Building and Environment: Sciencedirect

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 13

Building and Environment 156 (2019) 178–190

Contents lists available at ScienceDirect

Building and Environment


journal homepage: www.elsevier.com/locate/buildenv

Building facade multi-objective optimization for daylight and aesthetical T


perception
Yun Kyu Yi
School of Architecture, University of Illinois at Urbana-Champaign, 117 Temple Hoyne Buell Hall, 608 Lorado Taft Dr, Champaign, IL, 61820, USA

ARTICLE INFO ABSTRACT

Keywords: In recent years, especially in a building envelope, parametric design provides a method of continuous de-
Multi-objective optimization (MOO) formation of façade patterns until the architect finds interesting patterns or shapes that satisfy the desired
Expert system aesthetics. However, these new design methods pose a question regarding the reasoning behind them, and
Aesthetical perception sometimes the aesthetic interest dominates the true function of the envelope system and contrasts it. In opposite,
Daylight
too much engineering in the envelope system creates a problem with the identity of the façade.
Building performance
Parametric design
The purpose of this paper is to propose a method to integrate two different performances, quality and
quantity, into one measurable goal. Using an existing building's facade as a case study, the building's facade was
analyzed to understand the architect's logic behind its design. Then the found logic was programmed into the
tool that allowed for the morphing of the façade into a different configuration, which can be evaluated by both
quality and quantity performance to find the better solution to satisfy both goals.
For the purpose of this paper, multi-objective evolutionary algorithms were considered to find solutions. For a
quantitative objective, the indoor daylight availability was used as a measurement to allow for the best day-
lighting performance envelope system. For a qualitative objective, a matrix was developed to find the user's
design preference and used it to evaluate and find a quantitative performance goal. The proposed method
provides building facades that satisfy daylighting performance, and most importantly, it allows users to match
their aesthetic sensibilities with the design preference.

1. Introduction believe the provided method limits their creativity.


To resolve the limitation of evaluating the qualitative performance
Unlike in fine arts or engineering, the definition of performance in of a building, decision-making methods were used. Recently, it is
architecture contains both qualitative and quantitative measurements. common practice to find computational tools that help to identify a
Good architecture is judged by how well it performs in both qualitative user's preference and suggest the best items. This proposed method
and quantitative aspects and is comparatively well documented and understands the user's preference and proposes a new building design
developed to measure the quantitative performance of a building based choice.
on numerical measurements such as energy use, structural integrity, By integrating both qualitative and quantitative measures to find
thermal, acoustic, visual comfort, etc. However, it is difficult to mea- the high performance of the building façade, the purpose of this paper is
sure the qualitative performance of a building as it falls in the range of to join two different performances, quality and quantity, into one
the subjective. Furthermore, it is hard to evaluate the performance due measurable goal. The paper considers multi-objective, evolutionary
to each individual's own perspective on the architecture. algorithms to find the solution. For a quantitative objective, an indoor
Although the qualitative performance of a good or bad building is light condition was used as a measurement to allow the best daylighting
difficult to identify, the performance also shifts because of different performance envelope system. For a qualitative objective, the research
factors such as time, space, and culture that are dynamically hard to developed a methodology to find the user's design preference and used
capture in a numerical measure. Recent researches mainly focused on it to evaluate and find a quantitative performance goal.
the area of evaluating building performance have been constrained by
quantitative measures. Optimization of a façade in terms of daylight, 2. Method
energy use, LCA, is frequently published in the journals [1–5]. Most of
these studies are for engineers but rarely used by architects, as they Fig. 1 shows the overall process of the research approach for

E-mail address: ykyi@illinois.edu.

https://doi.org/10.1016/j.buildenv.2019.04.002
Received 4 October 2018; Received in revised form 2 April 2019; Accepted 3 April 2019
Available online 09 April 2019
0360-1323/ © 2019 Elsevier Ltd. All rights reserved.
Y.K. Yi Building and Environment 156 (2019) 178–190

Fig. 1. Overall process.

