Building and Environment 54 (2012) 186e196

Contents lists available at SciVerse ScienceDirect

Building and Environment

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

Verication and validation of EnergyPlus phase change material model

for opaque wall assemblies
Paulo Cesar Tabares-Velasco*, Craig Christensen 1, Marcus Bianchi 2
National Renewable Energy Laboratory, 1617 Cole Blvd. MS 5202, Golden, CO 80401, USA

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

Article history: Phase change materials (PCMs) represent a technology that may reduce peak loads and HVAC energy
Received 8 December 2011 consumption in buildings. A few building energy simulation programs have the capability to simulate
Received in revised form PCMs, but their accuracy has not been completely tested. This study shows the procedure used to verify
16 February 2012
and validate the PCM model in EnergyPlus using a similar approach as dictated by ASHRAE Standard 140,
Accepted 18 February 2012
which consists of analytical verication, comparative testing, and empirical validation. This process was
valuable, as two bugs were identied and xed in the PCM model, and version 7.1 of EnergyPlus will have
Keywords:
a validated PCM model. Preliminary results using whole-building energy analysis show that careful
Phase change materials
Validation
analysis should be done when designing PCMs in homes, as their thermal performance depends on
Building energy simulation several variables such as PCM properties and location in the building envelope.
Building envelope 2012 Elsevier Ltd. All rights reserved.
PCM
Storage

1. Introduction impregnated PCM wallboards [9,10]. More recent studies of PCM

Phase change materials (PCMs) represent a technology that may
reduce peak loads and HVAC energy consumption in buildings.
PCM research has been broad on many heat transfer applications
during recent decades, resulting in a considerable amount of
literature about PCM properties, indoor temperature stabilization
potential, and peak load reduction potential. PCM-oriented
research on buildings has investigated two primary applications:
passive and active building systems [1e5]. Previous PCM studies
have shown that important benets are related to thermal comfort,
energy savings, and perhaps HVAC downsizing when thermal
storage is added to buildings [6].
Early numerical studies analyzed PCM wallboards during late
1970s and 1980s, mainly for PCM-impregnated wallboard appli-
cations [7,8]. Later numerical studies used PCM thermal properties
obtained from differential scanning calorimetry (DSC) techniques
and from measurements of temperature proles across
* Corresponding author. Tel.: 1 303 384 7591; fax: 1 303 384 7495.
E-mail addresses: paulo.tabares@nrel.gov (P.C. Tabares-Velasco),
Craig.Christensen@nrel.gov (C. Christensen), Marcus.Bianchi@owenscorning.com
(M. Bianchi).
1 varied from early PCM models [21,24], to empirical models using an
Tel.: 1 303 384 7510; fax: 1 303 384 7540.
2
Present address: Owens Corning Science Technology Center, 2790 Columbus
Road, Granville, OH 43023-1200, USA. Tel.: 1 720 258 6822.
wallboards have analyzed building elements containing micro-
encapsulated PCMs. One compared the performance of a detailed
model considering a solideliquid interface against a simpler model
using equivalent heat capacity model. Overall, the equivalent heat
capacity method performed better than experimental data [11].
Another study found the optimum time and space steps for
a specic PCM using heat capacity method, DSC material property
data, and an implicit nite difference model [12]. A follow-up study
concluded that hysteresis should be considered in the numerical
simulations to improve the overall accuracy of the model [13].
Numerical research on PCM-enhanced building enclosure systems
has followed a similar approach to wallboards, analyzing building
enclosure systems containing PCM between two layers of insu-
lation [14,15].
These studies focused on understanding the physics behind
PCMs. They validated models and investigated heat ux reduction
potential using special laboratory and small-scale test rooms. Other
studies have attempted to estimate potential energy savings
through building energy simulation (BES). Energy simulation
studies analyzing energy and peak load benets from PCMs have
used commercial BES software such as CoDyBa [16], EnergyPlus
[17], ESP-r [18e20], and TRNSYS [21e23]. The implemented models
equivalent heat transfer coefcient [22], to fully implemented
nite difference models [17,23] and control volume models [18].

0360-1323/$ e see front matter 2012 Elsevier Ltd. All rights reserved.
doi:10.1016/j.buildenv.2012.02.019
 P.C. Tabares-Velasco et al. / Building and Environment 54 (2012) 186e196
Predictions from these studies included no signicant energy
benets [17]; improved thermal comfort and decreased peak load
[21,24]; and 90% reduction of heating energy demand during the
heating season [18]. Likewise, an energy simulation study focusing
on building enclosure systems predicted a 19e57% peak load
reduction for an attic system consisting of a PCM sandwiched
between two conventional insulation layers [15].
In conclusion, PCMs have different benets depending on
quantity and types (phase change temperature, energy storage
capacity), locations (drywall, walls, attic, and oor) and climates
(heating and/or cooling performance). Therefore, there are clear
differences between the methods and results from previous
research efforts as: (1) the reviewed building energy simulation
studies, except a TRNSYS model that works for certain exterior PCM
applications, did not perform comprehensive model PCM valida-
tion [25] and (2) previous studies do not cover a wide range of PCM
types, locations, and climates. Thus, there is a simulation and
analysis gap with respect to PCM benets and modeling. The
objective of this study is to verify, validate, and improve the Ener-
gyPlus PCM model, which has not been fully validated. This
procedure will be performed using analytical verication,
comparative verication, and empirical validation of three PCM
applications:
1) PCM distributed in drywall,
2) PCM distributed in brous insulation, and
3) Thin, concentrated PCM layers.
2. EnergyPlus PCM model
EnergyPlus PCM algorithm uses a one-dimensional conduction
nite difference (CondFD) solution algorithm.3 Recently, CondFD
was validated against multiple test suites [26]. This step was
necessary before any serious PCM validation efforts could be made,
as the PCM model relies on CondFD. Several xes were made during
the verication process.
The PCM model considered for verication and validation is
a modied version of EnergyPlus 6.0.0.037 that includes all xes
from the CondFD verication [26]. This version will be referred to as
v6 in this publication. The new and improved PCM/CondFD
model will be referred to as v7.1, as time constraints prevented
the xes, which will be described later, from being included in the
EnergyPlus version 7 release.
The CondFD algorithm in EnergyPlus uses an implicit nite
difference scheme, where the user can select Crank-Nicholson or
fully implicit. Eq. (1) shows the calculation method for the fully
implicit scheme inside a homogeneous material.



