Paper O.3.8.

1 -1-

HEAT-DRIVEN ADSORPTION CHILLER SYSTEMS FOR

SUSTAINABLE COOLING APPLICATIONS

Patrick Rucha, Sarmenio Salibaa, Chin Lee Onga, Yazid Al-Shehrib, Abdalrahman Al-Rihailib,
Ahmed Al-Mogbelb, and Bruno Michela
a
IBM Research – Zurich, Säumerstrasse 4, 8803 Rüschlikon, Switzerland
b
King Abdulaziz City for Science and Technology (KACST), Riyadh 11442, Saudi Arabia

Abstract: Thermally driven adsorption chillers enable utilization of low-grade heat below
100°C to drive refrigeration cycles at minimum electrical energy consumption. Therefore, this
technology is crucial to enable waste heat usage and substantially cut the power drawn from
electrical grids and thereby improve the energy efficiency of air-conditioning infrastructure.
Two scenarios for the implementation of adsorption heat pumps are considered: (i) hot-water
cooled datacenters and (ii) solar cooling. Simulation studies reveal distinct operating
windows for the adsorption chiller in each application and, accordingly, the need for
adsorbent materials tailored to the desired temperature lift to optimally use the driving heat.
Adsorption experiments were carried out in a custom-built apparatus to quantify the cooling
power per unit mass of adsorbent and its thermal response in relation to the applied
temperature swing. The results reveal poor adsorbent utilization in present adsorption chillers
due to a mismatch between the adsorption characteristics of the adsorbent and the boundary
conditions of the application as well as incomplete thermal cycling of the adsorbent due to
poor heat transport.

Key Words: adsorption, solar energy, datacenter, waste heat

1 INTRODUCTION

Space heating and cooling account for over one-third of energy consumption in buildings,
which in turn are responsible for more than one-third of global energy consumption [1]. By
2050, the cooling loads in hot developing countries will increase rapidly between three- and
more than six-fold resulting in a severe surge in electricity demand for cooling which already
accounts for more than 70% of domestic electricity consumption in hot countries during peak
demand [1, 2] In the course of the 21st century, it is likely that the share of cooling of the total
space-conditioning CO2 emissions will rise from 5% in 2000 to more than 70% by 2100 [3].
As a result, there is a strong need for efficient cooling technologies with minimal electricity
consumption and CO2 footprint in order to provide a sustainable air-conditioning
infrastructure, particularly in hot countries.

The abundance of low-grade heat from industrial processes [4] and solar heat [5] motivates
the use of thermally driven heat pumps [6] for air-conditioning applications. Adsorption heat
pumps rely on the reversible physisorption of a refrigerant vapor, the adsorptive, on a porous
material, the adsorbent [7]. The adsorption refrigeration cycle is driven mainly by temperature
swings applied to the adsorbent, thus requiring minimal amounts of electricity to operate.
Further, water can be readily used as an environmentally benign and highly effective
refrigerant due to its excellent latent heat of vaporization per unit volume and mass [8].
Therefore, adsorption cooling is a promising technology to make use of low-grade heat from
sustainable energy sources aimed at reducing electricity peaks for air-conditioning.

In contrast to vapor compression cooling, the adsorbent material in an adsorption chiller

assumes a critical role in driving the refrigeration cycle instead of a mechanical compressor.

11thIEA Heat Pump Conference 2014, May 12-16 2014, Montréal (Québec) Canada
Paper O.3.8.1 -2-

As will be shown in this contribution, the ideal adsorption characteristics of the

adsorbent/adsorptive working pair may vary considerably for different cooling scenarios.
Here, two distinct heat-driven cooling applications are investigated by simulation and
experiment, namely (i) hot-water cooled datacenters [9] and (ii) solar cooling in desert
climate [10].

Using direct water cooling, heat dissipated from servers can be recovered at sufficiently high
temperatures to enable free cooling during the entire year, thus eliminating the need for
compression cooling and air handling which accounts for 25%-50% datacenter power
consumption [9, 11]. With optimized cooling equipment temperature levels are high enough
to operate an adsorption chiller [12]. For heterogeneous datacenters consisting of hot-water
cooled servers as well as conventional air-cooled IT equipment, improved energy efficiency
may be realized through the integration of an adsorption chiller since the heat recovered from
the water-cooled servers can be used to provide chilled air for the air-cooled IT infrastructure.
Overall, this approach can reduce the energy cost of operating server hardware, which is
estimated to account for ca. 25 B$ of worldwide datacenter spending or about half the yearly
expenditure on server hardware [13]. The fraction of power cost increases with the transition
to cloud computing due to use of low-cost commodity hardware.

