Professional Documents
Culture Documents
A R T I C L E I N F O A B S T R A C T
Keywords: Energy saving in buildings is an important measure for mitigation of climate change. There exists a large po
Model validation tential for energy saving in buildings by improving the thermal performance of windows. For decisions on energy
Uncertainty analysis saving window retrofits, accurate estimation of the energy saved and its uncertainty is of importance. The ISO
Window thermal transmittance
15099 standard, which is normative for thermal modelling of windows within the building sector, does not give
Energy saving measure
Mitigation of climate change
uncertainty estimates. The main novelty of this study was to provide uncertainty analysis for model prediction of
Benchmarking energy models the thermal transmittance of windows, in the perspective of decisions on window retrofits. For this purpose, we
proposed a new simplified model, which facilitated uncertainty analysis, and still was similar to the ISO 15099
window model. The model was validated by application of a benchmark validation procedure to a set of pre
viously performed validation experiments. Main conclusions were: (i) The model was accurate within a pre
diction uncertainty equal to 0.20 Wm− 2K− 1; (ii) The domain where the model is valid was described using
existing well-documented validation experiments. This domain was restricted to windows with glazing thermal
transmittance corresponding to 2-layer glazing, and to windows where the frame area is a minor part of the total
window area. (iii) The prediction uncertainty was mainly determined by the measurement uncertainty in the
validation experiments; (iv) If a window retrofit is based on reduction of window thermal transmittance, then
this reduction has to be larger than 0.56 Wm− 2K− 1 in order to yield energy savings above the uncertainty limit.
* Corresponding author. Department of Applied Physics and Electronics, Umeå University, S-90187, Umeå, Sweden.
E-mail address: anders.ohlsson@umu.se (K.E.A. Ohlsson).
https://doi.org/10.1016/j.rser.2022.112748
Received 8 March 2022; Received in revised form 10 June 2022; Accepted 29 June 2022
Available online 12 July 2022
1364-0321/© 2022 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
K.E.A. Ohlsson et al. Renewable and Sustainable Energy Reviews 168 (2022) 112748
(in Germany) [18]. For a typical Passive House during the heating often based on calculations using a building energy model (BEM). BEMs
period, the total heat loss is about 40 kWh m− 2yr− 1, where 25 kWh are often categorized by the number of parameters included in the model
m− 2yr− 1 is lost through the windows, 8 kWh m− 2 yr− 1 with the venti equations, i.e. the model order, and also to what extent measurement
lation air, and the remainder through roof, walls, and ground [18]. The data are used for parameter estimation. For physics (white-box) models,
heating budget is balanced by 10 kWh m− 2 yr− 1 due to internal heat the model equations are obtained solely from physics theory, while the
gains from appliances and metabolic heat from occupants, and by 15 parameters of empirical (black-box) models are estimated from mea
kWh m− 2 yr− 1 solar heat gain through windows. From these figures we surement data only [23,24]. BEM physics models often include high
conclude that ESM focused on reducing heat transfer through windows resolution building geometry which leads to a model order in the range
is of major importance for the energy performance of nZEB. Further 102-104. There are also the so called grey-box models, which are reduced
examples of the importance of windows in the design of ESM is given in order physics models, where parameter estimation is performed on a
the following: Changing from 2-layer to 3-layer glazing windows of a series of measurements [25].
residential building (Afloor = 143 m2 and Aw = 20 m2) in the cold climate To support decisions on ESM based on BEM, the uncertainty analysis
of Madison, WI, USA, resulted in an energy saving of 29 kWh m− 2 yr− 1 (UA) is useful for several reasons: (i) It provides an estimate of uΨ . (ii)
[19]. Similar window replacement in a building (Afloor = 108 m2 and Aw The UA quantifies the fractional contribution χ of various sources of
= 18 m2) in Madagan, Russia, yielded a reduction in Φ equal to 77 kWh uncertainty to uΨ (See Eq. (A5)), and thereby indicates how to reduce
⋅m-2yr− 1 [20]. Variation of the thermal transmittance of the window the uncertainty. (iii) UA is an integral part of the process of physics
frame, and its width, could change the space heating demand by as much model validation, where the agreement between model simulation re
as 4–7 kWh m− 2 yr− 1, as was calculated for a single-family nZEB with sults (S) and physical reality, represented by measurement data (D), is
Afloor = 148 m2 located in Estonia [21]. quantified. For physics models, which are based only on theory, the
The aforementioned studies [18–21] indicate that ESMs designed to validation requires that the D values are obtained from a series of vali
decrease the window thermal transmittance Uw, including heat transfer dation experiments (VE). UA for BEM has been performed for individual
in both glazing and frame, have the potential to reduce the need for buildings [9,26–30], while [31] reviewed UA for BEM applied at the
supply of space heating energy. The Uw, is often in the range 1–3 WK− 1 level of building stock.
m− 2, while for energy efficient walls, the corresponding value is nor For physics models in general, Coleman and Stern (CS) [32,33], with
mally below unity, approaching 0.2 WK− 1 m− 2. The thermal trans contributions by Oberkampf et al. [34,35], provided a framework for UA
mittance of the glazing Ug could be reduced to below 1 WK− 1 m− 2 by and model validation, which is considered as the best practice for
using multiple glass panes, low-emissivity coatings, low conductivity gas models of high-risk safety-critical processes in e.g. the aerospace in
fills, or even vacuum, in between glass panes [22]. The thermal trans dustry [30]. This framework also includes the concept of validation
mittance of the frame Uf is determined to a large extent by the thermal domain (VD), which is defined in the model parameter space as the
conductivity of the solid material used for its construction: For wood and volume spanned by existing VEs [35]. It was recently proposed that the
plastic frames, Uf is approximatively between 2 and 3 WK− 1 m− 2, while current procedures for UA and validation of BEM should be bench
for aluminum frames the corresponding interval is between 5 and 8 marked using the CS framework as comparison reference practice [30].
