Building and Environment 197 (2021) 107828

Contents lists available at ScienceDirect

Building and Environment

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

A daylight-oriented multi-objective optimisation of complex

fenestration systems
Islam A. Mashaly a, *, Veronica Garcia-Hansen a, Michael E. Cholette b, Gillian Isoardi c
a
School of Architecture and Built Environment, Faculty of Engineering, Queensland University of Technology, 2 George St, GPO Box 2434, Brisbane, Qld, 4001,
Australia
b
School of Mechanical, Medical and Process Engineering, Queensland University of Technology, 2 George St. GPO Box 2434, Brisbane, Qld, 4001, Australia
c
Light Naturally, 1B/17 Peel St., South Brisbane, Qld, 4101, Australia

A R T I C L E I N F O A B S T R A C T

Keywords: Complex fenestration systems (CFS) are effective daylighting strategies that improve daylight performance in
Multi-objective optimisation indoor living spaces. Nevertheless, the design of CFS is complex as it requires specialised knowledge in optical
Complex fenestration systems physics and advanced building simulation. This paper presents an evaluation method (CFStrace) to optimise the
CFStrace
CFS design in a multi-objective optimisation setting. The method takes into account both architectural and
Forward raytracing
Daylighting
environmental conditions that maximise the effect of light redirection for complex fenestration systems to its full
Parametric design potential. The research aims to demonstrate a multi-objective and multi-scenario optimisation methodology for
designing complex fenestration systems. The parameters of the optimisation problem can be easily changed due
to the parametric nature of the modelling environment used. Thus, this gives designers the flexibility to adjust
the CFS parameters and decrease their discretisation. To support the presented methodology, a case study is
presented to demonstrate the design process and guide designers in selecting the best design option. The case
study involves the design optimisation of a sinewave-shaped light redirecting panel by tweaking its geometrical
parameters and comparing the optimised result with existing light redirecting panels. The case study results
indicate that optimising a CFS according to the architectural design can improve daylight metrics (such as
improving the useful daylight illuminance (UDI) by +10%). The case study reveals the importance of addressing
architectural and environmental constraints when designing CFS as the optimised design can differ with
changing parameters such as the room depth or panel height.

1. Introduction redirect the light using either reflection of light on their surfaces or
refraction of light within the medium of the device. Reflection of light is
Indoor daylight quality is a great focus for architects since adequate simpler in terms of calculation as it requires only the basic laws of
daylighting improves the health and wellbeing of occupants as well as reflection which depends on the incident light direction and power, as
increasing productivity [1,2]. However, excessive daylighting can lead well as the surface’s reflectance and scattering. Refraction is a more
to glare and discomfort issues as well as an increase in heat gains, which complicated process, which depends on Snell’s law for refraction and
results in the use of blinds that blocks the view to the outside and in­ involves the medium’s refractive index as well as the other factors
creases the reliance on electrical lighting [3]. Therefore, it is crucial to mentioned above for the reflectance. Successful examples of passive
optimise daylight performance inside the working and living spaces. light redirecting devices, such as 3 M light redirecting films and Serra­
One of the strategies that ensures suitable lighting in the space is the use Glaze redirecting films from SerraLux, have been commercially imple­
of complex fenestration systems [4]. CFS help control the amount of mented in several projects across the United States and Europe [6,7].
direct sunlight in the space as well as achieving better lighting quality Light redirecting films are micro structured prisms that tend to reflect
thought better light uniformity by redirecting it to poorly lit areas in the the light upwards at certain angles (3 M films direct 70% of light up­
space [5]. Some familiar examples of passive CFS are light redirecting wards at 50◦ solar altitude with a 66% transmission [6]). While the
devices, such as light shelves, blinds, films, and panels. Those devices maintenance and intervention of such systems is low; the design must be

* Corresponding author.
E-mail addresses: i.mashaly@qut.edu.au, islam.akm@hotmail.com (I.A. Mashaly).

https://doi.org/10.1016/j.buildenv.2021.107828
Received 3 November 2020; Received in revised form 16 March 2021; Accepted 20 March 2021
Available online 26 March 2021
0360-1323/© 2021 Elsevier Ltd. All rights reserved.
I.A. Mashaly et al. Building and Environment 197 (2021) 107828

