Applied Energy 266 (2020) 114864

Contents lists available at ScienceDirect

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

Modeling and performance analysis of high-efficiency thermally-localized T

multistage solar stills
⁎
Lenan Zhanga,1, Zhenyuan Xua,b,1, Bikram Bhatiaa, Bangjun Lib, Lin Zhaoa, Evelyn N. Wanga,
a
Department of Mechanical Engineering, Massachusetts Institute of Technology, Cambridge, MA 02139, USA
b
Institute of Refrigeration and Cryogenics, Shanghai Jiao Tong University, Shanghai 200240, China

H I GH L IG H T S

• Modeling framework of thermally-localized multistage solar stills (TMSS) is developed.

• Heat and mass transfer in the TMSS are analyzed in detail.
• Optimization strategies for the TMSS are presented.
• Ultrahigh solar-thermal cumulative efficiency over 700% is predicted.

A R T I C LE I N FO A B S T R A C T

Keywords: Seawater desalination is a promising solution to global water shortage. Commercially available desalination
Solar still technologies typically require large installations which can be impractical for developing regions without well-
Multi-stage developed infrastructure. Passive solar desalination promises a viable solution, but can suffer from low effi-
Heat localization ciencies. Recent advances in the thermal design of small-scale solar desalination systems have demonstrated the
Highly efficient
potential for high-efficiency solar desalination in portable systems. In particular, the concept of a thermally-
Desalination
Modeling
localized multistage solar still (TMSS) – which combines localized heating of a capillary flow with condensation
heat recycling – has been experimentally demonstrated very recently and achieved over 100% solar-thermal
cumulative efficiency. However, a fundamental understanding of the heat and mass transfer, efficiency limits
and optimization strategies are missing in the literature. This work presents a modeling framework that eval-
uates the thermal and vapor transport in a model TMSS system with varying device configuration and predicts its
solar desalination efficiency. We demonstrate that an ultrahigh solar-thermal cumulative efficiency, many times
higher than that of conventional solar stills, can be achieved by optimizing the number of stages and device
geometry. Specifically, our modeling shows that the efficiency of the capillary fed TMSS is limited by the dis-
sipation of thermal energy to the environment during condensation and significant gains in efficiency can be
achieved by minimizing this loss. This work provides insights into physical processes critical for thermally-
localized portable solar distillation which could lead to high-performance desalination or water purification
technologies.

1. Introduction These include reverse osmosis [4], multistage flash evaporation,

membrane distillation, multistage distillation [5], and humidification-
One third of the global population faces severe fresh water shortage dehumidification desalination [6]. However, these technologies require
[1]. Meanwhile, 70% of the total surface area of Earth is covered by centralized units and well-developed infrastructure, which are there-
oceans – an abundant natural resource for clean water. Therefore, ex- fore not suited for remote places experiencing severe fresh water
tracting clean water from the sea offers a promising solution to global shortage [7]. Small-scale passive desalination systems that can be de-
water scarcity [2,3]. Large-scale water desalination systems, driven by ployed easily and at a low cost in such areas could be a more practical
thermal or mechanical energy, are already commercially available. proposition to meet local fresh water demand and improve quality of

⁎
Corresponding author.
E-mail address: [email protected] (E.N. Wang).
1
Equal contribution to this work.

https://doi.org/10.1016/j.apenergy.2020.114864
Received 15 November 2019; Received in revised form 7 March 2020; Accepted 15 March 2020
0306-2619/ © 2020 Elsevier Ltd. All rights reserved.
L. Zhang, et al. Applied Energy 266 (2020) 114864

Fig. 1. Schematic showing the concept of the capillary fed thermally-localized multistage solar still (TMSS). Brine is driven into each stage due to capillary pressure
and evaporates at the capillary wick. Vapor diffuses to the right and condenses on the back wall (right) of each unit stage, where the purified water can be collected.
The heat released from condensation is then reused by the following stages. The condensation heat not utilized at the end of all stages is finally returned to the bulk
liquid (brine).

life. consumed for brine and fresh water pumping, which restricted its ap-
Solar stills, which convert sunlight into thermal energy and purify plications [24]. Very recently, the concept of capillary fed high-effi-
brine through distillation [8], are among the most reliable portable ciency thermally-localized multistage solar still (TMSS) was reported,
passive desalination devices that utilize the abundant solar energy re- which combines the advantages of condensation heat recycling and
source and have been used for centuries [9,10]. Conventional solar stills localized heating, and demonstrated a high efficiency of ~135% [27].
rely on solar absorption and heating a bulk liquid to induce evaporation While significant theoretical progress was made to enhance the un-
[11,12], which is then used to collect desalinated water through con- derstanding of the multistage desalination in previous studies
densation [13,14]. However, the solar-thermal cumulative efficiency, [25,28,29], a detailed mechanistic model and a specific design principle
defined as the total vaporization enthalpy of purified water over the for the capillary fed TMSS system are still missing in the literature.
total energy from sunlight input, of conventional solar stills is low This work provides a comprehensive theoretical framework that
(~35%) since a large volume of cold bulk liquid needs to be heated. The models thermal and vapor transport in a thermally-localized multistage
large optical and thermal losses in these systems also contribute to the solar still configuration to provide insight into the fundamental
low efficiency which often necessitates high optical concentration thermal-fluidic processes and offer specific design guidelines. Several
(10 ~ 1000×) and can be cost prohibitive [8,11]. Significantly higher- key mechanisms including the effect of interfacial heating, capillary fed
efficiency [15] and lower-cost solar vapor generation [16] can be and side wall heat loss were considered in the present model. We model
achieved using the recently-demonstrated concept of thermal localiza- representative single-stage and multistage solar still devices, and in-
tion [17] which localizes heat to the liquid-vapor interface [18] where vestigate the effect of the number of stages, thermal insulation and
the evaporation occurs and enables efficient operation even at dilute geometrical configuration on solar still efficiency. Our model predic-
solar flux (~1000 W/m2, 1 sun) [19,20]. Past works on thermally-lo- tions indicate that, using an optimized device configuration, solar-
calized solar vapor generation typically utilized floating porous struc- thermal cumulative efficiency many times higher than that of the
tures, such as double-layered structures (exfoliated graphite layer conventional solar still could be achieved. Such high efficiency, in
supported by a carbon foam) [17], carbon black paper [21], porous conjunction with a greater flexibility in the choice of materials com-
material made from natural wood [22,23] and hierarchically nanos- pared to some past works on thermally-localized solar stills, could lead
tructured gels [15], which have high solar absorption, low thermal to an affordable portable desalination solution to address the global
conductivity to avoid conductive heat loss to the bulk water and high fresh water shortage.
wickability to transport liquid from bulk to the heated interface [17].
Consequently, these thermally-localized solar vapor generators were 2. Concept of the thermal-localized multistage solar still
able to achieve efficiencies as high as ~95% under 1 sun [15,18], which
is promising for high-efficiency solar stills. Recent work, however, has We first elucidate the concept of the capillary fed TMSS device
shown that even higher efficiency and lower cost might be possible. shown schematically in Fig. 1. The device consists of several stages
Despite the application of thermally-localized technologies, the aligned in series. The front wall facing the sun (left) serves as the solar
demonstrated solar-thermal cumulative efficiency is typically limited to absorbing layer. The internal walls separating adjacent unit stages act
below 100% primarily because of the heat loss to the environment as the condensation heat recycling layers. The back wall (far right) has
through condensation. Efficiency higher than 100% is thus possible if high thermal conductivity and is partially immersed into the bulk
the heat released from condensation can be harvested and reused for water, which is used for maintaining a low vapor pressure at the last
vapor generation. Although some past works took advantage of con- stage. Otherwise, if the back wall is cooled by the ambient air only,
densation heat recycling and developed large-scale multistage solar still insufficient heat dissipation will lead to an elevated temperature of the
systems [24,25], the sub-optimal design caused large thermal losses and entire system and a low cumulative efficiency. The side walls are lined
restricted the experimental efficiency below 91% even when the with thermal insulation to reduce total heat loss. During operation,
nighttime yield was considered [26]. Additionally, electricity was capillarity draws the brine into the capillary wick from the bulk liquid,