WK− 1 m− 2, where the latter value is determined mainly by the intro The CS framework includes estimation of the component un
duction of low conductivity thermal break materials. certainties caused by: (i) uncertainty in the input variables used in the
For an ESM focused on reducing Ug and Uf, and optimizing the model calculation (uinput), (ii) uncertainty in the numerical solution of
relative sizes of glazing and frame, the Ψ and uΨ need to be calculated. model equations (unum), and (iii) measurement uncertainty (uD) origi
The link between Uw and Ψ is given below in section 6, and the models nating in the VE. The combination of these uncertainties yields the
for calculation of Uw are introduced in section 2. Design of ESMs for validation uncertainty (uval), which is compared to the model error E (=
buildings, building components (e.g. windows), or building stocks, is S – D). If E is significantly larger than uval, then the model is in error, and
2
K.E.A. Ohlsson et al. Renewable and Sustainable Energy Reviews 168 (2022) 112748
3
K.E.A. Ohlsson et al. Renewable and Sustainable Energy Reviews 168 (2022) 112748
modelling heat transfer through frame cavities using an effective con by Eq. (8), which express the internal heat flux of the frame as:
ductivity. Two such frame models were implemented in the ANSYS (LBL ∑( )
[43]) and FRAME (Enermodal Eng. Ltd., Canada [44]) software, but the Af ,j hj (θ1 − θ2 ) = Af h(θ1 − θ2 ) (8)
j
experimental evidence for their accuracy was incompletely docu
mented. More recently, Uf has been specifically measured, and where θ1 and θ2 are the frame surface temperatures. Rearrangement of
compared to simulated values obtained from the frame models given in Eq. (8) yields:
ISO 10077-2 and ISO 15099 [45–48]. ∑
In the present work, we changed the Rubin two-solid-layer (glass h= ∑
bj hj
(9)
panes) glazing model (Section 2.2 and Appendix B) to model instead the bj
1D conductive heat flux through the frame. This allowed reuse of the
Finally, this h-value is used as input variable in the frame model for
program code for the glazing model, with the following modifications:
calculation of Uf.
(i) The air between the two solid layers was replaced by the frame
material. (ii) The frame surface emissivity was substituted for glazing
3. Review of validation experiments
emissivity. Heat transfer through the frame material is by conduction,
while radiative and convective heat transfer occurs between the outer
Table 1 lists the VE which are relevant for validation of the present
frame surfaces and the environments, as in the glazing model. The
window model. We believe that some of the VEs have been used for
models for wood and metal (Al) frames differ slightly due the large
development of the area-based window model of §4.1.4. In ISO 15099
difference in their thermal conductivities (λ). Below, the wood frame
[38] since model Eq. (2) occurs in most of them. Note that the VEs in
and metal frame models are described (see also Fig. 2).
Table 1 mainly include windows which have 2-layers glazing and Aw
around 1 m2. Therefore, the area-based ISO 15099 model may have been
2.2.1. Wood frame
insufficiently validated for smaller windows, where relatively more heat
The frame width b is divided into sections, where λ of each section is
is transmitted through the frame, and for windows with 3-layer glazing.
either homogeneous (cf. section 1 in Fig. 2a), or piecewise homogeneous
For model validation, we needed a set of well-documented VE, which
(Fig. 2a, section 2). The heat transfer coefficients h1 and h2 (Wm− 2 K− 1)
allows the assignment of values for the input variables of the window
of the solid frame sections 1 and 2, respectively, are then given as:
model. For examples, there has to be sufficient information on window
λ geometry, radiative and convective boundary conditions (e.g. the
h1 = (4)
d convective heat transfer coefficients (CHTC) hex and hin), and material
properties (e.g. ε and λ). Table 1 gives an overview of such information
1
h 2 = ∑ di (5) which was provided by the VE documentation. However, it was found
λi that some important information was missing: For example, even in the
most well-documented set of VE, that given by Elmahdy in 1990 [49],
where d is the thickness of homogeneous frame material. In the next frame material properties were missing, and the convective boundary
step, Uf,j is calculated for section j, using the frame model. The sum of conditions were given only implicitly from the results for the 1-layer
heat fluxes Q (W) through all frame sections could be expressed in two glazing window. Moreover, complementary information on frame
ways (cf. Eq. (1)): areas (Af) had to be obtained from another paper [50].
∑( )
Q= Af ,j Uf ,j (θin − θex ) = Af Uf (θin − θex ) (6)
j
4. Uncertainty estimation
where the frame section area Af,j is defined as Af,j = L.bj , and Af = 4.1. Measurement uncertainty
∑
L bj , where L is the frame length measured along its width center.
Rearrangement of Eq. (6) yields: The measurement uncertainty of the single laboratory hotbox mea
∑ surement in Ref. [51] was estimated to be ±6.5%, with Uw around 3
Uf = ∑
bj Uf ,j
(7) WK− 1 m− 2 (we derived Uw from the given surface-to-surface resistance),
bj using a Type B procedure (see Appendix A). Assuming a 95% confidence
interval and Gaussian distribution of errors, we then calculated:
2.2.2. Metal frame
0.065Uw
For the metal frame illustrated in Fig. 2b, the interior and exterior uD1 = = 0.10 Wm− 2 K− 1
(10)
2
frame surfaces each consist of one continuous piece of metal. Due to high
thermal conductivity of metal, each of these frame surfaces are at In another study, the uD for hotbox measurements was estimated
approximately isothermal conditions. For this reason, Eq. (6) is replaced using Type A evaluation (see Appendix A) where 4 different test samples
4
K.E.A. Ohlsson et al. Renewable and Sustainable Energy Reviews 168 (2022) 112748
Table 1
Validation experiments used for development of window models.
Reference Year Tech Aw Geometry CHTC FMP Layers N Model validation Comment
Klems [52] 1979 HBox 1 Yes No – 2–4 Glazing VE for Rubin glazing model [39,56]
Hollands and Wright [40] 1983 GHP 0.017 – * – (3)* Glazing VE for HW glazing model [40]
Elmahdy and Bowen [57] 1988 HBox 0.8/1.6 Yes No – 2, 3 Glazing VE for glazing model of HW (Vision)
Dubrous and Harrison [58] 1989 HBox, SC 1.1 No No Type** 2, 2, 3 3 Frame/Glazing VE for frame models; Dark and sunlit
Elmahdy [49] 1990 HBox 1.1 Yes (YES)◦ Type** 1, 2, 3 7 Frame/Glazing JCUS#
Klems and Reilly [50] 1990 MoWiTT 1.1 Yes variable Type** 1, 2, 3 7 Frame/Glazing JCUS
Elmahdy [59] 1992 HBox, SC 1.1 No No Type** 2, 3, 4 9 Frame/Glazing JCUS
Arasteh et al. [53] 1994 HBox 1.1–2.2 No No Type** 1, 2 16 Frame/Glazing##
Measurement technique (Tech): HBox = Hotbox apparatus under nighttime conditions; GHP = Guarded hot plate method; MoWiTT (Mobile Window Thermal Test) =
Outdoor facility for twin test cell measurements at LBL, Nevada, USA; SC = Solar calorimeter. Aw (m2) is the window area including glazing and frame. Geometry:
Documentation on frame/glazing geometry published? CHTC is the convective heat transfer coefficients: (*) Hot plate method was used where boundary conditions are
set by metal plates, which also correspond to layers 1 and 3. (◦ ) In the present work, CHTC was derived from the 1-layer window VE. FMP is the frame material
properties (**) Material type (e.g. wood): only was given, without any property values. Layers: Number of glazing pane (glass or plastic) layers. N: Number of window
test samples. (##) Air and argon fill gas tested, with and without low-value emissivity coating. (#) JCUS is the Joint Canadian/US project on window performance.