optimised to the specific location in which it has to be installed. This luminaries and light transport systems benefiting from the realistic
means the profile of the micro structured prism has to vary from one behaviour of light. However, this comes at the cost of time and
location to the other to optimise the potential light redirected due to the computational power since realistic raytracing requires an infinite
changes of solar angle. Optimising such light redirecting devices is not a number of rays emitted from a source, as in the real life. Therefore, it is
simple task as it requires sophisticated calculations for optimal perfor­ difficult to utilise forward raytracing in the same way backward ray­
mance throughout the whole year. tracing is used to simulate the annual daylight performance and get
CFS are usually overlooked as a daylighting solution for buildings as similar results.
challenges arise in their design process due to the various performance The following table summarises the previous research done in CFS
factors that must be taken into consideration (e.g. reducing glare, uni­ design and the techniques, objective functions, decision variables and
fying illumination throughout the area, avoiding heating, allowing external variables the research included.
views to the outside) as well as the numerous fenestration system design As seen in Table 1, all research studies that involved forward ray­
options available nowadays [8]. These factors need a design optimisa­ tracing had a combination of transmitted flux direction and power in
tion methodology to balance potentially competing design requirements their optimisation objective function. These are the two main primitive
in order to improve wellbeing at work and at home in indoor spaces. The outputs of any forward raytracing simulation. In addition, most forward
emergence of technologies such as parametric design and building in­ raytracing engines are made for optical design in general and usually do
formation modelling (BIM), have created new workflows that make the not cover any architectural output features. On the other hand, back­
design process more efficient and simpler. One design workflow ward raytracing software (like Radiance) is more suited to architecture
improved by parametric technology is façade design and optimisation. design, which is why [12] its objective function involved several
Several researchers have enhanced façade designs to be responsive to architectural daylight matrices as the sDA and Annual Sunlight Exposure
the climate and daylight performances through parametric design and (ASE). Most research in CFS design optimisation incorporates similar
optimisation [9,10] With CFS, however, replicating such performances CFS geometric decision variables (such as the spacing between repetitive
using common architectural daylight raytracing software (e.g. Radi­ shapes, the profile, and the justification of the repetitive shape)
ance) is a challenge [5] although Radiance is one of the most commonly regardless of the simulation method used.
used daylight simulation software packages in the industry. To limit the The external variables for forward raytracing partially focus on the
daylight simulation time, Radiance uses backward raytracing, which is design’s geographical location through simulating certain angles that
inexact in tracing refractions and internal reflections in complex ge­ occur at special times of the year or favoured angles with more frequent
ometries [5]. To optimise the design variables of a CFS in Radiance, a occurrences [14–16]. However, research involving forward raytracing
bidirectional scattering distribution function (BSDF) has to be generated does not consider the geometrical and physical external environment
for each design iteration using forward raytracing, which can be that we are designing the panel for due to the requirement of higher
computationally intensive [11]. The combination of a backward ray­ computational power in comparison to the backward raytracing. One of
tracing technique in Radiance and the replacement of the fenestration the few exceptions is Isoardi’s design methodology which partially
system with a BSDF is called the Radiance Phase Method. This method considers the spatial design of the indoor space when redirecting light to
has its pros and cons: using a BSDF database to replace the need to the ceiling and avoiding glare on the occupant’s field of view [13].
generate a BSDF reduces the simulation time, on the other hand it will In summary, the current CFS design methodologies depend on either
limit the creation of new CFS possibilities [12]. utilised the Radiance forward raytracing or hybrid (forward-backward) raytracing simula­
Phase Method for optimising the design of a specific CFS shape. The tions such as the Radiance Phase Method. The gap in forward raytracing
researchers generated different design shapes with a fixed number of is the inapplicability of daylight analysis within architectural and
design variable combinations resulting in 1728 designs before running geographical contexts. While the downfall of hybrid raytracing is the
an optimisation algorithm to find the optimum solution within these complexity of the method as it involves multi-layered methodologies
1728 combinations [12]. The optimum solution was found within the and tools, which can be exhausting for non-experts [11]. Both methods
first 100 iterations in the optimisation process. have their merits, and the choice is dependent on the designer’s capa­
Other efforts in optimising CFS include the implementation of bilities with and knowledge of advanced daylight design systems. The
daylight simulation software that uses forward raytracing such as designer’s knowledge of CFS design poses potential limitations in the
TracePro, Zemax, and other simple forward raytracing tools [13–16]. design produced in terms of improvement in energy savings, daylight
Forward raytracing is used in the optical design of products such as comfort, and visual comfort.

Table 1
Summary of key research in CFS design.
Author(s) Isoardi, G. Tsangr-assoulis et al. Kostro et al. Uribe et al. Manzan et al. Mashaly et al.

Function Year 2009 2013 2016 2017 2017 2018

Simulation Method Forward Raytracing ● ● ● ●
Backward Raytracing ● ●
Optimisation GA GA N/A PSO-HJ FAST GA
Objective Function View to outside ● ●
Power emitted ● ● ●
Direction of out-angles ● ● ● ●
Daylighting metrics ● ●
Energy consumption ● ●
Glare ●
Light distribution ○

Decision Variables Spacing between rep. ● ● ● ● ●

Profile of rep. ● ● ●
Tilt/justification of rep. ● ● ● ● ●
Window to panel ratio
External Variables Location ● ○ ● ○

Space design ○ ●
Schedule & activity ● ●

● = considered; ○ = partially consideration; [empty cell] = not considered.

2
I.A. Mashaly et al. Building and Environment 197 (2021) 107828

Therefore, this research investigates the possibility of using a simpler Grasshopper for Rhino that consists of various components. A forward
and broader approach to include CFS design in the façade design pro­ raytracer was developed using Grasshopper and the Python scripting
cess. The researcher presents a new CFS design optimisation method­ language tool within Grasshopper. The Grasshopper script, shown in
ology that avoids complex processes and, by including the Fig. 2, contains the whole simulation and optimisation processes as well
environmental and spatial context, facilitates customisation. In addition, as the input, and visualization and numerical outputs that support the
a forward raytracing optimisation environment is proposed using a CFS design process in early design phases.
newly developed CFS raytracing simulation and analysis tool, called The raytracing tool, CFStrace, was validated against TracePro, a
CFStrace. The benefit of CFStrace is that it takes into account both the validated forward raytracing software. A model with a sine wave light
geographical and spatial contexts and simulates the direct component of redirecting panel was created in Rhino3D, a 3D CAD modelling soft­
daylight, which is the main target for several CFS [17]. By considering ware, through Grasshopper and imported to TracePro to test the validity
the geographical and spatial contexts, as well as direct daylight, the of the raytracing algorithm. The models in CFStrace (inside Grass­
customised CFS design will be optimised to each design case differently. hopper) and TracePro were tested using eight different solar altitudes
Moreover, the methodology tackles the time and computational power (10◦ –80◦ ) with equal number of rays with the same power directed to­
constraints that occur in a backward raytracing simulation by using a wards the panel. Around 88–90% of the rays in both models were
combination of forward raytracing and dynamic (annual) daylight directed in a similar direction and power with a root mean square error
interpolation, thus allowing a more flexible CFS design process. In short, for the rays power of 2.8% [17]. Fig. 3 shows a section view of the
the proposed method offers a simpler, faster, and more comprehensive raytracing occurring for both software for selected solar altitudes.
‘architect-friendly’ assessment process which is currently missing in the The following section provides a detailed description of the design
daylight design field. The proposed CFStrace tool will encourage de­ method, optimisation problem statement, and analysis method.
signers and architects to consider CFS in their design process and opens
the possibility for new CFS designs that could potentially improve
2.3. Design inputs for the simulation environment
daylighting harvesting, visual comfort, access to views while reducing
electrical light consumption.
The design conditions are set to consider the spatial and environ­
mental setting to approximate the CFS in a real environment. The room
2. Methodology
geometry is generated parametrically using Grasshopper’s sliders or the
room’s boundaries can be imported or drawn and extruded for the
2.1. Summary
height. The window area is determined along with the redirection panel
area. As for the climatic conditions, the location’s latitude and longi­
This paper presents a demonstration case study of the implementa­
tude, and the occupant schedule can be entered using numerical sliders.
tion of the methodology and the CFStrace tool. First, the methodology is
Then, the previous input is used in calculating the occurrences of the
explained: the design inputs, design criteria and optimisation process
range incident angles that will be associated with various daylight
are clarified. Afterwards, the case study procedure and case study results
metrics explained later to determine the performance of a certain CFS
are described and demonstrated.
design on a periodic level. (see Fig. 4)
ωi = occurance of a range of angles in a specified time in a location
2.2. Overview of the methodology
At the end of the input phase, the user can select the type of light
The proposed multi-objective optimisation methodology eases the redirecting system and based on their choice, the design variables are
CFS design process by simulating multiple design solutions by tweaking determined. Types of CFS that can be selected can range from sophis­
the decision variables based on the spatial needs and design conditions, ticated light redirecting panels that refract the daylight to light shelves
then displaying the optimum designs and showing the trade-off between and louvers that simply reflect the light.
the different optimum design solutions. The methodology is prepared
through Grasshopper, a visual scripting environment in Rhinoceros that
2.4. Design criteria
can model complex geometry and allows parameters and coding cus­
tomisation. Multi-objective functions are developed from daylight-
The user will then be able to select and prioritise the design criteria.
related performance criteria such as daylight illumination, uniformity,
These criteria are based on specific objectives from previous research’s
and glare probability. Fig. 1 details the methodology used to optimise
CFS designs and intents [18]. The main objectives of adding a redi­
CFS through CFStrace.
recting panel in the indoor space are:
CFStrace is a comprehensive simulation and analysis tool within

