Applied Energy 177 (2016) 740750

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

The calibration and validation of a model for predicting the performance

of gas-fired tankless water heaters in domestic hot water applications
Geoffrey Johnson , Ian Beausoleil-Morrison
Sustainable Building Energy Systems, Faculty of Engineering and Design, Carleton University, Ottawa, Canada

h i g h l i g h t s

A model of a gas-fired tankless water heater was derived.

Experiments were conducted to determine the calibration parameters of this model.

Model predictions were validated against data emanating from an earlier study.

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

Article history: Gas-fired tankless water heaters are gaining popularity as a means of producing domestic hot water for
Received 17 February 2016 residential applications. As they become more widespread, these devices are increasingly becoming
Received in revised form 4 May 2016 important to researchers who are seeking models of these devices that can be used to make reasonable
Accepted 21 May 2016
predictions of their energy consumption. This is especially true for the case of condensing tankless water
Available online 6 June 2016
heaters as they have been shown to be the most efficient method of producing domestic hot water.
Current models of these devices do exist, however, the uncertainties of their predictions are unclear
Keywords:
and require data collected from onerous field-trials for calibration.
Gas-fired tankless water heater
Building performance simulation
This article makes a contribution by introducing a new model of a condensing tankless water heater
Domestic hot water whose uncertainties are well described that can be calibrated by a less-onerous experimental program
Housing in a laboratory. The model was calibrated with data spanning a range of conditions: inlet temperatures
between 10 and 23 C, water flow rates between 0.08 and 0.27 L s1 and outlet temperatures between
36 and 48 C.
The model predictions were validated against data emanating from a previous field-trial. Above domes-
tic hot water consumption values of 15 MJ day1, the coefficient of determination (r2 value) was 0.98. The
average error (difference between the average model predictions and the measured values) was 3.0% (in
relative terms) while the root mean square error was 4.2%. The maximum error for a single point was
8.7% at a domestic hot water consumption of approximately 30.2 MJ day1.
2016 Elsevier Ltd. All rights reserved.

1. Introduction Advanced models of non-condensing TWHs exist [2,3]. How-

Gas-fired water heaters are widely used for domestic hot water
(DHW). They are used in approximately one half of all homes in the
United States [1]. There are several common configurations: stor-
age water heaters, boilers, and tankless water heaters (TWHs).
From 2005 to 2009, the number of homes in the United States
where a TWH was used as the main source of DHW increased from
approximately 1.3 to 2.6 million [1] suggesting that they are
gaining in popularity.
Corresponding author.
E-mail addresses: geoffreyjohnson@cmail.carleton.ca
Ian_Beausoleil-Morrison@carleton.ca (I. Beausoleil-Morrison).
ever, as was shown by Bohac et al. [4] and Hoeschele and Weitzel
[5], a condensing TWH is the most efficient conventional method
of generating heat from natural gas in a DHW application.
Notwithstanding their potential for energy savings, a simple model
of a condensing TWH for a residential DHW application capable of
predicting its energy consumption where uncertainties in
measurements are propagated through to model predictions is
unavailable. This is mainly due to the fact that a TWHs perfor-
mance is more dependent on its usage profile than either boilers
or storage water heaters are. This is why a steady-state model
(e.g. [6]) is inadequate for this application.
The transient period immediately following the activation of a
(G. Johnson),
TWH can affect its overall performance when infrequent and

http://dx.doi.org/10.1016/j.apenergy.2016.05.130
0306-2619/ 2016 Elsevier Ltd. All rights reserved.
 G. Johnson, I. Beausoleil-Morrison / Applied Energy 177 (2016) 740750
Nomenclature
t time (s)
Q_ in heat input to the heater (kW)
Q_ out heat output of the heater (kW)
Ein energy input to the heater (kJ)
Eout energy output of the heater (kJ)
T temperature of heater and outlet water (C)
T in temperature of inlet water (C)
DT T  T in (C)
T env ambient air temperature of environment (C)
V_ fuel volumetric flow-rate of fuel (natural gas) to the heater
(m3 s1)
V_ w volumetric flow-rate of water [m3 s1 ]
m_w mass flow-rate of water (kg s1)
Cw specific heat capacity of water (kJ kg1 C1)
C TWH heat capacity of heater (kJ C1)
intermittent use is common, such as in a DHW application.
Schoenbauer et al. [7] found, of the ten houses they monitored,
the average DHW draw was 5.7 L, the average draw duration was
66 s and, on average, there were 26 draws a day. This is a signifi-
cant reason why TWHs perform differently under different interna-
tional standard test conditions [8] and also why the performance of
a TWH is, in practice, different from its rated efficiency. It was
shown by Bohac et al. [4], that the North American Energy Factor
overstates the energy efficiency of a TWH in a DHW application.
The Energy Factor is determined from a standard DHW draw
pattern (6 draws of 40.6 L over a 24 h period) where overall
consumption (243 L day1) was nearly 70% greater than was found
in practice [7].
Previously [4,5] it has been observed that there is a strong linear
relationship between the energy input (Ein kJ) and the energy
output (Eout kJ) of a TWH in practice. Both Bohac et al. [4] and
Hoeschele and Weitzel [5] performed regression analyses to deter-
mine the coefficients ci in Eq. (1) from measured field-trial data.
Ein c1  Eout c2 1
For modeling, Butcher and Schoenbauer [9] suggested it might be
possible to derive the coefficients of this linear relationship for a
heater in a laboratory environment from several representative
DHW draws. The advantage of this approach was its simplicity. It
was noted by Bohac et al. [4] that although c1 clearly represents
the reciprocal of the steady-state efficiency fss , c2 is related to
the heat capacity of a TWH (C TWH kJ C1). Healy et al. [10] found
it difficult to robustly characterize c2 . Scatter in Ein and Eout mea-
surements at higher consumption levels in their experimental pro-
gram resulted in large uncertainty margins of the c2 values they
derived from regression. Consequently, Healy et al. [10] questioned
the validity of this approach. This also suggests that a detailed
accounting of experimental uncertainties is necessary in any newly
developed model forms.
1.1. Contributions
From the literature reviewed, it is clear that there are the fol-
lowing knowledge gaps in this field where there is an opportunity
to make a contribution. First, although Bohac et al. [4] and
Hoeschele and Weitzel [5] gathered an extensive amount of data
from their respective field-trials and presented models in the form
of Eq. (1), a thorough analysis of the uncertainties of their models
predictions was unavailable. A model for a condensing TWH whose
predictions uncertainties were well described would be an
advancement.
741
qw density of water (kg m3)
UA heat-loss coefficient between heater and environment
(kW C1)
s time constant (s)
f thermal efficiency of heater ()
HHV higher heating value of fuel (kJ m3)
c1 slope of linear relationship between Ein and Eout ()
c2 intercept of linear relationship between Ein and Eout (kJ)
ut unit step function ()
dt unit impulse ()
B bias error
S precision index
ts standard statistical Students t-value
U 95% uncertainty at 95% confidence level
Second, Glanville et al. [11], Healy et al. [10], Bohac et al. [4] and
Burch et al. [12] all described a need in this field to determine a
simple model form that can be calibrated with an experimental
program consisting of a small number of tests. Currently, the only
reliable method of calibrating Eq. (1) is with an extensive field-
trial. A more easily calibrated model would allow the performance
of a TWH to be simulated for any arbitrary DHW profile and facil-
itate comparisons between different methods of producing DHW.
Burch et al. [12] proposed that a model based on a lumped heat
capacity (Eq. (6)) would be suitable for calibration by such a simple
experimental program. However, whether the large uncertainty
margins they presented on their calibration parameters from their
experimental program would allow for accurate model predictions
is unclear. Glanville et al. [11] also derived some calibration
parameters that could be used with the model developed by Burch
et al. [12], however, the uncertainty of these parameters was not
described. Notwithstanding these limitations, a lumped heat
capacity approach to modeling could still be viable for a condens-
ing TWH if the uncertainty margins of the calibration parameters
could be shown to allow for reasonably accurate predictions. Such
a model should also be validated against a large set of actual DHW
use data to demonstrate its limitations.
This article makes a contribution by demonstrating that a sim-
pled lumped heat capacity condensing TWH model can be cali-
brated by an experimental program that is far less onerous than
an extensive field-trial. Throughout, the uncertainties of the model
calibration parameters are characterized. These uncertainties are
also propagated through to model predictions.
First, based on the analytical solutions to the governing differ-
ential equation of the TWH modeled as a lumped heat capacity, a
new model will be introduced based on the energy inputoutput
response of the TWH to only a step-input of heat. Second, Eq. (1)
will be shown to represent the energy inputoutput response of
the governing differential equation to the combination of an
initial-energy-impulse combined with a step-input of heat. Third,
another new model will be introduced as the average of the first
two models. Exponential decay will be introduced to account for
time between DHW draws. The experimental procedures required
to calibrate these models are demonstrated. The validity of the
simplifying assumptions inherent in the lumped-heat capacity
approach along with the assumed form of the heat-input function
are validated against data emanating from an earlier study [4].
The developed model is also intended to support researchers
studying more innovative systems for producing DHW. As an
example, applications where solar thermal collectors are used to
provide DHW for residential applications is an active area of
742 G. Johnson, I. Beausoleil-Morrison / Applied Energy 177 (2016) 740750