obtaining both the qualitative and quantitative high-performance multi-object function is satisfied, then that geometric setup is a possible
building facades. The process is divided into three main sections, where solution. If not, then a Genetic Algorithm (GA) will generate the next
the first step of this method is to prepare the Parametric Building population of agent points. By using the agent points in the hierarchical
Geometry (PBG) model and Agent-based Geometry Control System agent system, the geometry can be altered. The simulation should be re-
(AGCS) of a façade created with the use of a hierarchical agent system. run to see if the changed geometry will end up fulfilling the multi-ob-
This allows for specific points to be moved with fewer points. There- jective function. Until the geometry satisfies the multi-objective func-
fore, a much smoother transition can be made between geometry shifts tion, this process should be repeated until the goal is satisfied or reaches
compared to controlling each point separately. the maximum iteration.
After the geometry is set up, the next step is to simulate the model to In summary, based on a PBG model, initial random values (vari-
obtain both qualitative and quantitative performance. As seen in Fig. 1, ables) in an AGCS were generated and passed to the next step to update
the qualitative and quantitative measures were obtained differently. For the building geometry. Once the building geometry was changed, si-
the quantitative measurement, daylight conditions are tested with ad- mulation tools were used to calculate the performance. These perfor-
vanced simulation tools. These simulation tools calculate the dynamic mance outcomes were passed to the evaluation process where the ob-
daylight performance metrics used in the objective function in the next jective functions determine whether they satisfy the goals. If the
step. objective function values did not meet the requirements, the next po-
In terms of qualitative evaluation, a measurable index was first pulation is generated based on the selection process and passed to AGCS
developed to evaluate the aesthetic performance of each geometry. to generate a new building geometry to evaluate the next generation's
First, the test building's façade was analyzed to define its geometrical performance. This loop continues until it reaches the goal. The fol-
terms. After the analysis was performed, components of the façade lowing sections discuss this in more detail.
system were categorized to develop as questions for users. Their re-
sponses were analyzed to find numerical measurements for aesthetical
preferences on the façade systems. Once the measurement method is 2.1. Agent-based Geometry Control System (AGCS)
developed, it can be used to convert aesthetical values of façade geo-
metry into numeric values, which will again be used for the objective Generating the building geometry to test the performance consisted
function in the next step. of two main steps: 1) the creation of a Parametric Building Geometry
After all the numerical values are obtained from the simulations, the (PBG) model, and 2) the assignment of variables using the Agent-based
next step is the optimization phase. The data obtained from the pre- Geometry Control System (AGCS). A PBG model was created to produce
vious step is plugged into a Multi-Objective Optimization (MOO). If the the potential building façade based on user-defined variables and
constraints that include shade shape, opening, height, orientation, etc.

179
Y.K. Yi Building and Environment 156 (2019) 178–190

be applied to different parameters such as shade width (w), angle (a),


and height (h). Thus, the integration of the PBG model and AGCS allows
an efficient generation of the shading layout alternatives using few
variables, regardless of the number of shadings, and ultimately leads to
a successful optimization process. A more detailed explanation and its
application can be found in the author's previous papers [8].

2.2. Quantitative performance evaluation

Once the geometry is built, the next step is evaluating its perfor-
mance. For quantitative objectives, Spatial Daylight Autonomy (sDA)
and Annual Sun Exposure (ASE) as measurements for daylight perfor-
mance were used.
sDA is an upgraded and updated version of DA (Daylight Autonomy)
and shows the percentage of the area above 300 lux for 50% of the
occupied hours. Although sDA does not incorporate glare or direct sun
exposure, it has been verified to predict daylight availability showing a
single number for a building [9]. More specifically, sDA values can
range from 0% to 100% of the space. Regarding the interpretation of
the sDA value, a sDA value of 75% indicates a space in which daylight
Fig. 2. Variables and constraints in a conceptual PBG model.
availability is “preferred” by occupants; that is, occupants are able to
work comfortably without the use of electric lights and find the daylight
Fig. 2 is an example of the variables and constraints in a PBG model. levels (intensity) to be sufficient. A sDA value between 55% and 74%
Assume three rectangular shades are placed on a façade. At the sche- indicates a space in which daylight is “nominally accepted” by the
matic design stage each shade's width (w), depth (d), height (h), and occupants. Therefore, architects or lighting designers should try to
angle (a) are the main parameters for the layout of the shade. Other achieve sDA values of 75% or higher in regular spaces, such as in offices
shading shapes and their related parameters can be also integrated with or classrooms, and at least 55% in areas where some daylight is im-
the PBG model. A set of constraints for each parameter is required to portant. LEED (Leadership in Energy and Environmental Design) v4 is
generate practical building layout options. Consideration of building or required to achieve 50% or above to receive points for daylight criteria
zoning code requirements, minimum and maximum heights, and the [10].
range of width and depth should be well defined in order to be im- To compensate for the limitation of sDA regarding direct sunlight
plemented in practice. and glare, ASE was also used. ASE is the number of hours per year at a
Although various variables related to the shading layouts can be given point where direct sunlight is an incident on a surface that might
included in the PBG model, a large number of variables tend to result in cause discomfort, glare, or increase the cooling loads in summer. Based
the failure of the optimization process, inaccurate outcomes, and ad- on IES (Illuminating Engineering Society) LM-83 [11], ASE is defined as
ditional computational costs [6]. AGCS is a way to control multiple the percentage of square footage that has direct sunlight (at least 1000
variables efficiently by using a few agent points that define the hier- lux) for more than 250 h a year. According to LEED v4, it is re-
archical relationships between the agent points and the variables [7]. commended when designing a space with an ASE value of less than 10%
Fig. 3 illustrates the concept of AGCS for building layouts in which [10].
the coordinates of an agent point can have mathematical relationships
with the coordinate vector of each building (sub-agent point) to control 2.3. Qualitative performance evaluation
its location. For example, if the agent point moves horizontally with the
magnitude of displacement (U), the agent point dynamically updates To measure qualitative performance, a method from an expert
the coordinate vector of sub-agent points and alters the shape of the system was adapted. An expert system is a computer system that
shadings. Based on the user's strategy, displacement (U) can be mag- emulates the decision-making ability of a human expert [12]. Expert
nified with different weight factors (Uxa, Uxb and Uxc) for each in- systems are designed to solve complex problems by reasoning through
dividual or group of sub-agent points (shadings). The same process can bodies of knowledge, represented mainly as If–Then rules.
Due to logical reasoning, expert systems can be used for building
design decisions in general architectural practice composed of expert
consultants in specialized domains to execute a design. For that reason,
these systems have been developed for solving a variety of problems
related to architecture [13–15]. In the late 1980s to mid-1990s, expert
systems were frequently researched and developed; architectural code
compliance tools were especially popular early expert system types
[16–18]. More sophisticated and advanced expert systems combined
with other domains can be found in more recent studies, such as an
expert system with a genetic algorithm to optimize the layout planning
of multi-story buildings [19] and a tool developed in relation to day-
light performance [20–22].
Expert systems typically consist of a knowledge base containing
domain-specific knowledge, and an inference engine that applies the
knowledge to a user-specified problem to determine a solution. One
significant benefit of expert systems is their explanation capability.
Because any action or decision made by an expert system is based on
logical reasoning using a set of domain-specific rules, an expert system
Fig. 3. Concept of the AGCS in the PBG model. can explain why it chose to act in a certain way based on an initial set of

