University of Southern California Dissertations and Theses

Using bio-signals with smart windows

(USC Thesis Other)

Using bio-signals with smart windows

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content
Using Bio-signals with Smart Windows
Occupants’ Bio-signals as Potential Control Mechanism

by
Zihan Wang

A thesis presented to the
THE FACULTY OF THE USC SCHOOL OF ARCHITECTURE
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
MASTER OF BUILDING SCIENCE

AUGUST 2022

Copyright 2022 Zihan Wang
ii
Acknowledgements

I would like to thank all the participants to share their valuable information and their data with me
in the research. What’s more, I would like to thank my committee chair Professor Joon-Ho Choi,
committee members Professor Jose-Luis Ambite, Professor Douglas Noble, and Doctor Dong
Yoon Park for their instructions and comments on my thesis. And finally, I would thank all my
family members and friends for their assistance and support throughout my thesis.

iii
Table of Contents

Acknowledgements ....................................................................................................................... ii
List of Tables ................................................................................................................................. vi
List of Figures ............................................................................................................................. vii
Abstract ...................................................................................................................................... viii
Chapter 1. Introduction ................................................................................................................ 1
1.1 Introduction to Electrochromic Glass ..................................................................................................... 1
1.1.1 Basic Information of Electrochromic Glass ................................................................... 1
1.1.2 Advantages of EC Glass ................................................................................................. 4
1.1.3 Disadvantages of EC Glass ............................................................................................ 5
1.2 Smart Window and Control ....................................................................................................................... 6
1.2.1 Technical Features of Electrochromic Glass .................................................................. 6
1.2.2 Electrochromic Window Controls .................................................................................. 8
1.3 Thermal Performance Analysis .............................................................................................................. 10
1.3.1 Thermal Comfort .......................................................................................................... 10
1.3.2 Thermal Comfort Analysis ........................................................................................... 11
1.3.3 Visual Comfort ............................................................................................................. 12
1.3.4 Visual Comfort Analysis .............................................................................................. 13
1.4 Summary ....................................................................................................................................................... 13
Chapter 2. Literature Review ..................................................................................................... 15
2.1 Electrochromic Windows Control Technologies .............................................................................. 15
2.1.1 Different Control Strategies ......................................................................................... 15
2.1.2 Comparison of Different Control Strategies in Energy Consumption ......................... 16
2.1.3 Limitations of Current Control Technologies .............................................................. 17
2.2 Passive Environmental Control .............................................................................................................. 18
2.2.1 Different Heating and Cooling Control Strategies ....................................................... 18
2.2.2 Different Lighting Control Strategies ........................................................................... 19
2.3 Impacts of Electrochromic Windows ................................................................................................... 20
2.3.1 Impacts on Indoor Environmental Quality ................................................................... 20
2.3.2 Impacts on Occupants’ Thermal and Visual Comfort .................................................. 20
2.4 Thermal Comfort Prediction Models .................................................................................................... 21
2.4.1 Predicted Mean Vote and Predicted Percentage of Dissatisfied ................................... 22
2.4.2 Dynamic Mathematical Model ..................................................................................... 22
2.4.3 Other Thermal Comfort Models ................................................................................... 22
2.5 Summary ....................................................................................................................................................... 24
Chapter 3. Methodologies ........................................................................................................... 25
3.1 Methodologies Overview ......................................................................................................................... 25
3.1.1 Experiment Preparation ................................................................................................ 25
3.1.2 Chamber Preparation .................................................................................................... 26
iv
3.1.2.1 EC Glass Preparation ............................................................................................. 28
3.1.2.2 Other Devices Preparation ..................................................................................... 32
3.1.3 Sensors and Software ................................................................................................... 34
3.2 Experiment Process .................................................................................................................................... 38
3.2.1 Introduction of Experiment .......................................................................................... 38
3.2.2 General Process of the Experiment .............................................................................. 39
3.3 Data Collection ............................................................................................................................................ 40
3.3.1 Background Information of Subjects ............................................................................ 40
3.3.2 Subjects’ Bio-signals .................................................................................................... 41
3.3.3 Indoor Environmental Quality Data ............................................................................. 41
3.3.4 Participants’ Feedbacks ................................................................................................ 42
3.4 Data Analysis ............................................................................................................................................... 42
3.4.1 Data Cleaning and Processing ...................................................................................... 43
3.4.2 Data Analysis Process .................................................................................................. 44
3.4.3 Output of the Data Analysis ......................................................................................... 45
3.4.4 Result Verification ........................................................................................................ 45
3.5 Summary ....................................................................................................................................................... 46
Chapter 4. Data Analysis and Results ....................................................................................... 47
4.1 Data Processing ........................................................................................................................................... 47
4.1.1 Data Preprocess ............................................................................................................ 47
4.1.2 Database Information ................................................................................................... 48
4.2 Individual Data Analysis .......................................................................................................................... 49
4.2.1 ANOVA Analysis Between Window States and Human Bio-signals .......................... 49
4.2.2 ANOVA Analysis Between Thermal Sensation and Human Bio-signals .................... 50
4.2.3 ANOVA Tests Between Visual Sensation and Human Bio-signals ............................ 51
4.2.4 Correlation Between Thermal Sensations, Visual Sensations and Human Bio-signals
............................................................................................................................................... 52
4.3 Data Analysis Differences by Gender .................................................................................................. 53
4.3.1 Anova Analysis Result Differences by Gender ............................................................ 53
4.3.2 Correlation Analysis Differences by Gender ............................................................... 55
4.3.2.1 2-Sample T Test Between Visual Sensation and Heart Rate ................................ 55
4.3.2.2 2-Sample T Test Between Visual Sensation and Stress Level .............................. 56
4.3.2.3 2-Sample T Test Between Visual Sensation and EDA ......................................... 56
4.3.2.4 2-Sample T Test Between Visual Sensation and Skin Temperature ..................... 57
4.3.2.5 2-Sample T Test Between Thermal Sensation and Heart Rate ............................. 57
4.3.2.6 2-Sample T Test Between Thermal Sensation and Stress Level ........................... 58
4.3.2.7 2-Sample T Test Between Thermal Sensation and EDA ...................................... 58
4.3.2.8 2-Sample T Test Between Thermal Sensation and Skin Temperature .................. 59
4.4 Summary ....................................................................................................................................................... 59
Chapter 5. Development of Prediction Model .......................................................................... 61
5.1 Thermal Sensation Prediction Model ................................................................................................... 61
5.2 Visual Sensation Prediction Model ....................................................................................................... 65
5.3 Thermal Sensation Prediction Model by Gender .............................................................................. 69
5.4 Visual Sensation Prediction Model by Gender ................................................................................. 75
v
5.5 Summary ....................................................................................................................................................... 80
Chapter 6. Conclusions and Future Work ................................................................................ 82
6.1 Conclusions .................................................................................................................................................. 82
6.2 Limitations .................................................................................................................................................... 83
6.2.1 Short-term Problems ..................................................................................................... 84
6.2.2 Long-term Problems ..................................................................................................... 84
6.3 Future Work ................................................................................................................................................. 85
References .................................................................................................................................... 86

vi
List of Tables

Table 1. Background Information ................................................................................................. 40
Table 2. Summary of Window State Individual Anova Analysis Result ...................................... 50
Table 3. Thermal Sensation Scale ................................................................................................. 51
Table 4. Summary of Thermal Sensation Individual Anova Analysis Result ............................... 51
Table 5. Visual Sensation Scale .................................................................................................... 52
Table 6. Summary of Visual Sensation Individual Anova Analysis Result .................................. 52
Table 7. Correlation Analysis Results Summary .......................................................................... 53
Table 8. Anova Analysis Results Summary by Gender ................................................................ 54
Table 9. Descriptive Statistics Between Visual Sensation and Heart Rate ................................... 56
Table 10. P-value Between Visual Sensation and Heart Rate ....................................................... 56
Table 11. Descriptive Statistics Between Visual Sensation and Stress Level ............................... 56
Table 12. P-value Between Visual Sensation and Stress Level .................................................... 56
Table 13. Descriptive Statistics Between Visual Sensation and EDA Data .................................. 57
Table 14. P-value Between Visual Sensation and EDA Data ....................................................... 57
Table 15. Descriptive Statistics Between Visual Sensation and Skin Temperature ...................... 57
Table 16. P-value Between Visual Sensation and Skin Temperature ........................................... 57
Table 17. Descriptive Statistics Between Thermal Sensation and Heart Rate .............................. 57
Table 18. P-value Between Thermal Sensation and Heart Rate .................................................... 58
Table 19. Descriptive Statistics Between Thermal Sensation and Stress Level ............................ 58
Table 20. P-value Between Thermal Sensation and Stress Level ................................................. 58
Table 21. Descriptive Statistics Between Thermal Sensation and EDA Data ............................... 58
Table 22. P-value Between Thermal Sensation and EDA Data .................................................... 58
Table 23. Descriptive Statistics Between Thermal Sensation and Skin Temperature ................... 59
Table 24. P-value Between Thermal Sensation and Skin Temperature ........................................ 59
Table 25. Convert 7-Point Scale to 3-Point Scale ......................................................................... 62
Table 26. Result of 3-Point Thermal Sensation Model ................................................................. 65
Table 27. Result of 7-Point Thermal Sensation Model ................................................................. 65
Table 28. Result of 3-Point Visual Sensation Model .................................................................... 69
Table 29. Result of 7-Point Visual Sensation Model .................................................................... 69
Table 30. Result of 3-Point Thermal Sensation Model (Female and Male) .................................. 75
Table 31. Result of 7-Point Thermal Sensation Model (Female and Male) .................................. 75
Table 32. Result of 3-Point Visual Sensation Model (Female and Male) ..................................... 80
Table 33. Result of 7-Point Visual Sensation Model (Female and Male) ..................................... 80
Table 34. Summary Result of Each Prediction Model .................................................................. 80

vii
List of Figures

Figure 1. Electrochromic Glass [3] ................................................................................................. 2
Figure 2. Schematic Operation of EC Glass [4] .............................................................................. 3
Figure 3. Glare Control Example [12]. ............................................................................................ 9
Figure 4. Heating and cooling energy consumption for winter and summer [26] ......................... 17
Figure 5. Energy consumption of all control technologies [26] .................................................... 17
Figure 6. Overall Workflow .......................................................................................................... 25
Figure 7. Floor Plan of Chamber ................................................................................................... 27
Figure 8. Chamber Preparation ...................................................................................................... 28
Figure 9. EC Glass Clear State ...................................................................................................... 29
Figure 10. EC Glass Light Tint ..................................................................................................... 29
Figure 11. EC Glass Medium Tint ................................................................................................ 30
Figure 12. EC Glass Fully Tinted State ......................................................................................... 30
Figure 13. EC Glass Controller ..................................................................................................... 31
Figure 14. Installation Guide [42] ................................................................................................. 32
Figure 15. Features of Heat Lamp [43] ......................................................................................... 33
Figure 16. Infrared Heat Lamp ...................................................................................................... 33
Figure 17. Bulb Guard [44] ........................................................................................................... 34
Figure 18. Hanger Rack [45] ......................................................................................................... 34
Figure 19. Experimental and Wearable Sensors ............................................................................ 36
Figure 20. Sunlight Meter .............................................................................................................. 37
Figure 21. Infrared Camera [46] .................................................................................................... 38
Figure 22. Experiment Process ...................................................................................................... 39
Figure 23. Garmin Data Collection ............................................................................................... 41
Figure 24. Indoor Environmental Quality Data ............................................................................. 42
Figure 25. Sample Aggregated Data in Dataset ............................................................................ 48
Figure 26. Gender Rate of significant Result ................................................................................ 55
Figure 27. 3-Point Thermal Sensation Decision Tree Visualization ............................................. 63
Figure 28. 7-Point Thermal Sensation Decision Tree Visualization ............................................. 64
Figure 29. 3-Point Visual Sensation Decision Tree Visualization ................................................ 67
Figure 30. 7-Point Visual Sensation Decision Tree Visualization ................................................ 68
Figure 31. 3-Point Thermal Sensation Decision Tree Visualization (Female) .............................. 71
Figure 32. 7-Point Thermal Sensation Decision Tree Visualization (Female) .............................. 72
Figure 33. 3-Point Thermal Sensation Decision Tree Visualization (Male) ................................. 73
Figure 34. 7-Point Thermal Sensation Decision Tree Visualization (Male) ................................. 74
Figure 35. 3-Point Visual Sensation Decision Tree Visualization (Female) ................................. 76
Figure 36. 7-Point Visual Sensation Decision Tree Visualization (Female) ................................. 77
Figure 37. 3-Point Visual Sensation Decision Tree Visualization (Male) .................................... 78
Figure 38. 7-Point Visual Sensation Decision Tree Visualization (Male) .................................... 79

viii

Abstract

Nowadays, smart windows are more and more popular because of their high performance in
improving indoor environment and decreasing heating and cooling loads. Among different types
of smart windows, electrochromic windows are most common in office buildings because they are
easily installed and controlled. However, the control of electrochromic windows might cost more
energy than other smart windows like thermochromic windows and cause less thermal and visual
comfort; therefore, if it is possible to use bio-signals to control different states of the windows, this
situation will be benefitted a lot in the future.

The project aims to investigate the impacts of electrochromic windows on the users’ bio-signals,
to explore the potential of using them as a control strategy of the windows and decreasing energy
consumption and improving thermal comfort could be accomplished. This project collected indoor
environmental quality data and humans’ bio-signal data using different sensors. After collecting
data, the data was analyzed using machine learning and used to develop a control model of the
smart windows.