research (e.g. Refs. [1322]). In order to assess the energy savings where ut is the unit-step function given by the following
of this type of system, a well-justified model of a conventional definition.
method for producing DHW is required for comparison. The model

0 t<0
being proposed in this research is suitable for this. In particular, the ut 4
case study described in Section 4 demonstrates how the energy
1 tP0
savings of a solar DHW system change when the reference scenario where Q_ inss is the steady-state rate of heat input given by the follow-
is changed from a traditional boiler to the TWH modeled in this ing equation.
research. The next section derives the analytical solutions that
are the foundation of this article. Q_ inss HHV  V_ fuelss 5

where HHV is the higher heating value of the fuel (kJ m ) and V_ fuelss 3
2. TWH first-order models
is the steady-state volumetric flow-rate of the fuel (m3 s1) if both
values are at the same standard temperature and pressure condi-
The inspiration for the model forms investigated in this article
tions. This assumption neglects the initial heat-input overshoot that
was from experimental observations (Section 3.2). A typical heat-
was observed experimentally in Section 3.2. It is a significant depar-
input (Q_ in t kW) and temperature-difference (DT C) response ture from the model introduced by Burch et al. [12] but is necessary
from a TWH during a DHW draw is shown in Fig. 1. The heat input to allow for an analytical solution to the governing differential
is given by the following equation. equation shown in Eq. (6).
Q_ in t V_ fuel t  HHV 2 According to Burch et al. [12], the differential equation govern-
ing the TWH system can be reasonably well approximated by the
where V_ fuel t is the fuel flow rate that is also a function of time (t) following equation.
and HHV is the higher heating value of the fuel (discussed in
dT
Section 3). _ w  C w  T  T in C TWH 
m UA  T  T env Q_ in t  fc 6
dt
Also shown in Fig. 1 are two lines that were superimposed to
illustrate the inspiration for the model forms investigated in the where m _ w is the mass flow-rate of water (kg s1), fc is the
following sections. First, a horizontal line to represent a step- combustion efficiency, C w is the specific heat capacity of water
input whose magnitude is equal to the steady-state input of heat. (kJ kg1 C1), T is the temperature of the TWH which is assumed
This is the assumed form of the heat-input that is the inspiration equal to the outlet water temperature ( C), T in is the inlet water
for the model developed in Section 2.1. Second, a vertical line with temperature, T env is the air temperature in the ambient environ-
an arrow, at time equal to zero, to represent an initial impulse of ment, UA is the heat-loss coefficient of the TWH (kJ C1) and
energy. The combination of these two lines is the inspiration for C TWH is the heat capacity of the TWH (kJ C1).
the model form investigated in Section 2.2. Heat loss to the ambient environment is orders of magnitude
By inspection, it can be seen that the measured heat-input pro- less than the heat input during the initial firing stage. Therefore,
file resembles a step input, although there is significant overshoot it can be neglected during the step input (Eqs. (3) and (4)) of the
initially while the temperature difference profile resembles a first- initial firing stage, representing the governing equation in terms
order-step response that is delayed by several seconds. of the difference between outlet and inlet temperatures DT (Eq.
 DT

(7)). Note that a substitution for the derivative dT
dt
ddt has been
2.1. Step-response model made to simplify the solution procedure.

