Documentos de Académico
Documentos de Profesional
Documentos de Cultura
Applied Energy
journal homepage: www.elsevier.com/locate/apenergy
h i g h l i g h t s
a r t i c l e i n f o a b s t r a c t
Article history: This paper addresses the technical and economic issues associated with waste heat recovery in data
Received 4 June 2014 centers through the use of absorption cooling machines. The theoretical possibility of utilizing the heat
Received in revised form 20 October 2014 dissipated by a server, or a number of servers, to power an absorption system, which in turn produces
Accepted 24 October 2014
cooling for other servers in the data center, is investigated. For this purpose, a steady-state thermody-
Available online xxxx
namic model is developed to perform energy balance and exergy analyses for a novel conguration of
an on-chip two-phase cooling system and an absorption refrigeration system. This combination is created
Keywords:
by replacing the condenser in the on-chip cooling circuit with the generator of an absorption refrigeration
Data center
Heat recovery and reuse
cycle. The performance of the developed model in simulating both LiBrwater and waterammonia
Absorption refrigeration absorption cooling systems is examined through verication of the model results against the reference
Thermodynamics data available in the literature. The verication indicates the superiority of LiBrwater absorption system
for data center/server operating conditions. Therefore, a LiBrwater absorption refrigeration system is
modeled in the novel combined heat recovery system. For these systems it is shown that the traditional
denition for the coefcient of performance (COP) is not appropriate to evaluate the performance and, in
its place, introduce a new gure of merit. Through a sensitivity analysis, the effects of server waste heat
quality, server coolant type, solution peak concentration, solution heat exchanger effectiveness, evapora-
tor temperature, and operating pressures on the performance of the novel system are investigated.
Finally, using the thermodynamic model and cost information provided by the absorption refrigeration
industry, an economic analysis is carried out to calculate the payback period when this technology is used
for data center waste heat recovery.
2014 Elsevier Ltd. All rights reserved.
1. Introduction and literature review the worlds total electricity usage is related to data center industry.
In the United States, however, this proportion is a bit higher, with
As a consequence of the ongoing growth in the data center data centers responsible for 1.72.2% of total electricity consump-
industry, coupled with the increasing power density of server com- tion [3]. Considering the annual electricity usage in United States,
ponents, the share of electricity consumption by data centers is which is more than 3800 billion kW h [4], it is seen that more than
expected to grow over the coming years [1]. Industry predictions 75 billion kW h of electricity is consumed annually by data centers.
suggest that the annual increase in data center power demand In a typical data center, around 4050% of electric power supply is
can be as high as 1520% [2]. Statistics indicate that 1.11.5% of consumed by IT equipment; 3040% is consumed by cooling
system and the rest is consumed by other facilities such as power
conversion and distribution units, and lighting [57]. Almost all the
Corresponding author at: 800 Lancaster Avenue, Department of Mechanical energy consumed by the IT equipment is converted to waste
Engineering, Villanova, PA 19085, USA. Tel.: +1 610 519 4996.
heat. Therefore, in the United States alone, 32.2 billion kW h of
E-mail address: amy.eischer@villanova.edu (A.S. Fleischer).
http://dx.doi.org/10.1016/j.apenergy.2014.10.067
0306-2619/ 2014 Elsevier Ltd. All rights reserved.
Please cite this article in press as: Ebrahimi K et al. Thermo-economic analysis of steady state waste heat recovery in data centers using absorption refrig-
eration. Appl Energy (2014), http://dx.doi.org/10.1016/j.apenergy.2014.10.067
2 K. Ebrahimi et al. / Applied Energy xxx (2014) xxxxxx
electricity is turned to waste heat annually through data centers 35.2 kW was achievable. The corresponding Coefcient of Perfor-
operation and almost the same amount of energy is required to mance (COP) at the design point was 0.7. The minimum quality
operate data centers cooling system and remove that waste heat and quantity of the supply heat required for the system operation
from server components. From the above statistics, it is clear that were 14.1 kW and 70 C. The cooling capacity and COP correspond-
capturing and reusing the servers dissipated energy can result in ing to the minimum operation conditions were 8.8 kW and 0.62,
huge energy savings through not only repurposing the dissipated respectively. The maximum COP of the unit was 0.86 with a
heat but also eliminating or reducing the need for cooling and heat generator temperature of 80 C and a resultant cooling capacity
removal systems. The energy savings resulting from data center of 25 kW.
waste heat recovery can reduce not only the depletion rate of fossil Other research studies have focused on micro-scale systems in
fuels reservoirs but also the emission rate of carbon dioxide and which miniature AR systems are designed to directly cool a single
other contaminant gases. microprocessor using the energy dissipated by the same micropro-
The low quality (temperature) of the dissipated energy is the cessor. Kim et al. [10] designed and analyzed a micro-scale LiBr
most challenging issue in data center waste heat recovery. It Water AR unit. In their design, a hydrophobic membrane was used
restricts the number of heat recovery technology options that to place the generator and condenser micro-channel arrays in inti-
can be used for this purpose. In a previous study [1], we reviewed mate contact with each other. The membrane acted as a common
all the currently available and developmental low-grade waste interface between the generator and the condenser to separate
heat recovery techniques along with their operational require- the water vapor from LiBr solution. Depending on the micropro-
ments in order to assess their suitability and effectiveness for data cessor heating load, the evaporator temperature could vary
center applications. Based on a comparison between data centers between 25 C (corresponded to 100 W heat load) and 50 C (corre-
operational thermodynamic conditions and the operational sponded to 1000 W heat load), while the temperature of the gener-
requirements of the reviewed techniques, absorption refrigeration ator was xed at 90 C. The COP of the system for a heat load of
(AR) and Organic Rankine Cycle (ORC) were identied as the most 100 W was 0.74. Kim et al. [11] replaced the LiBr solution in their
promising technologies for data center waste heat reuse [1]. Since miniature AR design with an ionic liquid (IL) to lower the toxicity
AR systems can operate with generator temperatures as low as of the system and to eliminate the risk of crystallization that occurs
7090 C, they are compatible with the waste heat temperatures with LiBr/Water AR systems. Through this study, IL and ve HFC
from liquid-cooled and two phase cooled data centers [1]. How- refrigerants were evaluated as the system working uid pairs.