180
Y.K. Yi Building and Environment 156 (2019) 178–190

Figure 5. Concept of Pareto-optimal front.

measures are used to promote a diverse front of non-dominated solu-


tions [30].
The process starts with select parents for the next generation using
the selection function of the current population. The next generation is
then created from selected parents by mutation and crossover, scoring
the children by calculating their objective function values, and its
feasibility. Combining the current population and the children into one
matrix extends the population, and then the rank and crowding dis-
Fig. 4. Process of expert system. tance for all individuals in the extended population are computed,
followed by trimming the extended population to retain the appropriate
user input. number of individuals of each rank. The process will stop if the current
Fig. 4 shows the overall process of expert systems. To communicate generation average of the relative change in the value of the spread
with the user (non-expert), a user interface is required to present over the value is less than the tolerance, and the final spread is less than
questions to the user and then accepts the input. A knowledge base the mean spread over the past generations or the maximum number of
contains data and facts in the specific knowledge domain containing the generations is exceeded. If the process does not satisfy the condition,
knowledge base prepared by a knowledge engineer. From the input, the next step is to repeat the first step of the loop to continue the
inference engines match the user's input with data in the knowledge process until it satisfies the goal [31].
base to reach appropriate answers. This is done using If-Then rules. As in Fig. 6, the MOO process starts with the selection of the parents
for the next generation using the selection function on the current po-
pulation (k). The next generation is created from the selected parents by
2.4. Multi-Objective Optimization (MOO)
mutation and crossover, and by scoring the children by calculating their
objective function values and feasibility. The next process combines the
In the field of building performance simulation, various optimiza-
current population and the children into one matrix, the extended po-
tion methods have been used to solve single and multi-objective pro-
pulation. The extended population computes the rank and crowding
blems [23]. The multi-objective optimization (MOO) has two different
distance for all individuals; then the extended population is trimmed to
approaches: the classical methods and the non-traditional methods
retain the appropriate number of individuals of each rank. The process
[24]. In the classical methods, mathematical principles were used to
will stop if the current generation average of the relative change in the
convert MOO into a single scalar objective problem by integrating a
value of the spread over value is less than the tolerance, and the final
user-provided weight for each objective (e. g., the weighted sum
spread is less than the mean spread over the past generations or the
methods) [25].
maximum number of generations exceeded [32].
The non-traditional methods use stochastic rules to find a set of
Pareto fronts [24]. Pareto-optimal solutions are the non-dominated
solution sets within the entire feasible decision variable space. A
3. Application to design process
Pareto-optimal front corresponds to the boundary of the feasible ob-
jective function space, representing a set of good trade-offs between
3.1. Test case
two or more conflicting objectives (see Fig. 5) [26]. The non-traditional
methods could generate multiple trade-offs using a single optimization
Fig. 7 shows the test building, the Suites Avenue Building, selected
and address a large search space compared to the classical methods
by the author. Designed by Toyo Ito [33], it is located at Passeig de
[27]. The Genetic Algorithm (GA), an evolutionary algorithm, is widely
used in the field of multi-objective building performance optimization
by integrating parametric modeling platforms [28,29].
For this paper, Non-dominated Sorting Genetic Algorithm II (NSGA-
II) was used for the multi-objective optimization process. The NSGA
algorithm is an evolutionary MOO algorithm that improves the adap-
tive fit of a population of candidate solutions to a Pareto front con-
strained by a set of objective functions. The algorithm uses an evolu-
tionary process with surrogates for evolutionary operators including
selection, genetic crossover, and genetic mutation. The population is
sorted into a hierarchy of sub-populations based on the ordering of the
Pareto dominance. Similarity between members of each sub-group is
evaluated on the Pareto front, and the resulting groups and similarity Fig. 6. Evolution algorithm selection process.