The usage of solar thermal energy for adsorption cooling is an attractive sustainable cooling
option in hot countries. In Saudi Arabia, for example, the installed electricity generating
capacity is required to double in the timeframe 2007-2017 in order to meet the growth in
peak electricity demand, which is primarily governed by air-conditioning needs in
summer [14]. Adsorption cooling has the greatest potential for electricity reduction for air-
conditioning due to the absence of internal pumps and compressors, leading to a high
electrical coefficient of performance (COPel). At the same time, the high solar irradiance in
hot countries enables high solar thermal collector yields and thus substantial driving power
for the adsorption process. However, like all thermally driven cooling technologies,
adsorption cooling suffers from a large recooling need which is exacerbated due to the high
ambient temperatures in hot countries during cooling season. These conditions differ
markedly from the nominal conditions of commercial adsorption chillers, leading to an
underutilization of the adsorption capacity and performance degradation. Establishing a
better understanding of the operating environment is crucial for the design of improved
adsorption systems for solar adsorption cooling in hot climates.

2 METHODOLOGY

2.1 System simulation

Heat-driven adsorption cooling systems using a silica gel/water chiller with 15 kW nominal
cooling capacity [15, 16] were implemented in Polysun® 5.9 (Vela Solaris, Switzerland). The
simulation of server heat usage for adsorption cooling (Figure 1a) assumed a hot-water
cooled server rack with nominal outflow temperature of 65°C and a total thermal load of
30 kW/m2. In addition, an air-cooled server rack containing storage and network switches
with a total thermal load of 10 kW/m2 was added to the IT hardware configuration. While free
cooling can be carried out for the hot-water cooled servers throughout the year by direct heat
rejection to ambient in a dry recooler, the air-cooled servers can only make use of free
cooling for ambient temperatures lower than 18°C. To test the feasibility of all-year operation
without the need for vapor compression cooling, the heat recovered from the water-cooled
servers was used to drive the adsorption chiller to charge a chilled water buffer tank. During
periods in which elevated ambient temperatures do not allow free cooling for the air-cooled
servers, the chilled water from this tank provides the necessary air-conditioning of the rack
space. Simulations were carried out for two locations, Munich (Germany) and Dallas, TX

11thIEA Heat Pump Conference 2014, May 12-16 2014, Montréal (Québec) Canada
Paper O.3.8.1 -3-

(USA), in order to assess the feasibility of adsorption cooling in the datacenter environment
and to identify the operating conditions of the adsorption chiller for these two scenarios.

Solar cooling simulations were performed for a single-family household in Riyadh, Saudi
Arabia, with 150 m2 living area using flat-plate solar thermal collectors (135 m2 adsorber
area) with a hot water buffer (8 m3 storage volume) to drive the adsorption chiller. Heat
rejection was carried out using a dry recooler. Further details of the simulation methodology
for the solar cooling scenario can be found in [10].

(a)

(b)

Figure 1: Simulated heat-driven adsorption cooling scenarios:

(a) datacenter waste heat and (b) solar thermal cooling

2.2 Experimental investigation of adsorption characteristics

The adsorption isotherms of water on a micro-/mesoporous silica gel (RD-type, Fuji Silysia,
Japan) were measured using a gravimetric sorption analyzer (DVS Vacuum, Surface
Measurement Systems, United Kingdom). The adsorption characteristics were interpreted in
the context of Dubinin’s theory [17] based on the differential molar work of adsorption:

A = -RT ln(p/psat) (1)

where R is the universal gas constant, T the temperature of the adsorbent with the
corresponding saturation pressure psat(T), and p is the actual pressure. Using Equation (1),
different adsorption states (p, T) can be compared using the single variable A.