j1 j1 j1 j1
Ti
j1 j
 Ti Ti1  Ti Ti1  Ti
Cp rDX kW kE
Dt DX DX
where,


j1 j1
ki1 ki
kW
2 changed or tuned as a result of the verication. As a result, the PCM


j1
ki1 ki
j1
kE
2
3
The default conduction transfer function (CTF) solution algorithm in EnergyPlus
cannot simulate materials with variable properties such as PCMs.
187
where ki k(Tj1 i ) if thermal conductivity is variable,
T temperature, i node being modeled, i1 adjacent node to
interior of construction, i1 adjacent node to exterior of
construction, j1 new time step, j previous time step, Dt time
step, Dx nite difference layer thickness, Cp specic heat of
material, and r density of material.
In the CondFD algorithm, all elements are divided or discretized
automatically using Eq. (2), which depends on a space discretiza-
tion constant (c), the thermal diffusivity of the material (a), and the
time step. Users can leave the default space discretization value of 3
(equivalent to a Fourier number (Fo) of 1/3) or input other values.
r
p a$Dt
Dx c$a$Dt (2)
Fo
For the PCM algorithm, the CondFD method is coupled with an
enthalpyetemperature function (Eq. (3)) that the user inputs to
account for enthalpy changes during phase change [17]. The
enthalpyetemperature function is used to develop an equivalent
specic heat at each time step. The resulting model is a modied
version of the enthalpy method [17].
h hT (3)
hji  hij1
Cp* T j j1
(4)
Ti  Ti
where h enthalpy.
3. Verication and validation
Several studies have included validation exercises to assess the
credibility of PCM models [9e11,13,14]. Before validation, most
obtained the thermophysical material properties of the samples
through laboratory testing. However, there are differences between
the validation procedures followed in each study; approaches range
from qualitatively comparing surface temperature graphs
[9e11,13,14], heat uxes [14,27], or room air temperatures [28,29],
to validations that contained quantication of error caused by
spatial and time discretization [12] and quantication of an average
error [30].
This study follows a similar approach to ASHRAE Standard 140,
as the entire verication and validation scheme is divided into
analytical verication, comparative testing, and empirical valida-
tion [31]. Thus, the PCM model was veried using (1) an analytical
solution from the Stefan Problem as generalized by Franz Neumann
[32], (2) a numerical solution of a sine wave problem using a veri-
ed PCM model from Heating 7.3 [33], and (3) experimental data
from a hot box apparatus [34]. Moreover, the PCM model is tested
for building envelope applications only, as heat diffusion and
storage in the walls and attic were considered in the verication
(1)
and validation. This follows the previous approach used to verify
and validate the EnergyPlus CondFD algorithm [26]. It is important
to mention that: (1) the PCM model was not fully validated earlier,
and (2) the validation and verication process is not calibrating or
evaluating the model, because the model input would not be
verication and validation process found stability problems with
the v6 PCM model that were xed using an automatic under-
relaxation factor.
3.1. Analytical verication: Stefan Problem
The rst verication test was the analytical solution to the Ste-
fan Problem, which originally posed the problem for the thickness
188 P.C. Tabares-Velasco et al. / Building and Environment 54 (2012) 186e196

of polar ice. The solution was developed by Neumann for a semi-

innite wall. The assumptions are: (1) the wall is initially at
uniform temperature with PCM in liquid form above the melting
temperature (T(x,t) > Tmelting), (2) suddenly at time 0, the wall
outside surface temperature drops to 0  C (T(0,t) 0  C), (3) PCM
density is equal and constant in both the liquid and solid phases, (4)
PCM thermal conductivity and specic heat are constant in each
phase and can differ between the solid and liquid phases, and (5)
PCM has a xed melting temperature (no melting/freezing range)
[32].
The analytical solution was used to test the v6 PCM model for
the three PCM applications, assuming the properties and charac-
teristics shown in Table 1. Specic details include: (1) a zone with
one wall was used and (2) the walls were initially at a homogenous
temperature of 50  C and had a melting temperature range of
29.4e29.6  C to approximate the xed melting temperature
assumption in the Stefan Problem. The semi-innite wall was
simulated by a thick wall, so all surface temperatures were set in
EnergyPlus using the OtherSideCoefcient feature; the indoor air
temperature was set to the desired temperature and the inside
convective heat coefcient was set to 1000 W/m2C. More details
about similar approaches are in the literature [26].
As Table 1 shows, each PCM application was simulated in
EnergyPlus as a wall with a thickness representative of typical
material congurations, with further details as follows:
1. The distributed PCM in the insulation represents a 0.105 m
layer of insulation that could be installed in a wall cavity or
attic. In this study, the simulated wall was 1.5 m thick to meet
the semi-innite wall condition required by the Neumann
solution. However, only the rst 0.105 m was considered when
comparing simulations to the analytical solution, as this
represents the thickness of a PCM-enhanced portion in a wall
cavity.
2. The distributed PCM in drywall represents a 0.02 m layer of
drywall. In this study, the simulated wall was 0.3 m thick to
meet the semi-innite wall condition required by the Neu-
mann solution. However, only the rst 0.02 m was considered
when comparing simulations to the analytical solution, as this
represents the PCM-enhanced portion of the wall.
3. The concentrated PCM represents a 0.005 m layer that could be
installed between two layers of insulation or between insu-
lation and drywall. In this study, the simulated wall contained
only concentrated PCM and was 0.155 m thick to meet the
semi-innite wall condition required by the Neumann solution.
However, only the rst 0.005 m was considered when
comparing simulations to the analytical solution, as this
represent the PCM-enhanced portion of the wall.
Fig. 1. Outside surface heat ux calculated using analytical solution (Analytical) and
PCM model v6 for three node spacing values (dx, dx/3, dx/9).
All three numerical and analytical solutions show similar results
and the PCM model behaves similarly between cases. Thus, detailed
results are only shown here for the PCM-enhanced insulation.
Figs. 1and 2 show results for three node spacing values: 0.015 m
(the default, dx), 0.005 m (dx/3), and 0.00167 m (dx/9), which are
obtained using Fourier numbers equal to 1/3, 3, and 26 respectively.
These numbers seem high and would have caused problems in an
explicit scheme. However, this study used the fully implicit scheme,
which can work with higher Fourier numbers, and steadily decayed
to convergence [35,36]. Node spacing values refer to the distance
between adjacent nodes that discretized the analyzed wall. More
information about the discretization of walls in EnergyPlus can be
found in the literature [26,37].
Fig. 1 shows how the v6 PCM model solution for the default
node space (dx) oscillates around the analytical solution: the
temperature from the PCM model suddenly increases, then exhibits
almost constant or at behavior around the analytical solution. This
behavior was observed at the default node spacing value in each
solution for the three PCM applications, and in each solution for the
three PCM applications, but with a lesser magnitude, in the other
two node spacing cases (dx/3 and dx/9). The reason for this is
mainly numerical; the solution requires smaller node spacing than
the default node space in EnergyPlus to properly simulate heat
wave propagation through the PCM with a xed melting temper-
ature. In this case, the oscillations were almost eliminated with
node spacing nine times smaller than the default node spacing.
This study also analyzed the model stability and convergence by
recording the number of iterations required for the PCM model to
converge. In the three discretization cases, the solution did not

Table 1
PCM strategies tested in analytical verication.

