2018 10 Dopudi PDF

SCUOLA DI ARCHITETTURA URBANISTICA INGEGNERIA
DELLE COSTRUZIONI – BUILDING ENGINEERING
NATURALLY VENTILATED DOUBLE SKIN

FAÇADE: CFD AND SIMPLIFIED MODEL
FOR PARAMETRIC ENERGY SIMULATION
Relatore: Dr. Alessandro Dama

Correlatore: Prof. Federico Piscaglia
Tesi di laurea di:

Matteo Dopudi - 878123
Anno accademico 2017/2018

Acknowledgements
It took nearly one year and many hours of simulations to complete this work, and I
would like to thank all the people that took part in it and that made it possible.
In the first place, I would like to thank professor Andrea Mainini for giving me his
advice during the early phases of the project.
My gratitude goes also to Olena Kalyanova Larsen from the Aalborg University for her
feedbacks and suggestions regarding the experimental data.
Special thanks to Diego Angeli, from the University of Modena, whose precious
suggestions helped to fix the critical problems of the CFD model giving me a huge
boost in the advancement of the work.
Another special mention goes to Riccardo Mereu and Luigi Urbinati that let me
access to the computational cluster of the Energy Department of the Politecnico di
Milano whose power was crucial to perform the CFD analyses.
But most importantly I would like to thank Sonia for giving me the true motivation to
tackle such hard work and for inspiring me to overcome my limits and be a better
man every day.
I
Abstract
The Double Skin Façade (DSF) is a well-known technology that is useful to design
energy efficient buildings with all-glazed façades. However, evaluating the energy
and thermal performances of double skin constructions is still a difficult task,
especially in the case of naturally ventilated façades. Two main approaches can be
used to tackle this kind of problem, CFD analyses and simplified models. This work
aims at studying both approaches through the development and the validation of a
three-dimensional CFD model and of a simplified model based on a one-dimensional
temperature profile. The validation is performed by comparing the results of the two
models with the measurements of the mass flow rate and air temperatures obtained
from a full-scale test facility. The predictions of the two models are in good
agreement concerning the airflow rate and the vertical temperature profile
suggesting that the the hypotheses introduced in the simplified model are
reasonable. Furthermore, the simplified model shows good accuracy in the
prediction of the heat flux towards the inside, which is the parameter that mostly
matters when performing the energy balance of a building. The CFD model takes
approximately 30 hours to perform the analysis of a single timestep, while the
simplified model can simulate the behaviour of the DSF over a year in about 60
seconds. For this reason, the simplified model is a suitable tool for preliminary
parametric analyses. The last part of this work is an example of how the simplified
model could be used to perform a parametric study on a DSF in different climates
and for different orientations. To perform such analysis, the simplified model is
upgraded to feature an integrated shading system. The DSF is compared with a
double and a triple glazing system equipped with internal roller blinds. The output of
the analysis is that the DSF performs respectively 60% and 30% better in reducing
the total entering heat flux than the other two technologies during the cooling
season, while these values decrease respectively to 50% and 10% during the middle
season.
II
Sinossi
Le facciate a doppia pelle sono una tecnologia di involucro utile nella progettazione
di edifici trasparenti a basso consumo energetico. La valutazione termica ed
energetica di questa soluzione risulta essere tuttavia difficile da effettuare,
specialmente nei casi in cui la facciata è ventilata in modo naturale. Esistono due
approcci che possono essere impiegati per tentare di risolvere il problema: analisi
fluidodinamiche (CFD) o modelli semplificati. L’obiettivo di questo elaborato è quello
di sviluppare e validare un modello CFD tridimensionale e un modello semplificato
basato su un profilo lineare di temperatura. La validazione avviene comparando i
risultati dei modelli con le misure ottenute da una facciata in scala 1:1. I risultati dei
due modelli sono in linea tra loro rispetto al calcolo della portata d’aria e del profilo
verticale di temperatura nella cavità, suggerendo quindi che le ipotesi alla base del
modello semplificato sono ragionevoli. Inoltre, il modello semplificato ha dimostrato
di essere in grado di predire con un buon grado di accuratezza il valore del flusso
scambiato verso l’interno, che è il parametro più importante quando si considera il
bilancio energetico di un edificio. Il grande vantaggio del modello semplificato è la
sua abilità di condurre un’analisi annuale in 60 secondi contro le 30 ore che il
modello CFD impiega per calcolare un singolo istante di tempo. Questa caratteristica
rende il modello semplificato adatto per analisi parametriche e predimensionamenti.
L’ultima parte del lavoro è perciò dedicata ad offrire un esempio di un’analisi
parametrica tra una doppia pelle con schermatura integrata, un doppio vetro e un
triplo vetro con tenda interna in quattro differenti località e per i quattro
orientamenti cardinali. Il risultato dell’analisi mostra che nei casi considerati la
doppia pelle si comporta rispettivamente meglio per il 60% e per il 30%, rispetto alle
altre due soluzioni confrontate, nel ridurre il flusso termico entrante durante la
stagione di raffrescamento, mentre questi valori scendono a 50% e 10% durante la
stagione intermedia.
III
Table of contents
NATURALLY VENTILATED DOUBLE SKIN FAÇADE: CFD AND

SIMPLIFIED MODEL FOR PARAMETRIC ENERGY SIMULATION ................I
ACKNOWLEDGEMENTS .................................................................................I
ABSTRACT .....................................................................................................II
SINOSSI .........................................................................................................III
TABLE OF CONTENTS ................................................................................. IV
LIST OF FIGURES........................................................................................ VII
LIST OF TABLES .......................................................................................... XI
LIST OF CHARTS........................................................................................ XIII
CHAPTER 1 INTRODUCTION ........................................................................1
1.1 RESEARCH QUESTION ........................................................................................... 1

1.2 MOTIVATION ........................................................................................................ 4
1.3 OBJECTIVES ........................................................................................................ 4
CHAPTER 2 DOUBLE SKIN FAÇADES: STATE OF THE ART ......................5
2.1 TECHNOLOGY ...................................................................................................... 5

2.1.1 Influence of façade design............................................................................ 8
2.1.2 Influence of building design ........................................................................ 10
2.1.3 Influence of site conditions ......................................................................... 11
2.2 ENERGY PERFORMANCE MODELLING OF DSF ....................................................... 13
2.2.1 Simplified models ....................................................................................... 14
2.2.2 Zonal approach .......................................................................................... 22
IV
2.2.3 Computational Fluid Dynamics models (CFD) ............................................. 23
2.3 ENERGY MODELLING OF THE CASE-STUDY ........................................................... 28
CHAPTER 3 METHODOLOGY AND CASE STUDY .....................................33
3.1 METHODOLOGY ................................................................................................. 33

3.2 EXPERIMENTAL TEST FACILITY ............................................................................. 35
3.3 EXPERIMENTAL SETUP ........................................................................................ 38
3.3.1 Temperature measurements ...................................................................... 39
3.3.2 Incident solar radiation measurements ....................................................... 41
3.3.3 Airflow measurements ................................................................................ 41
CHAPTER 4 CFD MODEL ............................................................................45
4.1 FLUID DYNAMIC MODEL ...................................................................................... 45

4.2 MESH GENERATION ............................................................................................ 47
4.3 BOUNDARY CONDITIONS ..................................................................................... 51
4.3.1 Temperature boundary conditions .............................................................. 52
4.3.2 Velocity boundary conditions ..................................................................... 53
4.3.3 Pressure boundary conditions .................................................................... 53
4.4 SOLVER SELECTION AND SETUP........................................................................... 53
CHAPTER 5 SIMPLIFIED MODEL ................................................................57
5.1 MODEL DESCRIPTION ......................................................................................... 57

5.1.1 Algorithm workflow .................................................................................... 59
5.1.2 Solar-optical model .................................................................................... 61
5.1.3 Fluid dynamic model .................................................................................. 63
5.1.4 Thermal model ........................................................................................... 64
CHAPTER 6 MODEL VALIDATION...............................................................71
6.1 CFD MODEL: INVESTIGATION VS. MEASUREMENTS ................................................ 72

6.1.1 Case A ....................................................................................................... 73
6.1.2 Case B....................................................................................................... 79
6.1.3 Case C ...................................................................................................... 85
6.1.4 Case D ...................................................................................................... 91
V
6.2 SIMPLIFIED MODEL VS. EXPERIMENTAL DATA ........................................................ 96
6.2.1 Solar radiation and flow rate ....................................................................... 97
6.2.2 Temperatures and heat flux ........................................................................ 99
6.3 SIMPLIFIED MODEL, CFD AND MEASUREMENTS COMPARISON ............................. 103
6.3.1 Airflow rate ............................................................................................... 103
6.3.2 Vertical temperature profile....................................................................... 104
6.3.3 Convective heat transfer .......................................................................... 105
CHAPTER 7 PARAMETRIC ANALYSIS ......................................................109
7.1 EVALUATED SOLUTIONS .................................................................................... 110

7.2 CHOSEN CLIMATES .......................................................................................... 113
7.2.1 Milan ........................................................................................................ 114
7.2.2 Athens ..................................................................................................... 115
7.2.3 Paris ........................................................................................................ 116
7.2.4 Berlin ....................................................................................................... 117
7.3 RESULTS ......................................................................................................... 118
7.3.1 Daily simulations ...................................................................................... 120
7.3.2 Yearly simulations .................................................................................... 123
CHAPTER 8 CONCLUSIONS .....................................................................133
REFERENCES .............................................................................................137
VI
List of figures
Figure 2.1 DSF cavity of the Generali headquarter by Zaha Hadid, Milan, Italy. ... 6
Figure 2.2 DSF cavity of the Intesa-Sanpaolo headquarter by Renzo Piano, Turin,
Italy. ................................................................................................. 6
Figure 2.3 DSF classification (Barbosa et al. 2014): (a) Box Window, (b) Shaft-Box,
(c) Corridor, (d) Multi-Storey. ............................................................ 8
Figure 2.4 Volume averaged air temperature in the DSF (modelled vs.
experimental) (Larsen et al. 2008b). ................................................ 29
Figure 2.5 Hour averaged temperature of the glass surface facing the internal room
(modelled vs. experimental) (Larsen et al. 2008b). .......................... 30
Figure 2.6 Hour averaged mass flow rate in the air cavity, measured with the
velocity profile method (Larsen et al. 2008b)................................... 30
Figure 3.1 The Cube. Photo of Southern façade (left) and a photo of Northern
façade (right). ................................................................................. 35
Figure 3.2 Plan view of The Cube test facility (Larsen et al. 2008a). ................... 36
Figure 3.3 Plan of the experiment room and DSF (left). Section 1-1 of the
experiment room and DSF (right) (Larsen et al. 2014). .................... 36
Figure 3.4 Double skin façade section dimensions. ........................................... 37
Figure 3.5 Naming convention for the glass modules of the external (left) and of
the internal (right) layer. .................................................................. 37
Figure 3.6 Definition of openings. ...................................................................... 38
Figure 3.7 Measurements of air temperature in the DSF cavity. Positioning of the
sensors, view from outside (left). Photo of the experimental setup
(right), thermocouples are shielded with silver coated ventilated tube
(Larsen et al. 2008a). ..................................................................... 39
Figure 3.8 Measurement of outlet air temperature (Larsen et al. 2008a). ........... 40
VII
Figure 3.9 Shielding strategy from incident direct solar radiation. Left: Plan of DSF
cavity. Right: Section of DSF cavity. ............................................... 40
Figure 3.10 Positioning of pyranometers in the experimental set-up (Larsen et al.
2008a). .......................................................................................... 41
Figure 3.11 Positioning of anemometers in the DSF cavity (left). Photo of the
anemometers in the DSF cavity (right) (Larsen et al. 2008a). ........... 42
Figure 3.12 Positioning of the perforated tube for the release of the tracer gas at
the bottom of the DSF cavity (left). Positioning of the air intakes for
samples of tracer gas polluted air, at the top of the DSF cavity (right).
...................................................................................................... 43
Figure 4.1 CAD model of the DSF with measures. ............................................ 48
Figure 4.2 STL file of the DSF. .......................................................................... 48
Figure 4.3 Iterative workflow of the snappyHexMesh algorithm. ......................... 49
Figure 4.4 Background mesh. .......................................................................... 50
Figure 4.5 Castellated mesh. ............................................................................ 50
Figure 4.6 Snapped mesh. ............................................................................... 50
Figure 4.7 Layered mesh. ................................................................................. 50
Figure 5.1 Thermal network of the ventilated channel. The blue rectangles
correspond to the glass positions. ϕ1, ϕ2 and ϕ3 correspond to the
solar radiation absorbed respectively by the glass of the outer skin,
the external glass and the internal glass of the inner skin (double
glazed unit). Glass panes are represented in blue. .......................... 58
Figure 5.2 Iterative flowchart of the algorithm. green=inputs, orange=outputs. .. 60
Figure 5.3 Graphic representation of the temperature profile inside the cavity with
the change in height....................................................................... 68
Figure 6.1 Case A: horizontal section of velocity distribution along the z axis (Uz).
...................................................................................................... 74
Figure 6.2 Case A: vertical sections profiles of air velocity (magnitude values) with
velocity vectors (in white). ............................................................... 75
Figure 6.3 Case A: horizontal section of temperature distribution. ..................... 77
Figure 6.4 Case A: vertical section profile of temperature. ................................. 78
VIII
Figure 6.5 Case B: horizontal section of velocity distribution along the z axis (Uz).
...................................................................................................... 80
Figure 6.6 Case B: vertical sections profiles of air velocity (magnitude values) with
Figure 6.7 Case B: horizontal section of temperature distribution. ..................... 83
Figure 6.8 Case B: vertical section profile of temperature.................................. 84
Figure 6.9 Case C: horizontal section of velocity distribution along the z axis (Uz).
...................................................................................................... 86
Figure 6.10 Case C: vertical sections profiles of air velocity (magnitude values) with
Figure 6.11 Case C: horizontal section of temperature distribution. ................... 89
Figure 6.12 Case C: vertical section profile of temperature. .............................. 90
Figure 6.13 Case D: horizontal section of velocity distribution along the z axis (Uz).
...................................................................................................... 92
Figure 6.14 Case D: vertical sections profiles of air velocity (magnitude values) with
Figure 6.15 Case D: horizontal section of temperature distribution. ................... 95
Figure 6.16 Case D: vertical section profile of temperature. .............................. 96
Figure 6.17 Exchanged heat flux between the fluid and the surface for all the four
cases. Left column represents the heat exchanged by the internal
surface, while the right column the one exchanged by the external
one. ............................................................................................. 106
Figure 6.18 Visual representation of the averaging of the quantities necessary to
the calculation of hcv. .................................................................... 107
Figure 7.1 Airflow in the DSF cavity with integrated shading. The colouring of the
image and the white vectors are referred to the air speed along the z
axis. ............................................................................................. 111
correspond to the glass positions. ϕ1, ϕ2 , ϕ3 and ϕ4 correspond to
the solar radiation absorbed respectively by the glass of the outer skin,
the external glass of the inner skin, the internal glass and the shading.
Glass panes are represented in blue. ........................................... 111
IX
Figure 7.3 Selected solar shading. .................................................................. 112
X
List of tables
Table 2.1 Characteristics of the software employed for the validation. .............. 29
Table 3.1 Geographical coordinates of the site. ................................................ 35
Table 3.2 U-values and emissivity properties of the glazing system. .................. 37
Table 3.3 Spectral properties of the glazing system. ......................................... 38
Table 3.4 Discharge coefficients and dimension of the openings of the DSF. .... 38
Table 4.1 Consistency checking of the employed mesh. ................................... 51
Table 4.2 Cases chosen for the CFD simulations. ............................................. 52
Table 5.1 Characterisation of the glazing system for direct radiation. ................ 62
Table 5.2 Characterisation of the glazing system for diffuse radiation. ............... 62
Table 5.3. Heat transfer coefficients correlations. .............................................. 67
Table 6.1. Heat exchange towards the interior. Model versus experimental results.
Daily integrated values calculated separately for hours with flux
inwards (+) and hours with flux outwards (-). ................................ 101
Table 6.2 Airflow rate comparison between simplified model, CFD and
measurements for all the four cases. Asterisk values on the tracer gas
column mean that there is a false detection of this method due to the
downstream flow. ........................................................................ 103
Table 7.1 Shading system characteristics. ...................................................... 112
Table 7.2 Optical properties of DSF with shading system for direct radiation .. 112
Table 7.3 Optical properties of DSF with shading system for diffuse radiation . 112
Table 7.4 Properties of window system with triple glazing and internal roller blind.
.................................................................................................... 113
Table 7.5 Properties of window system with double glazing and internal roller blind.
.................................................................................................... 113
Table 7.6 Summary of DSF performances against double and triple glazing during
the middle season. ....................................................................... 124
XI
Table 7.7 Summary of DSF performances against double and triple glazing during
the cooling period. ....................................................................... 124
XII
List of charts
Chart 5.1 Glazing system characterisation for different incidence angles of direct
solar radiation. ............................................................................... 62
Chart 6.1 Case A: simulated air velocity on the z axis (Uz) for each horizontal
section. White circles represent the value of the magnitude of U in the
position of the anemometers. Black circles represent the experimental
measurements in the position of the anemometers. ....................... 73
Chart 6.2 Case A: Temperature profile of horizontal sections. Black circles
represent the experimental measure in the position of the
thermocouple. White circles highlight the values of the CFD data in the
position of the thermocouple. ......................................................... 76
Chart 6.3 Case B: simulated air velocity on the z axis (Uz) for each horizontal
Chart 6.4 Case B: Temperature profile of horizontal sections. Black circles
Chart 6.5 Case C: simulated air velocity on the z axis (Uz) for each horizontal
Chart 6.6 Case C: Temperature profile of horizontal sections. Black circles
XIII
Chart 6.7 Case D: simulated air velocity on the z axis (Uz) for each horizontal
Chart 6.8 Case D: Temperature profile of horizontal sections. Black circles
Chart 6.9 Experimental values (black dots) vs. model (black line) of solar radiation
impinging on the vertical façade. Red line represents the radiation
entering the building. ...................................................................... 97
Chart 6.10 Differential pressures and volume flow rates – simplified model vs.
experimental data. ......................................................................... 98
Chart 6.11 Glass surface temperatures. Black dots represent experimental values
while red line are the modelled ones............................................... 99
Chart 6.12 Outlet (red lines) and inlet (dotted line) air temperatures – model vs.
measures. .................................................................................... 100
Chart 6.13 Internal surface temperature of the inner skin, model (red line) vs.
measurements (white circles). ...................................................... 100
Chart 6.14 Glazing surface temperature and air temperature. Model versus
measurements for a single day (10th October). On the left the air and
the surface temperature of the glazing facing the ventilated cavity,
while on the right the internal surface temperature of the inner skin.
.................................................................................................... 102
Chart 6.15 Vertical profiles of bulk temperature for all the four cases computed by
CFD, simplified model and experimental data. The vertical coordinate
is normalized by the cavity height (H). ........................................... 104
Chart 6.16 Profiles of the heat flux along the external and internal glass for all the
simulated cases, as obtained by CFD, compared with the value
computed by the simplified model. ............................................... 108
XIV
Chart 7.1 Milan – whole year temperature profile. Red line represents the monthly
average, the grey line represents hourly values, while the yellow
rectangle is the comfort temperature zone. .................................. 114
Chart 7.2 Milan – whole year wind speed profile. Blue line represents monthly
average, while grey line represents hourly values. ......................... 114
Chart 7.3 Athens – whole year temperature profile. Red line represents the monthly
Chart 7.4 Athens – whole year wind speed profile. Blue line represents monthly
Chart 7.5 Paris – whole year temperature profile. Red line represents the monthly
average, the grey line represents hourly values while the yellow
Chart 7.6 Paris – whole year wind speed profile. Blue line represents monthly
average while grey line represents hourly values. .......................... 116
Chart 7.7 Berlin – whole year temperature profile. Red line represents the monthly
Chart 7.8 Berlin – whole year wind speed profile. Blue line represents monthly
Chart 7.9 Monthly average external temperatures in the four cities. ................ 118
Chart 7.10 Milan – Parametric analysis of DSF for four different orientations (solar
control activates at 200 W/m2) – daily focus on 21st March. .......... 120
control activates at 100 W/m2)– daily focus on 21stJune. .............. 121
Chart 7.12 Milan – Parametric analysis of DSF for four different orientations. .. 125
Chart 7.13 Athens – Parametric analysis of DSF for four different orientations. 127
Chart 7.14 Athens - Yearly profile of solar radiation hitting the south façade. .. 128
Chart 7.15 Paris – Parametric analysis of DSF for four different orientations. ... 129
Chart 7.16 Berlin – Parametric analysis of DSF for four different orientations. .. 131
XV
CHAPTER 1
INTRODUCTION
Climate change is one of the main challenges that humanity has to face in this century. One
of its leading causes is the energy consumption of buildings. This work concerns the study
of double skin façades, a well-known technology, which might offer interesting adaptive
features and improvements concerning comfort and energy efficiency in buildings with
large transparent façades. The evaluation of these benefits is still an open issue, due to
reappearing difficulties in the prediction of DSF thermal and energy performance as an
integrated part of building envelope through well-established building simulation tools. The
objectives of this thesis can be summed up as the development and validation of a CFD and
a simplified model for the energy modelling of naturally ventilated double skin façades
comparing them against an experimental dataset obtained from a full-scale test facility. The
advantage of a simplified model is that it can easily simulate the behaviour of the DSF over
the year with a good level of accuracy in the results and so, it is suitable for parametric
design.
1.1 Research question
Buildings are responsible for half of the energy consumption worldwide (Pomponi
2016). Energy waste, and the consequent production of CO2 emissions, is one of the
leading causes of the climate change effect that humanity is facing nowadays. The
projections state that climate change is going to lead to an increase of the average
world temperatures by 1.5°C, in the most favourable scenario, or by 5°C in the worst
1
1
Introduction
scenario within 2100. The consequences of this phenomenon are, on the one hand,
the thermal discomfort during the hot seasons and, more importantly, on the other
hand, the consequences that it has on people’s health. In fact, despite the different
outcomes outlined in the studies, scientists agree that there is an optimal
temperature at which mortality rates are the lowest (Akbari et al. 2016). Researchers
estimate that, the death rate increases in an interval between 1% and 4% for every
degree higher than the optimal condition (World Health Organization 2011). Hence, a
realistic increase of 3°C would bring up to +12% in the mortality rate. This number
would be fatherly increased by the higher probability that catastrophic events have
to occur under climate change effects.
The category of all-glazed buildings represents one of the main reasons for energy
waste among buildings. All-glazed façades have a lot of benefits, such as the ability
to capture the sunlight, a clear view towards the outside and a superior aesthetic
appeal for the customers. The drawback of this technology is that glass is not an
insulating material at all, and so, this kind of buildings have great thermal losses in
winter and high internal gains in summer. This aspect increases both the heating
loads and the thermal loads, making all glazed façades extremely energy inefficient.
During the past century, energy problems have been put aside in favour of the
aesthetic desire for transparency. However, the rising concern for environmental
protection has raised the awareness for energy efficiency among designers and
customers. Sustainable buildings are now seen as a status symbol and a marketing
boost for the companies that buy them. An example is the Bank of America tower,
built in New York City in 2010, which has been the first skyscraper in the world to
obtain the LEED platinum certificate.
Passive energy efficient solutions are the ones that allow maximising the drop on
the consumptions because they are not activated by mechanical or electrical
systems, but by the structural and architectural design of the building (Nicol et al.
2012). One of the passive solutions that have been proposed during the past years
to reduce the consumptions of all glazed buildings is the double skin façade (DSF).
The concept of a ventilated thermal buffer zone between two transparent layers was
introduced at the beginning of the XX century, but it has become popular starting
from the 1990s (Pomponi et al. 2013). In order to evaluate the effectiveness of this
solution, it is necessary to calculate its energy balance with the building. However,
2
1.1
Research question
predicting the thermal and energy performances of DSF remains an uncertain task,
even when using well-established commercial simulation tools or recently
developed models (Manz et al. 2005, Larsen 2008b). In fact, in contrast to traditional
building envelope technologies, naturally ventilated double skin façades
performances are very dynamic due to their high sensitivity to solar radiation
exposure (large glazing areas), outdoor temperature (high U-value of the outer
glazing) and ventilation conditions. The fluctuating boundary conditions and the
non-steadiness of the problem explain the vast complexity of this kind of problem.
Researchers have proposed many ways to categorize, DSF. The easiest one is by
separating them depending on whether they use natural or mechanical ventilation.
Mechanically ventilated façades are more straightforward to simulate because, in
these cases, the inlet velocity of air is known since it is a design choice. On the other
hand, naturally ventilated façades introduce another layer of complexity because
the models that aim to simulate them have to deal with the uncertainty of the outside
wind conditions in order to calculate the inlet velocity of the air inside the ventilated
channel.
At the present moment, computational fluid dynamic (CFD) methods are the ones
that can predict with the highest level of accuracy the conditions inside the cavity
of a DSF. However, this technique is costly from the computational point of view, and
so it is impossible to use it to simulate a whole year behaviour.
This work aims to present an extension of the simplified model presented by Dama
et al. (2017), that can predict the thermal behaviour of a double skin façade under
natural ventilation conditions with a reduced margin of error compared to the CFD
analysis and the experimental measurements. More specifically, the thermal model
discussed in this thesis is based on an equivalent thermal network and considers
both convection and long wave radiation. The convection inside the channel is
modelled through a simplified integral approach employing average bulk
temperatures and superficial heat transfer coefficient correlations. The thermal
model is coupled with a fluid-dynamic model, based on a pressure loop which takes
into account both buoyancy and wind differential pressure at the openings. An
optical module provides the parameters required to describe the interaction of short
wave radiation with façade components, i.e. the direct and diffuse solar and light
transmittance of the façade and the absorption of each glass. In the original thermal
3
1
Introduction
model (Angeli et al. 2015, Dama et al. 2017), the thermal and fluid dynamic ‘core’ of
the ventilated cavity was validated having the channel surface temperatures as
boundary conditions. In this work, a suitable thermal network has been implemented
to simulate the DSF having wind, solar radiation and environmental temperatures as
inputs. In order to validate the results of the simplified model, its results have been
confronted with an experimental dataset and with the outputs of a three dimensional
CFD model.
The advantage of having a simplified model is that it can easily simulate the yearly
behaviour of the DSF maintaining good accuracy in the results. For this reason, it is
useful for parametric analyses and optimisation of the geometry in every weather
condition.
1.2 Motivation
The interest in the topic of this research comes from the activities carried out during
the courses at the Politecnico di Milano that encourage the design of energy
efficient buildings. Mitigating climate change effects is one of the main challenges
of this century, and it is our duty as engineers to develop solutions, such as the
double skin façade, that can help humanity to achieve this goal.
This work intends to add another piece of knowledge to the understanding of the
DSF technology and to provide a helpful, yet not perfect, tool to support engineers in
their choices when designing double skin façades.
1.3 Objectives
The targets set by this study are:
• The development and validation of a 3D CFD model over an experimental
dataset.
• The development and validation of a simplified model, based on an equivalent
thermal network, over the results of the CFD model and the measurements of
the experimental dataset.
• The use of such developed model to perform a parametric analysis of a
hypothetic façade in different climates and for different orientations to draw
some design considerations that are helpful from the engineering point of
view.
4
CHAPTER 2
DOUBLE SKIN FAÇADES: STATE OF THE ART
The first part of the chapter analyses the development of the double skin façade (DSF)
technology and the main parameters that affect its performance. The overall behaviour of
this solution is driven by the choices made during the design of the façade and of the whole
building as well as from the climatic conditions of the site. The main issue during the DSF
design is the assessment of its energy performance since this is far from being a trivial
exercise, especially in the case of naturally ventilated skins. For this reason, the second part
of the chapter focuses on the analysis of the state of the art in DSF energy performance
analysis. Particular emphasis will be given to lumped parameters and CFD models of
naturally ventilated façades because they are of higher interest for the purpose of this
thesis.
2.1 Technology
In 2016, Pomponi et al. proposed the following definition of double skin façade:
“A glazed double skin façade is a hybrid system made of an external glazed skin and the
actual building façade, which constitutes the inner skin. The two layers are separated by
an air cavity which has fixed or controllable inlets and outlets and may or may not
incorporate fixed or controllable shading devices”.
The idea of the Double skin façade technology, also called Double-envelope façade,
was introduced at the beginning of the previous century.
5
2
Double skin façades: state of the art
Figure 2.1 DSF cavity of the Generali Figure 2.2 DSF cavity of the Intesa-
headquarter by Zaha Hadid, Milan, Italy. Sanpaolo headquarter by Renzo Piano,
Turin, Italy.
Many years had to pass before the golden period for Double skin façades started in
the 90s (Barbosa et al. 2014). Since this technique was conceived for cold climates,
it firstly spread to Europe, North America and Japan and only recently in the hot and
humid areas of the planet such as China (Zhou 2010). The main reasons behind the
adoption of this technology are the mass diffusion of computers and other
electronic appliances that generate a lot of internal heat and the introduction of the
idea of sustainable design (Gratia et al. 2007). Since the real estate market was
asking for all-glazed façades, the single-skin technology could not cope anymore
with the energy conservation standards. For this reason, designers started to look
for solar control passive strategies (Zhou 2010). Among them, DSF proved to be one
of the most attractive and so it started to be employed, especially for tall glazed
buildings where it protected the indoor environment while enhancing the daylighting
and thermal comfort (Elarga 2015).
There are many reasons behind the choice of this technology. First of all, it allows for
total transparency of the envelope, and this is usually the main architectural reason
for choosing a double skin façade. Moreover, from the client’s point of view,
6
2.1
Technology
transparency may represent a crucial factor in tall buildings since it raises the value
of the building. Beyond their appealing aesthetics, double skin façades can provide
many other advantages from the energetic point of view. In winter, the greenhouse
effect entraps the sun rays creating a thermal buffer zone between inside and
outside. During summertime, higher temperatures increase the buoyancy of the air,
generating more ventilation inside the cavity. In this way, the heat is discharged
more effectively from the cavity leading to more stable temperatures inside the
building. Furthermore, the integration of shading devices inside the cavity
guarantees solar control and enhances thermal comfort while decreasing the
cooling loads.
The outer layer protects the inside from weather conditions, pollution and external
noise (Zhou 2010). It also protects shading devices and integrated photovoltaic
systems (BIPV) from the weathering agents increasing their performances and their
durability over time.
As said before, a DSF mainly relies on the ventilation inside its cavity to thermally
insulate the building. Ventilation can be achieved through natural, mechanical or
mixed means (Ghaffarianhoseini 2016). In the first case, the pressure difference,
which determines the airflow, is given by two main factors: wind pressure and
thermal buoyancy. Thermal buoyancy depends on the air density: when air enters
the DSF (either from the bottom openings or the windows on the other side of the
building (Liddament 1996)) it warms up and loses density; this effect tends to move
the airflow upstream. On the other hand, the wind pressure direction is variable. In
fact, during night time, when thermal buoyancy is almost null, there can be cases in
which due to the wind pressure the airflow goes downstream. This effect does not
impact on the functioning of the technology because the critical factor is the actual
presence of a constant airflow inside the cavity and not its direction. Mechanical
ventilation can be employed to guarantee a constant airflow through the use of fans
that are usually positioned at the ceiling of the cavity.
Depending on the ventilation mode and the air flow path, Oesterle et al. (2001)
classified the DSF into four categories:
a) Box window: room by room ventilation. The inner layer can be opened to
allow for natural ventilation;
7
2
b) Shaft-box: ventilation occurs on vertical separated channels to enhance