For measurement of the adsorption dynamics, an in-house experimental setup was used
(Figure 2). The apparatus comprises a water-filled evaporator kept at constant temperature
and a chamber containing the adsorber heat exchanger. The temperature of the heat

11thIEA Heat Pump Conference 2014, May 12-16 2014, Montréal (Québec) Canada
Paper O.3.8.1 -4-

exchanger fluid flowing through the adsorber can be switched between a high temperature
(Th) and a medium temperature (Tm) reservoir in order to induce adsorption or desorption.
The rate of water sorption in g/s was quantified by measuring the flow rate of water vapor
using a flow sensor. Multiplication of the mass flow rate by the vaporization enthalpy, ΔvapH in
J/g, yielded the cooling power in W. Normalization of the cooling power with respect to the
dry adsorbent mass resulted in the specific cooling power (SCP) in W/kg.

Figure 2: Experimental scheme for characterization of adsorption rates using thermal swing
adsorption with online infrared thermography

The adsorber heat exchanger used in this work was a tube-lamella heat exchanger unit with
two aluminum lamellae (spacing 7 mm) enclosing the silica gel fixed bed adsorbent (Figure 2,
right). The diameter of the silica beads was between 1.8 and 2.0 mm and a total dry
adsorbent mass of 7.989 g was used. An infrared camera (Silver 420, FLIR Systems Inc.,
USA) was used to observe the thermal response of the adsorbent during cycling.

3 RESULTS AND DISCUSSION

3.1 Adsorption system simulation

The ambient temperatures as well as the water-cooled and air-cooled server temperatures
during free cooling operation are plotted in Figure 3 for the two investigated datacenter
locations. The shaded areas indicate the temperature span for each of the temperature
series. Due to the high outlet temperature of the hot-water cooled servers compared to the
ambient temperature, free cooling can be achieved at both locations during the entire year.
On the other hand, free cooling for the air-cooled servers was found to be feasible 90% of
the time in Munich and only 64% of the time in Dallas due to the higher ambient
temperatures of the latter location. In particular, during the months of June to August, no free
cooling is possible for the air-cooled servers in Dallas. During this time, all of the thermal
energy is dissipated via the cold buffer of the adsorption cooling system (Figure 4). Notably,
no vapor compression chiller is needed at either location at any time to provide cooling for
the air-cooled servers, as the entire cooling load is provided by the adsorption chiller driven
by the waste heat from the hot-water cooled servers.

The cold buffer is useful when the workloads on the water-cooled and air-cooled servers
differ substantially. Thus, when no chilled water production is required for air-conditioning of
the air-cooled servers, the cold buffer tank is charged to a temperature of 18°C in the top

11thIEA Heat Pump Conference 2014, May 12-16 2014, Montréal (Québec) Canada
 Paper O.3.8.1 -5-

stratified layer. During maximum chilled water production for air-conditioning, the tank outlet
temperature rises up to 18°C in Munich and 22°C in Dallas (Figure 5), which is still low
enough to maintain secure operation of the air-cooled IT infrastructure.

(a) (b)
70 70

60 60
Ambient Ambient
50 Water-cooled server outlet 50 Water-cooled server outlet
Temperature [°C]

Temperature [°C]
Air-cooled server inlet Air-cooled server inlet
40 40

30 30

20 20

10 10

0 0
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Month Month
Figure 3: Ambient temperature, water-cooled server outlet temperature and air-cooled server
intake temperature for (a) Munich and (b) Dallas

(a) (b)

Figure 4: Partitioning of air-cooled server heat dissipation between free cooling and cold buffer
for (a) Munich and (b) Dallas.

(a) (b)

Figure 5: Temperatures of the cold buffer tank stratified layers for (a) Munich and (b) Dallas

11thIEA Heat Pump Conference 2014, May 12-16 2014, Montréal (Québec) Canada
 Paper O.3.8.1 -6-

Scenario A: Datacenter Munich The temperature triple comprising the

110
driving temperature Th, heat sink
100 Hot water inlet temperature Tm and chilled water
Cooling water inlet
90
Chilled water outlet temperature Tl describes the
80 thermodynamic operating conditions of the
Temperature [°C]

70 adsorption chiller. The three scenarios (A)

