Applied Energy 107 (2013) 229–243

Contents lists available at SciVerse ScienceDirect

Applied Energy
journal homepage: [Link]/locate/apenergy

A building thermal bridges sensitivity analysis

Alfonso Capozzoli, Alice Gorrino, Vincenzo Corrado ⇑
Department of Energy, Politecnico di Torino, Corso Duca degli Abruzzi, 24, 10129 Torino, Italy

h i g h l i g h t s

" Heat transfer through the most typical thermal bridges is analyzed through a finite difference method.
" The linear thermal transmittance for a large number of design parameters is catalogued.
" Non-linear regression models of the most typical thermal bridges are identified.
" A sensitivity analysis of the most relevant design parameters is carried out.

a r t i c l e i n f o a b s t r a c t

Article history: Along with the entry into force of the new European Directive 2010/31/EU on the Energy Performance of
Received 26 July 2012 Buildings (EPBD recast), each Member State has the responsibility of supporting activities for the con-
Received in revised form 20 January 2013 struction of nearly zero energy buildings with a very high energy performance. In order to achieve the
Accepted 15 February 2013
new EU directive targets, designers, in addition to having to use innovative building components, also
Available online 16 March 2013
have to pay more attention to the construction details which mostly affect building envelope heat losses.
It is therefore necessary not only to properly design structural nodes, in order to minimize such energy
Keywords:
losses, but also to identify accurate numerical methods in order to appreciate the benefits of a proper
Thermal bridges
Transmission heat losses
design. A sensitivity analysis based on an extensive study of the linear thermal transmittance value of
Parametric analysis many types of thermal bridge, based on the methodology specified in EN ISO 10211, has been carried
Regression models out in the presented work. After having defined the input design variables and considering a range of var-
Sensitivity analysis iation for each of them for the linear thermal transmittance evaluation, a non-linear regression model has
Analysis of variance been specifically developed for each analyzed thermal bridge, considering the output values of a numer-
ical code as data set. In order to perform the sensitivity analysis a significant and representative number
of cases have been generated, using a sampling technique. The ANOVA–FAST method has been per-
formed, on the basis of the obtained results, in order to assess the contribution of each input design var-
iable to the deviation of the linear thermal transmittance for each kind of thermal bridge.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction components, while on the other hand to pay greater attention to

construction details.
1.1. Influence of thermal bridges on building energy performance In fact, the mere addition of an insulation layer only reduces the
one-dimensional heat flow, but does not significantly decrease the
With the entry into force of European Directive 2010/31 on the multi-dimensional one, if no attention has been paid to the heat
Energy Performance of Buildings (EPBD recast) [1], each Member flow through thermal bridges. Although it is not possible to obtain
State is required to draw up national plans to increase the number general results concerning the weight of thermal bridges on the en-
of nearly zero-energy buildings. Moreover, starting from European ergy needs of buildings, several studies have presented numerical
Directive 2002/91/CE, the Italian national legislation on building results for different cases. A study was conducted in Greece on a typ-
energy efficiency (Legislative Decree no. 192/2005, Legislative De- ical three-storey apartment building with an open ground-floor
cree no. 311/2006) has led to an improvement in opaque and trans- space (pilotis) and a flat roof; the façades are composed of two brick
parent envelope performance. layers with interposed insulation [2]. The study shows that the dou-
In order to achieve the objectives of the EPBD recast, the de- ble brick wall construction widely used in Greece is affected to a
signer is required on one hand to use innovative envelope great extent by thermal bridges. Even if the actual construction pre-
sents high insulation levels, the heating need can be 30% higher than
⇑ Corresponding author. Tel.: +39 011 090 4456; fax: +39 011 090 4499. the one calculated without taking into account the thermal bridge
E-mail address: [Link]@[Link] (V. Corrado). effects. Cappelletti et al. [3] have shown that the weight of thermal

0306-2619/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
[Link]
230 A. Capozzoli et al. / Applied Energy 107 (2013) 229–243

Nomenclature

Symbols b balcony
B0 characteristic dimension of the floor (m) calc calculated
btr temperature reduction factor (–) e external
e width (m) fl floor
H overall heat transfer coefficient (W K1) fr frame
l length (m) gr ground
L2D thermal coupling coefficient from bi-dimensional calcu- i internal
lation (W m1 K1) int intermediate
q heat flow per meter length (W m1) ins insulation layer
R thermal resistance (m2 K W1) m mean
S thickness (m) mas masonry
SI sensitivity index M1 internal masonry
U thermal transmittance of a building component M2 external masonry
(W m2 K1) p pillar
pred predicted
Greek symbols r roof
h air temperature (°C) se external surface
k thermal conductivity (W m1 K1) si internal surface
w linear thermal transmittance (W m1 K1) tr transmission
w wall
Subscripts
adj adjacent, adjusted

bridges on building energy needs for space heating can reach 67% for consistent reduction of up to 70–75% in linear thermal transmit-
a building with a double brick wall (U = 0,15 W/(m2 K)) in a typical tances when the window is moved from the internal to the exter-
Italian climatic zone. A collection of interesting studies has been re- nal position. This decrease mainly depends on the position of the
ported in [4]. The impact of thermal bridges on the energy quality of insulating layer installed in the window hole.
a building has also emerged from a study carried out in the Czech Hence, in order to minimize thermal bridge energy losses, it is
Republic: the case study was a residential building with brick walls necessary not only to design the structural nodes properly, but also
and wooden frame windows. From the same study it is highlighted to identify accurate methods to calculate heat losses, in order to
that the relative impact of the thermal bridges on the annual energy make it possible to appreciate the benefits induced by a correct
needs varies from 7%, for typical houses of the seventies to 28%, for design.
modern high-quality houses. The impact of thermal bridges on the
heating energy needs of different European Member States is gener- 1.2. Existing methods for the calculation of energy losses through
ally as high as 30% [4]. Moreover, the more a building envelope is thermal bridges
insulated, the more thermal bridges play a relatively increasing role
in the global heat losses and consequently in the energy needs for In this paper the thermal bridging effect is evaluated by means
space heating. of the linear thermal transmittance approach. Since this parameter
A study on the effects of thermal bridges in an Italian climate is calculated under stationary conditions, it is generally applied for
has been carried out on two building types (terraced houses and building energy need assessment through quasi-steady state meth-
semi-detached houses) and three envelope configurations by Evola ods in the framework of energy performance technical standards
et al. [5]. The results show that the correcting thermal bridges is an and regulations. The use of linear thermal transmittance approach
effective way of reducing the primary energy heating demand (25% presents some limitations for building energy simulations (BESs);
for terraced houses, 17.5% for semi-detached house), but only a therefore the aim of this paper is not to implement it in dynamic
slight improvement – about 3.5% – can be achieved in the cooling energy simulation codes.
performance of the building. The overall annual energy savings is The heat exchange through thermal bridges can be calculated
about 8.5%, but a cost analysis has shown that the savings deter- using the different methodologies specified in the relevant techni-
mined by correcting thermal bridges are not sufficient to recover cal standards [8,9], which present both simplified and detailed
the additional construction and refurbishment costs. methods to calculate thermal losses through thermal bridges un-
Some studies focus on window thermal bridges giving practical der steady-state conditions.
and technical solutions to minimize their effect on building ther- Italian Technical Specification UNI/TS 11300-1 [10], for existing
mal losses. Thermal bridging has been evaluated through three dif- buildings, in the absence of reliable project data or more accurate
ferent window systems commonly used in buildings in hot regions, information, for some building types, considers a percentage increase
by Ben-Nakhi [6] by means of the linear thermal transmittance (w) of the overall heat transfer coefficient by transmission (Htr,adj),
approach. The results show that the classical window system, while European Standard EN 12831 [11] introduces a corrected
which is the most common in Kuwait, is affected by significant thermal transmittance of building element, taking into account lin-
thermal bridging and that this effect should be considered in build- ear thermal bridges through an increase of the thermal transmit-
ing design. Some practical methods used to reduce the w value for tance of the element. Both documents mention EN ISO 14683
the classical window system were also evaluated in this paper. The standard as the main reference for the calculation of heat transfer
influence of thermal bridges on the performance of windows has through the thermal bridges in all the remaining cases.
also been analyzed by Cappelletti et al. [7] for the case of clay walls In the cases of linear thermal bridges, and in the absence of spe-
with external and cavity insulation. The results pointed out a cific data, EN ISO 14683 standard [8] provides the use of default
 A. Capozzoli et al. / Applied Energy 107 (2013) 229–243 231