The major assumption of the first-order-step-response model is 7

that the heat input Q_ in t can be represented by the step function
defined in Eq. (3). Eq. (7) also assumes that T in and m _ w are constant with time and
fc = fss. The solution to the differential equation shown in Eq. (7)
Q_ in t ut  Q_ inss 3 is the well-known first-order-step response. With an initial condi-
tion DT 0 , the solution is given by the following equation.
t
30
Initial-impulse of energy
50 DTt DT ss DT 0  DT ss es 8
where DT ss is the steady-state temperature difference between out-
25 let and inlet. The parameter s is termed the time constant and has
40
the same units as t (s). It is given by the following equation.
Temperature difference ( C)
o

20 C
s _ TWH 9
Heat input (kW)

30 mw  C w

15
Step-input of heat The following integral can be used to find Eout t.
Z t
20 Eout t _ w  C w  DTt  dt
m 10
10 0
Measured heat input
Substituting the step response for DTt from Eq. (8) into this inte-
Measured temperature difference
10 gral yields the following equation.
5
Z t  
_ w  C w DT ss DT 0  DT ss es  dt
t
Eout t m 11
0
0 0
0 100 200 300 400
Evaluating the integral in Eq. (11) yields the following equation.
Time from start of experiment (s)
h   it
_ w  C w  DT ss  t  DT 0  DT ss s  es
t
Fig. 1. A typical temperature-difference response of a TWH and heat input Eout t m 12
0
observed during the calibration experiments of Section 3.2.
 G. Johnson, I. Beausoleil-Morrison / Applied Energy 177 (2016) 740750 743

Z 0
Substituting s from Eq. (9) into Eq. (12) and evaluating the expres-
sion inside of the square brackets from Eq. (12) yields the following
dtdt 1 20
0
equation.
Note that dt can be regarded in this case as the derivative of ut
Eout t m_ w  C w  DT ss  t DT 0  DT ss with respect to time so the units on the right hand side of Eq. (18)
 _
 are consistent. Interested readers may find Ref. [23] to be a good
mw C w t
 C TWH 1  e CTWH 13 resource for describing how the Dirac distribution may be used
for modeling physical systems.
The energy input to the TWH can be found easily by integrating Eq. After an analogous development to that performed in
(3) according to the following equation. Section 2.1, the model energy inputoutput relationship is given
Z t by the following equation.
Ein t ut  Q_ inss dt Q_ inss  t 14
Eout t C TWH
0
Ein t  DT ss  DT 0 21
fss fss
The steady-state efficiency fss is given according to Eq. (15).
A more detailed derivation is provided as Supplementary informa-
_ w  C w  DT ss
m
fss 15 tion. Although this result is somewhat obvious, it shows that a sim-
Q_ in ss ple linear energy inputoutput relationship of the form in Eq. (1)
The definition of fss (from Eq. (15)) along with an expression for t in can be thought of as the TWHs response to a step-input of heat
terms of Ein t (from Eq. (14)) can be substituted into Eq. (13) to combined with an initial-impulse of energy. This insight assists in
express the energy input in terms of the energy output according understanding the selection for the third and final model form
to Eq. (16). described in the following section.
 
Eout t DT ss  DT 0  C TWH E tfss
 in 2.3. Average model
Ein t 1  e CTWH DT ss 16
fss fss
As has been shown at the beginning of Section 2, the heat input
An interesting observation can be made regarding Eq. (16). By tak-
can have some overshoot followed by a steady period. Essentially,
ing the limit as t approaches infinity, the first-order-step-response
the two models previously described represent extreme cases. The
model in Eq. (16) approaches the form of Eq. (1) as is shown below.
step-response model in Section 2.1 is a situation with no heat-
input overshoot and only a steady period. The step-function and
17 initial-impulse-response model in Section 2.2 represents a situa-
tion with a large, instantaneous overshoot followed by a steady
Eq. (17) is of the form Ein t c1  Eout t c2 . By comparison with period. By recognizing that in reality, the actual heat input poten-
Eq. (1), c1 is the reciprocal of the steady-state efficiency 1=fss tially lies somewhere between the two extremes, the arithmetic
and c2 is equal to the amount of energy input required to raise average of the two models presents itself as an interesting option.
the temperature of the TWH from its initial condition to its It is given by the following equation, formed by averaging Eqs. (16)
steady-state value DT ss  DT 0  C TWH =fss . This result is consistent and (21).
with the earlier works of Bohac et al. [4] and Hoeschele and Weitzel  
[5] who observed a simple linear energy inputoutput relationship.
Eout t DT ss  DT 0  C TWH 1  Ein tfss
Ein t 1   e CTWH DT ss 22
This result also satisfies the desire expressed by Healy et al. [10] for fss fss 2
a model that can account for the temporal distribution of a draw
All three of these defined model forms only account for the periods of
pattern. If the TWH is initially at an elevated temperature due to
time when the TWH is firing. The periods between firings are referred
recent use, the energy input required to raise it to its steady-state
to as environmental decay periods [11]. As all three models are based
value will be less than if the TWH was initially at the inlet temper-
on a lumped heat capacity assumption, environmental decay periods
ature value.
can be modeled using the same method for all three models.

2.2. Step-response and initial-impulse model

2.4. Environmental decay model

The only difference between the model developed here and the
According to Glanville et al. [11], the differential equation
one in Section 2.1 is the assumption of the functional form of the
governing the behavior of the TWH during environmental decay
heat input. In the preceding section, it was assumed that the func-
periods can be approximated by the following equation.
tional form was a step function whose magnitude was equal to the
steady-state heat input. In this section, the functional form will be dT
C TWH  UA  T  T env 0 23
assumed to be the combination of the earlier step function plus an dt
initial impulse of energy equal to the energy required to instantly The solution of this equation is the well-known exponential-decay
bring the TWH from its initial condition DT 0 to its steady-state equation shown below.
value DT ss . This functional form is shown in the following
   UA t
equation. T T 0  T env  e CTWH T env 24
Q_ in t Q_ inss  ut Ess  E0  dt 18 0
where T is the temperature of the TWH at the end of the previous
firing period. It will be shown, in Sections 3.5 and 3.8, that this
where Ess  E0 is the energy input required to raise the temperature
equation can be used reasonably well to predict the initial condition
of the TWH to its steady-state value from its initial condition and is
DT 0 required by the models defined by Eqs. (16), (21) and (22).
given according to the following equation.
Note that DT 0 can be determined from Eq. (24) by simply subtract-
Ess  E0 C TWH  DT ss  DT 0 =fss 19 ing T in from both sides as is shown below.
The unit impulse dt, given by the Dirac distribution, used in    UA t
DT 0 T 0  T env  e CTWH T env  T in 25
Eq. (18) has the following property.
744 G. Johnson, I. Beausoleil-Morrison / Applied Energy 177 (2016) 740750

The remainder of this article will be devoted to demonstrating the Table 1

validity of this and other aspects of these models through experi- Sensor types and measured parameters.

