Page 1
1 Copyright © 2014 by ASME
OPTIMUM SOLAR HDH DESALINATION FOR SEMI-ISOLATED COMMUNITIES
USING HGP AND GA’S
Khalid M. Abd El-Aziz Graduate Teaching & Research Assistant
Dept. of Mechanical Design and Production Cairo University, Cairo, Egypt
[email protected]
Karim Hamza Senior Research Fellow
Dept. of Mechanical Engineering University of Michigan, Ann Arbor, MI 48109-2102
[email protected]
Mohamed El Morsi Associate Professor
Dept. of Mechanical Engineering American University in Cairo, New Cairo, 11835, Egypt
Ain Shams University, Cairo, 1156, Egypt [email protected]
Ashraf O. Nassef Professor
Dept. of Mechanical Engineering American University in Cairo, New Cairo, 11835, Egypt
[email protected]
Sayed M. Metwalli Professor Emeritus
Dept. of Mechanical Design and Production Cairo University, Cairo, Egypt
[email protected]
Kazuhiro Saitou Professor
Dept. of Mechanical Engineering University of Michigan, Ann Arbor, MI 48109-2102
[email protected]
ABSTRACT Modeling and unit-cost optimization of a water-heated
humidification-dehumidification (HDH) desalination system
were presented in previous work of the authors. The system
controlled the saline water flow rate to prevent salts from
precipitating at higher water temperatures. It was then realized
that this scheme had a negative impact on condensation
performance when the controlled flow rate was not sufficiently
high. This work builds on the previous system by disconnecting
the condenser from the saline water cycle and by introducing a
solar air heater to further augment the humidification
performance. In addition, improved models for the condenser
and the humidifier were used to obtain more accurate
productivity estimations. The Heuristic Gradient Projection
(HGP) optimization procedure was also refactored to result in
reduced number of function evaluations to reach the global
optimum compared to Genetic Algorithms (GA’s). A case study
which assumes a desalination plant on the Red Sea near the city
of Hurghada is presented. The unit-cost of produced fresh water
for the new optimum system is $0.5/m3 compared to $5.9/m
3
for the HDH system from previous work and less than the
reported minimum cost of reverse osmoses systems.
NOMENCLATURE 𝐴 surface area [m
2]
𝑎 humidifier specific area [m2/m
3]
𝑏 incidence angle modifier coefficient
𝐶 cost [$]
𝑐 specific heat capacity [J/kg K]
𝑐𝑝 inflation and yearly maintenance cost factor
𝐸𝑂𝑇 equation of time [minute]
𝐹𝑅 heat removal effectiveness factor
ℎ heat transfer coefficient [W/ m2 K]
𝐻ℎ𝑢𝑚 humidifier packing height [m]
𝐼𝐴𝑀 incidence angle modifier
𝐼 irradiance [W/m2]
𝑖 specific enthalpy [kJ/kg dry air, kJ/kg water]
∆𝑖𝑣 latent heat of vaporization [kj/kg]
𝑘 thermal conductivity [W/m K]
𝐾 mass transfer coefficient [kg/ m2s]
𝐿𝐶 longitude correction [hour]
𝐿𝐶𝑇 local clock time [24-hour format]
�̇� mass flow rate [kg/s]
�̇�𝑠 specific fresh water production [kg/m2 day]
𝑚𝑟 mass flow ratio [kg air/kg water]
Proceedings of the ASME 2014 International Design Engineering Technical Conferences & Computers and Information in Engineering Conference
IDETC/CIE 2014 August 17-20, 2014, Buffalo, New York, USA
DETC2014-34598
Page 2
2 Copyright © 2014 by ASME
𝑁 day number [0-365]
�̇� heat transfer [W]
𝑟 tube radius [m]
𝑇 temperature [K]
𝑡𝑝 plant lifetime
𝑡𝑠 solar time [hour]
𝑈 overall heat transfer coefficient [W/m2K]
𝑈𝐿 heat loss coefficient [W/m2 ⁰C]
𝑉 volume [m3]
𝑉𝐷 daily fresh water production [m3/day]
�̇� volume flow rate [m3/s]
�̇�𝑠 specific air flow rate [m3/m
2 s]
𝑌 coordinate along humidifier height [m]
𝛽 collector tilt angle [rad]
𝛿 declination angle [rad]
𝛾 collector azimuth angle [rad]
𝜔 humidity ratio [kg vapor/kg dry air], hour angle [rad]
∅ latitude angle [rad]
𝜌 reflectance of surrounding surface
(𝜏𝛼) effective transmittance-absorptance product
𝜃𝑖 solar angle of incidence [rad]
Subscripts
1 distillate layer
𝑎 air
𝑎𝑚𝑏 ambient
𝐵 brine
𝑏, 𝑛 beam normal
𝑐 collector
𝑐𝑜 coolant
𝑐𝑜𝑛𝑑 condenser
𝐷 distillate
𝑑, ℎ diffuse horizontal
𝑑𝑏 dry bulb
𝑓, 𝑖 fluid at inlet
ℎ𝑢𝑚 humidifier
𝑖 interface
𝑠 saturation
𝑡 transferred heat
𝑡, ℎ total irradiance falling on a horizontal surface
𝑢 useful energy gain
𝑣 vapor
𝑤 water
𝑤𝑏 wet blub
1. INTRODUCTION Although a person requires about 2-5 liters of freshwater
per day for direct consumption, most of human water
requirement is consumed in agriculture and industry. Growth in
these sectors is, therefore, limited by the amount of available
fresh water, which is only 3% of Earth’s water content. This has
long motivated development of seawater desalination systems.
However, conventional desalination technologies are energy
intensive and rely mainly on fossil fuels which are becoming
less available over time.
Desalination using solar energy is friendly to the
environment and relies on a renewable energy source (RES).
The challenge remains for solar water desalination systems to
be cost-effective. Usually, the major cost component in these
systems is the initial investment, since their reliance on RES
allows for low running costs when compared with conventional
systems.
In Egypt, the Nile River is the main source of fresh water.
However, many locations on both the Red Sea and the
Mediterranean Sea are too far from the Nile River to have a
continuous fresh water supply. In this study, we focus on such
remote locations which require a small to medium-scale
desalination plant with a relatively low initial investment.
Several technologies may be utilized to effect water
desalination using solar energy. Reverse Osmosis (RO) is one
of the most widely-used technologies. It depends on pushing
the saline water through a nano-filter (membrane) by a high-
pressure pump, effectively allowing only the water molecules
to pass through the filter leaving salt and most unwanted
constituents behind [1]. RO is known to be a highly productive
process: reported unit-cost of produced fresh water is $2-3/m3
[2]. However, RO desalination systems require expertise in
their installation and maintenance, making them less suitable
for decentralized water production in remote regions. Also, RO
desalination systems are sensitive to inlet water salinity, further
discouraging their use, because the Red Sea is one of the most
saline open water surfaces in the world [3].