linear thermal transmittance values with reference to typical two- analysis, based on an extensive study carried out to calculate the
dimensional thermal bridges, and reports these values in Annex A, linear thermal transmittance of many types of thermal bridges
with a cautious overestimation. by means of detailed numerical methods, is presented.
The standard also specifies that these values can only be used if The main design variables of interest for the creation of thermal
no details are available on the particular thermal bridge or in the bridge catalogues were investigated, using the finite differences
case in which an approximate value of w is appropriate in relation calculation methodology, according to EN ISO 10211.
to the accuracy required to determine the overall heat transfer. Using the abovementioned methodology, many thermal bridges
As far as numerical calculations are concerned, EN ISO 14683 re- catalogues have been published till now. For the present study 36
fers to EN ISO 10211, which established the specifications for a three- thermal bridges, including the most typical junctions of the build-
dimensional and a two-dimensional geometric model of a thermal ing envelope, have been selected from a catalogue of about 100
bridge for the numerical calculation of heat flows, in order to assess thermal bridges drawn by the Authors [22].
the overall heat loss from a building or part of it, and the minimum sur- Having defined the main input variables pertaining to thermal
face temperatures in order to assess the risk of surface condensation. bridges and considering a range of variation that covers most of
EN ISO 14683 [8] suggest the following calculation methods, in the current building envelope design solutions for each of them,
descending order of accuracy: about 190 different calculated configurations have been consid-
ered for each thermal bridge, varying the values of the design vari-
– Numerical calculations according to EN ISO 10211 [9] (typical ables. An example of a thermal bridge form containing the
accuracy ±5%). considered configurations is presented in Fig. 1.
– Thermal bridge catalogues (typical accuracy of ±20%). From these configurations, for each thermal bridge a multiple
– Manual calculations (typical accuracy ±20%). non-linear regression model has been identified, considering the
– Default values provided by EN ISO 14683 itself (typical accuracy output values of the numerical code as the data set for the regres-
±50%). sion function.
In order to perform a sensitivity analysis, a significant number
The default w-values provide the most inaccurate results above of representative cases (about 1000) have been generated for each
all because the parameters used to calculate the linear thermal thermal bridge using a sampling technique. Each case consists of a
transmittance (boundary conditions, geometric parameters and set of input variables generated according to a uniform probability
material thermal properties) are set at default values that are close distribution. The linear thermal transmittance values of each ther-
to the maximum which is likely to occur in practice and are thus mal bridge have been predicted for each case by means of the iden-
cautious overestimates of the thermal bridging effects. tified regression equations.
Thermal bridges catalogues are much more accurate when the A sensitivity analysis of the results has been performed by means
parameters used to calculate the w values correspond to the real of the ANOVA (Analysis of Variance) technique [23], with the aim of
construction characteristics. In this case, the accuracy can be com- assessing the contribution of each input variable to the deviation of
parable to that of a numerical calculation [8,9]. the linear thermal transmittance for each analyzed thermal bridge.
As far as the numerical calculations according to EN ISO 10211, The aim of the sensitivity analysis is to provide designers and
which provides the most accurate results, are concerned, many policymakers with a simple and practical way to identify the main
simulation codes based on finite difference, finite element or finite important design variables affecting the heat losses through each
volume calculations have been developed. Current studies are thermal bridge and to address a more correct application of techni-
focusing on the development of a thermal parameter characteriz- cal standards and regulations.
ing thermal bridges under dynamic conditions [12] and its imple-
mentation in dynamic energy simulation codes [13–17], although 2. Calculation of linear thermal transmittance
some critical aspects still remain regarding the inertia of thermal
bridges, as can be seen in Martin et al. [18]. The sensitivity analysis in this paper starts with an extensive
Starting from numerical simulations, Dilmac et al. [19] have study which was carried out to evaluate the linear thermal transmit-
proposed a new method to calculate the parameters cited in ISO tance of many different types of thermal bridge, with a two dimen-
9164 for the floor/beam-wall intersections used in the equations sional geometric model under steady state conditions, according to
for the calculation of the energy transmission heat loss coefficient, EN ISO 10211. The procedure starts with the calculation of the ther-
HT, especially in the evaluation of thermal bridges. The proposed mal coupling coefficient (L2D), that is the two-dimensional heat
method is simpler than that specified in EN ISO 10211 and yields transfer coefficient, expressed in W m1 K1, obtained by means of
more accurate results for floors with web beams than EN ISO finite difference methods or finite element analysis:
14683. Another simple way to evaluate the effect of thermal q
bridges on the global building heat losses is defined in Asdrubali L2D ¼ ð1Þ
ðhi  he Þ
et al. [20] starting from results acquired by the thermographic
investigations and elaborated through an analytical process. where q is the heat flow per meter length, while hi and he are the
In order to provide linear thermal transmittance correlations of internal and external temperatures, respectively. The linear thermal
some thermal bridges, Ben Larbi [21] has calculated different w transmittance of the linear thermal bridge separating the two envi-
values by varying the thermal bridge properties, and has developed ronments, w, is calculated through the following equation:
a simple statistical model from results obtained through a numer-
Nj
ical resolution based on EN ISO 10211 in order to obtain w values X
w ¼ L2D  U j  lj ð2Þ
through non-linear regression equations. j¼1
Despite the intense research activity on this topic, only few
investigations concern the global influence of the envelope design where Uj is the thermal transmittance of the 1-D component j that
variables on thermal bridges linear thermal transmittance and on separates the considered environments, lj, is the length in the 2-D
the relevant heat losses. geometrical model over which the Uj value applies. The portion of
The purpose of this study is to evaluate the impact of the main the envelope taken into consideration in the calculation of L2D is
input design variables on the variance of the linear thermal trans- identified according to the specifications of the standard, and it is
mittance (w value) for several types of thermal bridge. A sensitivity extended to a suitable distance from the node under consideration,
232 A. Capozzoli et al. / Applied Energy 107 (2013) 229–243

Fig. 1. Example of a thermal bridge form including the considered configurations.