As a result, by using data analysis skills, human’s bio-signals were affected by the utilizing the
smart windows, which was shown when the EC windows are in the different states, occupants
would have different levels in heart rate and skin temperature. In the future, the research on the
utilization of bio-sensing control of electrochromic windows will be conducted further.

Keywords: Human Bio-signals, Electrochromic Windows, Thermal and Visual Sensations,
Machine Learning.
1
Chapter 1. Introduction

1.1 Introduction to Electrochromic Glass
Modern day high-rise building envelopes are often constituted with a high percentage of glazing.
Glass facades provide increased outdoor views, natural light, and solar heat into the interior space
of the building [1]. However, glazed facades are also responsible for about 50% of energy
consumption in a whole building through the heat gain or loss, and they can have a poorer thermal
performance compared to other building components. Therefore, improving the thermal
performance of the glazing area of a building is relatively important in reducing building energy
consumption [1]. In addition, occupants’ thermal comfort is also important, which could have a
significant effect on productivity and health. According to previous research by others, employees
who are more satisfied with thermal conditions like indoor temperature and relative humidity are
more productive. Thermal comfort can affect humans’ physiological signals, such as heart rate,
EDA, and skin temperature, these signals have been widely implemented to analyze thermal
comfort of humans, and even control the indoor environment in some studies.

Electrochromic glazing can be one façade strategy to address thermal performance and human
comfort. The definition and basic features of electrochromic (EC) glass will be introduced in this
section to provide an overview of the properties.

1.1.1 Basic Information of Electrochromic Glass
Electrochromic glass (EC) is also known as smart glass. EC is capable of changing tint by
administering an electrical voltage, allowing occupants to control the light transmittance just by
2
adjusting a controller. When using EC glass, there is no need to use blinds or curtains to block the
light and heat, because EC could be fully transparent or tinted [2].

Figure 1. Electrochromic Glass [3]

3
While EC glass has a different feature compared to ordinary glass, the primary difference between
them is the metal-oxide coatings which could accept and respond to electrons from the small
voltage applied by the controller. As Figure 1 shows, the EC glass panel is comprised of several
different layers. Above and beneath the two glass layers are the transparent conductive layers,
which were made from Indium Tin Oxide (ITO), which function as a battery to comprise the
electrodes. There is an electrochromic layer in the center of the glass panel structure, which is what
is capable of changing the tint and therefore also changing the light transmittance. When a small
voltage is applied to the glass, the charged particles (ions) then move to the electrolyte to the
electrochromic layer. Then the electrochromic layer is able to reflect more light, which makes the
glass become in dark color. After the voltage is removed or reversed, the transmittance of
electrochromic layer would increase, and the glass becomes clearer [3].
Figure 2. Schematic Operation of EC Glass [4]
4

The EC glass used in this study can be adjusted from fully clear to fully tinted and is able to
maintain some states in between them as well. The glass would absorb more solar radiation and
reflect more sunlight when is become more tinted. As the solar absorption increases, the glass will
become hotter, and thus a low-E coating could be useful for the glass panel such as shown in Figure
2. Although the glass could be adjusted into a fully tinted state, the light transmittance can never
fully go down to zero, but it can only go down as low as 1% [4].

1.1.2 Advantages of EC Glass
First, EC glass has a huge advantage for solar control and human comfort. Since it can reflect more
than 90% of sunlight and solar radiation when it is tinted, the need of air conditioning system could
be decreased, which could reduce the peak cooling and lighting load in buildings. In addition, it
could reduce the annual building energy consumption, because it absorbs excessive solar heat gain
in summer and provides passive heating and enough daylight in for the building occupants. What’s
more, using EC glass can reduce the need of curtains and blinds, which could improve occupants’
privacy, protect occupants’ upholstery and pictures from being burned by sunlight [2].
Furthermore, EC windows could also improve the security of indoor occupants because it can
make it difficult to see inside from outside, and EC glazing is typically harder to break than
standard glazing. Finally, using EC glass in the office building and schools could also improve
productivity and reduce eye fatigue. While using EC windows, enough daylighting could be
provided, and the solar heat gain and glare could be reduced [5].

5
Because of the advantages of EC for the indoor environment and energy consumption in buildings,
using EC glass can also help a project to get green building certifications. According to EC
manufacturer Sage Glass [6], because of benefits of EC glass, including reducing energy
consumption, improving the thermal and visual comfort of occupants, many credits could be
accessed for green building certifications like LEED.

1.1.3 Disadvantages of EC Glass
Although there are several benefits of using EC glass in buildings, there are still some drawbacks.
Understanding both the benefits and drawbacks could help designers to decide if it is appropriate
to use EC glass.

One important disadvantage is that the materials and installation costs of EC glass could be several
times more expensive than ordinary glass since EC glass uses electrodes and advanced metal
coatings. For example, a piece of EC glass is currently priced at about $500 to $1000 per square
foot, while even a piece of double pane low-E glass costs only from $40 to $55 per square foot.
Furthermore, the current EC glass can degrade in 10-20 years, which is significantly shorter than
the duration time of the ordinary glass. In addition, the time of transitioning the EC glass from one
tinted state to another would take several minutes, therefore, during the transferring time, the
occupants’ thermal and visual comfort might be worse than using the blinds or curtains [2].

Although using EC glass could improve the visual comfort of occupants, there can still be glare
issues being caused by EC glass even if EC glass is in the fully tinted state, and so the shading
device might also be needed. What’s more, if there is too much sunlight, and the EC glass is in the
6
fully tinted state, for occupants can work in enough light, the artificial lights indoor then will be
used. Additionally, electricity is essential while using control the EC glass. Therefore, the lighting
energy and electricity consumption of the building will be increased [7].

1.2 Smart Window and Control
To improve the thermal performance of the facade glazing area, some buildings use shading
devices to reduce the solar heat gain or loss to decrease cooling energy in summer. However, this
method is useless in winter, because it could increase heating energy due to the lack of solar heat
gain [8]. Therefore, integration with smart windows in buildings is a popular sustainable method
to save more energy and improve occupants thermal and visual comfort. A smart window is a kind
of glazing, which could alter the lighting transmission properties under the application of light,
voltage, or temperature. This kind of glass could change from transparent to translucent and vice
versa in specific conditions. There are several types of smart window technologies, including
electrochromic, photochromic, thermochromic, suspended-particle, micro-blind, polymer-
dispersed liquid-crystal devices [9]. Although they could each improve thermal performance in
some aspect, some of them such as electrochromic technologies are considered as active
technologies since they need to be controlled by occupants to accommodate for their demands [10].

1.2.1 Technical Features of Electrochromic Glass
Although there are several types of smart window technologies, electrochromic windows will be
used for the analysis of thermal and visual comfort in this research. Electrochromic glass is a kind
of glass used for windows, facades, and curtain walls, which could be changed to other colors
while being controlled by occupants. It could increase occupants thermal and visual comfort, and
7
saving cooling and heating energy, therefore, it is a popular smart window technology nowadays
[11]. Electrochromic glazing can be made of either glass or plastic with several thin layers coated
on the pane using a manufacturing method known as sputtering [2]. The inside surface of the
window includes a double-sandwich of five ultra-thin layers: the middle is a separator, on either
side the separator are two electrodes, and on either side of the electrodes are two transparent
electrical contact layers. An electrochromic window only needs power when transitioning from
one translucency state to another, and it does not need any power when it is not transitioning to a
different translucency state.

There are many benefits when using electrochromic windows as a smart solution for buildings
where the solar control is an important issue, such as high-rise offices, classrooms, healthcare
facilities, etc. In the dark state of electrochromic windows, they could reflect nearly all the light
falling on them, so the need of air-conditioning is reduced, and the cooling and heating energy
could be decreased. Electrochromic glazing can be easily controlled by a smart home system, a
sunlight sensor, or a temperature sensor, because it is electrically operated [2]. By using EC glass,
occupants could be protected from heat and glare, while provided with enough outdoor views and
thermal comfort; therefore, students and officers could be benefitted by increasing the productivity
and concentration on work, patients could get recovered faster and better emotional wellness, and
companies could get the less employee absenteeism [11]. However, there are also some drawbacks
of using electrochromic windows, including the higher cost of buying and installing them, possible
lower durability, and some the slow transition times from clear to dark and back [2].

8
1.2.2 Electrochromic Window Controls
The most common control for Electrochromic windows is by using an electronic controller. When
the glass receives a signal from the controller, then it could transition from one state to another
state. Currently, as the technology is still fairly new and developing, there are also many other
control strategies for the EC windows. For instance, one of the biggest companies of Smart
Window, SageGlass, has developed several control systems either by using sensors, or being a part
of building management system (BMS) [12]. In general, there are several control systems of
SageGlass, including Daylight Control, Schedule Control, and Glare Control.

For daylight control, a daylight sensor can used in the control system, which could detect the sky
conditions like sunny and cloudy, sun positions, and glass, then to trigger the EC glass to react
with the signals. Because daylight control could change the tints of EC glass depending on the
daylighting, which could let the maximum daylight in with overcast skies, and let the minimum
daylight in with clear skies, this control system is for the maximum natural light and optimum
energy management [12].

Schedule control strategy is a strategy that causes the EC glass to change tint states based on the
time of year and time of day and thus control the amount of the solar energy can pass into the space
based on solar positioning and expected outdoor thermal conditions. This control strategy could
reduce cooling energy in summer by blocking excessive solar heat gain, while also reduce heating
energy in winter by allowing solar heat gain into the space. Therefore, by using this control strategy,
combining the daylight control, improved energy performance of the building could be achieved
[12].
9

Since glare could cause visual discomfort and reduce occupants’ focus time at the workplace and
on the computer screen due to direct solar irradiation, EC glass could also be controlled by glare.
Like shown in Figure 3, an EC glass facade could reduce glare by tinting several panes of the glass,
while also allowing enough daylight into the space by maintaining the other panes of the glass
clear, which could reduce the artificial lighting energy in the daytime and heating energy in winter
[12].

Figure 3. Glare Control Example [12].

Besides the current control technologies, there are also some advanced researchers conducting
research on controlling EC glass by human physiological signals, based on the physiological
reactions while using the EC glass. For example, in the previous research of Professor Joon-Ho
10
Choi’s research team, they identified the potential of using physiological data such as heart rate,
skin temperature, brain signal etc., to predict the thermal and visual sensations and comfort of
occupants in real time by practicing the advanced sensing technologies including wearable devices
and remote sensors [13,14]. Using as estimation model of bio signals and occupants’ sensations
toward indoor environment, their team created a building environmental control system. Using
bio-sensing smart window controls could save more energy compared to using ordinary controls
and improve occupants thermal and visual comfort.
1.3 Thermal Performance Analysis
To evaluate the impact of electrochromic windows on humans’ bio-signals, the change of the
humans’ bio-signals before and after the application of electrochromic windows should be
analyzed.

1.3.1 Thermal Comfort
Thermal comfort is an important factor to evaluate the indoor environment, which is one of the
most important fields to represent indoor environmental quality. Thermal conditions are
potentially life-threatening for humans because at the extreme ranges it can cause physical
problems with higher temperature or lower temperature like hyperthermia and hypothermia.
Hyperthermia can happen when the core body temperature is above 37.5
o
C – 38.3
o
C, and the
hypothermia can happen when the temperature is below 35
o
C [15]. In building science studies,
thermal comfort is relatively related to the productivity and health of occupants. For example,
people feel more comfortable with the thermal environment and thus might have better
productivity [16].

11
There are six primary factors which could affect the thermal comfort directly, including metabolic
rate, clothing level, air temperature, mean radiant temperature, air speed and humidity. The first
two factors are person factors, and the others are environmental factors. Metabolic rate is a factor
related to human activity according to ASHRAE 55-2010 Standard, and it is expressed in met unit
[17]. Clothing level is the thermal insulation worn by a person, it is expressed in clo unit, where 1
clo is equal to 0.155m
2
K/W. The environmental factors could be measured with a thermal meter.

1.3.2 Thermal Comfort Analysis
In order to analyze indoor thermal comfort, several models have been designed to measure it. For
example, Fanger [18] has proposed the predicted mean vote (PMV) and predicted percentage
dissatisfied (PPD) models, and they have become the basis of many standards including ISO 7730
and ASHRAE 55. PMV estimates the thermal comfort of occupants on a seven-point thermal
sensation scale, while PPD estimates the potential of dissatisfaction of indoor thermal comfort.
PMV model is based on the six parameters as described in last paragraph, while PPD measures the
percentage of people who feel cold or hot. However, because the PMV model is based on both
human and environmental factors, it is complex to build the model. Therefore, the PMV model is
valuable for analyzing thermal comfort of occupants because the preciseness of the prediction.
Additionally, several studies have addressed the prediction of PMV value by using machine
learning algorithms like classification trees, logistic regression, etc. [19,20].

In order to analyze the thermal comfort of occupants using PMV model, data of occupants’ thermal
sensation should be collected. And smart wearable sensors like smart watches will be used to
collect the heart rate, skin temperature, eda data and stress level, remote sensors will be used to
12
collect indoor environmental factors like indoor temperature and humidity, both factors are needed
to measure thermal comfort of occupants. In this research, heart rate is represented by the number
of heart beats per minute (bpm). Skin temperature is the temperature of human skin, which is
measured in Celsius Degree. Eda data is electrodermal activity data, which represents the skin
conductance, telling researchers the state of body and mind when skin conducts electricity through
the sweat glands [21]. Stress level is the level of stress of participants based on their heart rate
variability, the smart watch could calculate the interval between each heartbeat to determine it.