the stack effect. This type of structure well protects against external
noise;
c) Corridor: each floor has its own ventilation. Careful design is necessary
not to have sound transmission from room to room;
d) Multi-storey: there is only one cavity. Highest potential concerning
energy savings, but poor performance when dealing with occupant
privacy and fire protection.
Figure 2.3 DSF classification (Barbosa et al. 2014): (a) Box Window, (b) Shaft-Box, (c)
Corridor, (d) Multi-Storey.
The following paragraphs explain the influence that façade design, building design
and site location have on the performances of double skin façades.
2.1.1 Influence of façade design
Cavity depth
The depth of the ventilated cavity usually has dimensions going from 10cm up to 2m.
The choice of this value depends upon the design strategies such as placing an
integrated solar shading system or adding a catwalk for maintenance purposes
(Pappas et al. 2008). Torres et al. (2007), Rahmani et al. (2012) and Radhi et al. (2013)
studied the pros and cons of using a broader or narrower air cavity. Having a smaller
cavity enhances the ventilation due to the Venturi effect while, on the other hand, a
bigger cavity leads to a slower airflow in the cavity but leaves the space to insert
shadings and other utilities inside the air channel. Radhi et al. (2013) recommended
a depth between 0.7 and 1.2 meters in order to balance the two extremes.
8
2.1
Technology
Shading device
The possibility of integrating shading devices inside the cavity is particularly useful
in tall buildings where high wind loads prevent from being able to install the shadings
externally. There are many types of shading devices suitable for integration that are
available nowadays: roller shades, louvred blinds and dynamic shadings are some of
the macro categories in which they can be gathered together. By reducing the direct
solar radiation impinging on the internal environment, the shading can reduce the
cooling needs and provide passive thermal comfort during the hot seasons.
Gratia et al. (2007) and Jiru et al. (2008) studied the positioning of the shadings
inside the cavity. These authors concluded that an optimised solution could save up
to 14.1% of cooling energy. The ideal location would be in the middle of the cavity, in
this way not only the energy savings are at their maximum, but the ventilation rate
is also enhanced (Ji et al. 2007). On the other hand, it should be avoided to place the
shading system close to the internal side otherwise multiple radiative reflections
increase and so does the temperature of the inner glass, causing a growth in the heat
flux going inside of the building.
Furthermore, the studies by Gratia et al. (2007) and by Haase et al. (2009)
investigated the impact that the colouring of the blind has from the energetic point
of view. Darker blinds contribute to raise the temperature in the cavity temperature
by 11°C compared with light coloured blinds.
Finally, it has been attempted to optimise the rotation of the lamellas in non-dynamic
systems. Ji et al. (2007) observed that horizontal inclination obstructs the air flow
rate, and thus it has to be avoided. On the other hand, for 80° inclination, the
buoyancy effect increased by almost 35%.
To sum up, the optimised shading devices should be light coloured and placed in the
middle of the cavity rotated by an angle that guarantees both adequate solar
protection and stack effect in the cavity.
Outer skin properties
The properties of reflectance, absorbance and transmittance of the external skin can
impact on the thermal load, entering a DSF by almost an order of magnitude (Pérez-
Grande et al. 2005). The fraction of solar radiation that passes through the outer skin
9
2
is absorbed by the inner layer that emits long wave radiation in all the directions
creating multiple reflections inside the cavity and warming it up (Gratia et al. 2007).
There are two optimal configurations depending upon the decision of minimising the
solar gains or maximising the ventilation rate in the cavity. In the first case, low
transmittance and high absorbance glazing on the outer skin and low-e internal
glazing are recommended (Guardo et al. 2009). In the second case, high-absorbing
inner material combined with an outer layer with the same absorbing and
transmitting power of 0.4 maximises the ventilation flow rate (Pérez-Grande et al.
2005).
Cavity openings
The dimension of the top and bottom openings should be carefully evaluated since
it can significantly change the air temperature and velocity inside the cavity. Results
in the study by Torres et al. (2007) highlighted a strong non-linear connection
between the dimension of the openings and the cooling loads inside the building.
During the same year, Gratia et al. (2007) showed how large openings could decrease
the cooling loads by 20% compared to narrower openings. Therefore, it is highly
recommended to perform an optimisation of this variable considering the outdoor
conditions and on the geometry of the DSF.
2.1.2 Influence of building design
The following parameters affect the behaviour of the DSF, but the decisions made
upon them can be related to other aspects of the design. However, in the framework
of an integrated building design approach their impact on the envelope performance
should be taken into account.
Inner skin properties
Windows provide beneficial effects such as daylight comfort and view towards the
outside. However, undesired effects can occur due to glare phenomena or excessive
heat gains due to the low thermal resistance of the glass. Studies on the optimal
ratio between opaque and transparent surfaces were made by Chou et al. (2009),
who highlighted that a window to wall ratio (WWR) of 0.3 is capable of reducing the
10
2.1
Technology
solar radiation gain component of the thermal transfer value (ETTV) by 45% whereas,
if the WWR tends to 0.9, it negatively impacts on the energy consumptions. In
another paper by Haase et al. (2009), decreasing the WWR from 0.9 to 0.32 led to a
26.4% increase in the annual energy savings. A lower WWR always has positive
effects from the thermal point of view. However, in a holistic design approach, other
variables should be considered, such as daylight autonomy and air changes. A large
WWR increments the airflow inside the building and thus reduces the heat gains
inside the rooms. Therefore, a balanced window to wall ratio should be adopted by
the designers.
Height of the cavity
The altitude difference between the inlet and outlet openings profoundly impacts on
the buoyancy effect that generates inside the cavity (Mingotti 2011). A taller cavity
creates a stronger stack effect extracting more heat (Oesterle 2001). Ding et al.
(2005) studied this effect and proposed a “thermal storage wall” that consisted into
an extension of the height of the envelope that strengthened the chimney effect
guaranteeing stable ventilation rates even with naturally ventilated cavities. The
authors recommended that the extension is at least two-floor high but, as said at the
beginning of the paragraph, this kind of decisions is not only driven by the envelope
needs but by many other considerations that the designer has to make.
2.1.3 Influence of site conditions
Two main weather parameters can influence the performance of double skin
façades, especially the naturally ventilated ones. They are solar radiation and wind
speed. The former is one of the factors that mostly drive the generation of the
airstream since it affects the variation of density of the air. Kim et al. (2009)
underlined how in the most extreme cases the temperature inside the cavity could
be 20°C higher than the outside one due to solar radiation which is dependant on the
orientation of the façade. East and West orientations should be avoided since they
provoke overheating respectively during the morning and the evening (Gratia et al.
2007, Hamza 2008). From the simulations of Haase et al. (2009), it turned out that
11
2
the best orientations in the northern hemisphere, regardless of surrounding

buildings and obstructions, are South, South-East and South-West.
The wind speed is the second parameter that affects the energy performance by
influencing the airflow inside the cavity (Stec et al. 2005). The ventilation rate inside
the cavity is directly proportional to the outside by a factor of 0.25. If no wind is
present, the differences between cavity air and external air can reach up to 50°C
while, on the other hand, it is sufficient a 4m/s wind to drop the difference at about
30°C (Gratia et al. 2007).
The analysis made up to know highlighted how DSF can have practical positive
environmental and economic effects when adequately designed. These benefits are
represented by energy consumption reduction, sound insulation, architectural
appeal and low long-term cost reduction (Larsen et al. 2008a, Shameri 2011).
However, they also come with some problems and uncertainties such as costs of
implementation (Poirazis 2006), structural load increase, fire propagation due to the
increased ventilation and reduction of useful office space.
These problems can be overcome only by performing a high-quality design phase
supported by thermal analyses that are as much precise as possible so that they can
prove the effectiveness of the chosen design.
However, predicting the behaviour of a DSF is not a straightforward task due to the
many unknowns like the optical modelling and the heat transfer convective
coefficients (Gratia et al. 2007). The solution of the problem becomes even more
tangled when natural ventilation is involved since the turbulence effects that are
generated inside the cavity are very hard to predict (Barbosa et al. 2014).
The scientific literature in this field of research has flourished since many authors
have tackled these problems in order to reach an adequate level of energy
performance prediction of DSF that is suitable for engineering purposes. The
following paragraph analyses the most important findings that have been obtained
during the past years.
12
2.2
Energy performance modelling of DSF
2.2 Energy performance modelling of DSF
The scientific literature regarding DSF is mainly oriented towards their energy
performance under different boundary conditions (De Gracia et al. 2013). The
calculation of the energy reduction provided by this technology is a crucial step
towards its optimal design and optimisation. Over the decades, many types of
approaches have been proposed and validated. The first filter between all these
approaches can be applied to the type of ventilation, natural or mechanical, that the
model is able to predict. A more consistent classification was proposed by De Gracia
et al. (2013):
a) Analytical and lumped parameters modelling
b) Computational Fluid Dynamics (CFD)
c) Airflow network analysis + Building energy simulation
d) Control volume approach
e) Zonal approach
f) Non-dimensional analysis
The following paragraphs aim to give an overview of these kinds of approaches. The
main difference between them is the way in which the air channel is modelled. For
the purpose of this work, the models will be divided into three categories from the
most simplified to the most complex:
1) Simplified models (lumped parameters models, airflow network models,
control volume models, non-dimensional analysis)
2) Zonal models
3) CFD models
The simplified models schematize the air cavity considering it as a single thermal
network or the sum of big control volumes solved through energy balances. The
zonal model is a solution halfway between a CFD simulation and a control volume
approach while the computational fluid dynamic approach is the one that nowadays
gives the most accurate results but it is also the one that takes the longest time to
run. Higher care will be taken to the lumped parameters and to the CFD models as
well as for the ones that tackled the problem of natural ventilation because these
topics are of higher interest for the purpose of the thesis. For all the other kinds of
approaches, only the most recent and significant papers will be discussed in detail.
13
2
2.2.1 Simplified models
Analytical and lumped parameters models
The need for rapid and robust simulation tools in the DSF modelling is the ultimate
achievement in this field of research (Dama 2017). Lumped parameter models have
been developed in order to reach the goal of rapid simulations. This kind of models
is usually based on the solution of a thermal network that assumes constant
temperatures on the surfaces of the DSF. This characteristic allows for high-speed
calculations and hence, they are very suitable especially:
• in the early design phase to make decisions about the overall strategies
• for simulations that cover long time spans that would not be completed in a
useful time using other approaches
The drawback of this kind of methods is that they are characterised by a lower
degree of accuracy of the outputs.
In 1991, Tarazi proposed a model of a Trombe wall, whose functioning principle is the
same as a DSF, basing on a previously developed method (Duffie et al. 1980). The
model took into account the wall and room properties, the hourly weather data and
the thermal inertia of the walls of the room. It was a mono-dimensional model and
sought to solve the problem of outputting the outlet air temperature at the end of
each hour as well as the heat delivered to the room through empirical correlations.
The Trombe wall was treated as solar plate collector, and the main output of the
analysis was that “the most important factors in the system are the cover transmissivity
and the room thermal resistance in addition to the wall thickness” (Tarazi 1991). This
model does not provide any calculation of the airflow rate or the air velocity in the
cavity, and moreover, it was not validated against any experimental data. Hence,
there is no way to quantify the degree of accuracy of the results.
The next step forward was made by von Grabe (2002), in his model he was one of the
first to tackle the problem of a naturally ventilated façade. He made some
simplifications to the problem. The most important one is that the fluid dynamic part
considered only buoyancy as the driving factor for the air movement. The analytical
solution is based on the resolution of a system of equations composed by the energy
transport and Bernoulli differential equations. This means that von Grabe tried to
14
2.2
apply the same type of calculation used for mechanically ventilated façades to a
naturally ventilated one. The author confronted the results provided by the algorithm
with the data obtained from a self-made experimental campaign. The experimental
setup was composed of four anemometers placed in the middle of the cavity to
measure the airflow rate and by thermocouples placed at three different heights to
measure the air and surface temperatures. The results showed an error of about 5%-
10% in the temperature predictions while no airflow calculation was presented. The
author states in the conclusion that “the main problem is caused by assuming the same
flow conditions for natural as those used for mechanical ventilation”.
The following year, Ong (2003) proposed and validated a model for a solar chimney
with three glazed façades based on the resolution of a thermal network. The main
assumptions were: uniform temperatures and air inlet temperature equal to room
temperature. The author employed some empirical correlations available from the
literature, such as Swinbank (1963) for the sky temperature and McAdams (1954) for
convective heat transfer coefficient, to get all the input necessary to solve the
thermal network. The outputs of the model were then compared with the
experimental results obtained by Hirunlabh et al. (1999). The results of the algorithm
were compared with the experimental data only at noon highlighting an
underestimation of the temperature of the external wall by 5°C, a good prediction of
the outlet temperature, an overestimation of the mean air temperature in the cavity
by 3°C and an underestimation of the airflow rate by 17%. The conclusion was that
the model could predict well the behaviour of the solar chimney but still needed
some upgrades.
Park et al. (2004a, 2004b) tried to simplify the complexity of the 3D heat transfer with
turbulent airflow proposing a 2D mathematical description of a DSF. The wind
pressure necessary to calculate the natural ventilation flow was modelled using
Bernoulli’s equations while the convective heat transfer inputs were drawn from the
experimental data. The test facility was composed by a 3m naturally ventilated DSF
equipped with thermocouples, hot sphere anemometers and pyranometers to
measure the temperatures at different heights, the air velocity in the middle of the
cavity and the solar incident radiation. As said before, the peculiarity of this model
is that some of the inputs are calibrated directly from the experimental data; this is
possible because the authors wanted to create a model able to monitor the
15
2
performance of the DSF during its life cycle. The model was then validated against
the air velocity in the cavity and the temperatures of both glazing surfaces and air.
The outputs of the calibrated model were an average error of 10.44 cm/s in the
estimation of the airflow and of 1.33°C in the prediction of temperatures. The authors
conclude that calibration allows to “compensate for errors introduced by the space
averaging and other model simplifications” and allows for accurate predictions of the
behaviour of the DSF.
Angeli et al. (2015) came up with another simplified model for a naturally ventilated
façade based on the structure proposed by Angelotti et al. (2007). The thermal
network is characterised by a fictitious thermal resistance that takes into account
the heat exchange of the inlet air with the air in the cavity. Following what was found
out by Saelens et al. (2004), which is that the inlet air temperature is sensibly
different from the external one, it was implemented a preheating of the inlet air. The
natural ventilation was modelled through the use of a pressure loop scheme
employing non-dimensional discharge coefficients at the openings and non-
dimensional wind pressure coefficients evaluated accordingly with the results of
Costola et al. (2009). Moreover, the glass temperatures facing the cavity were taken
as an input from the experimental data. The experimental dataset was obtained from
Larsen (2007) and will be discussed in the next chapter. The authors also a 2D,
steady-state computational fluid dynamic model to compare the analytical model
both with the CFD and the experimental data. The results demonstrated a very high
sensitivity to the discharge coefficients but the accurate predictions of the
algorithm (0.3% of error in the heat removed by the DSF and about 2% of error in the
calculation of temperatures and airflow rate) encouraged for further analyses with
an upgraded model built on a similar approach.
Xue et al. (2015) adopted a similar method to the one previously discussed. They
validated their CFD simulation against the experimental data obtained by Zeng et al.
(2012) and then used the CFD model to validate a lumped parameters algorithm. The
fast assessment method divides the indoor heat gain in the shortwave radiation
transmission part and another part that is a function of radiation intensity, heat
transfer coefficient, and the temperature difference between indoor and outdoor.
There is only one unknown in the function, the Radiation Conversion Index (RCI),
which represents the conversion factor from the radiation intensity on the outer
16
2.2
glass to the average level of heat gain through the inner glass for a given DSF. The
authors claim that this value can be calculated once and for all with CFD since it does
not depend on atmospheric boundary conditions. The CFD bi-dimensional model
employed the finite volume method and the SIMPLE algorithm through the use of the
commercial software STAR-CCM+. The validation of this process was made by Zeng
et al. (2012), whose work will be examined in detail in paragraph 2.2.3. The CFD
calculations were made over 20 different meteorological conditions. The indoor heat
gain predicted by the simplified model was then compared with the results of the
CFD showing good consistency in the results among all the 20 cases. The average
deviation consisted of an underestimation of the results by approximately 1.91 W/m2.
In the same year, Marques da Silva et al. (2015) validated a simplified airflow model
(Marques da Silva 2004) with the data obtained from the experimental campaign
performed in the framework of IEA-ECBCS Annex 44. The experiment considered
many types of ventilation: air supply, air exhaust, inside air curtain, outside air
curtain and buffer. The measurements were performed using T-type thermocouples,
irradiation sensors, anemometers and tracer gas to assess air velocity and airflow
rate. Moreover, a 1:100 scale model of the DSF was tested in a wind tunnel to assess
wind pressure coefficients for all the wind directions. The simplified model was able
to predict the experimental data with an error of ±10%. Such level of accuracy was
considered satisfying by the authors, considering the difficulties in measuring the
airflow under natural ventilation where there is a dominant presence of turbulence
phenomena.
Oliveira Panão et al. (2016) developed and validated a lumped RC model for natural
and mechanical ventilated DSF. In this case, both resistances and capacitances were
employed in the creation of the thermal network. The thermal network is an
extension, with a higher number of thermal resistances, of the 5R1C model used in
the EN ISO 13790 standard. The algorithm was then validated comparing the
temperatures inside the cavity and inside the room with the Energy plus simulations
and the measurements performed by Mateus (2014). The results showed an excellent
trend with a mean absolute error of 1-1.1°C in the prediction of room and cavity air
temperatures compared with the experimental data. However, no validation against
the airflow in the cavity was given in the paper.
17
2
In 2017, Dama et al. presented some improvements to the model by Angeli et al.
(2015) discussed before. In this work, the authors focused mainly on the convective
problem and the mass flow rate inside the cavity. The glass surface temperatures
were taken from the experimental data and so treated as an input. For the heat
transfer coefficients inside the cavity it was adopted McAdams (1954) correlation.
The wind pressure coefficients were drawn from the database of the Tokyo
Polytechnic University, which performed wind tunnel tests on a series of buildings
with different roof shapes obtaining their Cp. As in their previous work, the simplified
model was compared with the experimental data from “The Cube” test facility and
with CFD simulations obtained by implementing the RANS and energy conservation
equations inside the OpenFOAM® toolbox. The validation of the model was made by
comparing the air outlet temperatures and volumetric air flow rate with the
experimental dataset. Despite the complexity of the flow in the gap, fluctuating
boundary conditions and frequent flow reversals the outputs are quite satisfactorily
both for the temperatures and for the airflow rate. In particular, discrepancies of
about 2-3 °C were found for the outlet air temperature and about 0.02 m3/s in the
airflow calculation.
This simplified model constitutes the basis for the one presented in this thesis which
will be discussed in Chapter 5.
Airflow network analysis + Building energy simulation models
The airflow network was defined by Hensen et al. (2002) as a network of nodes
representing the physical parts of the system and their internodal connections. A
mass balance at each node gives birth to a set of nonlinear equations that are
integrated over the time to calculate the properties of the flow (Zhou 2010). Since
this type of models gives only information about the air behaviour, they are usually
coupled with building simulation tools (Zhou 2010), such as Energy Plus or TRNSYS
to evaluate also the energetic part of the problem and to increase the quality of the
boundary conditions of both models.
Validation of the combined calculation method was made by Haase et al. (2009). In
their paper, they have developed an algorithm that couples the potential of COMIS, a
nodal airflow simulator and TRNSYS. The objective of the study was to develop a
18
2.2
model for detailed calculation of the DSF performances. In order to do so, the results
were validated against the dataset of an experimental campaign performed in Lisbon
in 2005 on a mechanically ventilated DSF with integrated shading devices. According
to the authors, a comparison of the surface temperatures of the glasses, measured
with T-type thermocouples (error of 10%), was sufficient to provide a detailed
validation of the model. Apart from the results obtained for the external glazing,
there was a good fitting with the experimental data with errors limited to 2-3°C in
most of the cases. The authors decided then to use this model to evaluate the impact
of some design choices during the design of a DSF in a warm and humid climate such
as Hong Kong.
In 2013 Joe et al. combined EnergyPlus with the airflow network AIRNET. Moreover,
since they aimed to measure natural ventilation, they calculated the wind pressure
coefficients on the DSF through CFD. The experimental measurements were
performed from February to December on a DSF in South Korea. Both temperatures
and airflow inside the cavity were measured using respectively T-type
thermocouples (5 per each floor) and multi-point anemometers (one per floor). The
mean bias error (MBE), the root-mean-squared error (RMSE) and the R2 coefficient
were employed for the validation of the model. The simulation was able to fit with
the measured data with very high accuracy, on the temperature side the R2 never
went under 0.85. The main discrepancies were observed in the heating season
during mid-day hours where the data underestimated, sometimes with errors of
10°C, the surface temperatures. On the other hand, for air speed inside the cavity,
the errors were always below 0.3 m/s.
Mateus et al. (2014) performed a study with the aim of evaluating the degree of
precision that an engineer can expect while modelling a natural or a mechanically
ventilated DSF. The test cell was located in Lisbon, Portugal and was divided between
naturally and a mechanically ventilated DSF with integrated shadings. Temperature
values were measured through the use of 22 Hobo sensors, while the airflow rate was
measured with anemometers, but only for the mechanical ventilation. The
simulations were performed using EnergyPlus 7 and its built-in airflow network
solver (AirflowNetwork). The authors noticed that the results could be affected by
the choices made regarding thermal zones. In order to follow a common engineering
approach, there were modelled only a few zones. The validation of the results was
19
2
made on the surface temperatures and showed excellent results, both in the natural
and mechanical ventilation case. The average simulation error was 1.4°C, while the
maximum daily error was kept under 3.6°C. It is also interesting to notice a
recurrence in the underestimation of the data during midday hours. The authors
concluded that an EnergyPlus simulation could give results that are satisfying from
an engineering point of view.
The following year Khalifa et al. (2015) claimed to have obtained a better result than
Mateus et al. (2014). The approach of the authors was to couple TRNSYS and CONTAM
and validate a natural DSF model with the data provided by Saelens (2002).
Comparisons were made against the air temperature in the ventilated cavity showing
a maximum error of 3°C and an average error of 0.5°C. Moreover, a comparison of the
airflow rate was made with an error of 3 m3/h. Since there was no hour by hour
measurement, the comparison was made with averaged values of the output flow
rates in summer and winter.
Anđelković et al. performed in 2015 an experimental campaign on an office building
located in Belgrade, Serbia. Measurement setup comprised six T-type
thermocouples for air temperature and three anemometers for air velocity. The
following year Anđelković et al. (2016) aimed at modelling the same office building
by creating a model in EnergyPlus with its built-in airflow network engine. Standard
statistical values, MBE, RMSE, and R2 were employed to validate both the
temperatures and air velocities in the cavity. The model was slightly underestimating
the temperatures by approximately 2-3°C, but on overall, the predictions were very
good, with R2 always above 0.93. For the airflow velocity, the values were a little bit
lower but still considered as satisfying by the authors, considering that simulations
took approximately only 10 minutes.
Control volume approach models
In this kind of approach, the channel of the double skin is divided into control
volumes that are approximately one meter tall (Zhou 2010). The energy and mass
balances are solved for each control volume in order to get the temperature and
airflow profiles along the height of the cavity.
20
2.2
In his PhD thesis, Saelens (2002) developed a control volume method (CVM) to
simulate the behaviour of both mechanically and naturally ventilated double skin
façades with integrated roller blinds. The cavity was only divided orthogonally to the
façade, and a bulk temperature represented each of the control volumes. The second
simplification was that the airflow inside the cavity was considered vertical, thus
forbidding rotational flow simulations. Moreover, inlet temperatures were not
calculated but taken as an input from the experimental data. The pressure difference
to account for the buoyancy effect of air was taken from Liddament (1996) while the
term that considered the wind pressure was computed through the Bernoulli
equation. The results of the modelling were compared with the measurements taken
from the Vliet test facility in Belgium. The experimental setup was equipped with
thermocouples to measure the surface temperatures of the DSF and both
anemometers and tracer gas for the airflow measurements. The results of
temperature simulations showed good agreement with the measured data during
night time while the simulated temperature of the outside layer during the day was
underestimated up to 4°C. The airflow presented the same trend of the temperatures
with errors of 13.4% during the night and 14.6% during the day. The author also
developed a TRNSYS model of this kind of approach (Saelens et al. 2003) and used it
to perform energy calculations on office buildings in order to suggest improvement
strategies for the designers (Saelens et al. 2008).
Faggembauu et al. (2003a, 2003b) performed another significant study regarding
the control volume approach. Their model was validated against analytical solutions,
reference situations and test facilities from all the European climates. The
comparison with the data from Costa (2000a, 2000b) showed high overestimation
(of about 10°C) of the outer glazing temperature during the hottest hours and a very
good agreement in the inner layer temperatures. The algorithm was also used in this
case to withdraw design considerations such as that placing curtains inside the
cavity instead of placing them inside the building reduces the heat gains and that
cooling loads can be fatherly decreased in summer through the use of Low-e
coatings.
21
2
Non-dimensional analysis
The non-dimensional analysis can be found in literature in the studies of Balocco et