60 datacenter operation in Munich, (B)
50 datacenter operation in Dallas and (C) solar
40 cooling in Riyadh are compared in terms of
30
their temperature triples in Figure 6. The
20
data points represent the average monthly
value while the shaded areas indicate the
10
temperature span for each month.
0
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Month The yearly average temperature triples {Th,
Tm, Tl} for the three scenarios are {64°C,
Scenario B: Datacenter Dallas 32°C, 19°C} for Scenario A, {64°C, 35°C,
110
20°C} for Scenario B and {77°C, 35°C,
100 Hot water inlet 17°C} for Scenario C. The corresponding
Cooling water inlet
90
Chilled water outlet thermal coefficients of performance (COPth),
80 defined as the ratio of cooling energy to
driving energy, were 0.45 (Scenario A), 0.33
Temperature [°C]

70

60 (Scenario B) and 0.40 (Scenario C),

50 respectively. The most demanding
40
operation occurs in scenarios B and C
30
during the summer months when the heat
sink temperature is highest. The
20
temperature triples in July are {64°C, 38°C,
10
22°C} with a COPth of 0.21 for Scenario B
0
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec and {79°C, 38°C, 18°C} with a COPth of
Month 0.30 for Scenario C. Improved efficiencies
can be obtained by exploiting day/night
Scenario C: Solar cooling Riyadh temperature fluctuations in conjunction with
110
a larger cold storage.
100

90 The driving temperature in the solar cooling

80 scenario (C) exhibits a substantially larger
spread than in the two datacenter scenarios
Temperature [°C]

70

60 Hot water inlet

(A, B) due to the large variability in the solar
50 Cooling water inlet collector yield compared to the workload on
40
Chilled water outlet the servers. Also, server workloads may be
30
scheduled in a thermally aware manner so
as to obtain more uniform heat dissipation in
20
time. The variability in driving temperature
10
for the solar cooling scenario may be
0
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec minimized by implementing a variable pump
Month speed in the solar loop to target a constant
Figure 6: Monthly temperature triples for the collector outflow temperature or by
three investigated scenarios increasing the size of the hot water buffer.
However, a larger hot water buffer than the
presently implemented 8 m3 may prove impractical for a single-family home.

11thIEA Heat Pump Conference 2014, May 12-16 2014, Montréal (Québec) Canada
Paper O.3.8.1 -7-

3.2 Experimental investigation of adsorbent utilization

The adsorption isotherms of water on Fuji RD silica gel at temperatures between 30°C and
70°C are summarized in Figure 7a. According to Dubinin’s theorem for microporous
adsorbents, the water uptake curves should coincide when plotting the adsorbed amount
against the work of adsorption A. A curve fitting of all the data to the polynomial
ln(W) = a + b A was performed, where W is the volume of water adsorbed per unit adsorbent
mass and (a, b) are fitting parameters. The best fit yielded a = -0.8797 and b = -2.78 x 10-4,
which is in good agreement with previously published values [18]. A considerable spread of
about ±30% with respect to this fit was found for the present data (Figure 7b).

(a) (b)

Figure 7: Water adsorption isotherms on Fuji RD silica gel (a) and corresponding volume
adsorbed plotted against the work of adsorption A

The temperature triples for the three adsorption cooling scenarios investigated in section 3.1
were transformed into (p/psat)-values on the water adsorption isotherm measured at 50°C
based on Equation (1). Thus, the thermodynamic maximum amount of water that can be
cycled for each scenario is defined (Figure 8).

Figure 8: Water adsorption isotherm of Fuji RD silica gel at 50°C with corresponding
thermodynamic operating windows for the three investigated adsorption cooling scenarios

11thIEA Heat Pump Conference 2014, May 12-16 2014, Montréal (Québec) Canada
Paper O.3.8.1 -8-

The maximum possible amount of water cycled is ΔmA = 20 wt% for Scenario A (Datacenter
Munich), ΔmB = 14 wt% for Scenario B (Datacenter Dallas) and ΔmC = 15 wt% for Scenario C
(Solar cooling Riyadh). Note that these cycling amounts are significantly smaller than the
total water uptake capacity of the material near saturation (p/psat → 1), which is of the order
of 40 wt%. Due to the approximately linear adsorption isotherm of Fuji RD silica gel, the
adsorbent is able to cycle appreciable amounts of water within different p/psat windows.
However, only less than half of the total adsorption capacity can be used in a typical
application, which represents an ineffective use of the adsorbent material. The poor
utilization of the adsorbent is reflected in the low COPth values obtained by simulation of the
three scenarios in Section 3.1.