in order to include the entire region in which the heat flow is dis- wall–roof junctions, wall–balcony junctions, convex and concave
rupted from the one-dimensional conditions. The geometric model corners between walls with and without pillars, wall – intermedi-
for the calculation of the heat flow is delimited by shear plains and ate floor junctions, wall – slab on ground floor junctions, wall –
it is divided into cells by auxiliary plains (Fig. 2). window frame junctions, wall – pillar junctions. Three different
In the present study the system based on external dimensions, types of wall insulation were analyzed for each category of thermal
measured between the finished external faces of the external ele- bridge: external, intermediate and internal (in the ‘‘Description’’
ments of the building, has been assumed [8]. column of Table 1 a simplified scheme of each analyzed thermal
bridge is reported).
3. Analyzed thermal bridges Since a building could have a load-bearing wall structure rather
than a concrete structure, convex and concave corners were con-
Different junctions were chosen from the thermal bridge cata- sidered with and without pillars. Only corners with pillars were
logue developed by the Authors [22]. Thirty-six thermal bridges chosen for the double brick wall with intermediate insulation,
were considered in order to analyze the most typical junctions of which cannot be a load-bearing wall, considering both insulated
a building envelope. The following categories were considered: (C11 and C19) and non-insulated junctions (C10 and C18).
 A. Capozzoli et al. / Applied Energy 107 (2013) 229–243 233

been verified through different calculation tests. For example it

has been observed that frame U-values ranging from 2.0 to
3.0 W/(m2 K) do not give relevant variations of linear thermal
transmittance.

4. Identification of regression models

A non-linear regression model was developed from the linear

thermal transmittance values calculated by means of the TRISCO
simulation code, for each identified thermal bridge.
The statistical model used to fit the linear thermal transmit-
tance values obtained from the computer simulation is based on
the extra heat flow with respect to 1-D one, per meter and per tem-
perature difference between the inside and outside environment
through the part of the envelope components that constitute the
thermal bridge (Fig. 3).
The shape of the identified equations was adapted to each ther-
Fig. 2. Representation of the geometric model by means of shear plains (that mal bridge on the basis of the thermal characteristics of the bridge
delimitate the model) and auxiliary plains (that divide the model into cells). (heat flow lines) and the dependent variables. The general form of
the equation is:
Three different types of configuration were also considered for c1  e þ c2
double brick wall – pillar junctions: without an insulation layer w¼ þ f ðRadj ; eadj Þ ð3Þ
Rðbi  Ri Þ þ c3
(P2), with an insulation panel on the external surface (P7) and with
insulation surrounding the pillar (P9). where c1, c2, c3 and bi are the parameters estimated by the regres-
Another typical configuration was analyzed for a double brick sion analysis.
wall-intermediate floor junction: with insulation on the external This equation states that the linear thermal transmittance value
part of the floor (IF5) and without insulation (IF2). of each thermal bridge is equal to the ratio between the width of
Two different types of configuration were chosen for the three the thermal bridge and the thermal resistance, to which a function
types of wall – window frame junctions: with an interrupted insu- of the adjacent length and thermal resistances of the adjacent com-
lation layer, which means the insulation does not cover the win- ponents is added. The calculations have been limited to adjacent
dow frame (W7, W8 and W9) and with an insulation layer up to walls according to European Standard EN ISO 10211.
window frame (W12a, W11 and W12). A similar approach to identify the shape of equation was
About 190 configurations were calculated under steady-state adopted by Ben Larbi [21], in which the linear thermal transmit-
conditions for each thermal bridge by means of TRISCO simulation tance was expressed as the sum of two terms: the ratio of the ther-
code [24], varying the main design variables that influence the lin- mal bridge width to the total thermal resistance of the bridge
ear thermal transmittance value. (related to the main thermal flow through the thermal bridge)
This code applies the finite difference method to calculate the and a linear function of the adjacent wall thermal resistance.
global heat flows through the thermal bridge and the adjacent In the present study an improved formulation has been pro-
walls, according to the procedure specified in EN ISO 10211. posed, deducing each equation through the observation of the
About 190 values of linear thermal transmittance were there- thermal field distortion phenomenon for each node and the associ-
fore calculated for each selected thermal bridge as a function of ated heat flow lines.
the input variables (which differ for each thermal bridge) and of Let us examine, for example, the thermal bridge B1, shown in
Fig. 4. The first part of the equation w01 describes the heat flow


some fixed input parameters and boundary conditions, as reported
in Tables 2 and 3. per unit length that affect the node (in this case, the thermal bridge
Each input variable was set according to discrete steps within a is represented by balcony slab), which is a function of the thickness
variation range that includes most of the current building envelope of the slab (Sb) and of the relative thermal resistance (Smas + Sins,w)/
kb. The second part of the equation w02 describes the heat flow per


design solutions. The values considered for each input variable in
order to evaluate the linear thermal transmittance are reported unit length in proximity of the node. This flow is a function of the
in the ‘‘Values’’ column of Table 4. For each thermal bridge all thickness of the wall and the slab and their thermal resistances. In
analogy with the first two parts of the equation, the third part w03


the possible combinations of these values for the different vari-
ables have been considered. is the heat flow that only affects the wall.
As far as the wall–roof junction is concerned, the temperature Fig. 5 shows the correlation between the linear thermal trans-
reduction factor (btr) was also identified as a design parameter in mittance values obtained using the detailed method and the values
order to take into account heat losses through unconditioned predicted by the identified inverse model for the thermal bridge
spaces, according to EN ISO 13789 [25]. B1.
In general the design parameters were considered as input vari- A non-linear regression model was specifically developed for
ables if their values significantly influence the variation of w values each analyzed thermal bridge using the above described general
or if their range of variation in current design solutions is relevant. equation shape on the basis of the flux flow lines and the depen-
This is for example the case of thermal conductivity of masonry dent variables. The regression equations and the estimated param-
wall, whose value considerably varies from clay blocks to light- eters for all the analyzed thermal bridges are reported in Table 1. In
weight masonry. Otherwise, other input parameters (e.g. the ther- the same table the fixed input data are also reported.
mal conductivity of the insulation layer and of the ground, the Table 5 reports the standard deviation and the average values of
window frame U-value) were considered fixed as their ranges of the linear thermal transmittance calculated using the method
variation are very strict and/or their variation has a very low influ- adopted in EN ISO 10211 (wcalc) and the expected linear thermal
ence on the deviation of the w values. These assumptions have transmittance values obtained from the regression models (wpred),
234 A. Capozzoli et al. / Applied Energy 107 (2013) 229–243

Table 1
Identification of analyzed thermal bridges and regression equations.