al. (2004, 2006). This method relies on the Buckingham π-theorem that states that,
if a process can be described using a set of equations that depend on n physical
variables there is always the possibility to rewrite the original equations as a function
of dimensionless parameters πi that are obtained from the original variables. This
method aims to find correlations that depend on a-dimensional numbers so that the
problem can be scaled indefinitely using the same parameters. The author states
that the thermo-physical problem for a naturally ventilated façade can be expressed
as a function of fourteen non-dimensional variables that depend on eighteen
dimensional parameters which describe the heat flux through the walls. The authors
also presented another correlation based on 12 non-dimensional parameters to
describe the behaviour of mechanically ventilated façades. In both cases, the
thermo-physical properties are assumed to be constant except for air density.
Correlation validity was tested against the experimental measurements of heat flux
towards the inside by Moshfegh et al. (1996) and CFD simulation results.
The error on the χ2 value was 0.153 in the case of natural ventilation, whereas for
mechanical ventilation the fitting was a 0.92 value of R2.
The authors conclude that this kind of approach is suitable for the study of some
specific parameters during the design phase without using too many computational
resources.
2.2.2 Zonal approach
This way of modelling the DSF behaviour firstly appeared in Jiru and Haghighat
(2008) and is placed halfway between a CFD solution and a control volume approach.
The zonal approach seeks to join together these two methods to enhance their pros
and nullify the cons. The main idea behind this kind of computation is that the
solution of the fluid dynamic equations is made for larger cells than the ones used
for classic CFD reducing the computational time and allowing more accurate results
than for classic control volume approaches. The model was compared with an
experimental campaign performed at the Politecnico di Torino regarding a
mechanically ventilated DSF with venetian blinds. The authors also claim that their
22
2.2
model can be also used to simulate natural ventilation conditions just by changing
the correlation for the calculation of the Nusselt number. The results showed good
agreement during night time and an underestimation of the temperatures at noon
that reached 5°C in some cases.
In 2016, Wang et al. employed an improved zonal model to model a mechanically
ventilated DSF for hot summer and cold winter conditions in China. The
improvements consisted of a dynamic optical modelling of the shading devices and
an airflow network inspired by the work of Tanimoto et al. (1997) that was able to
calculate lateral cross airflow rate. The measurements were performed using T-type
thermocouples for temperature and hot-wire anemometers to measure inlet and
outlet air velocity. The algorithm was able to predict with high accuracy the
experimental data with errors always under 10%. The only discrepancies were found
at noon where the simulation was slightly underestimating the measurements by
roughly 1°C. The authors claim that their method is suitable for engineering
applications since it can simulate a whole year of performance of a DSF in
mechanical ventilation dynamically and with high accuracy.
To conclude, it is also interesting to notice that there are no traces about the use of
the zonal approach in literature to simulate naturally ventilated façades, even tough
Jiru et al. (2008) claim their model to be suitable for both the ventilation
configurations.
2.2.3 Computational Fluid Dynamics models (CFD)
The CFD approach represents an essential tool for predicting the airflow in DSF and
has been widely used during the years (De Gracia et al. 2013, Pasut et al. 2012). The
airflow modelling consists of a set of partial differential equations representing
continuity, momentum, turbulence, enthalpy, and concentration (Guohui, 1998) in
each point of the fluid dynamic mesh. These equations can be solved both by using
the Finite Difference Method (FDM) or the Finite Volume Method (FVM). Although
some researchers have developed their own CFD software (Onishi et al. 2001), the
majority of the papers use commercial CFD software such as Ansys FLUENT or Tas
Engineering.
23
2
As Hensen et al. (2002) pointed out, the CFD analysis can be only limited to short
period simulations due to the high computational load and so it is not suitable when
it comes to the calculation of the energy balance of a building over a whole year. This
kind of approach is the one that by now gives the most detailed information about
the airflow behaviour of a DSF (Fuliotto et al. 2009, De Gracia et al. 2013). However,
its implementation is not straightforward and does not ensure correct results (Chen
1997) even if they are obtained with commercial software. For this reason, it is
necessary to validate the obtained results against the experimental data to check
for their accuracy (Chen et al. 2002).
In 2004 Manz attempted a CFD analysis on a single-story DSF with natural
ventilation. His results were compared with the measurements obtained from an
experimental set-up that comprised, apart from the meteorological sensors,
shielded thermocouples located at the top, bottom and middle height of the façade.
The CFD implementation was a bi-dimensional model inside the FLOVENT
commercial software that solved the classical Reynolds-averaged Navier–Stokes
(RANS) equation using the FVM. The model employed a high-resolution rectangular
grid, a variant of the SIMPLE algorithm and a revised k-ε turbulence model that
allowed to solve computational problems close to the surfaces of the DSF. Air inlet
conditions were simulated with a piston flow as suggested by the most updated
normative of the time. The model simulated 24h on the 22nd of June and then the
results of air and surface temperatures of the glasses were compared with the
experimental data. Results showed that the difference between predicted and
measured temperatures could arrive up to 5°C in the hottest hours. The author
concluded that for natural ventilation façades the use of a piston flow to simulate
the air inlet is not sufficient to take into account the effects of recirculation that
happen in a real situation.
In 2005, Xamán et al. continued the work by Manz (2004) conducting a more detailed
study of fluid flow and heat transfer. The authors implemented four different 2D, tall
cavity, k-ε based models proposed by various authors and compared the results with
the data of Daffa’alla et al. (1991) to understand which of the models represented the
best fit to reality. The Nusselt number calculated by each model was used as the
mean for the comparison concluding that the algorithm proposed by Ince and
24
2.2
Launder (1989) was the one with the best accuracy. Moreover, since the simulations
were made considering three different aspect ratios (A=height over depth) of the
cavity, the authors observed that for A=20 the Nusselt numbers for natural and
turbulent flows were in good agreement with the results obtained applying the
correlations found in literature while, for A=80, the flow became almost one-
dimensional and so turbulence effects were faded.
During the same year, Ding et al. (2005) performed a validation campaign of a CFD
model using as experimental data collector a scaled building model with naturally
ventilated DSF. The 1:25 test model was placed in a laboratory and equipped with
panel heaters to simulate the solar radiation heating. The authors measured
temperature and pressure difference distribution as well as the airflow direction
near the openings using incense sticks. In this case, the model was tri-dimensional
and with a rectangular grid, but no other information was given about the algorithm
and the turbulence model employed. The validation of the model was made by
comparing the vertical profile of both temperature and pressure difference at the
openings. Although the results were not perfectly matching with the experimental
results (with errors of 1-2°C on temperature and less than 0.05 Pa for pressure
difference), the authors concluded that their CFD model was “appropriate to
reproduce the phenomena occurring in the experiments”.
Ji et al. (2007) evaluated the performances of a DSF in natural ventilation with blind
slats in the middle of the cavity. The test facility from which the authors obtained
the experimental data was located in an environmental chamber with an artificial
sun. The authors did not give any other detail on the experimental setup. The
commercial software Ansys CFX10 was selected to solve the 2D problem employing
a k-ω turbulence model that allowed for faster convergence of the solution. The
authors also tried to optimise the mesh using a hexagonal discretisation instead of
a rectangular one. The comparison with the experimental data was made by
comparing the surface temperatures of the glasses and of the shading. The best-
predicted values were the temperatures of the shading and of the inner glazing pane
with a maximum error of 2-3°C, while in the evaluation of temperature of the outside
glass the maximum errors reached 5°C.
25
2
Another naturally ventilated DSF with internal shading was modelled with CFD
methods by Xu et al. (2008). They coupled an optical model, an interface heat
balance model and a 2D CFD model validating them over the experimental data of
Manz (2004) that had already been discussed before. The experiment was made
under weak wind conditions; hence, the CFD model is built on windless condition.
The authors state that a good level of agreement was achieved in predicting natural
ventilation since the highest error in the temperature prediction was at 12% and the
lowest one at 2.5%.
Baldinelli (2009) developed and validated a CFD tri-dimensional model with the
purpose of evaluating the performances of a new type of DSF. The peculiarity of this
double skin system is that, in order to avoid greenhouse effect during the warm
months, it is equipped with hydraulic jacks that can open the outer skin changing
the configuration to an open one. The author modelled a 2D and a 3D version of the
configuration to check whether the conclusion by Manz (2003) that the additional
error brought by the 2D is negligible was correct even in this case. The simulations
were performed in winter and summer configuration and considered only free
convection instead of considering the wind effect. The experimental data was taken
by Chiu et al. (2001), whose setup was composed by two opaque surfaces were a
heated aluminium cable substituted the sun effect. The authors collected data about
the airflow rate inside the cavity using hot-wire anemometers. The CFD strategy
comprised the use of the k-ε model and a fine near-wall mesh. The validation was
performed comparing the air volumetric flow rate values (m3/s) showing the good
accuracy of the model since the error never overcome 4% in the case of a 3D model
and 6% in the 2D approach. The conclusion was in line with Manz (2003), and so the
2D modelling does not increase the error in any meaningful way.
In 2009, Fuliotto et al. studied a hybrid model where solar radiation effects were
evaluated separately from the airflow rate in order to reduce computational costs.
The experimental setup is described in Colombari (2002) and measured the
temperatures of a mechanically ventilated full-scale DSF with PT1000 temperature
sensors. Since the DSF was designed to be placed in a tropical climate, the
measurements were performed during summertime. The solar radiation model was
implemented in WIS that, by having the optical glass properties as inputs, calculated
26
2.2
the glazing heat sources at the different components. The CFD part of the proposed
method employed an RNG k-ε turbulence model and the SIMPLE algorithm using the
Fluent commercial software. As in the study by Baldinelli (2009), Fuliotto et al. also
investigated the differences between a 2D and 3D modelling of the DSF. The
decoupling of radiation and CFD analysis allowed to study the 3D flow with reduced
computational costs. The comparison between the measured and predicted
temperatures showed good agreement between the two with peak errors of 5°C. The
3D velocity field was very complicated. Nevertheless, the mean thermal flow was
almost bi-directional, and that was confirmed by the good accuracy obtained by the
2D models.
The primary objective of Zeng et al. (2012) was to model a naturally ventilated DSF
with venetian blinds. According to the authors, such problem had not been studied
in depth due to the small thickness and the large number of blind slats that increased
the computational load enormously. To try to optimise the modelling, they aimed to
validate the process that had already been used by Safer et al. (2005) proposing the
use of a porous layer to simulate the presence of the shading. This simplified CFD
model was then compared with a model that employed explicit blind slats and with a
full-scale series of measures. The measurements were performed using T type
thermocouples for temperature. The sensors were placed at eight different heights
and three different depths. On the other hand, the tracer gas (SF6) method was used
to measure the ventilation rate. The CFD model was implemented using the SIMPLE
algorithm and a k- ε turbulence model in the STAR-CCM+ software. Moreover, to
simulate the outdoor environment, a large area around the DSF was modelled to take
into account wind turbulence. Both the two CFD models remained under 2°C of
difference with the experimental data whereas, for the airflow rate (kg/s) the
maximum error was 22%. Moreover, the results showed that there was less
underestimation of airflow in the porous media model compared with the one where
the slats were explicitly modelled. The authors concluded that, since the results of
the porous media mesh were acceptable and given the high reduction of
computational costs that this method allows, this simplification is both acceptable
and suggested.
27
2
In order to try to reduce the computational load of the simulations, Pasut et al. (2012)
have performed a sensitivity analysis to understand which are the critical factors in
a simulation and which are the negligible ones. The authors developed and validated
a CFD model employing the SIMPLE algorithm and then compared their results with
the measurements obtained by Mei (2007). The sensitivity analysis was made
considering four variables: air properties, turbulence model, external environment
and bi or tridimensionality. Each model featuring a combination of alternatives in
these parameters was compared with the experimental data. In particular, it was
checked the accuracy in the calculation of the surface temperature of the glasses
and of the air velocity inside the naturally ventilated channel. The outputs of the
analysis were that:
• it is not worthy to consider the λ and the cp of air as functions of the
temperature since this increases a lot the computational time while not
giving an appreciable increase of accuracy.
• 2D models give almost the same results as 3D ones
• k-ε RNG turbulence models are more suitable than the k-ω SST ones
• The simulation of the outside environment is essential to increase the quality
of the simulation
Finally, it is remembered the two works by Xue et al. (2015) and by Dama et al. (2017),
whose CFD simulations were used to validate a simplified model and, for this reason,
they have already been discussed in paragraph 2.2.1.
2.3 Energy modelling of the case-study
This last paragraph is dedicated reports the results of a validation procedure

performed in 2008 (Larsen et al. 2008b) to evaluate the accuracy of four building
simulation software in predicting the energy behaviour of the case-study that will be
used in this thesis.
Table 2.1 reports the four software that were used as well as their main
characteristics and the simplifications that they introduce.
28
2.3
Energy modelling of the case-study
Table 2.1 Characteristics of the software employed for the validation.

Software BSim VA114 ESP-r TRNSYS-TUD
Wind force x x x x
Wind fluctuations - x - -
Buoyancy x x x x
Airflow model Experimental Network Network Network
Pressure-airflow relationship Not specified Power-law Orifice Power-law
Discharge coefficient 0.65 0.61 0.65/0.72 0.65
N° division of the DSF section 1 1 3 4
A set of sensitivity analyses was preformed on VA114 and ESP-r before the model
validation that led to demonstrate that the main parameters that affect the results
of the models are:
• The internal convective heat transfer coefficient
• The external long wave radiation heat transfer
These four software were used to predict the temperature of the air and of the
surfaces and the airflow rate inside the cavity. Both predictions were made in
external air curtain conditions and natural ventilation. In each case, their
estimations were compared with the experimental data.
Regarding the temperature, two results were compared: the air temperature inside
the DSF and the surface temperature of the glass facing the internal room. Figure
2.4 shows the results of the prediction of the air temperature inside the cavity. The
delta of temperature during the night is contained between 0 and 2 degrees. On the
other hand, during the day all the models, except BSim, hugely underestimate the
temperatures with errors up to 6°C.
Figure 2.4 Volume averaged air temperature in the DSF (modelled vs. experimental)
(Larsen et al. 2008b).
29
2
The other comparison was made on the surface temperature facing the internal side.
This measure is important because it allows to calculate the heat flow towards the
room. Again, the results were either underestimated (VA114) or overestimated (ESP-
r; TRNSYS-TUD), especially in the warm hours of the day. BSim did not give this
output and so it was not possible to verify if it could predict the surface temperature
correctly as it had already done for the cavity one.
Figure 2.5 Hour averaged temperature of the glass surface facing the internal room
(modelled vs. experimental) (Larsen et al. 2008b).
The measured mass flow rate in the cavity is very high. All the models calculate
different values and none of them is able to predict the real behaviour along all the
simulation period. Figure 2.6 shows the results compared with the measurements
obtained with the velocity profile method (hot-sphere anemometers), but the results
are very scattered also when compared with the tracer gas method.
Figure 2.6 Hour averaged mass flow rate in the air cavity, measured with the velocity
profile method (Larsen et al. 2008b).
30
2.3
Energy modelling of the case-study
The authors state that “the air flow rate is particularly interesting and influencing factor
for the DSF performance” and so it should be chosen as one of the validating
parameters while checking the accuracy of a DSF model. The conclusion of the
report was that “none of the models is consistent enough when compare results of
simulations with the experimental data: for every parameter considered – a different
model is closer to the experimental data”. Hence, the models were judged as too rough
and in need of further optimisation in order to be able to simulate the DSF behaviour.
31
CHAPTER 3
METHODOLOGY AND CASE STUDY
The first part of this chapter describes the methodology that is going to be used for the rest
of the work. The adopted approach includes the validation of a CFD and a simplified model
over the data obtained from an on-field measurement campaign and then a parametric
analysis using the simplified model to evaluate the impact of the design choices of a DSF in
different climates. The second part of the chapter describes in detail the characteristics of
the test facility “The Cube” on which the measurements have been taken, as well as the
experimental setup and the biases in the measurement of temperature, solar radiation and
airflow rate and wind speed inside the cavity.
3.1 Methodology
The previous chapter has highlighted some points that are not fully developed in the
scientific literature yet. First of all, while all the studies have validated their model
results on the experimental temperatures, only a few of them have considered the
validation of the airflow rate and wind speed inside the cavity. This lack is mainly due
to the fact that is not always easy to obtain accurate datasets on both temperature
and airflow profile. However, it is quite important to evaluate the accuracy in the
prediction of the air speed inside the cavity (Larsen et al. 2008b), especially when
modelling natural ventilation, in order to understand if the correlations and the
simplifications that have to be introduced in the model are correct or if they
introduce too much uncertainty in the final result.
33
3
Methodology and case study
The second observation that can be drawn from the literature review is that many
studies validated their models for short periods of time. In this work, it has been
chosen to perform the validation on a longer timespan to have many weather
conditions inside of it such as sunny and cloudy days or windy and calm days. The
evaluation of different weather conditions is helpful to evaluate the performance of
the algorithm in all the situations and to draw observations about it.
The last point is that nobody has ever modelled the window frames in 3D when
computing the CFD model. The CFD model that will be described in Chapter 4 has
three-dimensional frames to investigate whether their effect is negligible or not and
comparing it also with the simplified algorithm in which the frames are not modelled.
The method of investigation that will be used in this work is divided into three parts:
1. The validation of a three-dimensional CFD model on the experimental dataset
obtained from the DSF case-study locate in Denmark.
2. The validation of a simplified model that is based on the solution of an
equivalent thermal network. The validation of the simplified model will be
made both on the experimental data and on the validated CFD model. The
comparison with the experimental dataset is made to evaluate the accuracy
in the prediction of temperatures, airflow rate and heat transfer towards the
inside. This last parameter is particularly important because the heat that
passes through the envelope is the one that ultimately matters when making
the whole building energy simulation. On the other hand, comparing the
results of the simplified model with the outputs from the CFD helps to
understand the degree of correctness of the airflow rate prediction, of the
inlet and outlet temperatures and of the heat transfer coefficients.
3. The development, using the simplified model, of an example of a possible
parametric analysis performed on a DSF to compare it with other envelope
solutions located in different geographical positions to investigate the effect
of the different design choices on the heat flux passing through the skin.
The next paragraph will be dedicated to the explanation of the geometry and the
experimental setup that has been installed on the test facility in order to perform the
measurements, while Chapter 4 and Chapter 5 will be dedicated respectively to the
description of the CFD model and of the simplified model.
34
3.2
Experimental test facility
3.2 Experimental test facility
The experimental data on which the CFD model and the simplified algorithm will be
validated on comes from the experimental campaign conducted in 2006 in the frame
of IEA ECBCS ANNEX 43/SHC Task 34 “Testing and validation of Building Energy
Simulation Tools”. The experimental data on which this thesis relies were collected
between the 1st October and the 15th of the same month with a timestep of 10 minutes
(hourly measures are also available). The measurements were made at “The Cube”
test facility which is an outdoor full-scale test facility designed to be flexible in
changing between natural or mechanical flow conditions, different types of shading
devices etc. The experimental setup is located in a suburban area near the main
campus of Aalborg University, Denmark.
Table 3.1 Geographical coordinates of the site.