In an inter-laboratory comparison study, Uw for several different − 18 C Elmahdy [49] 0.3 C Uniform distribution
◦ ◦
θex
±0.5 ◦ C
windows were measured in up to 5 different laboratories (see Table 4 in
θin 21 C◦
Elmahdy [49] 0.3 C◦
Uniform distribution
Ref. [53]). We selected 19 measurement results for 5 windows tested ±0.5 ◦ C
(windows 3, 5, 7, 8, and 12, in Ref. [53], where each had been measured λwood 0.141 Kol [60] 0.020 Uniform distribution
in ≥3 laboratories), and with Uw ranging from 2.7 to 3.2 W K− 1m− 2. Wm− 1K− 1 Wm− 1 K− 1
±0.034 W m− 1K− 1
Assuming equal distribution of measurement errors for the selected λAl 215 Wm− 1K− 1
Clarke et al. 9W Uniform distribution
[61] m− 1K− 1 ±15 W m− 1K− 1
windows, we pooled their standard deviations according to the pro λspacer 0.24 Baker et al. 0.07 Uniform distribution
cedure in Ref. [54] to obtain the estimate: Wm− 1K− 1 [62] Wm− 1 K− 1
±50%
λTB 0.14 Baker et al. 0.04 Uniform distribution
uD3 = spooled = 0.19 W− 1Km− 2
(12) Wm− 1K− 1 [62] Wm− 1 K− 1
±50%
λglass 0.9 Wm− 1K− 1
Rubin [39] 0.03 Uniform distribution
With measurements performed in one single laboratory, a measure Wm− 1 K− 1
±0.05 W m− 1K− 1
ment error may be classified as being systematic because some variables εglass 0.84* Rubin [63] 0.01 Mean and s of 5
which influence the result are constant during the measurement period, reported values
0.90 Chen et al. 0.023 Not painted; Uniform
e.g. the same instrument or calibration were used throughout. When εwood
[64] distribution ±0.04
results were obtained from multiple laboratories, the former systematic εAl 0.05/0.2/0.9 Gustavsen 0.02 New polished/Aged,
measurement error may now show to be random for reasons such as use et al. [65] oxidized/Anodized
of different instruments and frequent recalibration. One possibility for εpaint 0.88 Fantucci et al. 0.02 See text.
reduction of uD is to perform difference measurements, where systematic [66]
0 Maiorov [67] For wavelengths >4
errors may cancel when two windows are measured simultaneously and
τ
μm
side-by-side [52,55]. b, d Various [49,50] 0.3 mm Uniform distribution
±0.5 mm
Af 0.124–0.296 Klems and 0.006 m2 Values given in their
4.2. Uncertainties of input variables
m2 Reilly [50] Table 4.
hex 26.7 Wm− 2 Present 1.3 Wm− 2 See note (***)
For validation of the present model, we selected a series of VE where K− 1 work** K− 1
Uw of six different windows had been measured in one single laboratory hin 4.2 Wm− 2 K− 1
Present 0.21 See note (***)
using the hotbox technique [49]. Table 2 shows the values of the input work** Wm− 2 K− 1
2
h 2.10 Wm− ElSherbiny 0.11 Values for window
variables xi, and their uncertainties ui, in this series of VE. K− 1 et al. [68] Wm− 2 K− 1 #F020 in [49]
The temperatures θex and θin were given in Ref. [49] as integer
values. Therefore, it was assumed that the errors in these values were (#) Selected from validation experiments in Ref. [49]. (*) Total hemispherical
emissivity of soda lime window glass. (**) See derivation below, using values for
represented by a uniform distribution ±a centered around the given
1-layer glazing window. (***) Gaussian distribution ±10% (95% confidence
values. In this case, a = 0.5 ◦ C (digital resolution in case of integer
level).
values), and the uncertainty could be evaluated as (cf. [69], §4.3.7):
√̅̅̅̅̅̅̅̅̅̅
uθ = a2 /3 = 0.29 ◦ C (13) the λwood –value estimated at the interval center, where MC = 11%.
Again using Eq. (13), we obtained uλwood = 0.020 Wm− 1 K− 1.
The thermal conductivity of wood (λwood), used in window frames,
Data on λAl were obtained from a review on thermal properties of
depends primarily on the wood species, the direction of heat flux rela
building materials, where doubt is expressed on the reliability and
tive to the axis of tree stem, and on the moisture content (MC). This
independency of reported values from different sources [61]. Eq. (13)
information was not documented for the frame wood material used in
was used for estimation of the uncertainty in λAl. The thermal conduc
Ref. [49]. For Pine wood (Pinus Silvestris) as frame material, and with
tivities of the foam spacer and the thermal break materials, λspacer and λTB
heat flux in the radial direction, Kol [60] measured λwood. It was found to
respectively, were unknown, and therefore estimated as in Ref. [62],
increase approximatively linear in the interval 0.141 ± 0.034 W m− 1K− 1
with the ±50% uncertainty interval. For the thermal conductivity of
for MC between 0 and 22%. Since in the present case MC was unknown,
glass, the uncertainty given in Table 2 derives from the digital resolution
we assumed a uniform distribution for MC over the interval 0–22%, with
5
K.E.A. Ohlsson et al. Renewable and Sustainable Energy Reviews 168 (2022) 112748
recently polished surfaces, εAl = 0.05. With age, oxidation in air leads to In the next step, the values of hrex and hrin were subtracted from the
an εAl-value around 0.2, and electrochemical oxidation (anodization) initial hex and hin, respectively, to yield updated values for these latter
yields εAl = ca 0.9 (cf [65]). The main uncertainty source is here the lack parameters. This calculation procedure was iterated until convergence
of knowledge on the degree of oxidation of the Al surface, while the of hr-values at hrex = 3.3 Wm− 2 K− 1 and hrin = 4.1 Wm− 2 K− 1. At this final
uncertainty in each of three values is of negligible importance, and point, we obtained hex = 26.7 Wm− 2 K− 1 and hin = 4.2 Wm− 2 K− 1, and
therefore was set arbitrary to equal 0.02 (cf. Table 2). Ug = 0.370 Wm− 2 K− 1 (close to the value obtained above with Eq. (15)).
The existence of coating on the frame surfaces is unknown, since Only for fex, the uncertainty level is given as ±10% [51], while it is
documentation was missing, although it seems a reasonable assumption unknown for the fin. Assuming that fin has the same uncertainty, and that
that at least the wooden frames were painted. The emissivity of uncoated the ±10% uncertainty interval gives 95% confidence with a Gaussian
wood, unspecified species, was around 0.90, as given from five different error distribution (±2s), we obtained the uncertainty estimates as in
sources [64]. The emissivity of paint coatings can vary across a wide Table 2.
range, as e.g. in Ref. [66] where the emissivity decreased linearly from The CHTC of the air layer between vertical plane glasses (h-value)
0.88 to 0.41, with addition of increasing amounts of low-emissivity Al were derived from the Nusselt (Nu) number as:
flake pigments in the coating.