Description Regression equations

(a) Externally insulated wall
0:153Smas 0:048Sins;w
λmas Sins,r wR5 ¼ Smas Sr
Sins;w þ Smas Sins;w
1
þ btr þ1:795  ð2:465S
Sins;r
mas 2:644Sr Þ
Sr
þ 0:201S
Sr
mas þ0:183
Sins;r  0:077
kr þ kins þ0:298 kmas 0:008 kins þ0:006 kins 0:068kr þ0:178 kins þ kmas þ1:089
Sr
Fixed input data for R5: kr, kins
Smas
Sins,w

R5 - Wall – roof junction

λ mas wB1 ¼ Smas SbSþ0:130
ins;w
þ ð0:785S mas 1:007Sb Þ
Smas Sins;w
þ Smas 0:504Smas
Sins;w  0:189
kb þ kb þ0:217 1:368kmas þ kins 0:044 kmas þ2:876 kb þ0:046
Smas S Fixed input data for B1: kb, kins
b

Sins,w

B1 - Wall – balcony junction

0:327Smur
Sins,w wC1 ¼ Smas1:381S
Sins;w
mas
þ Smas Sins;w  0:051
kmas þ kins 0:097 kmas þ0:287 kins 0:080

0:332Smas ð2:055Smas Þ
wC14 ¼ Smas Sins;w þ Smas Sins;w  0:060
þ0:188 þ0:108 kmas þ1:273 kins þ0:305
λ mas kpil kins

Fixed input data for C1 and C14: kpil, kins

Smas

C1 – C14 - Convex corner

between walls / Convex corner
between walls with pillar
Sins,w wC5 ¼ 0:0910:697S
Smas Sins;w
mas
þ Smas0:705SmasSins;w
0:071
þ 0:019
kmas þ kins 0:40 kmas þ0:710 kins 0:324
λ mas ð0:105Smas Þ
wC17 ¼ Smas0:399S
Sins;w
mas
þ Smas Sins;w þ 0:017
Smas kpil þ kins þ0:111 kmas þ0:261 kins þ0:169

Fixed input data for C5 and C17: kpil, kins

C5 – C17 –
Concave corner between walls /
Concave corner between walls
with pillar
Sfl 0:149Smas 1:850Sfl
λ mas wIF1 ¼ Smas Sins;w þ Smas Sins;w
kfl þ0:589 kins þ0:101 kmas þ kins þ0:267
Smas
Sfl Fixed input data for IF1: kfl, kins

Sins,w

IF1 - Wall – intermediate floor

junction
ð1:007Sfl 2:357Smas þ0:341Þ
λ mas wGF5 ¼ S S
3:056Smas
Sins;w þ S Sins;fl
ð0:296Þ
þ Sins;w  0:048
Sins,w 1:811 kins;fl þ fl þ3:561 Smas
kgr þ3:118kmas þ0:504 0:159kfl þ 1:783B0 þ0:223B02 þ3:052 2:729kSmas þ kins þ0:090
mas k fl fl kins mas
Smas Sins,fl 0
Fixed input data for GF5: kfl, kins,,kgr, B
Sfl

GF5 - Junction wall – slab-on-

ground floor
ð0:507Smas 0:134Þ
Sins,w wW7 ¼ SmasSmas þ Sins;w
kmas þ0:167 1:546Skmas
mas
þ kins 0:018

0:445Smas þ0:030 ð0:271Smas 0:241Þ

wW12A ¼ Smas þ Sins;w þ 0:059
λ mas kmas þ0:270 1:330Skmas
mas
þ kins þ1:025

Smas Fixed input data for W7 and W12a:kins, Ufr

W7 - W12a - Junction wall –
window frame with continuous
insulation
Spil ð0:838Spil Þ
Sins,w wP5 ¼ Smas Sins;w þ Smas Sins;w
þ0:864 kins þ0:166 kmas þ0:746 kins þ0:049
Smas kpil

Fixed input data for P5:kpil, kins

λ mas
Spil
P5 - Wall –pillar junction
 A. Capozzoli et al. / Applied Energy 107 (2013) 229–243 235

Table 1 (continued)

Description Regression equations

(b) Wall with intermediate insulation
SM1 8:306Sins;r 14:444Sins;w 7:035SM1 5:275SM2 þ1:72Sr þ3:941 1 ð2:676Þ 1
λ mas Sins,r wR6 ¼ S S SM1
þ S S Sins;w  btr þ3:719 þ Sins;r  btr þ1:821
M1 þ ins;r þ2:329
0:589Skrr þ0:553kmas kmas kr
M1 þ M2 þ
0:415kmas kmas kins þ0:461 kins 0:652kSrr þ0:469
Sr
Fixed input data for R6: kr, kins
SM1 SM2
Sins,w

R6 - Wall – roof junction

0:339SM1 þ0:339SM2 0:161Sins;w þ0:199Sb
λ mas wB2 ¼ SM1 þSins;w þSM2Sþ0:785S
b
b
þ ð0:333SM1 0:334SM2 1:035Sb Þ
Sins;w S
M1 þSM2
þ SM1 þSM2 Sins;w 0:186Sb
kb þ0:048 kins þ kmas kmas þ k þ k þ0:1
b b
Sins,w S Fixed input data for B2: kb, kins
b

SM1 SM2

B2 - Wall – balcony junction

6:061SM1
Sins,w wC10 ¼ M1 þ S Sins;w SM1 SM2 SM2
þ ð2:182S
Sins;w
M1 1:388SM2 0:114Þ
SM1 þSM2
þ 0:158
3:451kmas kpil þ3:274 kpil þ kpil þ0:795kmas þ0:445 kins þ1:117 kmas 0:089

SM1 ð2:279SM1 1:771SM2 2:497Sins;w 0:321Þ

wC11 ¼ SM1 Sins;w S
M1 SM2 SM2
þ Sins;w S
M1 SM2
þ 0:265
λ mas kmas þ0:31 kpil þ kpil 0:31 kpil þ0:205kmas þ0:057 kins þkmas þkmas þ0:28

SM1 Fixed input data for C10 and C11: kpil, kins
SM2

C10 – C11 - Convex corner

between walls with pillar
1:964SM1 þ0:882SM2
Sins,w wC18 ¼ M1 þ S Sins;w S
M1 SM2 SM2
þ Sins;w0:099SM2SM1
0:74SM1 0:215
SM2
þ 0:324
0:814kmas kpil þ kpil þ1:725 kpil þ1:108kmas þ0:087 þ0:569kmas þ2:622kmas þ0:196
λ mas kins

SM1 0:24SM2 0:051SM1 0:889Sins;w 0:637

SM1 wC19 ¼ SM1 Sins;w S
M1 SM2 SM2
þ Sins;w S
M1 SM2
þ 0:334
kmas þ0:34 kpil þ kpil 0:265 kpil þ0:082kmas þ0:057 kins þkmas þkmas þ1:034
SM2
Fixed input data for C18 and C19: kpil, kins

C18 – C19 - Concave corner

between walls with pillar
S þ0:075 ð2:396Sfl 0:214Þ
λ mas wIF2 ¼ fl
SM1 þSM2 Sins;w
þ Sins;w SM1 þSM2
þ 0:368SM1 þ0:311SM2 þ0:019
SM1 þSM2 Sins;w  0:166
0:847 þ k þ0:137 þ þ0:679 þ2:174 þ0:043
Sins,w kfl fl kins kmas kmas kfl

Sfl ð1:353Sfl Þ 0:151SM1 þ0:086SM2þ0:054

Sfl wIF5 ¼ SM1 1:016SM2 Sins;w SM2
þ Sins;w SM1 þSM2
þ S
M1 þ0:354 SM2 kmas Sins;w
 0:004
kfl þ kfl þ0:179kmas þ0:564 kins þ kmas 0:168 0:63kmas þ kfl þ0:037