HDH desalination is another technology that was studied in
various configurations by a lot of previous work in literature.
The main concept is that air capacity for water vapor increases
logarithmically with temperature [4]. So, if the air is heated to a
modest temperature and passed through a saline water stream, it
will carry water vapor free of salts. The air can then be cooled
so as the water vapor will condense into fresh water.
HDH desalination is usually coupled with solar-thermal
rather than thermal-electric technologies used in case of RO.
Solar-thermal technologies are usually less expensive and
easier to maintain. Also, solar HDH desalination systems
require less expertise in installation and maintenance making
them more suitable for remote regions. In addition, because the
effectiveness of the humidification process is only mildly
affected by inlet water salinity content [5], HDH desalination is
less sensitive to salinity than RO. However, Narayan et al. [6]
reported a unit-cost of $3-7/m3 for produced fresh water using
HDH desalination which is higher than unit-cost of RO. This
work is a continuation of effort to reduce the unit-cost of HDH
desalination in order to take advantage of its versatility.
This paper briefly describes a model of an HDH
desalination system from previous work of the authors [7].
Improvements to that system are then introduced along with
their respective reasons. The model equations for the modified
system are derived, and it is then optimized using HGP. Finally,
comparison of optimum points using HGP and GA’s for both
systems is presented.
Page 3
3 Copyright © 2014 by ASME
2. RELATED WORK
2.1 Variable Water Flow Rate El-Morsi et al. [8] modeled and optimized a single-stage
water-heated system. Similar to other systems, the condenser is
cooled using inlet saline water in order to also preheat it. A
genetic algorithm (GA) was used for optimizing the system.
The optimum system yielded 4.8 kg fresh water/m2 of solar
collector area per day, and the unit cost of produced fresh water
was $8/m3. The system addressed salt precipitation at high
saline water temperatures by using a solar water heater that will
at most reach the temperature limit for the design weather
conditions. However, in conditions other than maximum, the
water temperature will drop below the limit which had a
negative effect on humidification performance.
To address this issue, Abd El-Aziz et al. [7] modified this
system to control saline water temperature by variating its flow
rate. A closed-loop solar water heater was used to allow the
heat storage medium to reach its full capacity. The modified
system was optimized using a tailored optimization technique
based on HGP. The system yielded 8.3 kg/m2 day. And the
optimum unit-cost of produced fresh water was $5.9/m3. The
temperature control ensured the water will always be at the
temperature limit for maximum humidification performance.
However, when the saline water flow rate was not sufficiently
high, condensation performance was negatively affected. As
will be shown in section 3, this issue is also addressed in this
paper.
2.2 Air-heated Systems Narayan et al. [6] reviewed a wide range of system designs
including water and air-heated systems. The authors noted a
disadvantage of the air-heated cycle: Air temperature drops in
the humidifier as it exchanges heat with saline water.
Eventually, condensation takes place at a low temperature
which has a negative effect on condensation performance.
Farsad and Behzadmehr [9] modeled a hybrid cycle that
heats both air and saline water. This overcomes the
disadvantage of the air-heated cycle because air will leave the
humidifier at a higher temperature.
Orfi et al. [10] also modeled and simulated a system that
heats both air and saline water and reported daily production of
43 kg/m2 of solar collector area per day. It was also concluded
that an open air system is more productive than a closed air
system. However, the system had no mechanism of controlling
saline water temperature. Also, the condenser was being cooled
by the inlet saline water stream similar to many other
configurations [11, 12, 13, 14], but this does not necessarily
imply minimum unit-cost of production. In this paper, we will
model and optimize a similar system that addresses these
issues.
3. SYSTEM MODEL
3.1 Overview The diagram in Fig. 1 shows the system from previous
work of the authors [7]. It is similar to many other systems in
literature but adds a controller to change water and air flow
rates based on temperature reading from state 3 of Fig. 1. This
effectively allows the water temperature to stay within a preset
limit (In this case, 60 °C).
This model was used as basis for design modifications
which are shown in Fig. 2:
Condensation performance is negatively affected when
automatic saline water flow rate is not sufficiently
high. Therefore, the condenser was disconnected from
the inlet water stream to operate on an independent
coolant supply with constant flow rate.
Saline water temperature is limited to 60 °C, air
capacity for water vapor will be less than 0.153 kg
vapor/kg air (if we assume air is saturated at state 5).
Therefore, a solar air heater was added before the
humidifier to allow for higher outlet temperatures and
humidity ratios.
The solar tank of the solar water heater is an expensive
component, and it will not be very useful in
conjunction with the solar air heater because the latter
can’t operate at night. The solar tank was, therefore,
removed.
Fig. 1: Schematic diagram for original system
Fig. 2: Schematic diagram for modified system
Page 4
4 Copyright © 2014 by ASME
The components of the modified system are as follows:
i. Pump: pumps saline water through the system
ii. Solar water heater: a flat plate solar collector for
heating inlet water (state 1).
iii. Controller: controls water flow rate according to
temperature reading at state 2.
iv. Solar air heater: packed bed solar air heater that heats
inlet air (state 3).
v. Humidifier: uses hot saline water (state 2) to humidify
heated ambient air (state 4) to produce hot humid air
(state 5) and brine (state B).
vi. Condenser: uses a coolant circuit (e.g. water) to cool
humid air (state 5) to produce distillate (state D)
and cool air (state 6).
The following assumptions were made before writing the
system model equations:
1. Heat losses to the atmosphere by the condenser,
humidifier and piping are neglected.
2. Kinetic and potential energy changes for air and water
are neglected.
3. Temperature rise due to pumping and air circulation
are neglected.
4. Thermodynamic properties of seawater are
approximated to that of pure water [5].
Air is saturated at condenser exit.