1.3.3 Visual Comfort
According to several previous researchers, 65% of building occupants are exposed to inappropriate
lighting conditions, which could cause glare problems and visual stress in a workplace
environment [22]. This kind of environment is very likely to cause less productivity, employees’
absenteeism, and some physical issues like eye fatigue and headaches [23]. This inappropriate
indoor environment quality might contribute to up to $2700 loss per year of productivity and
absenteeism in the workplace [22]. Therefore, improving the lighting quality, and analyzing the
visual comfort of occupants are important issues in built environment problems.

There have been many researchers reporting that occupants might have different preferred lighting
intensity depending on several factors like age and gender [24]. Therefore, it could be concluded
that different individuals might have different visual sensations in similar light conditions and
might request different lighting intensity to perform the best work productivity [22]. Being
controlled by the pupillarity dilator and sphincter muscles, pupils could dilate and shrink in
response to different lighting intensity, and these muscles could be innervated by the sympathetic
13
systems, which is a part of the push/pull of the automatic nervous system [25]. Therefore,
conducting research on pupil size is a potential solution to analyze visual comfort of occupants.

1.3.4 Visual Comfort Analysis
Because of the reaction of pupil sizes toward the lighting intensity, several research projects have
been conducted to identify the relationship between the visual acuity and the pupil sizes [22].
However, most research topics related to pupil sizes have researched on physiological reactions to
visual stimuli, fatigue, and visual performance, but with little analysis on the visual comfort. In
building environment field, visual comfort is a significant factor related to occupants’ physical and
mental health. In addition, visual comfort of occupants is also an important field to test the
performance of electrochromic windows. Therefore, this research will investigate the analysis of
visual comfort.
In previous research by Professor Choi and his team, they conducted research on the relationship
between pupil sizes and visual comfort by conducting experiments with human subjects, using a
mobile pupilometer to measure the subjects’ pupil size and illuminance meters to measure the
illuminance levels [22].

1.4 Summary
Utilizing smart windows as a sustainable solution to improve occupants’ thermal comfort and
visual comfort is becoming more and more popular, because it helps with the heat gain and loss
through the glazing area of a building although it is more expensive to install than normal glazing.

14
This research will examine the thermal performance of application of smart windows
(electrochromic windows) in the building, including the thermal and visual comfort of occupants,
and energy consumption saving. The research will be conducted in two conditions to analyze the
performance, one is the room with smart window, and the other is with normal blinds. The goal of
the study is to analyze the impact of smart windows on the occupants’ indoor environmental
sensations based on Professor Choi’s previous research on the bio-sensing control for the smart
windows. The objective of this study is to understand the technical issue of integrating smart
windows with bio-sensing controls and establish methods to analyze the indoor environmental
quality with the bio-sensing controls for smart windows and learn the probability of the future of
the bio-sensing control and application of smart windows.

15
Chapter 2. Literature Review
2.1 Electrochromic Windows Control Technologies
In modern society, most high-rise buildings use highly glazed curtain walls as envelopes, which
offers outdoor views, natural light, and solar heat for the occupants indoors [1]. However, these
could cause more energy using and indoor environmental discomfort due to the use of HVAC
systems and solar heat gain and loss through the glazing area. Therefore, using electrochromic
windows has been a popular strategy to increase the indoor environmental quality and decrease
building energy consumption.

There are typically several conventional technologies to control electrochromic windows,
including static operation of glazing in either bleached state or colored state, on-off operation
depending on indoor light level or outdoor solar radiation, time scheduled operation, operation
based on either indoor light levels or solar radiation or luminance or indoor temperature [26].

2.1.1 Different Control Strategies
Assimakopoulos et al. [26] has conducted research on the energy consumption of using different
control strategies by building energy simulations. They used the SIBIL Building Toolbox to
implement the thermal and visual model of the PASSYS test cell. According to Assimakopoulos
et al., “The SIBIL Building Toolbox is a computational tool used for the thermal simulation of the
buildings. Its realization takes place in a MATLAB-SIMULINK environment”. Besides the control
strategies mentioned in section 2.1, Assimakopoulos et al. [26] separated the on-off control based
on indoor illuminance level and outdoor solar radiation into 3 parts, the first two were for indoor
illuminance levels with 400 lux and 500 lux being the thresholds to transfer the state of EC
16
windows, respectively, the last one was based on solar radiation level with 350 W/m
2
as the transfer
threshold. And the time scheduled control strategy was set to two time slots, one was from 11 am
to 3 pm to darken the window, while the other was from 12 pm to 4 pm. Another control strategy
has been analyzed was based on a dynamic controller (PID controller), which used the interior
average horizontal illuminance level as the input to be compared with a reference value to control
the states of the EC windows. And the last strategy was a fuzzy controller developed on the
principles of the adaptive neuro-fuzzy inference system, which used a backpropagation algorithm
based on the input/output dataset as the main function inside the fuzzy control. Two fuzzy
controllers were developed based on winter dataset and summer dataset.

2.1.2 Comparison of Different Control Strategies in Energy Consumption
According to Assimakopoulos et al. [26], the minimum energy consumption of each kind of
control technology were compared and analyzed. Figure 4 shows the simulation results of the
energy consumption in winter and summer, and Figure 5 shows the numerical data of the
simulation results for heating, cooling, and lighting energy consumption. As the result figures
show in terms of total energy consumption including heating and cooling, the strategy of the fuzzy
controller used a backpropagation algorithm for the summer period cost the least energy, which
was followed by the two on-off control strategies based on the illuminance level. However, when
including the lighting energy consumption, the on-off control strategies based on illuminance level
had the best performance. Overall, the energy consumption improvement of the electrochromic
windows compared to normal glazing, and the energy performance of the control strategies with
the capability to change for seasons would have more possibility to decrease more energy
consumption.
17

Figure 4. Heating and cooling energy consumption for winter and summer [26]

Figure 5. Energy consumption of all control technologies [26]
2.1.3 Limitations of Current Control Technologies
For the current control strategies listed previously, although the electrochromic windows could be
changed by different factors and cost less energy than the normal glazing systems in the buildings,
it seems that there is no significant advantage from using those control technologies. Furthermore,
the research on the impacts on the thermal visual comfort of different is limited. Therefore,
although the energy consumption is better in using some of the control strategies, it is difficult to
learn if the control strategy with the best energy performance would also make occupants feel the
18
best indoors. In this research, the impacts of electrochromic windows on the humans’ bio-signals
would be explored so that the potential of using bio-signals as a control strategy could be explored.

2.2 Passive Environmental Control
Nowadays, energy consumption and CO2 emission have increased considerably due to buildings,
human activities, and the burning of coal as fuel. According to the intergovernmental panel on
climate change (IPCC), building energy consumption accounts for about 40% of global energy
consumption, causing about 25% of carbon dioxide emissions [1]. In addition, people would spend
more than 90% of time indoor. Therefore, a comfort indoor environment is significant. To maintain
indoor environment comfort for occupants, almost all the buildings have installed heating,
ventilation, and air conditioning (HVAC) systems, which could account for a large amount of the
building energy consumption. However, the conventional HVAC systems could only provide a
static set level of indoor temperature and relative humidity [27] with the original control
mechanisms. By using the original control strategy, although some of occupants are satisfied with
the indoor environment, others would feel so cold or hot with the current set point of temperature,
which could also cause more energy consumed. Therefore, as one of the preferred methods, passive
environmental control is considered an effective control strategy which could not only increase
humans’ thermal and visual comfort but also decrease energy consumption in buildings.

2.2.1 Different Heating and Cooling Control Strategies
For different indoor environmental factors, there are different types of passive controls, including
passive thermal control, passive heating control, etc. Passive thermal control system is a control
strategy which controls the indoor temperature without energy cost, which includes the multilayer
19
insulation, wall surface coatings and paints, using phase change materials on envelops which
transfer heat with the exterior [28], and thermal control based on model predictive control (MPC)
[29], because there is some relationship between indoor environmental quality and humans’
thermal sensation based on previous research. For example, the Predicted Mean Vote (PMV)
model developed by Fanger [18] has been explored the potential of controlling thermal comfort of
occupants. Additionally, some researchers also explored using the occupants’ feedback to control
the thermal comfort [30], which could change the temperature based on occupants’ thermal
comfort vote.

2.2.2 Different Lighting Control Strategies
For commercial buildings, office lighting energy constitutes a large percentage in total energy
consumption. Artificial lighting control strategies, therefore, has a significant effect in terms of
energy consumption and occupants’ visual comfort. For the modern control strategies, the
appearance of light emitting diode (LED) based luminaires made it easy and flexible to control the
artificial lighting [31], so the system with sensor equipped luminaires has been more and more
popular in commercial buildings. There are several types of sensors equipped lighting controls for
modern commercial buildings, including occupant sensors-based controls and light-level sensors-
based controls [32]. By using occupant sensors equipped lighting controls, the light could be
automatically turned on and kept on without interruption when the room is occupied and turned
off when the room is vacated. There are several types of occupant sensor, such as passive infrared
sensors (PIR) which could detect the movement of a heat emitting body by their field of view,
ultrasonic sensors use an inaudible sound pattern to detect the movable objects, and dual-
technology occupant sensors use both the PIR and ultrasonic sensors. By using daylight sensors-
20
based control, the artificial lighting could be automatically dimmed or turned on / off, so that the
appropriate lighting level could be maintained indoors.

2.3 Impacts of Electrochromic Windows
According to the previous studies described in the following paragraphs, application of
electrochromic windows could not only improve energy consumption in buildings, but also
improve the indoor environmental quality and occupants’ thermal and visual comfort.

2.3.1 Impacts on Indoor Environmental Quality
According to Bo Rang Park et al.[33], by conducting simulation analysis of installation of
electrochromic windows, the results showed that when turning off the HVAC systems indoor, the
annual solar transmittance through EC windows was higher in winter months, while lower in
summer months comparing with other glazing types, which could highly reduce heating and
cooling energy consumption. For indoor temperature, the environment using EC windows also had
a slight improvement while comparing to other windows, which was a decrease of 0.99
o
C in
summer, and an increase of 1.2
o
C in winter[33]. According to the result, using EC windows could
have a huge impact on solar transmittance, which is directly related to energy consumption in
buildings, as well as a slight improvement in indoor temperature.

2.3.2 Impacts on Occupants’ Thermal and Visual Comfort
For the thermal comfort impacts of the electrochromic windows, there haven’ been many
technologies that can precisely tell, because it is difficult to measure the humans’ thermal comfort
concisely. According to Jean et al.’s research, after calculating the thermal comfort using the
21
Predicted Percentage of Dissatisfied (PPD) model, the using of smart window could improve
occupants’ visual comfort by up to 36.8%, while only improve the thermal comfort by up to
1.5%[34]. However, Jean’s team didn’t conduct research on the thermal comfort based on other
predictive models, which was a limit of their experiment. Furthermore, Ajaji[35] conducted
experiments in a chamber and used the operative indoor temperature to measure the thermal
comfort increased by the electrochromic windows. It showed that with the application of
electrochromic window, the indoor operative temperature has never exceeded 30
o
C, while the
indoor operative temperature exceeds 39
o
C at 43% of the time with the clear glazing in the same
condition. According to above research, both results show that using electrochromic windows
could improve the occupants’ thermal and visual comfort, comparing to the normal glazing, but
the result is not specific, because the first research was just based on the simulation, while the
second one used indoor operative temperature to represent the thermal comfort. Therefore, it is
important to conduct research to investigate the specific impact of the electrochromic windows on
the thermal comfort.

2.4 Thermal Comfort Prediction Models
In order to measure the impact of the electrochromic windows on the occupants’ thermal comfort,
the first thing is to know how to measure thermal comfort. Thermal comfort is a subjective
condition related to occupants’ feelings. Therefore, it is hard to measure the thermal comfort by
just one number. Nowadays, there have been different predictive models, which could predict the
humans’ thermal comfort by different input factors, like skin temperature, heart rate, brain signals,
and the feedback of the occupants.

22
2.4.1 Predicted Mean Vote and Predicted Percentage of Dissatisfied
Fanger[18] has developed two predicted models, Predicted Mean Vote (PMV) and Predicted
Percentage of Dissatisfied (PPD) to predict the thermal comfort. The former model uses air
temperature, air velocity, globe temperature, metabolic rate, clothing, and relative humidity as the
input parameters, developing an equation to calculate the thermal comfort in a determined equation,
representing the thermal comfort in the scale from -3 (cold) to +3 (hot). However, although PMV
model can predict the thermal comfort of one person, it can’t be used to predict the thermal comfort
of all occupants in a space. Then Fanger[18] developed PPD model, by which people could use
the scale calculated in PMV model to calculate the percentage of people in a space who are
dissatisfied with the thermal conditions.

2.4.2 Dynamic Mathematical Model
In 1971, Stolwijk[36] developed a dynamic mathematical model based on the 25 nodes in a body
of human. Those 25 nodes represent the thermal characters of the body, including the head, trunk,
arms, hands, legs, feet, and the central blood. The nodes represent the skin and exchange of heat
with the environment through the radiation, convection, and radiation. Therefore, this model used
the skin temperature of each part to develop the correlation of them and the humans’ thermal
comfort.

2.4.3 Other Thermal Comfort Models
The above two models are like the milestones in thermal predictive models, and most of the
following research and models were based on them. In 2002, the Adaptive Comfort Standard
model (ACS model) was developed and is suggested in ASHRAE 55 standard now[37]. This
23
adaptive model predicts that the occupants’ thermal expectations and thermal preferences could
be affected by the contextual factors and past thermal history.