1 Designing a redirecting panel that reduces direct sunlight falling on

occupants, especially in summer time [13,19,20].
2 Improving the daylight distribution and uniformity inside the space
[21].
3 Improving the useful daylight in deep spaces [17].
4 Retaining the view to the outside [22].

To sum up, it is important to design a CFS that is tailored to the

specific space, environment, location, and occupant’s preferences.
Therefore, based on the previous objectives a set of criteria and
metrics are selected to create a tailored CFS that addresses the
objectives:

1 Distribution of light within the depth of the space and useful daylight
availability, see Fig. 5 (left)
2 Glare protection for occupants, see Fig. 5 (right)
Fig. 1. Optimisation methodology workflow. 3 View to the outside, see Fig. 6

3
I.A. Mashaly et al. Building and Environment 197 (2021) 107828

Fig. 2. The developed Grasshopper script (CFStrace) for designing CFS and the output visualisation within the Rhinoceros viewport.

Fig. 3. Sample of the TracePro Results vs. the grasshopper raytracing script for the same settings [17].

4
I.A. Mashaly et al. Building and Environment 197 (2021) 107828

Fig. 4. Factors considered to determine the occurrences of a range of angles in a specific location for a specific time of the day and a specific portion of the year.

Fig. 5. Illuminance mean and standard deviation (left), Glare avoidance criteria (right).

the distribution and daylight uniformity in the space, using the

following equations (1) and (2) [24]:
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
1∑ n
SD(E)i = (Ex − Emean )2 (1)
n x=0

where

1∑ n
Emean = Ex (2)
n x=0

And to measure the light distribution and daylight uniformity per­

Fig. 6. View range. formance along the selected period, the occurrence is factored within
measuring the standard deviation for all solar angles as in equation (3).
Although the design criteria were discussed thoroughly in a previous ∑altitudes∘
rangeof solar
paper [18], for completeness these are briefly presented and used to SD(E) = SD(E)i × occurrence (ωi ) (3)
develop objective functions and constraints for design optimisation solar altitude steps∘

purposes.
2.4.2. Modified useful daylight illuminance (UDI)
2.4.1. The distribution of light within the depth of the space (standard The UDI is a metric commonly used in the daylight assessment of
deviation & illumination at work plane level) built spaces. UDI is a measure of the useful amount (from 100 to 2000
Daylight uniformity along the area of the occupied space is impor­ lux) of daylight throughout a whole year [25]. The metric’s range was
tant in assessing the quality of daylighting inside a space [23]. Having later shifted to between 100 and 3000 lux and refined to UDI autono­
significant variations in illumination values across the room’s depth can mous at 300–3000 lux, in which range most probably there would be no
cause a high contrast between the inside of the room and the area near need for artificial lighting in the space [26]. Here, a modified UDI
the window. This contrast is known as the cave effect where occupants (mUDI) is used since we are conducting an approximated climate
sitting further away from the window might experience discomfort glare analysis. The modified UDI uses the acceptable range of 300–3000 lux to
as well as poor lighting levels on their work planes. To quantify the assess the space; however, since the simulation does not run for each
measure of uniformity for a specific time (specific solar altitude), the hour of the year, an approximated analysis for the acceptable range is
standard deviation is calculated along the room’s depth as a measure of carried on by analysing the percentage of floor area between 300 and