ments. The following section describes the experimental apparatus Primary Sensed quantity Sensor type Bias
that was used to gather these data necessary for this purpose. measurement
T; T in ; T fuel Temperature Thermocouple 0:5 C
DT Temperature Thermopile 0:1 C
3. Experimental methods difference
P fuel Absolute Transducer 0:25% reading
The experiments conducted in this article were performed at pressure
V_ w Volumetric flow Oval-gear flow 1% reading and
Carleton University in the facility originally developed by Boucher rate meter 0:0019 L s1
[24] with some modifications. The TWH that was the subject of the V_ fuel Volumetric flow Diaphragm 1% reading and
experiments in this article had a burner capable of providing rate meter 0:0057 L s1
560 kW. For the experiments described in this article, Fig. 2
illustrates the hydronic configuration of the apparatus that was
used along with the location of the various primary measurements The pressure of the supply gas was measured with a transducer
used to derive the necessary parameters relevant to the models so the flow-rate of gas could be corrected to a consistent tempera-
described in the preceding sections. For each experiment, a fixed- ture and pressure (15 C and 101.325 kPA) for use with the HHV
speed pump was used to circulate water from two 1300 L water provided by the utility (38.15 MJ m3). This value has a bias of
tanks through the TWH. A throttling valve was manually adjusted 0:3 MJ m3 to account for its monthly variation over the period
to control the flow-rate of water in each experiment. that the experiments were conducted. It also accounts for the fact
The large thermal mass of the water tanks would ensure the that the utility determined this value by gas chromatograph, a
TWH inlet temperature would rise slowly (approximately measurement technique known to be accurate to within 0:25%
<0.05 C per minute was achieved in practice). An external chilled [25].
glycol stream (8 C) could be activated on-demand and used to cool The bias on the volumetric water and gas flow-rate measure-
the water stored in the tanks before, after or during experiments ments are presented as a percent reading combined with an abso-
via the heat exchanger at the TWH outlet. lute error in (L s1) to account for the resolution of the pulse output
The TWH was vented into the buildings exhaust system which of both meters for the sample period that was used (5 s). The
necessitated the use of an additional external exhaust fan not pre- impact of the instrumentation bias errors listed in Table 1 on the
sent in a typical residential application. However, the use of this parameters derived from them is explained in the following
external fan did not impact measurements during firings as this section.
fan was manually controlled in the lab to impose typical airflow
conditions as measured by a pressure sensor internal to the
TWH. Although this sensor would ensure normal operation during 3.1. Experimental uncertainty
firing, there were some implications related to its use between fir-
ings that are discussed in Sections 3.4 and 3.5. Experimental uncertainties were characterized with conven-
Table 1 describes the instrumentation of the apparatus. All tem- tional methods [26]. A review of these methods is provided as Sup-
peratures were measured with copperconstantan thermocouples plementary information. A description of how the uncertainty
that were calibrated beforehand in the laboratory. Temperature calculations were performed in practice is described in the follow-
difference was measured across the inlet and outlet of the TWH ing sections.
using a five-junction copperconstantan thermocouple (ther-
mopile) that was also calibrated beforehand in the laboratory. 3.1.1. Steady-state parameters
These sensors had exposed junctions that were submerged in the For any steady-state experiment, each measurement sampled
flow. every 5 s was viewed as a repeated observation. Here the repeated
observation of any parameter is denoted by /. For example, a
steady-state experiment of 120 s in duration was considered to
consist of 24 repeated observations of the same parameter
n 24.
External Exhaust Fan
There were several parameters derived from steady-state peri-
ods of operation that were relevant. First, according to Glanville
.
V T P et al. [11], UA can be derived from a steady-state experiment where
fuel fuel fuel the TWH is disabled and water at an elevated temperature is circu-
lated through it. Under these conditions, an instantaneous mea-
TWH surement of this parameter UAi is given by the following
Water Water equation.
tank 1 tank 2
qw  C w  V_ w;i  DT i
UAi 26
. T env ;i  T i
1300 L 1300 L
Pump V T
w in
T T
where qw is the density of water. Note that it was assumed
Valve T env T fuel . The calibration value of UA determined for modeling,
described in more detail in Section 3.4, was the sample mean of this
Chilled parameter /  measured over a steady period.
Heat
Glycol Exchanger A summary of other important steady-state derived parameters
and their uncertainties is described in Table 2. The instrumenta-
Fig. 2. Schematic representation of the experimental apparatus used to conduct the tion, in particular the flow meters, had greater accuracy at greater
experiments in this article. flow rates as was shown in Table 1. Also, experiments were more
 G. Johnson, I. Beausoleil-Morrison / Applied Energy 177 (2016) 740750 745

Table 2 The bias on the energy input is expressed as a product of the

uncertainty margin of the average value of Q_ inss and the duration
Range of derived parameters observed during steady-state periods of experiments
described in Sections 3.2, 3.3, 3.5 and 3.6 along with the bias, standard deviation and
uncertainty margins at a 95% confidence level. of the experiment m  Dt. Eq. (31) guarantees that BEin is the same

Derived Measurement Range of B Range of S

percentage of Ein as U _ is to Q_ in .
95%;Q inss ss
Range of U 95%
parameter range (% value) (% value) (% value)