181
Y.K. Yi Building and Environment 156 (2019) 178–190

Fig. 7. The suites avenue building by Toyo Ito [33].

Gracia in Barcelona, Catalonia, Spain. The complexity of Ito's façade


motivated the author's selection. With the advancements in computa-
tional power and algorithms, the appearance of recent building façade
design is opposed to orthodox buildings, thus making it difficult to
satisfy the quantitative building performance. For this reason Ito's fa-
çade design was adapted as a study case to showcase the proposed
design methods.
The first step was to develop a PBG model based on the variables to
create a façade geometry. As in Fig. 8, the author analyzed the façade
geometry and developed the following logic as a method to create the
façade geometry. However, this means of creating a façade geometry
can vary depending on the user's preference. A test façade (C) can be
created by subtracting it from the surface A with a group of blocks
labeled B. Surface A is a sine wave shape that has a period (pitch) a and
an amplitude b that can shift the shape of surface A. Block B can be
increased or shrink the hole size by changing the dimension of c and d.
Based on the four geometry elements (a, b, c, d), the test case can
morph to create various forms as shown in Fig. 9. If the period (pitch) a
value changes, it will create different number of waves and amplitude b Fig. 9. Geometry variables to shift the shape of the test surface.
can change the magnitude of the depth. Furthermore, the size of block
B, which is decided by the dimension of c and d, can create different

Fig. 8. Analyzing geometry elements of Suites Avenue Building by Toyo Ito.

182
Y.K. Yi Building and Environment 156 (2019) 178–190

Fig. 10. Graphic scripting to create the AGCS model of the test façade.

sizes and the number of holes in the subtracted surface C. If c and d complex surface with big openings can be defined by four variable
values are reduced, surface C will have more and smaller holes; in values as follows: a = 100, b = 100, c = 0, and d = 100, the reason the
opposite, lesser and bigger holes occur when c and d increase. c value is 0 for the big opening is that if c and d both are 100, it will
Based on the geometry logic, a PBG model was used to build the have a fewer number of opening areas in the same surface area. For that
Agent-based Geometry Control System (AGCS). To build the AGCS reason, the c value was not increased in order to keep a similar number
model, the author used a CAD (Computational Aided Design) tool called of openings in the façade area. With these values the six different cases
Grasshopper [34] to build the geometry model as shown in Fig. 10. This can be defined by the numeric values used to define the preferred look
model allows for the modification of a façade shape based on the four of the skin's objective functions in the next section (Fig. 12).
variables defined above.
3.3. Multi-objective optimization function