Time zone GMT +1hr
Longitude 9.93° E
Latitude 57.05° N
Altitude 19 m
A photo of the Southern and Northern façade of the facility is shown in Figure 3.1.
Figure 3.1 The Cube. Photo of Southern façade (left) and a photo of Northern façade
(right).
Key measurements are carried out in the DSF and in the room behind the DSF, named
as experiment room. The temperature in the experiment room is kept constant at
22°C by a cooling machine installed in the engine room and a ventilation system that
injects air at 0.2 m/s with the heating and cooling unit installed in the experiment
room Figure 3.2.
35
3
Figure 3.2 Plan view of The Cube test facility (Larsen et al. 2008a).
The airtightness of The Cube was measured to be 0.2 h-1 at 50 Pa of underpressure
and 0.35 h-1 at 50 Pa of overpressure in the experiment room.
The following image shows the exact dimensions of the DSF test case
Figure 3.3 Plan of the experiment room and DSF (left). Section 1-1 of the experiment
room and DSF (right) (Larsen et al. 2014).
The double skin façade is facing South and consists of an internal double-glazed
layer (4-Ar16-4) and an 8mm single-glazed clear glass exterior layer.
36
3.2
Experimental test facility
Figure 3.4 Double skin façade section dimensions.

The DSF is divided vertically into three sections, while horizontally there are two big
panes and two small panes which are located respectively at the top and the bottom
ventilation opening. The naming convention of the glass panes is illustrated in Figure
3.5.
Figure 3.5 Naming convention for the glass modules of the external (left) and of the
internal (right) layer.
Specifics of the material properties of all constructions in the Cube can be found in
Larsen et al. (2014), as well as spectral properties of finishing in the experimental
setup. Table 3.2 and Table 3.3 show a resume of the glass properties measured by
the EMPA Swiss Federal Laboratories for Materials Science and Technology.
Table 3.2 U-values and emissivity properties of the glazing system.

U-value [W/m2K]
Glazing Frame
External window 5.67 3.63
Internal window 1.12 3.63
Emissivity
Exterior side Interior side
External window 0.84 0.84
Internal window ext. pane 0.84 0.84
Internal window int. pane 0.037 0.84
37
3
Table 3.3 Spectral properties of the glazing system.

External window Internal window
Incident angle
τ ρ g τ ρfront ρback g
0 0.763 0.076 0.532 0.252 0.237 0.632
10 0.763 0.076 0.531 0.252 0.237 0.632
20 0.760 0.076 0.529 0.251 0.237 0.631
30 0.753 0.078 0.524 0.252 0.239 0.627
40 0.741 0.084 0.513 0.258 0.245 0.618
0.8
50 0.716 0.103 0.488 0.277 0.264 0.595
60 0.663 0.149 0.435 0.326 0.309 0.542
70 0.550 0.259 0.331 0.436 0.405 0.433
80 0.323 0.497 0.163 0.638 0.579 0.244
90 0 1 0 1 1 0
The discharge coefficients that describe a correlation between the ideal and the real
discharge of air through the openings were calculated as:
𝑄%
𝐶" ≡
2 · ∆𝑃 (3.1)
𝐴·
𝜌
Where 𝑄% is the volume flow rate, 𝐴 is the opening area, 𝜌 is the air density and ∆𝑃 is
the pressure difference across the opening.
The opening area and the discharge coefficients of the top and bottom openings are
given in Table 3.4.
Table 3.4 Discharge coefficients and dimension

of the openings of the DSF.
Top opening Bottom opening
Cd 0.72 0.65
“ab” distance [m] 0.09 0.11
α angle 11.5 14.0
Figure 3.6 Definition of

openings.
3.3 Experimental setup
The experiments were performed in the external air curtain functioning mode, i.e.
with a naturally ventilated air cavity. All the results are available both for a 10 minutes
timestep and a 1-hour timestep.
Particularly relevant for the task of this thesis is to mention the procedures for the
measurement of air and surface temperatures when the sensors are exposed to
38
3.3
Experimental setup
direct solar radiation, as well as the measurement of incident solar radiation and the
airflow rate inside the cavity.
3.3.1 Temperature measurements
Larsen (2007) explained that the presence of direct solar radiation is an essential
element for the façade operation, but it can profoundly affect measurements of air
temperature and may lead to errors of high magnitude using bare thermocouples.
Hence, the K-type thermocouples in DSF cavity were protected from the influence of
direct solar radiation by using a silver coated and ventilated air tube. The
thermocouples were placed in each DSF section at six different heights, in order to
build a vertical temperature gradient as shown in Figure 3.7.
Figure 3.7 Measurements of air temperature in the DSF cavity. Positioning of the
sensors, view from outside (left). Photo of the experimental setup (right),
thermocouples are shielded with silver coated ventilated tube (Larsen et al. 2008a).
The thermocouples located at the height of the small panes were necessary to
measure the inlet and outlet air temperature. These thermocouples were shielded
with the same method explained before, although in the case of the top ones the
shading provided by the ceiling of the double skin façade contributed to prevent
direct solar radiation from hitting them.
39
3
Figure 3.8 Measurement of outlet air temperature (Larsen et al. 2008a).

Measurement of glazing surface temperatures was performed in the centre of the
glazing panels for each large window section (BOL, BIL, BOH and BIH windows). The
measurements were carried on:
• The internal surface of the inner window (ii)
• The external surface of the inner window (ei)
• The internal surface of the outer window pane (ie)
Operatively, the surface temperatures were measured by thermocouples (type K)
glued to the glass surface using highly conductive paste and fixed to the support
using transparent tape. Surface temperature sensors were also protected from the
impact of incident solar radiation, using one of the two following strategies,
depending on the fact that the sensor was positioned on the outer skin or on the
inner skin (Figure 3.9):
• External skin (thermocouple 1): Highly reflective film of apx. size 20x20 mm.
• Internal skin (thermocouples 2 and 3): Highly reflective shield made out of
thin aluminium foil that shields both thermocouples together.
Figure 3.9 Shielding strategy from incident direct solar radiation. Left: Plan of DSF
cavity. Right: Section of DSF cavity.
40
3.3
Experimental setup
The uncertainty of equipment for the temperature measurement is estimated at a

maximum of ±0.14 °C. However, this uncertainty does not address the issues of the
solar exposure, as well as the limitations of the surface temperature measurements
related to the attachment of the sensor to the surface.
3.3.2 Incident solar radiation measurements
Solar radiation measurements were performed both on the rooftop of the test facility
and on the façade itself. The horizontally placed BF3 pyranometers measured the
global and the diffuse radiation on the top of the construction. The measurements
of the radiation on the vertical surface were performed through the use of three
pyranometers placed on the external surface on the internal surface and inside the
room.
Figure 3.10 Positioning of pyranometers in the experimental set-up (Larsen et al.

2008a).
The errors of the pyranometers’ measurements go from 0.01%, in the case of the
CM11 pyranometer positioned on the internal skin, to 10% of the BF3 pyranometer
positioned on the rooftop.
3.3.3 Airflow measurements
The airflow velocity profile is another essential measurement to be performed in

order to have a complete dataset that describes the DSF behaviour. The most known
experimental methods to perform this kind of measures are the velocity profile
method and the tracer gas method. Both methods have pros and cons. The first one
41
3
allows to have many measurements of the air velocity (one per each anemometer)
but is not very effective at measuring the flow rate and the flow direction. On the
other hand, the tracer gas method is able to predict the air flow rate and its direction,
but not the velocity profile. For this reason, both methods were implemented in the
experimental setup to take advantage of their complementary nature.
Velocity profile method

This method requires a set of anemometers to measure the velocity profile. The
shape of the determined velocity profile depends on the number of anemometers
installed. Therefore, the number of anemometers used for the measurement must
be sufficient to provide the minimum necessary resolution of the shape of the
velocity profile. However, the number of anemometers must not be too large
because otherwise, it can become an obstruction to the flow itself. Thus, the method
becomes a trade-off between the maximum desired amount of anemometers and
the minimum desired flow obstruction. In the case of double skin façades, the
method has another challenge that is the calibration of the hot-sphere
anemometers. The problem is that they have to be calibrated to detect velocities that
vary from 0m/s to 5m/s, which is a quite complex range because it involves both very
low velocities and medium-high velocities. The calculated error in the velocity
measurement of this method has been estimated to be 0.01 m/s.
46 hot-sphere anemometers were installed inside the cavity at six different heights
as shown in Figure 3.11.
Figure 3.11 Positioning of anemometers in the DSF cavity (left). Photo of the
anemometers in the DSF cavity (right) (Larsen et al. 2008a).
42
3.3
Experimental setup
Tracer gas method
This method employs a tracer gas (usually CO2 or SF6) that is mixed into the airflow,
by measuring the quantity of tracer gas that reaches the top of the cavity, it is
possible to calculate the air flow rate. This approach requires the minimum amount
of measurements and equipment, but it is characterized with frequent difficulties
caused by the wind wash-out effects, the time delay of signal caused by the time
constant of gas analyser and the difficulty in obtaining a uniform concentration of
the gas. In the measurements it has been used the constant injection method firstly
described Etheridge (1996), using CO2 as tracer gas.
Figure 3.12 Positioning of the perforated tube for the release of the tracer gas at the
bottom of the DSF cavity (left). Positioning of the air intakes for samples of tracer
gas polluted air, at the top of the DSF cavity (right).
It is not possible to give a precise error estimation for this kind of measurements
because of two main reasons:
• the dynamic nature of the measured phenomenon that induces the accuracy
of the results to change along with the boundary conditions
• the lack of information and experience for this kind of measurement in
outdoor test facilities
Assuming McWilliams’ (2002) hypothesis, the tracer gas concentration is constant
throughout the measured zone. The expected error of the tracer gas method is in the
range of 5-10% depending on the degree of mixing between the tracer gas and the
air.
43
CHAPTER 4
CFD MODEL
This chapter is dedicated to the explanation of the CFD model setup and in particular to its
three main features. The first one is the mesh, namely the group of cells that represent the
fluid domain. It is needed for the application of the finite volume method used to solve the
Navier-Stokes equations. The other two steps are the application of the boundary
conditions and the selection of the solver. The DSF problem has been treated by assuming
a buoyancy-driven, turbulent flow of an incompressible fluid under steady-state conditions.
The k-ω SST turbulence model proved to be the most suitable for this type of problem,
whereas the incompressible and the buoyancy effects can coexist thanks to the
Boussinesq approximation.
4.1 Fluid dynamic model
As Versteeg and Malalasekera (1995) stated: “Computational Fluid Dynamics (CFD) is

the analysis of systems involving fluid flow, heat transfer and associated phenomena
such as chemical reactions by means of computer-based simulations”. In other words,
this means that CFD is about solving the discretised form of the Navier-Stokes
equations with the aid of a computer.
Even though, as said in the previous chapters, it takes quite a long time to get to the
solution, CFD analyses are still faster and cheaper than using experiments when it
comes to studying complex flow phenomena such as external aerodynamics
problems or natural ventilation problems.
45
4
CFD Model
For Newtonian fluids such as air, the fluid motion is described by a system of these
five differential equations.
1. The equation of mass conservation
𝜕𝜌
+ 𝛻 𝜌𝑢 = 0 (4.1)
𝜕𝑡
2. The three equations of momentum conservation along the x, y and z axes
𝜕𝜌𝑢 2
+ 𝛻 · 𝜌𝑢𝑢 = −𝛻𝑝 + 𝜇𝛻𝑢𝐼 + 𝛻 · 𝜇 𝛻𝑢 + 𝛻𝑢 9
+ 𝜌𝑔 + 𝑆 (4.2)
𝜕𝑡 3
3. The equation of energy conservation
𝜕𝜌𝑒 2
+ 𝛻 · 𝜌𝑒𝑢 = −𝛻𝑝𝑢 + 𝜇 𝛻 · 𝑢 𝑢 + 𝛻 · 𝜇 𝛻𝑢 + 𝛻𝑢 9 𝑢 + 𝛻 𝜆𝑇 + 𝜌𝑄 + 𝜌𝑔𝑢 (4.3)
𝜕𝑡 3
To solve the system of equations, it is also necessary to define the constitutive

relationships that express the links between the variables. These relationships are:
• The internal energy equation
• The equation of state of perfect gases
• The Fourier’s law of heat conduction
• The generalised form of Newton's law of viscosity
Most of the commercial CFD software implement the finite volume method (FVM) for
spatial discretisation of these five differential equations. The FVM is a “numerical
technique that transforms the partial differential equations representing conservation
laws over differential volumes into discrete algebraic equations over finite volumes”
(Moukalled et al. 2016).
Three main things concur in the success of the CFD model:
• The mesh. Hence a discretisation into cell elements of the fluid domain.
These cells are the finite volumes on which the FVM will be applied
• The definition of the boundary conditions of the problem, namely pressure,
velocity and temperature, as well as the wall functions for the computation
of the turbulence models
• The selection of the appropriate solver
The following paragraphs explain the setup that was chosen for these three main
features, in order to perform the CFD analysis of the DSF case-study.
46
4.2
Mesh generation
Before proceeding with the description of the model, it is necessary to define the
assumptions that have been made:
1. The model is built in three dimensions as stated in Chapter 3 to evaluate if a
3D model is significantly more accurate than a 2D model. Since the boundary
conditions are three dimensional, there could be airflows in the z direction
but also in the x and y ones that could lead to different results compared with
a 2D model.
2. The atmospheric environment has been simulated with the use of air
plenums as in Dama et al. (2017)
3. Steady state simulation with boundary conditions averaged on a 10 minutes
time interval.
4. The flow is considered as turbulent to highlight the whirlwinds that can
generate inside the DSF
5. The density of the air is not considered as constant since the difference of
density activates the buoyancy effect which, together with the wind, moves
the air inside the cavity
6. The fluid is treated as incompressible since Mach’s number is far more lower
than 0.3. However, the variation of density needs to be taken into account,
and so it has been introduced Boussinesq approximation.
4.2 Mesh generation
The mesh is a group of non-overlapping cells that discretise the physical domain.
The conservation equations are solved by these discretised elements in a process
that is very similar to the finite element modelling.
There are many ways to obtain the most suitable mesh depending on the problem
that the user needs to face. In this case, it has been chosen to discretise the problem
using a block-structured grid generated with the snappyHexMesh utility that is part
of the OpenFOAM® package. snappyHexMesh is a fully automatic, parallel, octree-
refinement-based mesh generation app that passes through four steps to generate
the mesh. This tool generates 3D meshes containing hexahedra and split-hexahedra
47
4
CFD Model
elements, generated automatically from triangulated surface geometry in the

Stereolithography (STL) or Wavefront Object (OBJ) format (Moukalled et al. 2016).
Hence, the first thing that has been implemented is a 3D model of the DSF, that could
be then exported into the .stl file format. In order to have a parametric geometry that
could be easily changed by the user, the model was built into the Grasshopper®
environment.
Figure 4.1 CAD model of the DSF with Figure 4.2 STL file of the DSF.
measures.
The geometry is composed of the double skin façade domain and an air plenum
representing the external atmosphere. This plenum is partitioned into two parts to
allow the possibility of setting up a ∆?,ABC" that is necessary to simulate the effect of
natural ventilation.
Once the .stl file has been created, the next step is to generate the blockMesh that
is a type of base mesh in which the various refinements will take place. The only
48
4.2
Mesh generation
important thing about this step is to keep a cell ratio ≈ 1 otherwise the convergence
of the snapping procedure is slow and can even fail. The cell dimension of this coarse
mesh has been set to 13cm, which has proven to be the most balanced value between
a not-too-coarse mesh and a mesh that, being too refined, generated too many cells.
The mesh approximately conforms to the surfaces of the STL file by iteratively
refining the starting mesh that has been provided and morphing the resulting split-
hex mesh to the surface.
The algorithm works in three phases:
1. Generation of the Castellated mesh, which is obtained by splitting the cells in
correspondence of the surfaces and then removing the ones that remained
inside of the closed geometry that was provided. In this phase, it is also
possible to set some regions to get a higher refinement level by fatherly
splitting the starting cells. The castellated mesh is very rough, and it is not
suitable for simulations. However, it is crucial to obtain a good castellated
mesh because its quality will profoundly influence the next steps.
2. Snapping of the mesh to the surfaces to generate the snapped mesh. It is
obtained by moving the cell vertex points onto the surface geometry to
remove the jagged castellated surface from the mesh. This mesh can be
theoretically used for simulation but, in this case, since there was the need
to model the heat transfer between the fluid and the solid surfaces, it was
necessary to perform the last step
3. Optional layering of the mesh. It is created to simulate better the effect of the
fluid dynamic boundary layer. The layer addition produces longer cells with
different thicknesses at the boundary of the surfaces. In this case, it was
chosen to have five layers to have a fine simulation of the heat exchange.
Figure 4.3 Iterative workflow of the snappyHexMesh algorithm.
49
4
CFD Model
Figure 4.4 Background mesh. Figure 4.5 Castellated mesh.
Figure 4.6 Snapped mesh. Figure 4.7 Layered mesh.

As shown in Figure 4.5 and Figure 4.6, there is a buffer zone with refinement equal
to 2 between the coarse mesh of the plenum and the level-3 refined mesh that stays
inside the cavity. This area makes the passage between the highly refined zone and
the level-0 refined mesh smoother. In this way, convergence problems are avoided
during the simulation.
The structure of the mesh is important because it is strongly connected with the
computation of the operators, such as the gradient. For this reason, some
parameters have to be checked before proceeding with the use of the mesh for the
50
4.3
Boundary conditions
calculations. In particular, for this type of mesh, the most critical parameter that has
to be controlled is the skewness of the cells. This is important because, when the
grid is skewed, the line that connects the centres of two adjacent cells does not
necessarily pass through the centroid of the face. If this does not happen, it cannot
be guaranteed the precision of the second order discretisation. This means that
when the cells are too skewed, the solution might not reach convergence or even fail.
Table 4.1 reports the quantities and their relative value obtained using the
checkMesh utility.
Table 4.1 Consistency checking of the employed mesh.

Value Check
Max cell openness 3.86363e-15 OK
Max aspect ratio 89.7486 OK
Minimum face area 1.49288e-07 OK
Maximum face area 0.028636 OK
Min volume 1.49665e-08 OK
Max volume 0.00360174 OK
Max: 64.7178
Mesh non-orthogonality OK
Average: 3.86572
Max skewness 3.98888 OK
4.3 Boundary conditions
In order to solve the continuous problem, it is necessary to explicit the physical

conditions at the boundaries of the fluid domain. Generally, the boundary conditions
are either the value of a variable (Dirichlet conditions) or its gradient along one of
the three axis (Neumann conditions).
In this problem, the boundary conditions that have to be specified are the
temperature, the velocity and the pressure, as well as the wall functions that allow
to calculate the turbulence functions (αT, νT, k, ω). These values are taken from the
experimental dataset discussed in Chapter 3. More specifically, four configurations
of boundary conditions were chosen to run the CFD simulations. Case A and case D
are both temperature and wind driven while case B and case C are mainly subjected
to the delta of pressure generated by the wind rather than to the low difference of
temperatures. These four cases will be compared with the experimental data and the
simplified model. The boundary conditions of each case correspond to the average
of the 10 minutes measurements performed during the experimental campaign.
51
4
CFD Model
Table 4.2 Cases chosen for the CFD simulations.

Case Date & time Text [°C] Wind speed [m/s] Wind_dir Δp,wind
A Oct 1st, 13:40-13:50 19.0 4.2 183° -1.23
B Oct 2nd, 12:00-12:10 13.4 4.3 180° -1.38
C Oct 8th, 15:30-15:40 15.1 9.2 278° 5.00
D Oct 13th, 13:10-13:20 14.6 4.3 107° 0.47
4.3.1 Temperature boundary conditions
The temperature boundary conditions were obtained from the experimental data
since each of the glasses had its temperature measured experimentally. The
temperature was considered homogeneous throughout the whole area of each
glass. Moreover, since there was no measurement of the temperatures of the frames
they were hypothesised to be at the same temperature as the neighbouring glasses.
The Dirichlet conditions were applied on the floor and the ceiling of the DSF relying
on the experimental dataset. On the other hand, the surfaces of the plenum were
given a zeroGradient boundary condition. The zero gradient means that for a given
quantity ϕ the finite difference approximations leads to:
𝜕𝜙 𝜙I − 𝜙J
= 0 ⟹ = 0 ⟹ 𝜙I = 𝜙J (4.4)
𝜕𝑥 G 𝑥I − 𝑥J
This approximation makes the boundary value to be replaced with the value of the
cell next to the boundary. From the temperature point of view, this means that the
surfaces with the zeroGradient condition are adiabatic. Operatively, this condition was
set because the surfaces of the plenum are fictitious and so, it would be physically
incorrect to let the air exchange heat with them.
The inlet and the outlet surfaces that are the ones on which the difference of
pressure is applied were given the inletOutlet condition. This boundary condition
provides a generic outflow condition, with specified inflow for the case of return flow. In
this case, when the flow is going out from the surface, it takes the zeroGradient condition
while, if it goes towards the surface, it takes the value of the external temperature.
52
4.4
Solver selection and setup
4.3.2 Velocity boundary conditions
A graphic representation of the velocity boundary conditions is given by Figure 4.1.