ever, this technology is not suitable for waste heat from air cooled The effects of generator and absorber temperatures, waste-heat
data centers unless an additional heat booster is employed [1]. quality, and system design on the performance of miniaturized
In the standard vapor compression refrigeration systems used AR/heat pump were investigated. The study showed that for an
in data center Computer Room Air Conditioning (CRAC) units, a evaporator temperature of 41 C, the maximum COP is 0.35 with
considerable amount of power is consumed by the compressor. an evaporator cooling capability of 36 W and a generator temper-
Since liquids have much lower specic volumes than vapors, ature range of 50110 C. The addition of a solution heat exchanger
replacing the vapor compression system with an AR system, in increased the COP to 0.8. Kim et al. [12] extended their theoretical
which a liquid solution of absorbent uid and refrigerant is used analysis to explore and evaluate the effect of using 14 different
as the cycle working uid, can result in considerable power mixtures of refrigerants and imidazolium-based ILs as the working
savings. The application of AR to replace or supplement a CRAC uid pairs on the performance of the AR system. The saturation
not only reduces the power necessary to run the system, but also temperatures at the evaporator and condenser were set at 25 C
has the benet of using the data center waste heat as the AR gen- and 50 C, respectively, with a power dissipation of 100 W. The
erator heat source. Using data center waste heat in AR systems highest COP of 0.9 was achieved with a combination of water as
leads to direct reduction of the load on data center CRAC systems, the refrigerant and 1-ethyl-3-methylimidazolium tetrauorobo-
through two ways: rst by removing the servers heat and using rate as the absorbent uid at a generator temperature of 70 C.
that to operate the AR units and second by generating chilled water Chiriac and Chiriac [13,14] designed and developed a miniatur-
which can be used for cooling additional servers, which is of signif- ized ammoniawater AR system for smaller scale power electron-
icant economic benet to most data center operators. Another ics. However, the absorber was powered by an electric heater and
important benet of this technology is the fact that there are no not by waste heat. In the original system [13], a gravity-based
issues with site co-location and the technology can be retrot into thermosyphon was used to circulate the binary solution, but was
existing data centers if space permits [1]. replaced with a mini-pump in the next design [14]. This improve-
During the past decade, the feasibility of using AR systems for ment increased the COP of the system from 0.4 to 0.7. The system
data center or server waste heat recovery has been considered by was designed for heat removal capacity of 100 W and the temper-
a few researchers. We performed a comprehensive review of those atures at the generator, condenser and evaporator were set at
research works in our previous study [1], however, an updated 125 C, 40 C and 10 C, respectively. Table 1 provides a short
summary of that review is presented in this paper. Haywood summary of research in which the application of AR in data
et al. [8,9] focused on utilizing data center dissipated heat to oper- center/server/micro-processor dissipated heat removal and
ate a 10-ton single-effect lithium bromidewater AR chiller by recovery is investigated.
transferring the thermal energy from the highest power compo-
nents on a server blade to the generator of the AR chiller. Any
excess thermal energy was stored for use as a booster during 2. Objectives and conceptual framework
non-peak operating conditions. Through this approach the CRAC
load was reduced by using part of the waste heat that otherwise This study addresses the technical and economic issues associ-
would be cooled by the CRAC unit to run the AR unit, and also ated with integration of AR technology to a data center thermal
through cooling produced by AR chiller, which resulted in further management system. We develop steady-state thermodynamic
reduction in the CRAC load. The AR was able to operate using waste models to investigate the theoretical feasibility of using the heat
heat with temperature range of 7095 C and thus was consistent energy dissipated by a server or a number of servers in running
with liquid or two phase cooled data centers. At the design point of an AR system and producing cooling for other server or servers
the system where the quality and quantity of the supply heat were in the same center. Our main objective in this paper is to perform
88 C and 50.2 kW, respectively, the maximum cooling load of a thermo-economic analysis for data center waste heat recovery by
Please cite this article in press as: Ebrahimi K et al. Thermo-economic analysis of steady state waste heat recovery in data centers using absorption refrig-
eration. Appl Energy (2014), http://dx.doi.org/10.1016/j.apenergy.2014.10.067
K. Ebrahimi et al. / Applied Energy xxx (2014) xxxxxx 3
Table 1
The specications of AR units studied for data center/server/microprocessor heat removal and recovery.
developing a novel waste heat recovery system combined of a 3.1.1. Simulation of LiBrwater AR
simplied version of an on-chip two-phase cooling system [15] Fig. 1 shows a schematic view of the components and state
and an AR unit. These two systems are integrated through the points in a single-effect LiBrwater AR cycle. The system required
replacement of the condenser in the on-chip two phase cooling inputs are waste heat to the generator and t power to the pump.
system with the generator of the AR unit. The system output is the cooling load produced at the evaporator.
Our starting point is the creation of accurate thermodynamic The low pressure vapor refrigerant enters the absorber (state point
models to simulate LiBrwater and waterAmmonia AR systems, 10) where, through an exothermic process, the refrigerant is
which are veried against the reference data available in the litera- absorbed into a carrier liquid, which is called the absorbent. For
ture. The veried models can then be used to simulate data center this case, the refrigerant is water and the absorbent is LiBr. Then,
operating conditions to determine whether LiBrwater or water the strong liquid LiBrwater solution exiting the absorber (state
Ammonia shows better performance under data center operation 1) is pumped to a higher pressure and enters the solution heat
conditions. It will be shown that LiBrwater is a more efcient sys- exchanger (state 2). Recall that energy costs for pumping a liquid
tem for data center/server operating conditions. Thus the novel heat are considerably smaller than compressing a vapor as it is needed
recovery conguration is developed by replacing the condenser in in a conventional vapor compression cycle. By passing through the
the recently introduced on-chip two phase cooling system [15] with solution heat exchanger, the strong solution is preheated and then
the generator of a single effect LiBrwater AR machine. Then, the enters the generator (state 3). Through an endothermic process,
models are expanded to perform steady-state energy and exergy the refrigerant is separated from the absorbent. The exiting refrig-
analysis for the novel conguration. We explain why the traditional erant vapor (state 7) continues on to the condenser while the now
denition for the coefcient of performance (COP) is not appropri- weak liquid solution in the generator passes through an expansion
ate in evaluating the performance of the novel combined system
and, in its place, introduce a new gure of merit. Through a paramet-
ric analysis, the effect of server waste heat quality, server coolant
type, solution peak concentration, solution heat exchanger effec-
tiveness, evaporator temperature, and operating pressures on the
performance of the novel system is investigated. Through the para-
metric study, the requirements for the specications of AR chiller
and server operating condition for achieving highest coefcient of
performance are determined.