5
I.A. Mashaly et al. Building and Environment 197 (2021) 107828

3000 lux for each solar altitude simulated. According to the occurrences [29] study conducted in Germany and Denmark. Therefore, 40% is
of each solar altitude in the selected climate, the modified UDI is assumed in this study as the threshold for glare. To incorporate the DGPs
calculated. For any angle considered in the simulation, all locations in within the simulation nature of CFStrace the following equation is used.
the room that have illuminance between 300 and 3000 lux are 80∘
∑
accounted for and divided by the total number of daylight sensor points mDGPs(40%) =
No. ​ of ​ points ​ below ​ 40% ​ DGPs
(7)
in the room. The daylight sensor points translate to the percentage of a=10∘
total ​ number ​ of ​ glare ​ analysis ​ points
floor area that meets the illuminance criteria. The following equation
summarises how the mUDI is calculated 2.4.4. View to the outside
rangeof solar altitudes∘
Although the view to the outside can sometimes be a fixed external
∑
mUDI(300− =
No. ​ of ​ points ​ between ​ 300 − 3000 ​ lux variable that is bounded by the CFS redirecting dimensions and position,
3000lx)
solar altitude steps∘
total ​ number ​ of ​ daylight ​ sensor ​ points in other cases it can be considered an objective to maximise. The vari­
× occurrence (ωi ) ation in the panel to window ratio can affect the overall daylight per­
formance inside the space and competes with the previously discussed
∑ altitudes∘
range of solar
=
Floor area between ​ 300 − 3000 lux objectives.
solar altitude steps∘
Total floor ​ area
× occurrence (ωi ) 2.5. The multi-scenario multi-objective optimisation problem
(4)
Using the developed criteria, the design optimisation of a CFS can be
formulated as a four-objective optimisation problem:
2.4.3. Glare
As much as horizontal illuminance daylight metrics, such as the UDI min[ J1 (x) − J2 (x) J4 (x) J4 (x) ]
and daylight uniformity, inform designers with the overall daylight
x

performance in a space, they cannot be used as the only indicators of the subject ​ to :
occupants’ comfort level. Humans respond to the brightness in a visual (8)
ℓ≤x≤u
scene, which affects their visual comfort level if exposed to excessive
brightness or high contrast values for periods of time. The brightness in a where.
user’s field of view in the space cannot be measured using horizontal x ∈ Rn are the n design variables with lower and upper bounds and u,
illuminance alone. Therefore, glare-related metrics, such as the daylight respectively.
glare probability (DGP), are better to measure the occupants’ visual The objectives are computed as weighted sums at different solar
comfort [27,28]. DGP measures vertical daylight illuminance as well as angles:
the glare source luminance from the user’s perspective [28]. The DGP
∑
n
shows a strong correlation with the occupants’ perception of glare in J1 (x) = ωi ⋅J1,i (x) (9)
general. The DGP consists of the vertical illuminance (Ev ) at the analysis i=1

surface and the glare source luminance from the user’s point of view
L2 ωs,i
∑
n
(E1.87
s,i
P2
) and a correction factor. Using the user’s point of view to deter­ J2 (x) = ωi ⋅J2,i (x) (10)
v i
i=1
mine the glare source luminance requires an image-based analysis.
( ) ∑
n
∑ Ls,i (11)
2
ψ s,i J3 (x) = ωi ⋅J3,i (x)
− 5 − 2
DGP = 5.87 × 10 Ev + 9.8 × 10 log 1 + + 0.16 (5)
i
Ev1.87 P2i i=1

where.
where.
ωi = percentage of relevant timeframe spent at the particular solar
Ev = vertical illuminance (lux)
angle (occurrence),
Ls,i = glaresourceluminance(cd /m2 )
J1,i (x) = the standard deviation metric,
ψ s,i = solid angle subtended by the source(sr) J2,i (x) = the mUDI metric,
Pi = Guth’spositionindex J3,i (x) = the glare metric,
and J4 (x) = the panel to window height ratio.
Later, Wienold [29] derived a simplified DGP (DGPs) which takes
into consideration the vertical illuminance (Ev ) only as this has also In order to use standard optimisation solution techniques (e.g., a
shown a reasonable correlation to how occupants perceive glare [29]. genetic algorithm) the optimisation problem (8) can be scalarised. In
DGPs = 6.22 × 10− 5 Ev + 0.184 (6) this study, a combination of ε-constraint scalarisation and weighted sum
scalarisation is employed [31]; V. D [32]. Firstly, the panel size is
where assumed fixed a priori and J4 (x) is removed as an objective. This
assumption is not too restrictive since the panel size may be limited by
Ev = vertical illuminance (lux) other architectural considerations and the optimisation can be re-run at
The simplified DGP (DGPs) is included as one of the CFS assessment different panel sizes to examine its influence. Next, the glare objective is
metrics. Since the vertical illuminance can be calculated from the inci­ considered as a constraint and the main objectives (standard deviation
dent rays on a certain plane, DGPs can be evaluated without the need to of light along the depth of the room, mUDI) are scalarised as a weighted
create an image for glare analysis for the glare source luminance from sum. Adding the glare as a constraint ensures the selection of an opti­
the user’s point of view [30]. mum design with less visual discomfort for occupants but not necessarily
In order to obtain a climate-based and full annual analysis, each DGP the design with the least glare incidents. In this way, the multi-objective
analysis point should be below a certain threshold multiplied by the optimisation problem is scalarised by using 2 objectives as decision
weight of each angle. This threshold is to be determined by the user as objectives and 1 objective is used as a constraint. The resulting opti­
glare perception differs from one person to the other [28]. Initially, 40% misation problem is
DGP was the threshold for uncomfortable glare as stated in Wienold’s