3.2 Global solar irradiance In order to accurately determine the energy gain from a
solar collector, the global aperture irradiance 𝐼 is calculated
from [15]:
𝐼 = 𝐼𝐴𝑀 {𝐼𝑏,𝑛 cos 𝜃𝑖 + [𝐼𝑑,ℎ (1+cos 𝛽
2) + 𝜌𝐼𝑡,ℎ (
1−cos𝛽
2)]} (1)
Ib,n, Id,h and It,h are calculated from hourly weather data for
the location and 𝜌 may be taken as 0.2 (typical reflectivity for a
concrete ground). 𝛽 is set equal to the latitude angle of the
location for maximum performance. cos 𝜃𝑖 is calculated from
the following equation [15]:
cos 𝜃𝑖 = cos 𝛽 (sin 𝛿 sin ∅ + cos 𝛿 cos ∅ cos𝜔)− cos 𝛿 sin𝜔 sin 𝛽 sin 𝛾+ sin 𝛽 cos 𝛾 (sin 𝛿 cos∅− cos 𝛿 cos𝜔 sin ∅)
(2)
For a collector facing south, 𝛾 = 𝜋, and ∅ depends on the
location. 𝛿 and 𝜔 are given by [15]:
sin 𝛿 = 0.39795 cos[0.98563(𝑁 − 173)] (3)
𝜔 = 15(𝑡𝑠 − 12) (𝑑𝑒𝑔𝑟𝑒𝑒𝑠) (4)
The equation of time (𝐸𝑂𝑇) is given by [15]:
𝐸𝑂𝑇 = 0.258 cos 𝑥 − 7.416 sin 𝑥 + 3.648 cos 2𝑥− 9.228 sin 2𝑥
(5)
Where the angle 𝑥 is defined as a function of the day
number, 𝑁 [15]:
𝑥 =360(𝑁 − 1)
365.242 (𝑑𝑒𝑔𝑟𝑒𝑒𝑠) (6)
𝑡𝑠 can then be calculated from local clock time (𝐿𝐶𝑇) using
[15]:
𝐿𝐶𝑇 = 𝑡𝑠 −𝐸𝑂𝑇
60+ 𝐿𝐶 + 𝐷 (7)
Where longitude correction (𝐿𝐶) is calculated from [15]:
𝐿𝐶 =
(𝑙𝑜𝑐𝑎𝑙 𝑙𝑜𝑛𝑔𝑖𝑡𝑢𝑑𝑒) − (𝑙𝑜𝑛𝑔𝑖𝑡𝑢𝑑𝑒 𝑜𝑓 𝑠𝑡𝑎𝑛𝑑𝑎𝑟𝑑
𝑡𝑖𝑚𝑒 𝑧𝑜𝑛𝑒 𝑚𝑒𝑟𝑖𝑑𝑖𝑎𝑛)
15
(8)
Where 𝐷 = 1 if daylight saving time is in effect, and 0
otherwise. The incidence angle modifier (𝐼𝐴𝑀) can be
calculated from [15]:
𝐼𝐴𝑀 = 1 − 𝑏 (1
cos 𝜃𝑖− 1) (9)
The coefficient b is a property of the solar collector. This
effectively allows the formulation to represent tubular as well
as flat plate collectors.
3.3 Solar Water Heater The energy gain from a flat-plate solar collector is
calculated via the Hottel-Whillier-Bliss equation [16]:
�̇�𝑢 = 𝐹𝑅𝐴𝐶[(𝜏𝛼)𝐼 − 𝑈𝐿(𝑇𝑓,𝑖 − 𝑇𝑎𝑚𝑏)] (10)
𝐹𝑅, (𝜏𝛼) and 𝑈𝐿 are all specific to the solar collector. 𝑇𝑓,𝑖 is
taken as inlet saline water temperature and 𝐼 is calculated as
shown in section 3.2. The energy balance for the saline water
stream across the solar collector is given by:
�̇�𝑢 = �̇�𝑤(𝑖2 − 𝑖1) (11)
For variable flow rate, the controller will constantly change
�̇�𝑤 to satisfy eqn. (10) and (11), i.e.:
�̇�𝑤 =𝐹𝑅𝐴𝐶[(𝜏𝛼)𝐼 − 𝑈𝐿(𝑇𝑓,𝑖 − 𝑇𝑎𝑚𝑏)]
(𝑖2 − 𝑖1) (12)
In the above equation, 𝑖2 is the enthalpy of water at 60 ⁰C
and atmospheric pressure.
3.4 Solar Air Heater The performance parameters of the solar collector, 𝐹𝑅,
(𝜏𝛼) and 𝑈𝐿 are originally affected by fluid flow rate. In case of
liquid heaters, the variation in these parameters can be
neglected. For air heaters, on the other hand, the variation is
significant [17]. Therefore, an experimental correlation for a
packed bed solar air heater [18] was used instead of eqn. (10).
If we define �̇�𝑠 as the volume flow rate of air passing
through the collector per unit area of the collector, the
correlation is:
Page 5
5 Copyright © 2014 by ASME
(𝑇4 − 𝑇𝑎𝑚𝑏)
𝐼= 𝐶1�̇�𝑠 + 𝐶2 (13)
In the above equation, 𝐼 is calculated as shown in section
3.2. Also, for packed bed air heaters, 𝐶1 = −2.111 and
𝐶2 = 0.07822 [18].
3.5 Condenser The energy and mass balances across the condenser are:
�̇�𝑐𝑜(𝑖𝑐𝑜,𝑜𝑢𝑡 − 𝑖𝑐𝑜,𝑖𝑛) = �̇�𝑎𝑖𝑟(𝑖5 − 𝑖6) − �̇�𝐷𝑖𝐷 (14)
�̇�𝑎𝑖𝑟𝜔5 = �̇�𝑎𝑖𝑟𝜔6 + �̇�𝐷 (15)
In addition, the greatest amount of heat the condenser is
able to transfer is:
�̇�𝑐𝑜𝑛𝑑 = 𝐴𝑐𝑜𝑛𝑑𝑈𝑐𝑜𝑛𝑑(𝑇𝑑𝑏,5 − 𝑇𝑐𝑜,𝑜𝑢𝑡) − (𝑇𝑑𝑏,6 − 𝑇𝑐𝑜,𝑖𝑛)
ln(𝑇𝑑𝑏,5 − 𝑇𝑐𝑜,𝑜𝑢𝑡)
(𝑇𝑑𝑏,6 − 𝑇𝑐𝑜,𝑖𝑛)
(16)
�̇�𝑐𝑜𝑛𝑑 = �̇�𝑐𝑜(𝑖𝑐𝑜,𝑜𝑢𝑡 − 𝑖𝑐𝑜,𝑖𝑛) (17)
In eqn. (16), 𝑈𝑐𝑜𝑛𝑑 is determined by evaluating its three
components: the heat transfer coefficient on the water side, for
the condenser tubes and on the humid air side. For a shell and
tube condenser, it can be calculated from [19]:
1
𝑈𝑐𝑜𝑛𝑑=1
ℎ𝑤+ (
𝑟𝑜𝑘𝑤𝑎𝑙𝑙
) ln (𝑟𝑜𝑟𝑖) + (
𝑟𝑜𝑟𝑖⁄
ℎ𝑎) (18)
Where ℎ𝑤 is the heat transfer coefficient on the water side.