Values of the frame area Af were obtained for four windows (#F020, h=
Nu λair
(18)
#F022, #F024, and #F062) from Ref. [50] (their Table 4). For window w
#F015 [49], Af was assumed to equal that of window #F062. The un
where w is the air layer thickness, and λair is the air thermal conductivity.
certainty estimate for Af was initially set to 0.006 m2, which corresponds
For vertical gas layers, the Nu number as a function of the aspect ratio A
to ±10% uncertainty interval at 95% confidence level for the smallest
(= [Layer Height]/w) and the Rayleigh (Ra) number is given by Eq. (5)
frames, with Af = 0.120 m2. Values of the geometry variables (b-, d-type,
in Ref. [68]. The uncertainty in Nu was expressed as a standard deviation
as shown in Fig. 2) were obtained from drawings of frame cross sections
equal to 3.6% [68]. For the 2-glass layer window #F020, with wooden
given in Refs. [49,50]. The errors of these variables were assumed to be
frame, in Ref. [49], we estimated w = 12.7 mm, A = 79.7 and Ra = 6570,
uniformly distributed within ±0.5 mm, which yields an uncertainty
using property data of air at +5 ◦ C (approximate air layer temperature)
equal to 0.3 mm. The uncertainty in b-type variables was included in the
from Ref. [71]; An iterative solution was required since Ra depend on
uncertainty of Af. The contribution from the uncertainty in d-type var
the temperature difference across the air layer. This yields the value Nu
iables was negligible in comparison to the uncertainty in λ-type vari
= 1.13 (from Ref. [68]), and finally, using Eq. (18), h = 2.10 Wm− 2 K− 1.
ables for the corresponding piece of solid material (cf. Table 2 and Eq.
This result holds also for window #F015 since its glazing is similar.
(4)).
The uncertainty of h (= uh) was then obtained by propagation of
The glazing model of Rubin [39], and the present window model,
uncertainties of input variables of Eq. (18), as shown in Eq. (A3). For λair,
require input values for the convective heat transfer coefficients (CHTC;
the relative uncertainty was estimated to 4%, based on its temperature
hex, hin). In Ref. [49], the CHTCs are not given explicitly, but in the form
dependency, while the uncertainty of w was probably around 0.8%.
of “film coefficients” (i.e. heat transfer coefficients) for the exterior and
Eventually, for window #F020 (and similarly for #F015), we calculated:
interior glazing sides, fex and fin, respectively. For the external side, the
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
steady-state heat flux qs through a 1-layer glass pane is obtained as [51]:
uh = h (3.6%2 + 4%2 + 0.8%2 ) = 5.4%h = 0.11 Wm− 2 K− 1 (19)
qs = (hex + hrex )(θ1 − θex ) = fex (θ1 − θex ) (14)
5. Application of the benchmark validation procedure
where hrex is the infrared radiative component of fex, and θ1 is the
temperature of the exterior glass surface. With a similar definition of fin, In this section, the present window model was validated using the
we obtain from Ref. [51]: benchmark CS-procedure for validation described in Ref. [30], which is
1 1 d 1 based on the works by Coleman et al. [32,33] and by Oberkampf et al.
= + + (15) [34,35]. The VEs were selected from Ref. [49]. In the following we
Ug fin λ fex
describe and discuss the results for each procedural step as shown in
where d is thickness of the glass layer, and λ is its thermal conductivity. Fig. 3.
Assuming d = 3 mm and λ = 0.9 Wm− 1 K− 1 (value for glass given in In step 1, the uncertainties uD, unum, and uinput were estimated. For uD,
Ref. [39]), and with fin = 8.3 Wm− 2 K− 1 and fex = 30 Wm− 2 K− 1, as given we calculated three estimates, one based on the TSM method (see Ap
in Ref. [49], we calculated Ug = 6.363 Wm− 2 K− 1, using Eq. (15). pendix A), and two based on repeated measurements, performed either
We derived the values of hex and hin by the following procedure: First, in one single or in multiple laboratories (section 4.1). The estimate of
with the initial settings hex = fex and hin = fin, the glass surface temper unum is considered to be of negligible magnitude because the model
atures θ1 and θ2, and also a preliminary Ug-value, Ug0, were calculated equations were solved using highly accurate matrix algebra algorithms
using the present frame model (section 2.2). Then the radiative (MATLAB software). The uinput was estimated for two windows (#F015
component hrex (and similarly hrin) could be estimated as: and #F020 in Ref. [49]) using the TSM, with input data given in Table 2.
These two windows have the same 2-layer glass pane glazing, but
hrex = 4εeff σ θ3av (16) different frames; Window #F015 has an Al frame with thermal break
material built in, while the frame of window #F020 was made of wood.
where εeff is the effective emissivity between two planar surfaces with
The UA results for glazing and frames were presented in Tables 3–5, and
emissivity εa and εb, defined in Eq. (17), and θav is the average tem
the UA for the complete windows #F015 and #F020 was given in
perature of a window surface and its environment [70]. As an approx
Table 6.
imation, εeff for the radiative heat transfer between the exterior In step 2 of the validation procedure, the validation uncertainty uval
environment (εa = εex = 1) and exterior glass surface (εb = εglass) is
was calculated for windows #F015 and #F020 (see Table 7) as:
estimated by Ref. [70]:
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
uval = u2D + u2num + u2input (20)
6
K.E.A. Ohlsson et al. Renewable and Sustainable Energy Reviews 168 (2022) 112748
Table 3 Table 5
Uncertainty budget for 2-layer glazing of windows #F020 and #F015. Uncertainty budget for the Aluminum frame with two thermal breaks of window
Input quantity Value xi Uncertainty ui f (…, xi + ui , …) χ
#F015.
Input quantity Value xi Uncertainty ui f (…, xi + ui , …) χ
ε 0.84 0.01 3.0090 28.2%
θex − 18 ◦ C 0.3 2.9840 0.2% ε 0.90 0.02 4.9469 2.27%
θin 21 ◦ C 0.3 2.9844 0.3% θex − 18 ◦ C 0.3 4.9139 0.004%
hex 26.7 Wm− 2 K− 1 1.3 2.9928 4.7% θin 21 ◦ C 0.3 4.9158 0.02%
hin 4.2 Wm− 2 K− 1 0.21 3.0077 25.6% hex 26.7 Wm− 2 K− 1 1.3 4.9443 1.94%
h 2.10 Wm− 2 K− 1 0.11 3.0146 40.9% hin 4.2 Wm− 2 K− 1 0.21 4.9781 8.22%
λglass 0.9 Wm− 1 K− 1 0.03 4.9200 0.11%
Ug 2.9816 0.052 100% λAl 215 Wm− 1 K− 1 9 4.9124 0.000%
λspacer 0.24 Wm− 1 K− 1 0.07 5.0929 62.0%
Ug = f (…, xi , …) is the glazing thermal transmittance at original values of all xi,
λTB 0.14 Wm− 1 K− 1 0.04 5.0279 25.4%
while in f (…, xi +ui , …) the i:th input quantity xi has been increased by addition
of ui. Uf 4.9124 0.229 100%
7
K.E.A. Ohlsson et al. Renewable and Sustainable Energy Reviews 168 (2022) 112748
Table 7
Model error and validation uncertainty uval for the window model.