3.2. Implementation of expert system for measure of qualitative In practice, design options tend to be governed by building codes to
performance meet various environmental requirements, especially in commercial
buildings built in densely populated city areas, where closely built
The inference rules used in this test case asked for a preference look surrounding buildings cause limited access to the sky and direct sun-
of the test facade. Since the parameters to shift the surface was gov- light. To this extent, in this section, test cases for building skin opti-
erned by the two elements which were a morph of the waves on surface mization were conducted to present the potential application of the
A and the size of block B, the query composed two questions for the integration of both qualitative and quantitative performances in the
user. The first question concerned the number of waves and its ampli- design process in terms of daylight accessibility and appearance.
tude. A more complex shape means having a higher number of waves Six test cases were conducted by following the workflow as shown
and a bigger amplitude. In opposite, a simple shape means less waves in Fig. 1. The same PBG model was used to find better design options
and a smaller amplitude. The second question refers to the size of the regarding the daylight and preference of the surface look decided by the
opening; if the user's response to the opening is big, then the pattern user. Moreover, the test used the same MOO functions to determine the
will be enlarged to create bigger holes; conversely, the pattern will optimum building skin design used to optimize sDA, ASE, and the user's
shrink to create smaller holes if the answer is that the openings are preferred skin look. The objective function for all case studies can be
small. written as follows:
Fig. 11 shows the interface responding to the query. The top left
window shows a shape that is a complex form with big openings; the min f (x ) = [f1 (x ), f2 (x ), f3 (x ) ]
x X (1)
top right is a shape that has a complex shape with small openings. The
two bottom shapes are simple shapes with big openings on the left side Where,
and small openings on right side. The query comes with a 3D view that n
(rASE )n
i=1
allows the user to explore the façade shape decided by the queries. The f1 (x ) =
n (2)
3D view shows the shape from different view points and the ability to
zoom in and out to see more detail which helps to chose the user pre- n
(rsDA) n
ferences. This interface is built using a grasshopper plugin called f2 (x ) = i=1
n (3)
Human UI [35].
Once the user selects the preferred look, an expert system proceeds The goal of the case building was to find a solution that minimizes
to the inference engine to find the variable values for different re- all three objectives. The f1 (x ) is the average ASE value of all measured
sponses by following a decision tree used in the inference engine to floors (n) in the building and f2 (x ) is also the average value of sDA for
decide what the user's preferable skin look. Once the expert system is all floors (n). For sDA to maintain consistency with the other two ob-
able to find the user's qualitative preference, its preference look of the jectives ( f1 (x ) and ), the outcome of the simulation was multiplied by
skin must be coverted into numerical values used for optimization. For −1 to convert the function from maximization to minimization.
that reason, the four variables defined in Fig. 12 were used to define An objective function for the preference look of a building skin ()
each of the different cases. All four variables (a, b, c, d) were normalized has six different objective functions and is based on the response from
to the same range from 0 to 100. As in Fig. 8, the elements a and b the user's preference of the building façade appearance. As shown in
change the wave of the surface to create a complex or simple surface. Fig. 12, each different choice has a different assigned numeric value of
Elements c and d control the size of the opening. For instance, a a, b, c, and d. For instance, if the user's choice is a Complex and Big

183
Y.K. Yi Building and Environment 156 (2019) 178–190

opening then the assigned variable values are as follows: a = 100,


b = 100, c = 0, and d = 100. Based on the variable values, the fol-
lowing equations are developed for each case to minimize the objective
function. For a Complex and Big opening, the sum of a, b, d is sub-
tracted by 300 to make a minimum value that is closed to the choice.
The same logic is used for the other five cases, and one of following six
functions is used for the objective function for based on the user's
choice. If the preference skin look of the skin is complex and big,
f3 (x ) = 300 (a + b + d ) (4)
Else, if the skin look is complex and small,
f3 (x ) = a +b 200 + c + d (5)
Else, if the skin look is medium and big,
f3 (x ) = b + d 200 (6)
Else, if the skin look is medium and small,
f3 (x ) = b 100 + d (7)
Else, if it is simple and big,
f3 (x ) = b + d 100 (8)
Else, if it is simple and small,
f3 (x ) = b + d (9)

3.4. Test case boundary condition and model setup

To simulate daylight conditions to measure ASE and sDA, a com-


putational tool called DIVA [36] to simulate both measurements is
used. DIVA uses Radiance [37] as its daylight simulation engine to
calculate several dynamic daylight performance metrics based on sky
conditions acquired from specific location weather data. In addition to
the computed two values from DIVA, the user's preference look of the
skin, which is received from the interface created by Human UI, was
calculated using the script in Grasshopper. Theses calculation results
were passed to Matlab for MOO and the results were sent back to Rhino
to show the solution geometry.
The actual building is located in Barcelona, Spain where the test
skin faces a northeast orientation without any direct sun throughout the
year (41.3851° N). For this paper, the test building is relocated to
Champaign, IL, USA (40.1164° N, 88.2434° W) with a south orientation
so that the skin could be utilized as a daylight control system to im-
prove the indoor light condition. The test façade is attached to the four-
story building with a building size 14 × 18.5 × 23.2 m. It has 100%
window coverage on the south façade where the test skin is attached
(see Fig. 13).
Table 1 shows the daylight model setup and simulation parameters
used for the daylight simulation. The base case was set up with the same
skin shape as the current building and was compared to find im-
provements in terms of sDA and ASE results.

4. Test results

The six tests were conducted and included all the possible choices
the user could select for a preferable appearance of skin (see Fig. 11).
Appendixes A to F show each test result's total scatter plot of the col-
umns of one matrix against another measure.
Fig. 14 shows six different cases of the Pareto optimal front of the
Fig. 11. User interface shows the questions and skin shape based on user's three objectives and the shape of the skin. Overall, the solution shows a
choice.

