You are on page 1of 13

Shaneza Fatma

Rahmadhanty
Department of Mechanical Engineering,
National Taiwan University of Science and
Technology,
No. 43, Keelung Road, Sec. 4,
Design Optimization and
Taipei 10607, Taiwan
email: shanezafr@gmail.com Manufacturing of A Self-Heated
Brijesh Patel
Department of Mechanical Engineering,
Vacuum Membrane Distillation
National Taiwan University of Science and
Technology,
No. 43, Keelung Road, Sec. 4,
System
Taipei 10607, Taiwan
email: aero.brijesh@gmail.com At present, there is a notable advancement in technology that has spread to various do-
mains, including the application of membrane distillation (MD) in water treatment. This
Subrahmanya T. M. technology holds particular significance in the purification of saline water, given the es-
Graduate Institute of Applied Science and calating scarcity of potable water required for consumption across domestic, agricultural,
Technology, and industrial spheres. The utilization of a self-heated vacuum membrane distillation
National Taiwan University of Science and (VMD) system represents a noteworthy approach in the establishment of membrane distil-
Technology, lation (MD) systems. The process involves the utilization of a graphene-PVDF membrane
No. 43, Keelung Road, Sec. 4, that is subjected to heating through a power supply, The problems brought on by pre-feed
Taipei 10607, Taiwan heating are intended to be resolved by this method. The temperature polarization may be
email: tmssubrahmanya@gmail.com managed in related research and observations. Additionally, several designs are created
and then contrasted in order to enhance performance indicators including permeate flux
(𝐽𝑊 ), temperature polarization factor (TPF), specific heating energy (𝑄 𝑆𝐻 ), and gain
Wei-Song Hung output ratio (GOR). The present research examines four parameters: the voltage of the
Graduate Institute of Applied Science and DC power supply, the feed flow rate, and the dimensions of the cell body’s slot, including
Technology, its length and width. The system is simulated using Ansys FLUENT software, and the
National Taiwan University of Science and results are afterwards analyzed using Minitab software with the application of the Re-
Technology, sponse Surface Method (RSM). The objective of implementing RSM is to attain the optimal
No. 43, Keelung Road, Sec. 4, design parameters. The accuracy of the simulation data was confirmed through exper-
Taipei 10607, Taiwan imental data, and the identification of the optimal conditions for the self-heated VMD
email: wshung@mail.ntust.edu.tw system resulted in optimum outcomes, including the maximization of 𝐽𝑊 , TPF, and GOR,
while minimizing 𝑄 𝑆𝐻 .
Po Ting Lin1 Keywords: Vacuum Membrane Distillation, optimization, Response Surface Methodology,
High-Speed 3D Printing Research Center,
RSM
National Taiwan University of Science and
Technology,
No. 43, Keelung Road, Sec. 4,
Taipei 10607, Taiwan
email: potinglin@mail.ntust.edu.tw

1 Introduction saline water. VMD employs a vacuum pump to create a vacuum


The freshwater demand that rises over years [1–3], leads to a lot on the permeate side, thereby generating a vapour pressure differ-
of curiosity in technology that has access to a recently undiscovered ential across the membrane. This differential effectively reduces
saline water supply, such as Membrane Distillation (MD) [4,5]. heat loss at low temperatures [11,12].
MD is a separation process that operates based on thermal princi- The present investigation employs a superhydrophobic mem-
ples, wherein solely water vapour molecules are allowed to traverse brane composed of graphene and PVDF as a joule heater within a
through a microporous hydrophobic membrane, separating the hot self-heated VMD system, which is directly linked to a direct cur-
feed from the cold permeate. The primary driving force in the rent (DC) power supply. Therefore, the integration of feed stream
MD process is the vapour pressure gradient resulting from the interfacial heating with conventional vapour membrane distillation
temperature disparity across the hydrophobic membrane [6]. Typi- (VMD) leads to enhanced energy efficiency. One notable bene-
cally, a hydrophobic membrane is comprised of synthetic materials, fit of this system lies in its ability to eliminate the need for feed
such as polyvinylidene fluoride (PVDF), polytetrafluoroethylene fluid preheating and mitigate temperature polarization (TP). This is
(PTFE), and polypropylene (PP) [7–10]. achieved by incorporating the membrane within a filtration module,
The utilization of MD technology, particularly in the treatment such as the VMD module assembly depicted in Figure 1. Conse-
of saline water, is regarded as a promising strategy for addressing quently, the fluid on the feed side can be efficiently heated at the
the substantial need for freshwater. This study will primarily in- interface between the feed and the membrane. In addition, there is
vestigate the application of self-heated VMD for the treatment of an increase in the permeate water flux, resulting in a reduction in
the energy demand for the desalination process [13–15].
1 Corresponding Author. Several studies are documenting the progress of implementing
Version 1.18, December 19, 2023 optimization in water treatment and purification using membrane