2
L. Zhang, et al. Applied Energy 266 (2020) 114864

Fig. 2. (a) Schematic of the single-stage solar still device with thermal localization and (b) the corresponding iterative algorithm used to evaluate heat and mass
transfer. The model determines the front wall temperature and corresponding vapor flux and accounts for heat loss due to conduction, convection and radiation on
the front and side walls.

solar radiation is converted to heat and transferred to the capillary wick losses evaluated using this mathematical model enabled us to quantify
containing the brine, resulting in vapor generation. As the solar thermal the solar vapor generation efficiency and explore strategies to maximize
energy is localized where the evaporation occurs, energy is not lost performance.
heating the bulk liquid, thereby minimizing losses. The vapor diffuses In order to simplify the analysis, we model a representative capillary
to the right and condenses on the internal wall. Latent heat released fed TMSS device (schematically shown in Fig. 1) with the following
from condensation is transferred to the next capillary wick which in- assumptions: (1) The device operates at a steady state. (2) Water
duces vapor generation in the second stage. At each stage, heat gen- transport provided by the capillary wick is sufficiently high and no
erated from the condensation is harvested to evaporate the brine in the dryout occurs since the input solar flux is dilute (~1 sun). (3) The in-
subsequent stage, which enables high-efficiency solar vapor generation. ternal walls are thin, impermeable to vapor and have high thermal
In addition to the thermal localization and condensation heat recycling, conductivity. Since the thermal resistance due to the air gap between
the capillary fed TMSS has two other potential advantages: two stages is typically more than two orders of magnitude larger than
the total thermal resistance of the internal wall, the temperature dif-
(1) In conventional thermally-localized solar evaporators, both solar ference across the wall thickness is negligible and therefore, the in-
absorption and evaporation occur at the front side, so the solar plane temperature gradients within each wall can be neglected (see
absorber should also act as an evaporator and thermal insulator. For Appendix A1 for assessments). (4) There is no convective heat transfer
this reason, this solar absorber should have high wickability and inside each stage as the size of each stage is typically small (~1 cm).
low thermal conductivity. However, in TMSS, the solar absorber Radiative heat transfer inside each stage is negligible since the tem-
and the evaporator (capillary wick) are located on different sides of perature difference between two neighboring stages is small (see
the front wall, which relaxes the design constraints and provides Appendix A1 for assessments). (5) There is no convective mass trans-
more flexibility in material choices. For example, the commonly port within a stage because of the dilute solar flux and small size of each
used commercial selective solar absorbers can be incorporated for unit cell. Vapor transport relies on diffusion driven by a concentration
TMSS application. (vapor pressure) gradient, which can be determined by pure water
(2) As the bulk brine contacts the solar still device only through the properties due to the relatively low salinity (35 g/l) of the typical
thin capillary wick, the parasitic heat loss from the solar energy seawater (see Appendix A1 for assessments). (6) One-dimensional
absorption layer to the bulk liquid can be minimal leading to higher transport of heat and mass is dominant due to the small side wall losses
efficiency. and large aspect ratio (the device dimensions perpendicular to vapor
flow are large in comparison to the gap between the stages). (7) The
3. Theoretical model solar still device is sealed and there is no vapor or liquid loss.

This section outlines the modeling framework used to elucidate the 3.1. Mathematical modeling for the thermally-localized single-stage solar
working principle and predict the performance of a single-stage as well still
as a multi-stage solar still with thermal-localization. We considered the
coupling of mass and energy transport inside each stage, and modeled We first consider the single-stage solar still device with thermal-
the corresponding phase change and resulting vapor diffusion. Heat localization schematically shown in Fig. 2. The device is a cuboid with
losses through conduction, convection and radiation were also ac- front wall area dimensions a × a and thickness b. The incident sunlight
counted for in the model. The temperature, vapor mass flux and heat (absorbed heat flux qsun ''
) is absorbed by the front absorber and