2 1 2 1
Window Frame* Glazing S D E=S-D uD1 = 0.10 Wm− K− uD3 = 0.19 Wm− K−
#F020 Wood 2L, UC 2.66 2.70 − 0.04 0.12 − 0.17 0.20 − 0.10
#F022 Al + TB 2L, LowE 2.96 3.12 − 0.16
#F062 Al + TB 2L, UC 3.36 3.51 − 0.15
#F015 Al + 2 TB 2L, UC 3.20 3.55 − 0.35 0.11 − 1.55 0.20 − 0.89
#F024 Al 2L, UC 3.40 3.75 − 0.35
(*) TB is thermal break. (**) Double layer (2L) glazing, either uncoated (UC) or with low emissivity (LowE) coating (ε = 0.44). Values of the window thermal
transmittance (Wm− 2 K− 1) were obtained from simulation (S) using the present window model, or from measurement results (D) reported in Ref. [49] using the hotbox
technique. The coverage factor k = 2 corresponds approximatively to the 95% confidence level.
In step 3, the model error for each VE is estimated as E ± k⋅uval, #F020, a regression fit was not possible. For both these windows, uval is
where E = S – D, and k is the coverage factor. With k = 2 and a Gaussian determined mainly by uD, especially when the multiple laboratory es
error distribution, the interval ± k⋅uval corresponds to the 95% confi timate uD3 is used. Since the uncertainty of hotbox measurement does
dence level. The results in Table 7 shows that, for uD3 = 0.19 Wm− 2 K− 1, not depend on the window design, it seems reasonable to believe that the
and for the windows #F020 and #F015, the model error lies within the estimate of uval applies to all windows in the present work. This
uncertainty interval of the respective VE, i.e. − k.uval < E < k.uval . conclusion is supported by the results shown in Fig. 5, which shows
However, with uD1 = 0.10 Wm− 2 K− 1, there exists a significant model results from the present window model as well as from previously
error for window #F015, i.e. |E| > k.uval . published results, where the frames were modelled using a 2D model
In step 4, the model error (E ± k⋅uval) was compared with the desired [49].
level of accuracy in the model prediction, with accuracy defined as Step 7 involves estimation of the prediction uncertainty upred by
closeness between prediction and true value. In our case, we wanted the extrapolation of the model error (here uval), using the regression equa
accuracy to be as good as uval allows for, i.e. the requirement is that |E| < tion, to the point of application of the model, where no experimental
k.uval . In cases where this requirement is not fulfilled, the model should data are available [35]. In general, upred includes both uval and the un
first be revised and then validated again starting at step 1. If |E| > k.uval , certainties arising from the calculation of the regression equation pa
then it is still possible that E is within the required level of accuracy. In rameters. In the present case, uval was constant across the VD (see step
such cases validation proceeds with step 5, but with E substituted for 6), and the regression fit was not needed. We therefore estimated upred =
uval. uval, and assumed that this estimate was valid for model applications
In step 5, the validation domain (VD) is defined in the model within and in the vicinity of the VD.
parameter space by the VEs which have passed step 4. In our case, we The UA included in the validation procedure provides information
have the two such VEs for windows #F020 and #F015, with their model on the sources of uncertainty, which could be useful for future model
parameters values shown in Table 6 and in Eq. (3). The total window improvement or planning of new VEs. For example, the uncertainty in
area Aw = 1.10 m2 and the Ug = 3.0 Wm− 2 K− 1 (2-layer glazing; un the (simulated) Uf –value is much larger than the uncertainty in Ug (see
coated glass panes) are the same for all five windows in this study, Table 6). If it is desired to reduce the former uncertainty in a new VE
except for window #F022 where Ug = 2.6 Wm− 2 K− 1 (due to low then the main contributing input variables of uinput of the frame need to
emissivity coating). The VD could therefore be defined in the 2D be measured more accurately. In the present case, this means that the
parameter space formed by the model parameters Uf and the ratio Af/Aw uncertainties of λwood, λspacer, and λTB should be reduced (see Tables 4 and
(see Fig. 4). Since Ug and Aw are approximatively constant for all win 5).
dows, the present VD does not include variation in these model
parameters. 6. Comparative analysis of energy saving measures
Step 6 includes modelling of the model error structure within the VD,
usually by a regression fit to model errors found in the set of VEs [35]. In In the introduction it was stated that any ESM considered must fulfill
our case, the error structure was given by the variation in uval over the
VD. However, since uval was only estimated for two windows, #F015 and
8
K.E.A. Ohlsson et al. Renewable and Sustainable Energy Reviews 168 (2022) 112748
the minimum requirement of yielding a saving of energy that is above information on validation and uncertainties, as listed in Section 1, for
the uncertainty limit. This condition could be expressed as: Ψ > k.uΨ . window retrofit ESMs involving the reduction of Uw. Based on findings
For the window ESM based on reduction of Uw by replacing the existing in the present study we drew the following main and novel conclusions:
window (Uw0) with a retrofit window (Uw1) of equal size, the energy First, we provided quantitative estimates of the model error (E) and
saved could be calculated as: uncertainty (upred) in the values of Uw predicted at the design stage. This
∫ was performed using an area-based model similar to the window model
Ψ = (Uw0 − Uw1 )Aw (Θin − Θex )dt (21) of ISO 15099, and a benchmark validation CS-procedure applied to a set
of existing VEs. The similarity between the models makes the obtained
where the integration is over the time period of the year where space estimates of E and upred relevant for the ISO 15099 model. Since the ISO
heating is required, i.e. when θin > θex. Note that the integral in Eq. (21) 15099 is normative for window thermal design within the building
is similar to the well-known concept of “degree days”, which is often sector, these estimates are useful in window retrofit design for MCC.
used for calculation of building heating/cooling energy demand. This Second, we found that the VD, defined using the five VEs selected
integral and Aw remain constant, while Uw is changed. Therefore, with from Ref. [49], is described by only two model variables: the Af-to-Aw
the prediction uncertainties for Uw0 and Uw1 equal to upred0 and upred1, ratio and the Uf. Since window model application should be located
respectively, we could estimate the uncertainty in the energy saved as inside or in the vicinity of the VD, this means that there are serious re
(cf. Eq. (A3)): strictions to the use of the model. For examples, windows with Ug much
∫ less than 3 W m − 2K− 1 (e.g. triple pane glazing), and small windows with
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
uΨ = u2pred0 + u2pred1 Aw (Θin − Θex )dt (22) relatively large Af, are located outside the VD, which makes model
predictions unreliable. The review of existing VEs (see Section 3)
For all window designs of the present study, we estimated above that showed that the VD could not be extended further because of lack of
upred = uval = 0.20 Wm− 2 K− 1, and insertion in Eq. (22) gives: well-documented VEs.