Journal of Thermal Science and Engineering Applications PREPRINT FOR REVIEW / 1


Copyright © 2023 by ASME
technology, especially in distillation desalination systems. In 2017, rate, and the dimension of the slot in terms of length and width.
Dudchenko et al. [16] invented a self-heating membrane by creat- The performance of the design is assessed using Ansys software in
ing a thin-film structure using conductive, hydrophilic, and porous this study. Response Surface Methodology (RSM) is also used as
carbon nanotubes (CNTs) and polyvinyl alcohol (PVA). This struc- the technique for optimizing the performance of self-heated VMD
ture was achieved through a layer-by-layer spray coating technique, systems, including current designs and newly proposed parametric
which was applied onto a hydrophobic porous membrane support designs. In this case, RSM is a suitable statistical method to evalu-
made of polytetrafluoroethylene (PTFE). The performance results ate and simultaneously resolve multi-variable equations correlating
were 3.5 L/m2 h of JW , 1.25±0.02 kWh/L of QSH , and 0.55 of dependent parameters (response) [24].
GOR, with input variables of 50 W of power supply, 6.6 mm/s of
feed flow rate, and NaCl 100 g/L of saline water feed. Another
study of self-heating VMD was presented in 2019 by Anvari et
al. [17] using spray coating technique involving the deposition of
a mixture of iron oxide and carbon nanotubes onto a hydrophobic
polytetrafluoroethylene membrane. With the saline water feed of
35-100 g/L NaCl, it resulted in 4 L/m2 h of JW , 0.2 kWh/L of QSH ,
and 3.45 of GOR at low inlet flow velocity (2.33 cm/min) and low
vacuum (20 kPa) conditions. And for the latest research, in 2021,
Subrahmanya et al. [18] found that by delivering 10 V of DC power
to a membrane made of graphene-PVDF with 1.0 mL/min of 3.5
wt% NaCl saline water solution feed, it achieved 23.44 L/m2 h of
JW , 0.109 kWh/L of QSH , and 5.72 of GOR.
For optimization in membrane distillation, especially for the
DCMD configuration, Boubakri et al. did it by using RSM in
2013 [19]. The researchers incorporated four variables into their
studies, namely vapour pressure differential, feed flow rate, per-
meate flow rate, and feed ionic strength. In 2018, Ali et al. [20]
also found that the optimization of the length and thickness of the
module also affects the performance of the MD system. Further- Fig. 1 The self-heated VMD module assembly
more, Eykens et al. [21]. conducted a study that focused on the
enhancement of membrane characteristics in MD with regard to
hydrophobicity. While optimization of desalination by VMD is
proposed by Mohammadi, et al. [22], using the Taguchi method to 2 Response Surface Methodology
strategically design a minimal number of tests utilizing a flat sheet
Response Surface Methodology (RSM) refers to a set of statisti-
PP membrane as the primary component in 2009. It selected three
cal tools that may be used for the purpose of designing, enhancing,
levels of parameters, resulting in the optimal levels of 55°C (tem-
and optimizing a range of processes. These processes may include
perature), 30 mL/s (flow rate), 30 mbar (vacuum pressure), and 50
both novel and pre-existing ones [25]. Typically, RSM uses two
g/L (concentration). In 2020, Yang et al. [23] completed a study
variables: response and independent variables. Response is the
to assess the efficacy of the VMD desalination process, as docu-