3
L. Zhang, et al. Applied Energy 266 (2020) 114864

converted to heat which is then input into the device. A fraction of the '' dc cf − c b
Jevap = −Da ≈ Da
solar energy incident at this front surface is lost due to radiation and dx b (8)
'' ''
convection, given by qrad + qconv . The heat input into the device causes where cf = cf (Tf ) and cb = cb (Tb) are temperature dependent.
the liquid water in the capillary wick to evaporate. The moist air on the Similarly, an energy balance on the back wall is given by,
right-side of the thin wick is in a saturated state with a water vapor
'' '' ''
concentration of cf which is determined by the absorber temperature Tf . qb, in = a2 (qevap + qcond ) − 4abqside = qb, out (9)
The vapor then condenses on the left-side surface of the back wall
(Fig. 2). The vapor concentration cb at this internal surface corresponds where qb, in and qb, out represent the total heat input and output at the
''
to the saturation temperature Tb which depends on the heat dissipation back wall, respectively. The heat leaking through the side walls qside can
through the back wall. Mass diffusion (mass flux Jevap ''
) is thus driven by be estimated as,
this concentration gradient. Meanwhile, heat loss through conduction
'' '' T¯ − T∞
qcond occurs inside the device due to the temperature difference between qside =
Rside (10)
the front and back walls, and heat can also leak through the side wall
Tf + Tb
because of imperfect thermal insulation. where T̄ is the average side wall temperature and Rside is the total
2
In this problem, the incident solar flux qsun''
and the back-wall tem- thermal resistance per unit area of the side wall insulation and the
perature Tb are known. We assume Tb is equal to the temperature of the ambient air,
bulk water due to the high thermal conductivity of the condenser (as-
sumption 3). Apart from these two parameters, the geometry of the 1 t
Rside = +
device (a and b), ambient air temperature T∞, and the thickness of the ha k (11)
insulation layer t are also known. We assume all of the transport and The net heat flux ''
qnet associated with condensation on the back wall
thermophysical properties (i.e., diffusivity of vapor in air Da , thermal is thus given by,
conductivity of air ka and insulation layer k , emissivity of the solar
''
absorber ε , and convective heat transfer coefficient of air ha ) are con- ''
qb, out − a2qcond
qnet =
stant as the device operates within a relatively narrow temperature a2 (12)
range. For convenience, we define a set P containing all of these input
When the solar still device is working, vapor can condense on both
parameters as,
the back wall and the side walls. Although the liquid water on both the
''
P = {qsun , Tb, T∞, Da , ha, t , k , ka, a, b, ε }. (1) back and side walls is desalinated and can be collected, only the heat
''
'' '' ''
flux qnet on the back wall can be potentially reused. Therefore, to better
The temperature of the front wall Tf , heat losses qcond (qrad , qconv , evaluate the performance of vapor generation, we defined two effi-
''
and qside ), overall efficiency η, vapor flux Jevap and corresponding va-
''
ciencies. The total efficiency ηtot which accounts for all of the condensed
porization heat flux qevap
''
are unknown and need to be solved for. We liquid water from the input solar flux qsun
''
which is given by,
define the output set O containing all unknown quantities,
'' ''
a2qnet + 4abqside
''
O = {qnet ''
, qside ''
, qrad ''
, qconv ''
, qcond ''
, Jevap , Tf , cf , η}. ηtot =
(2) ''
a2qsun (13)
Heat and mass transfer are coupled within the device. Specifically,
In addition, we define the recyclable efficiency ηrec to evaluate the
vapor concentration on the capillary wick cf is a function of the front
ability of the device to recycle heat from condensation,
wall temperature Tf , which is determined by heat transfer within the
''
device. In steady state, the energy balance at the front wall can be qnet
expressed as, ηrec = ''
qsun (14)
'' '' '' '' ''
qf , in = a2qsun = qf , out = a2 (qrad + qconv + qcond + qevap ) (3) ''
where only contributes to ηrec . If the heat loss through the side wall
qnet
'' ''
where qf , in and qf , out represents the total heat input and output at the qside is large, ηrec will decrease as qside cannot be reused by the next stage,
''
front wall respectively. The radiation heat loss qrad from the solar ab- but ηtot is relatively unchanged as the side wall contribution is ac-
sorber is given by, counted in the total efficiency.
In order to solve the heat and mass transfer model expressed in Eqs.
''
qrad = εσ (T f4 − T∞4 ) (4) (1)–(13), we developed an iterative solver taking set P as the input and
computing set O. Fig. 2(b) shows the flow chart of the algorithm. As
where σ = 5.67 × 10−8 Wm−2 K−4
is the Stefan-Boltzmann constant. both heat loss and vapor flux depend on the front wall temperature Tf ,
The convection heat loss on the front wall is expressed as, which is unknown, we initially guess Tf in the initialization step. To
''
qconv = ha (Tf − T∞). (5) pick a reasonable initial value and accelerate convergence, we assumed
'' ''
qevap = qsun , and evaluated the resulting Tf using Eqs. (7) and (8). We
We assume the conduction mainly occurs along the x-direction as then calculate the heat loss terms qrad'' ''
, qconv ''
, qcond ''
and qside based on the
the temperature gradient in x-direction is much larger than that in y- values of Tf and Tb (Eqs. (4), (5), (6) and (10) respectively), and update
and z-directions (assumption 6). For this reason, conduction heat loss is ''
qevap using Eq. (7). An updated Tf can then be recalculated using Eqs. (7)
estimated by one-dimensional Fourier’s law,
and (8). We then iterate until Tf converges (i.e. |Tf − T| < δ , where T is
'' dT Tf − Tb the front wall temperature from last iteration and δ = 0.01o C is the
qcond = −k ≈k .
dx b (6) tolerance used in this work).
The heat carried by vaporization ''
qevap is related to the vapor flux
'' 3.2. Mathematical modeling for the thermally-localized multistage solar still
Jevap through the latent heat hfg and molecular weight of water M,
'' '' We next model the heat and mass transfer within a capillary fed
qevap = Mhfg Jevap . (7)
TMSS based on the methodology described above for a single-stage
In the low flux regime (qevap
''
< 105 W/m2 ) [30,31], vapor transport device. We assume an n-stage device shown in Fig. 3. The solar absorber
relies on diffusion which is governed by Fick’s law (assumptions 5 and ''
with qsun input is located at the front and the condenser at a constant
6), temperature Tb is at the back. The device is divided into several unit

4
L. Zhang, et al. Applied Energy 266 (2020) 114864

Fig. 3. (a) Schematic of the multistage solar still device and (b) the corresponding iterative algorithm to evaluate heat and mass transfer. The model solves for the
coupling of heat and mass transfer within each stage as well as the coupling between adjacent stages. The wall temperature and corresponding fluxes were solved
iteratively from stage n to stage 1.