∫ Third, the UA provided quantification of the contribution of various
uΨ = 0.28Aw (Θin − Θex )dt (23) sources of uncertainty to upred. In the present case with model prediction
of Uw, it was found that uD is the major source of uncertainty in the
Eventually, we reformulate the minimum requirement of the win predicted value. The comparison between the uD2 (single laboratory)
dow ESM as follows, using Eqs. (21)–(23): and uD3 (multiple laboratories) estimates, showed the importance of
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ performing all VEs under closely similar conditions.
(Uw0 − Uw1 ) > k u2pred0 + u2pred1 (24) Fourth, for the window ESM retrofit, we derived Eq. (22) which
provided the needed relationship between uΨ and the prediction un
In the present work, assuming the Gaussian error distribution and the certainties of the two Uw-values involved. For the window retrofit to
95% confidence level, we set k = 2, and obtain the requirement for the yield an energy saving above the uncertainty limit, it is therefore
ESM: concluded that the reduction in Uw must be larger than 0.56 Wm− 2 K− 1,
given that both upred are equal to 0.20 Wm− 2 K− 1.
(Uw0 − Uw1 ) > 0.28k = 0.56 Wm− 2 K− 1
(25)
This requirement, expressed by Eq. (25), could now be applied to Funding
pairwise comparisons of simulated Uw-values (S) for different windows
shown in Table 7. Among those windows, we could predict that there is a This work was supported by the Kolarctic CBC project: KO1089
significant energy saving only when the Al-frame-window #F024 is Green Arctic Building (GrAB).
replaced by the wood-frame window #F020 (then the difference in S-
values equals 0.74 W m− 2K− 1, which satisfies Eq. (25)). In this example, CRediT
uD3 was selected to be used in Eq. (23) for estimation of upred. This is
appropriate if the hotbox measurements in the VEs have been performed Anders Ohlsson: Conceptualization; Software; Investigation;
in different laboratories or under varying conditions. Writing – review & editing; Gireesh Nair: Writing – review & editing;
For future design of VEs for modelling of Uw, with the intention to Thomas Olofsson: Writing – review & editing; Funding acquisition.
compare model predictions between two windows considered for ESM,
there is the possibility to measure the two windows simultaneously, and
side-by-side. In such comparative measurement of the difference in the Declaration of competing interest
Uw-values for two windows, systematic errors, that are equal for both
windows, will cancel. This possibility was discussed in Ref. [55] for the The authors declare that they have no known competing financial
outdoor MoWiTT facility, but to our knowledge no further published interests or personal relationships that could have appeared to influence
works are available. the work reported in this paper.
The objectives of present case study was to meet the needs for Data will be made available on request.
The present reduced order window model based on physics theory was assumed to have uncorrelated model input variables. Therefore it was
justified to use the Taylor Series Method (TSM) for estimation of uinput. The TSM is described in the GUM standard [32,69] for estimation of uD.
GUM defines measurement uncertainty as a “parameter that characterizes the dispersion of the values that could reasonably be attributed to the
measured quantity” (see §B.2.18 in Ref. [69]). The measurement uncertainty is usually expressed as a standard deviation (uD), or a multiple thereof.
GUM describes two types of evaluation of uD: (1) In Type A evaluation, uD is estimated as the standard deviation s of a series of N measurements
repeated under the same conditions, while (2) in Type B evaluation, the estimation is based on scientific judgement of other available information on
9
K.E.A. Ohlsson et al. Renewable and Sustainable Energy Reviews 168 (2022) 112748
the variability of the measured quantity. Both types were used in the present study, and their applications are exemplified below.
Type A evaluation quantifies the uncertainty caused by random errors, which occur during a measurement time period, where many conditions are
constant, e.g. the instrument used, the calibration or the environmental parameters. The systematic error β in the measurement is estimated as the
difference between the mean of the N measurements and the true value of the measured quantity, the latter obtained using a reference sample [69]. If
instead conditions vary between individual measurements, then often β is reduced and the spread of the random errors increases. Therefore, the
systematic uncertainty is often expressed as the standard deviation bsys of the distribution of systematic errors [32]. If β is significant in the series of N
measurements, then
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
uD = b2sys + s2 (A1)
In the present work, TSM is applied for estimation of uinput, with the function f given implicitly by the numerical algorithm for calculation of S. It
was then convenient to use the following approximation in Eq. (A3) (cf. Chap. 3–1.5 in Ref. [32]):
∂f
u ≈ f (x1 , x2 , …, xi + ui , …) − f (x1 , x2 , …, xi , …) (A4)
∂xi i
In UA, it was useful to express the fractional contribution to uinput from uncertainty in input variable xi by:
( )2 ∑( )2
∂f / ∂f
χi = ui ui (A5)
∂xi i
∂xi
The Rubin model [39] was applied to the 2-layer glass pane glazing, with pane spacing w, and with the four glass surfaces indexed from 1 to 4
starting from the exterior side (see Fig. 1 in Ref. [39]). The model allows for non-zero infrared transmittance τ glazing layers, but in our case with glass
layers, τ = 0.
2
The net infrared radiation fluxes from the i:th glass surfaces, qri (Wm− K− 1), are given by (cf. Eqs. (1) and (2) in Ref. [39]:
qr1 = S1 + ρ r
1 qex +τ r
1 q3 (B1)
qr4 = S4 + ρ r
4 qin +τ r
2 q2 (B4)
where Si is the emitted flux, ρi is the infrared reflectance, and τi is the infrared transmittance. The emitted energy fluxes are given by:
Si = εi σ θ4n (B5)
8
where i = 1, 2, 3, or 4, and layer n = 1 or 2. The boundary radiative fluxes are given by qrex = σθ4ex and qrin = σθ4in (σ = 5.67⋅10− Wm− 2K− 4; Stefan-
Boltzmann′ s constant). The Eqs. (B1)-(B4) could be rearranged into matrix form:
10
K.E.A. Ohlsson et al. Renewable and Sustainable Energy Reviews 168 (2022) 112748
⎡ ⎤⎡ ⎤ ⎡ ⎤
1 0 − τ1 0 ⎢ qr1 ⎥ ⎢ S1 + ρ1 qrex ⎥
⎢ ⎥ ⎢ ⎥
⎢0 1 − ρ2 0 ⎥ ⎢
⎢ qr2 ⎥
⎥
⎢
⎢ S2 + τ1 qrex ⎥
⎥
Mqr − r = ⎣0 ⎦ .⎢ ⎥− ⎢ ⎥=0 (B6)
− ρ3 1 0 ⎢ qr3 ⎥ ⎢ S3 + τ2 qrin ⎥
⎢ ⎥ ⎢ ⎥
⎣ q r ⎦ ⎣ S +ρ q r ⎦
0 − τ2 0 1 4 4 4 in
Eq. (B6) is the corrected version of Eq. (7) in Ref. [39], where in the latter the terms in r containing qrex and qrin , are missing. Given the layer
[ ]
θ
temperatures θ = 1 , we solve Eq. (B6) as:
θ2
qr = M − 1 r (B7)
Following [39], the heat flux due to natural convection or air conduction between layers 1 and 2 could be expressed as:
q = h(θ1 − θ2 ) = hθ1 − hθ2 = qc2 − qc3 (B8)
where h is the heat conductance, and qci is the net convective/conductive heat flux at the i:th surface (i = 2 and 3). With hex and hin defined as the
exterior and interior conductance respectively, we have
⎡ ⎤ ⎡ ⎤
c
⎢ q1 ⎥ hex 0
⎢ ⎥ [ ]
⎢ qc ⎥
⎢ ⎥ h 0⎢ θ ⎥
qc = ⎢ 2c ⎥ = ⎣. 1 = Hθ ⎦ (B9)
⎢ q3 ⎥ 0 h θ2
⎢ ⎥
⎣ qc ⎦
4 0 hin
For each layer (glass pane), at steady-state, no net energy is absorbed, which is expressed as:
⎡ ⎤
q1
[ ] [ ] [ ]
Δ1 1 1 − 1 0 ⎢ q2 ⎥ qex
Δ= = .⎣ ⎦− = Cq − b = 0 (B10)
Δ2 0 − 1 1 1 q 3
qin
q4
where
q = qr + qc (B11)
Solving Eq. (B10) yields the layer temperature vector θ, and from these the q is calculated. The solution was obtained by the iterative procedure
described in Ref. [39], and with ∂Δi /∂Θj approximated by the first term in a Taylor series expansion.