6
I.A. Mashaly et al. Building and Environment 197 (2021) 107828

minαJ1 (x) − (1 − α)J2 (x) 3. Case study: optimisation with a basic design shape
x
subject ​ to (12)
ℓ≤x≤u The purpose of the methodology is to provide designers and archi­
J3 (x) ≤ ε tects with the tools and methods needed to design and optimise the
performance of a CFS according to the design conditions and settings.
for α ∈ [0, 1]. The case study tests the methodology and the multi-objective opti­
misation problem. Light redirecting panels are optimised and tested for
where whether an optimised CFS will give better results than using an out-of-
α = a weight for each objective (explained in the next paragraph) the-box CFS by comparing the optimised CFS with other CFS.
x ∈ Rn are the n design variables with lower and upper bounds ℓ and The case study will act as a guide to designers on how to read the
u, respectively. optimisation results, interpret the daylight metrics, and select the best
design option according to the objectives set.
ε = the glare limit using the modified simplified DGP metric from equation (7)
3.1. Room setup and location
The parameter 0 ≤ α ≤ 1 can be interpreted as the user’s preference
in favouring one objective over the other. If α = 0, objective J2 (x) is
The dimensions of the room were determined from the dimensions of
optimised exclusively, while for α = 1, only objective J1 (x) is consid­
a reference office room that has been used in several daylight simulation
ered. A value between 0 and 1 will optimise a weighted sum of the
analysis research studies before [33–35]. The location selected was
objectives. To obtain a set of Pareto-optimal designs, the optimisation
Brisbane, which has a subtropical climate with a high percentage of sun
problem in equation (11) is solved for a sample of values of α ∈ [0, 1].
visibility.
The resulting set of samples from the Pareto front will be used to assess
the trade-off between the daylight distribution and overall light power
3.2. Sensors’ placement
inside the space for the optimised designs.
The glare constraint is addressed by using the well-known penalty
Fig. 9 illustrates the location of daylight and direct glare analysis
function approach, which results in the non-smooth augmented objec­
points in the room for this case study. The illuminance sensors are placed
tive function:
at 0.4 m along the depth of the room and at a plane height of 0.75 m. The
min αJ1 (x) − (1 − α)J2 (x) + β(max(ε, J3 (x) ) − ε (13) glare analysis points were placed at a sitting height of 1.2 m in three
x
generic locations in the room: in the middle of the room, near the
where β is a large factor to eliminate solutions with any glare factor and window, and further away from the window.
ε is the glare limit. Fig. 7 illustrates the glare penalty factor graph.
While there are many solution techniques available for single 3.3. Window configuration and light redirecting panel design
objective problems, e.g., global optimisation, pattern search method,
genetic algorithm (GA), particle swarm optimisation (PSO), and Hooke- The room’s window facing the equatorial façade (for Brisbane this is
Jeeves (HJ), a GA is selected here due to its simplicity and the fact that north) and has a window to wall ratio of 55%. Table 2 lists all the pa­
the objective function is non-differentiable, which prevents the use of rameters considered for the case study.
some well-known optimisation techniques. The selection of the best The CFS that is used in the case study is an acrylic light redirecting
solver is considered outside the scope of this paper and is left to future panel that has a prismatic shape. The decision variables, shown in
research. Table 3 and Fig. 10, are the justification of the profile, the amplitude of
the prism wave, the fillet of the prism’s corners and the spacing between
prisms. Nevertheless, as the panel size is considered as a constraint,
2.6. Optimisation procedure
separate optimisation runs will be performed for each panel size.
Moreover, the panel-window height and the room depth are vari­
The optimisation procedure is described below and shown in Fig. 8:
ables for multi-scenario optimisation runs to compare the effect of
changing such variables on the optimised designs and pareto front.
• Inputs are split into CFS design inputs and other external inputs such
Fig. 11 illustrates the multi-scenario variables utilised.
as room dimensions, location, and occupant schedule. Those are
The optimised panel for the 0.6 m panel case will be compared
determined in the beginning.
against the existing CFS similar to it and with similar objectives, such as
• The decision variables are also outlined and set as objective functions
better daylighting in the space and less glare, when designed. Fig. 12
or constraints. The decision variables are factored with the angle
illustrates the shapes of the existing CFS that will be compared against
occurrence which is determined by the location and occupancy
the optimised CFS. All existing CFS have the same simulation conditions
schedule.
as the optimised panel as well as the same solar angle range and
• A series of simulation, analysis and automated decisions is carried
schedule.
out through CFStrace and the optimisation engine.
• Several solutions from the Pareto Front and their performances are
3.4. Optimisation parameters
presented for the user to select the desired shape.

3.4.1. Objective function

There are 3 objectives that are addressed in this case study, illumi­
nation uniformity, UDI, and glare. The formulated objective function
aims to minimise the deviation in illuminance (i.e., increasing illumi­
nation uniformity) and maximise the UDI in the space. The glare is set as
a constraint to simplify the multi objective optimisation problem,
therefore becoming a bi-objective optimisation with a third objective as
a constraint. The glare constraint value is set to ε = 0.5 (i.e., the DGP
must not exceed a value of 0.40 for more than 50% of the time) which
allows for some discomfort for some time, as constraining the glare
Fig. 7. Glare penalty graph. further results in very limited valid solutions. The weights for the

7
I.A. Mashaly et al. Building and Environment 197 (2021) 107828

Fig. 8. Detailed optimisation procedure with inputs, processes, and outputs.

Table 2
Summary of the main parameters of the case study.
Parameter Details Value

Location Brisbane, Australia S33◦ , E50◦

Room dimensions Width 3.6 m
Height 2.8 m
Depth (variable) 5, 8, 12 m
Room material properties Ceiling Reflectance 50%
Walls Reflectance 50%
Floor Reflectance 20%
Window Transmission 89%
Window orientation Equatorial Façade N
WWR 55%
Window dimensions Width 2.8 m
Height 2.0 m
Panel dimensions Width 2.8 m
Thickness Variable
Height (variable) 0.6, 1.2 m
Schedule Regular Office Schedule 8:00–18:00 h
Type of simulation Annual simulation 365 days
Occupants’ behaviour Blinds on Clear Glazing Closed

Fig. 9. Room and fenestration system dimensions and sensor locations. lution of the resulting Pareto front and the considerable optimisation
computation time. If a more accurate Pareto font is desired, additional α
samples may be used. Table 4 details the GA parameters that would be
objective functions are set to α ∈ {0, 0.3, 0.5, 0.7, 1} to sample the
used for each α.
Pareto Front and examine the trade-off in the optimal values of the
objectives. Since this case study is only a proof of concept, the selection
of the α samples between 0 and 1 are set to balance between the reso­

8
I.A. Mashaly et al. Building and Environment 197 (2021) 107828

Table 3 4.2. Comparing the optimised design with existing CFS designs
The design variables, boundaries, and steps.
Steps Upper Bound Lower Bound One of the optimised designs on the Pareto Front was chosen to
compare against the existing designs. The chosen optimised design is
Justification 1% 100% 0%
Amplitude 0.1 mm 20.0 mm 0.0 mm α = 0.7, which is near the utopia point.
Fillet 0.1 mm 5.0 mm 0.0 mm The optimised CFS showed better values in UDI and standard devi­
Spacing 0.01 mm 20.00 mm 10.00 mm ation when compared to all existing CFS used for comparison in this case
Panel Size 0.6 m 0.6 m 1.2 m study (see Fig. 15). The highest UDI achieved by an existing CFS was
56% whereas the UDI of the optimised design was 67%. The deviation in
the illuminance of the optimised design was 7.4% with an UDI of 66.9%,
4. Results which performed better than the best existing CFS which was at 8.9%
and the UDI was only 45%. The optimised design is also compliant with
4.1. The pareto outcome the glare constraint added at DGPs above 0.4 for 16% of the time.