Finally, using the thermodynamic model and cost information
provided by the AR industry, an economic analysis is carried out
to calculate the return of investment (ROI) and payback period
when this technology is used for data center waste heat recovery.
3. Thermodynamic analysis
Please cite this article in press as: Ebrahimi K et al. Thermo-economic analysis of steady state waste heat recovery in data centers using absorption refrig-
eration. Appl Energy (2014), http://dx.doi.org/10.1016/j.apenergy.2014.10.067
4 K. Ebrahimi et al. / Applied Energy xxx (2014) xxxxxx
valve and returns to the absorber (states 4, 5 and 6). The loop of the 3. Knowing the pressure and temperature of water at the outlet of
absorber, pump, generator and expansion valve are together generator (state 7), condenser (state 8) and evaporator (state
referred to as a chemical compressor. 10), the enthalpy is calculated through the REFPROP subroutine.
In order to develop our thermodynamic model for this system, This subroutine was developed by the National Institute of
the governing equations for the conservation of mass and energy Standards and Technology (NIST) and more information in this
must be applied to each component as well as to the whole system. regard can be found in [18].
The general forms of these equations are 4. Considering assumptions (1), (3) and (4), the energy equation
for the expansion valve can be simplied as
dm X X
_
m _
m 1
dt in out h8 h9 6
The temperature at the absorber, generator, condenser, and 7. Considering assumptions (1), (2), (4) and (5) and knowing the
evaporator. heat transfer rate (cooling capacity) at the evaporator and
The effectiveness of the solution heat exchanger. the refrigerant properties before and after this component, the
The heat transfer rate at evaporator (the cooling capacity of the mass ow rate is calculated through an energy balance
system). Q_ e
_r
m 8
h10 h9
Considering the input variables and implementing the assump- where Q_ e (kW) is the heat transfer rate at the evaporator and m
_r
tions, the steady state thermodynamic model consists of the fol- (kg/s) is the refrigerant mass ow rate.
lowing steps:
8. Knowing the refrigerant mass ow rate from the previous step
1. Knowing the temperatures at condenser and evaporator, the and considering the fact that the mass ow rate of LiBr in both
corresponding pressures are calculated through the use of the weak and strong solution streams are the same (m _ 1X1 m_ 6 X 6 ),
following empirical equation developed by Sun [16]. knowing the concentration of LiBr in the weak and strong solu-
tions (from step 2) and taking assumption (1) into account, the
3892:7
p exp 9:48654 3 weak and strong solution mass ow rates can be found through
42:6776 T a mass balance at the absorber
where p(MPa) is pressure and T(K) is temperature. m_ r X1
_6
m 9
X6 X1
2. Knowing the temperatures at absorber, evaporator, generator,
and condenser the concentration of LiBr in the strong and weak m_ r X6
solutions is calculated using equations presented in Lansing _1
m 10
X6 X1
[17].
In the above equations m _ 6 (kg/s) and m_ 1 (kg/s) denote mass
LiBr concentration in strong solution: ow rates for week and strong solutions, respectively.
49:04 1:125T 1 T 10
X1 X2 X3 4 9. Knowing the LiBr concentration in the strong solution from step
134:65 0:47T 1
2 and the temperature of absorber, the density of strong solu-
LiBr concentration in weak solution: tion exiting the absorber (state 1) is calculated using the follow-
ing equation [20]
49:04 1:125T 4 T 8
X4 X5 X6 5
134:65 0:47T 4
q 1145:36 470:84X 1374:79X 2 0:333393 0:571749XT
In Eqs. (4) and (5), X (kg LiBr/kg solution) denotes absorbent
concentration and T (C) temperature. 11
Please cite this article in press as: Ebrahimi K et al. Thermo-economic analysis of steady state waste heat recovery in data centers using absorption refrig-
eration. Appl Energy (2014), http://dx.doi.org/10.1016/j.apenergy.2014.10.067
K. Ebrahimi et al. / Applied Energy xxx (2014) xxxxxx 5
Table 2
Input data from the reference models.
Variable/parameter Symbol (unit) Data series #1 Rubio-Maya et al. [21] Data series #2 Aphornratana and Eames [22]
Evaporator cooling load Q_ e (kW) 201.29 Free to choose
Generator temperature Tg (C) 84.8 90
Condenser temperature Tc (C) 39.8 35
Evaporator temperature Te (C) 8.6 5
Absorber temperature Ta (C) 35.5 35
Solution heat exchanger effectiveness e (%) 70.7 60
where q (kg/m3) is the solution density, T(K) is the solution temper- 15. Knowing the enthalpy at all state points of the cycle the heat
ature, and X (kg LiBr/ kg solution) is the concentration of absorbent transfer rate at absorber, generator and condenser can be
in the solution. calculated through the application of energy balance for
each component:
10. Knowing the pressures (from step 1), mass ow rate of
strong solution (from step 8), and density of strong solution Absorber Q_ a m
_ r h10 m
_ 6 h6 m
_ 1 h1 19
(from step 9), the pump work is calculated by [22]
Generator Q_ g m
_ r h7 m
_ 6 h4 m
_ 1 h3 20
_P _ 1 p2 p1
m
W 12
q1 Condenser Q_ c m
_ r h7 h8 21
_ 1 cp1 < m
If m _ 4 cp4 T3 is calculated by
3.1.2. Simulation of waterammonia AR
In contrast to LiBrwater AR systems in which water works as
T 3 T 2 eT 4 T 2 16
the cycle refrigerant, in waterammonia systems, water plays the
role of absorbent. These cycles also require a rectication column
Following the calculation of T5 or T3, the enthalpy at states 3 or due to the existence of water in the ammonia vapor leaving the
5 is calculated using Eq. (7). generator (state 8 in Fig. 2), which adversely affects the perfor-
mance of the system. The rectier removes water from the exiting
13. The only remaining enthalpy not calculated from the previ- vapor [26]. Passing through the rectier, a portion of the heat of the
ous steps is determined by the energy balance at the solu- vapor exiting the generator is removed by the lower-temperature,
tion heat exchanger waterammonia solution stream leaving the absorber and pumped
to the high-side pressure (process 12 in Fig. 2). Therefore the
_ 4 h4 m
m _ 2 h2 m
_ 3 h3 m
_ 5 h5 17 majority of water content in the vapor exiting the generator is con-
densed to liquid and drips back to the generator (state 9 in Fig. 2).