Human skin is a mediator of heat transfer; therefore, skin temperature and skin conductance are
essential in the thermoregulation process[38]. According to Changzhi et.al[39], after gathering the
data of thermal sensation of the occupants, the indoor environmental quality, and the 13 occupants’
real-time local skin temperature, which were measured in 4 body parts: head, trunk, upper limbs
and lower limbs, the most accurate prediction of the thermal comfort was using the combination
of the skin temperature in chest, upper arm, and shin. As the technology is developing, the skin
temperature could now be measured by some wearable sensors like smart watch, and clothes, etc.
This model is also updated from the Stolwijk’s dynamic model, but it only measured 13 nodes of
the human body.

In addition, Choi et al.[40] also conducted several different research projects to investigate the
relationship between occupant’s thermal comfort and their physiological signals. The input factors
in this research are the physiological signals including heart rate and seven local body skin
temperatures, and human factors including the BMI, gender, and the age. As a result, the
correlation between heart rate and thermal comfort is negative, and among the skin temperatures
in different parts, forehead has the most accuracy of predicting the thermal comfort.

Besides, as the technology is developing, there are also researchers using advanced technology to
predict the thermal comfort, like infrared thermography. Because there is a high density of blood
vessels which could be used to monitor the individual’s thermoregulation performance,
24
Ghahramani[41] conducted research on using infrared thermography sensing based method to
predict thermal comfort of occupants. The infrared thermography tool they used was a pair of
eyeglass frame with infrared sensors, which could monitor the facial skin temperature. After
Ghahramani[41] collected data of thermoregulation performance during hot and cold thermal
stresses and the thermal comfort votes from 15 participants and concluding the correlation between
facial skin temperature and the thermal comfort, the possibility of using the thermoregulation
performance to predict individual’s thermal comfort was demonstrated.

2.5 Summary
There has been considerable previous research about how the application of electrochromic
windows could reduce energy consumption in buildings and improve the visual comfort; however,
research on how electrochromic windows could affect occupants’ thermal comfort was rarely
investigated. What’s more, almost all the control strategies of electrochromic windows need
occupants to actively control the windows, which could not only decrease the thermal comfort of
the occupants, but also detract the occupants from their work and to control the windows.
Therefore, if the correlation between the electrochromic windows and humans’ physiological
signals could be demonstrated, then the potential of using them as a control mechanism could both
save more energy using in the building and increase the productivity of the indoor occupants.

25
Chapter 3. Methodologies
3.1 Methodologies Overview
This chapter introduces the overall methodologies used in the research process of analyzing the
relationship between electrochromic windows and humans’ bio-signals. The main research process
included several steps: 1) prepare two environments for the experiments, one with electrochromic
glazing and the other with the normal clear glazing; 2) collect the indoor environmental quality
data and bio-signal data of the human subjects in both environments; and finally, 3), analyze all
the data using machine learning methods to find out the relationship between the utilization of
electrochromic windows and the humans’ bio-signals.

Figure 6. Overall Workflow
3.1.1 Experiment Preparation

For this research, 30 human subjects would participate in the experiments, 15 for the first-round,
and 15 for the second-round experiment. After the collection of data, machine learning was used
to conduct data analysis, and finding out the relationship between human bio-signals and
electrochromic windows.
26

Because there was no appropriate office space with electrochromic glazing available for the
research, a sample facade of electrochromic glazing was used for the experiment by being exposed
to simulated sunlight. To mimic sunlight in the chamber, 5 infrared heat lamps were used to
simulate the solar heat and sunlight. For the experiment, the simulated outdoor environmental
quality was maintained the same all the time, so that the impact of outdoor environment was
eliminated. For the first-round experiment, the electrochromic window was turned on in different
states during one experiment, and each state stayed for about 30 minutes. For the second-round
experiment, the electrochromic windows were controlled by the human subjects’ bio-signals, to
verify if it could be controlled correctly.

There were 30 student subjects of different genders and at the age from 20 to 40 participated in the
experiments, and each participant participated in only one experiments to increase the validation
of the second-round experiment. During the experiments, all the subjects were asked to wear the
sensors all the time, so that the bio-signals could be collected in real time. After finishing the data
collection of all the subjects, machine learning skills like random forest and linear regression were
used to analyze the data and explore the relationship between the electrochromic windows and
human bio-signals.

3.1.2 Chamber Preparation
The room in the lower level of Watt Hall at the University of Southern California was used as the
experimental chamber in this research. The floor plan is shown in Figure 6. There are no windows
27
in the chamber, so the EC glass could be used as the glazing in this area. The EC glass for research
was put beside the desk, just mimic the environment in the office with EC glazing.

Besides the EC glass, this research also needs 5 heat lamps to simulate the daylight and solar heat,
and a hanger rack to fix the lamps.

Figure 7. Floor Plan of Chamber
28

Figure 8. Chamber Preparation
3.1.2.1 EC Glass Preparation
In the research, a piece of EC glass made by SageGlass was used. Before the research process,
some basic features of the glass and how to use it should be introduced.

The type of EC glass used in this research has 4 states per insulated glass unit (IGU), which could
be controlled by a controller shown in Figure 8. This EC glass has 4 states related to different
visible light transmittance, including fully clear (60%), light tint (20%), medium tint (6%), and
fully tinted (1%) [42]. By pushing the black or the white button on the controller, the EC glass
could become darker or lighter. If there is a daylight sensor, this glass could also be controlled by
the daylighting. The four pictures below show 4 states of the EC glass.

29

Figure 9. EC Glass Clear State

Figure 10. EC Glass Light Tint
30

Figure 11. EC Glass Medium Tint

Figure 12. EC Glass Fully Tinted State
31
According to the user guide [42], this type of EC glass will take 5-10 minutes to transfer from one
state to another state, and the time might be longer in the winter or there isn’t direct sunlight. For
this research, before conducting the research, the transferring time was measured by a timer for
each state.

Figure 13. EC Glass Controller
After understanding how to use the EC glass, the controller and the glass should be installed
correctly. The installation guide was shown in Figure 9. First, the IGU pigtail should be connected
to the window frame, which could connect the EC glass to the wire cable, which is used to connect
to the controller. Then, connect the cable wires and the controller, by connect each with the specific
color (black, white, and red). After this step, the controller has been connected to the EC glass.
Next, connect the power supply extension and the controller. Then, before starting controlling the
EC glass, connect the power extension cable and the power supply, which means the controller
32
was connected to the power and ready to be used. Finally, just power up the glass, then the
controller could successfully control the glass [42]. Like Figure 9 shows, there are 4 channels on
the back of the controller, it means one controller can control up to 4 IGUs and each IGU should
connect to only one channel. However, in this research, only one IGU will be used, so only one
channel was connected.

Figure 14. Installation Guide [42]
3.1.2.2 Other Devices Preparation
Besides the preparation of the EC glass, this research also needs 5 infrared heat lamps to simulate
the sunlight and solar radiation. The heat lamps used in the research are from Philips, with features
of 250 Watt, BR40 (the diameter of the lamp is 5”), and 120 Volt [43]. The actual heat lamp was
shown in Figure 11.
33

Figure 15. Features of Heat Lamp [43]

Figure 16. Infrared Heat Lamp

In addition, five bulb guards are needed as shown in Figure 12 to hold the heat lamps and provide
power. A hanger rack was also needed to fix the heat lamps behind the EC glass. The hanger rack
was shown in Figure 13, and the height of it is 64 inches.
34

Figure 17. Bulb Guard [44]

Figure 18. Hanger Rack [45]

3.1.3 Sensors and Software
In this research, both environmental sensors and wearable sensors as shown in Figure 19 were used
to monitor the environmental data and physiological data. Remote sensor HOBO MX1102 was
used for measuring the indoor environmental data including ambient temperature, relative
humidity, and carbon dioxide concentration, where the data collection interval time is 1 minute.
Wearable sensors including Garmin Vivosmart 3 and Empatica Embrace were used to monitor and
35
collect the human subjects’ bio-signals in real time. Garmin Vivosmart 3 was used to measure
heart rate and stress level, while Empatica Embrace was used to measure the electrodermal activity
data (EDA) and skin temperature. The data collection interval time is also 1 minute when collecting
the bio-signal data.

With human subjects wearing and installing the sensors, the data was collected in real time and
exported to their mobile application including Garmin Connect Mobile, and Mate App. To collect
indoor environmental quality data, HOBO ware was used. After collecting data each time, the data
was automatically exported the database on the computer. Since the data collected from Garmin
Vivosmart 3 and Empatica Embrace was not in csv format, it was transferred into a CSV file first
using Python, to prepare for the data analysis process later. Besides transferring data into different
formats, Python was also used to classify data, analyze data, and machine learning, so that no other
software would be used in this research.

36

Figure 19. Experimental and Wearable Sensors

Besides the wearable sensors and indoor environmental sensor, a sunlight meter and a solar
radiation meter were also used in this research. Although the aim of this study isn’t related to the
visual comfort and visual sensation, they were just used as a reference of the EC glass. Sunlight
meter used in this research is Omega HHLM-1. The accuracy is 3%, the measurement rate is
2.5/second, the light range it can measure is from 20 to 200000 Lux. While using this light meter,
the sensor part of the meter should be put on the work plane.

37

Figure 20. Sunlight Meter

This research also involved the use of a radiation meter, which could be used to measure the solar
radiation on the work plane. In this research, an infrared camera was used, which could take photos
with the radiation temperature. The infrared camera used was shown in Figure 21, which was made
by Flir, providing detail-rich 320 * 240 pixels thermal images and storing up to 200 photos or
videos [46].
38

Figure 21. Infrared Camera [46]
3.2 Experiment Process
To explore the relationship of the electrochromic windows and humans’ bio-signals, two rounds
of experiments should be conducted, and each experiment should be conducted several times in
same environment with different states of the electrochromic windows.

3.2.1 Introduction of Experiment
The experiment was conducted with 15 participants to collect data of the participants, and the data
then was analyzed and used to develop a prediction model. During the experiment, the human
subjects were allowed to do their own work but asked to wear the sensors in the experimental
chamber for about 2 hours, so that the environmental quality and their bio-signals data including
skin temperature, heart rate, and EDA data could be recorded in real time. In the experiments, the
electrochromic windows were changed to different states to explore if they could impact the
39
subjects’ physiological data. As shown in Figure 17, the participants should wear the sensors all
the time during the experiments with the HOBO sensor being put on the workplace.

Figure 22. Experiment Process

3.2.2 General Process of the Experiment
For each experiment for each participant, the background information of them was collected,
including their gender, age, height, weight, and BMI index. After collecting the information, the
researcher waited until the EC glass was fully transferred to another state, which took 10 minutes.
Then, the heat radiation and light irradiation were collected as a reference, and keep all things
maintain for 10 minutes. After 10 minutes, transfer the glass to another state, and repeat all the
above steps. Because the glass has 4 states in total, each experiment was lasted for about 2 hours.

40
3.3 Data Collection
In this process, both the indoor environmental quality data and subjects’ bio-signals were collected
using the sensors. Besides, the subjects’ background information was collected as well.

3.3.1 Background Information of Subjects
To decrease the impacts of factors other than the electrochromic windows, the background
information of the human subjects should be collected as well. Because the subjects’ age, gender,
and nationality are not changed, they were asked to input those kinds of data in a questionnaire
before the experiments. After collecting all the data, it was analyzed by Python, Weka, and Minitab
using several data analysis methodologies. Table 1 shows the background information of all human
subjects.
Table 1. Background Information
ID Gender Age Nationality
1 Female 25 China
2 Female 24 China
3 Female 23 China
4 Male 23 China
5 Male 31 Bangladesh
6 Male 26 China
7 Female 25 China
8 Female 23 China
9 Female 26 China
10 Female 22 China
11 Male 22 China
12 Female 28 United States
13 Male 34 China
14 Male 25 China
15 Male 23 China

41
3.3.2 Subjects’ Bio-signals
In this step, the subjects’ bio-signals were collected in real time by wearing the sensors, including
Garmin Vivosmart 3 and Empatica Embrace. Garmin Vivosmart 3 was used to measure heart rate
and stress level of the human subjects, while Empatica Embrace measured EDA data and skin
temperature. The pictures shown below were heart rate data collected from Garmin Connected
web application. After installing the software like “Garmin Connect Mobile” and “Mate App”,
data from Garmin and Empatica were able to be seen in real time in the smart phone. Because the
data collected from Garmin watch and Empatica watch were not in CSV format, which was a
convenient format for data analysis, the data were transferred into CSV format by using Python to
prepare for analyzing. The data collected by Garmin watch was in 1 minute interval, which means
the data was collected per minute, while the data collected by Embrace watch was in milliseconds.

Figure 23. Garmin Data Collection

3.3.3 Indoor Environmental Quality Data
To explore the impacts of the indoor environmental quality on the subjects’ bio-signals and the
impacts of EC glass on the indoor environment, indoor temperature, relative humidity, and CO2
concentration were collected by HOBO. By using HOBO MX1102, temperature between 0
o
C and
42
50
o
C could be measured and collected. All data measured were stored in HOBO ware web
application. There is also a screen on the HOBO MX1102, where the data measured in real time
could be seen. Because indoor ambient temperature could be affected by several factors like
individual activities and locations, the HOBO sensor should not be put close to the HVAC system.
Data collected by HOBO sensor was in 1 minute

Figure 24. Indoor Environmental Quality Data

3.3.4 Participants’ Feedbacks
After the EC glass was fully transferred to another state, the participants’ feedbacks were also
collected, which was in the survey format. The survey included several questions regarding the
participants’ visual comfort, visual sensation, thermal comfort, and thermal sensation. All the data
was in 7 scales.