''
stages (each with thickness b), separated by internal walls with negli- heat loss through the side wall of ith stage. qside , i is calculated by plug-
gible thickness. For each of the internal walls, vapor condenses on the Tf , i + Tb, i
ging the average side wall temperature of ith stage T̄i = 2
into Eq.
left side and liquid water on the capillary wick evaporates on the right (10). For the first stage at the front, radiation and convection heat losses
side. To predict the overall performance, the heat and mass transfer ''
(qrad ''
and qconv ) from the solar absorber are also included into the energy
inside each unit stage is solved. The energy balance within each stage ''
balance equation. The outgoing heat flux qout th
, i from the back of i stage
can be expressed as,
is given by,
'' '' '' '' ''
⎧ a2qin, i = a2 (qrad + qconv + qout , i ) + 4abqside, i , (i = 1) ''
a2qout 2 '' '' 2 '' '' ''
, i = a (qcond, i + qnet , i ) = a (qcond, i + qevap, i ) − 4abqside, i . (16)
⎨ a2q '' = a2q '' + 4abq '' , (i > 1)
⎩ in, i out , i side, i (15) ''
where qcond , i is the conduction heat loss along x-direction determined by
where qin'' , i is the incoming heat flux at the front wall of the ith stage, qout ''
,i
the front and back wall temperatures (i.e., Tf , i and Tb, i respectively), and
''
qnet , i is the heat flux released by condensation. Similar to Eq. (8), the
''
is the outgoing heat flux at the back wall of the ith stage, and qside, i is the

5
L. Zhang, et al. Applied Energy 266 (2020) 114864

vapor flux Jevap

''
, i is calculated by the one-dimensional (1D) Fick’s law, addition to validation with recent past experiments, our proposed
cf , i − cb, i model shows how to further improve the performance of TMSS with
''
Jevap , i = Da proper optimization, which is discussed in Section 4.
b (17)
where cf , i = cf , i (Tf , i ) and cb, i = cb, i (Tb, i ) are the vapor concentrations at 4. Results and discussion
the front and back of the ith stage respectively, and the evaporative heat
flux qevap
''
, i is given by Eq. (7). 4.1. Device performance and optimization of the single-stage solar still
Two levels of coupling exist in this capillary fed TMSS model: intra-
coupling between the conjugate heat and mass transfer within each We first analyzed the effect of device configuration on the perfor-
stage similar to the single-stage model, and inter-coupling due to in- mance of the single-stage solar still. To show the effect of device geo-
teraction between neighboring stages leading to the following con- metry on the performance, we varied the unit stage thickness b from 0
tinuity relations, to 6 cm and calculated the corresponding vapor flux (Fig. 5(a)), wall
'' '' temperature (Fig. 5(b)), different heat loss contributions (Fig. 5(c)), and
qout , i − 1 = qin, i , andTb, i − 1 = Tf , i. (18)
the overall efficiency (Fig. 5(d)). The simulation parameters used for
We developed an iterative algorithm to deal with the inter-coupling the modeling are listed in Table 1.
effect between adjacent stages. Specifically, the energy and mass con- As shown in Fig. 5(a), the vapor flux at the front wall increases first
servation equations are solved stage-by-stage from right to left (see and then decreases with increasing unit stage thickness. The relatively
Fig. 3). The unit stage solver Φ is formulated according to the model low vapor flux when b is small is due to the low front-wall temperature
''
presented in Section 3.1, which takes the outgoing heat flux qout , i and
(Tf < 45 °C when b < 0.5 cm, see Fig. 5(b)) and large conductive heat
back-wall temperature Tb, i as the input parameters and calculates the loss (see Fig. 5(c)) due to the small thermal resistance. When
incoming heat flux qin'' , i and front-wall temperature Tf , i of the ith stage. b > 0.5 cm, the diffusive transport resistance, which scales with b/ Da ,
For convenience, we defined the input and output subsets as becomes dominant and vapor flux decreases as b increases (Fig. 5(a)).
'' '' Consequently, the front-wall temperature increases to supply the higher
Pi = {qout i
, i , Tb, i} and O = {qin, i , Tf , i} , respectively. Combining with the
inter-coupling relations shown in Eq. (18), we obtain the following vapor concentration (Fig. 5(b)). In order to avoid complexity associated
recursive relationship, with boiling in the capillary wick and significant internal convection
(which invalidates assumption 4), modeling was limited to unit stage
Pi − 1 = Oi = Φ(Pi). (19)
thickness b < 5.5 cm when the wall temperature approaches the sa-
Eq. (19) is then solved recursively starting at i = n , where Tb, n and turation temperature of water (100 °C at 1 atm). The different heat loss
'' mechanisms also vary significantly with b. As shown in Fig. 5(c), vapor
, n are required. As shown in the flowchart Fig. 3(b), we assigned qsun
''
qout
'' heat conduction loss is the dominant mechanism when b < 0.7 cm,
, n as the initial value for the first iteration.
to qout Oi of each stage is then
'' '' '' because the thermal resistance (scaled by b/ ka ) is small in this regime.
computed from Eq. (19). To make qin,1 converge to qsun , qout , n is rescaled When 0.7 cm < b < 3 cm, Tf increases and convection and radiation
after iteration as,
at the front wall contribute most to the total heat loss. When b > 3 cm,
''
qout '' heat loss from the side wall becomes the most significant loss me-
, n = γqout , n (20)
chanism due to the increase in total side wall area. Since the conductive
''
where the scaling factor γ = qsun / qin'' ,1.
This rescaled ''
is then used as
qout ,n heat loss decreases with increasing b while the front and side wall heat
the input for the next iteration. Similar to the single stage device, we losses increase with increasing b, a minimum in the total heat loss oc-
can define and calculate the total and recyclable efficiency using Eqs. curs at b ≈ 0.5 cm (see Fig. 5(c)). As shown in Fig. 5(d), the optimized
(21) and (22), respectively, overall efficiency reaches its maximum when the total heat loss is
'' '' minimum. However, there is a significant difference between ηtot and
∑allstages (a2qnet + 4abqside )
ηtot = ηrec when b is large. Specifically, ηtot decreases slowly because the
''
a2qsun (21) summation of conduction and front-wall loss increases slightly with b,
whereas ηrec decreases significantly with b because the heat loss at the
''
∑allstages qnet side wall is dominant which cannot be utilized.
ηrec = ''
qsun (22) To validate the single-stage simulation, we compared the compu-
tation results with finite element method (FEM) solution obtained using
COMSOL Multiphysics (see Appendix A2 for details of the FEM simu-
3.3. Model validations lation). Note that it is not straight forward to solve the heat and mass
transfer directly with P as the input using FEM. Thus we used Tf and Tb
To validate the proposed model, the theoretical prediction was given by the present solver as the boundary conditions for FEM and
''
compared with two very recent experimental studies, which demon- calculated qsun instead. As shown in Fig. 5(d), the overall efficiency
strated a membrane-based multistage solar still technique using capil- predicted by FEM is in good agreement with the present method. Minor
lary fed and interfacial heating [27,32]. The simulation parameters discrepancies are observed when b is large, because the side wall loss is
including the geometry of the device, number of stages as well as the significant and the 1D vapor transport assumption is not strictly valid.
thermal properties of the material were given by the corresponding Nevertheless, our solver can still be used to provide accurate prediction
experiments. Fig. 4(a) shows the comparison between our theoretical for the modeled solar still device since a large unit stage thickness
results and Chiavazzo et al.’s experimental measurements of a one- (b > 3 cm) is not commonly used in the practice in order to achieve
stage, three-stage and ten-stage device [27]. Fig. 4(b) shows the theo- high efficiency.
retical predictions of a three-stage and five-stage device were compared
with the corresponding experimental results given by Wang et al. [32]. 4.2. Device performance and optimization of the multistage solar still
Reasonably good agreement between the theory and experiments was
seen. In general, the simulation slightly over-predicted the cumulative To understand the effect of coupling multiple stages, we analyze the
efficiency, which can be a result of the inevitable loss of vapor and heat and mass transfer within a capillary fed TMSS when the number of
distilled water during the experiment. According to our simulation, we stages n is 16 and the unit stage thickness b is 1 cm. All of the other
found the difference between these two experiments is mainly attrib- parameters are the same as listed in Table 1. The predicted total effi-
uted to the different geometrical configuration of the device. In ciency ηtot of this device is 223% while the recyclable efficiency ηrec is