184
Y.K. Yi Building and Environment 156 (2019) 178–190

Fig. 12. Decision tree used in the inference engine.

Table 1
Radiance setup.
Boundary Condition

Name Set up

Window void glass Tau_vis = 0.47


Glazing_TriplePane_Krypton_47
0 SHGC = 0.23
0 visual transmittance:
47%
3 0.5135 0.5135 0.5135 visual transmissivity:
51.35%
Wall/Roof void plastic GenericInteriorWall_50 standard grey wall
0 reflectivity of 50%.
0
5 0.5 0.5 0.5 0 0
Floor void plastic TileFloor_40 standard floor
0 reflectivity of 40%
0
5 0.4 0.4 0.4 0 0
Skin (Shade) void plastic OutsideFacade_35 a standard reflectivity
0 of 35%
0
5 0.35 0.35 0.35 0 0
Simulation -aa .15 -ab 2 -ad 512 -ar 256 -as 128 -dr 2 -ds .2 -lr 6 -lw .004 -dj 0
Setup -lr 6 -sj 1 -st 0.15
Fig. 13. Test building dimensions (unit: m).

very close shape appearance to what the user selected at the beginning. closest to the shape the architect designed. As mentioned in the pre-
The simple and complex cases show clear differences between the ap- vious section, the actual building faces north, whereas the building was
pearances with size openings. Some of solutions from the medium relocated and oriented so that the test case faced south. Compared to
shape show less success compared to the complex and simple cases. The the base case, except for the simple and small case, sDA values were
reason can be found from the preference look of the skin objective increased. Moreover, the ASE value was increased; this is related to
function ( f3 (x )) , as its definition for medium shapes was not able to having more sunlight that also increases the possibility of thermal and
capture the user's preference look of the skin. Although the medium visual discomfort. For simple and small cases, the shape's opening area
shape solution shows a less successful outcome, it clearly shows the is less than the base case, which is the reason for lower sDA and ASE
difference of having big or small openings. Furthermore, compared to compared to the base case. The preference look of the skin value shows
the other four cases, these two cases can be indefinable as a medium the magnitude in how close it is to the initially selected shape based on
shape. the objective function . A closer zero means more similarities to the
Fig. 15 more visibly shows the differences between the six solutions initially selected shape. The result indicates that the complex and big
and the initial shape the user selected. It shows one of the solutions opening case is the most different from the other cases.
(right) from the Pareto-front and compares it to a selected shape (left)
in the query interface. Overall, the comparison shows a very close si-
5. Discussion and conclusions
milarity between the solution and the initial choice. Table 2 shows the
solution's objective values. The base case is a simulation result with the
This paper proposes a design process that can satisfy two different

185
Y.K. Yi Building and Environment 156 (2019) 178–190

Fig. 14. Six test cases of a Pareto optimal front with a selected solution appearance.

186
Y.K. Yi Building and Environment 156 (2019) 178–190

rational basis spline) system that allowed the manipulation and control
of the geometry. Based on the design principles, the building skin was
identified to six different design preference looks of the skins from an
expert system. These six preferences looks of the skins were converted
into an equation that was used as the objective function for the test.
The proposed methodology then tested the six different design
preference looks of the skins to see how the solutions were similar to
the user's initial choice. As shown in Table 2 and Fig. 15, overall the test
shows promising results; the Pareto-front solutions show a similar shape
to the initial choices. Some cases, such as the medium with a big
opening, showed less agreeable shapes compared to the other cases as
shown in Fig. 15. However, medium and big opening shape are clear
enough to identify the other cases as in Fig. 15.
In terms of other two objectives (sDA and ASE), the test result shows
improvement compared to the base case that was the same shape as the
architect first designed in terms of the sDA result. However, for a simple
and small opening the results of sDA were not better than the base case
because the openings were smaller than the base case, thus making it
difficult to achieve a better result. For ASE, the results show an increase
as it received more daylight compared to the base case selected.
It is worthwhile to observe the appendices that show each case's
scattered plot of the columns of one measure against another measure.
Plots can be used to inform an educated user which configuration will
perform better than others. For example, the plots can indicate a
complex and big style that could perform better than other configura-
tions. This could be further developed into proposed methods, not just
finding one close to the initial preference but also suggesting a better
performance configuration.
It is important to note that this is the author's first attempts to
measuring qualitative performance in a numeric value using an expert
system method proposal that contains certain limitations and requires
more research to elaborate the benefit of the proposed method.
However, the scope and several limitations of this work offer oppor-
tunities for future research.