Properties PCM- PCM-Drywall Concentrated

insulation thin PCM layer
% by weight of 20% 30% 100%
microencapsulated
PCM in wall layer
Equivalent latent heat of wall 34 33 80
layer (kJ/kg)
Thermal conductivity (W/m K) 0.03365 0.2300 0.1600
Density (kg/m3) 29 1000 850
Specic heat (J/kg K) 960 1400 2500
Thickness (m) 0.105 0.020 0.005
Default node spacing 0.015 0.005 0.005 Fig. 2. Node temperature 0.105 m from outside surface calculated using analytical
solution (Analytical) and PCM model v6 for three node spacing values (dx, dx/3, dx/9).
 P.C. Tabares-Velasco et al. / Building and Environment 54 (2012) 186e196 189

converge when a node entered or left the melting range. Thus, after
CondFD passed the maximum allowable iterations, the solution did
not meet the convergence criteria. Therefore, the PCM model used
the values from the last iterations and moved to the next time step,
even though it did not converge. This problem did not cause any
signicant accuracy issues for the three cases tested, because of the
small time step used, but did increase the run time signicantly,
taking 6, 15, and 66 min to simulate the wall for 2 days in the dx, dx/
3, dx/9 scenarios, respectively.
The convergence problem was solved by adding an automatic
and dynamic under-relaxation factor after a determined number of
iterations. This new under-relaxation factor did not affect the
accuracy of the PCM model and reduced the run time 60e90%
relative to the v6 model. It is important to mention that the
semi-innite wall problem stated by the Neumann solution is Fig. 3. Calculated PCM-Insulation middle node temperature (Tmid) using comparative
demanding and challenging from a numerical point of view. A very software (H73), EnergyPlus PCM model with default node spacing (E), smaller node
space (Edx/3), 2- and 4-min time steps (E2 min, E4 min) and without PCM
close match was obtained only with Heating 7.3 using a time step of
(ENoPCM).
0.01 s and node spacing roughly 0.01 times the default size in
EnergyPlus. These values are stricter than those recommended in
previous studies [25]; they are likely overly strict and impractical  Indoor convective heat transfer coefcient: hi 5 W/m2 K
for modeling more realistic PCM-enhanced wall congurations in  Outside air temperature: 25  C at t 0; Varying sinusoidal over
annual building simulations. As a result, the PCM model must be 24 h with a range 10e40  C
veried with more realistic wall and boundary conditions. There is  Outdoor convective heat transfer coefcient: hout 20 W/m2 K
no analytical solution for this case, so another veried numerical  Wall initial conditions: homogenous temperature of 25  C
model needs to be used to conduct comparative testing.  Ideal/pure PCM with a xed melting temperature

3.2. Comparative testing against Heating 7.3
The analytical verication enabled us to detect run time issues
and identify node spacing requirements for an extreme circum-
stance with a large and sudden temperature drop. The last test is
useful as a quality assurance tool but not as a performance-based
tool. Thus, the next step in the verication process was compara-
tive testing relative to the ideal PCM model in Heating 7.3. This is
a multidimensional, nite difference heat conduction program that
can simulate materials with variable thermal properties and other
features [33]. This software has been used before in multidimen-
sional heat transfer problems with and without PCMs [14,38]. It
was also tested with the Neumann solution, which proved its
accuracy at small time steps (0.01 s). The comparative verication
consisted of a 24-h sine temperature wave test with amplitude of
30  C representing outside temperature variation during a day, and
a 1-dimensional wall consisting of 1 cm wood, 10 cm ber insu-
lation, and 1.5 cm drywall. This test represents a more realistic wall
conguration and boundary conditions than the analytical solution.
Model space discretization in Heating 7.3 and boundary conditions
are:
The two wall applications veried in this test are shown in
Table 2. This test evaluates only two applications, because no
signicant information was obtained from conducting three tests in
the analytical test. In addition, the concentrated thin PCM layer
application will be tested with experimental data in Section 3.3.
Fig. 3 through Fig. 5 shows the results of the comparative veri-
cation for the distributed PCM in insulation. All gures show
calculated values using Heating 7.3 (H73), EnergyPlus v7.1 PCM
model with default space discretization and 1-min time step (E),
EnergyPlus v7.1 with no PCM and 1-min time step (ENoPCM),
EnergyPlus v7.1 PCM model with smaller node spacing (Edx/3)
and 1-min time step, EnergyPlus v7.1 PCM model using a 2-min
time step (E2 min) and EnergyPlus v7.1 PCM model using a 4-min
time step (E4 min). Simulations with 2- and 4-min time steps
used the default space discretization values. All EnergyPlus simu-
lations were performed using the updated model (v7.1). Fig. 3
through Fig. 5 shows that discrepancies increase substantially
when time steps of 4 min (E4 min) or longer are used. Thus, all
simulations should use time steps shorter than 4 min.

 0.01 s time steps

 Node spacing 0.01 times the EnergyPlus default value (see
Table 2)
 Indoor air temperature: 25  C t  0

Table 2
PCMs tested in comparative verication.

Properties PCM-insulation PCM-Drywall

% by weight of 20 30
microencapsulated PCM
in wall layer
Equivalent latent heat (kJ/kg) 34 33
Melting temperature ( C) 26 25.05
Thermal conductivity (W/m K) 0.03365 0.1600
Density (kg/m3) 50 950
Specic heat (J/kg K) 960 840
Thickness (m) 0.100 0.015 Fig. 4. Inside surface temperature (Tin) calculated using comparative software (H73),
Default node spacing 0.015 0.005 EnergyPlus PCM model with default node spacing (E), smaller node space (Edx/3),
2- and 4-min time steps (E2 min, E4 min) and without PCM (ENoPCM).
190 P.C. Tabares-Velasco et al. / Building and Environment 54 (2012) 186e196

Table 4
Differences in net heat gain over 12 h between Heating 7.3 and other models.