14. Considering assumptions 1, 3, and 4 the enthalpy of weak However, there still exists some water in the liquid ammonia exit-
solution at the outlet of solution valve (state 6) is deter- ing the condenser (state 11) and when its pressure falls by passing
mined by an energy balance through the expansion valve, the liquid ashes to vapor which is
almost pure ammonia. Therefore, the liquid has a higher fraction
h6 h5 18 of water after than before the expansion valve. As the mixture of
Please cite this article in press as: Ebrahimi K et al. Thermo-economic analysis of steady state waste heat recovery in data centers using absorption refrig-
eration. Appl Energy (2014), http://dx.doi.org/10.1016/j.apenergy.2014.10.067
6 K. Ebrahimi et al. / Applied Energy xxx (2014) xxxxxx
Table 3
Comparison between our model and reference models regarding the methods applied for the calculation of refrigerant and solution properties.
Type of property Present study Rubio-Maya et al. [21] Aphornratana and Eames [22]
Pressures of the system Sun [16] Sun [16] Not specied
Refrigerant (water) properties REFPROP [18] Lansing [17] Irvine and Liley [24]
Absorbent (LiBr) concentration Lansing [17] Lansing [17] Not specied
Solution (LiBrwater) properties Density from Lee et al. [20], specic heat from Kaita [19] Patterson and Perez-Blanco [25]
ASHRAE handbook [23], rest of properties from Kaita [19]
Table 4
Verication of our model by comparing our results with those of Rubio-Maya et al.
[21]. The input data are the same as used by Rubio-Maya et al. [21] (see Table 2).
Table 5
Verication of model by comparing results with those of Aphornratana and Eames [22]. The input data are the same as used by Aphornratana and Eames [22] (see Table 2).
Variable MATLAB Model-present study Aphornratana and Eames [22] Deviation (%)
Generator heat input (%) 100 100 0
Solution pump work (%) 0.0001898 0.001 81% (the large deviation is because the
difference of two very small values are
divided by a very small value)
Evaporator heat input (%) 75.561 75.865 0.4
Absorber heat reject (%) 95.212 94.885 0.3
Condenser heat reject (%) 80.349 80.981 0.7
Solution heat exchanger load (%) 10.443 11.785 11.3
Please cite this article in press as: Ebrahimi K et al. Thermo-economic analysis of steady state waste heat recovery in data centers using absorption refrig-
eration. Appl Energy (2014), http://dx.doi.org/10.1016/j.apenergy.2014.10.067
K. Ebrahimi et al. / Applied Energy xxx (2014) xxxxxx 7
concentration of the refrigerant before and after the condenser is With the high and low-side pressures known and knowing the
constant, the ammonia fraction in the liquid stream exiting the con- temperatures of absorber and generator, as well as the tempera-
denser at state 11 will be the same as ammonia fraction in the vapor ture at the outlet of rectier, the ammonia molar fraction and
stream after rectier at state 10. Then, knowing the temperature enthalpy of solution at the state 1 (liquid solution exiting the
and ammonia fraction at state 11, the high-side pressure is calcu- absorber), 8 (vapor refrigerant exiting the generator), 9 (dripped
lated through the following equilibrium equation [28] liquid exiting the rectier and returning to the generator), and 5
(the returning solution from generator to the absorber) are calcu-
p pT; x 23
lated using Eqs. (22), (22)(25).
where x is the ammonia molar fraction of the liquid solution. Using In order to determine the total enthalpy at the 13 (after the
the updated pressure from Eq. (23), another iteration is performed evaporator), we need to calculate the enthalpy of liquid and vapor
and the ammonia fraction at state 10 is recalculated through Eq. solutions at this state as well as the mass ow rates of purge liquid
(22). The recalculated ammonia fraction at state 10 is then used and vapor leaving the evaporator. From conservation of mass for
for updating the calculations at state 11 and nding a new value ammonia between the states 10 and 13 it can be concluded that
for the high-side pressure through Eq. (23). This loop of calculations
x13 ym;10 30
is repeated until the prediction for high-side pressure converges.
A similar iterative technique is used in the next step for the cal- where x is the total ammonia mass fraction and ym is the ammonia
culation of the low-side pressure. As explained above, when the mass fraction in the vapor solution. Knowing the value of y at the
condenser exiting liquid solution passes through the expansion state 10, ym,10 can be determined through the following equation
valve some of the liquid solution ashes to vapor. Therefore, there
y10 M Ammonia
is a mixture of liquid and vapor at state 12. Making an initial guess ym;10 31
for the pressure at the inlet of evaporator (state 12), and knowing
y10 MAmmonia 1 y10 M Water
the temperature at this state, which is equal to the evaporator inlet Knowing the low-side pressure and assuming the temperature
temperature, the molar fractions of ammonia in the vapor and at the outlet of evaporator is 5.5 C higher than at the inlet, the
liquid solution at the inlet of evaporator is calculated through ammonia molar fractions in the liquid and vapor leaving the evap-
Eqs. (22) and (23). Following the calculation of ammonia molar orator can be found through Eqs. (22) and (23). Subsequently, the
fractions, the enthalpies of liquid and vapor solutions at state 12 ammonia mass fractions at the state-point 13 can be calculated in a
are determined using [28]: similar fashion presented in Eqs. (27) and (31). Then, the enthal-
hL hL T; x 24 pies of liquid and vapor phases at state 13 are determined using
Eqs. (24) and (25). The mass ow rate of ammonia at this state
hG hG T; y 25 can be calculated using either the total ammonia mass fraction
(x13) or the ammonia mass fractions in the liquid and vapor
where hL (kJ/kg) and hG (kJ/kg) represent the enthalpy of liquid and streams. Since these results will be the same, we develop the fol-
vapor solutions, respectively. Since the refrigerant expansion valve lowing equation
is adiabatic, the enthalpy across this valve is unchanged, and there-
fore h11 = h12. This equality is used to calculate the mass fraction of
x13 wpl xm;13 1 wpl ym;13 32
liquid solution at the state 12. Knowing that the solution at the state where wpl is the mass ow rate ratio of the purge liquid leaving the
11 (before passing through the valve) is completely liquid, the evaporator:
enthalpy of the solution at this state is easily calculated using Eq. m_ pl
(24). Then, considering the equality of enthalpies and mass-ow wpl 33
_ total
m
rates across the valve, the mass fraction of liquid solution (fL) at
the state 12 is calculated [26]: The total enthalpy at the state 13 is calculated next
x11 M Ammonia X X
xm;11 27 m_x m_x 36
x11 M Ammonia 1 x11 M Water in out
Please cite this article in press as: Ebrahimi K et al. Thermo-economic analysis of steady state waste heat recovery in data centers using absorption refrig-
eration. Appl Energy (2014), http://dx.doi.org/10.1016/j.apenergy.2014.10.067
8 K. Ebrahimi et al. / Applied Energy xxx (2014) xxxxxx
In order to nd mass ow rates at states 4 and 5 in terms of Eq. (24). Knowing the enthalpy values at states 3, 5, and 6, the
mass ow rate at state 8, conservation of mass is applied around enthalpy at state 4 is calculated by energy balance on the solution
the generator for both the solution and ammonia streams heat exchanger
_ 9m
m _4m
_ 5m
_8 41 _ 3 h3 m
m _ 5 h5 m
_ 6 h6
h4 50
m_4
m _ 4 x4 m
_ 9 xm;9 m _ 5 xm;5 m
_ 8 ym;8 42 Finally, the heat transfer rates at the evaporator and the gener-
ator are calculated from Eqs. (51) and (52).