3.4 Data Analysis
In this process, machine learning skills like linear regression, decision tree and random forest were
used to analyze the data collected from the sensors. And most analysis process were conducted
43
using Python, and some additional software like Weka and Minitab were used to assist with the
correctness of data analysis.

3.4.1 Data Cleaning and Processing
Before conducting data analysis for the research, the collected data was cleaned first. In the
beginning of data cleaning, the missing data needs to be found. Because there are a lot of reasons
could cause the wearable sensors not collecting data, like the wearing position on the wrist, or they
were out of charge, the collected data would include a lot of missing data. Therefore, the first step
was to find all the missing data and decided process them in different situations. For example, if
the missing data was caused by wearing the sensor too loosely, then all the data in different sensor
at the same time should be deleted. The second step was to delete all the irrelevant and unnecessary
data. Because if the data collected was too small or too big, which was obviously abnormal, then
the data should be deleted without consideration.

Besides, the data collected also needs to be transferred and maintained in normal format. First, for
data collected by HOBO sensor and Garmin watch, because they all collected data per minute, the
data was cleaned in an easy way, just by maintaining the data used for the research, which was in
10 minutes for each state from the EC glass was fully transferred into another state. However, the
data cleaning for Embrace was more complicated because it collected data per millisecond, which
was a huge number of data compared to the others. What’s more, the time stamp to collect data of
Embrace was represented in the total seconds from January 1st, 1970. Therefore, in this case, the
timestamp in the file was first transferred into the current time in normal time format, which was
44
Year-Month-Day. Then, the data collected was used Python script to maintain the data in the time
range was needed for the research.

3.4.2 Data Analysis Process
In this research, each participant had a collection of data, and data analysis was conducted for each
participant to see the correlations between each human bio-signals and visual and thermal
sensations. Besides, machine learning algorithm was also used to create a control model using the
aggregated dataset to predict the visual and thermal sensations of participants. The individual data
analysis didn’t consider the individual background information. Besides the individual data
analysis, an overall impact model was also developed, with which could predict in such indoor
environmental quality, which visual sensation and thermal sensation a participant could be in, and
which state of EC glass should be transferred to. Because the overall impact model was a prediction
model for all participants, the background information was considered.

After processing the collected data, the data was analyzed for individual participant first. In
individual data analysis part, ANOVA analysis and correlation were used to define the correlations
between each type of bio-signals and visual sensation, thermal sensation, and window states. After
analyzing all the individual data, a summary chart was created to explain each participant data
analysis more clearly. And correlation and ANOVA could be used as a baseline which was easier
to introduce the correlation between the parameters. After finishing the individual data analysis
part, the aggregated dataset including the background information of each participant was
separated in training set and testing set. Then several machine learning algorithms were used to
train and test the data. In this research, random forest was used to create the model to use those
45
parameters to predict visual or thermal sensations, and decision tree was used to create the model
to decide the EC window state using the visual and thermal sensations, because random forest and
decision tree has high prediction accuracy. The decision tree model used J48 decision tree
algorithm, and random forest model was developed by using the "Scikit-learn Ensemble" package
in python.

In this part, the correlation between different pairs of variables like bio-signals and visual
sensations, and bio-signals and thermal sensations were analyzed using Minitab software. After
getting the correlations. And after getting the correlations of them, random forest and decision tree
were used to create the prediction model.

3.4.3 Output of the Data Analysis
After analyzing all the collected data and developing the prediction models, the output in this part
should be the thermal sensation and visual sensation, respectively, which were used to predict the
thermal sensation and visual sensation based on human subjects’ bio-signals. Then in the future
when the model is used, the EC glass could transfer to correct state just based on prediction of the
thermal sensation and visual sensation.

3.4.4 Result Verification
After developing the prediction model, a second-round experiment was conducted to test the
accuracy of the prediction model and optimize it. In this step, for each participant, the experiments
should be conducted the same way as the first round, and then do the data analysis to check if the
thermal sensation and visual sensation predicted by models are correct. After the development of
46
the control model of the window states, an operation experiment also should be conducted to verify
if the control model works for residents.

3.5 Summary
Chapter 3 mainly explained the overall methodologies used in the research part, including the
preparation of experiment environment and experimental devices, major process to conduct the
experiment, data collection and analysis process. In chapter 4, the details of developing prediction
model and the results of data analysis was explained.

47
Chapter 4. Data Analysis and Results
In this chapter, the process of data cleansing, data preprocess, and data analysis was explained in
detail. For data analysis methodologies, Anova analysis and Correlation analysis was used to
analyze the relationship between window states, thermal sensations, visual sensations and different
human bio-signals data collected from the experiment.

4.1 Data Processing
About 15 tests with the electrochromic (EC) glass and human subjects were conducted to evaluate
the relationships between humans’ bio-signals including heart rate, stress level, skin temperature,
and EDA data, and different states of the window. In this section, the detailed process of data
analysis was described, and the results for each analysis was shown.

4.1.1 Data Preprocess
The raw data collected from the sensors had to be preprocessed before analysis, because different
types of data had different formats. First, the heart rate and stress level data from Garmin watch
were in Fit format, which can’t be analyzed directly; therefore, those data were transformed to
CSV format using Python. What’s more, although EDA and skin temperature data collected from
Embrace watch were in CSV format, which could be analyzed directly, they were collected in
millisecond intervals and the time was represented in Unix Timestamp (UTC). The analysis for
this work needed to use one-minute intervals and be in PST time, so the data were transferred into
one data point per minute by using Python.

48
After transforming all the data of the participants into target formats, because there are some data
were unreasonably far from the average data of all people, then those data were removed from the
database. For example, if the stress level and EDA data was negative value, then it means the
wearable sensors were not attached to the skin tightly some time; therefore, the negative stress
level and EDA data were removed from the dataset. What’s more, the normal heart rate data is
expected to range from 50 BPM to 130 BPM, so the heart rate data out of this range were deemed
inaccurate and were removed. Furthermore, there were also some data lost due to irregular wearing
of the watches, then those data were filled with the average data of this participant within the same
range of time. The figure below shows some sample data in the dataset.

Figure 25. Sample Aggregated Data in Dataset

4.1.2 Database Information
Fifteen human subjects participated in the chamber experiment located in the basement of Watt
Hall. After preprocessing, the data were transferred into the smart window experiment database.
The data were collected per minute for the research, and total number of data points was 9900,
including the background data of the participants, real time bio-signals data, real time thermal and
49
visual sensitivity feedbacks, and real time indoor environmental quality data, where each
participant had 11-minute data for bio-signals per state of the EC window.

4.2 Individual Data Analysis
This section describes the methods used for individual data analysis and the results for the
relationship between EC glass, human subjects’ bio-signals, and thermal and visual sensitivities.
All the individual data analysis were conducted using Minitab software.

4.2.1 ANOVA Analysis Between Window States and Human Bio-signals
ANOVA test refers to Analysis of Variance, which could be used to check if there is any statistical
difference between the means of 3 or more independent groups. Therefore, in this section, ANOVA
tests were utilized to compare the differences among the human bio-signals related to different EC
glass states for each human subject using Minitab.

By conducting ANOVA tests, significant differences of mean of bio-signals related to the various
window states could be analyzed and showed. Some participants have the decreasing trends
between the EDA data related to different window states. That means, as the window state turned
to more and more dark, the EDA data of the participants became higher and higher. Similarly,
some participants also have the decreasing trends with skin temperature, stress level, and heart rate
related to different window states. While there are also other participants have the increasing trends
of different bio-signals related to different states of the EC glass. There are 4 states for EC glass,
where state 1 represents the clear state with 60% visible light transmission, state 2 represents the
second clear state, with 20% visible light transmission, state 3 represents the medium tinted state
50
with 6% visible light transmission, and state 4 represents the fully tinted state with 1% visible light
transmission. As shown in the summary table, all the Anova analysis between different human
bio-signals and window states has a high significant rate, where more than 50% human subjects
have a significant result in the Anova analysis. For the Anova analysis between EDA data and skin
temperature, all the participants have a significant result.
Table 2. Summary of Window State Individual Anova Analysis Result
ID 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Significant
Rate
Window and HR 1 1 1 0 1 0 1 1 0 1 1 1 1 0 0 66.67%
Window and SL 1 1 0 0 1 1 0 1 1 1 1 0 1 1 0 66.67%
Window and
EDA
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 100%
Window and ST 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 100%

4.2.2 ANOVA Analysis Between Thermal Sensation and Human Bio-signals
Similar to the ANOVA tests between window states and human bio-signals, the changes of bio-
signals related to thermal sensations of each participant were also conducted. Thermal sensation
was in 7-point scale, which could represent the feeling of the thermal conditions of the participants.
The thermal sensation scale was shown in Table 3 below. In this analysis, some significant results
showed that each human bio-signal was decreased while the thermal sensation was increased;
however, others had the increased bio-signals while their thermal sensation was increased. It means,
some participants’ heart rate, EDA data, stress level and skin temperature were increased when
they felt hotter and hotter, while others were decreased during the same situations. As the summary
of all individual Anava Analysis between thermal sensation and bio-signals showed, 1 represents
there are significant differences of bio-signals among different thermal sensations. All the Anova
analysis for the human subjects have a high significant rate, which is higher than 50%, where the
EDA has highest significant rate for the Anova analysis.
51
Table 3. Thermal Sensation Scale
Scale Thermal Sensation
-3 Cold
-2 Cool
-1 Slightly Cool
0 Neutral
1 Slightly Warm
2 Warm
3 Hot

Table 4. Summary of Thermal Sensation Individual Anova Analysis Result
ID 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Significant
Rate
Thermal and
HR
1 0 1 0 0 0 1 1 0 1 1 1 1 0 0 53.3%
Thermal and
SL
1 1 0 0 0 1 0 1 0 1 1 0 1 1 0 53.3%
Thermal and
EDA
1 1 1 1 1 1 1 1 0 1 1 1 1 1 0 86.67%
Thermal and
ST
1 1 1 1 1 1 1 1 0 1 0 1 1 1 0 80%

4.2.3 ANOVA Tests Between Visual Sensation and Human Bio-signals
This section analyzed the relationship between the human bio-signals and difference visual
sensations. Visual sensations were also represented by a 7-point scale, which showed the feeling
of participants toward the visual condition while working. The details of the scale of visual
sensations were shown in the below. There were also some significant results for this analysis,
some participants had a decreased bio-signals pattern while their visual sensations were increased,
while others had an increased pattern in the same conditions. As shown in the summary table, more
than 50% of human subjects have a significant result in the Anova analysis between visual
sensation and bio-signals, where 100% of human subjects have a significant result for the Anova
analysis between visual sensation and EDA data and skin temperature.
52
Table 5. Visual Sensation Scale
Scale Visual Sensation
-3 Very Dark
-2 Dark
-1 Slightly Dark
0 Neutral
1 Slightly Bright
2 Bright
3 Very Bright

Table 6. Summary of Visual Sensation Individual Anova Analysis Result
ID 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 Significant
Rate
Visual and HR 1 0 1 0 1 0 1 1 0 1 1 1 1 0 0 60%
Visual and SL 1 1 0 0 0 1 0 1 1 1 1 0 1 1 0 60%
Visual and EDA 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 100%
Visual and ST 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 100%

4.2.4 Correlation Between Thermal Sensations, Visual Sensations and Human Bio-signals
In this section, the correlation between the human bio-signals and thermal sensations, and visual
sensations were analyzed for each participant. The correlation could show the strength of the
relationship between two variables, and it is expressed in coefficient value, which is between -1 to
1. A positive coefficient value means if one variable becomes larger or smaller, then the other
variable turns to the same direction. And a negative coefficient value means the two variables will
go to different directions when they change. Below are the results for each participant. All the
correlation results are statically significant with a P-value less than 0.05. The table below shows
the coefficient value of all human subjects in this research, almost all the participants have a high
coefficient value of larger than 0.3.
53
Table 7. Correlation Analysis Results Summary
ID Gender Visual
and HR
Thermal
and HR
Visual
and SL
Thermal
and SL
Visual
and
EDA
Thermal
and
EDA
Visual
and ST
Thermal
and ST
1 Female 0.557 0.585 0.796 0.854 -0.688 -0.807 -0.944 0.988
2 Female 0.175 0.147 0.477 0.443 -0.504 -0.38 -0.863 -0.855
3 Female 0.484 0.419 -0.263 -0.244 0.739 0.692 -0.818 -0.826
4 Male 0.281 0.155 0.109 0.009 -0.967 -0.854 0.679 0.54
5 Male 0.463 -0.179 0.293 -0.072 -0.857 0.8 -0.781 0.871
6 Male 0.093 0.038 -0.496 -0.393 -0.313 -0.519 -0.169 -0.254
7 Female 0.479 0.43 -0.247 -0.18 -0.628 -0.717 -0.777 -0.829
8 Female -0.327 -0.564 -0.229 -0.393 -0.793 -0.626 0.492 0.626
9 Female 0.189 0 0.883 0 -0.961 0 -0.838 0
10 Female 0.5 0.54 0.209 0.276 0.79 0.688 -0.643 -0.546
11 Male 0.776 0.64 0.748 0.667 -0.815 -0.366 0.366 0.07
12 Female 0.451 0.451 0.056 0.056 0.367 0.367 -0.786 -0.786
13 Male 0.427 0.669 0.186 0.446 0.252 0.484 0.568 0.582
14 Male 0.36 -0.308 0.514 -0.467 -0.959 0.927 -0.961 0.843
15 Male 0.203 0 0.016 0 -0.654 0 -0.293 0

4.3 Data Analysis Differences by Gender
This section is mainly about analyzing the differences of the above data analysis results caused by
gender, to explore if different genders of human subjects could have a huge impact on different
data analysis. Besides, when comparing the gender differences of correlation analysis, 2-Sample
T Test was used as the data analysis method. Because all participants were at the age range from
20 to 35, which are very similar, therefore, this study only considers the impact of genders.