The operating window of the adsorbent is narrowed even further due to incomplete thermal
cycling. Figure 9 shows the cooling power and temperature excursions of the adsorber
loaded with Fuji RD silica gel and cycled between temperature extremes of 90°C and 40°C at
an evaporator temperature of 20°C (vapor pressure 23 mbar) with 10 minutes between
switching events. The adsorbent temperature was derived from infrared thermography and
represents an average value of the entire fixed bed. The specific cooling power goes through
a maximum shortly after each switching event when the largest temperature gradient and
hence driving force for adsorption is achieved during the experiment [19]. However, the
temperature response of the adsorbent lags behind that of the heat exchanger significantly
(Figure 9, bottom) due to the poor heat transfer between the heat exchanger and the silica
gel and the poor thermal conductivity within the silica gel fixed bed itself.

Figure 9: Specific cooling power (top) and temperature excursions of heat exchanger and
adsorbent (bottom) of a tube-lamella heat exchanger loaded with Fuji RD silica

From an adsorbent utilization perspective, it is favorable to employ long cycling times which
result in higher COPth for a given thermodynamic operation window. On the other hand, the
SCP value decreases markedly with prolonged cycling time (Figure 9, top). Thus, cycling
times for commercial adsorption chillers are selected close to the time for maximum SCP, but
shifted to longer times to obtain improved COPth values. For a commercial adsorption chiller
utilizing silica gel/water as the working pair, a cycle time of 450 s has been reported [20]. It is
obvious from Figure 9 that such short cycling times result in a significant reduction in the
effective temperature swing of the adsorbent material. In fact, the actual change in the
amount of water adsorbed for the cycling conditions in Figure 9 corresponds to ca. 5 wt% in

11thIEA Heat Pump Conference 2014, May 12-16 2014, Montréal (Québec) Canada
Paper O.3.8.1 -9-

each cycle for the full 10 minutes cycling time. For comparison, the thermodynamic limit for
the temperature triple {87°C, 40°C, 20°C} is 12.5 wt% based on Equation (1) and the
adsorption isotherm of Fuji RD silica. Thus, there is a two-fold underutilization of the
adsorbent in this case: (i) the operating boundary conditions only allow for a maximum
utilization of ca. 31% of the total adsorption capacity of the adsorbent, and (ii) the poor
thermal performance of the system further reduces the utilization of the adsorbent down to
ca. 13% of the total adsorption capacity for a cycling time of 10 minutes. An improved
adsorption cooling system in terms of adsorbent utilization, SCP and COPth can be obtained
by two main routes, namely the selection of an adsorbent material which specifically matches
the boundary conditions of the application [21] or as process intensification through
improvement of heat and mass transport within the adsorber heat exchanger [7].

4 CONCLUSIONS

Adsorption cooling is attractive to make use of the vast pool of low-grade heat available from
industrial processes or solar heat. Such utilization of otherwise unused heat reduces the
primary fuel consumption and carbon footprint for air-conditioning and can help meeting the
projected rise in global cooling demand while minimizing the imposed strain on power grids
associated with vapor compression cooling. The implementation of adsorption cooling in
datacenter applications with low-grade driving heat below 70°C is feasible to reduce or even
eliminate the need for vapor compression cooling, thereby supporting server
implementations with minimum overhead energy cost over their lifetime. Also, solar-driven
cooling for residential air-conditioning is feasible even in regions with very high ambient
temperatures, such as Saudi Arabia. Present adsorption cooling technology is limited in
terms of efficiency and effectiveness due to the underutilization of the adsorbent materials in
specific application scenarios. This was demonstrated experimentally in terms of the
adsorption characteristics of widely employed micro-/mesoporous silica gel as well as the
limited temperature swing undergone by a fixed-bed adsorber consisting of this adsorbent in
a widely employed adsorber configuration. Advances in tailored adsorbent materials and
enhancement of heat and mass transport within the adsorbent can boost the attractiveness
of adsorption cooling for a wide range of industrial and residential applications.