Case Outside Inside Outside Inside

heat gain (kJ) heat gain (kJ) discrepancy discrepancy
with H7.3 (%) with H7.3 (%)
H73 216 43 e e
E 213 43 1.3 1.5
Edx/3 215 43 0.6 0.7
E2 min 214 46 0.9 7.8
E4 min 297 18 37.6 58.5
E noPCM 127 115 41.3 169.8

Fig. 5. Inside heat ux (Qin) calculated using comparative software (H73), EnergyPlus
PCM model with default node spacing (E), smaller node space (Edx/3), 2- and 4-min
time steps (E2 min, E4 min) and without PCM (ENoPCM).
The PCM model for the case with PCM distributed in the drywall
has performance similar to the one shown in Table 4. Table 5 shows
that all cases agree, having an almost zero heat ux coming into the
inside wall over the 24-h test.

Fig. 3 also shows the node temperature in the middle of the
insulation layer. All node locations for the middle node in the
insulation are in the middle of the insulation except the nodes for
the 2-min (E2 min) and 4-min (E4 min) time steps. These were
linearly interpolated, as the automatic node spacing in EnergyPlus
did not generate a node in the middle of the insulation. Figs. 3 and 4
show that the node temperature calculated by the EnergyPlus PCM
model oscillates around the temperature calculated by Heating 7.3.
This is the same behavior observed in the analytical test. The
oscillations attenuated and disappeared once smaller node spacing
was used, which in this case was about 1/3 smaller than the default
size. Fig. 3 shows that during approximately the rst 4 h the middle
wall node temperature with PCM remained at almost constant
temperature around the melting temperature (26  C). Once all the
PCM melted, the temperature started to increase as no more latent
storage was available, delaying the peak temperature by almost 2 h.
Overall, the benets of PCM in this particular example are reected
in Figs. 4 and 5, where the peak inside surface temperature and
heat ux are attenuated by 0.5  C and 50%, respectively, and the
peak time is delayed almost 4 h.
Figs. 4 and 5 and show the impact distributing PCM in the
insulation has on the inner surface temperature and inside heat
ux for this example. The PCM model correctly calculates peak load
reduction and shift. Despite the small oscillations around the values
calculated from Heating 7.3, the PCM model with default space
discretization obtains values close to Heating 7.3 (see Table 3).
Moreover, using smaller node spacing improves the agreement
with Heating 7.3 and using a 2-min time step still has close
agreement with Heating 7.3.
Finally, Table 4 summarizes the discrepancies in the rst 12-h
cycle using the same caption labels as in Fig. 3. The results agree
with Table 3: the default space discretization (E) and smaller node
space (Edx/3) models with PCM agree with Heating 7.3 very
closely. The 2-min time step (E2 min) has looser agreement but is
still close compared to the case without PCM which is not the case
when using the 4-min time step.
3.3. Experimental: DuPont hot box experiment
The last step in the verication and validation of the PCM model
was empirical validation. This step compared the PCM model with
experimental data from the literature [34,39] for the concentrated
PCM application. Fig. 6 is a schematic of the wall tested with the
PCM layer representing the DuPont Energain PCM product with
a melting temperature range centered around 21.7  C, a latent heat
of 70 kJ/kg, and a variable thermal conductivity. Properties for this
PCM are shown in the Appendix [39]. PCM properties in the
experimental study were provided by the manufacturer; thus,
there were no degree of freedom to calibrate EnergyPlus results
to match measured values. The tests were performed in a hot box
apparatus, where initially the cold box side was kept at 20  C and
the hot side was kept at 20  C. At t 0, a heater in the hot box
started heating the air temperature inside the hot box for 7 h
(heating stage). The nal inside wall temperature reached 24  C.
After that, the heater was turned off and the hot box slowly cooled
to the initial temperature, 20  C (cooling stage) [34,39].
Initial testing of the PCM model, in this case with variable
thermal conductivity revealed another bug that occurred when
these two features were simulated at the same time. The thermal
conductivityetemperature array had larger dimensions than the
allowable inputs for users. Thus, the variable thermal conductivity
was lling the empty values with temporary data from the PCM
enthalpy array. This bug was xed by sizing the variable thermal
conductivity array correctly. Fig. 7 through Fig. 10 compares the
results for the xed version to the original, which is marked Ev6.
Figs. 7e10 relate to the measurement points shown in Fig. 6. Results
from the empirical validation are shown in Fig. 7 through Fig. 10 for
experimental data (Experiment) with the uncertainty bars, Ener-
gyPlus v7.1 PCM model with variable thermal conductivity (E),
EnergyPlus v7.1 PCM model without PCM (EnoPCM), previous
version of EnergyPlus (v6), and EnergyPlus v7.1 PCM model with
variable thermal conductivity (E) and 4-min time step. The PCM
model was further evaluated using a 2-min time step and a smaller
node space (Edx/3). These are not shown in Fig. 7 through Fig. 10

Table 3
Root Mean Square Error (RMSE) compared to Heating 7.3 values for PCM distributed Table 5
in Insulation. RMSE for PCM distributed in drywall compared to Heating 7.3.

Tin ( C) Tmin ( C) Qin (W/m2) Qout (W/m2) Tin ( C) Tmin ( C) Qin (W/m2) Qout (W/m2)
ENoPCM 0.45290 2.6470 2.2940 4.370 ENoPCM 5.6180 3.066 4.610 1.801
E 0.01919 0.1384 0.1406 1.885 E 0.0254 0.138 0.019 1.800
Edx/3 0.01564 0.0741 0.0771 0.357 Edx/3 0.0255 0.114 0.017 0.347
E2 min 0.03715 0.2180 0.2698 0.717 E2 min 0.0257 0.015 0.193 1.777
E4 min 0.30770 2.3040 1.4590 2.844 E4 min 0.0340 0.753 0.197 4.394
 P.C. Tabares-Velasco et al. / Building and Environment 54 (2012) 186e196 191

Fig. 6. Tested wall conguration with dimensions (in cm) and location of temperature
(Temp) and heat ux measurements (HF).