4.3.1 Anova Analysis Result Differences by Gender
Table 8 shows the significance for each test for each participant, where 1 represents the differences
of the variables among different conditions are significant, and 0 represents the differences are not
significant. And the differences between different genders were represented by the rate of
significant results among all participant’s results. As illustrated in Figure 26, the different rate of
54
significant results compared by genders had some differences. For the Anova test between thermal
sensation and EDA data, visual sensation and EDA data, visual sensation and skin temperature,
window states and EDA data, and window states and skin temperature, there are no differences
between female and male, all the results are significant. For Anova test between thermal sensation
and stress level, and window states and stress level, the results are similar for female and male,
where the significant rate for male is slightly higher than female. For Anova test between thermal
sensation and heart rate, thermal sensation and skin temperature, visual sensation and heart rate,
visual sensation and stress level, and window states and heart rate, the significant rate for female
is higher than male.
Table 8. Anova Analysis Results Summary by Gender
ID 1 2 3 7 8 9 1
0
1
2
Femal
e Rate
4 5 6 1
1
1
3
1
4
1
5
Male
Rate
Significan
ce Rate
Thermal
and HR
1 0 1 1 1 0 1 1 0.75 0 0 0 1 1 0 0 0.29 53.33%
Thermal
and SL
1 1 0 0 1 0 1 0 0.5 0 0 1 1 1 1 0 0.57
1
53.33%
Thermal
and EDA
1 1 1 1 1 0 1 1 0.875 1 1 1 1 1 1 0 0.85
7
86.67%
Thermal
and ST
1 1 1 1 1 0 1 1 0.875 1 1 1 0 1 1 0 0.71
4
80%
Visual
and HR
1 0 1 1 1 0 1 1 0.75 0 1 0 1 1 0 0 0.42
9
60%
Visual
and SL
1 1 0 0 1 1 1 0 0.625 0 0 1 1 1 1 0 0.57
1
60%
Visual
and EDA
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 100%
Visual
and ST
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 100%
Window
and HR
1 1 1 1 1 0 1 1 0.875 0 1 0 1 1 0 0 0.42
9
66.67%
Window
and SL
1 1 0 0 1 1 1 0 0.625 0 1 1 1 1 1 0 0.71
4
66.67%
Window
and EDA
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 100%
55
Window
and ST
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 100%

Figure 26. Gender Rate of significant Result
4.3.2 Correlation Analysis Differences by Gender
Table 7 shows all the results of correlation analysis for each participant. The numbers in the cell
are the coefficient value of the correlation results, which represents the degree of the correlation
between two variables. In order to show the difference of correlation between different genders,
2-Sample T test was used for the analysis.

For each correlation between different variables, the 2-Sample T Test P-value is shown in the
below tables.

4.3.2.1 2-Sample T Test Between Visual Sensation and Heart Rate
As shown in the below tables, the mean coefficient values for female and male are similar, and
both represent the visual sensation and heart rate is in a positive correlation. And the P-value for
the 2-Sample T Test shows that there are no significant differences between genders in the
correlation.
0
0.2
0.4
0.6
0.8
1
1.2
1 2 3 4 5 6 7 8 9 10 11 12
Rate of Siginificant Result
Analysis ID
Anova Test Result Compared By Gender
Female Rate Male Rate
56
Table 9. Descriptive Statistics Between Visual Sensation and Heart Rate
Gender N Mean StDev SE Mean
F 8 0.314 0.296 0.10
M 7 0.372 0.220 0.083

Table 10. P-value Between Visual Sensation and Heart Rate
T-Value DF P-Value
-0.44 12 0.670

4.3.2.2 2-Sample T Test Between Visual Sensation and Stress Level
As shown in the below tables, the mean coefficient values for female and male are similar, and
both represent the visual sensation and stress level has a little correlation. And the P-value for the
2-Sample T Test shows that there are no significant differences between genders in the correlation.
Table 11. Descriptive Statistics Between Visual Sensation and Stress Level
Gender N Mean StDev SE Mean
F 8 0.210 0.465 0.16
M 7 0.196 0.395 0.15

Table 12. P-value Between Visual Sensation and Stress Level
T-Value DF P-Value
0.07 12 0.949

4.3.2.3 2-Sample T Test Between Visual Sensation and EDA
As shown in the tables below, the mean coefficient values for female and male are similar, both
representing there is a negative correlation between the 2 variables, and the 0.210 P-value indicates
there are no significant differences between genders on the correlation.

57
Table 13. Descriptive Statistics Between Visual Sensation and EDA Data
Gender N Mean StDev SE Mean
F 8 -0.210 0.720 0.25
M 7 -0.616 0.444 0.17

Table 14. P-value Between Visual Sensation and EDA Data
T-Value DF P-Value
1.33 11 0.210

4.3.2.4 2-Sample T Test Between Visual Sensation and Skin Temperature
As shown in the tables below, the mean coefficient value for female shows there is negative
correlation between two variables, while male shows there is no correlation. But the 0.086 P-value
indicates there are no significant differences between genders.
Table 15. Descriptive Statistics Between Visual Sensation and Skin Temperature
Gender N Mean StDev SE Mean
F 8 -0.647 0.468 0.17
M 7 -0.084 0.648 0.24

Table 16. P-value Between Visual Sensation and Skin Temperature
T-Value DF P-Value
-1.90 10 0.086

4.3.2.5 2-Sample T Test Between Thermal Sensation and Heart Rate
As shown in the tables below, the mean coefficient values for female and male are similar, both
representing there is little correlation between thermal sensation and heart rate. And the 0.601 P-
value indicates there are no significant differences between genders.
Table 17. Descriptive Statistics Between Thermal Sensation and Heart Rate
Gender N Mean StDev SE Mean
F 8 0.251 0.384 0.14
M 7 0.145 0.379 0.14

58
Table 18. P-value Between Thermal Sensation and Heart Rate
T-Value DF P-Value
0.54 12 0.601

4.3.2.6 2-Sample T Test Between Thermal Sensation and Stress Level
As shown in the tables below, there are little correlation between thermal sensation and stress level
both for female and male, and the 0.732 P-value indicates there are no significant differences
between genders.
Table 19. Descriptive Statistics Between Thermal Sensation and Stress Level
Gender N Mean StDev SE Mean
F 8 0.102 0.409 0.14
M 7 0.027 0.412 0.16

Table 20. P-value Between Thermal Sensation and Stress Level
T-Value DF P-Value
0.35 12 0.732

4.3.2.7 2-Sample T Test Between Thermal Sensation and EDA
As shown in the tables below, the mean coefficient values show there is little correlation between
thermal sensation and EDA data both for female and male, and 0.636 P-value indicates there are
no significant differences between genders.
Table 21. Descriptive Statistics Between Thermal Sensation and EDA Data
Gender N Mean StDev SE Mean
F 8 -0.098 0.622 0.22
M 7 0.067 0.687 0.26

Table 22. P-value Between Thermal Sensation and EDA Data
T-Value DF P-Value
-0.49 12 0.636

59
4.3.2.8 2-Sample T Test Between Thermal Sensation and Skin Temperature
As shown in the tables below, the mean coefficient values show there is a negative correlation
between thermal sensation and skin temperature for female, while there is a positive correlation
for male. But the 0.056 P-value indicates there are no significant differences between genders.
Table 23. Descriptive Statistics Between Thermal Sensation and Skin Temperature
Gender N Mean StDev SE Mean
F 8 -0.278 0.733 0.26
M 7 0.379 0.441 0.17

Table 24. P-value Between Thermal Sensation and Skin Temperature
T-Value DF P-Value
-2.13 11 0.056

4.4 Summary
Chapter 4 showed the discussion of data preprocessing, data analysis, and results for data analysis.
First, because the raw data extracted from the variable sensors are in a large amount, and the
formats of data are different, which is hard for data analysis, and developing a prediction model in
the future, the data were preprocessed by Python programming to transfer different formats of data
into CSV format, which was easier to be processed later. In addition, there are a lot of invalid data
in the processed data, then data cleaning process was conducted to remove the invalid data from
the dataset.
• Anova Analysis Result

After cleaning the data, Anova Analysis for each participant between thermal sensation and bio-
signals, visual sensation and bio-signals, and window states and bio-signals were conducted to see
if there are significant differences of different bio-signals under different conditions. Almost all
60
the groups of analysis results showed there were significant differences of bio-signals among the
thermal sensations, visual sensations, and window states. And according to the aggregated analysis,
all tests have more than 50% of participants had a significant result. In addition, when compared
the significant rate between genders, more tests of female had a higher significant rate of
differences than male.
• Correlation Analysis Result
Then, the Correlation between thermal sensation and bio-signals, and visual sensation and bio-
signals were conducted to see if there is any correlation between the bio-signals and participant’s
sensation towards thermal and visual conditions. All the correlation showed there were significant
differences for the results, and almost all tests had a correlation with a high coefficient value. And
2-Sample T Test was used for analyzing the difference between genders among the correlation
results, and the results showed there were no significant differences of correlation between genders.

In the Chapter 5, several different prediction models will be developed and their results will be
analyzed.

61
Chapter 5. Development of Prediction Model
This section is the process and analysis of the development of the prediction model of participants’
thermal and visual sensations. The models were developed by using Decision Tree machine
learning algorithm, and the process was conducted by Weka. In this study, the prediction models
used 15 participants’ dataset, which includes 638 instances. All the modes used a 10-fold cross-
validation test option to simulate in Weka. Because this model is used to classify the levels of
thermal visual sensation, the most significant result is the accuracy rate, which means the
percentage of correctly classified attributes in the total output attributes. Therefore, the researcher
used accuracy rate to analyze the accuracy of the model, and the higher rate means a better model
accuracy.

5.1 Thermal Sensation Prediction Model
All the prediction models were developed by using Decision Tree machine learning algorithm,
which is a prediction algorithm based on several difference input attributes. The result of Decision
Tree could be visualized by a tree, including the root, branches, and leaves, which could also show
the significant level of different input attributes related to the predicted attributes. This section is
describing the process and result of the thermal sensation prediction model.

In order to develop the prediction model and explore the significant levels of different participants’
information related to the thermal sensations, the participants’ heart rate, stress level, skin
temperature, EDA data, and their gender information were selected as the input attributes. Because
Decision Tree is a classification algorithm, and in order to develop a more accurate prediction
model, the thermal sensation was changed from numeric type to nominal type, and two models
62
which predicted 7-point scale and 3-point scale respectively was developed to determine a more
accuracy model.
Table 25. Convert 7-Point Scale to 3-Point Scale
7-Point Scale 3-Point Scale
-3 - (-1) -1
0 0
1 - 3 1

As the Figure 27 shows, the root of the 3-point thermal sensation decision tree is gender, which
means gender is the most significant input attribute correlated to participants’ thermal sensation
among all attributes. While as illustrated in Figure 28, the root of the 7-point thermal sensation
decision tree is skin temperature. According to Weka, the results for 3-point and 7-point prediction
model were shown in Table 26 and Table 27, where the overall accuracy rate is 78.9969% and
71.6301%, respectively, representing both prediction models have a high prediction accuracy,
while the 3-point thermal sensation model has a better accuracy. And for different classes, true
positive rate means the percentage of correctly predicted classification in all the predicted results.
In the 3-point prediction model, the prediction of 1 has the highest true positive rate among all 3
classes, which might because there is less data for class -1 and 0, representing this model could
better predict when the thermal sensation level is 1.

63

Figure 27. 3-Point Thermal Sensation Decision Tree Visualization
64

Figure 28. 7-Point Thermal Sensation Decision Tree Visualization
65
Table 26. Result of 3-Point Thermal Sensation Model
Class True Positive Rate
Overall 78.9969%
-1 70.5%
0 74.7%
1 85.4%

Table 27. Result of 7-Point Thermal Sensation Model
Class True Positive Rate
Overall 71.6301%
-2 65.5%
-1 55.8%
0 75.3%
1 79.8%
2 71.1%
3 0.00%

5.2 Visual Sensation Prediction Model
In order to explore develop the prediction model of visual sensation using humans’ information
and bio-signals, the input attributes of the model are also the gender, heart rate, stress level, skin
temperature, and EDA data. For the visual sensation prediction model development, the 3-point
model and 7-point models are used for comparison regarding the accuracy of the models. The
converting methodology is the same as the thermal prediction models.

In the 3-point visual sensation decision tree, as shown in Figure 29 the root is EDA data,
representing the EDA data is the most significant attribute to predict the visual sensation in 3-point
scale. However, as shown in Figure 30, gender is the most significant attribute to predict the 7-
point scale visual sensation. According to the result of the models, the overall accuracy rate is
80.4075% and 73.35425 for 3-point visual sensation model and 7-point visual sensation model,
respectively, representing the 3-point model has a better classification accuracy. The true positive
66
rate is high for every class in the model, where class 1 is more correctly predicted in 3-point
prediction model, and class -2 and 1 are more correctly predicted in 7-point prediction model.