5 REFERENCES

[1] International Energy Agency 2013, “Technology Rodmap: Energy efficient building
envelopes.”

[2] Booz & Company 2012, “Unlocking the Potential of District Cooling: The Need for
GCC Governments to Take Action.”

[3] Isaac, M., and Van Vuuren, D. P. 2009, “Modeling global residential sector energy
demand for heating and air conditioning in the context of climate change,” Energy
Policy, Vol. 37, pp. 507–521.

[4] U.S. Department of Energy 2008, “Waste Heat Recovery: Technology and
Opportunities in U.S. Industry.”

[5] International Energy Agency 2012, “Technology Roadmap: Solar Heating and
Cooling.”

[6] Núñez, T. 2010, “Thermally Driven Cooling: Technologies, Developments and

Applications,” Journal of Sustainable Energy, Vol. 1, pp. 16–24.

11thIEA Heat Pump Conference 2014, May 12-16 2014, Montréal (Québec) Canada
Paper O.3.8.1 - 10 -

[7] Critoph, R., and Zhong, Y. 2005, “Review of trends in solid sorption refrigeration and
heat pumping technology,” Proceedings of the Institution of Mechanical Engineers,
Part E: Journal of Process Mechanical Engineering, Vol. 219, pp. 285–300.

[8] Critoph, R. E. 1988, “Performance limitations of adsorption cycles for solar cooling,”
Solar Energy, Vol. 41, pp. 21–31.

[9] Brunschwiler, T., Smith, B., Ruetsche, E., and Michel, B. 2009, “Toward zero-emission
data centers through direct reuse of thermal energy,” IBM Journal of Research and
Development, Vol. 53, pp. 1–13.

[10] Al-Mogbel, A., Ruch, P., Al-Rihaili, A., Al-Ajlan, S., Gantenbein, P., Witzig, A., and
Michel, B. “The Potential of Solar Adsorption Air-Conditioning in Saudi Arabia: A
Simulation Study,” in Proceedings of the 5th International Conference on Solar Air-
Conditioning, 2013, pp. 256–261.

[11] Rasmussen, N. 2010, “Electrical Efficiency Modeling of Data Centers,” APC White
Paper #113 Revision 2.

[12] Meyer, N., Ries, M., Solbrig, S., and Wettig, T. 2013, “iDataCool: HPC with Hot-Water
Cooling and Energy Reuse,” Lecture Notes in Computer Science, Vol. 7905, pp. 383–
394.

[13] IDC 2011, “Market Analysis Perspective: Worldwide Datacenter Trends and Strategies
2011.”

[14] Business Monitor International 2012, “Saudi Arabia Power Report Q1 2012.”

[15] Núñez, T., Mittelbach, W., and Henning, H. M. 2007, “Development of an adsorption
chiller and heat pump for domestic heating and air-conditioning applications,” Applied
Thermal Engineering, Vol. 27, pp. 2205–2212.

[16] http://www.sortech.de. [Accessed: 31-Jan-2014].

[17] Dubinin, M. M. 1967, “Adsorption in Micropores,” Journal of Colloid and Interface

Science, Vol. 23, pp. 487–499.

[18] Aristov, Y. I., Tokarev, M. M., Freni, A., Glaznev, I. S., and Restuccia, G. 2006,
“Kinetics of water adsorption on silica Fuji Davison RD,” Microporous and Mesoporous
Materials, Vol. 96, pp. 65–71.

[19] Aristov, Y. I. 2009, “Optimal adsorbent for adsorptive heat transformers: Dynamic
considerations,” International Journal of Refrigeration, Vol. 32, pp. 675–686.

[20] Chua, H., Ng, K., Malek, A., Kashiwagi, T., Akisawa, A., and Saha, B. 1999, “Modeling
the performance of two-bed, sillica gel-water adsorption chillers,” International Journal
of Refrigeration, Vol. 22, pp. 194–204.

[21] Aristov, Y. I. 2013, “Challenging offers of material science for adsorption heat
transformation: A review,” Applied Thermal Engineering, Vol. 50, pp. 1610–1618.

11thIEA Heat Pump Conference 2014, May 12-16 2014, Montréal (Québec) Canada