It can be obtained from the Dittus-Boelter correlation [19]:
ℎ𝑤 = 0.023(𝑅𝑒𝑤)0.8(𝑃𝑟𝑤)
0.333(𝑘𝑤𝐷𝑖) (19)
A condensate layer will separate the humid air from the
condenser tubes. Therefore, ℎ𝑎 will have two components: one
for the condensate layer and another for the air stream [19]:
1
ℎ𝑎=1
ℎ1+ (
𝑍𝑔
ℎ𝑔) (20)
Where ℎ1 is the coefficient of the condensate layer which
is given by [19]:
ℎ1 = 0.943 [𝑘1
3𝜌1(𝜌1 − 𝜌𝑔)𝑔∆𝑖𝜗
𝜇1(𝑇𝑠 − 𝑇𝑤)�̇�𝑤
]
14⁄
(21)
The parameter 𝑍𝑔is given by [19]:
𝑍𝑔 = 𝐶𝑝 𝑑𝑇
𝑑𝑖 (22)
Where 𝑇(𝑖) for humid air is given by [4]:
𝑇 =𝑖 − 2501𝜔
1 + 1.805𝜔 (23)
Finally, ℎ𝑔 is given by [19]:
ℎ𝑔 = 𝑁𝑢𝑔 (𝑘𝑔
𝐷𝑡) (24)
Where the air Nusselt number 𝑁𝑢𝑔 is given by [19]:
𝑁𝑢𝑔 = 1.019 × 10−14(𝑅𝑒𝑎)0.4806(𝑃𝑟𝑎)
−95.69 (25)
In the above equations calculation of Reynolds and Prandtl
numbers for air and water require calculation of thermal
conductivity and dynamic viscosity. Correlations from [20]
were used for this purpose. Value of 𝑈𝑐𝑜𝑛𝑑 will vary along the
length of the condenser. Therefore, an average value is used
[19]:
𝑈𝑐𝑜𝑛𝑑 =1
2(𝑈𝑖𝑛𝑙𝑒𝑡 + 𝑈𝑜𝑢𝑡𝑙𝑒𝑡) (26)
3.6 Humidifier To determine the humidity content at state 5 accurately,
consider a volume element (Fig. 3) of the humidifier, and
assume that temperature and humidity gradients are only in the
vertical direction. The mass balance for the control volume of
the whole element becomes:
𝑑�̇�𝑤 = 𝑑�̇�𝑣 = �̇�𝑎𝑑𝜔 (27)
Fig. 3: volume element of humidifier The energy balance for the control volume of the water
side is:
𝑑𝑇𝑤 =ℎ𝑤𝑎
�̇�𝑤𝑐𝑤(𝑇𝑤 − 𝑇𝑖) 𝑑𝑦 (28)
The energy and mass balances for the control volume of
the air side are:
𝑑𝑇𝑎 =ℎ𝑎𝑎
�̇�𝑎(𝑐𝑎 + 𝜔𝑐𝑣)(𝑇𝑖 − 𝑇𝑎) 𝑑𝑦 (29)
𝑑𝜔 =𝐾𝑎
�̇�𝑎
(𝜔𝑖 − 𝜔) 𝑑𝑦 (30)
The interface between air and water is assumed to be
saturated air at temperature 𝑇𝑖 , that is, 𝜔𝑖 can be calculated
knowing 𝑇𝑖 . And the energy balance for the interface is:
Page 6
6 Copyright © 2014 by ASME
ℎ𝑤𝑎(𝑇𝑤 − 𝑇𝑖)𝑑𝑦 = ℎ𝑎𝑎(𝑇𝑖 − 𝑇𝑎) 𝑑𝑦+ ∆𝑖𝜗𝐾𝑎(𝜔𝑖 − 𝜔) 𝑑𝑦
(31)
For small 𝑑𝑦, equations (27) to (31) can be solved
progressively for 𝑑𝑇𝑤, 𝑑𝑇𝑎 and 𝑑𝜔 along the length of the
humidifier. The boundary conditions are:
𝑇𝑤(0) = 𝑇2, 𝑇𝑎(0) = 𝑇4, 𝜔(0) = 𝜔4 (32)
Heat and mass transfer coefficients are calculated from
[21]:
𝐾 =2.09 �̇�𝑎
0.11515 �̇�𝑤0.45
𝑎
(33)
ℎ𝑤 =5900 �̇�𝑎
0.5894 �̇�𝑤0.169
𝑎 (34)
ℎ𝑎 = 𝑐𝑎𝐾 (35)
Also, a type of humidifier packing was used that has
𝑎 = 157.7 𝑚2 𝑚3⁄ and a cross sectional area of 0.56 𝑚2. The
cost per unit height was obtained from the manufacturer of that
packing type [22].
3.7 Cost Model The following cost functions were obtained by
interpolating actual component prices listed on an online trade
portal [23]:
𝐶𝑐𝑜𝑛𝑑 = 50𝐴𝑐𝑜𝑛𝑑 (36)
𝐶𝑓𝑎𝑛 = 73.4�̇�𝑎 (37)
𝐶𝑝𝑢𝑚𝑝 = 25�̇�𝑤 + 25�̇�𝑐𝑜 (38)
𝐶𝑐𝑜𝑙,𝑤 = 120𝐴𝑐,𝑤 + 1100𝑉𝑡𝑎𝑛𝑘 (39)
The cost for the solar air heater was obtained from [18]:
𝐶𝑐𝑜𝑙,𝑎 = 120𝐴𝑐,𝑎 (40)
The cost for the humidifier was approximated from [22]:
𝐶ℎ𝑢𝑚 = 110𝐻ℎ𝑢𝑚 (41)
4. OPTIMIZATION
4.1 Heuristic Gradient Projection Heuristic Gradient Projection (HGP) is first described by
Metwalli [24, 25]. Abd El-Malek et al. [26] utilized HGP for
3D space frame weight minimization subject to bending stress
constraints [27]. It was later used by [28, 29, 30] to handle
stress constraints of some form because activation of stress
constraints translates to maximum material utilization, hence
the optimum solution.
The method aims at activating the constraints while
maximizing (or minimizing) the objective function value at the
same time. From a general search point, the procedure is to:
1. Find the objective function gradient (𝛻𝑓) at the current
point. To minimize, use the negative of the gradient
(−𝛻𝑓).
2. If there is a violated constraint, project the negative of
its gradient onto 𝛻𝑓:
𝑆 = (−∇𝑔 𝑇∇𝑓)∇𝑓 (42)
3. If there are no violated constraints, but there is an
inactive constraint, project its gradient onto 𝛻𝑓:
𝑆 = (∇𝑔 𝑇∇𝑓)∇𝑓 (43)
4. Project 𝑆 onto the constraint gradient tangents for
active constraints. If we have ‘r’ active constraints,
define:
𝐺𝑟 = [∇𝑔1 ∇𝑔2 … ∇𝑔𝑟] (44)
As per the gradient projection method, the final search
direction becomes [24, 31]:
𝑓𝑝 = {𝐼 − 𝐺𝑟[𝐺𝑟𝑇 𝐺𝑟]
−1𝐺𝑟𝑇}𝑆 (45)
If the active constraint is linear, step 4 will make it remain
active for the rest of the search. If it is convex, step 4 will cause
the search to violate the constraint, and it will be handled in the
next iteration by step 2. If it is concave, step 4 will cause the
search to deactivate the constraint, and it will be handled in the
next iteration by step 3.