Q_ inss 7.244.6 kW 1.43.3 2.213.8 1.65.4 3.2. Calibration of the TWH heat capacity
Q_ outss 5.942.3 kW 1.22.8 0.05.6 1.32.9
fss 84.695.9% 1.94.4 1.813.8 2.06.2 The heat capacity C TWH is assumed to be a characteristic con-
(HHV) stant of each individual TWH system. To derive this value, a series
of calibration experiments, with initial conditions equal to zero,
were conducted. Each experiment was conducted in the following
unsteady at lower heat-input rates. This is where the greater stan-
manner. Before the experiment, in order to ensure the initial con-
dard deviations (S) shown in Table 2 were measured.
dition was zero DT 0 0, water was circulated through the TWH
For experiments, instantaneous measurements of Q_ in;i ; Q_ out;i until DT  0 was observed. This would ensure the time between
and fi were derived according to the following equations. experiments was not a factor. All boundary conditions (T in ; T out
Q_ in;i V_ fuel;i  HHV 27 and V_ w ) were held steady for the duration of the experiment.
Between experiments boundary conditions were changed by
Q_ out;i V_ w;i  qw  C w  DT i 28 adjusting the set point of the TWH T out , adjusting the throttling
valve V_ w or heating the tank for an extended period of time
Q_ out;i T in . Data from one such experiment are shown in Figs. 3a and 3b.
fi 29 The measured energy input Ein t and output Eout t are each
Q_ in;i
plotted in Fig. 3a against time. The error bars shown are the bias
The steady-state values for these parameters (Q_ inss ; Q_ outss and fss errors derived according to Eq. (31). For the same experiment,
reported in Table 2) were defined as their respective samples mean Eout is plotted against Ein in Fig. 3b. After the TWH reaches a
 measured from steady periods of operation.
/ steady-state period of operation, the linearity of the relationship
between Eout t and Ein t is apparent. For the calibration experi-
3.1.2. Integrated parameters ments, 100 s of operation was observed to be sufficient for the
Aside from the steady-state derived parameters in the preced- steady-state period to be reached. The goal of each calibration
ing discussion, for the experiments conducted in this article, there experiment was to derive C TWH according to the following
were several important parameters that were integrated over the equation.
duration of the steady-state and startup transient periods of c2  fss
C TWH 32
TWH operation. One such parameter was the energy input to the DT ss
TWH Ein which was derived from the following equation.
where c2 is the coefficient from Eq. (1). It was derived by a linear
X
m
least-squares regression fit to the data shown in Fig. 3b during
Ein Q_ in;i  Dt 30
i1
the steady-state period. To estimate the bias on c2 for the uncer-
tainty analysis, the bias on Eout and the bias on Ein (both as a % of
where Q_ in;i is the heat input measured during the ith sample inter- value) along with the maximum residual of the least-squares
val of duration Dt. Summing the quantity on the right-hand side of regression fit were combined according to standard methods [26].
Eq. (30) from the first to last sample interval yields the total energy In total, 14 tests similar to the one shown in Figs. 3a and 3b
input over that period. It is important to note that for Ein , and other were conducted to calibrate C TWH . A summary of the boundary con-
integrated parameters, the experimental uncertainty had to be han- ditions and results of these experiments is shown in Table 3. Note
dled differently than the steady-state parameters in the preceding that the V_ w values investigated are at the upper range of what
discussion. would be expected in a DHW application [7].
For example, consider the experiment shown in Fig. 1. This
experiment is composed of a transient startup period followed by
7000
a steady-state period. If, for example, Ein was calculated after
120 s m 24, this Ein could only be considered as a single obser-
6000 TWH Energy Input
vation of /. This experiment would have to be repeated several
times under the same boundary conditions then, for each repeated
5000
experiment, each Ein calculated after 120 s could be considered as a
repeated observation of /.
Energy (kJ)

4000
For the experiments in this article, Eq. (31) was used to estimate
the bias on the energy input as the experiments were predomi-
3000
nantly steady-state. A similar approach was taken to estimate
the bias on the energy output as well. This approach is an approx-
2000
imation for the first several measurements during the initial tran-
sient stage of a firing. However, the most important use of these 1000
uncertainty margins was to estimate the uncertainty of the TWHs TWH Energy Output
heat capacity (Section 3.2) and it will later be shown (Section 3.8) 0
that the uncertainty of the TWHs heat capacity is only a small 0 100 200
component of the overall uncertainty of model predictions so this Time from start of experiment (s)
approach was found to be adequate.
Fig. 3a. Energy input and output of the TWH measured during one of several
BEin U 95%;Q_ in  m  Dt 31 calibration experiments used to derive C TWH each plotted against time from the start
ss of experiment as a separate series.
746 G. Johnson, I. Beausoleil-Morrison / Applied Energy 177 (2016) 740750

deviation (S) of the entire sample. It is noteworthy that the esti-

mate of U 95% explains the range of values observed for C TWH . This
6000
validates the approach that was used to estimate experimental
uncertainty in this article. In addition to being more amenable
5000
for use with generally accepted methods of deriving experimental
Energy Input (kJ)

uncertainty parameters, using this methodology to derive the heat

4000
capacity described here produced a narrower range of observed
values than was achieved by Glanville et al. [11] and Burch et al.
3000
[12].
2000
3.3. Validation of the TWH heat capacity
1000
To validate the value for C TWH derived during calibration, 4 addi-
0 tional validation tests were conducted in the same manner as the
0 1000 2000 3000 4000 5000
calibration experiments in the preceding discussion with one
Energy Output (kJ)
important exception. Although the range of temperatures spanned
Fig. 3b. Energy input plotted against energy output of the TWH measured during by the validation experiments were similar to the calibration
one of several calibration experiments used to derive C TWH . experiments, the water flow rates of the validation experiments
were significantly less at 0.08 L s1 for each of these experiments.
The reasoning behind this was that these lower flow rates could
Table 3
potentially be challenging for the models to make accurate predic-
Summary of 14 experiments used to calibrate the heat capacity of the TWH.
tions. In general, lower heat-input rates, relative to the specific
Parameter Range investigated heat of the TWH, will result in a slower response and a longer tran-
T in 21.330.3 C sient period of operation. These flow rates are also more typical of
T out 43.048.1 C what would be expected in a DHW application [7].
DT ss 13.326.3 C
Fig. 4 compares model predictions from the 4 validation exper-
V_ w 0.280.3 L s1
iments to observed experimental values. Each of the 3 graphs
Q_ inss 17.135.0 kW
shown in Fig. 4 compares a different models predictions to exper-
Q_ out 15.132.9 kW
fss
ss

87.994.1% (HHV)
iments. Each series in each graph represents the percent difference
between a models predictions and observed experimental values
C TWH Range Range of B S U 95%
for a single validation experiment. A positive percent difference
1
23.121 kJ C 21.725.3 0.882.17 1.03 2.25 indicates an over-prediction of energy consumption. Model predic-
tions were made using the value of C TWH at the bottom of Table 3.
For clarity, in Fig. 4 the range of model predictions shown is
The average value for the heat capacity observed from the 14
restricted to be between 10% of the measured value, therefore,
experiments C TWH is shown at the bottom of Table 3. The range predictions at the beginning of the DHW draw outside of this range
of specific heat values along with the corresponding range of biases are not shown.
observed are shown to the right of C TWH . The maximum bias It can be seen that the average model (in the center of Fig. 4) is
observed was used to calculate U 95% along with the standard more accurate for shorter DHW draw durations. Within 60 s, this

10

0
Step-Response and Initial-Impulse Model (eq: 21)
-5

-10

5
% error

0
Average Model (eq: 22)
-5

-10

0
Step-Response Model (eq: 16)
-5

-10
0 100 200 300 400
Duration of DHW draw (s)