process performance/quality that is influenced by the input vari-
mented in their publication. Although it is unable to determine the
ables. Those are assessed on a continuous scale. In the context
appropriate parameter for achieving optimal performance, an arti-
of a test/experiment, independent variables are the input variables
ficial neural network (ANN) may be used to anticipate and demon-
that the scientist/supervisor can choose. One of the advantages of
strate the potential for effectively evaluating the performance of
RSM is that it can optimize the response, and in addition, contour
the VMD. The evaluation of performance parameters is conducted
plots and surface plots can be generated.
over a range of temperatures at the feed inlet, different feed flow
If a response-y product, method, or device that depends on the
rates, and varying membrane lengths.
controllable input variables ζ1 , ζ2 , . . . , ζ𝑛 is built, an appropriate
However, the existing literature hasn’t talked about how to op- regression model can be written as shown in Eq. (1) below,
timize a self-heated VMD using a graphene-PVDF membrane as
a joule heater in a filtration module. This is because this system 𝑦 = 𝑓 (ζ1 , ζ2 , . . . , ζ 𝑘 ) + 𝜖 (1)
is a new technology that hasn’t been used before. Through this
study, it is hoped that the latest and most useful information and where the form of the true response function, denoted as 𝑓 is
insights can be found about self-heating VMD systems that use a currently unknown and potentially intricate. Additionally, the word
graphene-PVDF membrane as the joule heater. This will help to 𝜖 represents sources of unpredictability that are not accounted for
enhance the system’s performance by maximizing the water perme- by 𝑓 .
ate flux (JW ), temperature polarization factor (TPF), gain output In this study, the regression model uses a second-order approx-
ratio (GOR), and minimizing the specific heating energy (QSH ) imating function, which is a full-quadratic regression model func-
with the optimal desired energy, feed flow rate, and design param- tion, as its proximity to the response area. The objective functions
eters. And make the energy used more efficient, so production shown in Eq. (2) and for the model using full quadratic regression
costs can also be minimized. are shown in Eq. (3).
In terms of the design parameter itself, this research also makes
several new designs of cell-body models by adding slots in them so 𝑌 = 𝑓 (𝑥1 , 𝑥2 , 𝑥3 , 𝑥4 ) (2)
that the water flow can experience maximum evaporation. Theoret-
ically, the longer and greater the number of slots, the more tortuous 𝑘
∑︂ 𝑘
∑︂ 𝑘
∑︂ ∑︂
the fluid flow, which can lead to maximal evaporation with a min- 𝑌 = 𝛽0 + 𝛽𝑖 𝑥 𝑖 + 𝛽𝑖𝑖 𝑥𝑖2 + 𝛽𝑖 𝑗 𝑥 𝑖 𝑥 𝑗 (3)
imal energy source. Moreover, by increasing the width of the slot, 𝑖=1 𝑖=1 𝑖< 𝑗=2
there is less space for fluid to flow, so less water will evaporate. To
overcome this problem, the parameters of the length and width of The response function is denoted as 𝑌 , while the coefficients of
the slot will be added. Then, the performance results will be ana- the linear, quadratic, and interaction factors are represented by 𝛽𝑖 ,
lyzed for each design having a different number of slots, with the 𝛽𝑖𝑖 , and 𝛽𝑖 𝑗 , respectively. The variables 𝑥𝑖 and 𝑥𝑗 are considered
input variables being the DC power source’s voltage, the feed flow to be independent variables in this context.