Fixed input data for IF2 and IF5: kfl, kins

SM1
SM2
IF2 – IF5 - Wall – intermediate
floor junction
ð1:253SM1 0:693SM2Þ 0:6440:016B0 0:373Sfl þ2:938
Sins,w λ mas wGF6 ¼ SM1
SM1 SM1 þSins;w þSM2 kfl Sfl
þ SM1 SM2 kmas Sins;w
þ Sins;fl S þS SM1 þ13:652SM2þS kmas  0:196
1:2kfl 1:069B0 þ11:573 ins;fl Sfl ins;w þ15:306
kmas þ þ kfl þ0:318 kmas þ þ kins 0:03 kins fl kins þkfl þ  4:153
SM2 SM1 Sins,fl
0
Sfl Fixed input data for GF6: kfl, kins, kgr, B

GF6 - Junction wall – slab-on-

ground floor
0:523SM1 þ0:602SM2 þ0:181Sins;w þ0:057
SM2 wW8 ¼ SM1 þSM2 S þ 0:116SM10:012
þSins;w þ0:054SM2  0:012
þ0:015 ins;w
Sins,w kmas kins kmas

0:178Sins;w 0:137SM1 0:142SM2 0:058 0:024Sins;w þ0:427SM1 þ0:441SM2

wW11 ¼ SM1 þSM2 S þ Sins;w SM1 þSM2
þ 0:022
kmas þ0:174 ins;w
kins þ1:169 1:693 kins þ kmas

λ mas Fixed input data for W8 and W11: kins, Ufr

SM1
W8 - W11 - Junction wall –
window frame with interrupted
insulation/ with continuous
insulation
pil S ð1:028Spil 0:383SM1 0:299SM2 0:396Sins;w Þ 0:49SM1 þ0:285SM2 0:16Sins;w þ0:058
wP2 ¼ SM1 þSM2 þSins;w þ SM1 þSM2 Sins;w
þ SM1 þSM2 Sins;w
Sins,w kpil þ0:158 kmas þk kmas þ k þ0:157
ins pil
λ mas Spil ð1:457Spil Þ
SM2 wP7 ¼ SM1 1:019SM2 þSins;w þ 1:348SM1 þ0:809SM2 0:321SM1 þ0:177SM2 þ0:075
þ 0:96S  0:031
Spil kpil
S
M2 þ0:713
þ0:148kmas kmas þ
Sins;w
kins
M1 þ0:369SM2
kmas
Sins;w
þ kpil þ0:068
SM1
ð1:324Spil 0:494SM2 Þ Spil þ0:713SM2
wP9 ¼ SM1 þSM2 Sins;w
þ SM1 þSins;w S
M2 þ0:85
kmas þ k þ0:261 kpil þ0:31kmas
ins

P2 - P7 - P9 - Wall –pillar Fixed input data for P2, P7 and P9: kins, kpil
junction
(continued on next page)
236 A. Capozzoli et al. / Applied Energy 107 (2013) 229–243

Table 1 (continued)

Description Regression equations

(c) Internally insulated wall

0:12Smas 0:021Sins;w
λ mas Sins,r wR7 ¼ Sins;w
Sr
Smas
þ Smas Sins;w þ b1tr  ð0:123Smas 0:336Sr Þ
Sins;r þ btr  ð0:036Sr Þ
Sins;w  0:063
0:712kSrr þ kr þ0:892 kr þ0:086 kmas þ0:006 kins þ0:018
Sr
kr þ0:324 kins þ0:029 0:026Skmas
mas
þ Sr
kr 0:007kr
Sr
Fixed input data for R7: kr, kins
Smas
Sins,w

R7 - Wall – roof junction

λ mas wB3 ¼ Smas SbSþ0:136 þ Smas 0:413Smas
Sins;w þ ð0:014Smas 0:02Sb Þ
Sins;w
kb
ins;w
þ kb þ0:233 kmas þ0:056 kins þ0:035 0:029kSmas
mas
þ kb 0:188
Smas S Fixed input data for B3: kb, kins
b

Sins,w

B3 - Wall – balcony junction

1:539Smas 0:954Sins;w
Sins,w wC3 ¼ Smas Sins;w
kmas þ0:713 kins þ0:167

1:93Smas 0:216Smas 0:554Sins;w

wC9 ¼ Smas Sins;w þ Sins;w S
mas
kmas þ0:75 kins þ0:145 0:421 kins þ kpil þ0:038
λ mas
Fixed input data for C3 and C9: kpil, kins
Smas

C3 – C9 - Convex corner between walls

/ Convex corner between walls with
pillar
0:458Smas
Sins,w wC7 ¼ Smas Sins;w
kmas þ0:234 kins þ0:02
λ mas 0:237Smas þ0:114Sins;w
0:54Smas
wC22 ¼ Smas Sins;w þ Sins;w S
Smas kmas þ0:581 kins þ0:175 0:215 mas
kins þ kpil þ0:05

Fixed input data for C7 and C22: kpil, kins

C7 – C22 -
Concave corner between walls / Concave
corner between walls with pillar
Sfl þ0:086 0:27Smas 0:84Sfl
λ mas wIF3 ¼ Smas Sins;w þ Smas Sins;w
þ Smas0:263SmasSins;w  0:127
kfl þ kfl þ0:184 kmas þ kins 0:082 kmas þ1:825 kfl
Smas
Sfl Fixed input data for IF3: kfl, kins

Sins,w

IF3 - Wall – intermediate floor junction

1:008Smas 0:972Sfl 0:113Smas þ0:082Sfl
λ mas wGF7 ¼ S Sins;fl þ Sins;w  0:055
Smas 1:594kfl þ kins þ0:664 kins 0:057Skmas
mas
þ0:563
fl
Sins,w Sins,fl
Fixed input data for GF7: kfl, kins, kgr, B0
Sfl

GF7 - Junction wall – slab-on-ground

floor
λ mas wW9 ¼ 0:475Smas 0:134
Sins;w þ SmasSmas
1:219Skmas
mas þ
kins þ0:478 kmas þ0:292
Smas
wW12 ¼ 0:127S mas 0:241
Smas Sins;w
þ 0:414S
Smas
mas þ0:024
þ 0:055
kmas þ kins þ1:412 kmas þ0:448

Fixed input data for W9 and W12:kins, Ufr

Sins,w
W9 - W12 - Junction wall – window
frame with interrupted insulation/
Junction wall – window frame with
continuous insulation
Spil þ0:04 0:706Spil
λ mas wP6 ¼ Smas Sins;w þ Smas Sins;w
þ kins þ0:202 kmas þ0:655 kins
Spil kpil

Fixed input data for P6:kpil, kins

Sins,w
Smas
P6 - Wall –pillar junction
 A. Capozzoli et al. / Applied Energy 107 (2013) 229–243 237

Table 2
Fixed input data used for the calculation of the linear thermal transmittance.