Fig. 8. Point 5 temperature obtained from experimental data (Experiment), Ener-

to avoid overpopulating the graphs. No new results were observed
gyPlus v7.1 PCM model with variable thermal conductivity (E), EnergyPlus v7.1 PCM
besides those that were already found in Section 3.2 for the 2 and 4 model without PCM (EnoPCM), previous version of EnergyPlus (v6).
min time steps. The 2-min time steps and smaller node space
behaved similarly to the 1-min time step with default space dis-
cretization. In contrast, using a time step of 4 min introduces
show heat ux at measurement Points 1 and 5, respectively,
important errors in all gures that are consistent with the results
using the same legend as in Fig. 7. All models in Fig. 9 started with
shown in the analytical and numerical cases. All cases with
similar heat uxes and followed the same trend, but are outside the
different time steps used the same Fourier number, which makes
experimental uncertainty, a behavior that was also observed
the node space increase (Eq. (2)). This problem with the 4-min time
previously using WUFI and COMSOL [34], even when a smaller
step will be further analyzed in the future.
node space was used. However, measuring heat ux in this type of
Figs. 7 and 8 show the temperatures at Points 3 and 5. In both
interface is very challenging because radiation and convective
gures, Ev7.1 with PCMs agrees well with the experimental data
uxes are present in the air-wall surface interface. This might not
for the rst 8 h (heating stage). However, all models disagree with
be correctly simulated in EnergyPlus because the surface temper-
the experimental data after 8 h during the cooling stage of the
ature was input as a boundary condition.
problem. The disagreement shown is mainly due to the PCM
In contrast with Fig. 9, Fig. 10 shows that most models results
hysteresis, as the current PCM model in EnergyPlus does not
are within the uncertainty range of the measured heat ux in Point
simulate hysteresis. Thus, only the enthalpyetemperature infor-
5, located between the PCM and mineral wood during the heating
mation for the heating mode was input. This is also consistent with
process. As with the results shown in Figs. 7 and 8 the disagreement
the calculated values from Haavi et al., which simulated the same
increases once the PCM starts to cool down and hysteresis occurs. In
problem using PCM models in COMSOL and WUFI [34].
this case, the differences in the cooling range (starting around
Figs. 7 and 8 show that Ev6 completely disagrees with the data
Time 8 h) are on the order of 10e15%. In all cases, the previous
for roughly half the test because of the bug related to variable
PCM model completely missed the behavior of the PCM because of
thermal conductivity. The EnergyPlus simulation without PCM
the bug with variable thermal conductivity.
shows a peak reduction and time shift of roughly 0.5  C and 30 min,
Overall, this comparison to empirical data was important
respectively. These differences are small and caused mainly by the
because it used a real PCM product with variable thermal
slight excitation in the experiments (4  C).
conductivity to test the PCM model. Combined, the results from
In addition to temperature data used for empirical validation,
Sections 3.1e3.3 indicate that the v7.1 PCM model is performing
the experiments measured heat ux at two points. Figs. 9 and 10
well for PCM wall applications, except when the materials present

Fig. 7. Point 3 temperature obtained from experimental data (Experiment), EnergyPlus Fig. 9. Point 1 heat ux obtained from: experimental data (Experiment), EnergyPlus
v7.1 PCM model with variable thermal conductivity (E), EnergyPlus v7.1 PCM model v7.1 PCM model with variable thermal conductivity (E), EnergyPlus v7.1 PCM model
without PCM (EnoPCM), previous version of EnergyPlus with PCM (v6). without PCM (EnoPCM), previous version of EnergyPlus (v6).
192 P.C. Tabares-Velasco et al. / Building and Environment 54 (2012) 186e196

Table 7
PCMs used in modied ASHRAE Standard 140, Case 600.

Properties PCM- PCM-Drywall Concentrated

insulation thin PCM layer
% weight of 20% 30% 100%
microencapsulated
PCM in wall layer
Equivalent latent heat 34 33 130
(kJ/kg)
Melting temperature 28.5e30.5  C 22.5e24.5  C 28.5e30.5  C
range
Thermal conductivity 0.03365 0.23 0.16
(W/m K)
Density (kg/m3) 29 1000 850
Specic heat (J/kg K) 960 1400 2500
Location Walls and roof Walls Walls and roof
(5 cm insulation
0.5 cm PCM 5 cm
insulation)
Fig. 10. Point 5 heat ux obtained from experimental data (Experiment), EnergyPlus
v7.1 PCM model with variable thermal conductivity (E), EnergyPlus v7.1 PCM model
without PCM (EnoPCM), and previous version of EnergyPlus (v6).

building has interior loads equal to 200 W (60% radiative, 40%

strong hysteresis. The scope of this verication and the hysteresis
limitation should be kept in mind when conducting whole-building
simulations using EnergyPlus, such as those presented in the next
section.
4. Annual whole-building performance evaluation of veried
PCM model
After verication, the ideal step is to assess the performance of
the PCM model with a simple and a more complicated building and
mechanical system. The model was evaluated in Phoenix using
a lightweight building described in case 600 of ASHRAE Standard
140 [31] as well as a more realistic house using BEopt [40] with
EnergyPlus as the underlying simulation engine. The main purpose
of this initial evaluation was to test and demonstrate the simulation
capabilities of the PCM model in the context of whole buildings, not
to draw conclusions about the performance of specic PCM tech-
nologies. Future research will include detailed parametric analyses
where different PCM melting temperatures and enthalpies are
analyzed.
4.1. Annual performance model evaluation: modied ASHRAE
Standard 140, Case 600
The Case 600 model was selected because it is a simple, well-
referenced, and well-understood building that has been simu-
lated by several major simulation engines [31]. The basic test
building is a lightweight, rectangular, single-zone building, with
dimensions of 8 m wide by 6 m long by 2.7 m high, no interior
partitions, and a total window area of 12 m2 on the south wall. The
convective) and a highly insulated slab to essentially eliminate
thermal ground coupling. The inltration was set to 0.5 air changes
per hour. The building mechanical system is 100% efcient with
100% convective air, no duct losses, and no capacity limitation. The
thermostat is set with dead band so heating takes place for
temperatures below 20  C and cooling for temperatures above
27  C. Insulation properties have been slightly modied to incor-
porate PCMs (Table 6 and 7). In contrast with the PCMs modeled in
Sections 3.1e3.3, all PCMs in Table 7 are assumed to have a 2  C
melting temperature range.
The modied building was simulated using Phoenix TMY3
weather conditions with the three PCM applications described in
Table 7. The same building was also analyzed without PCMs using
CondFD, and then using conduction transfer functions under
different time steps and smaller node spacing for comparative and
time-dependency analysis. Fig. 11 shows the run time for all
simulated cases using EnergyPlus v6 (Ev6), v7.1 (Ev7.1), v7.1
with a smaller node space (Ev7.1 dx/3), v7.1 with a 2-min time
step (Ev7.1 2 min), and v7.1 with a 4-min time step (Ev7.1 4 min).
For each setting, the following simulations were performed: CTFs
without PCM (CTF), CondFD without PCM (CondFD), CondFD with
PCM distributed in insulation (Insulation), CondFD with PCM
between two insulation layers (Ins/PCM/Ins) and, CondFD with
PCM distributed in drywall (Drywall). There are several important
aspects to highlight here:

Table 6
Modied Case 600 materials properties.