Fig. 4. Percent error between energy input to the TWH observed during 4 validation experiments and predictions made by model Eqs. (16), (21) and (22).
 G. Johnson, I. Beausoleil-Morrison / Applied Energy 177 (2016) 740750
model can make predictions to within 5% of the measured energy
input. The other models can take over 100 s to make predictions as
accurate. It can also be seen that the step-response and initial-
impulse model (at the top of Fig. 4), converges to its final value
from large positive values, while the step-response model (at the
bottom of Fig. 4) converges to its final value from lower values.
This indicates that the logic behind the selection of the average
model (the actual response lies somewhere between the other
two models) is somewhat justified by these experiments. All 3
models converge to within 1% of the actual measured value after
nearly 400 s.
3.4. Calibration of the TWH heat-loss coefficient
The calibration value for UA along with its associated uncer-
tainty margins were determined according to the methods
described in Section 3.1.1. This type of experiment was difficult
to conduct accurately due to the small DT that was measured
across the TWH while it was disabled. While the external fan
was operating, UA was determined to be 0.0076 kW C1 33%
based upon the measured data. In practice, an external fan is most
often unnecessary so the value of UA was also determined when
this fan was off. With this fan off, UA was determined to be
0.0023 kW C1 46% based upon the measured data. It will be
demonstrated in Section 3.8 that these large uncertainty margins
still allow for reasonably accurate model predictions.
It should be noted that UA is not only a characteristic of the each
individual TWH, but also of its installation. Local ambient condi-
tions and heat-transfer characteristics could result in differences
between the heat-loss coefficient derived here and in another
location.
3.5. Validation of the heat-loss coefficient
The experimental program used to validate the calibration
value determined for UA in the preceding section is described as
follows. For this program, initial conditions DT 0 were estimated
by comparing the c2 coefficients from a series of 5 min DHW draw
experiments where first there was a preceding experiment where
DT 0 was equal to zero. This preceding experiment was then fol-
lowed by several other 5 min DHW experiments where DT 0 was
not equal to zero. Between any two experiments there was an idle
period where the temperature of the TWH was allowed to decay
while the external fan was left on. For this experimental program
the following equation was used to derive DT 0 .
 
c2  c02  fss
DT 0
C TWH
where c2 was derived from the preceding experiment where DT 0
was zero (according to the methods discussed in Section 3.2) and
c02 was the y-intercept coefficient derived from each experiment
that followed. The advantage of performing the experimental pro-
gram in this way, by using Eq. (33), was that initial conditions could
be derived rather than measured directly. Measuring the initial con-
dition directly is difficult to perform accurately as it is unclear how
to define precisely when the initial condition occurs during this
highly transient period. The results from 9 separate validation
experiments conducted in this manner are shown in Fig. 5.
In Fig. 5, the initial temperature of the TWH T 0 is shown rather
than DT 0 . The derived value of T 0 of a single experiment is repre-
sented by the individual markers. The error bars represent the bias
on T 0 calculated using analogous methods to those described in
Section 3.2. Shown as a solid line is the environmental decay model
given by Eq. (24) with T env 21 C, UA 0:0076 kW C1,
C TWH 23:121 kJ C1 and T 0 48 C. Also shown, as broken lines,
747
50
45
Initial temperature of TWH ( C)
o
40
o -1
UA = 0.0051 kW C
o -1
35 UA = 0.0076 kW C
30 o -1
UA = 0.0100 kW C
25
20
0 500 1000 1500 2000 2500 3000
Time since last DHW draw (s)
Fig. 5. Initial temperature of the TWH derived from experimental data compared to
the exponential decay model for validation when the external exhaust fan was left
on between firings.
is the environmental decay equation with UA 0:0100 kW C1
and UA 0:0057 kW C1 to represent the large uncertainty mar-
gins derived in Section 3.4. As can be seen, the value of UA from cal-
ibration while the external fan was left on estimated T 0 reasonably
well.
As the external exhaust fan was manually controlled it was not
possible to guarantee that it would be off for an entire period
between firings so a similar experimental program for when the
fan was off was not attempted. However, these results still demon-
strate that exponential decay is potentially suitable for modeling
the period between firings. The UA determined for when the fan
was off (0.0023 kW C1) is more typical of a residential applica-
tion where an external exhaust fan is not needed. For this reason
this lower UA value is validated against actual DHW use data in
Section 3.8.
3.6. Calibration of the steady-state efficiency
In this section, the fss parameter required to calibrate model
Eqs. (16), (21) and (22) has been derived through an additional
separate series of 31 experiments according to the methods pre-
sented in Section 3.1.1. The experiments spanned a wide range of
boundary conditions, (10 < T in < 23 C, 0:08 < V_ w < 0:27 L s1,
36 < T out < 48 C). Although in practice there are certainly DHW
33
draws that will have lower flow rates, the lower limit is still repre-
sentative of an average flow rate (0.076 L s1 Schoenbauer et al.
[7]). The impact of this limitation of the calibration procedure is
addressed by the validation in Section 3.8. The results from these
experiments are shown in Fig. 6.
As fss was relatively insensitive to the boundary conditions over
the ranges investigated, only the average fss from the 31 experi-
ments of 93.9% is recommended for use. The maximum residual
was 3.6% of value.
3.7. TWH electric power consumption
According to Bohac et al. [4] a significant source of electric con-
sumption, that could be measured and is relevant to other installa-
tions, is the standby electric consumption of the TWH. This was
determined to be 5.5  0.3 W based upon the measured data.
Another source of electric power consumption is known as freeze
protection [4]. This aspect has not been investigated in this paper.
748 G. Johnson, I. Beausoleil-Morrison / Applied Energy 177 (2016) 740750

1 70

60
Steady-State Efficiency (%HHV)

0.95

Modelled Ein (MJ Day )

-1
50

40

0.9
30

20
0.85
10

0
0.8 0 10 20 30 40 50 60 70
0 10 20 30 40 50 -1
Measured Ein (MJ day )
Heat Output (kW)

Fig. 6. The steady-state efficiency of the TWH versus its heat output during a series Fig. 7. Comparison of the average model predictions against the measured data
of 31 calibration experiments. provided by Bohac et al. [4].