There are mainly three different velocity boundary conditions that have been
applied:
• noSlip: it is a Dirichlet boundary condition which states that the velocity of the fluid
is equal to the wall velocity. It follows from the fact viscous fluids stick to solid
boundaries. This condition is applied to all the surfaces that are artificial plus
the ground to simulate their friction effect.
• Slip: it is the opposite of the noSlip condition. In this case, the surface does not
limit the flow of the air. This condition is applied to the plenum surfaces that being
fictitious should no exert any friction force.
• pressureInletOutletVelocity: this condition was applied to the intlet and outlet
patches. A zeroGradient condition is applied for outflow; for inflow, the
velocity is obtained from the patch-face normal component of the internal-
cell value.
4.3.3 Pressure boundary conditions
The pressure boundary condition was very easy to set up. There is a fixedFluxPressure
boundary condition everywhere apart from the inlet and outlet patches. This
constrain adjusts the pressure gradient, such that the flux on the boundary is the
one specified by the velocity boundary condition. On the inlet and outlet patches,
there were respectively set a zero pressure Dirichlet condition and a fixed value of
pressure representing the difference of pressure generated by the wind (∆?,ABC" ).
4.4 Solver selection and setup
The solver is the part of the code that actually solves the system of the Navier-Stokes
partial differential equations. The most accurate and simple approach to the solution
of the problem is to solve the system of equations without introducing any
simplification nor approximation. This approach is the direct numerical simulation
(DNS). The results of this kind of simulations are very detailed, too detailed from the
53
4
CFD Model
engineering point of view (Ferziger et al. 2002). The most suitable approach for the
DSF fluid dynamic modelling is to employ the Reynolds-averaged approach to
turbulence in which all the unsteadiness is considered as part of the turbulence.
Hence, the model presented in this chapter employs the Reynolds-Averaged Navier-
Stokes (RANS) equations coupled with a turbulence model. In its simplest form,
turbulence is described by k, the kinetic energy, and L, the length scale. This means
that the solution of the problem is achieved by a system of two equations, the one
describing k can be easily derived, while, the equation needed to derive L is not
obvious. Many relationships have been proposed to solve this problem. The most
employed is to write L as a function of k and ε, the dissipation, creating the so-called
k-ε model. The second most commonly used two-equation model is the k-ω one
(Ferziger et al. 2002) developed by Komolgorov (1942). In the present work it has
been chosen to use a variant of the k-ω model, the k-ω SST (Shear-Stress transport
model) developed by Menter (1992, 1993) that is particularly suitable for external
aerodynamic applications.
The selected turbulence model will operate, as said before, inside the solver
algorithm. OpenFOAM® does not feature a generic solver that is suitable for all the
cases; instead, the user can choose the one that better fits the problem that needs
to be solved. The choice fell upon the buoyantBoussinesqSimpleFoam segregated
solver. This decision was driven by the assumptions and simplifications that were
already stated in paragraph 4.1. To resume, they are:
• Turbulent flow
• Incompressible fluid but with buoyancy effects
• Steady-state problem
The solver is suited to deal with heat transfer and buoyancy-driven flows. Since the
fluid is treated as incompressible, because the velocities of the problem are very
slow compared to the speed of sound, the buoyancy effect is introduced through the
use of the Boussinesq hypothesis. This approximation states that, if the density
variation is not too large, it can be considered constant in the unsteady and
convection terms, while it can be solved as a variable in the gravitational term,
assuming it linearly dependent from the temperature. The introduced error is in the
54
4.4
Solver selection and setup
order of 1% if the temperature difference is lower than 15°C (Ferziger et al. 2002) as
it is in the case of a DSF.
Finally, the steady-state problem is modelled through the use of the SIMPLE
algorithm that corrects the pressure and velocity terms contained the momentum
equations by unlinking them from the temporal derivatives.
55
CHAPTER 5
SIMPLIFIED MODEL
This chapter deals with the description of the simplified model’s functioning, which is based
on the solution of a steady-state energy balance over a thermal network. In this way, it
simulates the behaviour of a double skin façade under natural ventilation reducing the
computational time and maintaining a good level of precision. There are three main
modules in the code that contribute to get to the solution: the radiative model, the fluid
dynamic model and the thermal model. The first one calculates the radiative heat fluxes to
be put in the thermal network at each timestep thanks to two inputs: the solar radiation
impinging on the façade and the glazing optical characterisation. The fluid dynamic module
calculates the wind speed inside the air channel using the differences of pressure
generated by the buoyancy of the air and by the wind. Finally, the thermal model uses the
outputs from the previous two to calculate the thermal resistances and the convective
coefficients at each timestep with these values it solves the network outputting all the
temperatures and the heat flux.
5.1 Model description
The thermal model described in this chapter aims to be a simplified model that can
simulate the behaviour of the DSF in a small amount of time. The results provided by
this algorithm are going to be validated against the experimental data and the CFD
results. The model is based on steady-state energy balance equations on a thermal
network solved through an iterative method. It considers both convection and long
57
5
Simplified model
wave radiation. The convection inside the channel is modelled by a simplified

integral approach employing average bulk temperatures and superficial heat
transfer coefficient correlations. The thermal model is coupled with a radiative
model that calculates the heat flux provided by the sun and a fluid-dynamic model,
based on a pressure loop which takes into account both buoyancy and wind
differential pressure at the openings, as described by Dama (2017). Figure 5.1 shows
the implemented DSF thermal network.
correspond to the glass positions. ϕ1, ϕ2 and ϕ3 correspond to the solar radiation
absorbed respectively by the glass of the outer skin, the external glass and the
internal glass of the inner skin (double glazed unit). Glass panes are represented in
blue.
This model is based on the one proposed by Angelotti et al. (2007) and is divided into
a thermal and fluid-dynamic part. The main differences with the model quoted above
are that the ventilation mode is natural and not mechanical introducing a higher level
of complexity because wind conditions are harder to simulate that the mechanical
ones where the mass flow rate of air is always known. Natural ventilation has been
modelled with the equations and the approach described by Dama (2017). Secondly,
the definition of the 𝑅% parameter, namely the fictitious thermal resistance that
takes into account the quantity of heat extracted by the natural ventilation of the
DSF, has been simplified and been linked only to the air velocity and not to the
surface temperatures. The following paragraphs explain in detail the functioning and
the equations that are behind the main sections of the analytical model.
58
5.1
Model description
5.1.1 Algorithm workflow
This paragraph explains briefly how the iterative process of the algorithm works. It
starts by reading the static inputs such as the geometric and geographic data as well
as the physical properties of the glasses obtained from the optical model (LBNL
Window). The following step is entering inside the for loop that solves the thermal
network at each timestep (every hour) of the simulation period. The model reads the
experimental inputs at that timestep: date and hour, external and internal air
temperature and external air velocity. The solar model uses the geopositioning data
and the date to compute the solar radiation on the vertical surfaces. The heat fluxes
that have to be placed on the glasses are calculated coupling this data with the
glazing properties. Before proceeding with the fluid dynamic model, it is performed
an initialisation of the temperatures. This is necessary because the value of 𝑇L (air
temperature inside the channel) enters in the term of ∆?,MNOPQ and thus, in the
calculation of the air velocity. The initialisation of the temperatures is made
considering a fully closed cavity with no air circulation. Using this input, the fluid
dynamic model is then able to calculate the average air speed inside the channel and
then to pass this information to the thermal model, that calculates all the
temperatures at the nodes of the thermal network. The thermal network is solved by
superposition of effects. This means that it is firstly solved with imposed
temperatures, then with imposed fluxes and finally the two results are summed up.
The calculation of air velocity and temperatures needs to be repeated iteratively until
the value of 𝑇L converges. A new definition of 𝑅% , described in paragraph 5.1.4, allows
to make the convergence of this iteration loop easier because it does not depend on
any temperature; this makes the algorithm more robust. The model finally writes the
outputs that are:
• Tf: air cavity temperature [°C]
• Tin: Inlet air temperature [°C]
• Tout : Outlet air temperature [°C]
• Tsi : surface temperature of the internal glass (internal side) [°C]
• Ts1: surface temperature of the external glass (internal side) [°C]
• Ts2: surface temperature of the internal glass (external side) [°C]
• Average air speed inside the cavity [m/s]
• Heat flux towards the internal room [W/m2]
59
5
Simplified model
It then starts iterating again on the next timestep until the whole simulation is
concluded.
Figure 5.2 Iterative flowchart of the algorithm. green=inputs, orange=outputs.
60
5.1
Model description
The next paragraphs explain in detail the functioning of the core parts of the
algorithm.
5.1.2 Solar-optical model
The calculation of the glass temperatures is one of the main additions of the present
work to the existing model. In the previous versions of the algorithm, the glass
surface temperatures were taken as an input from the experimental data. However,
from the point of view of creating a predictive model of the double skin façade
energy behaviour, these temperatures have to be computed directly by the code. In
order to calculate them, it has been necessary to couple the solar model with the
optical model. The solar model takes into consideration the sun position, the direct
and diffuse radiation on the DSF to evaluate the radiation impinging on the vertical
surface, while the optical model gives the optical properties of the glasses. The final
purpose is to pair these data to calculate the heat fluxes ϕ1, ϕ2 and ϕ3 that will be
inserted in the thermal network of Figure 5.1.
The sun position is calculated using the number of the day, the latitude, the
longitude and the orientation of the façade. All of these parameters contribute to
calculate the incident angle (ϑ) and the zenith angle (ϑz).
It is also calculated a reduction factor equal to:
𝑠 ∙ 𝑡𝑎𝑛(𝛼? )
𝑅𝐹 ≡ (5.1)
𝐻
to take into account the shade cast by the roof of the channel.
Where 𝑠 is the depth of the cavity, αp is the profile angle of the sun and 𝐻 is the height
of the façade.
The most critical part of this section is the glass characterisation, namely the
definition of transmittance, absorbance and reflectance of the glasses because the
model has shown to be quite sensitive to these values. The complete angular
characterisation of the DSF glazing systems has been obtained from the optical
model of the LBNL Window software. Chart 5.1 shows the glazing system
transmittance (which is the visual transmittance across all the four glasses that
compose the DSF) and the absorbance of each glass. Table 5.1 reports the normal
incidence properties of the glazing adopted in the experimental report and in
61
5
Simplified model
Window calculation while Table 5.2 reports the same properties but for the diffuse
radiation.
Chart 5.1 Glazing system characterisation for different incidence angles of direct
solar radiation.
Table 5.1 Characterisation of the glazing system for direct radiation.
External glazing Internal glazing

τ sol ρ sol τ sol ρ sol,f ρ sol,b
Experimental report 0.763 0.076 0.532 0.252 0.237
LBLN Window 0.771 0.071 0.536 0.277 0.267
Table 5.2 reports the coefficients that were used to deal with the diffuse radiation
part.
Table 5.2 Characterisation of the glazing system for diffuse radiation.
External glazing Internal glazing

τ sol ρ sol τ sol ρ sol,f ρ sol,b
LBLN Window 0.69 0.097 0.447 0.327 0.312
As it can be seen from Table 5.1, the coefficients used in the model are very similar
to the real ones and so they have been considered acceptable.
The glass characterisation is important because it is used for the calculation of the
heat flux that has to be assigned to each glass of the skin in order to simulate the
radiative contribution of the sun. The other parameters that appear in the equation
62
5.1
Model description
are direct solar radiation (𝐺"BP ) and diffuse solar radiation (𝐺"BLL ). They are calculated
with an if statement considering cos(ϑz) < 0.05 as the condition in which the sun has
set or it still has to dawn, and so the only radiation that can be, if present, is the
diffused one. Hence, if it is daytime the equations are:
𝐺N,"BP ∙ 𝑐𝑜𝑠 𝜗
𝐺"BP = 𝑚𝑎𝑥 0;
𝑐𝑜𝑠 𝜗a (5.2)
𝐺"BLL = 0.5 ∙ 𝐺N,"BLL + 𝑎𝑙𝑏𝑒𝑑𝑜 ∙ 0.5 ∙ 𝐺N,"BLL
If it the sun has set or it is dawning:
𝐺"BP = 0
𝐺"BLL = 𝐺N,MgM Gh,tot is zero except at dawn and twilight (5.3)
Where 𝐺N,"BP is the direct radiation on the horizontal plane, 𝐺N,"BLL is the diffuse
radiation on the horizontal plane, 𝐺N,MgM is the total radiation on the horizontal plane,
ϑ is the incident angle, 𝜗a is the zenith angle and the albedo is the one of the ground
which is set to 0.2.
With the values obtained by the equations described in this paragraph it is possible
to calculate the heat fluxes ϕ1, ϕ2 and ϕ3 to be placed in the thermal network of Figure
5.1
𝜑B = 𝐺"BP · 𝛼B,"BP + 𝐺"BLL · 𝛼B,"BLL (5.4)
5.1.3 Fluid dynamic model
The fluid dynamic model has the ultimate purpose of calculating the wind speed
inside the air channel. It does it by implementing an iterative calculation (on the 𝜒
value) of the pressure loop presented in Dama (2017).
2 ∙ ∆?,MgM
𝑣Q = I
1 𝜒(𝑅𝑒) ∙ 𝐻 𝑠
𝜌AL ∙ + + 𝐴gnM (5.5)
I 𝐷N
𝑠 𝐶",Mg?
𝐴BC
𝐶",lgMMgQ
Where:
63
5
Simplified model
• 𝜌AL is the density of air inside the channel calculated with the perfect gas
model equation.
• 𝑠 is the thickness of the air channel
• 𝐻 is the height of the façade
• 𝐴BC , 𝐴gnM are respectively the areas per meter of width of the bottom and top
opening
• 𝐶",lgMMgQ , 𝐶",Mg? are the bottom and top pressure drops coefficients of the
façade, obtained from the experimental report
• 𝐷N is the hydraulic diameter, set to 2 · 𝑠
• 𝜒 is a parameter introduced by Colebrook and White (1937) in their formula.
J
It is calculated as s | where 𝜀 is the superficial roughness,
tu |.z||y
oI·pqr {
v.wxyz (}~· •)
set to 0.01
• ∆?,MgM is the most important parameter of the formula, it represents the
difference of pressure that drives the air movement inside the cavity. Two
terms compose it: the delta pressure generated by the presence of the wind
(∆?,ABC" ) and the delta pressure generated by the buoyancy of hot air
(∆?,MNOPQ ). ∆?,ABC" was calculated with the following formula:
𝑤ƒI
∆?,ABC" = 𝜌O•M · · ∆𝐶? (5.6)
2
The ∆?,MNOPQ component is calculated using Archimedes’ relationship for an

ideal gas at constant pressure:
∆?,MNOPQ = 𝑔 · ∆𝐻 · 𝜌O•M − 𝜌AL (5.7)
5.1.4 Thermal model
The main issues for the solution of the thermal network are the calculation of the
inlet air temperature and of the heat transfer coefficients.
As said before, it has been introduced 𝑅% with the purpose of modelling the heat
removed by the natural ventilation. Hereafter, it is reported the demonstration of the
formula for the new definition of 𝑅% .
64
5.1
Model description
The ventilation flux is written as,
𝛷% ≡ 𝑚 ∙ 𝑐? ∙ 𝛥𝑇 = 𝐿 ∙ 𝑠 ∙ 𝜌 ∙ 𝑣 ∙ 𝑐? ∙ (𝑇gnM − 𝑇BC ) (5.8)
Where 𝑇gnM and 𝑇BC are respectively the outlet and the inlet air temperatures of the
ventilation channel. Dividing by the area of the façade it is obtained the ventilation
flux per unit area.
𝛷% 𝐿 ∙ 𝑠 ∙ 𝜌 ∙ 𝑣 ∙ 𝑐? ∙ (𝑇gnM − 𝑇BC )
𝜑% ≡ = (5.9)
𝐻∙𝐿 𝐻∙𝐿
A fictitious thermal resistance 𝑅% is introduced to couple the air in the channel with
outdoor air, and is defined as:
𝑇L − 𝑇BC
𝑅% ≡ (5.10)
𝜑%
Where TL is the bulk temperature inside the air channel. By joining together the
definitions of φ% and 𝑅% :
𝑇L − 𝑇BC
𝑅% = (5.11)
𝑠 ∙ 𝜌 ∙ 𝑣 ∙ 𝑐? ∙ (𝑇gnM − 𝑇BC )
𝐻
By solving the differential energy balance on the channel, as described in the papers
by Angelotti et al. (2007) and by Angeli et al. (2015), the following equations are
drawn:
1 − 𝑒 o‰ƒ
𝑐≡ 𝑇gnM = 𝑇BC + (𝑇‹ŒC − 𝑇BC )(1 − 𝑒 o‰ƒ )
𝑘𝐻
ℎŽ%J + ℎŽ%I (5.12)
𝑘≡ 𝑇L = 𝑐𝑇BC + 1 − 𝑐 𝑇‹ŒC
𝑠 ∙ 𝜌 ∙ 𝑣 ∙ 𝑐?
By substituting (5.12) into (5.11) it is possible to obtain
1 − 𝑒 o‰ƒ 𝑘𝐻 − 1 + 𝑒 o‰ƒ
𝑇BC + 𝑇Œ − 𝑇BC
𝑘𝐻 𝑘𝐻 (5.13)
𝑅% =
ℎŽ%J + ℎŽ%I
∙ (𝑇Œ + 𝑇BC 𝑒 o‰ƒ − 𝑇Œ 𝑒 o‰ƒ − 𝑇BC )
𝑘𝐻
After some other algebraic passages it is obtained the final form of R %
𝑒 o‰ƒ + 𝑘𝐻 − 1
𝑅% = (5.14)
ℎŽ%J + ℎŽ%I 1 − 𝑒 o‰ƒ
65
5
Simplified model
This definition of 𝑅% is more efficient because it depends only on the air velocity (the
𝑘 parameter) and not - at least explicitly - on the temperature of the surfaces. This
allows to have a more stable model that converges faster and is less subjected to
instability.
Another important part in the definition of the model is to select the most suitable
correlations to be used for the calculation of the heat convective and radiative
coefficients since the results are very sensitive to those values. The important
coefficients that are present in the model are the convective and radiative ones both
for the internal room, for the air channel and for the external side.
From the inside to the outside, the convective coefficient has been calculated as:
𝑁𝑢 ∙ 𝜆
ℎŽ%,BCM = 𝑊/𝑚 I 𝐾 (5.15)
𝐻
Where:
𝑁𝑢 is the Nusselt’s number calculated with the correlation by Churchill and Chu
(1975):
J I
𝑁𝑢 = 0.825 + 0.325𝑅𝑎 • 𝑃𝑟 = 0.72
(5.16)
𝑔𝛽(𝑇ŒB − 𝑇™ )𝐻 š 1
𝑅𝑎 = (𝛽 = ; 𝑇 = 𝑇BCM ; 𝜈 = 1.3 ∗ 10o• ; 𝛼 = 1.9 ∗ 10o• )
𝜈𝛼 𝑇BCM ™
𝜆 is the thermal conductivity of air [W/mK]

𝐻 is the height of the glazed wall facing the room [m]
Inside the air cavity it has been employed McAdams (1954) correlation:
ℎŽ%,ŽN‹CCOŸ = 5.62 + 3.9 ∙ 𝑣‹BP 𝑊/𝑚 I 𝐾 (5.17)
Where 𝑣‹BP is the air velocity inside the air channel that is iteratively calculated using
the Colebrook and White (1937) formula by the fluid dynamic component of the
model.
For the external part, it has been used Sharples’ (1984) correlation that considers
two different values of 𝑣Œ depending on the degree of exposure to the wind of the
façade.
1.8 ∙ 𝑣J¡ + 0.2 directly exposed façade

𝑣Œ = 𝑚/𝑠 (5.18)
0.2 ∙ 𝑣J¡ + 1.7 not directly exposed façade
66
5.1
Model description
The value of 𝑣Œ was obtained by interpolating the two cases with respect to the
cosine of the wind angle of attack. This allowed to take into account the different
gradients of wind exposure of the façade at each timestep
ℎŽ%,O•M = 1.7 ∙ 𝑣Œ + 5.1 𝑊/𝑚 I 𝐾 (5.19)
For the radiative coefficients in the interior and in the channel, it has been used the
common grey body assumption. While for the calculation of the radiative heat
transfer towards it has been necessary to calculate the sky temperature using
Swinbank’s (1963) formula:
J.•
𝑇Œ‰² = 0.05532 ∙ 𝑇O•M 𝐾 (5.20)
The following table shows a summary of the correlations used to calculate all the
heat coefficients.
Table 5.3. Heat transfer coefficients correlations.
External side Channel Internal side
Convective part Sharples (1984) McAdams (1954) Churchill and Chu (1975)
Tsky based on
Radiative part Grey surfaces Grey surfaces
Swinbank (1963)
The other important variable to be calculated in order to be able to solve the thermal
network is the inlet air temperature. As Saelens et al. (2004) observed, using the
external air temperature is not an effective approximation, because the inlet air
exchanges heat with the floor of the cavity and so it enters at a higher temperature
compared with the external one. For this reason, it has been introduced a correction
that considers the thermal exchange between the basement (heated by the solar
radiation) and the inlet air imposing the following energy balance:
lO‹Q "BLL "BLL

𝑚𝑐? 𝑇BC − 𝑇O•M = 𝑞Ž% 𝛼l‹Œ 𝜏O•MoŒ‰BC lO‹Q
𝐺%OPM. tan 𝑎? + 𝜏O•MoŒ‰BC 𝐺%OPM. 𝐹l‹Œ→%OPM 𝐴l‹Œ (5.21)
This energy balance is made in steady state conditions and does not take into
account the thermal inertia effect of the basement.
𝑞Ž% is a term introduced to take into account the amount of absorbed radiation
exchanged by convection towards the inlet air. It has been set to 0.5.
𝛼l‹Œ is the absorption coefficient of the basement and has been set to 1 since there
were lots of black cables on the floor during the measurements
67
5
Simplified model
The last interesting feature implemented in the thermal model is how it calculates
the temperature of the air inside the channel and the temperature of the outlet air.
The equation comes from the energy balance of the cavity (also adopted in EN 13362-
2 standard).
ℎŽ%,J 𝑇Œ,J − 𝑇 𝑦 + ℎŽ%,I 𝑇Œ,I − 𝑇 𝑦 · 𝐿 · 𝑑𝑦 = 𝑚𝑐? 𝑑𝑇

(5.22)
𝑇 0 = 𝑇BC 0 ≤ 𝑦 ≤ 𝐻
From these equations, it is possible to obtain
𝑇‹ 𝑧 = 𝑇BC + 𝑇‹ŒC − 𝑇BC · (1 − 𝑒 o‰a ) (5.23)
Where 𝑇‹ŒC is the asymptotic temperature that can be theoretically reached by the
air inside the cavity. This value is the weighted average of the surface temperatures
of the glasses facing the cavity over their convective coefficients.
𝑇Œ,J ∙ ℎŽ%,J + 𝑇Œ,I ∙ ℎŽ%,I