Material Thermal Thickness Density Specic

conductivity (m) (kg/m3) heat (J/kg K)
(W/m K)
Wall Plasterboard 0.1600 0.0120 950 840
Insulation 0.03365 0.0660 29 960
Wood siding 0.1400 0.0090 530 900
Roof Plasterboard 0.1600 0.0100 950 840 Fig. 11. Run time for lightweight construction using EnergyPlus: v6 (Ev6), v7.1
Insulation 0.03365 0.1118 29 960 (Ev7.1), v7.1 with a smaller node space (Ev7.1 dx/3), v7.1 with a 2-min (Ev7.1
Roof deck 0.1400 0.0190 530 900 2 min), and a 4-min time step (Ev7.1 4 min). The y-axis represents: CTFs without PCM
Floor Timber ooring 0.1600 0.0100 950 840 (CTF), CondFD without PCM (CondFD), CondFD with PCM distributed in insulation
Insulation 0.0400 0.2500 12 840 (Insulation), CondFD with PCM between two insulation layers (Ins/PCM/Ins), and
CondFD with PCM distributed in drywall (Drywall).
 P.C. Tabares-Velasco et al. / Building and Environment 54 (2012) 186e196
Fig. 12. Annual cooling energy difference for lightweight construction using captions
as Fig. 11.
 Overall, CondFD takes about 2e3 times longer to run than CTF
when PCMs are not simulated,
 If PCMs are simulated, the run time doubles in comparison with
the case without PCMs and using CondFD,
 The new automatic under-relaxation described in Section 2
reduces run time by a factor up to 4e5, and
 Using a node space size three times smaller than the default
size increases run time by a factor of 2e3.
Fig. 12 shows the annual cooling energy difference, using CTF
without PCM as the baseline. Key highlights include:
 CondFD agrees with CTF within less than 0.2% when no PCM is
present,
 Decreasing the node space size by a factor of 3 does not cause
signicant differences,
 Increasing the time step to 2 min does not cause signicant
differences, and
 Increasing the time step to 4 min or longer causes signicant
differences, and should not be used.
Overall, using the default space discretization and time steps up
to 2e3 min yields very similar results. This information can inform
many parametric and optimization analyses where computational
Fig. 13. BEopt rendering of example house simulated with various PCM applications.
193
Table 8
Thermal storage of composite PCMs for BEopt new house.
Properties PCM distributed PCM distributed Concentrated
insulation drywall PCM layer
Thickness wall 0.089 m 0.0127 m 0.005 m
Walls latent storage 87 kJ/m2 420 kJ/m2 550 kJ/m2
Amount of PCM 0.5 kg/m2 3.8 kg/m2 4.2 kg/m2
speed is a key concern. Although not shown in the paper, the same
conclusions were found with an annual heating energy difference,
but with higher differences with respect to the case without PCMs.
Energy Difference was used instead of Energy Savings in
Fig. 12 because the authors main objective in using this simple
building was to test the CondFD and PCM models, not to assess the
energy benets of the PCM applications. Predicted benets may be
smaller when more realistic homes with higher internal gains and
inltration loads are modeled. Results should not be extrapolated
from Fig. 12 to state which PCM application is best, as more detailed
analysis is needed to address that problem.
4.2. Annual performance model evaluation: BEopt house
A second house was analyzed in EnergyPlus using the same
weather data. This is a two-story, slab-on-grade, 231 m2 (2500 ft2)
house with three bedrooms, two bathrooms, an unconditioned
attic, garage and a more realistic HVAC. This EnergyPlus model was
generated using BEopt (see Fig. 13) according to the Building
America 2010 Benchmark [41]. Thus, this is a realistic example of
a newly constructed house. The 2010 Benchmark represents
a house built using 2009 International Energy Conservation Code
(IECC) and federal appliance standards in effect as of January 1, 2010
home. More details about the house can be found in the literature
[42].
A few modications were made to the original EnergyPlus
model generated by BEopt to accommodate PCMs:
 The house was created in BEopt and the input EnergyPlus le
(idf) was exported to EnergyPlus where PCMs were introduced,
 The properties of the insulation used in the walls and attic were
modied to the properties for PCM-distributed insulation
shown in Table 7, and
194 P.C. Tabares-Velasco et al. / Building and Environment 54 (2012) 186e196
Fig. 14. Monthly cooling electric energy difference for Phoenix, Arizona: upper half
(Upper Attic) and lower half (Lower Attic) of the attic insulation with PCM, PCM
concentrated at the middle of the attic insulation (RCRAttic), wall cavity with PCM
distributed in insulation (Walls), PCM concentrated at the middle of the wall cavity
insulation (RCRWall), PCM distributed drywall (Drywall) and combination of RCRWall
and PCM-Drywall (RCRWall-DW).
 The walls and attic were assumed to be studless and trussless.
This assumption was made to simplify the problem, as a rst
approximation as BEopt cannot model PCMs at this time.