Parameter Symbol Value Unit

Insulation thermal conductivity kins 0.04 W m1 K1
Concrete thermal conductivity of the floor, roof, balcony or pillar kfl, kr, kb, kpil 2.0 W m1 K1
Ground thermal conductivity kgr 2.0 W m1 K1
Thermal transmittance of the window frame Ufr 2.0 W m2 K1
Thickness of the window frame Sfr 0.06 m
Characteristic dimension of the floor B0 2.0 m

Table 3 set of transformations functions Gi and integer frequencies xi using

Fixed boundary conditions used to calculate the linear thermal transmittance. the Monte Carlo method [27].
Parameter Symbol Value Unit About 1000 cases were generated for each thermal bridge using
External surface thermal resistance Rse 0.04 m2 K W1
a combination of input variables and considering a variation for
Internal surface thermal resistance Rsi 0.13 m2 K W1 each of them within the range specified in the ‘‘Range’’ column
Indoor temperature hi 20.0 °C of Table 4.
Outdoor temperature he 0.0 °C The linear thermal transmittance values of the thermal bridges
were identified for each case through the developed regression
models. All the combinations of the input variables were consid-
as well as of the residual values of all the considered thermal
ered excluding only the cases for which the resulting envelope
bridges. The relevant correlation coefficient adjusted R-squared is
thermal transmittance was out of the validity range.
also reported for each thermal bridge. Examining these indices, it
The ANOVA technique is used to split the variance of the output
is possible to appreciate that the models developed are well able
variable between the different input variables. The variance is a mea-
to represent the 2D-w values.
sure of the dispersion of the output variable, which in our case, is rep-
Only the thermal bridges in the case of externally insulated wall
resented by the linear thermal transmittance of each thermal bridge.
are reported in Table 5. However, similar results were obtained for
Using this technique, it is then possible to evaluate the contri-
the inverse models developed to fit the 2D-w values for the other
bution of each input variable to the variability of the linear thermal
thermal bridges considered in this paper (intermediate and inter-
transmittance from a quantitative point of view. Furthermore,
nal insulation).
using the ANOVA method, it is possible to evaluate the effect of
The adjusted R-squared of the identified regression models are
the interactions between the input variables on the variability of
on average very close to 100% for all the analyzed thermal bridges.
the output. Consequently, adopting this methodology, it has been
The proposed models are a simple way to obtain w values with rel-
possible to carry out an analysis with the aim of assessing the rel-
ative errors very similar to those obtained by numerical
evance of each input design variable. The technique is based on the
calculations.
decomposition of the model variance into:

 A first order sensitivity index, SIi, caused by each input factor.

5. Sensitivity analysis  A second order sensitivity index, SIi,j caused by the interactions
between any pair of input factors i and j (–i), which cannot be
A sensitivity analysis was conducted by generating a random explained by the sum of the individual effects due to the two
sample of the possible combinations of physical parameters that factors.
determine the linear thermal transmittance value for each ana-  Analogously, higher order sensitivity indexes (SI1. . .i. . .k) caused
lyzed thermal bridge. by the interactions between more than two factors.
The sample was generated and the sensitivity analysis was con-
ducted using the ANOVA–FAST technique [23,26,27] by means of A total sensitivity index, SITi, can be defined for each factor,
SimLab 2.2 code [28]. which is the sum of the first order sensitivity index of the factor
The FAST method is based on a transformation that converts the under investigation and the higher sensitivity indices involving
variance of a variable Y, which is a k-dimensional integral, to a sin- this factor.
gle dimensional integral with respect to a scalar variable s, by The first order sensitivity index related to each input factor
transforming each input factor Xi into the form Xi = Gi(sin (xis)). represents the impact of the variation of that factor on the
The different sensitivity indexes can be evaluated for an appropriate variance of the linear thermal transmittance. Some of the effects

Table 4
Input design variables: values considered to calculate the linear thermal transmittance and ranges considered to generate the sample.

Variable Symbol Values Range Unit

Thickness of the intermediate floor, roof or balcony Sfl = Sr = Sb [0.15, 0.20, 0.25] [0.15–0.25] m
Thickness of the pillar Spil [0.30, 0.40] [0.30–0.40] m
Thickness of the masonry Smas [0.20, 0.30, 0.40] [0.20–0.40] m
Thickness of the masonries for a double brick wall with intermediate insulation SM1 [0.10, 0.20, 0.25, 0.30] [0.1–0.3] m
SM2 [0.10, 0.20, 0.25, 0.30] [0.1–0.3]
Thickness of the floor or roof insulation Sins,fl, Sins,r [0.0453–0.3902] m
? Thermal transmittance of the floor or roof Ufl, Ur [0.10, 0.20, 0.30, 0.40, 0.50, 0.60, 0.70] ? [0.10–0.70] W m2 K1
Wall insulation thickness Sins,w [0.0023–0.3843] m
? Wall thermal transmittance Uw [0.10, 0.20, 0.30, 0.40, 0.50, 0.60, 0.70] ? [0.10–0.70] W m2 K1
Thermal conductivity of the masonry kmas [0.25, 0.50, 0.90] [0.25–0.90] W m1 K1
Temperature reduction factor btr [0.50, 0.65, 0.80, 1.0] [0.50–1.0] –
238 A. Capozzoli et al. / Applied Energy 107 (2013) 229–243

Fig. 3. Geometrical model of a 2D thermal bridge.

Fig. 4. Geometrical model of thermal bridge B1 and explanation of the regression equation.

Table 5
Statistical indices for the calculated and predicted linear thermal transmittance referring to the thermal bridges in the case of the externally insulated wall.

Thermal bridge code Mean value (W/(m K)) Standard deviation (W/(m K)) Adjusted R2
wcalc wpred Residuals wcalc wpred Residuals
B1 0.637 0.630 0.000 0.136 0.137 0.008 0.997
C1 0.147 0.147 0.000 0.078 0.078 0.005 0.997
C5 0.040 0.039 0.000 0.020 0.021 0.002 0.989
W7 0.301 0.300 0.001 0.154 0.154 0.008 0.998
R5 0.063 0.063 0.000 0.068 0.067 0.006 0.994
GF5 0.009 0.008 0.000 0.108 0.108 0.008 0.995
C14 0.090 0.090 0.000 0.042 0.041 0.003 0.995
C17 0.067 0.067 0.000 0.065 0.065 0.004 0.995
W12A 0.166 0.167 0.000 0.069 0.069 0.006 0.992
P5 0.077 0.077 0.000 0.128 0.128 0.005 0.998

on the output variable are due to the interactions between is not due to linear effects and which can therefore be interpreted
the input variables and are therefore not explained by the first or- as a measure of the non-linearity of the model.
der index. The share allocated to the interaction item represents The sensitivity analysis was performed for each thermal bridge
the portion of the variance of the linear thermal transmittance that according to the following phases:
 A. Capozzoli et al. / Applied Energy 107 (2013) 229–243 239

2
R = 0.997

Calculated linear thermal transmittance

1.20

1.00

0.80

(W m -1 K-1)
0.60

0.40

0.20

0.00
0 0.2 0.4 0.6 0.8 1 1.2
-1 -1
Predicted linear thermal transmittance (W m K )

Fig. 5. Correlation between the calculated linear thermal transmittance and that predicted by the regression model. Thermal bridge B1.

Table 6
First-order sensitivity indices – linear thermal transmittance.