4.2 Problem Formulation The objective is to minimize the unit-cost of produced
fresh water ($/𝑚3):
𝑓(𝑋) = (1 + 𝑐𝑝)(𝐶 +𝐶 +𝐶 +𝐶 +𝐶 , +𝐶 , )
𝑉 𝑡 (46)
Where
𝑋 = [𝐴𝑐,𝑎 𝐻ℎ𝑢𝑚 𝐴𝑐𝑜𝑛𝑑 �̇�𝑠 �̇�𝑐𝑜 𝐴𝑐,𝑤]𝑇 (47)
Because �̇�𝑤 is controlled based on 𝐴𝑐,𝑤, system capacity
is strongly related to the value of 𝐴𝑐,𝑤. Therefore, to be able to
have control on the size of the system generated, 𝐴𝑐,𝑤 is
assumed to be a constant input. All other design variables are
taken relative to the input 𝐴𝑐,𝑤 to generate adequate component
sizing.
Also, �̇�𝑠 directly defines air temperature for a given
weather condition. The fact that saline water is kept at 60 ⁰C
suggests existence of an air temperature that provides for
optimum system throughput regardless of all other parameters.
We can, therefore, specify �̇�𝑠 as an input to the algorithm whose
optimum value will be obtained later.
In addition, because the parameters have different units,
they are scaled by an appropriate factor to be in the same order
of magnitude.
𝑋 = [𝐴𝑐,𝑎𝐴𝑐,𝑤
100 𝐻ℎ𝑢𝑚𝐴𝑐,𝑤
𝐴𝑐𝑜𝑛𝑑𝐴𝑐,𝑤
10 �̇�𝑐𝑜
𝐴𝑐,𝑤]
𝑇
(48)
𝑋 = [𝑥1 𝑥2 𝑥3 𝑥4]𝑇 (49)
The cost functions, 𝐶𝑋, are closed-form of the design
variables. However, the yearly production, 𝑉𝑦𝑒𝑎𝑟 , requires
Page 7
7 Copyright © 2014 by ASME
simulation to be evaluated. A year-long simulation is a time-
consuming task. Therefore, 𝑉𝑦𝑒𝑎𝑟 is approximated as follows:
𝑉𝑦𝑒𝑎𝑟 = 𝑉𝑑𝑎𝑦 × 365 (50)
This approximation is sufficiently accurate for comparative
purposes because the system has no solar tank, which means
that performance in a day is independent of the performance in
the day before it. Also, a day with average weather conditions is
used to evaluate 𝑉𝑑𝑎𝑦 (day 100 in case of Hurghada), so as the
generated parameters will be near-optimum for the whole year.
After the optimum point is found, productivity and cost results
are calculated from a year-long simulation.
4.3 Performance Constraints To assess the nature of the problem, the method of steepest
descent was used to find the optimum. Table 1 shows the
optimization result from two different starting points. Each
starting point converges to a different solution which indicates
existence of multiple local optima.
Table 1: Optimization result from different starting points
Variable Start 1 Result 1 Start 2 Result 2
𝑥1 1 0.7 1.5 0.8
𝑥2 2 7.6 5 1
𝑥3 1 4.3 1 2.6
𝑥4 1 3.9 1.2 0.5
To find the global optimum, performance constraints were
used. These constraints depend on understanding the nature of
the system being optimized and identifying the bottlenecks
which may hinder its performance. The system at hand transfers
vapor, with air being the container that carries it from the
humidifier to the condenser. In a multi-stage dynamic
programming sense [24], optimality in each of the subsequent
stages necessarily implies optimality of the overall system. A
constraint is formulated for each stage as follows.
4.3.1 Vapor capacity constraint It is essential to maximize the capacity of air to carry
water vapor before the humidification stage (i.e. across the
solar air heater). In reference to Fig. 2, a constraint could
be formulated as:
𝑔1 = �̇�𝑤 − �̇�𝑣,4 ≤ 0 (51)
Where �̇�𝑤 = �̇�𝑤(𝐴𝑐,𝑤). Also, �̇�𝑣,4 represents the
maximum capacity of air to carry water vapor at state 4:
�̇�𝑣,4 = �̇�𝑎𝜔𝑠,4 = 𝜌𝑎𝐴𝑐,𝑎�̇�𝑠𝜔𝑠,4(𝑇4(�̇�𝑠)) (52)
Where 𝜔𝑠,4 is the absolute humidity for saturated air at
conditions of state 4. Since �̇�𝑠 and 𝐴𝑐,𝑤 are constant, 𝑔1 is a
linear function of 𝐴𝑐,𝑎 . Also, 𝑔1 requires simulation to be
evaluated. After that simulation is performed, daily average
values for both �̇�𝑤 and �̇�𝑣,4 are used. Since the global
optimum is assumed to activate 𝑔1, ∇𝑔 𝑇∇𝑓 will be
positive if 𝑔1 is inactive, and will be negative if 𝑔1 was
violated. This makes step 2 or 3 of the HGP procedure
reduce to:
S = +𝑘 [𝜕𝑓
𝜕𝑥1
𝜕𝑓
𝜕𝑥2
𝜕𝑓
𝜕𝑥3
𝜕𝑓
𝜕𝑥4]𝑇
(53)
Where 𝑘 is constant. However, for activating 𝑔1 =𝑔1(𝑥1), and since 𝑥1 is independent of all other 𝑥,
activation can be achieved by moving only in 𝑥1:
S = +𝑘[1 0 0 0]𝑇 (54)
To find out 𝑘 that activates 𝑔1, eqn. (51) is used to
quantify the slackness variable 𝑠 as follows:
�̇�𝑤 − �̇�𝑣,4 − 𝑠 = 0 (55)
For activation according to HGP, 𝑠 = 0, and it follows
that:
�̇�𝑤 = �̇�𝑣,4 = 𝜌𝑎𝐴𝑐,𝑎�̇�𝑠𝜔𝑠,4 (56)
Therefore:
𝐴𝑐,𝑎 =�̇�𝑤
𝜌𝑎�̇�𝑠𝜔𝑠,4 (57)
Dividing by 𝐴𝑐,𝑎𝑜𝑙𝑑 to obtain 𝐴𝑐,𝑎𝑛𝑒𝑤 which should
activate 𝑔1:
𝐴𝑐,𝑎𝑛𝑒𝑤𝐴𝑐,𝑎𝑜𝑙𝑑
=𝑥1𝑛𝑒𝑤𝑥1𝑜𝑙𝑑
=�̇�𝑤
�̇�𝑣,4
(58)
Since 𝑔1 is linear in terms of 𝐴𝑐,𝑎, it is activated in a single
evaluation of eqn. (58). This is the clout of HGP.