These assumptions resulted in a well-insulated building enve-
lope, as the equivalent R-values of the studless wall and trussless
attic are RSI  2.7 (RIP  15.3) and RSI  6.7 (RIP  40), respectively.
Thus, this case represents a home having walls with an R-value
close to a wall assembly consisting of RIP  19 batts, 5  15 cm
(2  6 in) studs, 60 cm (24 in) apart from center having a RSI  2.9
(RIP  16.3). In contrast a typical wall consisting of RIP  13 batts,
5  10 cm (2  4 in) studs, 40 cm (16 in) apart from center have
a RSI  2 (RIP  11.4).
Table 8 shows storage capacity for the PCM assemblies in the
walls using PCMs composite properties shown in Table 7. As with
the ASHRAE modied Case 600, the PCM distributed in insulation
has the lowest storage capacity due to the low density of the
insulation. Higher latent storage values could be achieved by
increasing the percentage of PCM, the thickness of insulation layer
or the density of insulation.
Fig. 14 shows the predicted monthly cooling electric energy
reduction for the following PCM applications: upper half of the attic
insulation enhanced with PCM (Upper Attic), lower half of the attic
insulation enhanced with PCM (Lower Attic), PCM concentrated at
the middle of the attic insulation (RCRAttic), wall cavity with PCM
distributed in insulation, PCM concentrated at the middle of the
wall cavity insulation (RCRWall), PCM distributed in the ceiling and
Fig. 15. Monthly peak cooling electric difference for Phoenix, Arizona using same
captions as in Fig. 14.
wall drywall (Drywall) and combination of RCRWall with PCM
distributed in Drywall (RCRWall-DW). The last PCM application was
chosen to evaluate the ability of the PCM model to simulate walls
with different PCMs in a single wall. The predicted annual electric
cooling consumption for the house without PCM was 5500 kWh,
less than half the 2009 average for homes in Arizona, which is
12,900 [43]. In all cases, the EnergyPlus run time was about 20 min
using a 2-min time step, about twice the run time of CTF. This
suggests that the CondFD and PCM models are working properly, as
previously shown with the modied ASHRAE Standard 140
building. Moreover, no instability problems were detected.
The predicted annual peak cooling electric load was 3.83 kW for
the house without PCM. Fig. 15 shows the predicted peak load
reduction for the PCM applications analyzed, which was about 4.3%
for the best PCM analyzed in the cooling peak month of July.
Figs. 14 and 15 show very modest predicted energy benets.
However, predicted savings could be greater for a well-designed
PCM strategy that implements multiple technologies in a single
home and optimizes PCM properties and locations within the walls
or attic.
5. Conclusions
The PCM model in EnergyPlus was veried and validated using
an approach similar to ASHRAE Standard 140, which consists of
analytical verication, comparative testing, and empirical valida-
tion [31]. During this process, two bugs were identied and xed in
the PCM model. Version 8 of EnergyPlus will include the validated
PCM model and xes, which speed up the run time and allow for
simulation of PCMs with variable thermal conductivity.
This study also identied a few key limitations of, and guidelines
for using, the EnergyPlus PCM model:
 Time steps equal to or shorter than three minutes should be
used,
 Accuracy issues can arise when modeling PCMs with strong
hysteresis,
 Default CondFD can be used with acceptable monthly and
annual results. However, if accurate hourly performance and
analysis is required, smaller node space (1/3 of the default
value in EnergyPlus) should be used at the expense of longer
run times.
The modied Case 600 with PCM proposed in this study could
also be added into ASHRAE Standard 140 now that PCMs are
becoming more common and other building energy simulation
programs have PCM models that might not have been validated.
Finally, this research opens multiple dimensions for future
research on PCMs, including the ability to condently investigate
different PCM material properties, congurations, and locations
within a house. Future research will include more detailed para-
metric analyses where PCM melting temperatures and enthalpies
will be analyzed for specic U.S. climate zones. Finally, integration
of the PCM model into BEopt would allow for the optimization of
PCMs using BEopts proven optimization schemes for new home
construction and existing home retrots.
Acknowledgements
This work was supported by the U.S. Department of Energy
under Contract No. DE-AC36-08-GO28308 with the National
Renewable Energy Laboratory. The authors would like to thank Ben
Polly and Dane Christensen of NREL for their critical reviews and
Brent Grifth of NREL for his technical support.
 P.C. Tabares-Velasco et al. / Building and Environment 54 (2012) 186e196 195