 Definition of the range of each variable to cover most of the cur- validity of each model regression was analyzed; the uniform
rent building envelope design solutions, considering the same distribution is a bounded continuous distribution over an inter-
probability of occurrence for each selected amplitude. val [a, b] so that each variable, with a value between the bounds,
 Definition of a probability density function for the input has an equal probability.
variables, choosing a uniform function in the range of variability  Definition of fixed input parameters and boundary conditions
considered in the ‘‘Range’’ column of Table 4 for which the during the simulations (see Tables 2 and 3).
240 A. Capozzoli et al. / Applied Energy 107 (2013) 229–243

 Generation of a matrix whose number of rows (n) corresponds variables that can be considered irrelevant according to literature
the number of cases generated while the number of columns indications [27]. For the same thermal bridges, the most effective
(k) corresponds to the number of input variables. variables are also illustrated in Table 8 with different colors in or-
 Application of the developed regression model (for each ther- der to better identify the ranking evaluated by the total order sen-
mal bridge) to generate an output vector of the linear thermal sitivity indices analysis. This Table can be useful for the designers
transmittance values with a number of rows equal to n. and policymakers, because it identifies for each thermal bridge the
 Evaluation of the first order sensitivity indices for each input design variables that need more attention for their effect on ther-
variable using the ANOVA–FAST analysis. mal bridging and on building heat losses and energy performance.
 Evaluation of the total order sensitivity indices for each input As far as the junction between the wall and roof, with both an
variable using the ANOVA–FAST analysis. external and intermediate insulated wall (R5 and R6) is concerned,
the thermal insulation thickness of the roof and the wall, and
6. Results and discussion hence their relative thermal resistance, have the most effect on
the variance of the linear thermal transmittance. In the case of
The results relative to the sensitivity analysis are presented in internal insulation (R7), the thermal insulation roof and the roof
this section. The weight that the uncertainty of each input variable thickness have an important impact.
has on the uncertainty of the output variable (linear thermal trans- The variables that have the greatest weight in the wall – bal-
mittance w) has been evaluated for each type of thermal bridge. cony junction (B1, B2 and B3) are the balcony thickness and subse-
The first and total order indices of each variable are reported in quently the thermal resistance of the massive part of the floor, as
Tables 6 and 7 for all the thermal bridges analyzed with the ANO- well as the thermal insulation thickness of the wall. Moreover
VA–FAST technique using the developed inverse models to gener- the internal masonry part of the wall (SM1) has an important im-
ate the output vector. In Table 6 the first-order sensitivity indices pact for B2.
related to each thermal bridge higher than 0.20 have been The most important variable for the convex and concave cor-
highlighted, while in Table 7 total order sensitivity indices related ners between the walls, with and without pillar (C1, C5, C14,
to each thermal bridge higher than 0.25 have been highlighted. C17, C10, C11, C3, C9, C7, and C22) is the insulation thickness of
These values have been conventionally selected to exclude those the wall and hence the thermal resistance insulation layer. The

Table 7
Total sensitivity indices – linear thermal transmittance.
Table 8
Identification of thermal bridges most effective variables (in red, orange and yellow colors are shown the first, the second and the third most important variables respectively, with relation to total sensitivity indices).

Wall–roof junction Sins,fl Sins,fl

Sins,fl

Sins,w Sfl
Sins,w

R5 R6 R7
Wall – balcony junction SM1
Sfl Sfl Sins,w

Sins,w Sfl
Sins,w

A. Capozzoli et al. / Applied Energy 107 (2013) 229–243

B1 B2 B3
Convex corner between walls/convex corner between walls with pillar

Sins,w
Sins,w Sins,w Sins,w Sins,w Sins,w

C1 – C14 C10 – C11 C3 – C9

Concave corner between walls/concave corner between walls with pillar
λ mas λ mas Sins,w
Sins,w Sins,w Sins,w Sins,w

C5 – C17 C18 – C19 C7 – C22

Wall – intermediate floor junction
Sins,w Sins,w Sins,w
Sins,w

SM1 λ mas Sfl

IF1 IF2 – IF5 IF3

Junction wall – slab-on-ground floor λ mas
Sins,w
Sins,w

λmas Sins,fl

GF5 GF7
GF6

(continued on next page)

241
 242 A. Capozzoli et al. / Applied Energy 107 (2013) 229–243

thermal conductivity of the masonry has the most important im-

pact on the variance of linear thermal transmittance in the concave
Sins,w λ mas

corner between the walls with a pillar (C18 and C19). The thickness
of insulation layer is also non-negligible for thermal bridge C19.
The most important variable for the junction wall – insulated
W9 - W12

pillar with an externally and internally insulated wall (P5 and

Sins,w P6) is the insulation thickness of the wall and hence the thermal
Sins,w λ mas

resistance insulation layer. The most significant variable for the

P6
case of a double brick wall with intermediate insulation in the
thermal bridge P2 is the thickness of the pillar and of the insulation
layer. When the same junction is thermally corrected with an
external insulation panel on the pillar (P7), the thermal conductiv-
ity of the masonry becomes the most important variable followed
by the insulation layer. The thermal bridge P9 characterized by a

P2 - P7 - P9
correction all around the pillar with a continuous insulation layer
Sins,w

λ mas Sins,w
Spil Sins,w
W8 - W11

Sins,w
λ mas

is almost exclusively influenced by the wall insulation layer.

The most important variable for the wall – intermediate floor
junction in the case of an externally insulated wall (IF1) is the insu-
lation thickness of the wall. The thickness of the internal masonry
part of the wall (SM1) also has a large impact on the variance of lin-
ear thermal transmittance for thermal bridge IF2 as well as the
Sins,w

λ mas

wall insulation layer; in the case of the junction between the wall
and the intermediate floor with a correction (IF5), the most signif-
W7 - W12a

icant variables are the conductivity of the masonry and the wall
insulation layer. In the case of an internally insulated wall (IF3),
the variables which show the greatest impact are the thickness
Sins,w

P5
Sins,w

λ mas

of the insulation layer and the thickness of the floor.

The variables that have the greatest impact on the variance of
the linear thermal transmittance for the wall–slab-on-the ground
floor thermal bridge, in the case of an externally insulated wall
and double brick masonry, with intermediate insulation (GF5 and
Junction wall – window frame with interrupted insulation/junction wall – window frame with continuous insulation

GF6), are the insulation thickness of the wall and the thermal con-
ductivity of the masonry. In the case of an internally insulated wall
(GF7), the most important variable is the insulation thickness of
the slab on the ground floor.
The variables that have most impact on the junction wall – win-
dow frame, in the case of an externally and intermediate insulated
wall (W9, W12, W12a, and W7) are the insulation thickness of the
wall and the thermal conductivity of the masonry. In thermal
bridge W8 (wall – window frame with intermediate interrupted
insulation), the most significant variable is the thermal conductiv-
ity of the masonry; the only significant variable in the case of
intermediate continuous insulation (W11) is the insulation thick-
ness of the wall. In the analysis carried out the thermal transmit-
tance of the window frame has been considered constant as it
has been found that it plays a negligible role on the linear thermal
transmittance variation. For this reason the regression equations to
calculate the linear thermal transmittance of the junction
wall–window frame would be also valid for any frame in the range
from 2 to 3 W m2 K1.
The results in general show that in the case of an externally and
internally insulated wall, when the insulation layer is continuous
at the junction, the most important design variable is the insula-
tion thickness of the wall, while, in the case of interrupted
insulation, the variables that affect the thermal resistance of a
non-insulated junction (thermal conductivity of masonry, floor or
roof thickness, wall thickness) are also important. Moreover,
among the variables that affect the thermal resistance of a wall
junction, masonry conductivity has more effect than thickness on
Wall –pillar junction

the linear thermal transmittance variation. When the node is rep-

resented by a floor or a roof, the thickness is more important, be-
Table 8 (continued)

cause the concrete thermal conductivity is constant.