where x4 xm;1 . Replacing m _ 9 and m
_ 10 in Eqs. (41) and (42) with
the right hand side of Eqs. (39) and (40) and solving Eqs. (41) and Q_ e m
_ 12 h13 h12 51
_ 4 and m
(42) for m _ 5 will result in
" ! # Q_ g m
_ 8 h8 m
_ 5 h5 m
_ 4 h4 52
ym;8 xm;5 ym;8 ym;10 xm;9 xm;5
_4
m m_8 43
xm;1 xm;5 xm;9 ym;10 xm;1 xm;5 The COP is calculated using Eq. (22).
Fig. 3 shows a comparison of the MATLAB model developed
" ! # through the above steps to results in the archival literature [26].
ym;8 xm;1 ym;8 ym;10 xm;1 xm;9
_5
m m_8 44 As we see, for the temperatures below 105 C there is an excellent
xm;1 xm;5 xm;9 ym;10 xm;1 xm;5 agreement between the two models, however, for the higher tem-
perature the deviation between two models is increased to as high
Since the mass ow rate through pump, m _ 4 , is known m
_ 8 can be
as 11%. This can be attributed to the difference between the data
easily calculated using Eq. (43) and subsequently all the mass ow
sets used for the thermodynamic properties.
rates are calculated through Eqs. (39), (40) and (44). In order to cal-
culate the temperature at state 3, we apply an energy balance
3.1.3. Superiority of LiBrwater AR
around rectier
Fig. 4 shows a comparison of COP values for single-effect LiBr
_ 8 h8 m
m _ 2 h2 m
_ 2 h3 m
_ 9 h9 m
_ 10 h10 0 45 water and waterAmmonia AR cycles. The high and low pressures
for waterAmmonia AR cycle are 1.1 and 0.3 MPa, respectively
In this, equation, enthalpies at states 8 and 9 have been calcu-
while the corresponding values for LiBrwater are 4.8 103and
lated from the above. Knowing the temperature and molar fraction
9 104MPa. It can be seen that for the range of data center waste
of Ammonia at state 10, the enthalpy at this state can be found
heat temperatures (temperatures less than 80 C), the COP of LiBr
through Eq. (25). With pump work is negligible, we see that
water AR cycle is considerably higher than waterAmmonia (up to
h2 = h1, where h1 is known from ae previous step. Since the mass
14% at 65 C). The higher COP under data center operating condi-
ow rates are also known from a previous step, the only unknown
tions is not the only advantage of LiBrwater over waterAmmonia
in Eq. (45), h3, is calculated. The temperature at state 3 can now
AR systems. The other advantages are a simpler system (as LiBr
calculated by Eq. (24).
water does not need the rectication column, while this compo-
The temperature at states 4 and 6 are calculated using e-NTU
nent is necessary for the waterAmmonia) and less pump work
method [29]. The NTU is from
(due to operation at very low pressures). On the other hand, the
UA most challenging issue is the risk of crystallization, which imposes
NTU 46
C min an upper limit on the LiBr concentration (65%). However, because
of the better performance of LiBrwater AR systems for data center
_ p ) for the
where Cmin is the smaller of the two heat capacity rates (mc
operating conditions, this type of AR system is selected for integra-
cold and hot streams entering the heat exchanger at states 3 and 5
tion into the data center design, and the system is designed to
(see Fig. 2). Following the calculation of NTU, the effectiveness for
avoid the risk of crystallization.
the solution heat exchanger (counter-ow type) is
Fig. 4 also shows that for both systems, the COP decreases at
1 expNTU1 C r higher generator temperatures. Our hypothesis for this behavior
for C r < 1 : e 47a is as follows. When operating at higher generator temperatures,
1 C r expNTU1 C r
the vapor exiting the generator and entering the condenser is at
NTU elevated temperatures and as such requires a larger heat transfer
for C r 1 : e 47b surface at the condenser in order to be completely turned into a
1 NTU
saturated liquid at the condenser exit. As the heat transfer surface
where Cr is the heat capacity ratio is xed, therefore instead of completely condensing, there may be
C min
Cr 48
C max
where Cmin is as above and Cmax is the maximum of the two heat
capacity rate for the cold and hot streams entering the heat exchan-
ger at states 3 and 5 (see Fig. 2). After the calculation of effective-
ness, the unknown temperatures T4 and T6 are calculated by solving
_ 3 cp;3 T 4 T 3
m
e 49a
C min T 5 T 3
_ 5 cp;5 T 5 T 6
m
e 49b
C min T 5 T 3
After nding the temperature at state 6 and assuming that the Fig. 3. Verication of the MATLAB model developed in this study for the steady-
solution at this state is completely liquid phase [26] (which state thermodynamic simulation of the rened waterammonia AR system (see
resulted in x6 = x5), the enthalpy at this state is calculated using Fig. 2) by comparing results with those of Stoecker and Reed [26].
Please cite this article in press as: Ebrahimi K et al. Thermo-economic analysis of steady state waste heat recovery in data centers using absorption refrig-
eration. Appl Energy (2014), http://dx.doi.org/10.1016/j.apenergy.2014.10.067
K. Ebrahimi et al. / Applied Energy xxx (2014) xxxxxx 9
Table 7
Verication of our model expanded for exergy analysis through the comparison of our
results with those of Rubio-Maya et al. [21]. The input data are the same as used by
Rubio-Maya et al. [21] (see Table 2).
Table 6
Comparison between our and reference models regarding the adopted parameters and methods in the 2nd law analysis.