Appendix [10] Kedl RJ. Conventional wallboard with latent heat storage for passive solar
applications; 1990. Medium: X; Size: Pages: (5 p.).
[11] Ahmad M, Bontemps A, Salle H, Quenard D. Experimental investigation and
PCM properties used in Experimental Validation computer simulation of thermal behaviour of wallboards containing a phase
change material. Energ Build 2006;38(4):357e66.
Thickness: 0.0053 m [12] Kuznik F, Virgone J, Roux J-J. Energetic efciency of room wall containing PCM
wallboard: a full-scale experimental investigation. Energ Build 2008;40(2):
Thermal conductivity: Solid (T < 21.6  C): 0.18 W/m K 148e56.
Liquid (T > 21.6  C): 0.14 W/m K [13] Kuznik F, Virgone J. Experimental investigation of wallboard containing phase
Density: 855 kg/m3 change material: data for validation of numerical modeling. Energ Build 2009;
41(5):561e70.
Specic heat: 2500 J/kg-K [14] Petrie TW, Childs KW, Childs PW, Christian JE, Shramo DJ. Thermal behavior of
mixtures of perlite and phase change material in simulated climate. Quebec
Enthalpy vs temperature data for DuPont Energain PCM used in City, USA: ASTM; 1997.
[15] Halford CK, Boehm RF. Modeling of phase change material peak load shifting.
Experimental Validation. Data obtained from differential scanning calo- Energ Build 2007;39(3):298e305.
rimeter (DSC) measurements with a heating rate of 0.05  C/min [39]. [16] Virgone, J, Nol, J, Reisdorf, R. Numerical study of the inuence of the thick-
ness and melting point on the effectiveness of phase change materials:
application to the renovation of a low inertia school, In: Eleventh interna-
tional IBPSA conference; 2009.
T ( C) H (J/kg  C) [17] Pedersen, CO. Advanced zone simulation in EnergyPlus: incorporation of
variable properties and phase change material (PCM) capability, In: Building
9.0 0.001 simulation 2007; 2007. Beijing, China.
7.0 5200 [18] Heim D, Clarke JA. Numerical modelling and thermal simulation of PCM-
5.0 10,800 gypsum composites with ESP-r. Energ Build 2004;36(8):795e805.
4.0 13,750 [19] Heim D. Isothermal storage of solar energy in building construction. Renew
3.0 16,850 Energy 2010;35(4):788e96.
2.0 20,350 [20] Schossig P, Henning HM, Gschwander S, Haussmann T. Micro-encapsulated
1.0 24,750 phase-change materials integrated into construction materials. Solar Energy
0.2 30,030 Mat Solar Cells 2005;89(2e3):297e306.
0.0 31,610 [21] Stovall TK, Tomlinson JJ. What are the potential benets of including latent
1.0 37,160 storage in common wallboard? J Solar Energy Eng, Trans ASME 1995;117(4):
2.0 40,510 318e25.
[22] Ibez M, et al. An approach to the simulation of PCMs in building applica-
2.5 42,160
tions using TRNSYS. Appl Therm Eng 2005;25(11e12):1796e807.
4.0 47,335
[23] Koschenz M, Lehmann B. Development of a thermally activated ceiling panel
5.0 50,885
with PCM for application in lightweight and retrotted buildings. Energ Build
7.5 60,135 2004;36(6):567e78.
10.0 70,010 [24] Tomlinson JJ, Heberle DD. Analysis of wallboard containing a phase change
12.5 81,010 material. Reno, NV, USA: IEEE; 1990.
15.0 93,760 [25] Kuznik F, Virgone J, Johannes K. Development and validation of a new TRNSYS
17.5 109,385 type for the simulation of external building walls containing PCM. Energ Build
20.0 129,635 2010;42(7):1004e9.
22.5 157,385 [26] Tabares-Velasco PC, Grifth B. Diagnostic test cases for verifying surface heat
23.5 170,985 transfer algorithms and boundary conditions in building energy simulation
24.0 177,535 programs. J Building Performance Simul; 2011:1e18. doi:10.1080/
25.0 186,185 19401493.2011.595501.
[27] Kosny J, Yarbrough D, Wilkes K, Leuthold D, Syad A. PCM-enhanced cellulose
26.0 191,185
insulation e Thermal mass in lightweight natural bers. In: 2006 ECOSTOCK
27.0 195,535
conference. Richard Stockton College of New Jersey: IEA, DOE; 2006.
28.0 199,485
[28] Lin K, Zhang Y, Xu X, Di H, Yang R, Qin P. Modeling and simulation of under-
29.0 203,135 oor electric heating system with shape-stabilized PCM plates. Building
30.0 206,335 Environ 2004;39(12):1427e34.
31.5 210,535 [29] Zhang YP, Lin KP, Yang R, Di HF, Jiang Y. Preparation, thermal performance
45.0 244,960 and application of shape-stabilized PCM in energy efcient buildings. Energ
80.0 332,460 Build 2006;38(10):1262e9.
[30] Huang MJ, Eames PC, Norton B. Thermal regulation of building-integrated
photovoltaics using phase change materials. Int J Heat Mass Transfer 2004;
References 47(12e13):2715e33.
[31] ASHRAE. ANSI/ASHRAE Standard 140-2004. In: Standard method of test for
[1] Khudhair AM, Farid MM. A review on energy conservation in building appli- the evaluation of building energy analysis computer programs. Atlanta, GA:
cations with thermal storage by latent heat using phase change materials. ASHRAE; 2004. p. 151.
Energy Convers Manag 2004;45(2):263e75. [32] Carslaw HS, Jaeger JC. Conduction of heat in solids. 2nd ed. Oxford University
[2] Tyagi VV, Buddhi D. PCM thermal storage in buildings: a state of art. Renew Press; 1959.
Sustainable Energy Rev 2007;11(6):1146e66. [33] Childs KW. Heating 7.2 users manual. Oak Ridge National Laboratory; 2005.
[3] Pasupathy A, Velraj R, Seeniraj RV. Phase change material-based building p. 242.
architecture for thermal management in residential and commercial estab- [34] Haavi T, Gustavsen A, Cao S, Uvslkk S and Jelle BP. Numerical simulations of
lishments. Renew Sustainable Energy Rev 2008;12(1):39e64. a well-insulated wall assembly with integrated phase change material panels
[4] Sharma A, Tyagi VV, Chen CR, Buddhi, D. Review on thermal energy storage with eComparison with hot box experiments, In: The international conference on
phase change materials and applications. Renew Sustainable Energy Rev 2009; sustainable systems and the environment; 2011. Sharjah, Sharjah, United
13(2):318e45. Arab Emirates.
[5] Wang X, Zhang YP, Xiao W, Zeng RL, Zhang QL, Di HF. Review on thermal [35] Hensen JLM, Nakhi A. Fourier and Biot numbers and the accuracy of
performance of phase change energy storage building envelope. Chinese Sci conduction modelling. In: BEP 94 conference facing the future.
Bull 2009;54(6):920e8. York: Building Environmental Performance Analysis Club (BEPAC);
[6] Zhu N, Ma Z, Wang S. Dynamic characteristics and energy performance of 1994.
buildings using phase change materials: a review. Energy Convers Manag [36] Waters JR, Wright AJ. Criteria for the distribution of nodes in multilayer
2009;50(12):3169e81. walls in nite-difference thermal modelling. Building Environ 1985;20(3):
[7] Drake JB. A study of the optima; transition temperature of PCM wallboard for 151e62.
solar energy storage. Oak Ridge National Laboratory; 1987. [37] EnergyPlus. EnergyPlus engineering reference: the reference to EnergyPlus
[8] Solomon AD. Design criteria in PCM wall thermal storage. Energy 1979;4(4): calculations. Ernest Orlando Lawrence Berkeley National Laboratory; 2011.
701e9. p. 1130.
[9] Athienitis AK, Liu C, Hawes D, Banu D, Feldman D. Investigation of the thermal [38] Kosny J, Christian JE. Thermal evaluation of several congurations of insu-
performance of a passive solar test-room with wall latent heat storage. lation and structural materials for some metal stud walls. Energ Build 1995;
Building Environ 1997;32(5):405e10. 22(2):157e63.
196 P.C. Tabares-Velasco et al. / Building and Environment 54 (2012) 186e196

[39] Cao S, et al. The effect of wall-integrated phase change material panels on the
indoor air and wall temperature e Hot box experiments, In: Zero emission
buildings e Proceedings of renewable energy conference 2010; 2010:
Trondheim, Norway. p. 15e26.
[40] Christensen C, Horowitz S. Building energy optimization. Available from:
http://beopt.nrel.gov/; 2011.
[41] Hendron R, Engebrecht C. Building America house simulation protocols.
Golden, CO: National Renewable Energy Laboratory; 2010. p. 80.
[42] Casey S, Booten C. Energy savings measure packages: new homes. Golden, CO:
National Renewable Energy Laboratory; 2011. p. 205.
[43] EIA. 2009 residential energy consumption survey [cited 2011, 10/21/2011];
Available from: http://205.254.135.24/consumption/residential/index.cfm; 2011.