6
L. Zhang, et al. Applied Energy 266 (2020) 114864

Fig. 4. Comparison of the theoretically predicted cumulative efficiency with (a) Chiavazzo et al.’s experiments [27] and (b) Wang et al.’s experiments [32].

''
Fig. 5. Model results showing (a) front-wall vapor flux Jevap , (b) front-wall temperature Tf , (c) heat losses, and (d) overall efficiency η of the single-stage solar still
device as a function of the unit stage thickness b. The efficiency can be optimized by minimizing heat losses, which can be controlled through the device geometry.
The energy that can be reused by the next stage, quantified by the recyclable efficiency, is very sensitive to the device geometry, indicating that device geometry
optimization is essential to maximize solar still performance.

Table 1 164%. Fig. 5(a) and (b) show the vapor flux and wall temperature in
Simulation parameters used in this study. each individual stage. The vapor generation and transport mainly occur
Parameter Value Parameter Value in the first several stages (n < 8) and decreases rapidly at later stages
(Fig. 6(a)) due to a diminished temperature gradient (Fig. 6(b)).
''
qsun (W/m2) 1000 ha (W/m2K) 10 Fig. 6(c) shows the heat loss and transport in each stage. Specifically,
Tb (°C) 20 k a (W/mK) 0.026 only 60% of the total energy input into the first stage (1.5 W out of
T∞ (°C) 20 t (cm) 1 2.5 W) can be reused by the second stage because the heat loss at the
a (cm) 5 k (W/mK) 0.05 front wall is significant at high temperature (~85 °C). The energy reuse
b (cm) 0–6 ε 0.03
rate increases to about 80% for the third stage due to the absence of
Da (m2/s) 3.0 × 10−5 hfg (kJ/kg) 2357
radiative and convective heat loss on the front wall. The input energy is

7
L. Zhang, et al. Applied Energy 266 (2020) 114864

Fig. 6. (a)–(c) Heat and mass transfer within a 16-stage solar still device. (a) Front-wall vapor flux at each unit stage. (b) Front-wall temperature at each unit stage.
(c) Heat transfer rate and various heat losses at each unit stage. (d) Effect of side wall thermal insulation on the overall efficiency of the 16-stage solar still device. (e)
Effect of the number of stages on the multistage device efficiency for different geometrical configurations. For a fixed geometrical configuration, TMSS is limited in its
efficiency indicating that an increase in the number of stages is not always helpful to increase the efficiency. Proper choice of the number of stages is important to
maximize the efficiency and avoid wastage of material. In addition, enhancing the side wall thermal insulation and careful geometry optimization can significantly
improve efficiency.

almost completely dissipated through the side wall when n = 8, in- completely exposed to ambient air (ha = 10 W/m2K ) to ~1000% when
dicating that the contribution of the rest of the stages (n > 8) to the the side wall is well-insulated (t = 10 cm ). It should be noted that even
overall efficiency is negligible. This result is evident by the fact that, for a thin thermal insulation layer is very helpful. For example, the overall
example, the total efficiency of an 8-stage device (ηtot = 221%) is nearly efficiency is doubled when a 1 cm thick insulation is used compared to
the same as that of a 16-stage device (ηtot = 223%) even though the the no insulation case. As the side wall thermal resistance is increased,
number of stages is doubled. ηtot converges gradually to ηrec because the side wall heat loss becomes
Another interesting result from our analysis is related to the effect of negligible.
side wall loss on the overall efficiency of the capillary fed TMSS. We further calculated the total efficiency ηtot as a function of
Fig. 6(d) shows the calculated efficiency of this 16 stage-device as the number of stages n for different geometries. As shown in Fig. 6(e), for
side wall thermal resistance is increased. We observe that the side wall each fixed combination of a and b, there is an efficiency limit for the
insulation plays an important role in efficiency optimization. The device, i.e., ηlim = ηlim (a, b) , which is independent of the number of
overall efficiency increases from ~100% when the side wall is stages n. The minimum number of stages nlim to approach this efficiency