In the case of cavity insulation, since there is no continuous
insulation, the variance of the linear thermal transmittance is also
influenced by the thermal resistance of the junction and, above all,
by the thermal conductivity of the wall masonry.
 A. Capozzoli et al. / Applied Energy 107 (2013) 229–243 243

A thermal correction in a junction generally induces a change in validation of simplified building energy performance calculation
the most important variables affecting the variance of linear models’’.
thermal transmittance. In fact for thermal bridges C18–C19, IF2–
IF5, P2–P7–P9 the thermal correction influences the thermal field References
distortion phenomenon and the associated heat flow lines.
It is important to underline that the results of the sensitivity [1] European Commission. Directive 2010/31/EU of the European Parliament and
of the council of 19th May 2010 on the energy performance of buildings, OJEU,
analysis are affected by the relative amplitude of the selected L 153/13, 18/06/2010.
ranges for each input variable. For this reason the set of the most [2] Theodosiou TG, Papadopoulos AM. The impact of thermal bridges on the
typical design envelope solutions have been taken into account energy demand of buildings with double brick wall constructions. Energy Build
2008;40:2083–9.
to select the input variable range. [3] Cappelletti F, Corrado V, Gasparella A, Gorrino A. Detailed assessment of
thermal bridges in highly insulated buildings. In: CLIMA 2010, 10th Rehva
World Congress ‘‘Sustainable Energy Use in Buildings’’, Antalya, 9–12 May,
7. Conclusion
2010.
[4] Citterio M, Cocco M, Erhorn-Kluttig H. Thermal bridges in the EBPD context:
The paper presents a sensitivity analysis carried out on thirty- overview on MS approaches in regulations. In: ASIEPI information paper; 2008.
six thermal bridges, including the most typical junctions of the [5] Evola G, Margani G, Marletta L. Energy and cost evaluation of thermal bridge
correction in Mediterranean climate. Energy Build 2011;43:2385–93.
building envelope. The adopted methodology has allowed non-lin- [6] Ben-Nakhi AE. Minimizing thermal bridging through window systems in
ear regression models to be identified as a function of the design buildings of hot regions. Appl Therm Eng 2002;22:989–98.
variables of each thermal bridge, on the basis of linear thermal [7] Cappelletti F, Gasparella A, Romagnoni P, Baggio P. Analysis of the influence of
installation thermal bridges on windows performance: the case of clay block
transmittance values calculated using a numerical method accord- walls. Energy Build 2011;43:1435–42.
ing to EN ISO 10211. The results show that the statistics of the [8] EN ISO 14683. Thermal bridges in building construction – linear thermal
identified regression models are very good with an adjusted R- transmittance – simplified methods and default values; 2007.
[9] EN ISO 10211. Thermal bridges in building construction – heat flows and
squared on average very close to 100% for all the analyzed thermal surface temperatures – detailed calculations; 2007.
bridges. Under the validity range of each variable, all the proposed [10] UNI/TS 11300-1. Energy performance of buildings Part 1: evaluation of energy
models generally show relative errors very similar to those ob- need for space heating and cooling; 2008 [in Italian].
[11] EN 12831. Heating systems in buildings – method for calculation of the design
tained by numerical calculations. The equations can be used by heat load; 2003.
designers and practitioners for their reliability and simplicity com- [12] Martin K, Escudero C, Erkoreka A, Flores I, Sala JM. Equivalent wall method for
pared to numerical methods with a calculation error lower than dynamic characterisation of thermal bridges. Energy Build 2012;55:704–14.
[13] Dèquè F, Ollivier F, Roux JJ. Effect of 2D modelling of thermal bridges on the
using default w values provided by EN ISO 14683.
energy performance of buildings: numerical application on the Matisse
The variance of the linear thermal transmittance has been apartment. Energy Build 2000;33:583–7.
decomposed through the ANOVA–FAST analysis into the input [14] Gao Y, Roux JJ, Zhao LH, Jiang Y. Dynamical building simulation: a low order
variables considered for each type of thermal bridge. model for thermal bridges losses. Energy Build 2008;40:2236–43.
[15] Ben-Nakhi AE. Development of an integrated dynamic thermal bridging
The impact of each design variable on the variance of the linear assessment environment. Energy Build 2003;35:375–82.
thermal transmittance through the calculation of the first and total [16] Kosny J, Kossecka E. Multi-dimensional heat transfer through complex
order indices has been evaluated. It has been observed in a large building envelope assemblies in hourly energy simulation programs. Energy
Build 2002;34:445–54.
number of simulations that the linear thermal transmittance val- [17] Ascione F, Bianco N, de’ Rossi F, Turni G, Vanoli GP. Different methods for the
ues vary significantly for changes in the considered design modelling of thermal bridges into energy simulation programs: comparisons
variables. of accuracy for flat heterogeneous roofs in Italian climates. Appl Energy
2012;97:405–18.
For almost each analyzed thermal bridge, the thickness of the [18] Martin K, Erkoreka A, Flores I, Odriozola M, Sala JM. Problems in the
insulation layer is one of the most important variables affecting calculation of thermal bridges in dynamic conditions. Energy Build
the deviation of linear thermal transmittance. Whenever the insu- 2011;43:529–35.
[19] Dilmac S, Guner A, Senkal F, Kartal S. Simple method for calculation of heat loss
lation layer does not cover the junction, the thermal resistance of through floor/beam-wall intersections according to ISO 9164. Energy Convers
the non-insulated junction is also significant. Moreover, among Manage 2007;48:826–35.
the variables affecting the thermal resistance of a wall junction, [20] Asdrubali F, Baldinelli G, Bianchi F. A quantitative methodology to evaluate
thermal bridges in buildings. Appl Energy 2012;97:365–73.
the masonry thermal conductivity has a greater effect on the linear
[21] Ben Larbi A. Statistical modelling of heat transfer for thermal bridges of
thermal transmittance variation than its thickness. buildings. Energy Build 2005;37:945–51.
A thermal correction of a junction generally determines a [22] Capozzoli A, Corrado V, Gorrino A, Soma P. Atlante nazionale dei ponti termici.
change in the most w value influencing design variables. Conforme alle UNI EN ISO 14683 e UNI EN ISO 10211. Borgomanero (Italy):
Edizioni Edilclima; 2011.
Since a variation of the linear thermal transmittance affects the [23] Mechri HE, Capozzoli A, Corrado V. Use of the ANOVA approach for sensitive
building heating energy need, the performed sensitivity analysis building energy design. Appl Energy 2010;87:3073–83.
has allowed the design variables that have the highest influence [24] Trisco, Kobru86. Computer program to calculate 2D and 3D steady state heat
transfer in rectangular objects. PHYSIBEL; 2007.
on building energy performance to be identified. Moreover, to this [25] EN ISO 13789. Thermal performance of buildings – transmission and
purpose a Table has been presented as a practical and useful tool ventilation heat transfer coefficients – calculation method; 2007.
for designers and policymakers. [26] Fang S, George Z, et al. Improved generalized Fourier amplitude sensitivity test
for model assessment. Stat Comput 2003;13:221–6.
[27] Saltelli A, Chan K, Scott EM. Sensitivity analysis: gauging the worth of scientific
Acknowledgement models. 1st ed. England: John Wiley & Sons; 2000.
[28] JRC. European Commission Joint Research Centre; 2002. <http://
[Link]>.
This work has been carried out as part of a research activity
supported by Edilclima S.p.A. concerning the ‘‘Development and