67

Figure 29. 3-Point Visual Sensation Decision Tree Visualization
68

Figure 30. 7-Point Visual Sensation Decision Tree Visualization
69
Table 28. Result of 3-Point Visual Sensation Model
Class True Positive Rate
Overall 80.4075%
-1 77%
0 70.9%
1 80.4%

Table 29. Result of 7-Point Visual Sensation Model
Class True Positive Rate
Overall 73.3542%
-2 83.3%
-1 71.9%
0 66.4%
1 83.1%
2 69.4%
3 63.1%

5.3 Thermal Sensation Prediction Model by Gender
Because gender could be a significant factor in the model development, thermal sensation
prediction models by female and male were also developed using Weka, to analyze the difference
between genders. The prediction models for both female and male are developed in 3-point scale
and 7-point scale to compare their accuracy rate.

For female dataset, there are 341 attributes, including 7 females’ bio-signals data, while male data
set has 297 instances. As shown in Figure 31 and Figure 33, the most significant input to predict
3-point scale thermal sensation for both female and male is heart rate. However, as Figure 32 and
Figure 34 show, the most significant input to predict 7-point scale thermal sensation for both
female and male is skin temperature. As Table 30 shows, the overall accuracy rate for female
model is 77.7125%, which is higher than male model, with 75.0842% accuracy rate. And for each
class in models, the true positive rate for all classes in female model are higher than them in male
70
model, while class 1 has the highest true positive rate in both models. And for male model, class -
1 has a low true positive rate, which might because there is less dataset for this class than other
classes. For the 7-point scale thermal sensation prediction models, there are 5-point scale actually
in the output, because the data for -3 and 3 is rare in both female and male dataset. And the overall
accuracy rate of 7-point scale prediction model for both female and male are less than 3-point scale
prediction models. For both female and male 7-point models result, the class 1 has the highest true
positive rate.

71

Figure 31. 3-Point Thermal Sensation Decision Tree Visualization (Female)
72

Figure 32. 7-Point Thermal Sensation Decision Tree Visualization (Female)
73

Figure 33. 3-Point Thermal Sensation Decision Tree Visualization (Male)
74

Figure 34. 7-Point Thermal Sensation Decision Tree Visualization (Male)
75
Table 30. Result of 3-Point Thermal Sensation Model (Female and Male)
Class True Positive Rate (Female) True Positive Rate (Male)
Overall 77.7126% 75.0842%
-1 70% 45.4%
0 79.5% 70%
1 82.5% 82.4%

Table 31. Result of 7-Point Thermal Sensation Model (Female and Male)
Class True Positive Rate (Female) True Positive Rate (Male)
Overall 74.7801% 71.7172%
-2 67.3%
-1 70.9% 63.6%
0 73.9% 76.4%
1 82.7% 77.3%
2 69.7% 71.2%

5.4 Visual Sensation Prediction Model by Gender
Same as the thermal prediction model, the visual prediction models also used 3-point scale visual
sensation as the output attributes. As shown in Figure 35 and Figure 37, heart rate is the most
significant input attribute in both female and male decision tree models. While as shown in Figure
36, the most significant input to predict 7-point scale visual sensation for female is skin
temperature, and as shown in Figure 38, the most significant input to predict 7-point scale visual
sensation for male is EDA data. As the result of the prediction model shows, the overall accuracy
rate for male model is 86.8586%, which is higher than 77.4194% for female model. Therefore, the
male morel is more accurate in predicting the visual sensation. The true positive rates for all classes
in male model are higher than female model, where class 1 has the highest true positive rate in
both models, representing the models predict class 1 the most correctly. For 7-point visual
sensation prediction model, the overall accuracy rate is higher for female, where the true positive
rate for all classes except class 2 is higher compared to male model.
76

Figure 35. 3-Point Visual Sensation Decision Tree Visualization (Female)
77

Figure 36. 7-Point Visual Sensation Decision Tree Visualization (Female)
78

Figure 37. 3-Point Visual Sensation Decision Tree Visualization (Male)
79

Figure 38. 7-Point Visual Sensation Decision Tree Visualization (Male)
80
Table 32. Result of 3-Point Visual Sensation Model (Female and Male)
Class True Positive Rate (Female) True Positive Rate (Male)
Overall 77.4194% 85.8586%
-1 77.3% 81.8%
0 67.3% 72.7%
1 81.2% 90.9%

Table 33. Result of 7-Point Visual Sensation Model (Female and Male)
Class True Positive Rate (Female) True Positive Rate (Male)
Overall 74.7801% 69.697%
-2 83.3%
-1 74.2% 72.7%
0 72.75 61.8%
1 81.8% 72.7%
2 62.3% 72.7%
3 81.8% 65.9%

5.5 Summary
Table 34. Summary Result of Each Prediction Model
Prediction Model Accuracy Rate (3-Point) Accuracy Rate (7-Point)
Thermal Sensation (Compiled) 78.9969% 71.6301%
Visual Sensation (Compiled) 80.4075% 73.3542%
Thermal Sensation (Female) 77.7126% 74.7801%
Thermal Sensation (Male) 75.0842% 71.7172%
Visual Sensation (Female) 77.4194% 74.7801%
Visual Sensation (Male) 85.8586% 69.697%

This section focuses on the development of the prediction model for aggregated dataset and for
different genders’ dataset using Decision Tree machine learning algorithm. For aggregated thermal
sensation and visual sensation prediction models, 3-point scale and 7-point scale prediction models
were developed to compare the accuracy rate and the most significant input attribute. By
comparing the overall accuracy rate, the 3-point prediction models have higher accuracy rate than
7-point models. And the most significant input attribute for 3-point thermal sensation prediction
model is gender, while that for 7-point thermal prediction model is skin temperature. For visual
81
sensation models, the most significant input attribute is EDA data for 3-point model, while which
is gender for 7-point model.

The prediction models by different genders were also developed and analyzed. Both visual
sensation and thermal sensation prediction model for female and male also included both 3-point
scale and 3-point scale models. For both female and male prediction models, 3-point scale models
have a higher accuracy rate than 7-point scale models. For thermal sensation model, female
prediction model has a higher accuracy rate than male prediction model, while for visual sensation
model, male prediction model has a higher accuracy rate. Heart rate is the most significant input
attribute for all 3-point prediction models.

82
Chapter 6. Conclusions and Future Work
6.1 Conclusions
The objective of this research is to explore the correlation between the changing of smart window
states and human’s bio-signals, in order to develop the prediction model based on the human’s
bio-signals. For this research, the prediction model of the window states could only be explored
by the prediction model of human’s thermal and visual sensation, then based on those predictions,
the window states could be modified in the future with further research.

First, the author conducted the experiments with 15 human subjects in the basement chamber in
the Watt Hall of at the University of Southern California. In the experiments, Hobo sensors were
used to collect the indoor environmental quality data in the chamber, and 2 wearable sensors
Garmin Vivosmart3 and Empatica Embrace were used to measure the human bio-signals including
heart rate, stress level, skin temperature, and EDA data. And the thermal and visual sensation was
collected by questionnaires finished by the human subjects during the experiments in 7 scales,
from cold to hot for thermal sensation, and from too dark to too bright for visual sensation.

After collecting the experiment data, different data analysis methods were used to analyze the
relationship between each human bio-signal and the thermal sensation, visual sensation, and the
window state. Anova test analysis was used to visualize the patterns and analyze the significance
of each bio-signal in different levels of thermal sensation, visual sensation, and window state,
although the results show different human subjects might have different patterns in the bio-signals,
and different bio-signals also have different patterns related to one same output, but almost all the
results have a P-value less than 0.05, representing there are significant differences of each bio-
83
signal among different levels of outputs. Besides, the Correlation analysis was also conducted to
analyze the correlation between each bio-signal and visual and thermal sensation. Almost all the
results showed that there was significant correlation between each bio-signal and the visual and
thermal sensation with a high correlation coefficient value.

Because the correlation between each bio-signal and the thermal and visual sensation is very
significant, it means the prediction models of visual sensation and thermal sensation could be
developed based on the human gender data and bio-signals as the input attributes. Both the
prediction models were developed using Weka software, which helps researcher build up models
based on different machine learning algorithms. In this research, all prediction models were
developed using Decision Tree algorithm. And comparing the results of different models, the 7-
point thermal sensation and visual sensation models are the most accurate models. And for the 7-
point thermal sensation prediction model, the visualization tree showed that skin temperature was
the most significant attribute, while in the 7-point visual sensation prediction model, the gender
information is the most significant attribute.

6.2 Limitations
The data collected by conducting the experiments for the human subjects have provided some
results of the exploration of the relationship between the use of EC Glass and the thermal sensation
of the participants, which influences on their physiological signals like heart rate, skin temperature,
stress level and EDA. However, there are still some limitations of this research, like the limitations
of the number of human subjects and the diversity of them, and a second test of the human subjects
is needed to verify the results drown by the experiments so far.
84
6.2.1 Short-term Problems
In this study, 15 human subjects participated in the experiment exploring the relationship between
the EC glass and their thermal sensation and visual sensation, while this number was limited for a
very significant result of an experiment, if more time could be provided for the tests, then the more
human subjects could participate in this research, which could draw a more accurate result for the
research. Besides the number of the subjects is limited, the diversity of them is also a limitation.
Almost all the participants were from the USC School of Architecture with similar ages in a certain
range, and from limited diversity of nationality, where the majority of them are from China.

What’s more, because of the limit of time for the research, only one round of experiment was
conducted for data analysis and prediction models development; therefore, the accuracy of the
prediction model can’t be verified. If there is more time for the research, a second-round
experiment could be conducted to verify the result of the data analysis and the accuracy of the
prediction model. In addition, a control model of the window states could then be developed based
on the prediction of the thermal sensation and visual sensation.

6.2.2 Long-term Problems
In this research, 4 types of human bio-signals were collected by 2 wearable sensors, where some
of the wearable sensors might have the data loss problems caused by the human subjects, including
they wore the watches too loosely, or they developed a wrong sign in information. Therefore, if
the research could be conducted in the future, there might be more advanced technologies which
could measure the human bio-signals more accurately.

85
What’s more, all the experiments conducted for this research were taken place in the basement of
Watt Hall in the University of Southern California, where the heat lamps were used to mimic the
natural sunlight. Therefore, if the experiments were conducted in the real office with the
electrochromic glass as the glazing area, the data analysis and the prediction model would be more
accurate.

6.3 Future Work
Because of the time limited for this research, the control model of the EC window states wasn’t
developed. However, the prediction models for thermal sensation and visual sensation were
developed, and now the window states could be changed based on only one condition, which is
when the thermal sensation is low, then the window could be changed to a lighter tinted state, or
if the visual sensation is low, then the window could be changed to a lighter tinted state to let more
solar radiation and sunlight into the office area. But if there are conflicts between thermal sensation
and visual sensation, for example, the visual sensation is high, while the thermal sensation is low,
then how the window should be changed will be uncertain without an accurate algorithm.
Therefore, in the future, the algorithm to analyze different conditions of thermal sensation and
visual sensation to decide how to change the window states should be developed.

And in the future, more human subjects with different ages, or nationalities, could be included to
the experiments to contribute to a more accurate result for this research. And after the development
of the prediction models and control model of the window states, a second and third round of
experiments should be conducted to verify the result of the models.
86
References