𝑇‹ŒC = (5.24)
ℎŽ%,J + ℎŽ%,I
From equation (5.24) it is possible to draw the definitions of the cavity air
temperature (𝑇L ) and the outlet air temperature (𝑇gnM ) that are already expressed in
equation (5.12). So the temperature grows exponentially inside the cavity starting
from the inlet value and exiting at an outlet temperature, that is lower than the
asymptotic one.
Figure 5.3 Graphic representation of the temperature profile inside the cavity with
the change in height.
68
5.1
Model description
By having solved the thermal network, which means to have calculated all the values
of the temperatures and the heat transfer convective coefficients, the algorithm is
finally able to compute the heat flux coming into the internal ambient at each
timestep with the following formula:
𝑄BCM = 𝐴 · ℎŽP,BCM · (𝑇ŒB − 𝑇BCM ) + 𝜏"BP,¹º» · 𝐺"BP,"ŒL + 𝜏"BLL,¹º» · 𝐺"BLL,"ŒL (5.25)
Where 𝐴 is the area of façade, ℎŽP,BCM is the radiative-convective heat transfer

coefficient of the internal ambient, 𝜏¹º» is the direct and diffuse solar transmittance
of the DSF system considering the position of the sun, 𝑇ŒB is the temperature of the
the internal glass, 𝑇BCM is the internal temperature and 𝐺"ŒL is the amount of radiation
both direct and diffuse hitting on the vertical façade.
69
CHAPTER 6
MODEL VALIDATION
In this chapter the two models are validated by comparing their results with the
experimental data. The comparison between the CFD model and the measurements is
performed for four cases, each representative of a particular flow condition. The degree of
agreement between the outputs, temperature and air velocity, and the measurements gave
acceptable results in each of the four cases. The CFD model is also effective in highlighting
the turbulent and flow-reverse conditions that are established inside the cavity. The next
step is to test the simplified model against the measures of solar radiation, airflow rate,
surface temperatures and, most importantly, of the heat exchanged towards the inside,
which is the ultimate parameter that matters when performing a building energy simulation.
While the model has issues in identifying the peaks of the surface temperatures of the
glasses facing the cavity, it predicts very well the glass temperature on the room side with
a maximum error of 1.5°C.
Consequently the prediction of the outward heat flux is overestimated by only 6%, whereas
for the inward heat flux between 15% and 25%. The final part of the chapter puts to test the
two models, comparing airflow rate and vertical temperature profile. In both cases the
simplified model gives similar results to the CFD one demonstrating that the hypotheses at
its core are reasonable. The last parameter to be compared is the convective heat transfer
coefficient, which on average is overestimated by the simplified model by 30-40%.
This chapter is devoted to validate the results of the CFD and of the simplified model.
As said in Chapter 3, the first part is dedicated to the comparison of the results
71
6
Model validation
obtained with the CFD against the experimental data. Secondly, it is made the same
comparison for the simplified model and, at last, there is a cross comparison
between the three sets of data to evaluate the degree of precision of both
approaches. The main parameters that are important for validation purposes are: the
air velocity and the air flow rate inside the cavity, the outlet temperature of the air
and most importantly the heat flux exchanged towards the interior. This last variable
is the most important because is the one that matters when making the energy
balance of the whole building to which the DSF is connected to.
6.1 CFD model: investigation vs. measurements
The results of this paragraph are divided into four parts. Each part proposes the
results of one of the four cases chosen for the CFD analysis, as specified in Chapter
4. The four cases are representative of four different flow conditions that depend on
the wind and temperature boundary conditions.
• Case A: the difference of pressure of -1.23Pa creates a downward wind
pressure, however buoyancy is dominant, and the air stream goes upwards
• Case B: the downward wind pressure is very high and dominates against weak
buoyancy producing a downstream flow of air
• Case C: weak buoyancy and strong upward wind differential pressure
• Case D: buoyancy and upward wind differential pressure have the same sign
and generate an upstream flow
The solution converged after 10000 iterations that correspond to approximately 30
hours of calculation. The CFD data tested in this paragraph is the air velocity inside
the channel and the horizontal temperature profiles to check if three-dimensional
velocity patterns are generated inside the cavity. The air velocity simulations rely on
the experimental data obtained by the anemometers, six for each level of altitude,
while the thermocouples were placed singularly at each level and so there is a single
experimental measure for each height in the temperature charts.
The left, middle and right tags that are present in the following charts refer to the
data drawn from the centre of the three glasses in which the façade is vertically
divided. These labels apply to the façade when looked from the outside to the inside.
72
6.1
CFD model: investigation vs. measurements
6.1.1 Case A
Air velocity
Chart 6.1 Case A: simulated air velocity on the z axis (Uz) for each horizontal section.
White circles represent the value of the magnitude of U in the position of the
anemometers. Black circles represent the experimental measurements in the
position of the anemometers.
73
6
Model validation
Chart 6.1 shows the velocity profile on the horizontal section at the different heights.
The continuous lines refer to the z component of the velocity but, since the hot-
sphere anemometer measures the magnitude of the vector and not only a single
component, it has been also plotted the magnitude value of velocity in the position
of the anemometers to be confronted with the experimental data (black circles). The
velocity profile shows higher values on the borders and values tending to zero in the
middle, with errors comprised between 0.05 and 0.2 m/s.
h=0.96m h=1.91m h=2.5m
h=4.36m h=4.7m h=5.15m
Figure 6.1 Case A: horizontal section of velocity distribution along the z axis (Uz).
74
6.1
Chart 6.1 and Figure 6.1 show that in the horizontal sections the profile is very
homogeneous, since the line plots of the left, middle and right profiles look alike and
the 2D plots present similar distributions of flow in the three sections. The areal
profiles better highlight the presence of higher velocities, around 0.5 m/s, close to
the surfaces of the DSF. In particular, these spots could be generated by the
irregularities generated by the 3D discontinuities introduced by the mullions and by
the transoms.
Middle section Parallel to the y axis
Figure 6.2 Case A: vertical sections profiles of air velocity (magnitude values) with
velocity vectors (in white).
The analysis shows that, during case A, the flow is very turbulent inside the cavity
probably due to opposite forces generated by the buoyancy and by the wind.
The vertical sections also highlight the acceleration of the wind when entering and
exiting the DSF as expected and moreover, they display, once again, the acceleration
of the flow close to the mullions and the transoms.
75
6
Model validation
Temperature
Chart 6.2 Case A: Temperature profile of horizontal sections. Black circles represent
the experimental measure in the position of the thermocouple. White circles
highlight the values of the CFD data in the position of the thermocouple.
By looking at Chart 6.2 it can be stated that the predictions of the temperature of the
CFD model in case A have a high degree of agreement with the experimental data,
except for the inlet and outlet portions. The underestimation of temperature at the
76
6.1
opening is probably due to the particular flow condition close to the bottom of the
cavity, where recirculation forms, that drags hot fluid from the irradiated bottom wall
upwards.
h=0.1m h=1.5m h=2.5m
h=3.5m h=4.5m h=5.5m
Figure 6.3 Case A: horizontal section of temperature distribution.

The flat lines of the plots of Chart 6.2 and the uniform colours of the sections of
Figure 6.3 suggest a high mixing of the air. This observation is in agreement for what
has been said regarding the air velocity. The presence of air movement along the x
77
6
Model validation
and y direction create a homogeneous distribution of the temperature along the

section of the cavity.
Figure 6.4 Case A: vertical section profile of temperature.

Given the homogeneity of the temperatures Figure 6.4 only portraits the
temperature profile in the middle vertical section. In case A, there is a high
temperature increase (from approximately 19°C to 30°C), with a stratification pattern
typical of natural convection in vertical cavities.
78
6.1
6.1.2 Case B
Air velocity
Chart 6.3 Case B: simulated air velocity on the z axis (Uz) for each horizontal section.
79
6
Model validation
Case B velocity profile is different compared with the first case, the wind pressure is
very strong and creates a strong downstream. It is noteworthy that while the Uz
profile is very far from the measurements, the data of the magnitude profile is in
good agreement with the measures in three of the five cases even if the case is a
flow reversal one. This means that the stream flow is very turbulent and so the vector
of the air is not only directed along the z axis but in all the three directions.
h=0.96m h=1.5m h=2.5m
h=4.36m h=4.7m h=5.15m
Figure 6.5 Case B: horizontal section of velocity distribution along the z axis (Uz).
80
6.1
The wind velocity in the horizontal sections reaches very high values both positive
and negative. Moreover, the areas of higher velocity are not always concentrated in
the same positions. These two observations suggest the presence of whirlwinds that
move air along x and y. This hypothesis is confirmed by looking at Figure 6.6, where
the velocity vectors highlight a big recirculation pattern that reaches speed values
up to 0.5 m/s. These values increase the efficiency of the cooling effect of air on the
glazing because they increase the thermal exchange between the air and the
surfaces.
Figure 6.6 Case B: vertical sections profiles of air velocity (magnitude values) with
81
6
Model validation
Temperature
Chart 6.4 Case B: Temperature profile of horizontal sections. Black circles represent
82
6.1
The charts of the temperature profile show a slight underestimation of about 1/2 °C
in the prediction of the temperature. This output is in accordance with the results of
Chart 6.3, where it can be seen that the velocity profile computed by the CFD is very
close to the experimental data and, when the two are not in accordance, it is always
the CFD that is underestimating the measurements.
h=0.1m h=1.5m h=2.5m
h=3.5m h=4.5m h=5.5m
Figure 6.7 Case B: horizontal section of temperature distribution.
83
6
Model validation
Figure 6.8 Case B: vertical section profile of temperature.

The relatively high values of velocity result in a meagre rise from the outside value
because the temperature of the glass is close to the external one. The air enters the
cavity at 13.4°C warms up to 16°C and exits at approximately the same temperature
at which enters. This situation would be different in summer because there would be
higher glass temperatures and so, the high ventilation in the cavity would bring away
the heat thus cooling down the building.
84
6.1
6.1.3 Case C
Air velocity
Chart 6.5 Case C: simulated air velocity on the z axis (Uz) for each horizontal section.
85
6
Model validation
Case C velocity profile has higher values of air velocity similar to the ones of the
previous case. The simulated values have their peak at around 0.5 m/s on the left
side of the cavity and then fade to zero towards the exterior side. Except for the top
of the cavity the Uz profile has high differences compared with the experimental
data. The alignment between the white circles and the black line means that the
magnitude values are in accordance with the Uz ones, meaning that in this simulation
there are basically no turbulent movements along the x and y directions.
h=0.96m h=1.91m h=2.5m
h=4.36m h=4.7m h=5.15m
Figure 6.9 Case C: horizontal section of velocity distribution along the z axis (Uz).
86
6.1
The coloured sections highlight that the airflow movement occurs mainly against the
internal side of the cavity, while on the other side it moves very slow until 2.5m of
height. The sections of Figure 6.10 clearly show the laminar and fast flow that is
established inside the cavity. This output is opposite to what have been observed in
case B, where the flow was highly turbulent. The explanation for this phenomenon
could be that, while in case B the temperature of the glass surfaces is homogeneous
between the lower and higher part, in case C there is a thermal difference between
the lower glass and the higher glass that enhances the vertical flow of air.
Figure 6.10 Case C: vertical sections profiles of air velocity (magnitude values) with
87
6
Model validation
Temperature
Chart 6.6 Case C: Temperature profile of horizontal sections. Black circles represent
As for the previous cases, the CFD temperature profile has a very good degree of
agreement with the data measured by the thermocouples. Being the flow mostly
88
6.1
laminar, there is almost no difference between the left, right and middle temperature
profiles.
h=0.1m h=1.5m h=2.5m
h=3.5m h=4.5m h=5.5m
Figure 6.11 Case C: horizontal section of temperature distribution.

As in the previous case, even if the velocities very high, there is a little thermal
gradient on the z axis (from 15.1°C to 16°C) because the glazing temperatures are
close to the external one.
89
6
Model validation
Figure 6.12 Case C: vertical section profile of temperature.
90
6.1
6.1.4 Case D
Air velocity
Chart 6.7 Case D: simulated air velocity on the z axis (Uz) for each horizontal section.
91
6
Model validation
Case D is characterised by a velocity profile that is similar to the one of case A with
high values on the borders, typical of natural convection boundary layers, and
stagnant bulk. Moreover, as in the previous case, the white circles of the chart are on
the black lines indicating that the flow is mostly laminar. In this case, the predictions
of the CFD model present errors up to 0.4 m/s especially in the highest part. It is
observed that the CFD model overestimates the real values close to the internal
surface, while there is a very good level of agreement close to the external skin.
h=0.96m h=1.91m h=2.5m
h=4.36m h=4.7m h=5.15m
Figure 6.13 Case D: horizontal section of velocity distribution along the z axis (Uz).
92
6.1
Figure 6.13 shows that the air flow slows down close to the mullions and transoms,
whereas its peak values are the ones where it is in contact with the glass. The other
observation that can be drawn is that there is no negative value except for the top
section meaning that the flow is directed upwards. This is accordance with the
characteristics of case D, whose buoyancy and wind pressure have the same sign
and contribute to generate an upstream flow.
Figure 6.14 Case D: vertical sections profiles of air velocity (magnitude values) with
93
6
Model validation
Temperature
Chart 6.8 Case D: Temperature profile of horizontal sections. Black circles represent
Even though there are spotted errors on the velocity profile, the simulated
temperature at the position of thermocouples has errors of about 0.2°C. Differences
in the left, middle and right temperature profiles can be noticed in the top and
94
6.1
bottom sections, while in the middle of the cavity there is no difference between the
three charts.
h=0.1m h=1.5m h=2.5m
h=3.5m h=4.5m h=5.5m
Figure 6.15 Case D: horizontal section of temperature distribution.

Except for the last section, where some hotter spots can be noticed, all the sections
have homogeneous temperature values rising from 14.6°C up to 25°C.
95
6
Model validation
Figure 6.16 Case D: vertical section profile of temperature.
6.2 Simplified model vs. experimental data
An experimental validation of the thermal and fluid dynamic model of the ventilated
channel, using the glass surface temperatures as boundary conditions, was already
presented by Dama at al. (2017). As said in the previous chapters, this work focuses
on upgrading the thermal network model to accept as input the environmental
temperatures and the solar irradiance allowing to calculate the convective and
radiative (long wave) heat transfer through the DSF towards the interior. In this way,
the model is able to solve the problem using only the data that are available from the
common weather data files. The exchanged heat flux strictly depends on the surface
temperature of the internal glass of the inner skin (double glazed unit) and is the
most important parameter to be evaluated from the building energy simulation point
of view. However, all the outputs are reported for the sake of completeness.
96
6.2
Simplified model vs. experimental data
6.2.1 Solar radiation and flow rate
Chart 6.9 Experimental values (black dots) vs. model (black line) of solar radiation
impinging on the vertical façade. Red line represents the radiation entering the
building.
Chart 6.9 highlights the good prediction of the solar radiation hitting on the vertical
surface of the façade. The maximum values on the horizontal plane peak at 450 W/m2
that become around 800 W/m2 when translated to the vertical plane.
Even though there are some underestimations, such as in day 1, and some errors
when there are sensible fluctuations, such as in day 3, the solar model is very good
at predicting the amount of solar radiation that contributes to warm up the test
façade.
The red line in the chart represents the percentage of solar radiation that passes
through the air curtain and enters the building. This value is sensible both to the
optical characterisation of the glasses and to the angle of incidence of the sun. In
this case, the amount of radiation that enters varies between 40% at the sun peaks
and around 20% when the impinging radiation is at its minimum.
97
6
Model validation
Chart 6.10 Differential pressures and volume flow rates – simplified model vs.
experimental data.
Chart 6.10 shows the trends of the driving pressure and of the air flow rate during
the days of the experimental campaign. The driving pressure remains between -5
and 5 Pa except during days 7 and 8 when it goes up to 10 Pa, due to the strong wind
that hit the façade during those days. The volume flow rate ranges between 0 and
0.2 m3/s/m considering the anemometers measurements and the predictions of the
model, while, the tracer gas method has a wider range that reaches up to 0.4 m3/s/m.
The tracer gas method detected the flow rate peaks in correspondence both of the
maximum values of the modelled total differential pressure and of the negative
values of differential pressure. During the 7th and 8th day, when the driving pressure
is high, the model has a good degree of accordance with the measurements taken
from the anemometers. It can be also noticed that, during this same period, there is
a large gap between the measurements obtained with the CO2 method and the ones
obtained with the anemometers. This error could be explained by the fact that, when
the wind speed is high and orthogonal to the façade, the tracer gas is affected by
wash-out at the openings that alters the measurement.
During day 6, 9 and 10 the tracer gas method is again subjected to false detections
due to the reverse stream that is generated. The air enters from the top and exits
from the bottom but, since the CO2 injection is placed at the bottom of the cavity,
98
6.2
poor concentrations of CO2 detected at the top could mislead to the presence of high
mass flow rate.
From the 11th to the last day, the driving force is lower and so is the probability of
errors in the tracer gas method. During these days there is a remarkable degree of
accordance between the model and the CO2 method, while the anemometers
measured higher values. A possible explanation for this could be the fact that the
anemometers measure the velocity magnitude and not only its vertical component.
6.2.2 Temperatures and heat flux
Chart 6.11 Glass surface temperatures. Black dots represent experimental values
while red line are the modelled ones.
The prediction of the glass surface temperatures is not accurate during the central
hours of the day with errors up to 7°C. On the other hand, during the night the model
predictions are well aligned with the measurements.
The error concerning the temperature of the surfaces needs further studies both on
the model and on the experimental point of view. The model is sensible to the heat
transfer coefficient and to the absorbances of the glazings but this sensibility does
not seem high enough to justify such large errors. On the experimental side it should
99
6
Model validation
be evaluate the effectiveness of the shielding apparatus protecting the

thermocouples.
The same observations can be applied to Chart 6.12 that is relative to the outlet
temperature prediction.
Chart 6.12 Outlet (red lines) and inlet (dotted line) air temperatures – model vs.
measures.
Chart 6.13 Internal surface temperature of the inner skin, model (red line) vs.
measurements (white circles).
100
6.2
Chart 6.13 shows the measured and modelled internal surface temperatures during
the 15 days of the experimentation. The results show that the internal surface
temperature is predicted with acceptable accuracy, even though a general
overestimation with errors up to 1.5 °C is observed when solar irradiance heats up
the DSF. Since the heat flux entering (or exiting) the internal room is calculated with
equation (5.25), the surface temperature is a crucial parameter for its calculation.
Table 6.1. Heat exchange towards the interior. Model versus experimental results.
Daily integrated values calculated separately for hours with flux inwards (+) and
hours with flux outwards (-).
Day Qint (+) Model Qint (+) Exp. Rel error (+) Qint (-) Model Qint (-) Exp. Rel. error (-)
[Wh/m] [Wh/m] [Wh/m] [Wh/m]
1 1380 1238 11% 670 684 -2%
2 620 392 58% 723 684 6%
3 590 386 53% 708 703 1%
4 526 227 132% 850 827 3%
5 1415 1035 37% 891 900 -1%
6 481 367 31% 727 765 -5%
7 1016 790 29% 715 761 -6%
8 1309 1130 16% 784 819 -4%
9 808 697 16% 644 658 -2%
10 1857 1515 23% 753 717 5%
11 142 27 429% 1038 1176 -12%
12 0 0 - 1004 1144 -12%
13 1216 1073 13% 835 832 0%
14 1228 1040 18% 1157 1208 -4%
15 1133 988 15% 1079 1142 -6%
Table 6.1 reports the daily integrated values of the surface heat fluxes from the DSF
toward the inside of the Cube, separating the positive contributes to negative ones.
For outward heat fluxes, the relative daily errors are below 6% during night time and
around 12% in cloudy days. For inward fluxes, the average overestimation is about
25%, and reduces to 15% in the last three days of the experimentation, with more
stable solar and wind conditions.
101
6
Model validation
Chart 6.14 Glazing surface temperature and air temperature. Model versus
measurements for a single day (10th October). On the left the air and the surface
temperature of the glazing facing the ventilated cavity, while on the right the internal
surface temperature of the inner skin.
In order to have a better picture of model predictions versus measurements, Chart
6.14 shows a focus on the temperature profiles of day 10. It can be noticed that, while
the internal surface temperature is slightly overestimated, the outer surface
temperatures, facing the ventilated channel, are underestimated.
It can also be observed that, while the modelled temperatures have the same
dynamic profile, which responds instantaneously to the solar irradiance, the
measured values have more complex dynamics. The internal surface temperature
increases slower than the modelled one and the surface temperatures 1 and 2
(respectively of the outer glass and of the external surface of the inner skin)
decrease slower than modelled. This phenomenon suggests that the thermal inertia
of the system could have a non-negligible impact on the overall performance.
Component capacities are often omitted when modelling DSF performances; in this
case, it might have led to an overestimation of the heat transmitted to the inside.
The introduction of the heat capacity could increase the accuracy and the
applicability of the model for better prediction of complex dynamic processes in the
DSF.
102
6.3
Simplified model, CFD and measurements comparison
6.3 Simplified model, CFD and measurements comparison

This paragraph makes a comparison between the three sets of data in order to
evaluate in parallel both ways of modelling. The analysis is performed considering
the four cases studied with the CFD that are representative of all the flow conditions.
6.3.1 Airflow rate
The first parameter to be evaluated is the airflow rate at the different heights. The
value is constant throughout all the z axis, except at the top opening where the air is
discharged. This loss of mass can be seen only with the CFD analysis since the
simplified model calculates only a single value for the bulk velocity of the air.
However, calculating this effect is not very useful from the engineering point of view,
and so Table 6.2 reports only a single value for each of the cases. This value it is
representative of the airflow rate in the middle of the cavity and not at its
boundaries.
Table 6.2 Airflow rate comparison between simplified model, CFD and measurements
for all the four cases. Asterisk values on the tracer gas column mean that there is a
false detection of this method due to the downstream flow.
Uz [m/s] Air flow rate [m3/s/m]
Simplified CFD Simplified CFD Velocity Tracer
model (average) model profile gas
Case A 0.042 0.029 0.028 0.020 0.078 0.060
Case B -0.134 -0.091 (-)0.087 (-)0.060 0.078 0.732(*)
Case C 0.284 0.252 0.185 0.154 0.121 0.155
Case D 0.127 0.112 0.083 0.073 0.098 0.064
By observing the results it can be seen a good degree of agreement between the CFD
and the simplified model results (around 10-20%). The two models share the same
boundary condition in terms of wind differential pressure imposed at the openings
but different temperature boundary conditions. In case A where the wind differential
pressure is opposite and of the same magnitude of the buoyancy the error of both
models is probably due to the wrong dpwind boundary condition. Both models are able
to identify the condition of flow reversal that is created in case B with an error of
10%-20%. Flow reversal condition can be hypothesised with a high degree of
confidence by looking at measurement of the tracer gas which is hugely
overestimated compared with the anemometer. Chart 6.3 highlighted how the CFD
predicted with quite good accuracy the velocity profile inside the cavity. Hence, in
103
6
Model validation
this case the model calculates both the general and the detailed measure in a
satisfying way. In case C and case D, the computed values falls in the range between
the anemometers measure and the tracer gas one, meaning that it can be considered
an very good result. Moreover, the high degree of accordance between the
predictions of the two model proves that the hypothesis that were made for the
natural ventilation inside the channel of the simplified model are reasonable.
6.3.2 Vertical temperature profile

The following charts compare the vertical temperature profile of the simplified
model (blue line) of the CFD model (red line and white dots) and of the experimental
data (black dots). The CFD data is taken both as the integral value at each section
(red line) and as the local value in the position of the thermocouple (white dots).
Chart 6.15 Vertical profiles of bulk temperature for all the four cases computed by
CFD, simplified model and experimental data. The vertical coordinate is normalized
by the cavity height (H).
104
6.3
The simplified model and the CFD one have again similar trends among all the cases
with slight differences. The measured temperature close to the bottom opening is
significantly higher than the bulk temperature value because close to the irradiated
bottom of the cavity the recirculation that forms drags the hot fluid towards the
thermocouple. The simplified model seems to predict with higher accuracy the inlet
temperature because of the correction made on the inlet temperature to take into
account the pre-heating of the basement. The CFD has as a boundary condition the
temperature of the floor but this does not seem enough to heat the inlet air up to the
measured point. On the other hand, the CFD temperature profile grows faster due to
the lower air flow rate. On average, the CFD values taken in the position of the
thermocouples underestimate the measured values between 1°C and 3°C.
6.3.3 Convective heat transfer
The last comparison presented in this chapter uses the CFD model results for the
heat transfer coefficient and compares them with the ones of the simplified model.
Hence, the last comparison presented in this chapter is only between the two models
since no experimental data was available. The output of this analysis could also give
some hints to improve the simplified model.
The convective heat transfer coefficient (ℎŽ% ) can be directly extracted in the
simplified model (see Eq.(5.17)), whereas OpenFOAM® does not provide its
calculation directly. Hence, the first step to calculate the ℎŽ% is the extraction of the
heat flux exchanged by the surfaces of the DSF.
Internal surface External surface
Case A
105
6
Model validation
Case B
Case C
Case D
Figure 6.17 Exchanged heat flux between the fluid and the surface for all the four
cases. Left column represents the heat exchanged by the internal surface, while the
right column the one exchanged by the external one.
106
6.3
The plots of Figure 6.17 show that the airflow that generates inside the cavity makes
air exchange more heat with the internal façade than with the external one.
Moreover, in these charts, it is highlighted the tri-dimensionality of the model, since
the frame that is protruding from the glass layer exchanges significantly more heat
than the glass itself. This phenomenon happens because air slows down in the
proximity of the frame and so it can be more heated up.
The convective heat transfer coefficient has been calculated using this formula
𝜑
ℎŽ% ≡ (6.1)
𝑇Œ − 𝑇L
Where:
• 𝜑 [W/m2] is the heat flux per unit area exchanged by the internal and external
surfaces.
• 𝑇Œ is the linear average of the temperature of the surface at every height
(highlighted in pink in Figure 6.18)
• 𝑇L is the integral value of the temperature of the air at every section
(highlighted in green in Figure 6.18)
Figure 6.18 Visual representation of the averaging of the quantities necessary to the
calculation of hcv.
107
6
Model validation
Chart 6.16 reports the profiles of the convective coefficient hcv, for all the four cases.
Chart 6.16 Profiles of the heat flux along the external and internal glass for all the
simulated cases, as obtained by CFD, compared with the value computed by the
simplified model.
As already observed during the analysis of Figure 6.17 the heat flux is significantly
different in the transoms positions and so is the convective heat transfer coefficient.
Except for the local peaks and valleys introduced by the frame, the convective
coefficient is mainly constant throughout the whole vertical section in CFD results.
The average superficial value of the heat transfer coefficient for the CFD lies
between 4 and 4.8 W/m2K. For the simplified model the values range between 5.7 and
6.7 W/m2K Hence, for both cases the hcv value is actually only slightly dependent on
the air velocities in the considered range.
The comparison of the two models shows that, on average, the simplified model
predicts 30-40% more heat exchanged than the CFD does.
108
CHAPTER 7
PARAMETRIC ANALYSIS
The last chapter is dedicated to the parametric energy analysis of a DSF with integrated
roller blinds compared with a double and a triple glazing system equipped with internal
shadings. The simulation has been performed in four different cities chosen for their
temperature and wind conditions as well as for the four cardinal orientations. To do so, the
model presented in the previous chapter has undergone some small changes such as the
ability to accept data from .epw files and the adjustment of the thermal network to take into
account the presence of the solar shading. The double skin façade with integrated
shadings proves to be a better solution than the other two by 30-60% during the cooling
season. These values decrease to a 10-50% during the middle season. South orientation is
the one in which the DSF gives the higher benefit concerning solar gains control. On the
other hand, lower radiation values decrease the gap with the other possible solutions in the
northern façade.
In Chapter 6 the simplified model has been validated against the set of experimental
data and the CFD results giving a good feedback. Since every 30-hours-long CFD
analysis allows to simulate only one timestep of the year, the simplified model, which
takes about 60 seconds to perform a whole-year analysis, can be a useful tool in the
early design phases to assess the feasibility of a DSF in a particular climate.
Hence, this last chapter is dedicated to show the potential of the simplified model
by developing a preliminary parametric analysis of a DSF under four different
109
7
Parametric analysis
climates and for the four cardinal orientations to compare it with other envelope
solutions.
7.1 Evaluated solutions