4.3.2 Humidity content constraint The constraint 𝑔1 ensured the air had enough capacity
to carry all the �̇�𝑤 available. The constraint to ensure the
humidifier is able to fully saturate the air passing through it
could be written as:
𝑔2 = 𝜔5 − 𝜔𝑠,5 ≤ 0 (59)
While:
𝜔𝑠,5 = 𝜔𝑠,5{𝑇4(�̇�𝑠), 𝑇3, �̇�𝑎(𝐴𝑐,𝑎, �̇�𝑠), �̇�𝑤(𝐴𝑐,𝑤), 𝑥2} (60)
and:
𝜔5 = 𝜔5(𝜔𝑠,5, 𝑥2) (61)
𝐴𝑐,𝑎 was determined by 𝑔1 (in the previous stage),
𝐴𝑐,𝑤; �̇�𝑠 are known. Therefore, 𝑔2 = 𝑔2(𝑥2). In a fashion
Page 8
8 Copyright © 2014 by ASME
similar to 𝑔1, the following relation can be used for
activating 𝑔2:
𝐻ℎ𝑢𝑚𝑛𝑒𝑤
𝐻ℎ𝑢𝑚𝑜𝑙𝑑
=𝑥2𝑛𝑒𝑤𝑥2𝑜𝑙𝑑
=𝜔𝑠,5
2
𝜔52
(62)
However, because 𝑔2 is not quadratic in terms of 𝑥2,
activation will require iterative application of eqn. (62). Each
evaluation will get closer to 𝑔2 until the separation distance is
small enough that 𝑔2(𝑥2) can be approximated quadratically.
4.3.3 Distillation mass rate constraint The final stage is actual distillation of fresh water in
the condenser. The condenser has to be large enough for
distillation of all the vapor content in the humid air (state
5):
𝑔3 = (�̇�𝐷 + �̇�𝑎𝜔𝑠,𝑇 , ) − �̇�𝑎𝜔5 ≤ 0 (63)
Where 𝜔𝑠,𝑇 , is absolute humidity at saturation at
inlet water temperature. 𝑔3 is a function of 𝑥3 and 𝑥4 and
the following relations could be used to achieve activation
(similar to 𝑔2):
𝑥3𝑛𝑒𝑤𝑥3𝑜𝑙𝑑
=𝑥4𝑛𝑒𝑤𝑥4𝑜𝑙𝑑
=(�̇�𝑎𝜔5)
2
(�̇�𝐷 + �̇�𝑎𝜔𝑠,𝑇 , )2 (64)
This relation scales 𝑥3 and 𝑥4 by the same factor. However,
their optimum values could be in different scales relative to
each other. This will be remedied as will be shown in the next
section. Fig. 4 shows the HGP activation procedure for 𝑔1, 𝑔2
and 𝑔3.
4.4 Constrained Search Now that all the performance constraints are active, a
search is performed to find the constrained optimum. This step
corrects the scale between 𝑥3 and 𝑥4. To proceed, ∇𝑓 and 𝐺𝑟
[32] are numerically calculated to obtain the direction 𝑓𝑝 from
eqn. (45) of the HGP procedure using 𝑆 = −∇𝑓. Small steps are
taken in 𝑓𝑝 until:
𝑓 starts to increase: 𝑓𝑝 is recalculated to step
again.
Any of the constraints deactivates: the constraint
is reactivated as before, then 𝑓𝑝 is recalculated to
step again.
The process is repeated until an 𝑓𝑝 calculation fails to take
a single minimizing step. Fig. 5 shows the stepping procedure.
4.5 Constraint Relaxation By activating the performance constraints and performing
the constrained minimization, the search is localized to the
closest vicinity of the global optimum. The assumption that the
global optimum exactly activates all of the performance
constraints can now be relaxed, and the global optimum can be
found by using the procedure illustrated in Fig. 5 by setting
𝑓𝑝 = 𝑆 directly, and by removing the check and the procedure
for constraint activation (i.e. deactivate all the constraints).
Fig. 4: Constraint activation procedure
Fig. 5: Stepping procedure
4.6 Optimum Specific Air Flow Rate For a given value of �̇�𝑠, optimum values for the rest of the
parameters can now be determined. This allows one-
dimensional optimization (e.g. Fibonacci search) on top of the
optimization procedure discussed before. That is, minimize:
𝑓2(�̇�𝑠) = 𝑓(�̇�𝑠 , [𝐴𝑐,𝑎 𝐻ℎ𝑢𝑚 𝐴𝑐𝑜𝑛𝑑 �̇�𝑐𝑜 𝐴𝑐,𝑤]𝑇) (65)
For different values of 𝐴𝑐,𝑤, �̇�𝑠 always converged to 0.011
𝑚3 𝑚2𝑠⁄ . This value was later used to obtain all results. This
process is an implementation of ridge path optimization. That
𝑆 = −𝛻𝑓
𝑋 = 𝑋 + 𝜖𝑓
𝑓 increased No
𝑓𝑝 from eqn. (45)
𝑋0 = 𝑋
𝑋 = 𝑋0 No
𝑋 = 𝑋 − 𝜖𝑓
𝑔
inactive?
No
Activation procedure
𝑥1𝑛𝑒𝑤 from eqn. (58)
𝑔1 active ?
𝑥2𝑛𝑒𝑤 from eqn. (62)
No
𝑔2 active ? No
Eqn. (64)
𝑔3 active ? No
Page 9
9 Copyright © 2014 by ASME
is, optimum values of parameters given a certain value of �̇�𝑠, are used to determine the optimum value of �̇�𝑠 [33, 34].
4.7 Genetic Algorithm The optimization problem has been solved by “Optimera”
[35] which is a GA optimization package for C# [36]. The
package was activated from the developed program with 0.9
crossover rate and 0.2 mutation rate. The number of generations
was 40, each containing 20 individuals. The termination
threshold was 0.001 $/𝑚3. Several runs were carried out to
make sure that the global optimum has been reached. Several
very close outputs were attained with the best being reported.
That is why the number of function evaluations was varying
between 1000 and 1600. This was monitored to compare the
GA performance to the HGP as indicated in the next results and
discussion section.
5. DEVELOPED PROGRAM A program was written by the authors in C# [36] to
simulate and optimize the system. It has a simulation as well as
optimization user interfaces.
The simulation interface allows the user to input custom
values for all the system parameters and then simulates the
system behavior in the chosen period of the year. After the
simulation completes, an output chart draws variation of the
different state measures of the system along the simulation
period. Accumulative system output (e.g. average daily
production, unit cost of produced fresh water, total production)
can also be displayed [37].
Finally, the optimization parameter values are obtained
from the input to the simulation interface. The optimization
interface gives log messages for each step the optimizer
performs. When the optimum point is found, the input fields are
updated to reflect its value so the yearly simulation can be run
right away [37].
The following libraries were written by the authors to serve
the purpose of the user interface:
Thermodynamic property calculator for water
which implements the IAPWS industrial
formulation [38].