8
L. Zhang, et al. Applied Energy 266 (2020) 114864

Fig. 7. Global efficiency optimization by changing the geometrical configuration of the multistage solar still device. (a) Efficiency limit as a function of the unit stage
thickness for different device widths. (b) Efficiency limit as a function of the device width for different unit stage thicknesses. Efficiency limit can be optimized for a
fixed unit stage thickness or width. For a fixed unit stage thickness, a peak of efficiency limit was found, indicating the globally optimized point. This peak shifts to
the right with the unit stage width increase. In a real device design, the stage width can be first determined by the wicking length, and the unit stage thickness
corresponding to the global optimization is then calculated. The minimum number of stages to reach the efficiency limit can finally be calculated.

is also determined by device geometry a and b. For example, when transport and corresponding efficiency of a single-stage device. We then
a = 5 cm and b = 0.5 cm, ηlim is 377%, and nlim is 12. Fig. 6(e) also extended the analysis to the multistage device and observed that the
indicates that a change in device geometry can significantly affect ηlim . number of stages, side wall insulation and geometric configuration of
For example, ηlim is almost doubled when b decreases from 1 cm to the device significantly affect the overall efficiency. Specifically, we
0.5 cm. Therefore, the device geometry should be carefully optimized to found that the increase in efficiency by increasing the number of stages
maximize device performance. is bounded by an upper limit which is determined by the geometric
We maximize the efficiency limit ηlim by optimizing the device configuration of the device, since only the first several stages contribute
geometry. Fig. 7(a) shows how ηlim varies with unit stage thickness b the most to vapor generation. Increasing the side wall thermal insula-
when a = 5 cm, 10 cm and 15 cm. For each fixed a chosen in Fig. 7(a), tion can also result in an order-of-magnitude enhancement in the
there exists a maximum efficiency ηpeak = ηpeak (a) and a corresponding overall efficiency by increasing the amount of the recyclable energy.
peak position bpeak = bpeak (a) . Similar to the single-stage results, this Additionally, for each fixed device width, constrained by the wicking
optimal device geometry arises from minimizing the total heat loss of length, we found the optimized unit stage thickness to reach the global
conduction, front and side wall convection and radiation. We observe maximum efficiency of the device. This optimized thickness and cor-
that the peak position of the limit efficiency bpeak shifts to the right with responding peak efficiency increase with the device width. For ex-
a, and has the displacement relationship ηpeak ≈ 3000bpeak (see the grey- ample, we predict that a device with 15 cm width with appropriate
dashed line in Fig. 7(a)). However, from a practical consideration, the thermal insulation can achieve a solar still efficiency over 700% when
choice of a heavily relies on the wicking length of the capillary wick. the unit stage thickness is equal to 0.25 cm. The proposed modelling
For example, if the wicking length is 15 cm, at a maximum a is 15 cm framework not only provides a mechanistic understanding of capillary
and the corresponding optimized efficiency is ~ 750% when fed TMSS, but can be widely applied to guide the design and optimi-
b = 0.25 cm. Finally, we predicted the efficiency limit ηlim as a function zation of future passive desalination technologies.
of the device width a at several fixed b (0.25 cm, 0.5 cm and 1 cm). As
shown in Fig. 7(b), the efficiency limit plateaus when a is sufficiently CRediT authorship contribution statement
large (e.g., a ≈ 35 cm when b = 0.25 cm). This plateau occurs because
when b ≪ bpeak , 1D heat conduction becomes the dominant loss and the Lenan Zhang: Conceptualization, Methodology, Software,
increase in the efficiency limit due to increase in a is negligible (see
Validation, Formal analysis, Investigation, Data curation, Writing -
Fig. 7(a)). original draft, Writing - review & editing, Visualization. Zhenyuan Xu:
We performed an economic analysis to estimate the cost and pay-
Conceptualization, Formal analysis, Investigation, Writing - review &
back period of the TMSS device. Based on the material cost of the solar editing. Bikram Bhatia: Investigation, Writing - review & editing.
absorber, evaporator, condenser, and thermal insulation provided by
Bangjun Li: Investigation. Lin Zhao: Investigation, Writing - review &
recent studies [16,33], the estimated total material cost of a ten-stage
editing. Evelyn N. Wang: Writing - review & editing, Supervision,
TMSS device ranges from $30/m2 to $46/m2. If this ten-stage device is
Project administration, Funding acquisition.
operated at 600% solar thermal cumulative efficiency, the corre-
sponding estimated payback period is 1.3–1.9 years (see Appendix A3
for details). The payback period can be further reduced if a higher ef- Declaration of Competing Interest
ficiency or lower material cost can be achieved in the future.
The authors declare that they have no known competing financial
5. Conclusions interests or personal relationships that could have appeared to influ-
ence the work reported in this paper.
This work analyzes the performance of high-efficiency solar still
devices that combine thermal localization with condensation heat re- Acknowledgement
cycling. We developed a comprehensive model to understand the heat
and mass transfer during solar vapor generation and optimize the This work is supported by the National Natural Science Foundation
performance of a thermally-localized multistage solar still. First, we of China (Grant No. 51976123), Shanghai Pujiang Program (Grant No.
analyzed the working principle and quantified the heat loss, vapor 2019PJD022), the Singapore-MIT Alliance for Research and

9
L. Zhang, et al. Applied Energy 266 (2020) 114864

Technology (SMART) LEES Program.