2 / PREPRINT FOR REVIEW Transactions of the ASME


Before doing optimization, defining the problem needs to be minimization case, it is obtained from the lowest value. And for
done first. Eq. (4) shows the optimization problem. The function the weight of 𝑤, is chosen to be 1.
of 𝑦 corresponds to the response which are JW , TPF, and GOR Such equations’ graphical representations are called response
that must be maximized, while QSH is the opposite. The factors surfaces. It may analyze the mutual interactions among the input
𝑥1 , 𝑥2 , 𝑥3 , and 𝑥4 represent the input variables, namely the DC factors and their impacts on the response variable. To analyze
voltage, the feed flow rate, the length, and the width of the slot, the data from the experiment, this research uses Minitab software,
respectively. which can analyze statistics.

Max/Min 𝑦 = 𝑓 (𝑥1 , 𝑥2 , 𝑥3 , 𝑥4 ) 3 Method and Application of Response Surface


This section provides a comprehensive overview of the method-
Subject to: ology used in the current investigation, as seen in Figure 2.

𝑥1𝐿 ≤ 𝑥1 ≤ 𝑥𝑈
1

𝑥2𝐿 ≤ 𝑥2 ≤ 𝑥𝑈
2 (4)

𝑥3𝐿 ≤ 𝑥3 ≤ 𝑥𝑈
3

𝑥5𝐿 ≤ 𝑥5 ≤ 𝑥𝑈
5

Then the optimization is calculated using the desirability func-


tion, whose equation is shown in Eq. (5) below,

𝑛
(︃∏︂ )︃ 1
𝑛
𝐷= 𝑑𝑖 (𝑌𝑖 ) (5)
𝑖=1

where 𝐷 is the overall composite desirability, 𝑑𝑖 is the individ-


ual desirability, 𝑌𝑖 is the predicted value of 𝑖 𝑡 ℎ response, and 𝑛
is the number of responses. A desirability function 𝑑𝑖 (𝑌𝑖 ) gives Fig. 2 Flow chart of methodology
numerical values ranging from 0 to 1. A value of 𝑑𝑖 (𝑌𝑖 ) = 0 in-
dicates a response value that is entirely unwanted, while a value
of 𝑑𝑖 (𝑌𝑖 ) = 1 represents a response value that is wholly desired or After reviewing the literature, the process starts with the selec-
perfect. tion of variables, which are the DC voltage of the power source,
To calculate the individual desirability, Derringer and Suich [26] the feed flow rate, and the size of the slot added to the cell body
proposed forms of the related desirability function. If a response is model, including the length and width.
maximized. The individual desirability is defined as Eq. (6) when The selection of DC voltage is based on consideration of the
a response is maximized. energy used, which is proportional to the cost consumed. While
the feed flow was chosen because it considers the time consumed.
In addition, 11 new designs that add slots, whose design concept
⎧ 0(︃ if 𝑌𝑖 < 𝐿 𝑖


⎨ 𝑌 −𝐿 𝑤


)︃ can be seen in Figure 3 below, are introduced, with the dimension
𝑑𝑖 (𝑌𝑖 ) = 𝑖 𝑖
if 𝐿 𝑖 ≤ 𝑌𝑖 ≤ 𝑇𝑖 (6) of the fluid module being 30 x 40 mm and the inlet and outlet
𝑇𝑖 −𝐿𝑖


⎪ channels being 5mm in diameter. The design is analyzed by de-
⎪1 if 𝑌𝑖 > 𝑇𝑖
⎩ termining the number of slots as well as the length and width of
the slot [27]. This is because the slot’s dimensions affect the path
The individual desirability is defined as Eq. (7) below to min- of fluid flow, which will also affect the evaporation process. The
imise a response. selected performances and the factors are based on existing theo-
retical references, recommendations from experts, and limitations

⎪ 1(︃ if 𝑌𝑖 < 𝑇𝑖 to be manufactured. The value and unit of the selected factors can
be seen in Table 1 below. They are represented as 𝑥1 , 𝑥2 , 𝑥3 , and

⎨ 𝑈 −𝑌 𝑤
⎪ )︃

𝑑𝑖 (𝑌𝑖 ) = 𝑈𝑖𝑖 −𝑇𝑖𝑖 if 𝑇𝑖 ≤ 𝑌𝑖 ≤ 𝑈𝑖 (7) 𝑥4 . Table 1 also shows the description of all the cases, which con-



⎪0 sist of 12 cases in total, including the original one, and the design
⎩ if 𝑌𝑖 > 𝑈𝑖 of the experiment used in each case as well.
After designing the factors and values, the subsequent stage in-
𝑇𝑖 is the target value of 𝑖 𝑡 ℎ in this context, 𝐿 𝑖 is the lowest ac- volves the design of experiment (DOE). DOE is an effective method
ceptable value of 𝑖 𝑡 ℎ response, 𝑈𝑖 is the highest acceptable value of for organizing experiments for the production of new products and
𝑖 𝑡 ℎ response, and 𝑤 signifies the weight assigned to the desirability process modifications. It is carried out based on suitable statis-
function. A value of 𝑤 = 1 indicates that the desirability function tical analysis to produce the best possible combinations of fac-
is linear, when 𝑤 > 1, the desirability function tends to approach tors/characteristics of process parameters that have an impact on
the target value more closely. On the other hand, when 0 < 𝑤 < 1, the quality of a product or process [28].
the importance of the desirability function is relatively diminished. In this work, both Central Composite Design (CCD) and Box-
In this case, the basis for the selection of the values of 𝐿 𝑖 and Behnken Design (BBD) are used. CCD contains factorial design
𝑈𝑖 are established as the minimum and maximum values of the from cube point, centre runs, and star runs/axial runs. CCD of
response, respectively. The 𝑇𝑖 value in the maximization case is three-level two-factor factorial design (32 ) is used to make the
obtained from the highest value of the response since there is no DOE of the original design case. This original design only has 2
limitation on the acceptable value of the response, while in the factors: x1 for the feed flow rate and x2 for the DC voltage.