The net heat flux from exterior to interior environments is calculated as:
qnet = q4 − qin (B14)
Eventually, the thermal transmittance of the glazing is obtained as:
qnet
Ug = (B15)
θex − θin
11
K.E.A. Ohlsson et al. Renewable and Sustainable Energy Reviews 168 (2022) 112748
[5] Ürge-Vorsatz D, Khosla R, Bernhardt R, Chan YC, Vérez D, Hu S, et al. Advances [37] ISO 10077-1. Thermal performance of windows, doors and shutters - calculation of
toward a Net-Zero global building sector. Annu Rev Environ Resour 2020;45: thermal transmittance - Part 1: General. Brussels: CEN; 2006.
227–69. [38] ISO 15099. Thermal performance of windows, doors and shading devices - detailed
[6] Mills E. Building commissioning: a golden opportunity for reducing energy costs calculations. Geneva: Switzerland ISO; 2003.
and greenhouse gas emissions in the United States. Energy Efficiency 2011;4: [39] Rubin M. Calculating heat transfer through windows. Energy Res 1982;6:341–9.
145–73. [40] Hollands KGT, Wright JL. Heat loss coefficients and effective Πα products for flate-
[7] Mata É, Sasic Kalagasidis A, Johnsson F. Energy usage and technical potential for plate collectors with diathermanous covers. Sol Energy 1983;30:211–6.
energy saving measures in the Swedish residential building stock. Energy Pol 2013; [41] Curcija D, Ambs LL, Goss WP. A comparison of European and North American
55:404–14. window U-value calculation procedure. ASHRAE Trans 1989;95:575–91.
[8] Mata É, Sasic Kalagasidis A, Johnsson F. Contributions of building retrofitting in [42] Gustavsen A, Arasteh D, Jelle BP, Curcija C, Kohler C. Developing low-conductance
five member states to EU targets for energy savings. Renew Sustain Energy Rev window frames: capabilities and limitations of current window heat transfer design
2018;93:759–74. tools - state-of-the-art review. J Build Phys 2008;32:131–53.
[9] Burhenne S, Tsvetkova O, Jacob D, Henze GP, Wagner A. Uncertainty [43] Arasteh D. An analysis of edge heat transfer in residential windows. J Therm Insul
quantification for combined building performance and cost-benefit analyses. Build 1991;14:295–310.
Environ 2013;62:143–54. [44] Carpenter S, McGowan A. Effect of framing systems on the thermal performance of
[10] Bertoldi P, Economidou M, Palermo V, Boza-Kiss B, Todeschi V. How to finance windows. ASHRAE Trans 1993;99:907–14.
energy renovation of residential buildings: review of current and emerging [45] Asdrubali F, Baldinelli G, Bianchi F. Influence of cavities geometric and emissivity
financing instruments in the EU. WIREs Energy Environ 2021;10:1–26. e384. properties on the overall thermal performance of aluminum frames for windows.
[11] Grillone B, Danov S, Sumper A, Cipriano J, Mor G. A review of deterministic and Energy Build 2013;60:298–309.
data-driven methods to quantify energy efficiency savings and to predict [46] Lechowska AA, Schnotale JA, Baldinelli G. Window frame thermal transmittance
retrofitting scenarios in buildings. Renew Sustain Energy Rev 2020;131:110027. improvements without frame geometry variations: an experimentally validated
[12] Touzani S, Granderson J, Jump D, Rebello D. Evaluation of methods to assess the CFD analysis. Energy Build 2017;145:188–99.
uncertainty in estimated energy savings. Energy Build 2019;193:216–25. [47] Noyé P, Laustsen S, Svendsen S. Calculating the heat transfer coefficient of frame
[13] Johnson KA, Hamer W, Vosloo JS. Structuring uncertainty management for energy profiles with internal cavities. Nord J Build Phys 2004;3:1–7.
savings calculations. S Afr J Ind Eng 2019;30:149–62. [48] Cardinale N, Rospi G, Cardinale T. Numerical and experimental thermal analysis
[14] Rabani M, Madessa HB, Nord N. A state-of-art review of retrofit interventions in for the improvement of various types of window frames and rolling-shutter boxes.
buildings towards nearly zero energy level. Energy Proc 2017;134:317–26. Int J Energy Environ Eng 2015;6:101–10.
[15] Wu W, Skye HM. Residential net-zero energy buildings: review and perspective. [49] Elmahdy AH. Joint Canadian/U.S. research project on window performance:
Renew Sustain Energy Rev 2021;142:110859. project outline and preliminary results. ASHRAE Trans 1990;96:896–900.
[16] Ruggeri AG, Gabrielli L, Scarpa M. Energy retrofit in European building portfolios: [50] Klems JH, Reilly S. Window nighttime U-values: a comparison between computer
a review of five key aspects. Sustainability 2020;12:7465. calculations and MoWiTT measurements. ASHRAE Trans 1990;96:907–11.
[17] He Q, Hossain MU, Ng ST, Augenbroe G. Identifying practical sustainable retrofit [51] Elmahdy AH. Heat transmission and R-value of fenestration systems using IRC hot
measures for existing high-rise residential buildings in various climate zones box: procedure and uncertainty analysis. ASHRAE Trans 1992;98:630–7.
through an integrated energy-cost model. Renew Sustain Energy Rev 2021;151: [52] Klems JH. In: A calibrated hotbox for testing window systems - construction,
111578. calibration, and measurements on prototype high-performance windows.