The parametric analysis will be performed comparing three different alternatives:
1. A double skin façade with an integrated shading system
2. A triple glazing with internal blinds
3. A double glazing with internal blinds
In the context of this comparison there were not considered solutions with an
external shading system even if that would guarantee a better solar control
compared with solution 2 and 3. This comes from the fact that, in a realistic situation
of integrated building design, there would be also architectural constrains to be
taken into account rather than just the energetic performances. For instance, the
architect might want to have a whole glazed façade without any protruding parts.
The DSF is one of the appropriate alternatives to reach this target, and the other
solutions that have to be considered should have the same appearance of the DSF
when viewed from the outside. This would not be possible with external shadings.
The first solution is an upgrade of the one discussed in Chapter 6, a 6m tall DSF with
60 cm of cavity depth. The enhancement is represented by the introduction of a
movable roller blind in the middle of the cavity that controls inlet solar gains. To take
into account this improvement the thermal network at the core of the model had to
be changed. The hypothesis that was made is that, as observed in Dama et al. (2015)
the introduction of a roller blind in the cavity is comparable to splitting the channel
into two parallel channels, each of them being separated from the other until the
outlet opening area where air mixes up again.
The double channel hypothesis, found in the literature, is in agreement with the
performed CFD analysis, whose result is shown by Figure 7.1. In this image the
splitting of the airflow into two channels can be clearly observed. It is important to
state that this conclusion can be drawn only in the case of a roller blind, for other
types of shading, such as lamellas, the modelling of the air mixing in the channel is
different and so it requires further investigation using CFD and experimental data.
110
7.1
Evaluated solutions
Figure 7.1 Airflow in the DSF cavity with integrated shading. The colouring of the
image and the white vectors are referred to the air speed along the z axis.
The double-channel thermal network that takes into account the integrated shading
system inside the cavity is depicted in Figure 7.2.
correspond to the glass positions. ϕ1, ϕ2 , ϕ3 and ϕ4 correspond to the solar radiation
absorbed respectively by the glass of the outer skin, the external glass of the inner
skin, the internal glass and the shading. Glass panes are represented in blue.
The main difference between with the not-shaded model is the addition of the node
at temperature Ts3. It represents the roller blind, which absorbs part of the solar
radiation and introduces it in the network in the form of ϕ4.
111
7
Parametric analysis
Table 7.1 sums up the characteristics of the shading system selected for the purpose
of this analysis.
Table 7.1 Shading system characteristics.

Typology Roller blind Solar transmittance 8%
OF 5% Solar absorbance 31%
Thickness 0.77 mm Solar reflectance 61%
Colour Grey Visual transmittance 7%
Figure 7.3 Selected solar

shading.
Table 5.1 and Table 5.2 reported the optical properties of the DSF without shading
system, while Table 7.2 and Table 7.3 report the optical properties of the DSF with
the shading system for direct and diffuse radiation computed with LBNL Window:
Table 7.2 Optical properties of DSF with shading system for direct radiation
α1 α2 α3 α4 τsol
0° 0.2 0.503 0.008 0.007 0.03
10° 0.201 0.502 0.008 0.008 0.031
20° 0.204 0.501 0.008 0.008 0.03
30° 0.209 0.498 0.008 0.008 0.029
40° 0.215 0.491 0.008 0.008 0.028
50° 0.222 0.478 0.008 0.008 0.026
60° 0.226 0.449 0.008 0.009 0.023
70° 0.223 0.381 0.008 0.008 0.016
80° 0.198 0.227 0.005 0.004 0.006
90° 0 0 0 0 0
Table 7.3 Optical properties of DSF with shading system for diffuse radiation
α1 α2 α 3 α4 τsol
Diffuse radiation 0.213 0.452 0.008 0.008 0.025
Where α1, α3 and α4 are the absorbances of the three glazing panes, α2 is the
absorbance of the shading system and τsol is the total solar transmittance of the
whole system.
The other two technical solutions that are evaluated for the sake of comparison are
a triple glazing and a double glazing, both having the same shading system
presented in Table 7.1 placed inside the internal ambient. The characteristics of the
triple glazing and of the double glazing are given in Table 7.4 and in Table 7.5.
112
7.2
Chosen climates
Table 7.4 Properties of window system with triple glazing and internal roller blind.
Thickness 4-16-4-16-4
Gas 90% Argon - 10% Air
Low-e coating Face 5 (emissivity 0.038)
U [W/m2K] 1.005
gdir,sha (0°) 0.434
gdiff,sha 0.425
gdir,nosha (0°) 0.559
gdiff,nosha 0.477
Table 7.5 Properties of window system with double glazing and internal roller blind.
Thickness 4-16-4
Gas 90% Argon - 10% Air
Low-e coating Face 3 (emissivity 0.038)
U [W/m2K] 1.456
gdir,sha (0°) 0.460
gdiff,sha 0.453
gdir,nosha (0°) 0.617
gdiff,nosha 0.545
7.2 Chosen climates
To perform the parametric analysis, the code of the simplified model has been
slightly changed so that, instead of reading as inputs the weather data related to the
experimental values, it is able to read the inputs (temperature, pressure,
geopositioning data, solar radiation and wind speed) from a generic .epw type of file.
This makes it able to analyse every climate whose weather data has been recorded.
The three envelope alternatives will be evaluated under four different climates
chosen for their temperature and wind conditions. The cities under evaluation are
Milan, Athens, Paris and Berlin.
A different comfort temperature zone has been calculated for each climate following
the prescriptions of the PMV (Predicted Mean Vote) model. This experimentally
derived algorithm takes into account dry bulb temperature, humidity, air velocity and
metabolic activity for the computation of comfort internal temperatures.
113
7
Parametric analysis
7.2.1 Milan
Milan’s climate is a temperate one with cold winters that reach -5°C and hot
summers with a maximum temperature of 33°C. Given this type of climate, the
boundaries of comfort temperature have been set between 22.3°C and 26.7°C. Wind
speed is very low in Milan with an almost constant value floating around 1 m/s.
Chart 7.1 Milan – whole year temperature profile. Red line represents the monthly
average, the grey line represents hourly values, while the yellow rectangle is the
comfort temperature zone.
Chart 7.2 Milan – whole year wind speed profile. Blue line represents monthly
average, while grey line represents hourly values.
114
7.2
Chosen climates
7.2.2 Athens
Athens climate is the hottest between the chosen ones with average temperatures
in winter around 10°C and very hot summers reaching up to 35°C. Wind speed is
higher than Milan with average velocities between 3 m/s and 4 m/s. The comfort
range has been calculated between 23.8°C and 26.7°C.
Chart 7.3 Athens – whole year temperature profile. Red line represents the monthly
Chart 7.4 Athens – whole year wind speed profile. Blue line represents monthly
115
7
Parametric analysis
7.2.3 Paris
Paris climate is similar to Milan’s temperate one with the exception that its wind
conditions are stronger thus giving a better environment for natural ventilation to
happen. Comfort temperature range is the same as Milan, which is between 22.3°C
and 26.7°C.
Chart 7.5 Paris – whole year temperature profile. Red line represents the monthly
average, the grey line represents hourly values while the yellow rectangle is the
Chart 7.6 Paris – whole year wind speed profile. Blue line represents monthly
average while grey line represents hourly values.
116
7.2
Chosen climates
7.2.4 Berlin
Berlin’s climate is the coldest one with rigid winters reaching -10°C and temperate
summers. Low temperatures pair with quite strong wind conditions reaching up to
50 km/h. Comfort temperature range has been calculated between 19.3°C and
25.3°C.
Chart 7.7 Berlin – whole year temperature profile. Red line represents the monthly
Chart 7.8 Berlin – whole year wind speed profile. Blue line represents monthly
117
7
Parametric analysis
7.3 Results
The results are given in terms of the monthly average inward heat flux entering the
building during working hours (8-18).
Since the purpose of the analysis is to perform a preliminary comparative analysis
of the DSF solar control potential independently from the building case, the
hypotheses made regarding the building were reduced at the minimum. Seasons
definitions are introduced to module the internal conditions and the shading control
in order to simulate a realistic coupling between them and the external conditions.
Chart 7.9 Monthly average external temperatures in the four cities.
The method to define the seasons in which the HVAC should work consists in
computing the external air temperature monthly average during working hours (8-
18) for each city and then test it against two temperatures (an upper limit and a lower
limit). When the monthly average is lower than the lower limit, that in this case was
set to 13°C, the period is considered a heating one. The longest heating seasons are
in Paris and Berlin where it lasts until April and restarts in September, while in Athens
the heating period is between December and March.
118
7.3
Results
On the other hand, when the monthly average is higher than the upper limit (set to
18°C), that period is considered to be a cooling one. All the other cases that fall in-
between 13°C and 18°C are in the middle season where there is no need for heating
and cooling needs might be completely avoided with the adoption of an effective
solar control.
Thirteen and eighteen have been selected after a qualitative analysis of the four
curves that led to understand where the HVAC system activity period could
reasonably fall.
The internal temperature of the room is dependent from the season in the following
way:
• When in heating season T internal is set at the lower comfort boundary
• When in cooling season T internal is set at the higher comfort boundary
• When in middle season the internal temperature is placed in the middle of
the comfort boundaries
The heat flux is confronted with the hypothesised heating (red) and cooling (blue)
seasons. Theoretically, the best envelope solution is the one that is able to maximise
the heat gains during winter while reducing as much as possible the heat gains
during cooling season.
Before going through the results of the annual simulations, paragraph 7.3.1 reports
the results of daily simulations performed in Milan that are useful to understand the
functioning of the solar shading.
The control strategy that opens or closes the shading depends upon the season. In
winter, when solar gains are beneficial, the model leaves the shading up when the
solar radiation hitting the façade is lower than 300 W/m2, while after this threshold
is overcome it automatically starts to close. The shading system is fully closed when
the impinging radiation reaches 500 W/m2. During springtime, the boundaries are
set to 200 W/m2 and 400 W/m2, while in summer they are reduced again, 100 W/m2
and 300 W/m2.
119
7
Parametric analysis
7.3.1 Daily simulations
control activates at 200 W/m2) – daily focus on 21st March.
120
7.3
Results
control activates at 100 W/m2)– daily focus on 21stJune.
121
7
Parametric analysis
Chart 7.10 and Chart 7.11 report the results of daily simulations performed in Milan
during the 21st March and the 21st June for all the orientations. Particularly, they show
the behaviours of the DSF with and without shading during the day comparing it with
the solar radiation hitting the façade.
On the 21st March, Milan has just got out from the heating season to enter the middle
one. Hence, the shading system starts to be used at 200 W/m2 and is fully closed at
400 W/m2. The four charts show how the shading is necessary only when facing
south, because the other orientations almost do not reach 200 W/m2 and, when they
do, it is only for a short period of time. Hence, the double skin façade with integrated
shading behaves in the same way as the not-shaded one. On the other hand, the
south orientation chart clearly highlights the activation of the shading system. At
200 W/m2 the roller blind starts to go down and so does the inlet heat flux, while the
not shaded DSF curve keeps on increasing following the trend of the solar radiation.
At 11 a.m. o’clock, the shading is fully closed, and so the entering flux starts growing
again until 1 p.m. when it starts reducing again accordingly to the solar radiation
trend.
During summer, the impact of the solar shading is much more prominent since the
radiation values are higher and the threshold for shading activation is lower. The
inlet flux is up to 8 times lower in the shaded DSF for south, east and west
orientations, whereas for the north one it is half the value, which is still a very good
improvement.
122
7.3
Results
7.3.2 Yearly simulations
Summary of results
Before going through the detailed results, Table 7.6 and Table 7.7 aim to give a
complete overview of the results of the parametric analysis performed on the four
climates for every orientation. A performance value, namely a single number that
defines whether the DSF is better or worse than the compared strategy at controlling
the unwanted inward heat flux in the middle and in the cooling season, has been
calculated for each orientation in every climate. The performance value is expressed
in terms of percentage difference of the integral entering flux between the DSF and
the other two technologies during the considered period. The formula for the
performance value is given by Eq. (7.1):
𝑄BCM,¹º»
𝑝𝑒𝑟𝑓𝑜𝑟𝑚𝑎𝑛𝑐𝑒 𝑣𝑎𝑙𝑢𝑒 = 1 − (7.1)
𝑄BCM,½Ÿ‹aBC½
Positive percentages mean that the DSF is better than the considered strategy at
keeping out the solar gains by the corresponding value on the table whereas
negative values mean that the not-ventilated system is better in that situation.
𝑄BCM,¹º» is calculated as in Eq. (7.2):
𝐴CgŒN‹ 𝜏CgŒN‹ (𝑡) + 𝐴ŒN‹ 𝜏ŒN‹ (𝑡)

𝑄BCM,¹º» = 𝐴 · 𝐺"ŒL + ℎŽP,BCM (𝑡) 𝑇ŒB (𝑡) − 𝑇BCM (𝑡) 𝑑𝑡 (7.2)
?OPBg" 𝐴
Where 𝐴CgŒN‹ is the area of façade not shaded, 𝜏CgŒN‹ is the solar transmittance of
the DSF system without shading, 𝐴ŒN‹ is the area of the shaded façade, 𝜏ŒN‹ is the
solar transmittance of the DSF system with shading, 𝐴 is the total area of the façade,
𝑇ŒB is the temperature of the the internal glass, 𝑇BCM is the internal temperature and
𝐺"ŒL is the amount of radiation hitting on the vertical façade.
𝑄BCM,½Ÿ‹aBC½ is calculated as in Eq. (7.3):
𝐴CgŒN‹ 𝑔CgŒN‹ (𝑡) + 𝐴ŒN‹ 𝑔ŒN‹ (𝑡)

𝑄BCM,½Ÿa½ = 𝐴 · 𝐺"ŒL + 𝑈½Ÿa½ (𝑡) 𝑇O•M (𝑡) − 𝑇BCM (𝑡) 𝑑𝑡 (7.3)
?OPBg" 𝐴
123
7
Parametric analysis
Where 𝐴CgŒN‹ , 𝐴ŒN‹ , 𝐺"ŒL , 𝑇BCM and 𝐴 have the same meaning as before, 𝑔ŒN‹ and 𝑔CgŒN‹
are the g-values of the shaded and of the not shaded window, the solar
transmittance of the DSF system without shading, 𝑇O•M is the external temperature
and 𝑈½Ÿa½ is the thermal transmittance of the glazing system. 𝑄BCM,½Ÿ‹aBC½ is calculated
both for the double and the triple glazing in order to draw a performance value
relative to both these two technologies.
The integration period is the cooling one for the values presented in the tables, while
it is the single month in the charts presented in the following pages.
Since the performance values depend on the entering flux they are consequently
dependent on the control strategy and so they do not represent an absolute value.
For this reason, they can be compared only if the employed control strategy is the
same in all the evaluated solutions as it is in the case of these simulations.
Table 7.6 Summary of DSF performances against double and triple glazing during the
middle season.
DSF vs. Double glazing DSF vs. Triple glazing
S W N E Avg. S W N E Avg.
Milan 61% 43% 34% 49% 47% 43% 6% -33% 19% 9%
Athens 65% 53% 26% 57% 50% 50% 16% -75% 31% 6%
Paris 57% 49% 33% 52% 48% 35% 17% -33% 25% 11%
Berlin 57% 45% 25% 44% 43% 37% 10% -7% 6% 12%
Table 7.7 Summary of DSF performances against double and triple glazing during the
cooling period.
DSF vs. Double glazing DSF vs. Triple glazing
S W N E Avg. S W N E Avg.
Milan 71% 60% 51% 63% 61% 60% 42% 14% 47% 41%
Athens 68% 66% 36% 67% 59% 55% 50% -55% 52% 26%
Paris 69% 64% 51% 45% 57% 56% 45% -16% 18% 26%
Berlin 69% 66% 51% 66% 63% 57% 48% -5% 46% 37%
124
7.3
Results
Milan
Chart 7.12 Milan – Parametric analysis of DSF for four different orientations.
125
7
Parametric analysis
The parametric analysis in Milan’s case shows that a double skin façade with
integrated shadings can be a good solution in this climate. It can cut off internal
gains by 60% during summer when oriented to the south and by 40% for east and
west orientations compared to a triple glazing. These values increase by another 15%
when compared with the double glazing performances. The benefit of having a DSF
decreases during the middle season, especially compared with the triple glazing
where the average value is just a 9% in favour of the DSF.
All in all, the ventilation and shading effects combined together prove to be a better
solution in this kind of climate. Economic, architectural, acoustic and daylighting
considerations should be performed for the northern façade since the performances
are quite similar to the ones of the triple glazing during summer and are less
performing in the middle season. Therefore, the choice of having a DSF façade with
an integrated shading on the northern orientation should be carefully evaluated
considering other factors besides the energetic one.
To sum up, the simplified model states that, on average, a DSF with integrated
shading system in Milan is respectively 40% and 60% better at keeping out the heat
flux during summer, compared to a triple and a double glazing system with internal
roller blind, while these values decrease to a 45% and 10% during the middle season.
126
7.3
Results
Athens
Chart 7.13 Athens – Parametric analysis of DSF for four different orientations.
127
7
Parametric analysis
In Athens’ hot climate there is a longer cooling period that spans between April and
October. During this period, the DSF is approximately 26% better of the triple glazing
during the cooling season and just 6% better during the middle seasons. On the other
hand, the use of a DSF oriented towards the north in Athens proves to be a worse
choice from the energetic point of view than a triple glazing both in the summer and
in the middle season where it is 75% less performing. This happens because at
Greece’s latitude the northern side almost never reaches the threshold for shading
activation meaning that the DSF behaves as if it was not shaded throughout the
whole period. Hence, it behaves like the triple glazing whose insulating power is a
little lower because the external wind speed is not enough to prevent heat from
going inside. Being the cooling season far more big than the middle one, the DSF can
be still considered a good choice in Athens.
Another interesting observation regards the behaviour in September for the south
orientation. The double and triple glazing systems have a peak in the flux, while the
DSF does not. This can be explained by looking at Chart 7.14, that shows that 300
W/m2 are reached in July, meaning that the solar shading is being fully used, but the
solar radiation keeps on increasing until September. In the DSF, this increment of
the solar radiation is balanced by the heat flux going out due to the internal-external
delta of temperature, whereas in the not-ventilated system there is no balancing and
so the inlet flux increases during that period.
Chart 7.14 Athens - Yearly profile of solar radiation hitting the south façade.
128
7.3
Results
Paris
Chart 7.15 Paris – Parametric analysis of DSF for four different orientations.
129
7
Parametric analysis
In Paris’ cold climate the differences between the double skin and the triple glazing
are less evident, especially for east and north orientation. The DSF is still better than
the double glazing by 60% while, when compared with a triple glazing, the benefit is
less evident. The middle season is again the one in which the gap between the triple
glazing and the DSF is the lowest. On average the DSF is just 11% better than the triple
glazing for 25% of the year (which is the length of the middle season in Paris).
Such observation should give rise to a careful analysis to evaluate whether the DSF
is a cost effective solution since its energy savings are not that important when
compared to a triple glazing.
On average, the output of the simplified model is that a DSF with integrated shading
system in Paris is respectively 26% and 57% better at keeping out the heat flux during
summer compared to a triple and a double glazing system with internal roller blind,
whereas these values decrease to 11% and 48% during the middle season.
130
7.3
Results
Berlin
Chart 7.16 Berlin – Parametric analysis of DSF for four different orientations.
131
7
Parametric analysis
Berlin has a similar behaviour than Paris since the two climates are quite similar.
South, East and West orientations behave better by 50% when confronted with a
triple glazing, and by 65% in comparison with the double glazing. These values drop
by 20-30% when considering the middle season. On the other hand, north orientation
performs worse when equipped with a DSF with an average of -6% during the
considered periods.
132
CHAPTER 8
CONCLUSIONS
Double skin façades are a technology on which engineers can rely upon while
creating energy efficient transparent buildings. This study aimed at addressing the
complex problem of modelling the energy behaviour of a naturally ventilated double
skin façade. To do so, a CFD and a simplified model were developed, and then
validated against an experimental dataset obtained from a full-scale test facility. The
former considers a 3D environment for the DSF cavity, while the other is based on a
one-dimensional temperature profile along the vertical channel.
The two models were tested under four different cases representing different
combinations of thermal and pressure boundary conditions. Concerning the air flow
rate, the results showed that the predictions given by the two models are in good
agreement, with the simplified model giving mass flow rate values 10-20% higher
than CFD. The two models share the same boundary condition in terms of wind
differential pressure imposed at the openings. Nevertheless, the temperature
boundary conditions are quite different between the two models because the CFD
model has as input the surface glass temperatures, taken from the experimental
data, whereas the simplified model inputs are the environmental temperatures and
the solar irradiance. Therefore, the fact that the predictions of the two agree
suggests that the hypotheses introduced in the simplified model are reasonable.
Focusing one by one on the four cases that were evaluated during the analysis, it
turned out that:
133
8
Conclusions
• In case A, where the wind differential pressure is opposite and of the same
magnitude of the buoyancy it was found the highest differences between the
models and the measurements
• In case B, both models predict a flow reversal condition. The experimental
measurements seem to justify this output.
• In case C and D, the CFD has a difference of 5-15% compared to the tracer gas
and of 30-35% compared to the velocity profile
The two models shared also a similar trend in the prediction of the vertical
temperature profile with slight differences. The simplified model predicts better the
inlet temperature by using a correction that takes into account the thermal
exchange between the inlet air and the basement. On the other hand, the vertical
profile of the CFD grows faster because of the lower air flow rate.
The comparison of the heat transfer convective coefficients showed that the
simplified model overestimates by 30-40% the value computed by the CFD and that,
with such low air velocities, this parameter is not highly dependent on the speed of
air inside the cavity.
When tested alone against the whole experimental dataset, the simplified model
predicted the surface internal temperature with good accuracy over the whole
period of the experimental campaign, except for some spotted overestimations
when solar irradiance heats up the DSF. This values led to an overestimation of the
heat transmitted inwards of about 15%, when considering the three days with more
stable wind and irradiance, and around 25% averaging over the 15 days of the
experimental campaign. The analysis of the results of the simplified model results
suggested also the presence of dynamic effects, which might be associated to the
DSF component capacities. This could be one of the next steps to be performed in
order to make the model more accurate.
The CFD model takes approximately 30 hours to perform the analysis of a single
timestep, while the simplified model is able to simulate the behaviour of the DSF over
a year in about 60 seconds. Given that, it is clear that the simplified model is a
suitable tool for preliminary parametric analyses because it can be used to assess
whether the DSF is the best choice under variable boundary conditions.
134
The final part of this work presents an example of how the simplified model could be
used to perform a parametric study on a DSF, integrating a shading device, for
different climates: Milan, Athens, Paris and Berlin. The purpose of the analysis was
to perform a preliminary comparative analysis of the DSF solar control potential. Two
solutions have been used as reference cases, a triple and a double glazing system,
both equipped with an internal roller blind. The results were presented in terms of
total heat flux transmitted to the inside of the building.
The results show that the DSF with an integrated roller blind reduces, over the
cooling season, the total entering heat flux by 30% compared to the triple glazing
system, and 60% compared to the double glazing system. These value decrease
respectively to 10% and 50% during the middle season. For the northern façade, the
solar thermal control performances of the DSF during the working hours are very
similar to the ones of the triple glazing system.
To conclude, the simplified model reveals to be a useful tool to quantify yearly energy
performance of a DSF compared to other solutions and thus, to support the design
optioneering choices in the early design phases of a project.
Further studies should aim at increasing the reliability, accuracy and applicability of
the simplified model. Apart from the addition of thermal capacities in the thermal
network that has already been discussed before, there are other steps that could be
performed to increase the completeness of this model. For instance, it should be
validated with different kind of shading devices integrated in the DSF. The
implementation of this feature might be also assisted by further CFD analyses to
study the mixing of air between the two vertical channels.
135
References
[1] Agathokleous R.A., Kalogirou S.A., 2016. Double skin facades (DSF) and
building integrated photovoltaics (BIPV): A review of configurations and
heat transfer characteristics. Renewable Energy, Vol.89, pp.743–756.
[2] Akbari H., Cartalis C., Kolokotsa D., Muscio A., Pisello A.L., Rossi F.,
Santamouris M., Synnefa A., Wong N.H., Zinzi M., 2016. Local climate change
and urban heat island mitigation techniques – the state of the art. Journal
of Civil Engineering and Management, Vol.22, pp.1–16.
[3] Anđelković A.S., Gvozdenac-Urošević B., Kljajić M., Ignjatović M.G., 2015.
Experimental research of the thermal characteristics of a multi-storey
naturally ventilated double skin façade. Energy and Buildings, Vol.86,
pp.766–781.
[4] Anđelković A.S., Mujan I., Dakić S., 2016. Experimental validation of a
EnergyPlus model: Application of a multi-storey naturally ventilated double
skin façade. Energy and Buildings, Vol.118, pp.27–36.
[5] Angeli D., Dama A., 2015. Modelling natural ventilation in double skin facade.
Energy Procedia, Vol. 78, pp. 1537-1542.
[6] Angelotti A., Dama A., Mazzarella L., Perino M., 2007. Validazione
sperimentale di un modello per facciate a “doppia pelle” in ventilazione
meccanica. Proceedings of 62nd National Congress ATI, Vol. 1, pp. 306-311
(in Italian).
[7] Baldinelli G., 2009. Double skin façades for warm climate regions: Analysis
of a solution with an integrated movable shading system. Building and
Environment, Vol.44, pp.1107–1118.
[8] Balocco C., 2004. A non-dimensional analysis of a ventilated double façade