Type of parameter Present study Rubio-Maya et al. [21] Aphornratana and Eames [22]
Reference point used for the calculation of exergy (Eq. (53)) T0 = 25 C and T0 = 25 C and Water at T0 = 20 C (reference pressure not specied)
p0 = 101.3 kPa p0 = 101.3 kPa
Entropy of water REFPROP [18] IAPWS [30] Irvine and Liley [24]
Entropy of LiBrwater solution Kaita [19] Kaita [19] Their own novel method [22]
Please cite this article in press as: Ebrahimi K et al. Thermo-economic analysis of steady state waste heat recovery in data centers using absorption refrig-
eration. Appl Energy (2014), http://dx.doi.org/10.1016/j.apenergy.2014.10.067
10 K. Ebrahimi et al. / Applied Energy xxx (2014) xxxxxx
Q_ e
COPnew 55
W_P
Fig. 5. Schematic of a waste heat recovery system where the server dissipated heat Fig. 7 shows the results of our analysis, when the minimum
is converted to useful cooling. temperature of the waste heat (generator temperature) is 65 C.
The upper limit for the range of upgraded waste heat temperature
is restricted by the second law of thermodynamics. It is seen that
and nally delivered to the generator of AR (through a heat the upper limit for the temperature of the upgraded waste heat
exchanger). AR is operated using the heat removed from the can be as high as 100 C (for HFC 245fa) and as low as 85 C (for
micro-processor as the input at the generator. The cooling load HFC 1234ZE). We also see that using HFC 1234ZE as the working
produced by the AR can then be used to remove heat from addi- uid in the server cooling cycle leads to the highest COP value of
tional micro-processors. The indicated state points in the simpli- 5. Another very useful gure of merit (FoM) in analyzing the sys-
ed on-chip two-phase cooling cycle illustrated in Fig. 5 are tem is dened as:
shown in Fig. 6, which is the pressure-enthalpy diagram for HFC-
245fa. We see that through the energy added by the compressor Q_ MP
FoM 56
the quality of waste heat is upgraded and reaches temperatures Q_ e
as high as 90 C [15]. As mentioned earlier, the risk of crystalliza-
tion only occurs when the concentration of the strong absorbent where Q_ MP is rate of heat dissipated by the micro-processors and Q_ e
exceeds 65%. As is shown in Fig 6 (a) in this conguration, the con- is the cooling produced at the evaporator (see Fig. 5). The signi-
centration of the strong line remains below this threshold. cance of this gure of merit comes from the fact that it indicates
the number of servers or racks required to be coupled with an AR
3.2.1. Upgrading the model for the steady-state thermodynamic in order to produce sufcient cooling load for another single server
simulation of the novel system or rack. Fig. 8 shows the results using FoM for various generator and
In addition to the assumptions listed in Section 3.1.1, the fol- microprocessor temperatures. For the results presented in Fig. 8,
lowing assumptions were also considered when upgrading the HFC 245fa was used as the coolant in the server cooling cycle, the
model for the novel system. reect worst case conditions. The upper limit on the coolant tem-
perature at the microprocessor is determined by the 2nd law anal-
1. At state 1 the coolant is a saturated liquid (Fig. 6(b)). ysis. It is seen that in the worst case, the heat dissipated by 5 servers
2. At states 2 and 3 the coolant is a saturated vapor (Fig. 6(b)). or racks can operate an AR machine that will produce sufcient
3. At state 4 the coolant is sub-cooled (Fig. 6(b)). cooling for another single server or rack of the same type.
Please cite this article in press as: Ebrahimi K et al. Thermo-economic analysis of steady state waste heat recovery in data centers using absorption refrig-
eration. Appl Energy (2014), http://dx.doi.org/10.1016/j.apenergy.2014.10.067
K. Ebrahimi et al. / Applied Energy xxx (2014) xxxxxx 11
Fig. 6. Different state-points of refrigerant (water), solution (LiBrwater) and coolant (HFC-245fa) for the waste heat recovery system depicted in Fig. 5. (a) P-T diagram for
LiBrwater AR [32]. (b) P-h diagram for the coolant (HFC-245fa) in the simplied on-chip two-phase cooling cycle.
Fig. 7. The effect of upgraded waste heat temperature on the COP of the rst type of
the novel heat recovery system. Fig. 8. The behavior of the newly dened gure of merit (see Eq. (56)) when the
generator and microprocessor temperatures vary and the refrigerant is HFC 245fa.
Please cite this article in press as: Ebrahimi K et al. Thermo-economic analysis of steady state waste heat recovery in data centers using absorption refrig-
eration. Appl Energy (2014), http://dx.doi.org/10.1016/j.apenergy.2014.10.067
12 K. Ebrahimi et al. / Applied Energy xxx (2014) xxxxxx
Table 9
Specications of Yazaki AR products which are designed to operate with low-quality heat [34]. A ton of cooling is 3517 W (12,000 Btu/h).
Table 10
The number of 10 kW racks provides heat and cooled by each type of Yazaki AR chillers.
5-Ton (17.6 kW) 10-Ton (35.17 kW) 20-Ton (70.34 kW) 30-Ton (105.5 kW) 50-Ton (175.8 kW)
Number of racks provide required heat for the 3 6 11 16 26
operation of ONE machine
Number of racks cooled by ONE machine 1 3 7 10 17
Number of racks removed from CRAC unit using 4 9 18 26 43
ONE machine
Water ow-rate supplied by the racks provide 1530 (Max 3060 (Max 55110 (Max 80160 (Max 156260 (Max
heat for AR machine (GPM) allowable: 22.8) allowable: 45.6) allowable: 91.32) allowable: 136.92) allowable: 228.48)
Comparison of the cooling capacity and heat input of every AR A 10 MW data center features 1000 racks of 10 kW each. Con-
unit from Table 9 with the amount of heat dissipated by a single sidering Table 10, we see that every 43 racks can be managed by
rack leads us to the numbers presented in Table 10. For each type a 50-ton AR chiller. Of these 43 racks, 26 of them dissipate their
of AR, Table 10 shows the number of racks that are required to pro- heat directly into the generator of the AR system and another 17
vide sufcient heat at the generator and how many racks can be have their heat removed using the chilled water produced by the
cooled using the cooling produced by the same AR unit. Thus, AR system. Therefore, using 23 chillers of 50-ton capacity, 989
Table 10 shows the total number of racks removed from the CRAC racks can be removed from the CRAC unit and the CRAC load is
load through the use of a specic AR unit. decreased from 1000 to 11 (98.9% CRAC load reduction). Of course
For a 2 MW data center, which consists of 200 racks, the infor- the data center can be designed to feature racks in multiples of 43,
mation presented in Table 10 shows that 18 racks can be managed potentially eliminating any need for server cooling using the CRAC,
with a 20-ton AR chiller. Of these 18 racks, 11 of them dissipate depending on specic data center operational needs. The arrange-
their heat directly into the generator of the AR system. Another 7 ment of racks and chiller, in every group of 43 racks is shown in
have their heat removed using the evaporator of the AR system. Fig. 10. The footprint area of the 50 ton chiller is approximately
Therefore using ten 20-ton capacity chillers, 180 racks can be that of 3 racks. The chiller is located at the center in this gure,
removed from the CRAC load (18 managed by each chiller). There- but many different congurations are possible.