Appendix A1. . Assessment of key assumptions

In this section, we assess several key assumptions made in this study. Assumption 3 neglects the thermal resistance of the internal wall between
neighboring two stages, which consists of a condensation plate and a capillary wick. As an example, if the material of the condensation plate is a
typical solid (such as a metal or semiconductor) with a 1 mm thickness and a thermal conductivity of ~10–300 W/mK, the thermal resistance R con
due to the condensation plate is given by,
δcon
R con = ≈ 5 × 10−6 m2K/W
kcon (A.1)
where δcon is the thickness of the plate and kcon is its thermal conductivity. Since the capillary wick is filled by water, its thermal conductivity is
mainly determined by the thermal conductivity of water. Considering a 1 mm thick capillary wick (δ wick ) which is typically used in the experiments
[27,32], the thermal resistance due to the wick (Rwick ) is given by,
δ wick
Rwick = ≈ 1.5 × 10−3 m2K/W
k wick (A.2)
where k wick is the thermal conductivity of the wick. Since the thickness of unit stage is between a millimeter and centimeter scale, and the thermal
conductivity is ka , the corresponding thermal resistance due to the air gap is given by,
b
Rgap = ≈ 0.3 m2K/W.
ka (A.3)
Comparing Eq. (A.3) with Eqs. (A.1) and (A.2), the thermal resistance of the air gap between two stages limits thermal transport, which allows
the thermal resistance due to the condensation plate and the capillary wick to be neglected in this study.
Next, we estimated the maximum error that can be introduced by assumption 3. The maximum temperature drop due to the thermal resistance of
the condensation plate and the capillary wick occurs at the first stage. When the incident solar flux is one sun, this temperature drop is about 1.5 °C,
which is typically much smaller than the temperature drop due to the air gap (≈10 °C) and the total temperature drop of the system (≈40 ~ 80 °C).
For assumption 4, we analyzed the radiation heat transfer between the first two stages where the radiation is largest compared to other stages.
According to our simulation, the temperature of the first two stages is about 87 °C and 78 °C, respectively, leading to a radiative heat flux about
13 W/m2, which is within 1.5% of the total energy flux. For this reason, neglecting the internal radiative heat transfer is a reasonable assumption.
For assumption 5, we assumed the vapor pressure of the salt aqueous solution is mainly determined by the water properties due to the low
salinity of seawater (35 g/l). Here, we justify this assumption. The vapor pressure difference Δpv between the evaporator and the condensation plate
is estimated by Raoult’s law [27],
Δpv = a (x e ) pv (Te ) − a (x c ) pv (Tc ) (A.4)
where a is the activity of the solution, which is a function of the mass fraction ( x e or x c ) of the salt in the solution. The subscript e and c represent the
evaporator and condensation plate, respectively. In this study, x e = 0.035 for the seawater and x c = 0 for the distilled water. The value of a can be
estimated as [27],
MNaCl (1 − x )
a (x ) =
MNaCl (1 − x ) + Nion MH 2O x (A.5)
where Nion = 2 for sodium chloride. MNaCl (=58.44 g/mol) and MH 2O (=18.02 g/mol) are the molar masses of sodium chloride and water re-
spectively. Using Eq. (A.5), we obtained a (x e ) = 0.98 and a (x c ) = 1, indicating the vapor transport due to the presence of salt aqueous solution is
approximately the same as that determined by the pure water properties.

Appendix A2. . Finite element method simulation using COMSOL Multiphysics

In this section, we show more details about the FEM simulation. We simulated the heat and mass transfer in the single-stage solar still device with
different unit stage thicknesses (from 1 cm to 5 cm). Fig. A1 shows an example of the simulation domain when a = 5 cm and b = 1 cm, and the
corresponding mesh used for modeling. The average element size of the mesh is about 1 mm where no meshing dependence was observed. In the
simulation, heat conduction was first solved by setting constant boundary temperatures Tf and Tb at the front wall (1) and the back wall (2),
respectively. As the heat loss through the side walls (3) is primarily due to vapor condensation, the side walls were assumed to be thermally
insulating for solving the heat conduction problem. The side wall heat loss was then calculated using Eq. (10) with the temperature distribution
provided by the heat conduction simulation as the input. The convection and radiation losses from the front wall were determined by Eqs. (5) and
(6). The evaporation heat flux was obtained by solving the mass transport in this unit stage. The vapor concentration at the front wall (1) and back
wall (2) were determined based on Tf and Tb respectively (see Fig. A1). On the side walls (3), there exists a small mass flux due to condensation driven
by side wall heat loss. This mass flux boundary condition was derived from Eq. (10), which was coupled with the heat transfer model. The heat flux
'' '' '' '' ''
due to evaporation was calculated using Eq. (7). This FEM simulation can predict qcond , qconv , qrad , qside and qevap separately, and therefore we obtained
''
qsun from Eq. (3). The FEM simulation was also used to calculate the total efficiency ηtot and the recyclable efficiency ηrec . We used the results obtained
from this FEM simulation to validate the results obtained from the model presented in this work in Fig. 5.

Appendix A3. . Economic analysis of TMSS

In this section, we provide details for the economic analysis. In previous studies [16,33], a black fabric or commercial solar absorber was typically

10
L. Zhang, et al. Applied Energy 266 (2020) 114864

Fig. A1. Finite element model using COMSOL. (a) Simulation domain with boundary conditions. The simulation domain represents a unit stage as shown in Fig. 2(a).
Brine evaporates on the wall (1) and then condensates on the wall (2). Boundaries (1) and (2) were maintained at a constant wall temperature and vapor con-
centration. The heat and mass loss due to condensation were applied on the boundary (3). (b) Domain with mesh. (c) Temperature distribution in the unit stage. (d)
Vapor concentration in the unit stage. The vapor concentration is linearly distributed along the vapor flow, indicating the side wall effect is negligible and the one-
dimensional transport is dominant.

Table A1
Data used for the economic analysis.
Items Material cost Daily solar Water unit Solar thermal
irradiation price cumulative efficiency

Values ~$30–46/m2 4.5 kWh $0.0016/L 600%

used to absorb the solar flux. The fabric wick, the polyester film or aluminum plate, and the polystyrene foam were commonly used as the eva-
porator, condenser and thermal insulation, respectively. Therefore, the approximate material cost of a ten-stage TMSS device ranges from $30/m2 to
$46/m2. The average daily production rate is approximately 41 L/m2/day considering a solar thermal cumulative efficiency of 600% and an average
daily solar irradiation in the US of 4.5 kWh/m2 [34]. The average water price in the US is approximately $0.0016/L [35], and therefore, the resulting
payback period of the TMSS device is ~1.3–1.9 years. The data used for our analysis is listed in Table A1.

References dehumidification cycle: performance of the unit. Desalination 1998;120(3):273–80.