energy performance. Energy and Buildings, Vol.36, pp.35–40.
[9] Balocco C., Colombari M., 2006. Thermal behaviour of interactive

mechanically ventilated double glazed façade: non-dimensional analysis.
Energy and Buildings, Vol.38, pp.1–7.
137
[10] Barbosa S., Ip K., 2014. Perspectives of double skin façades for naturally
ventilated buildings: A review. Renewable and Sustainable Energy Reviews,
Vol.40, pp.1019–1029.
[11] Chen Q., 1997. Computational fluid dynamics for HVAC-successes and
failures. ASHRAE Transactions, Vol.103, pp.178-187.
[12] Chen Q., Srebric J., 2002. A procedure for verification, validation, and
reporting of indoor environment CFD analyses. HVAC&R Research, Vol.8,
pp.201-216.
[13] Chiu Y., Shao L., 2001. An investigation into the effect of solar double skin
façade with buoyancy-driven natural ventilation. Proceedings of the CIBSE
National Conference, London, UK, p. 9.
[14] Chou S.K., Chua K.J., Ho J.C., 2009. A study on the effects of double skin
façades on the energy management in buildings. Energy Conversion and
Management, Vol.50, pp.2275-2281.
[15] Churchill S.W., Chu H.H.S., 1975. Correlating equations for laminar and
turbulent free convection from a vertical plate. International Journal of Heat
and Mass Transfer, Vol. 18 No. 11, pp. 1323-1329.
[16] Colebrook C. F., White C. M., 1937. Experiments with fluid friction in
roughened pipes. The Royal Society.
[17] Colombari M., 2002. Interactive wall tests – Test Rooms 5-12 –preliminary
results of full scale monitoring. Internal Report Permasteelisa Research &
Engineering.
[18] Costa M., Aceves O., Sen F., Platzer W., Haller A., Indetzki M., Ojanen T.,
2000a. Analysis of multifunctional ventilated facades. An European Joule
Project. Proceedings of Eurosun Conference, Copenhagen, Denmark.
[19] Costa M., Oliva A., Soria M., Faggembauu D., 2000b. Optimal design of
multifunctional ventilated facades. Publishable Final Report. European
Commission, Non Nuclear Energy Programme JOULE III.
[20] Costola D., Blocken B., Hensen J.L.M., 2009. Overview of pressure
coefficient data in building energy simulation and airflow network
programs. Building and Environment. Vol. 44, pp. 2027-2036.
[21] Daffa’alla A., Betts P., 1991. Experimental study for turbulent natural
convection in a tall air cavity. Report TFD/91/6, UMIST, UK.
[22] Dama A., Angeli D., 2015. Modelling mechanically ventilated double skin
facades with integrated shading device. Proceedings of BS2015: 14th
Conference of International Building Performance Simulation Association,
Hyderabad, India, Dec. 7-9, 2015, pp.883–890.
138
[23] Dama A., Angeli D., Larsen O.K., 2017. Naturally ventilated double skin facade
in modelling and experiments. Energy and Buildings, Vol. 144, pp. 17-29.
[24] De Gracia, A., Castell, A., Navarro, L., Oró, E., Cabeza, L. F., 2013. Numerical
modelling of ventilated facades: A review. Renewable and Sustainable
Energy Reviews, Vol. 22, pp. 539–549.
[25] Ding W., Hasemi Y., Yamada T., 2005. Natural ventilation performance of a
double skin façade with a solar chimney. Energy and Buildings, Vol.37,
pp.411–418.
[26] Duffie J. A., Beckman W. A., 1980. Solar Engineering of Thermal Processes.
John Wiley, New York.
[27] Elarga H., De Carli M., Zarrella A., 2015. A simplified mathematical model for
transient simulation of thermal performance and energy assessment for
active façades. Energy and Buildings, Vol.104, pp. 97-107.
[28] Etheridge, D. 1996. Building ventilation: theory and measurement.

Chichester : John Wiley.
[29] Faggembauu D., Costa M., Soria M., Oliva A., 2003a. Numerical analysis of
the thermal behaviour of ventilated glazed facades in Mediterranean
climates. Part I: Development and validation of a numerical model. Solar
Energy, Vol.75, pp.217–228.
[30] Faggembauu D., Costa M., Soria M., Oliva A., 2003b. Numerical analysis of
the thermal behaviour of ventilated glazed facades in Mediterranean
climates. Part II: Applications and analysis of results. Solar Energy, Vol.75,
229–239
[31] Ferziger J.H., Perić M., 2002. Computational Methods for Fluid Dynamics.
Springer International Publishing Switzerland.
[32] Fuliotto R., Cambuli F., Mandas N., Bacchin N., Manara G., Chen, Q., 2010.
Experimental and numerical analysis of heat transfer and airflow on an
interactive building facade. Energy and Buildings, Vol.42, pp.23–28.
[33] Gan G., 1998. A parametric study of Trombe walls for passive cooling of
buildings. Energy and Buildings, Vol. 27, pp. 37–43.
[34] Ghaffarianhoseini A., GhaffarianHoseini A., Berardi U., Tookey J., Li D.H.W.,
Kariminia S., 2016. Exploring the advantages and challenges of double skin
facądes (DSFs). Renewable and Sustainable Energy Reviews, Vol.60,
pp.1052–1065.
[35] Grasshopper, 2014. Version 0.9.0076 https://www.grasshopper3d.com
139
[36] Gratia E, De Herde A., 2007. The most efficient position of shading devices
in a double skin façade. Energy and Buildings, Vol.39, pp.364–373.
[37] Guardo A., Coussirat M., Egusquiza E., Alavedra P., Castilla R.A., 2009. CFD
approach to evaluate the influence of construction and operation
parameters on the performance of Active Transparent Façades in
Mediterranean climates. Energy and Buildings, Vol.41, pp.534–542.
[38] Haase M., Marques da Silva F., Amato A., 2009. Simulation of ventilated
facades in hot and humid climates. Energy and Buildings, Vol.41, pp.361–
373.
[39] Hamza N., 2008. Double versus single skin facades in hot arid areas. Energy
and Buildings, Vol.40, pp.240–248.
[40] Hensen J, Bartak M, Drkal F., 2002. Modelling and simulation of a double skin
facade system. ASHRAE Transactions, Vol.108, pp.1251–1259.
[41] Hirunlabh J., Kongduang W., Namprakai P., & Khedari J., 1999. Study of
natural ventilation of houses by a metallic solar wall under tropical climate.
Renewable Energy, Vol.18, pp.109–119.
[42] Ince N., Launder B., 1989. On the computation of buoyancy-driven turbulent
flows in rectangular enclosures. International Journal of Heat Fluid Flow,
Vol.10, pp.110–117.
[43] Ji Y., Cook, M. J. Hanby, V. I. Infield D. G., Loveday D. L., Mei L., 2007. CFD
modelling of double skin façades with venetian blinds. Building Simulation
2007, Beijing, China, pp.1491–1498.
[44] Jiru T.E., Haghighat F., 2008. Modelling ventilated double skin facade - A
zonal approach. Energy and Buildings, Vol. 40, pp.1567–1576.
[45] Joe J., Choi W., Kwak Y.J., Huh H., 2013. Load characteristics and operation
strategies of building integrated with multi-storey double skin façade.
[46] Khalifa I., Ernez L.G., Znouda E., Bouden C., 2015. Coupling TRNSYS 17 and
CONTAM: Simulation of a naturally ventilated double skin facade. Advances
in Building Energy Research, Vol.9, pp.293–304.
[47] Kim Y.M., Kim S.Y., Shin S.W., Sohn J.Y., 2009. Contribution of natural
ventilation in a double skin envelope to heating load reduction in winter.
Buildings and Environment, Vol.44, pp.2236–2244.
[48] Kolmogorov A.N., 1942. Equations of Turbulent Motion of an Incompressible

Fluid. Physics, Vol. 6, pp. 56-58.
[49] Larsen O.K., 2007. The Cube. VENTInet, Vol.19, pp.4-5.
140
[50] Larsen O.K., Heiselberg P., 2008a. Experimental Set-up and Full-scale
Measurements in ‘The Cube'. Aalborg University, Department of Civil
Engineering. Technical Report No.34.
[51] Larsen O.K., Heiselberg P.K., 2008b. Empirical Validation of Building

Simulation Software: Modelling of Double Facades - Final Report, IEA ECBCS
Annex43/SHC Task 34 subtask E. Aalborg University, Department of Civil
Engineering. Technical Report No.27.
[52] Larsen O.K., Jensen R.L., Heiselberg P.K., 2014. Experimental Data and
Boundary Conditions for a Double skin Facade Building in External Air
Curtain Mode, Aalborg, Aalborg University.
[53] Liddament M.W., 1996. A guide to energy efficient ventilation, Warwick: Air
Infiltration and Ventilation Centre.
[54] Manz H., 2003. Numerical simulation of heat transfer by natural convection
in cavities of façade elements. Energy and Buildings, Vol.33, pp.305–311.
[55] Manz H., 2004. Total solar energy transmittance of glass double façades
with free convection. Energy and Buildings, Vol.36, pp.127–136.
[56] Manz H., Frank T., 2005. Thermal simulation of buildings with double skin
façades. Energy and Buildings, Vol.37, pp.1114–1121.
[57] Marques da Silva F., 2004. Buildings natural ventilation. Atmospheric

turbulence [PhD. thesis]. Lisboa, TPI 33, LNEC, 972-49-2020-8. (in
Portuguese)
[58] Marques da Silva F., Gomes M. G., Rodrigues A. M., 2015. Measuring and
estimating airflow in naturally ventilated double skin facades. Building and
Environment, Vol.87, pp.292–301.
[59] Mateus N.M., Pinto A., Da Graça G.C., 2014. Validation of EnergyPlus thermal
simulation of a double skin naturally and mechanically ventilated test cell.
Energy and Buildings, Vol.75, pp. 511–522.
[60] McAdams W.H. 1954. Heat Transmission. McGraw-Hill, New York, NY.
[61] McWilliams J., 2002. Review of Airflow Measurement Techniques. Lawrence

Berkeley National Laboratory, California.
[62] Mei L., Loveday D.L., Infield D.G., Hanby V., Cook M., Li Y., Holmes M., Bates
J., 2007. The influence of blinds on temperatures and air flows within
ventilated double skin façades. Proceedings of clima, WellBeing Indoors.
[63] Menter F.R., 1992. Improved Two-Equation k-ω Turbulence Models for
Aerodynamic Flows. NASA Technical memorandum 103975.
141
[64] Menter F.R., 1993. Zonal two-equation k−ω turbulence model for
aerodynamic flows. AIAA proceedings, Orlando, pp. 1993–2906.
[65] Michalak, P., 2017. The development and validation of the linear time varying
Simulink-based model for the dynamic simulation of the thermal
performance of buildings. Energy and Buildings, Vol. 141, pp.333–340.
[66] Mingotti N., Chenvidyakarn T., Woods A.W., 2011. The fluid mechanics of the
natural ventilation of a narrow-cavity double skin facade. Building and
Environment, Vol.46, pp.807-823.
[67] Moshfegh B., Sandberg M., 1996. Investigation of fluid flow and heat
transfer in a vertical channel heated from one side by PV elements. Part I.
Numerical study. Part II. Experimental study, WREC, pp.248–258.
[68] Moukalled F., Mangani L., Darwish M., 2016. The Finite Volume Method in
Computational Fluid Dynamics: An Advanced Introduction with OpenFOAM®
and Matlab®. Springer International Publishing Switzerland.
[69] Nicol F., Humphreys M., 2012. Adaptive thermal comfort: principles and
practice. London: Routledge.
[70] Oesterle E., Lieb R., Lutz M., Heusler W., 2001. Double skin Façades -
Integrated Planning. Prestel Verlag, Munich, Germany.
[71] Oliveira Panão, M.J.N., Santos, C.A.P., Mateus, N.M., Carrilho Da Graça, G.,
2016. Validation of a lumped RC model for thermal simulation of a double
skin natural and mechanical ventilated test cell. Energy and Buildings, Vol.
121, pp.92–103.
[72] Ong K.S., 2003. A mathematical model of a solar chimney. Renewable

Energy, Vol. 28, pp.1047–1060.
[73] Onishi J., Soeda H., Mizuno M., 2001. Numerical study on a low energy
architecture based upon distributed heat storage system. Renewable
Energy, Vol.22, 61–66.
[74] OpenFOAM, 2018. Version 6 http://www.openfoam.org
[75] Pappas A., Zhai Z., 2008. Numerical investigation on thermal performance
and correlations of double skin façade with buoyancy-driven airflow.
[76] Park C.S., Augenbroe G., Messadi T., Thitisawat M., Sadegh N., 2004a.
Calibration of a lumped simulation model for double skin façade systems.
142
[77] Park C.S., Augenbroe G., Sadegh N., Thitisawat M., Messadi T., 2004b. Real-
time optimization of a double skin façade based on lumped modelling and
occupant preference. Building and Environment, Vol. 39, pp.939–948.
[78] Pasut W., De Carli M., 2012. Evaluation of various CFD modelling strategies
in predicting airflow and temperature in a naturally ventilated double skin
facade. Applied Thermal Engineering, Vol.37, pp.267–274.
[79] Pérez-Grande I., Meseguer J., Alonso G., 2005. Influence of glass properties
on the performance of double-glazed facades. Applied Thermal
Engineering, Vol.25, pp.3163–3175.
[80] Poirazis H., 2006. Double skin façades: a literature review. Sweden: IEA SHC
Task 34 ECBCS Annex.
[81] Pomponi F., Ip D.K., Piroozfar D.P., 2013. Assessment of double skin façade
technologies for office refurbishments in the United Kingdom. In: Passer A.,
editor. International Sustainable Building Conference Graz. Graz 2013
[82] Pomponi F., Piroozfar P.A.E., Southall R., Ashton P., Farr E.R.P., 2016. Energy
performance of Double skin Façades in temperate climates: A systematic
review and meta-analysis. Renewable and Sustainable Energy Reviews,
Vol.54, pp.1525–1536.
[83] Radhi H., Sharples S., Fikiry F., 2013. Will multi-facade systems reduce
cooling energy in fully glazed buildings? A scoping study of UAE buildings.
[84] Rahmani B., Kandar M.Z., Rahmani P., 2012. How double skin Façade’s air-
gap sizes effect on lowering solar heat gain in tropical climate? World
Applied Science Journal, Vol.18, pp.774–778.
[85] Saelens D., 2002. Energy performance assessment of single storey

multiple-skin façades. PhD Thesis. Catholic University of Leuven, Belgium.
[86] Saelens D., Carmeliet J., Hens H., 2003. Energy performance assessment of
multiple-skin facades. HVAC&R Research, Vol.9, pp.167–185.
[87] Saelens D., Roels S., Hens H., 2004. The inlet temperature as a boundary
condition for multiple-skin façade modelling. Energy and Buildings, Vol.36,
pp. 825-883.
[88] Saelens D., Roels S., Hens H., 2008. Strategies to improve the energy
performance of multiple-skin facades. Building and Environment, Vol.43,
pp.638–650.
[89] Safer N., Woloszyn M., Roux J.J., 2005. Three-dimensional simulation with a
CFD tool of the airflow phenomena in single floor double skin facade
equipped with a venetian blind. Solar Energy, Vol.79, pp.193-203.
143
[90] Shameri M., Alghoul M., Sopian K., Zain M.F.M., Elayeb O., 2011. Perspectives
of double skin façade systems in buildings and energy saving. Renewable
and Sustainable Energy Reviews, Vol.15, pp.1468–1475.
[91] Sharples S. 1984. Full-scale Measurements of Convective Energy Losses

from Exterior Building Surfaces. Building and Environment, Vol. 19, No. I, pp.
31-39.
[92] Stec W., Paassen D.V., 2005. Sensitivity of the double skin facade on the
outdoor conditions. Proceedings of the international conference on air
quality and climate. Indoor Air Beijing.
[93] Swinbank W.C., 1963. Long-wave radiation from clear skies. Quart. J. R.
Meteorol. Soc., Vol. 89, pp. 339-348.
[94] Tanimoto J., Kimura K., 1997. Simulation study on an airflow window system
with an integrated roll screen, Energy and Buildings, Vol.26, pp.317-325.
[95] Tarazi N. K., 1991. A model of a Trombe wall. Renewable Energy, Vol. 1,
pp.533–541.
[96] Torres M., Alavedra P., Guzmán A., Cuerva E., Planas C., Clemente R.,
Escalona V., 2007. Double Skin Façades-Cavity and Exterior Openings
Dimensions for Saving Energy on Mediterranean Climate. Proceedings
Building Simulation, pp.198–205.
[97] Versteeg H. K., Malalasekera W., 1995. An Introduction to Computational

Fluid Dynamics: The Finite Volume Method. Longman Scientific and
Technical.
[98] Von Grabe J., 2002. A prediction tool for the temperature field of double
facades. Energy and Buildings, Vol. 34, pp.891–899.
[99] Wang, Y., Chen, Y., & Zhou, J. (2016). Dynamic modelling of the ventilated
double skin façade in hot summer and cold winter zone in China. Building
and Environment, 106, 365–377.
[100] World Health Organization (WHO), 2011. Air quality and health. Geneva:
World Health Organization. Fact sheet No 313.
[101] Xamán J., Álvarez G., Lira L., Estrada C., 2005. Numerical study of heat
transfer by laminar and turbulent natural convection in tall cavities of
façade elements. Energy and Buildings, Vol.37, pp.787–794.
[102] Xu X. li, Yang Z., 2008. Natural ventilation in the double skin facade with
venetian blind. Energy and Buildings, Vol.40, pp.1498–1504.
144
[103] Xue, F., Li, X., 2015. A fast assessment method for thermal performance of
naturally ventilated double skin façades during cooling season. Solar
Energy, Vol.114, pp.303–313.
[104] Zeng Z., Li X., Li C., Zhu Y., 2012. Modelling ventilation in naturally ventilated
double skin facade with a venetian blind. Building and Environment, Vol.57,
pp.1–6.
[105] Zhou J., Chen Y., 2010. A review on applying ventilated double skin facade
to buildings in hot-summer and cold-winter zone in China. Renewable and
Sustainable Energy Reviews, Vol.14, pp.1321–1328.
145

2018 10 Dopudi PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2018 10 Dopudi PDF

Uploaded by

Copyright:

Available Formats

SCUOLA DI ARCHITETTURA URBANISTICA INGEGNERIA

DELLE COSTRUZIONI – BUILDING ENGINEERING

NATURALLY VENTILATED DOUBLE SKIN

Relatore: Dr. Alessandro Dama

Tesi di laurea di:

Anno accademico 2017/2018

NATURALLY VENTILATED DOUBLE SKIN FAÇADE: CFD AND

TABLE OF CONTENTS ................................................................................. IV

LIST OF FIGURES........................................................................................ VII

LIST OF TABLES .......................................................................................... XI

LIST OF CHARTS........................................................................................ XIII

CHAPTER 1 INTRODUCTION ........................................................................1

1.1 RESEARCH QUESTION ........................................................................................... 1

CHAPTER 2 DOUBLE SKIN FAÇADES: STATE OF THE ART ......................5

2.1 TECHNOLOGY ...................................................................................................... 5

CHAPTER 3 METHODOLOGY AND CASE STUDY .....................................33

3.1 METHODOLOGY ................................................................................................. 33

CHAPTER 4 CFD MODEL ............................................................................45

4.1 FLUID DYNAMIC MODEL ...................................................................................... 45

CHAPTER 5 SIMPLIFIED MODEL ................................................................57

5.1 MODEL DESCRIPTION ......................................................................................... 57

CHAPTER 6 MODEL VALIDATION...............................................................71

6.1 CFD MODEL: INVESTIGATION VS. MEASUREMENTS ................................................ 72

CHAPTER 7 PARAMETRIC ANALYSIS ......................................................109

7.1 EVALUATED SOLUTIONS .................................................................................... 110

CHAPTER 8 CONCLUSIONS .....................................................................133

1.1 Research question

b) Shaft-box: ventilation occurs on vertical separated channels to enhance

2.1.1 Influence of façade design

Outer skin properties

2.1.2 Influence of building design

Inner skin properties

Height of the cavity

2.1.3 Influence of site conditions

the best orientations in the northern hemisphere, regardless of surrounding

2.2 Energy performance modelling of DSF

2.2.1 Simplified models

Analytical and lumped parameters models

Airflow network analysis + Building energy simulation models

Control volume approach models

The non-dimensional analysis can be found in literature in the studies of Balocco et

2.2.2 Zonal approach

2.2.3 Computational Fluid Dynamics models (CFD)

2.3 Energy modelling of the case-study

This last paragraph is dedicated reports the results of a validation procedure

Table 2.1 Characteristics of the software employed for the validation.

3.2 Experimental test facility

Table 3.1 Geographical coordinates of the site.

Figure 3.4 Double skin façade section dimensions.

Table 3.2 U-values and emissivity properties of the glazing system.

Table 3.3 Spectral properties of the glazing system.

Table 3.4 Discharge coefficients and dimension

Figure 3.6 Definition of

3.3 Experimental setup

3.3.1 Temperature measurements

Figure 3.8 Measurement of outlet air temperature (Larsen et al. 2008a).

The uncertainty of equipment for the temperature measurement is estimated at a

3.3.2 Incident solar radiation measurements

Figure 3.10 Positioning of pyranometers in the experimental set-up (Larsen et al.

3.3.3 Airflow measurements

The airflow velocity profile is another essential measurement to be performed in

Velocity profile method

Tracer gas method