Thermodynamic property calculator for humid air
which implements the ASHRAE formulation [4,
39].
Fibonacci optimizer.
Sun angle and irradiance calculator which
implements the equations in [15].
Numerical function gradient calculator [32].
The program makes use of the following libraries and
codes:
ZedGraph: charting library.
Math .NET: linear algebra package.
CSVReader: reads comma separated value files
(csv) for reading weather data.
Optiemera: a genetic algorithm optimization
library to optimize the system at hand.
6 RESULTS AND DISCUSSION The study considers a small desalination plant operating
near the city of Hurghada. Weather data for the location were
obtained from [40]. The study assumes a small-scale
desalination plant with required productivity between 5 and 10
m3 per day. Inflation and yearly maintenance cost factor (cp)
was assumed to be 10% and plant lifetime was assumed to be
30 years, which are typical values for the type of equipment
considered [41].
Table 2 shows a comparison between the optimum system
from this work and from previous work by the authors [7, 8].
For the new systems, the objective function value between
brackets is based on the previous cost function from [7]. The
unit-cost of produced fresh water relative to previous work was
reduced by about 90% if we use the new cost function and by
83% if we use the original cost function. The much smaller
condenser and the absence of a solar tank considerably reduced
the system’s overall cost as well as unit-cost of produced fresh
water.
Table 2 also lists the optimization and simulation result for
a system with larger input 𝐴𝑐,𝑤 (1000 𝑚2) and smaller input
𝐴𝑐,𝑤 (100 𝑚2). The value of 𝑓 is only slightly affected by the
system size.
The last two columns in Table 2 list the results of
optimization runs that use a larger (𝐴𝑐,𝑤 = 1000 𝑚2)
The condenser area in this system is much smaller than
previously reported [7] because the coolant flow rate is now
constant at a high value and also because the simulation
calculates 𝑈𝑐𝑜𝑛𝑑 during each iteration based on fluid properties
rather than using a single average value.
The values of 𝑔1, 𝑔2 and 𝑔3 at the optimum points are
always close to 0 which means that the final search converged
in their vicinity. This supports the concept of using performance
constraints to guide the search for the global optimum using
HGP. To further verify the optimization procedure, a genetic
algorithm (GA) optimization was performed for a system with
𝐴𝑐,𝑤 = 500 𝑚2. Table 2 shows no significant difference in unit-
cost between HGP and GA results. However, there’s a small
difference in specific fresh water production in favor of the
HGP-optimized system. In terms of function evaluations,
however, the HGP optimization converges in about 10% the
number of function evaluations required by the GA
optimization.
Page 10
10 Copyright © 2014 by ASME
Table 2: Optimization Results and Comparison
[8]
(GA)
[7]
(HGP)
This work
(HGP)
This work
(GA)
This work
(larger)
This work
(smaller)
𝐴𝑐,𝑤[𝑚2] 1434.7 1247.9 500 500 1000 100
𝐴𝑐,𝑎[𝑚2] - - 518.1 525.5 1036.2 103.3
𝐴𝑐𝑜𝑛𝑑[𝑚2] 494.0 972.7 922 303.5 586.9 77.6
𝐴ℎ𝑢𝑚[𝑚2] 1051.7 1072.9 19.262 1373.4 1963.5 662.1
𝐻ℎ𝑢𝑚[m] - - 17.8 15.6 22.3 7.5
𝑉𝑡𝑎𝑛𝑘[𝑚3] 69.3 158.8 - - - -
𝐴ℎ𝑒[𝑚2] - 8.9 - - - -
�̇�𝑠[𝑚3 𝑚2. 𝑠⁄ ] - - 0.011 0.011 0.011 0.011
�̇�𝑐𝑜[𝑘𝑔 s⁄ ] - - 2.67 100.9 215.8 9.9
𝑉𝐷[𝑚3] 6.9 10.3 26.2 27 54.3 5.1
𝑉𝑦𝑒𝑎𝑟[𝑚3] 2518.5 3773.2 9891.4 9836.4 19803.1 1872.5
�̇�𝑠[𝑘𝑔 𝑚2. 𝑑𝑎𝑦⁄ ] 4.8 8.3 26.4 24.3 26.7 25.2
Function evaluations 940 6204 99 1000-1600 101 103
Running time [minutes] - 90-120 2 10 2 2
𝑔1[𝑘𝑔/𝑠] - - 0.047 0.038 0.093 0.01
𝑔2[𝑘𝑔/𝑘𝑔] - - -0.003 -0.004 -0.003 -0.006
𝑔3[𝑘𝑔/𝑠] - - -0.002 -0.008 -0.01 -0.006
𝑓[$ 𝑚3⁄ ] 8.0 5.9 0.5 (1.2) 0.5 (1.2) 0.5 (1.1) 0.6 (1.8)
7. CONCLUSION This work modified a previously optimized system by
disconnecting the condenser from the saline water cycle and by
introducing a solar air heater to further augment the
humidification performance. Using further improved models
for the condenser and the humidifier obtained more accurate
productivity estimations. The Heuristic Gradient Projection
(HGP) optimization procedure reduced the number of function
evaluations to reach the global optimum compared to Genetic
Algorithms (GA’s). The developed procedure can simulate and
optimize the new model based on selected weather data file,
among various other inputs in any other prospective locations.
A case study, however, which assumes a desalination plant on
the Red Sea near the city of Hurghada is presented. The unit-
cost of produced fresh water for the new optimum system is
$0.5/𝑚3 compared to $5.9/𝑚3 for the HDH system from
previous work and less than the cost of other reported reverse
osmoses systems. The HGP optimization procedure proved to
be robust and efficient. It depends, however, on the total
understanding of the system and its constraints and their effect
on the design parameters and performance.
ACKNOWLEDGMENTS This research was supported by the U.S. Department of
Agriculture Grant # 58-3148-1-162 and Egypt Science and
Technology Development Fund, STDF Project # 3832.
REFERENCES [1] Kucera, J., 2010, Reverse Osmosis: Design, Processes,
and Applications for Engineers, Scrivener Publishing,
Salem, Massachusetts, pp. 15-19, Chap. 2.
[2] Ghermandi, A. and Messalem, R., 2009, "Solar-Driven
Desalination with Reverse Osmosis: The State of the Art,"
Desalination and Water Treatment, 7, pp. 285–296.
[3] Wikipedia, 2013, "Red Sea", http://en.wikipedia.org/
wiki/Red_Sea.
[4] Devres, Y. O., 1994, "Psychrometric Properties of Humid
Air: Calculation Procedures," Applied Energy, 48, pp. 1-
18.
[5] Sharqawy, M. H., Lienhard , J. H. and Zubair , S. M.,
2011, "On Thermal Performance of Seawater Cooling
Towers," J Eng Gas Turb Power,133.