This section presents the optimum solutions for the sine-wave shaped
4.3. Comparing different panel heights
panel in terms of UDI and light distribution (standard deviation) in the
space. We assessed the results with different weights in the trade-offs
The following section presents the effect of changing the panel height
between the daylight distribution and overall annual daylight illumi­
on the optimum results. The first observation is that when increasing the
nance through the UDI. In this way, designers can assess their prefer­
panel height, both the Pareto Front UDI and standard deviation
ences with either higher daylight values or better daylight uniformity.
decreased. These reductions were due to the glare constraint which
Designers must consider the trade-off when preferring one feature over
eliminate any designs that caused glare more than 50% of the time.
the other.
Since the panel’s height was increased, the glare probability rose, which
The results for a window height equalling 0.6 m and a room depth
forced the optimum and valid results to a lower uniformity and illumi­
equalling 8 m are presented in the following section. The Pareto Front is
nance (see Fig. 16 and Fig. 17).
formed by drawing a curved line intersecting the most optimum solution
for each weight (α = 0, 0.3, 0.5, 0.7, 1). The α = 0 and α = 1 represent the
two extremes where the optimisation occurs for either UDI or standard 4.4. Comparing different room depths
deviation, respectively. α = 0.3, 0.5, 0.7 represent the balanced objec­
tive functions. The utopia point is close to the α = 0.5 and α = 0.7, which Changing the depth of the room also affected the optimum perfor­
means using those specific weights in those objective functions renders mance and the final CFS design. As shown in Fig. 18, when the depth of
an optimal solution near utopia. the room increased, the optimum UDI and the optimum SD decreased (i.
When looking at the Pareto Front design parameters in Fig. 13 and e., uniformity increased) (see Fig. 19).
Table 5, the justification of panels’ wave peak had values ranging from
73.4% to 86.2%. These values mean it is better to have the wave peak 5. Discussion
oriented upwards to increase uniformity and daylight illuminance at all
angles. The fillet values also indicate that, in order to have more uni­ 5.1. Results discussion
formity, the fillet must be near a value of zero, which also means fewer
glare issues. This is due to the fact that designs with larger fillets would The results show that the optimised CFS performance in a specific
disperse the light into the whole space (upwards and downwards), setting exceeds the daylight performance of other CFS with the same
which would cause glare, but at the same time, the light dispersion in all category and settings. This is partially due to the focus on architectural
directions would increase the UDI on the working plane. daylight performance related metrics when optimising the design. The
Fig. 14 represents the convergence plot for the UDI and Relative SD other reason a better design was found is due to CFStrace’s high number
(uniformity) objectives along the optimisation time for α = 0.3. This is to of simulations which can be run in a shorter time.
show that the solver attempted to reach a near optimal value for both The results also show that the Pareto optimal designs can vary in
objectives. mUDI performance drastically (from 25% to 75% compared to the full
range of mUDI which was from 2% to 75%) while the changes in SD are

Fig. 10. CFS design variables.

9
I.A. Mashaly et al. Building and Environment 197 (2021) 107828

Fig. 11. Room depth and panel-window height as multi-scenarios for other optimisation runs.

Fig. 12. Shapes of existing CFS in literature.

within 6%, which is a low range (Pareto optimal from 5% to 11%

Table 4 compared to the full range from 5% to 45%). These ranges in results
Optimisation settings. illustrate the importance of multi-objective optimisation to determine
an optimal solution.
Parameter Value
Factoring direct glare into the optimisation process as a constraint is
Population size 50 genomes (2 × initial population boost)
crucial as it ensures the occupant’s visual comfort.
Max no. of stagnant populations 50
Mutation 5%
As for the importance of including the architectural context in the
Inbreeding 75% CFS initial design process, the results demonstrate the different opti­
mised solutions for different architectural designs when changing the
room depth and window height.

10
I.A. Mashaly et al. Building and Environment 197 (2021) 107828

Fig. 13. The Pareto Front and valid solutions from the multi-objective optimisation.

5.2. Comparison to other design methodologies

Table 5
The pareto front designs, variables, objectives and results.
The simplicity of CFStrace’s design workflow resulted in a thorough
α α=0 α = 0.3 α = 0.5 α = 0.7 α=1 optimisation process with multi-objectives and multiple constraints
Simulation no. 3752 890 4153 2414 3601

Objectives
UDI 76% 75% 73% 67% 27%
SD 10% 10% 9% 7% 5%
Glare 38% 46% 42% 17% 4%

Variables
Amplitude (mm) 5.1 4.9 4.6 8.1 8.1
Justification (%) 86% 83% 81% 74% 73%
Spacing (mm) 10.8 11 10.9 17.1 17.3
Fillet (mm) 2.8 2.3 2.3 0.1 0

Fig. 15. Comparison of existing CFS with an optimised design.

Fig. 14. Convergence Plot for each objective for the first 50 iterations.

11
I.A. Mashaly et al. Building and Environment 197 (2021) 107828

Fig. 16. Pareto Front results for optimising the UDI and SD for different panel heights (non-optimised valid solutions are hidden from the graph for the clarity of
the comparison).

Fig. 17. Pareto Front CFS shapes and parameters for different panel heights.

other hand, CFStrace does not require any pre-modelling or BSDF gen­
eration as the designs are generated as a part of the optimisation process,
depending on the objective function and how close it is from the opti­
mum solution.
The proposed method can be easily generalised to include multiple
types of complex fenestration systems without the need to generate the
designs beforehand, thus giving flexibility in adjusting the parameters
and decreasing their discretisation. While other methods can provide a
thorough daylight performance analysis to spaces with CFSs, CFStrace is
limited to providing an approximate, yet quick, annual daylight anal­
ysis, making it more suitable for a preliminary design phase.