[7] Viala E. Water for food, water for life a comprehensive assessment of water man-
agement in agriculture. Springer; 2008.
[1] Elimelech M, Phillip WA. The future of seawater desalination: energy, technology, [8] Fath HE. Solar distillation: a promising alternative for water provision with free
and the environment. Science 2011;333(6043):712–7. energy, simple technology and a clean environment. Desalination
[2] Service RF. Desalination freshens up. Science 2006; 313(5790): 1088. 1998;116(1):45–56.
[3] Shannon MA, Bohn PW, Elimelech M, Georgiadis JG, Marinas BJ, Mayes AM. [9] Badran AA, Al-Hallaq IA, Salman IAE, Odat MZ. A solar still augmented with a flat-
Science and technology for water purification in the coming decades. Nanosci plate collector. Desalination 2005;172(3):227–34.
Technol: A Collection Rev Nat J: World Sci 2010:337–46. [10] Qiblawey HM, Banat F. Solar thermal desalination technologies. Desalination
[4] Fritzmann C, Löwenberg J, Wintgens T, Melin T. State-of-the-art of reverse osmosis 2008;220(1–3):633–44.
desalination. Desalination 2007;216(1–3):1–76. [11] Kabeel A. Performance of solar still with a concave wick evaporation surface.
[5] Khawaji AD, Kutubkhanah IK, Wie J-M. Advances in seawater desalination tech- Energy 2009;34(10):1504–9.
nologies. Desalination 2008;221(1–3):47–69. [12] Bouchekima B, Gros B, Ouahes R, Diboun M. The performance of the capillary film
[6] Al-Hallaj S, Farid MM, Tamimi AR. Solar desalination with a humidification- solar still installed in South Algeria. Desalination 2001;137(1–3):31–8.

11
L. Zhang, et al. Applied Energy 266 (2020) 114864

[13] Sakthivel M, Shanmugasundaram S, Alwarsamy T. An experimental study on a re- [24] Rajaseenivasan T, Murugavel KK, Elango T, Hansen RS. A review of different
generative solar still with energy storage medium—Jute cloth. Desalination methods to enhance the productivity of the multi-effect solar still. Renew Sustain
2010;264(1–2):24–31. Energy Rev 2013;17:248–59.
[14] Velmurugan V, Gopalakrishnan M, Raghu R, Srithar K. Single basin solar still with [25] Tanaka H, Nosoko T, Nagata T. A highly productive basin-type-multiple-effect
fin for enhancing productivity. Energy Convers Manage 2008;49(10):2602–8. coupled solar still. Desalination 2000;130(3):279–93.
[15] Zhao F, Zhou X, Shi Y, Qian X, Alexander M, Zhao X, et al. Highly efficient solar [26] Xiong J, Xie G, Zheng H. Experimental and numerical study on a new multi-effect
vapour generation via hierarchically nanostructured gels. Nat Nanotechnol solar still with enhanced condensation surface. Energy Convers Manage
2018;13(6):489. 2013;73:176–85.
[16] Ni G, Zandavi SH, Javid SM, Boriskina SV, Cooper TA, Chen G. A salt-rejecting [27] Chiavazzo E, Morciano M, Viglino F, Fasano M, Asinari P. Passive solar high-yield
floating solar still for low-cost desalination. Energy Environ Sci 2018;11(6):1510–9. seawater desalination by modular and low-cost distillation. Nat Sustain
[17] Ghasemi H, Ni G, Marconnet AM, Loomis J, Yerci S, Miljkovic N, et al. Solar steam 2018;1(12):763.
generation by heat localization. Nat Commun 2014;5. ncomms5449. [28] Toyama S, Aragaki T, Murase K, Tsumura K. Simulation of a multieffect solar dis-
[18] Tao P, Ni G, Song C, Shang W, Wu J, Zhu J, et al. Solar-driven interfacial eva- tillator. Desalination 1983;45(2):101–8.
poration. Nat Energy 2018. [29] Jubran BA, Ahmed M, Ismail AF, Abakar Y. Numerical modelling of a multi-stage
[19] Ni G, Li G, Boriskina SV, Li H, Yang W, Zhang T, et al. Steam generation under one solar still. Energy Convers Manage 2000;41(11):1107–21.
sun enabled by a floating structure with thermal concentration. Nat Energy [30] Lu Z, Kinefuchi I, Wilke KL, Vaartstra G, Wang EN. A unified relationship for
2016;1(9). nenergy2016126. evaporation kinetics at low Mach numbers. Nat Commun 2019;10(1):2368.
[20] Zhou L, Tan Y, Wang J, Xu W, Yuan Y, Cai W, et al. 3D self-assembly of aluminium [31] Zhang L, Zhao L, Wang EN. Stefan flow induced natural convection suppression on
nanoparticles for plasmon-enhanced solar desalination. Nat Photon high-flux evaporators 104255 Int Commun Heat Mass Transfer2019.
2016;10(6):393. [32] Wang W, Shi Y, Zhang C, Hong S, Shi L, Chang J, et al. Simultaneous production of
[21] Liu Z, Song H, Ji D, Li C, Cheney A, Liu Y, et al. Extremely cost-effective and effi- fresh water and electricity via multistage solar photovoltaic membrane distillation.
cient solar vapor generation under nonconcentrated illumination using thermally Nat Commun 2019;10(1):3012.
isolated black paper. Global Challenges 2017;1(2):1600003. [33] Xu Z, Zhang L, Zhao L, Li B, Bhatia B, Wang C, et al. Ultrahigh-efficiency desali-
[22] Jia C, Li Y, Yang Z, Chen G, Yao Y, Jiang F, et al. Rich mesostructures derived from nation via a thermally-localized multistage solar still. Energy Environ Sci 2020.
natural woods for solar steam generation. Joule 2017;1(3):588–99. [34] Sengupta M, Xie Y, Lopez A, Habte A, Maclaurin G, Shelby J. The National Solar
[23] Liu H, Chen C, Chen G, Kuang Y, Zhao X, Song J, et al. High-performance solar Radiation Data Base (NSRDB). Renew Sustain Energy Rev 2018;89:51–60.
steam device with layered channels: artificial tree with a reversed design. Adv [35] Statista. Average monthly residential cost of water in the U.S. from 2010 to 2019 (in
Energy Mater 2018;8(8):1701616. U.S. dollars).

12