Fig. 15. Initial appearance (left) and one of the Pareto front solution appear-
• This study utilizes an expert system for testing the quantitative as-
pect of performance. However, the result shows limitations in an
ances (right). expert system that could be improved. Proposed equations need
further research to find a method for making equations more
Table 2 straightforward and generalized for various conditions.
Selected cases objectives value. • The sDA and ASE results did not clearly show significant improve-
32 sDA 33 ASE 34 Preference look of the skin
ment in terms of daylight condition from the base case. The test case
façade typology might not be suitable for the South orientation and
35 Base case 36 24.3 37 20.7 by using these shading options the best achievable sDA and ASE
39 Complex and big 40 42.9 41 36.4 42 70 values can be found in Table 2. Use of the other indexes is desirable
43 Complex and small 44 33.6 45 27.9 46 12
47 Medium and big 48 36.4 49 35.0 50 5
to improve the methodology's strength, and it will be beneficial to
51 Medium and small 52 32.1 53 27.9 54 22 use other measures.
55 Simple and big 56 35.0 57 33.6 58 5 • This work focused on daylight conditions. Other environmental
59 Simple and small 60 10.0 61 16.4 62 1 factors can be included such as natural ventilation, thermal comfort,
and building energy use.

types of building performance, that is, the qualitative and quantitative


This research serves as a starting point for implementing qualitative
performances of the building. Quantitative performance measures are
performance as a measurement for building optimization. Therefore,
well defined and popularly used in architecture; however, qualitative
the study focused on using a qualitative measure to overcome the
performance has rarely been studied. For that reason, this paper pro-
shortfalls of currently available methods. Furthermore, the proposed
poses utilizing an expert system to evaluate the appearance of the
qualitative performance measure was integrated with a parametric
building as a qualitative performance.
modeling environment. This method could simultaneously be used to
As mentioned, the design objectives were to satisfy the user's visual
calculate both qualitative and quantitative performances that can then
favor but also to effectively filter the sunlight from entering the
be applied at the early stages of the building design process with sup-
building by adjusting the building skin geometry in response to varying
port from MOO. This research requires further development if the new
outdoor and sky conditions. To measure these design objectives, sDA
method is to be applied to more practical and complex test cases, thus
and ASE were used to measure quantitative performance. For the user's
showing its ability to support design and refining its measurement
visual preference look of the skin of the building design, the test
method to improve its accuracy.
building design was analyzed, and the design principles identified.
These design principles were programed to the NURBS (Non-uniform

187
Y.K. Yi Building and Environment 156 (2019) 178–190

Appendix A

Appendix 1Complex and big style

Appendix 2Complex and small style

Appendix 3Medium and big style

188
Y.K. Yi Building and Environment 156 (2019) 178–190

Appendix 4Medium and small style

Appendix 5Simple and big style

Appendix 6Simple and small style

References [4] Elarga Hagar, Andrea Dal Monte, Rune Korsholm Andersen, Ernesto Benini, PV-
PCM integration in glazed building: Co-simulation and genetic optimization study,
Build. Environ. 126 (2017) 161–175.
[1] Doris A. Chi, David Moreno, Jaime Navarro, Design optimisation of perforated solar [5] J. Sargent, J. Niemasz, C. Reinhart Shaderade, Combining rhinoceros and
façades in order to balance daylighting with thermal performance, Build. Environ. EnergyPlus for the design of static exterior shading devices, Proceedings of Building
125 (2017) 383–400. Simulation 2011, 12th Conference of International Building Performance
[2] Klodian Gradeci, Nathalie Labonnote, Berit Time, Jochen Köhler, A probabilistic- Simulation Association, Sydney, 14-16 November, 2011, pp. 310–317 (Sydney,
based methodology for predicting mould growth in façade constructions, Build. Australia).
Environ. 128 (2018) 33–45. [6] A.-T. Nguyen, S. Reiter, P. Rigo, A review on simulation-based optimization
[3] Alex Vlachokostas, Nicholas Madamopoulos, Daylight and thermal harvesting methods applied to building performance analysis, Appl. Energy 113 (2014)
performance evaluation of a liquid filled prismatic façade using the radiance five- 1043–1058.
phase method and EnergyPlus, Build. Environ. 126 (2017) 396–409. [7] Y.K. Yi, A.M. Malkawi, Optimizing building form for energy performance based on