During firings, typically the electric consumption of the fan Table 4

Summary of the sensitivity analysis of the model predictions to each model
would initially increase quickly to approximately 80 W for a period
parameters uncertainty margin.
of several seconds before stabilizing at a lower level for the rest of
the firing at a consumption level that was less than 0.5% of Q_ out . ss
UA fss C TWH U 95%tot
1
This aspect was not modeled in detail, the electric consumption xi 0.0023 kW C 0.939 23.121 kJ C1
during firings was only measured to ensure it was small. U 95% 0.0011 kW C1 0.042 2.25 kJ C1
@Ein
 DEinxi 2.9% 4.7% 0.4% 5.5%
@xi

3.8. Whole model validation

To demonstrate the validity of the whole average model along
with the exponential decay equation, measured residential DHW
profiles from TWHs emanating from an earlier study [4] were
made available to the authors of this study and the following anal-
ysis was performed. In this analysis, model predictions for Ein were
made for 43 separate DHW profiles and compared to the values
measured by Bohac et al. [4]. Each profile was 1 week of data
where sampling was performed every second. The profiles were
from 4 separate sets of occupants, who used 3 different condensing
TWHs. Weekly average DHW usage spanned a range of 3.9
58.5 MJ day1 for the profiles investigated. For this range of DHW
consumption, a range of 5.969.3 MJ day1 of TWH energy con-
sumption was measured [4]. The comparison of modeled and mea-
sured energy consumption is illustrated in Fig. 7. The measured
over a reasonably wide range. This indicates that a comparison
between the performance of different TWHs in different environ-
ments is reasonable even if the model parameters are significantly
different.
To estimate the uncertainty margin for each of the 43 DHW pro-
files, model predictions were made for the base case, where the
model parameters values xi were those determined by calibra-
tion in this paper. Model predictions were then repeated for each
DHW profile for cases where each model parameter was perturbed
by its uncertainty margin. Note that the U 95% value shown for fss is
a combination (according to Moffat [26]) of the maximum residual
reported in Section 3.6 along with the U 95% values determined
experimentally shown on Fig. 6 for Q_ out greater than 15 kW.
ss

values from Bohac et al. [4] are plotted against the abscissa while In Table 4, @E
@x
in
 DE xi was calculated as the mean percent difference
i in

