.p., 2011.

[20] Roudsari, Mostapha Sadeghipour, Michelle Pak, and Adrian Smith. “Ladybug: A Parametric

Environmental Plugin for Grasshopper to Help Designers Create an Environmentally-Conscious

Design.” Proceedings of the 13th International IBPSA Conference Held in Lyon, France Aug. N.p.,

2013.

[21] "Octopus." Food4Rhino. Web. 28 June 2016.

[22] Preisinger, Clemens, and Moritz Heimrath. “Karamba—A Toolkit for Parametric Structural

Design.” Structural Engineering International 24.2 (2014): 217–221.

[23] Flager, Forest, Benjamin Welle, Prasun Bansal, Grant Soremenku, and John Haymaker.

“Multidisciplinary Process Integration and Design Optimization of a Classroom Building.” Journal of

Information Technology in Construction 14 (2009): 595–612.

[24] Shao, Yunming, Philipp Geyer, and Werner Lang. “Integrating Requirement Analysis and Multi-

Objective Optimization for Office Building Energy Retrofit Strategies.” Energy and Buildings 82

(2014): 356–368.

[25] Welle, Benjamin, John Haymaker, and Zack Rogers. “ThermalOpt: A Methodology for

Automated BIM-Based Multidisciplinary Thermal Simulation for Use in Optimization

Environments.” Building Simulation. Vol. 4. Springer, 2011. 293–313.

[26] Jones, Nathaniel L., McCrone, Colin J., Walter, Bruce J., Pratt, Kevin B., and Greenberg, Donald

P. "Automated translation and thermal zoning of digital building models for energy analysis."

Building Simulation Conference. 2013.

[27] Kensek, Karen. "Visual programming for building information modeling: energy and shading

analysis case studies." Journal of Green Building 10.4 (2015): 28-43.

[28] Georgescu, Michael, and Igor Mezić. "Building energy modeling: A systematic approach to

zoning and model reduction using Koopman Mode Analysis."Energy and buildings 86 (2015): 794-

802.

[29] Kim, Jong Bum, WooSeong Jeong, Mark J. Clayton, and Jeff S. Haberl. "Developing a physical

BIM library for building thermal energy simulation."Automation in construction 50 (2015): 16-28.

[30] Jeong, WoonSeong, and JeongWook Son. "An Algorithm to Translate Building Topology in

Building Information Modeling into Object-Oriented Physical Modeling-Based Building Energy

Modeling." Energies 9.1 (2016): 50.

[31] Dogan, Timur, Christoph Reinhart, and Panagiotis Michalatos. "Autozoner: an algorithm for

automatic thermal zoning of buildings with unknown interior space definitions." Journal of Building

Performance Simulation 9.2 (2016): 176-189.

[32] Pottmann, Helmut, Jacques Raynaud, and Alexander Schiftner. "New strategies and

developments in transparent free-form design: from facetted to nearly smooth envelopes."

International Journal of Space Structures 25.3 (2010): 185-197.

[33] Eigensatz, Michael, Mario Deuss, Alexander Schiftner, Martin Kilian, Niloy J. Mitra, and

Helmut Pottman. "Case studies in cost-optimized paneling of architectural freeform surfaces."

Advances in Architectural Geometry 2010 (2010): 49-72.

[34] Deng, Bailin, Sofien Bouaziz, Mario Deuss, Alexandre Kaspar, Yuliy Schwartzburg, and Mark

Pauly. "Interactive design exploration for constrained meshes." Computer-Aided Design 61 (2015):

13-23.

[35] Pottmann, Helmut, Michael Eigensatz, Amir Vaxman, and Johannes Wallner. "Architectural

geometry." Computers & graphics 47 (2015): 145-164.

[36] Sechelmann, Stefan, Thilo Rörig, and Alexander I. Bobenko. "Quasiisothermic mesh layout."

Advances in Architectural Geometry 2012. Springer Vienna, 2013. 243-258.

[37] Rörig, Thilo, Stefan Sechelmann, Agata Kycia, and Moritz Fleischmann. "Surface Panelization

Using Periodic Conformal Maps."Advances in Architectural Geometry 2014. Springer International

Publishing, 2015. 199-214.

[38] Krieg, Oliver David, Tobias Schwinn, Achim Menges, Jian-Min Li, Jan Knippers, Annette

Schmitt, and Volker Schwieger. "Biomimetic lightweight timber plate shells: Computational

integration of robotic fabrication, architectural geometry and structural design." Advances in

Architectural Geometry 2014. Springer International Publishing, 2015. 109-125.

[39] Negendahl, Kristoffer. “Building Performance Simulation in the Early Design Stage: An

Introduction to Integrated Dynamic Models.” Automation in Construction 54 (2015): 39–53.

[40] Piker, Daniel. "Kangaroo: form finding with computational physics."Architectural Design 83.2

(2013): 136-137.

[41] Zitzler, Eckart, Marco Laumanns, and Lothar Thiele. “SPEA2: Improving the Strength Pareto

Evolutionary Algorithm.” Eurogen. Vol. 3242. 2001. 95–100.