[1] C. Liu, Y. Wu, Y. Zhu, D. Li, L. Ma, Experimental investigation of optical and thermal
performance of a PCM-glazed unit for building applications, Energy and Buildings. 158
(2018) 794–800. https://doi.org/10.1016/J.ENBUILD.2017.10.069.
[2] How do electrochromic (smart glass) windows work? - Explain that Stuff, (n.d.).
https://www.explainthatstuff.com/electrochromic-windows.html (accessed September 15,
2021).
[3] What is Electrochromic Smart Glass? | Smart Glass World, (n.d.).
https://www.smartglassworld.net/what-is-electrochromic-glass (accessed December 12,
2021).
[4] M. Burdis, N. Sbar, Electrochromic Windows: Process and Fabrication Improvements for
Lower Total Costs , 2009. https://www.osti.gov/servlets/purl/993475 (accessed December
13, 2021).
[5] 5 Key Benefits of Installing Smart Glass Windows in Your Home Office, (n.d.).
https://www.archiscene.net/residential/key-benefits-of-installing-smart-glass-windows/
(accessed December 13, 2021).
[6] LEED v4 Green Building Standard, (n.d.).
[7] Y. Cheng, ELECTROCHROMIC GLASS THERMAL COMFORT Proposals for
Improving Existing Workflow, 2017.
[8] A.I. Palmero-Marrero, A.C. Oliveira, Effect of louver shading devices on building energy
requirements, Applied Energy. 87 (2010) 2040–2049.
https://doi.org/10.1016/J.APENERGY.2009.11.020.
[9] R. Baetens, B.P. Jelle, A. Gustavsen, Properties, requirements and possibilities of smart
windows for dynamic daylight and solar energy control in buildings: A state-of-the-art
review, Solar Energy Materials and Solar Cells. 94 (2010) 87–105.
https://doi.org/10.1016/J.SOLMAT.2009.08.021.
[10] D. Cao, C. Xu, W. Lu, C. Qin, S. Cheng, Sunlight-Driven Photo-Thermochromic Smart
Windows, Solar RRL. 2 (2018) 1700219. https://doi.org/10.1002/SOLR.201700219.
[11] What Is Electrochromic Glass? | SageGlass, (n.d.).
https://www.sageglass.com/en/article/what-electrochromic-glass (accessed September 15,
2021).
[12] Performance assessment of sageglass® electrochromic coatings and control scenarios,
(n.d.). www.paladinoandco.com (accessed December 14, 2021).
[13] J.-H. Choi, Human-Environment Interaction: Potential Use of Pupil Size for Visual
Environmental Controls Human-Building Integration Research View project, (2013).
www.ashrae.org (accessed September 15, 2021).
[14] J.H. Choi, D. Yeom, Study of data-driven thermal sensation prediction model as a
function of local body skin temperatures in a built environment, Building and
Environment. 121 (2017) 130–147. https://doi.org/10.1016/J.BUILDENV.2017.05.004.
[15] Y.K. Axelrod, M.N. Diringer, Temperature Management in Acute Neurologic Disorders,
Neurologic Clinics. 26 (2008) 585–603. https://doi.org/10.1016/J.NCL.2008.02.005.
[16] D.P. Wyon, I. Andersen, G.R. Lundqvist, Chapter 16 the Effects of Moderate Heat Stress
on Mental Performance, Studies in Environmental Science. 10 (1981) 251–267.
https://doi.org/10.1016/S0166-1116(08)71093-8.
87
[17] ANSI/ASHRAE Addendum a to ANSI/ASHRAE Standard 55-2020, (2021).
www.ashrae.org (accessed September 15, 2021).
[18] P.O. Fanger, Thermal comfort: Analysis and applications in environmental engineering,
Applied Ergonomics. 3 (1972) 181–181. https://doi.org/info:doi/.
[19] T. Chaudhuri, Y.C. Soh, H. Li, L. Xie, Machine learning based prediction of thermal
comfort in buildings of equatorial Singapore, 2017 IEEE International Conference on
Smart Grid and Smart Cities, ICSGSC 2017. (2017) 72–77.
https://doi.org/10.1109/ICSGSC.2017.8038552.
[20] B. Brik, M. Esseghir, L. Merghem-Boulahia, H. Snoussi, An IoT-based deep learning
approach to analyse indoor thermal comfort of disabled people, Building and
Environment. 203 (2021) 108056. https://doi.org/10.1016/J.BUILDENV.2021.108056.
[21] What is electrodermal activity? – Empatica Support, (n.d.).
https://support.empatica.com/hc/en-us/articles/115000198283-What-is-electrodermal-
activity- (accessed February 7, 2022).
[22] J.H. Choi, R. Zhu, Investigation of the potential use of human eye pupil sizes to estimate
visual sensations in the workplace environment, Building and Environment. 88 (2015) 73–
81. https://doi.org/10.1016/J.BUILDENV.2014.11.025.
[23] W. van Bommel, G. van den Beld, Lighting for work: a review of visual and biological
effects:, Http://Dx.Doi.Org/10.1191/1365782804li122oa. 36 (2016) 255–269.
https://doi.org/10.1191/1365782804LI122OA.
[24] J. Kim, R. de Dear, C. Cândido, H. Zhang, E. Arens, Gender differences in office
occupant perception of indoor environmental quality (IEQ), Building and Environment. 70
(2013) 245–256. https://doi.org/10.1016/J.BUILDENV.2013.08.022.
[25] B. Cassin, M.L. Rubin, Dictionary of eye terminology, (2006) 288.
https://books.google.com/books/about/Dictionary_of_Eye_Terminology.html?id=3XKpP
wAACAAJ (accessed September 16, 2021).
[26] M.N. Assimakopoulos, A. Tsangrassoulis, M. Santamouris, G. Guarracino, Comparing the
energy performance of an electrochromic window under various control strategies,
Building and Environment. 42 (2007) 2829–2834.
https://doi.org/10.1016/j.buildenv.2006.04.004.
[27] C. Pearson, S. King, Passive Environmental Control for Small Cultural Institutions in
Australia, Http://Dx.Doi.Org/10.1080/00048623.2000.10755117. 31 (2013) 69–78.
https://doi.org/10.1080/00048623.2000.10755117.
[28] Passive Thermal Control System - Satellite Wiki, (n.d.).
https://www.aero.iitb.ac.in/satelliteWiki/index.php/Passive_Thermal_Control_System
(accessed September 16, 2021).
[29] K. Kuzuhara, H. Nishi, Accurate indoor condition control based on PMV prediction in
BEMS environments, IECON Proceedings (Industrial Electronics Conference). (2013)
8142–8147. https://doi.org/10.1109/IECON.2013.6700495.
[30] S. Purdon, B. Kusy, R. Jurdak, G. Challen, Model-free HVAC control using occupant
feedback, Proceedings - Conference on Local Computer Networks, LCN. (2013) 84–92.
https://doi.org/10.1109/LCNW.2013.6758502.
[31] A. Peruffo, A. Pandharipande, D. Caicedo, L. Schenato, Lighting control with distributed
wireless sensing and actuation for daylight and occupancy adaptation, Energy and
Buildings. 97 (2015) 13–20. https://doi.org/10.1016/J.ENBUILD.2015.03.049.
88
[32] Electric Lighting Controls | WBDG - Whole Building Design Guide, (n.d.).
https://www.wbdg.org/resources/electric-lighting-controls (accessed September 16, 2021).
[33] B.R. Park, J. Hong, E.J. Choi, Y.J. Choi, C. Lee, J.W. Moon, Improvement in energy
performance of building envelope incorporating electrochromic windows (ECWs),
Energies (Basel). 12 (2019). https://doi.org/10.3390/en12061181.
[34] J.M. Dussault, L. Gosselin, Office buildings with electrochromic windows: A sensitivity
analysis of design parameters on energy performance, and thermal and visual comfort,
Energy and Buildings. 153 (2017) 50–62.
https://doi.org/10.1016/J.ENBUILD.2017.07.046.
[35] Y. Ajaji, P. André, Thermal Comfort and Visual Comfort in an Office Building Equipped
with Smart Electrochromic Glazing: An Experimental Study, Energy Procedia. 78 (2015)
2464–2469. https://doi.org/10.1016/J.EGYPRO.2015.11.230.
[36] J.A.J. Stolwijk, J.B. Pierce, A MATHEMATICAL MODEL OF PHYSIOLOGICAL
TEMPERATURE REGULATION IN MAN 4. Title and Subtitle A MATKElrlATICALp
MODEL OF PHYSIOLQGICAL "F’ERATUHE IGGULATION I N b@. N 7. Author(sJ
13. Type of Report and Period Covered 14. Sponsoring Agency Code, (n.d.).
[37] P. Antoniadou, A.M. Papadopoulos, Occupants’ thermal comfort: State of the art and the
prospects of personalized assessment in office buildings, Energy and Buildings. 153
(2017) 136–149. https://doi.org/10.1016/J.ENBUILD.2017.08.001.
[38] T. Chaudhuri, D. Zhai, Y.C. Soh, H. Li, L. Xie, Random forest based thermal comfort
prediction from gender-specific physiological parameters using wearable sensing
technology, Energy and Buildings. 166 (2018) 391–406.
https://doi.org/10.1016/J.ENBUILD.2018.02.035.
[39] C. Dai, H. Zhang, E. Arens, Z. Lian, Machine learning approaches to predict thermal
demands using skin temperatures: Steady-state conditions, Building and Environment. 114
(2017) 1–10. https://doi.org/10.1016/J.BUILDENV.2016.12.005.
[40] J.H. Choi, D. Yeom, Development of the data-driven thermal satisfaction prediction
model as a function of human physiological responses in a built environment, Building
and Environment. 150 (2019) 206–218. https://doi.org/10.1016/j.buildenv.2019.01.007.
[41] A. Ghahramani, G. Castro, B. Becerik-Gerber, X. Yu, Infrared thermography of human
face for monitoring thermoregulation performance and estimating personal thermal
comfort, (2016). https://doi.org/10.1016/j.buildenv.2016.09.005.
[42] SageGlass, SCS-023.1 SOLO Controller, Install and User Guide, n.d.
www.sageglass.com.
[43] Phillips 416743 Heat Lamp 250-Watt BR40 Clear Flood Light Bulb 4 Pack - -
Amazon.com, (n.d.). https://www.amazon.com/Phillips-416743-250-Watt-Clear-
Flood/dp/B00OJ31X5U/ref=sr_1_26?keywords=heat+lamp&qid=1639595056&s=industri
al&sr=1-26 (accessed December 14, 2021).
[44] Amazon.com : Simple Deluxe 250 Watt Clamp Lamp Light with 10.5 Inch Aluminum
Reflector and Bulb Guard (no Bulb Included) E26 Socket 6 Feet Cord, Silver : Pet
Supplies, (n.d.). https://www.amazon.com/Simple-Deluxe-Watt-Clamp-
Light/dp/B098N4C9KN/ref=sr_1_9?keywords=clamp+lamp+light+250w&qid=16365069
47&sr=8-9 (accessed December 14, 2021).
[45] Amazon.com: Simple Houseware Heavy Duty Clothing Garment Rack, Chrome : Home &
Kitchen, (n.d.).
https://www.amazon.com/dp/B01MSAVQJO/ref=redir_mobile_desktop?_encoding=UTF
89
8&aaxitk=d2ef844b1f464eeb15c183117e7d485f&hsa_cr_id=1038199186&pd_rd_plhdr=t
&pd_rd_r=8461e100-7248-4a0b-9fc4-
7a159f2d0e62&pd_rd_w=G5dsq&pd_rd_wg=x0KR9&ref_=sbx_be_s_sparkle_lsi4d_asin
_1_img&th=1 (accessed December 14, 2021).
[46] FLIR K65 NFPA Compliant 320x240 TIC | Teledyne FLIR, (n.d.).
https://www.flir.com/products/k65/ (accessed December 14, 2021).

Abstract (if available)

Abstract Nowadays, smart windows are more and more popular because of their high performance in improving indoor environment and decreasing heating and cooling loads. Among different types of smart windows, electrochromic windows are most common in office buildings because they are easily installed and controlled. However, the control of electrochromic windows might cost more energy than other smart windows like thermochromic windows and cause less thermal and visual comfort; therefore, if it is possible to use bio-signals to control different states of the windows, this situation will be benefitted a lot in the future.
The project aims to investigate the impacts of electrochromic windows on the users’ bio-signals, to explore the potential of using them as a control strategy of the windows and decreasing energy consumption and improving thermal comfort could be accomplished. This project collected indoor environmental quality data and humans’ bio-signal data using different sensors. After collecting data, the data was analyzed using machine learning and used to develop a control model of the smart windows.
As a result, by using data analysis skills, human’s bio-signals were affected by the utilizing the smart windows, which was shown when the EC windows are in the different states, occupants would have different levels in heart rate and skin temperature. In the future, the research on the utilization of bio-sensing control of electrochromic windows will be conducted further.

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Multi-occupancy environmental control for smart connected communities

PDF

Developing a data-driven model of overall thermal sensation based on the use of human physiological information in a built environment

PDF

Indoor air quality for human health in residential buildings

PDF

Enhancing thermal comfort: air temperature control based on human facial skin temperature

PDF

Exploration for the prediction of thermal comfort & sensation with application of building HVAC automation

PDF

Indoor environmental quality and comfort: IEQ adaptation and human physiological responses in commercial buildings

PDF

Exploring participatory sensing and the Internet of things to evaluate temperature setpoint policy and potential of overheating/overcooling of spaces on the USC campus

PDF

IEQ, sleep quality, and IoT: meta-analysis on improving IEQ and sleep quality using IoT

PDF

Building energy performance estimation approach: facade visual information-driven benchmark performance model

PDF

Developing environmental controls using a data-driven approach for enhancing environmental comfort and energy performance

PDF

Dynamic shading and glazing technologies: improve energy, visual, and thermal performance

PDF

Mitigating thermal bridging in ventilated rainscreen envelope construction: Methods to reduce thermal transfer in net-zero envelope optimization

PDF

Data visualization in VR/AR: static data analysis in buildings

PDF

Human-building integration: Investigation of human eye pupil sizes as a measure of visual sensation in the workstation environment

PDF

Enhancing thermal comfort: data-driven approach to control air temperature based on facial skin temperature

PDF

Energy performance of different building forms: HEED simulations of equivalent massing models in diverse building surface aspect ratios and locations in the US

PDF

Human–building integration: machine learning–based and occupant eye pupil size–driven lighting control as an applicable visual comfort tool in the office environment

PDF

Impacts of building performance on occupants' work productivity: a post occupancy evaluation study

PDF

Human-building integration based on biometric signal analysis: investigation of the relationships between human comfort and IEQ in a multi-occupancy condition

PDF

Natural ventilation in tall buildings: development of design guidelines based on climate and building height

Asset Metadata

Creator Wang, Zihan (author)

Core Title Using bio-signals with smart windows

School School of Architecture

Degree Master of Building Science

Degree Program Building Science

Degree Conferral Date 2022-08

Publication Date 05/11/2022

Defense Date 04/20/2022

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag electrochromic windows,human bio-signals,machine learning.,OAI-PMH Harvest,thermal and visual sensations

Format application/pdf (imt)

Language English

Contributor Electronically uploaded by the author (provenance)

Advisor Choi, Joon-Ho (committee chair), Ambite, Jose-Luis (committee member), Noble, Douglas E. (committee member), Park, Dong Yoon (committee member)

Creator Email adonis8233@gmail.com,zwang052@usc.edu

Permanent Link (DOI) https://doi.org/10.25549/usctheses-oUC111313296

Unique identifier UC111313296

Document Type Thesis

Format application/pdf (imt)

Rights Wang, Zihan

Type texts

Source 20220517-usctheses-batch-942 (batch), University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the author, as the original true and official version of the work, but does not grant the reader permission to use the work if the desired use is covered by copyright. It is the author, as rights holder, who must provide use permission if such use is covered by copyright. The original signature page accompanying the original submission of the work to the USC Libraries is retained by the USC Libraries and a copy of it may be obtained by authorized requesters contacting the repository e-mail address given.

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus MC 2810, 3434 South Grand Avenue, 2nd Floor, Los Angeles, California 90089-2810, USA

Repository Email cisadmin@lib.usc.edu