6. Conclusions
Fig. 18. Pareto Front results for optimising the UDI and SD at different room
depths (non-optimised valid solutions are hidden from the graph for the clarity
of the comparison). In this study, the current state-of-the-art CFS design processes were
assessed and the CFS design simulation processes shortcomings and gaps
were highlighted. The research was conducted with a case study
possible. In the proposed methodology, each simulation took around an
approach featuring the developed design optimisation methodology.
average of 2–5 s with the Grasshopper plugin, (simulations were per­
The aim of the newly developed method was to optimise CFS designs
formed on a desktop computer with an 8-core Intel® Core™ i7-6700
while considering both the architectural and environmental context.
processor clocked at a maximum of 3.4 GHz, 16 GB RAM, 1 TB hard
This method, using the CFStrace tool, includes the new daylight metrics
drive, an Intel® HD Graphics 530 Card and a Windows© 7 Enterprise 64-
developed specifically for assessing CFS performance in a working or
bit operating system). In comparison, for one of the state-of-the-art
living space.
workflows using Radiance’s five phase method [36], the average
A virtual case study was used, where a light redirecting device would
computing time of each simulation within the optimisation process was
be designed as part of the CFS to enhance the daylight performance of a
about 3:45 min [36]. The researchers had to generate 1728 BSDF design
simple room. A genetic algorithm method was used to optimise a multi-
files to run an optimisation process on selecting the optimum solution
objective function with several constraints. A trade-off between the
from within those 1728 design solutions. However, it is undocumented
objective functions was demonstrated through weighing the objective
how much time was taken to create the 1728 BSDF files for the opti­
functions and plotting the Pareto solutions. The trade-off demonstrates
misation. This can take a significant amount of time compared to the
the importance of optimising CFS designs and of the user’s input and
main simulation, depending on the complexity of the geometry. On the

12
I.A. Mashaly et al. Building and Environment 197 (2021) 107828

Fig. 19. Pareto Front CFS shapes and parameters for different room depths.

preferences for the design solution. The case study also demonstrated systems, Journal of Building Performance Simulation 6 (1) (2012) 24–37, https://
doi.org/10.1080/19401493.2012.671852.
that changing the architectural environment would result in different
[12] D. Uribe, S. Veraand, W. Bustamante, Optimization of Complex Fenestration
optimum CFS designs. Systems Using an Artificial Neural Network. Back To the Future: the Next 50 Years,
The metrics developed and their usage in the proposed methodology (51st International Conference Of the Architectural Science Association (ANZAScA),
and tool open the door to new possibilities of CFS performance analysis. 2017, pp. 177–185. http://anzasca.net/wp-content/uploads/2017/11/ASA_2017
_Uribe_Veraand_Bustamante.pdf.
The tool is most likely to be used in the early phases of design as well [13] G. Isoardi, The Design and Testing of a Daylighting Device: Optimizing the Energy
as retrofitting phases for post occupancy. For further analysis, more and Optical Performance of Australian Commercial Buildings, Queensland
comprehensive tools such as Radiance and genBSDF can be used to University of Technology, 2009.
[14] A. Kostro, M. Geiger, J.-L. Scartezzini, A. Schüler, CFSpro: ray tracing for design
analyse the optimised design. and optimization of complex fenestration systems using mixed dimensionality
approach, Appl. Opt. 55 (19) (2016) 5127–5134, https://doi.org/10.1364/
AO.55.005127.
Declaration of competing interest [15] I.A. Mashaly, K. Nassar, S. El-Haggar, Mathematical model for designing a light
redirecting prismatic panel, Sol. Energy 159 (2018) 638–649, https://doi.org/
The authors declare that they have no known competing financial 10.1016/j.solener.2017.11.014.
[16] A. Tsangrassoulis, V. Machairas, C. Axarli, Simplified design of a specular slat
interests or personal relationships that could have appeared to influence profile curve using 2 ray tracing and genetic algorithms, Proceedings of Building
the work reported in this paper. Simulation BS2013 (2013) 3667–3672.
[17] I.A. Mashaly, V. Garcia-Hansen, G. Isoardi, M. Chamorro-Koc, A design method for
complex fenestration systems using forward raytracing, in: V. Corrado, E. Fabrizio,
Acknowledgements A. Gasparella, F. Patuzzi (Eds.), Proceedings of Building Simulation 2019: 16th
Conference of IBPSA, vol. 16, International Building Performance Association
This research is a part of a PhD research supported by QUT with a (IBPSA), 2020, pp. 1060–1067, https://doi.org/10.26868/
25222708.2019.210599.
QUTPRA scholarship. [18] I.A. Mashaly, V. Garcia-Hansen, G. Isoardi, M.E. Cholette, CFStrace: An evaluation
method to include complex fenestration systems in the façade design process, Solar
References Energy 217 (2021) 253–262, https://doi.org/10.1016/j.solener.2021.01.073. In
press, https://www.sciencedirect.com/science/article/abs/pii/S003809
2X21000992?via%3Dihub.
[1] P. Boyce, C. Hunter, O. Howlett, The benefits of daylight through windows, Light.
[19] S.I. El-Henawy, M.W.N. Mohamed, I.A. Mashaly, O.N. Mohamed, O. Galal, I. Taha,
Res. Cent (2003) 9–68. http://thedaylightsite.com/wp-content/uploads/papers/D
K. Nassar, A.M.E. Safwat, Illumination of dense urban areas by light redirecting
aylightBenefits.pdf.
panels, Opt Express 22 (S3) (2014) A895, https://doi.org/10.1364/OE.22.00A895.
[2] M.J. Farley, J.A. Veitch, A Room with a View: A Review of the Effects of Windows
[20] S. Klammt, A. Neyer, H.F.O. Müller, Microoptics for efficient redirection of
on Work and Well-Being, 2001.
sunlight, Appl. Opt. 51 (12) (2012) 2051–2056, https://doi.org/10.1364/
[3] K. Pham, A. Wagdy, G. Isoardi, A. Allan, V. Garcia-Hansen, A methodology to
AO.51.002051.
simulate annual blind use in large open plan offices, Proceedings of Building
[21] M. Fontoynont, Perceived performance of daylighting systems: lighting efficacy
Simulation 2019: 16th Conference of IBPSA 16 (2020) 1091–1097, https://doi.
and agreeableness, Sol. Energy 73 (2) (2002) 83–94, https://doi.org/10.1016/
org/10.26868/25222708.2019.210742.
S0038-092X(02)00035-X.
[4] W. O’Brien, K. Kapsis, A.K. Athienitis, Manually-operated window shade patterns
[22] P.D. Robinson, M.G. Hutchins, Advanced glazing technology for low energy
in office buildings: a critical review, in: Build. Environ., 60, 2013, pp. 319–338,
buildings in the UK, Renew. Energy 5 (1–4) (1994) 298–309, https://doi.org/
https://doi.org/10.1016/j.buildenv.2012.10.003. Pergamon.
10.1016/0960-1481(94)90387-5.
[5] N. Jakica, State-of-the-art review of solar design tools and methods for assessing
[23] A. Eltaweel, S. Yuehong, Using integrated parametric control to achieve better
daylighting and solar potential for building integrated photovoltaics, Renew.
daylighting uniformity in an office room: a multi-Step comparison study, Energy
Sustain. Energy Rev. 81 (1) (2018) 1296–1328, https://doi.org/10.1016/j.
Build. 152 (2017) 137–148, https://doi.org/10.1016/j.enbuild.2017.07.033.
rser.2017.05.080.
[24] V. Garcia-hansen, G. Isoardi, M. Hirning, J. Bell, An assessment tool for selection of
[6] 3 M, 3MTM Daylight Redirecting Film Product Bulletin, 2018. www.3mgraphics.com.
appropriate daylighting solutions for buildings in tropical and subtropical Regions :
[7] SerraLux, SerraLux Project, 2020. https://serraluxinc.com/projects/.
validation using radiance simulation, Wref 1–8 (2012).
[8] G. Molina, Integrated thermal and lighting analysis of spaces with controled
[25] A. Nabil, J. Mardaljevic, Useful daylight illuminances: a replacement for daylight
complex fenestration systems and artificial lighting during the design stage
factors, Energy Build. 38 (7) (2006) 905–913, https://doi.org/10.1016/j.
[Pontificia Universidad Católica de Chile], 2014. http://repositorio.uc.cl/h
enbuild.2006.03.013.
andle/11534/3275.
[26] J. Mardaljevic, Daylight daylight/daylighting , indoor illumination indoor
[9] G.C. Henriques, J.P. Duarte, V. Leal, Strategies to control daylight in a responsive
illumination , and human behavior, in: Encyclopedia of Sustainability Science and
skylight system, Autom. ConStruct. 28 (2012) 91–105, https://doi.org/10.1016/j.
Technology, Springer, New York, 2012, pp. 2804–2846, https://doi.org/10.1007/
autcon.2012.06.002.
978-1-4419-0851-3_456.
[10] A. Wagdy, F. Fathy, A Parametric Approach for Achieving Optimum Daylighting
[27] R.G. Hopkinson, P. Petherbridge, J. Longmore, Daylighting, Heinemann, 1966.
Performance through Solar Screens in Desert Climates, 2015, https://doi.org/
https://qut.primo.exlibrisgroup.com/discovery/fulldisplay?docid=alma
10.1016/j.jobe.2015.07.007.
991000321829704001&context=L&vid=61QUT_INST:61QUT&lang=en&search
[11] A. McNeil, E.S. Lee, A validation of the Radiance three-phase simulation method
for modelling annual daylight performance of optically complex fenestration