fore, the number of racks cooled by CRAC unit is decreased from An economic analysis for both the 2 and 10 MW data centers is
200 to 20 (90% CRAC load reduction). Of course the data center shown in Tables 11 and 12, respectively. In calculating the num-
can be designed to have racks in multiples of 18, potentially elim- bers presented in these tables the following assumptions were
inating any need for server cooling using the CRAC, depending on made:
specic data center operational needs. The footprint area of each
20-ton AR unit is about that of two racks. The arrangement of racks The number of data center operating hours in one year is 7500.
and AR units in groups of 18 racks is shown in Fig. 9. In this partic- The cost of electricity is $0.08 per kW h [15].
ular arrangement, the AR unit is located at the center, but congu- The prices of the chillers are: 50-ton: $75,000, 20-ton: $50,000
ration can vary for each data center. [35].
Please cite this article in press as: Ebrahimi K et al. Thermo-economic analysis of steady state waste heat recovery in data centers using absorption refrig-
eration. Appl Energy (2014), http://dx.doi.org/10.1016/j.apenergy.2014.10.067
K. Ebrahimi et al. / Applied Energy xxx (2014) xxxxxx 13
Table 11
Table of energy savings for a 2 MW data center when ten 20-ton AR chillers are installed.
Before being retrotted with Yazaki 20-Ton absorption After being retrotted with Yazaki 20-Ton absorption machines
machines
Annual electricity 200 (number of racks) 10 kW (power of each 200 (number of racks) 10 kW (power of each rack) 7500 h (total working
consumption rack) 7500 h (total working hours in a year = 15,000,000 hours in a year = 15,000,000
(kW h/year)
Annual electricity cost 15,000,000 kW h (annual electricity consumption) 0.08 15,000,000 kW h (annual electricity consumption) 0.08 (electricity cost per
($/year) (electricity cost per kW h) = $1,200,000 kW h) = $1,200,000
Annual cooling cost $1,200,000 for a typical data center energy usage of cooling $1,200,000 (cooling cost before retrot) (1/10) = $120,000 the number of
($/year) system is equivalent to energy usage of servers racks cooled by data center cooling system decreased from 200 to 20
(reduction factor of 1/10)
Total annual cost 2,400,000 1,320,000
($/year)
Table 12
Table of energy savings for a 10 MW data center when twenty-three 50-ton chillers are used in the waste heat recovery.
Before being retrotted with Yazaki 20-Ton absorption machines After being retrotted with Yazaki 20-Ton absorption machines
Annual electricity 1000 (number of racks) 10 kW (power of each 200 (number of racks) 10 kW (power of each rack) 7500 h (total working
consumption rack) 7500 h (total working hours in a hours in a year = 15,000,000
(kW h/year) year = 75,000,000
Annual electricity 75,000,000 kW h (annual electricity 75,000,000 kW h (annual electricity consumption) 0.08 (electricity cost
cost ($/year) consumption) 0.08 (electricity cost per per
kW h) = $6,000,000 kW h) = $6,000,000
Annual cooling cost $6,000,000 for a typical data center energy usage of $6,000,000 (cooling cost before retrot) (11/1000) = $66,000 the number
($/year) cooling system is equivalent to energy usage of of
servers racks cooled by data center cooling system decreased from 1000 to 11
(reduction factor of 11/1000)
Total annual cost 12,000,000 6,066,000
($/year)
The price of 20-ton cooling tower is $20,000 [35]. two phase cooling cycle developed by Marcinichen et al. [15,32] we
The cooling cost and power cost are the same. introduced a novel conguration for waste heat recovery and usage
from electronic components. In developing this conguration some
With the above, an approximate payback period can be estimated. simplications were made in the original on-chip, two phase cool-
We say approximate because the calculations do not include the cost ing cycle and the condenser was replaced by the generator of a
of operation and maintenance of the AR system. The approximate absorption refrigeration system to produce cooling. A detailed
payback period of an AR system for a 2 MW data center is between thermodynamic rst and second law analysis is developed and val-
78 months, and for a 10 MW data center is between 45 months. idated. The analysis showed that LiBrwater systems outperform
waterAmmonia AR systems when applied to data center operat-
5. Summary and conclusions ing conditions.
For waste heat conversion to cooling, the selection of the server
Considering the fact that liquid/two phase cooled data centers coolant type, the quality of heat delivered to the generator and the
are able to deliver higher quality waste heat and using the on-chip temperature of the micro-processor are shown to have signicant
Please cite this article in press as: Ebrahimi K et al. Thermo-economic analysis of steady state waste heat recovery in data centers using absorption refrig-
eration. Appl Energy (2014), http://dx.doi.org/10.1016/j.apenergy.2014.10.067
14 K. Ebrahimi et al. / Applied Energy xxx (2014) xxxxxx
effects on the performance of the system. Also, it is shown that the [13] Chiriac V, Chiriac F. Miniaturized refrigeration system with absorption
application to microelectronics cooling, In: Proceedings of ASME
2nd law of thermodynamics imposes limits not only on the COP of
InterPACK07, 812 July 2007, Vancouver, Canada.
the system but also on the maximum allowable temperature of [14] Chiriac V, Chiriac F. Absorption refrigeration method with alternative water
micro-processor and generator. Results from the model show by ammonia solution circulation system for microelectronics cooling. In:
dissipating the waste heat from 3 to 5 racks, the absorption refrig- Proceedings of ITHERM 2010, 12th intersociety conference: 2010 June 25;
Las Vegas, NV, USA.
eration system can provide sufcient chilled water to cool an addi- [15] Marcinichen JB, Olivier JA, Thome JR. On-chip two-phase cooling of
tional rack of the same type. Results from an economic analysis for datacenters: cooling systems and energy recovery evaluation. Appl Therm
the AR system showed the payback period can be as short as Eng 2012;41:3651.