189
Y.K. Yi Building and Environment 156 (2019) 178–190

hierarchical geometry relation, Autom. ConStruct. 18 (6) (2009) 825–833. (Eindhoven, Netherlands).
[8] Y.K. Yi, H. Kim, Agent-based geometry optimization with genetic algorithm (GA) [22] J. Gagne, An Interactive Performance Based Expert System for Daylighting in
for tall apartment's solar right, Sol. Energy 113 (2015) 236–250. Architectural Design, Ph. D. thesis MIT, 2011.
[9] J. Mardaljevic, L. Heschong, E. Lee, Daylight metrics and energy savings, Light. Res. [23] V. Machairas, A. Tsangrassoulis, K. Axarli, Algorithms for optimization of building
Technol. 41 (3) (2009) 261–283. design: a review, Renew. Sustain. Energy Rev. 31 (2014) 101–112.
[10] Council, U.S.G.B., LEED v4 user guide, https://www.usgbc.org/resources/leed-v4- [24] P.K. Shukla, K. Deb, S. Tiwari, Comparing classical generating methods with an
user-guide , Accessed date: December 2014. evolutionary multi-objective optimization method, in: C.A.C. Coello Coello,
[11] illuminating Engineering Society of North America, IES LM-83-12, illuminating A. Hernández Aguirre, E. Zitzler (Eds.), Proceedings from the Evolutionary Multi-
Engineering Society of North America, 120 Wall Street, New York, New York, 2012 Criterion Optimization: Third International Conference, EMO 2005, Springer,
10005. Berlin, Heidelberg, 2005, pp. 311–325 Guanajuato, Mexico, March 9-11, 2005.
[12] Peter Jackson, Introduction to Expert Systems, third ed., Addison Wesley, Boston, [25] R.T. Marler, J.S. Arora, The weighted sum method for multi-objective optimization:
MA, 1998. new insights, Struct. Multidiscip. Optim. 41 (6) (2010) 853–862.
[13] J.G. Doheny, P.F. Monaghan, IDABES: an expert system for the preliminary stages [26] A.L. Jaimes, S.Z. Martınez, C.A.C. Coello, An introduction to multiobjective opti-
of conceptual design of building energy systems, Artif. Intell. Eng. 2 (4) (1987) mization techniques, Optim. Polym. Process. (2009) 29–57.
54–64. [27] E. Zitzler, Evolutionary Algorithms for Multiobjective Optimization: Methods and
[14] P. Fazio, C. Bedard, K. Gowri, Knowledge-based system Approach to building en- Applications, Swiss Federal Institute Technology, Zurich, Switzerland, 1999.
velope design, Comput. Aided Des. 21 (10) (1989) 519–527. [28] M. Rahmani Asl, S. Zarrinmehr, M. Bergin, W. Yan, BPOpt: a framework for BIM-
[15] C. Robin, J. Brau, J. Roux, Integration of expert knowledge and simulation tools for based performance optimization, Energy Build. 108 (2015) 401–412.
the thermal design of buildings and energy systems, Energy Build. 20 (2) (1993) [29] R. Vierlinger, K. Bollinger, Accommodating change in parametric design,
167–175. Proceedings of Association for Computer-Aided Design in Architecture 2014
[16] M.A. Rosenman, J.S. Gero, Design codes as expert systems, Comput. Aided Des. 17 International Conference, 2014, pp. 609–618 (Los Angeles, CA, USA).
(9) (1985) 399–409. [30] J. Brownlee, Clever algorithms: nature-inspired programming recipes, Available at:
[17] C.L. Dym, R.P. Henchey, E.A. Delis, S. Gonick, A knowledge-based system for au- http://www.cleveralgorithms.com/.
tomated architectural code checking, Comput. Aided Des. 20 (4) (1988) 137–145. [31] MathWorks, gamultiobj, Available at: https://www.mathworks.com/help/gads/
[18] H.W. Chun, E.M.K. Lai, Intelligent critic system for architectural design, IEEE Trans. gamultiobj.html#bvg1lh0-5.
Knowl. Data Eng. 9 (8) (1997) 625–639. [32] K. Deb, A. Pratap, S. Agarwal, T. Meyarivan, A fast and elitist multiobjective genetic
[19] P.K. Mashood, C.S. Krishnamoorthy, K. Ramamurthy, KB-GA-Based hybrid system algorithm: NSGA-II, IEEE Trans. Evol. Comput. 6 (2) (2002) 182–197.
for layout planning of multistory buildings, J. Comput. Civ. Eng. 21 (7) (2007) [33] http://www.toyo-ito.co.jp/WWW/Project_Descript/2005-/2005-p_16/2005-p_16_
229–237. en.html.
[20] B. Paule, J. Scartezzini, Leso-dial, a new computer-based daylighting design tool, [34] https://www.grasshopper3d.com/.
Proceedings Of Right Light 4, 1997, pp. 93–97 (Copenhagen, Denmark). [35] https://www.food4rhino.com/app/human-ui.
[21] E. de Groot, L. Zonneveldt, B. Paule, Dial europe: a decision support tool for early [36] https://www.solemma.com/Diva.html.
lighting design, Proceedings of Building Simulation, 2003, pp. 421–426 [37] https://www.radiance-online.org/3w.

190

You might also like