13
I.A. Mashaly et al. Building and Environment 197 (2021) 107828

_scope=MyInst_and_CI&adaptor=Local Search Engine&tab=Everyth [33] B. Bueno, J. Wienold, A. Katsifaraki, T.E. Kuhn, Fener: a Radiance-based modelling
ing&query=any,contains,R.G. Hopkinson Daylighting, He. approach to assess the thermal and daylighting performance of complex
[28] J. Wienold, J. Christoffersen, Evaluation methods and development of a new glare fenestration systems in office spaces, Energy Build. 94 (2015) 10–20, https://doi.
prediction model for daylight environments with the use of CCD cameras, Energy org/10.1016/j.enbuild.2015.02.038.
Build. 38 (7) (2006) 743–757, https://doi.org/10.1016/j.enbuild.2006.03.017. [34] J.A. Jakubiec, Validation of Simplified Visual Discomfort Calculations, 2018.
[29] J. Wienold, Dynamic daylight glare evaluation, IBPSA 2009 - International https://www.researchgate.net/publication/327823262.
Building Performance Simulation Association 2009 (2009) 944–951. [35] C.F. Reinhart, J.A. Jakubiec, D. Ibarra, Definition of A reference office for
[30] N.L. Jones, Fast climate-based glare analysis and spatial mapping, in: Proceedings standardized evaluations of dynamic façade and lighting technologies, in: 13th
of Building Simulation 2019: 16th Conference of IBPSA, September, 2019. Conference of International Building Performance Simulation Association, 2013,
[31] B.Y. Qu, P.N. Suganthan, Constrained multi-objective optimization algorithm with pp. 3645–3652.
an ensemble of constraint handling methods, Eng. Optim. 43 (4) (2011) 403–416, [36] S. Vera, D. Uribe, W. Bustamante, Optimization of a fixed exterior complex
https://doi.org/10.1080/0305215X.2010.493937. fenestration system considering visual comfort and energy performance criteria,
[32] V.D. Noghin, Linear scalarization in multi-criterion optimization, Sci. Tech. Inf. Build. Environ. 113 (2017) 163–174, https://doi.org/10.1016/j.
Process. 42 (6) (2015) 463–469, https://doi.org/10.3103/S014768821506009X. egypro.2017.09.676.

14