[18] Feist W, Schnieders J. Energy efficiency - a key to sustainable housing. Eur Phys J ASHRAE/DOE-ORNL Conference on the Thermal Performance of the Exterior
Spec Top 2009;176:141–53. Envelopes of Buildings; 1979. Orlanda, FL, USA.
[19] Sullivan R, Selkowitz S. Energy performance analysis of fenestration in a single- [53] Arasteh D, Beck FA, Stone NI, duPont WC, Mathis RC, Koenig MJ. Phase I results of
family residence. CA, USA: Lawrence Berkeley National Laboratory; 1984. the NFRC U-value procedure validation project. ASHRAE Trans 1994;100:
[20] Pásztory Z, Gorbacheva G, Czimondor D, Rébék-Nagy P, Sanaev V, Rykunin S, et al. 1724–31.
The effects of windows and reduced night temperature on the heating energy [54] Box GEP, Hunter WG, Hunter JS. Statistics for experimenters. New York: Wiley;
demand in different regions of Russia. IOP Conf Ser Mater Sci Eng 2018;371: 1978.
012050. [55] Klems JH. Method of measuring nighttime U-values using the mobile window
[21] Kalbe K, Kalamees T. Influence of window details on the energy performance of an thermal test facility. ASHRAE Trans 1992;98(pt 2):619–29.
nZEB. J Sustain Architect Civ Eng 2019;1:61–70. [56] Arasteh D, Hartmann J, Rubin M. Experimental verification of a model of heat
[22] Manz H. On minimizing heat transport in architectural glazing. Renew Energy transfer through windows. ASHRAE Trans 1987;93:1425–31.
2008;33:119–28. [57] Elmahdy AH, Bowen RP. Laboratory determination of the thermal resistance of
[23] Foucquier A, Robert S, Suard F, Stéphan L, Jay A. State of the art in building glazing units. ASHRAE Trans 1988;2:1301–16.
modelling and energy performances prediction: a review. Renew Sustain Energy [58] Dubrous FM, Harrison SJ. Comparison of experimental test results and analytical
Rev 2013;23:272–88. calculations of window thermal performance. ASHRAE Trans 1989;95:747–54.
[24] Harish VSKV, Kumar A. A review on modeling and simulation of building energy [59] Elmahdy AH. In: Testing and simulation of high-performance windows - phase II of
systems. Renew Sustain Energy Rev 2016;56:1272–92. the Canadian/U.S. joint research project on window performance. BTECC
[25] Déqué F, Ollivier F, Poblador A. Grey boxes used to represent buildings with a Conference; 1992. Clearwater, FL.
minimum number of geometric and thermal parameters. Energy Build 2000;31: [60] Kol HS. Thermal and dielectric properties of pine wood in the transverse direction.
29–35. Bioresources 2009;4:1663–9.
[26] Hopfe CJ, Hensen JLM. Uncertainty analysis in building performance simulation [61] Clarke JA, Yaneske PP, Pinney AA. The harmonisation of thermal properties of
for design support. Energy Build 2011;43:2798–805. building materials. UK: BEPAC; 1990. Technical Note 91/6.
[27] Ostergård T, Jensen RL, Maagaard SE. Building simulations supporting decision [62] Baker JA, Sullivan HF, Wright JL. Window (glazing and frame) heat transfer
making in early design - a review. Renew Sustain Energy Rev 2016;61:187–201. modeling. ASHRAE Trans 1990;96:901–6.
[28] Tian W, Heo Y, de Wilde P, Li Z, Yan D, Park CS, et al. A review of uncertainty [63] Rubin M. Optical properties of soda lime silica glasses. Sol Energy Mater 1985;12:
analysis in building energy assessment. Renew Sustain Energy Rev 2018;93: 275–8.
285–301. [64] Chen H-Y, Chen C. Determining the emissivity and temperature of building
[29] Shamsi MH, Ali U, Mangina E, O′ Donnell J. A framework for uncertainty materials by infrared thermometer. Construct Build Mater 2016;126:130–7.
quantification in building heat demand simulations using reduced-order grey-box [65] Gustavsen A, Berdahl P. Spectral emissivity of anodized aluminum and the thermal
energy models. Appl Energy 2020;275:115141. transmittance of aluminum window frames. Nord J Build Phys 2003;3:1–12.
[30] Ohlsson KEA, Olofsson T. Benchmarking the practice of validation and uncertainty [66] Fantucci S, Serra V. Experimental assessment of the effects of low-emissivity paints
analysis of building energy models. Renew Sustain Energy Rev 2021;142:110842. as interior radiation control coatings. Appl Sci 2020;10:842. https://doi.org/
[31] Fennell PJ, Ruyssevelt PA, Mata É, Jakob M. In: A review of the status of 10.3390/app10030842.
uncertainty and sensitivity analysis in building-stock energy models. Proceedings [67] Maiorov VA. Window glasses: state and prospects. Opt Spectrosc 2018;124:
of the16th IBPSA International Conference and Exhibition; 2019. p. 3353–60. 594–608.
[32] Coleman HW, Steele WG. Experimentation, validation, and uncertainty analysis for [68] ElSherbiny SM, Raithby GD, Hollands KGT. Heat transfer by natural convection
engineers. third ed. Hoboken, NJ: Wiley; 2009. across vertical and inclined air layers. Trans ASME, J Heat Transfer. 1982;104:
[33] Coleman HW, Stern F. Uncertainties and CFD code validation. J Fluid Eng 1997; 96–102.
119:795–803. [69] Evaluation of measurement data - guide to the expression of uncertainty in
[34] Oberkampf WL, Trucano TG, Hirsch C. Verification, validation, and predictive measurement (GUM 1995 with minor correction). Geneva: JCGM; 2008.
capability in computational engineering and physics. Appl Mech Rev 2004;57: [70] Simko TM, Elmahdy AH, Collins RE. Determination of the overall heat transmission
345–84. coefficient (U value) of vacuum glazing. ASHRAE Trans 1999;105:891–9.
[35] Roy CJ, Oberkampf WL. A comprehensive framework for verification, validation, [71] Cengel YA, Cimbala JM, Turner RH. Fundamentals of thermal-fluid sciences. New
and uncertainty quantification in scientific computing. Comput Methods Appl York: McGraw-Hill Education; 2017.
Mech Eng 2011;200:2131–44. [72] Roache PJ. Verification and validation in computational science and engineering.
[36] ISO 10077-2. Thermal performance of windows, doors and shutters - calculation of Albuquerque, USA: Hermosa; 1998.
thermal transmittance - Part 2: numerical method for frame. Brussels: CEN; 2012. [73] Oberkampf WL, Roy CJ. Verification and validation in scientific computing.
Cambridge: Cambridge University Press; 2010.
12