the average model predictions are plotted against the ordinate.
In this analysis, the temperature of the TWH Tt was modeled
according to the following equation.
Ein tfss
C
DT 0  DT ss e TWH DT ss
Tt DT ss T in
2
This equation is consistent with the average model and is necessary
to determine the temperature of the TWH at the end of a firing per-
iod from which the environmental decay period begins (Eq. (25)).
The model parameters that were used are shown in Table 4
along with a summary of how the error bars shown in Fig. 7 were
derived from a perturbation analysis. The parameters that were
used for modeling were those determined by calibration in Sec-
tions 3.2, 3.4 and 3.6 in a laboratory environment and are different
than those that would be determined from the TWHs studied by
Bohac et al. [4]. However, as can be seen from the results of the
perturbation analysis at the bottom of Table 4, the model predic-
tions are relatively insensitive to changes in the parameters values
between the base case and a perturbed case, for each parameter xi ,
over the 43 DHW profiles analyzed. The mean result of this analy-
sis is summarized at the bottom of Table 4. Note that it was unrea-
sonable to derive uncertainty margins from the data measured by
Bohac et al. [4] due to the highly transient nature inherent in field-
34 trial data.
Notwithstanding this limitation, the result of this comparison
indicates that the average models predictions are commensurate
with what has been measured in practice from other condensing
TWHs. The solid line in the center of Fig. 7 represents the line of
perfect agreement. Above and below this line are two other lines
that represent a 5% deviation from perfect agreement. For the
entire range of the average models predictions, the coefficient of
determination (r2 value) was 0.990. The average error (difference
between the average model predictions and the measured values)
was 4.1% (in relative terms) while the root-mean-square error was
4.5%. The maximum error for a single point was 16.8% at a DHW
consumption of approximately 3.9 MJ day1. Above DHW con-
sumption values of 15 MJ day1, the coefficient of determination
 G. Johnson, I. Beausoleil-Morrison / Applied Energy 177 (2016) 740750
Table 5
Case study: primary energy savings of a solar DHW system with glazed collectors [16] compared to a traditional boiler and the condensing TWH modeled in this research.
Solar DHW [16] (MJ day1) Gas energy savings (MJ day1)
Traditional boiler [16]
Rome 31.8 45.4
Madrid 29.0 41.4
Munich 23.2 33.2
(r2 value) was 0.981. The average error (difference between the
average model predictions and the measured values) was 3.0%
(in relative terms) while the root-mean-square error was 4.2%.
The maximum error for a single point was 8.7% at a DHW con-
sumption of approximately 30.2 MJ day1.
4. Case study: solar DHW
One potential use for this TWH model is as a reference case
against which the performance of a more innovative system for
providing DHW may be compared to. To demonstrate this, the fol-
lowing case study was conducted that compares the performance
of the solar DHW system simulated by Comodi et al. [16] with
glazed collectors to the TWH modeled in this research.
In their research, Comodi et al. [16] estimated the payback per-
iod for a solar DHW system in different climatic regions. As part of
their analyses, they estimated the gas energy savings of the solar
DHW system by comparison with a traditional boiler (70% effi-
cient) as a reference scenario. Table 5 shows the gas energy savings
calculated with their reference scenario along with the energy sav-
ings relative to the TWH modeled in this research.
The solar DHW consumption from several locations is shown on
the left side of Table 5. Next to this are the gas energy savings of
the solar DHW system relative to the traditional boiler. The final
three columns show a range of gas energy savings based on the
TWH modeled in this research. Using boundary conditions
described by Comodi et al. [16] (T in 11:5 C, Tout = 45 C) and an
assumed ambient temperature (Tamb = 18 C), simulations were
conducted with the energy use profiles provided by Bohac et al.
[4] to estimate the range of efficiencies that are appropriate for
the solar DHW consumption shown in Table 5. In this region of
usage the efficiency varied between 82% and 89.5% due to the tem-
poral distribution of occupant usage patterns. The average between
these two boundaries (85.7%) was selected to estimate the average
model predictions in Table 5.
One advantage of using the TWH model developed in this
research is that uncertainty margins may be estimated. The lower
uncertainty margin was estimated as the combination of the
uncertainty caused by the temporal distribution of occupant usage
patterns (4.4%), and the uncertainty of the models calibration
parameters (5.5% from Table 4). The upper uncertainty margin
also considers these two sources but additionally considers a fur-
ther uncertainty of (8.7%) to account for the maximum under-
prediction of energy consumption described in Section 3.8.
As can be seen from Table 5, the ranges of gas energy savings
relative to the TWH modeled in this research are significantly less
than the gas energy savings relative to the traditional boiler.
5. Conclusions
In this article, a model to describe the energy inputoutput rela-
tionship based on the analytical solutions to the TWH modeled
with a lumped heat capacity was developed. The model inspired
a new method of measuring the heat capacity from experiments.
This method of measuring the heat capacity was shown to produce
repeatable results that were amenable to conventional methods of
749
Tankless water heater (average model)
Upper uncertainty margin Average model prediction Lower uncertainty margin
41.7 37.1 34.6
38.1 33.8 31.6
30.5 27.1 25.3
investigating experimental uncertainty. An analysis revealed that
the calibration parameters required by this model could be deter-
mined from an experimental program in a laboratory that is less
onerous than an extensive field trial and would not propagate an
unreasonable amount of uncertainty to model predictions.
All aspects of the developed model were validated through a
comparison between model predictions and data emanating from
a previous study. From this, it was found that the model predic-
tions of TWH energy consumption were commensurate with data
measured in a field trial for most cases. The models major limita-
tion is that it can under-predict energy consumption by as much as
8.7% in a typical region of DHW usage. This model is most suitable
for use in applications where it is important to quantify the uncer-
tainty of model predictions such as a well-justified reference sce-
nario for comparison with a more innovative method of
producing DHW. The results presented in this article could hope-
fully inspire other researchers to investigate the validity of using
simple analytical solutions to the lumped-heat-capacity differen-
tial equations as models to represent the energy performance of
other TWHs.
Acknowledgments
The authors of this study are grateful to Ben Schoenbauer and
the Center for Energy and Environment for sharing the data col-
lected in their TWH field-trial study.
Appendix A. Supplementary material
Supplementary data associated with this article can be found, in
the online version, at http://dx.doi.org/10.1016/j.apenergy.2016.
05.130.
References
[1] EIA. Residential energy consumption survey: Water heating in U.S. homes, by
housing unit type - Tables HC8.1 (2009) and HC2.8 (2005). Tech. Rep., U.S.
Energy Information Administration; http://www.eia.gov/consumption/
residential/ Accessed May 27 2016.
[2] Lutz J, Grant P, Kloss M. Simulation models for improved water heating
systems. Tech. rep., Lawrence Berkeley National Laboratory; 2013.
[3] Grant P. Evaluation of the performance of gas tankless water heaters. Masters
thesis, United States: University of Colorado Boulder; 2010.
[4] Bohac D, Schoenbauer B, Hewett M, Lobenstein M, Butcher T. Actual savings
and performance of natural gas tankless water heaters. Tech. rep., Center for
Energy and Environment; 2010.
[5] Hoeschele M, Weitzel E. Monitored performance of advanced gas water
heaters in California homes. ASHRAE Trans 2013;119:21425.
[6] Bourke G, Bansal P. New test method for gas boosters with domestic solar
water heaters. Sol Energy 2012;86:7886.
[7] Schoenbauer B, Bohac D, Hewett M. Measured residential hot water end use.
ASHRAE Trans 2012;118:87289.
[8] Bourke G, Bansal P, Raine R. Performance of gas tankless (instantaneous) water
heaters under various international standards. Appl Energy 2014;131:46878.
[9] Butcher T, Schoenbauer B. Application of a linear input/output model to
tankless water heaters. ASHRAE Trans 2011;117:6839.
[10] Healy W, Ullah T, Roller J. Inputoutput approach to predicting the energy
efficiency of residential water heaters testing of gas tankless and electric
storage water heaters. ASHRAE Trans 2011;117:52132.
[11] Glanville P, Kosar D, Stair J. Short-term performance of gas-fired tankless
water heater: laboratory characterization. ASHRAE Trans 2013;119:24869.
750 G. Johnson, I. Beausoleil-Morrison / Applied Energy 177 (2016) 740750
[12] Burch J, Thornton J, Hoeschele M, Springer D, Rudd A. Preliminary modeling,
testing and analysis of a gas tankless water heater. In: Proc. SOLAR, San Diego,
United States, 2008.
[13] Asaee R, Ugursal I, Beausoleil-Morrison I, Ben-Abdallah N. Preliminary study
for solar combisystem potential in Canadian houses. Appl Energy
2014;130:5108.
[14] Chow T, Pei G, Fong K, Lin Z, Chan A, He M. Modelling and application of direct-
expansion solar-assisted heat pump for water heating in subtropical Hong
Kong. Appl Energy 2010;87:6439.
[15] Colclough S, McGrath T. Net energy analysis of a solar combi system with
seasonal thermal energy store. Appl Energy 2015;147:6116.
[16] Comodi G, Bevilacqua M, Caresana F, Paciarotti C, Pelagalli L, Venella P. Life
cycle assessment and energy-CO2-economic payback analyses of renewable
domestic hot water systems with unglazed and glazed solar thermal panels.
Appl Energy 2016;164:94455.
[17] Haillot D, Franquet E, Gibout S, Bdcarrats J. Optimization of solar DHW
system including PCM media. Appl Energy 2013;109:4705.
[18] Uday Kumar N, Mohan G, Martin A. Performance analysis of solar cogeneration
system with different integration strategies for potable water and domestic
hot water production. Appl Energy 2016;170:46675.
[19] Panaras G, Mathioulakis E, Belessiotis V. A method for the dynamic testing and
evaluation of the performance of combined solar thermal heat pump hot water
systems. Appl Energy 2014;114:12434.
[20] Wei H, Liu J, Yang B. Cost-benefit comparison between domestic solar water
heater (DSHW) and building integrated photovoltaic (BIPV) systems for
households in urban China. Appl Energy 2014;126:4755.
[21] Rodrguez-Hidalgo M, Rodrguez-Aumente P, Lecuona A, Legrand M, Ventas R.
Domestic hot water consumption vs. solar thermal energy storage: the
optimum size of storage tank. Appl Energy 2012;97:897906.
[22] Sterling S, Collins M. Feasibility analysis of an indirect heat pump assisted
solar domestic hot water system. Appl Energy 2012;93:117.
[23] Kamen E, Heck B. Fundamentals of signals and systems using the Web and
Matlab. Prentice Hall; 2000 [ISBN No.: 0-13-017293-6].
[24] Boucher E. The design and commissioning of a micro-cogeneration testing
facility. Masters thesis, Canada: Carleton University Ottawa; 2013.
[25] LaNasa P. Overall measurement accuracy determination and influence. Tech.
rep., American School of Gas Measurement Technology; 2003.
[26] Moffat R. Describing the uncertainties in experimental results. Exp Therm
Fluid Sci 1988;1:317.