[6] Narayan, G., Sharqawy, M., Summers, E., Lienhard, J.,
Zubair, S. and Antar, M., 2010, "The Potential of Solar-
Driven Humidification–Dehumidification Desalination for
Small-Scale Decentralized Water Production," Renewable
and Sustainable Energy Reviews, 14, pp. 1187-1201.
[7] Abd El-Aziz, K. M., El-Morsi, M., Hamza, K., Nassef, A.
O., Metwalli, S. M., and Saitou, K., 2013, "Optimum
Solar-powered HDH Desalination System for Semi-
isolated Communities," ASME IDETC/CIE-12876,
Portland, Oregon.
[8] El-Morsi, M., Hamza, K., Nassef, A. O., Metwalli, S. M.,
and Saitou, K., 2012, "Integrated Optimization of a Solar-
Powered Humidification Dehumidification Desalination
System for Small Communities," ASME IDETC/CIE-
70783, Chicago, Ill.
[9] Farsad. S. and Behzadmehr, A., 2011, "Analysis of a Solar
Desalination Unit with Humidification-Dehumidification
Cycle Using Doe Method," Desalination, 278, pp. 70-76.
[10] Orfi, J., Galanis, N. and Laplante, M., 2007 "Air
Page 11
11 Copyright © 2014 by ASME
Humidification–Dehumidification for a Water Desalination
System Using Solar Energy," Desalination, 203, pp. 471-
481.
[11] Yuan , G. and Zhang, H., 2007, "Mathematical Modeling
of a Closed Circulation Solar Desalination," Desalination,
205, pp. 156-162.
[12] Al-Hallaj, S., Farid, M. M. and Tamimi, A., 1998, "Solar
Desalination with a Humidification-Dehumidification
Cycle: Performance of the Unit," Desalination, 120, pp.
273-280.
[13] Zamen, M., Amidpour, M. and Soufari, S. M., 2009, "Cost
Optimization of a Solar Humidification–Dehumidification
Desalination Unit Using Mathematical Programming,"
Desalination, 239, pp. 92-99.
[14] Farid, M. M., Parekh, S., Selman, J. R. and Al-Hallag, S.,
2002, "Solar Desalination with a Humidification-
Dehumidification Cycle: Mathematical Modeling of the
Unit," Desalination, 151, pp. 153-164.
[15] Stine, W. and Geyer, M., 2001, Power From The Sun,
Chap. 3.
[16] Hottel, H. C. and Whillier, A., 1958, "Evaluation of Flat-
Plate Collector Performance," Transcripts of the
Conference on the Use of Solar Energy, Arizona.
[17] Duffie, J. A. and Beckman, W., 1974, Solar Engineering of
Thermal Processes, Second Edition, Wiley-Interscience,
pp. 250-296, Chap. 6.
[18] Gill, R., Singh, S. and Singh, P. P., 2012, "Low Cost Solar
Air Heater," Energy Conversion and Management, 57, pp.
131-142.
[19] Mhergoo, M. and Amidpour, M., 2011, "Derivation of
Optimal Geometry of a Multi-Effect Humidification-
Dehumidification Desalination Unit: A Constructal
Design," Desalination, 281, pp. 234-242.
[20] Al-Sahali, M. and Ettouney , M. H., 2008, "Humidification
Dehumidification Desalination Process: Design and
Performance Evaluation," Chemical Engineering Journal,
143, pp. 257-264.
[21] Amer, E., Kotb, H., Mostafa, G. and El-Ghalban, A., 2009,
"Theoretical and Experimental Investigation of
Humidification–Dehumidification Desalination Unit,"
Desalination, 249, pp. 949-959.
[22] Cooling Tower Depot, Inc., 2013, "Cooling Tower Depot
Parts Warehouse: CTD-19MA10,"
http://www.coolingtowerdepot.com/content/parts/product-
detail/1608.
[23] Alibaba.com Hong Kong Limited, 2013,
http://www.alibaba.com/.
[24] Metwalli, S. M., 2002, Optimum Design: Advanced
Lecture Notes, Cairo University.
[25] Metwalli, S. M., 2004, "Synthesis Paradigm in Computer
Aided Design and Optimization of Mechanical
Components and Systems," Current Advances in
Mechanical Design and Production, 3, pp. 3-11.
[26] Abd El Malek, M. R., Senousy, M. S., Hegazi, H. A. and
Metwalli, S. M., 2005, "Heuristic Gradient Projection for
3d Space Frame Optimization,", ASME IDETC/CIE85348,
California.
[27] Metwalli, S. M., 1999, 3D Frame: Computer Aided Design
of Space Frames Using Beam Elements, Privately
Developed Program.
[28] Abbas, M. H. and Metwalli, S. M., 2011,
"Elastohydrodynamic Ball Bearing Optimization Using
Genetic Algorithm and Heuristic Gradient Projection,"
ASME IDETC/CIE-47624, Washington DC.
[29] Hanafy, M. M. A. and Metwalli, S. M., 2009, "Hybrid
General Heuristic Gradient Projection for Frame
Optimization of Micro and Macro Applications," ASME
IDETC/CIE-87012, , San Diego.
[30] Senousy, M., Hegazi, H. and Metwalli, S. M., 2005,
"Fuzzy Heuristic Gradient Projection for Frame Topology
Optimization,", ASME IDETC/CIE-85353, Long Beach,
California.
[31] Rao, S. S., 1996, Engineering Optimization: Theory and
Practice, Third Edition, Wiley-Interscience, pp. 455-459,
Chap. 7.
[32] Chapra, S. and Canale, R., 2001, Numerical Methods for
Engineers, McGraw-Hill, pp. 632-635, Chap. 23.
[33] Metwalli, S. M. and Mayne, R. W., 1977, "New
Optimization Techniques," ASME DAC, 77-DAC-9.
[34] Elzoghby, A. A., Metwalli, S. M. and Shawki, G. S. A.,
1980, "Linearized Ridge-Path Method for Function
Minimization," Journal of Optimization Theory and
Application, 30(2), pp. 161-179.
[35] Egan , C., 2013, "Optimera – a multithreaded optimization
library in C#," http://cosmobomb.com/wp/?page_id=656 .
[36] Microsoft Corp., 2010, "Microsoft Visual C# Express
(2010)," http://www.microsoft.com/visualstudio/eng/
downloads.
[37] Abd El-Aziz, K. M., 2014, "Optimum Design of Solar
Water Desalination Systems by Humidification and
Dehumidification," M.S Thesis, Cairo University.
[38] IAPWS, 1997, "Release on the IAPWS Industrial
Formulation 1997."
[39] Parsons, R. A., 1993, ASHRAE Handbook—Fundamentals,
ASHRAE, Chap. 6.
[40] US Department of Energy, 2011, "EnergyPlus Simulation
Software, Weather Data."
[41] National Renewable Energy Laboratory, 2011, "System
Advisor Model."