[16] Sun DW. Thermodynamic design data and optimum design maps for
45 months. absorption refrigeration systems. Appl Therm Eng 1997;17(3):21121.
[17] Lansing FL. Computer modeling of a single-stage lithium bromide/water
Acknowledgements absorption refrigeration unit. JPL Deep Space Network Progress Report, 1976;
DSN 42-32: p. 247257.
[18] Lemmon EW, Huber ML, McLinden MO. NIST reference uid thermodynamic
This material is based upon work supported by the National and transport propertiesREFPROP version 8 -Users Guide. National Institute
Science Foundation under Grant No. IIP-1134810. Any opinions, of Standards and Technology Standard Reference Data Program, 2007;
Maryland 20899.
ndings, and conclusions or recommendations expressed in this [19] Kaita Y. Thermodynamic properties of lithium bromidewater solutions at
material are those of the author(s) and do not necessarily reect high temperatures. Int J Refrig 2001;24:37490.
the views of the National Science Foundation. [20] Lee RJ, DiGuilio RM, Jeter SM, Teja AS. Properties of lithium bromidewater
solutions at high temperatures and concentration. II. Density and viscosity.
ASHRAE Trans 1990;96(Pt. 1):70928.
References [21] Rubio-Maya C, Pachero-Ibarra JJ, Belman-Flores JM, Galvan-Gonzalez SR,
Mendoza-Covarrubias C. NLP model of a LiBrH2O absorption refrigeration
[1] Ebrahimi K, Jones GF, Fleischer AS. A review of data center cooling technology, system for the minimization of the annual operating cost. Appl Therm Eng
operating conditions and the corresponding low-grade waste heat recovery 2012;37:108.
opportunities. Renew Sust Energy Rev 2014;31:62238. [22] Aphornratana S, Eames IW. Thermodynamic analysis of absorption
[2] Brunschwiler T, Smith B, Ruetsche E, Michel B. Toward zero-emission data refrigeration cycle using the second law of thermodynamics method. Int J
centers through direct reuse of thermal energy. IBM J Res Dev 2009;53(3): Refrig 1995;18(4):24452.
11:113. [23] ASHRAE. Handbook of Fundamentals. Atlanta, 1997.
[3] Koomey JG. Growth in data center electricity use 2005 to 2010. A report by [24] Irvine TF, Lily PE. Steam and Gas Tables with Computer Equations. USA: Academic
Analytics Press, completed at the request of The New York Times; August 2011. Press; 1984.
[4] Independent Statistics & Analysis. U.S. Energy Information Administration. [25] Patterson MR, Perez-Blanco H. Numerical ts of properties of Lithium
November 11, 2013 <http://www.eia.gov/electricity/>. bromidewater solutions. ASHRAE Trans 1998;96(Part 2):205977.
[5] Brown R, Masanet E, Nordman B, Tschudi B, Shehabi A, Stanley J, et al. Report to [26] Stoecker WF, Reed LD. Effect of operating temperatures on the coefcient of
congress on server and data center energy efciency: public law 109- performance of aquaammonia refrigeration systems. ASHRAE Trans 1971;77:
431. Berkeley, CA, USA: Lawrence Berkeley National Laboratory; 2008. 16370.
[6] Nozaki Y. Development of higher-voltage direct current power feeding system [27] Kaushik SC, Bhardwaj SC. Theoretical analysis of ammoniawater absorption
for ICT equipment. NTT Tech Rev 2009;7(10). cycles for refrigeration and space conditioning systems. Int J Energ Res
[7] Iyengar M, Schmidt R. Analytical modeling for thermodynamic 1982;6:20525.
characterization of data center cooling Systems. J Electron Packaging 2009;131 [28] Patek J, Klomfar J. Simple functions for fast calculations of selected
. EP-08-1048. thermodynamic properties of the ammoniawater system. Int J Refrig 1995;18(4):
[8] Haywood A, Sherbeck J, Phelan P, Varsamopoulos G, Gupta, SKS. A sustainable 22834.
data center with heat-activated cooling. In: Proceedings of ITHERM 2010, 12th [29] Incropera FP, DeWitt DP, Bergman TL, Lavine AS. Fundamentals of heat and
intersociety conference: 2010 June 25; Las Vegas, NV, USA. mass transfer. 6th ed. US: John Wiley & Sons; 2006. p. 686688.
[9] Haywood A, Sherbeck J, Phelan P, Varsamopoulos G, Gupta SKS. [30] Wagner W et al. The IAPWS industrial formulation 1997 for the thermodynamic
Thermodynamic feasibility of harvesting data center waste heat to drive an properties of water and steam. J Eng Gas Turb Power 2000;122(1):15084.
absorption chiller. Energ Convers Manage 2012;58:2634. [31] Marcinichen JB, Thome JR, Michel B. Cooling of microprocessors with micro-
[10] Kim YJ, Joshi YK, Fedorov AG. An absorption based miniature heat pump evaporation: a novel two-phase cooling cycle. Int J Refrig 2010;33(7):126476.
system for electronics cooling. Int J Refrig 2008;31(1):2333. [32] Florides GA, Kalogirou SA, Tassou SA, Wrobel LC. Design and construction of a
[11] Kim S, Kim YJ, Joshi YK, Fedorov AG, Kohl PA. Absorption heat pump/ LiBrwater absorption machine. Energy Convers Manage 2003;44:2483508.
refrigeration system utilizing ionic liquid and hydrouorocarbon refrigerants. [33] The ofcial webpage of Broad U.S.A. Inc: <http://www.broadusa.com>.
J Electron Packaging 2012;134(031009):19. [34] Yazaki Manual for water red chiller/chiller-heater-WFC-S series.
[12] Kim YJ, Kim S, Joshi YK, Fedorov AG, Kohl PA. Thermodynamic analysis of an [35] Personal Communication. Steve A.Cohen, Yazaki Energy Systems, Inc.
absorption refrigeration system with ionic-liquid/refrigerant mixture as a <scohen@yazakienergy.com>.
working uid. Energy 2012;44:100516.
Please cite this article in press as: Ebrahimi K et al. Thermo-economic analysis of steady state waste heat recovery in data centers using absorption refrig-
eration. Appl Energy (2014), http://dx.doi.org/10.1016/j.apenergy.2014.10.067