Journal of Thermal Science and Engineering Applications PREPRINT FOR REVIEW / 3


Table 1 The factors and values

Flow rate Voltage Length Width


DOE
Case Slot Orientation (mL/min) (V) (mm) (mm)
𝑥1 𝑥2 𝑥3 𝑥4
1 0 (original) - 5 - 25 10 - 30 - - CCD 32
2 1 Vertical 5 - 25 10 - 30 15 - 25 2-6 BBD 34
3 1 Horizontal 5 - 25 10 - 30 20 - 35 2-6 BBD 34
4 2 Vertical 5 - 25 10 - 30 15 - 25 2-6 BBD 34
5 2 Horizontal 5 - 25 10 - 30 20 - 35 2 - 3.5 BBD 34
6 3 Vertical 5 - 25 10 - 30 15 - 25 2-5 BBD 34
7 3 Horizontal 5 - 25 10 - 30 20 - 35 2 - 2.5 BBD 34
8 4 Vertical 5 - 25 10 - 30 15 - 25 2 - 3.5 BBD 34
9 4 Horizontal 5 - 25 10 - 30 20 - 35 2 BBD 33
10 5 Vertical 5 - 25 10 - 30 15 - 25 2-3 BBD 34
11 5 Horizontal 5 - 25 10 - 30 20 - 35 2 BBD 33
11 6 Vertical 5 - 25 10 - 30 15 - 25 2 - 2.15 BBD 34

Table 2 Design of experiment of CCD and BBD

CCD 32 BBD 33 BBD 34


Run
𝑥1 𝑥2 𝑥1 𝑥2 𝑥3 𝑥1 𝑥2 𝑥3 𝑥4
1 -1 -1 -1 0 -1 0 0 -1 -1
2 1 -1 -1 0 1 0 0 -1 1
3 -1 1 1 0 -1 0 0 1 -1
4 1 1 1 0 1 0 0 1 1
5 0 0 0 -1 -1 -1 -1 0 0
6 0 0 0 -1 1 1 -1 0 0
(a) (b) 7 0 0 0 1 -1 -1 1 0 0
8 -1.414 0 0 1 1 1 1 0 0
Fig. 3 The new design: (a) vertical (b) horizontal shape 9 1.414 0 -1 -1 0 0 -1 0 -1
10 0 -1.414 1 -1 0 0 -1 0 1
11 0 1.414 -1 1 0 0 1 0 -1
Other cases for the new designs containing more than two factors 12 0 0 1 1 0 0 1 0 1
use BBD as DOE, considering BBD has fewer runs of experimental 13 0 0 0 0 0 -1 0 -1 0
units. This is because BBD only has a factorial incomplete block 14 0 0 -1 0 1 0
design and centre runs without star/axial runs. For 4- and 5-slot 15 1 0 -1 0
design cases in which the width of the slot has no range and is set 16 1 0 1 0
to be only 2 mm, the factors are 3: 𝑥1 for the feed flow rate, 𝑥2 for 17 -1 0 0 -1
the DC voltage, and 𝑥3 for the length of the slot. These cases use 18 -1 0 0 1
BBD of a three-level three-factor factorial design (33 ). 19 1 0 0 -1
While for the remaining cases, they have 𝑥1 , 𝑥2 , 𝑥3 , and 𝑥4 for 20 1 0 0 1
the feed flow rate, DC voltage, and the length and width of the slot, 21 0 -1 -1 0
respectively. BBD of three-level four-factor factorial design (34 ) is 22 0 -1 1 0
used regarding these remaining cases. To facilitate understanding, 23 0 1 -1 0
a summary of the use of DOE in each case in this study can be 24 0 1 1 0
seen in Table 1. 25 0 0 0 0
The design table for DOE is shown in Table 2. The three-
level is often denoted as the low, intermediate, and high levels,
corresponding to numerical values of -1, 0, and 1, respectively.
where 𝑣 is feed flow velocity (m/s); 𝑄 is flow rate (m3 /s); and 𝐴 is
And since this is a simulation approach, the centre points are not
repeated. cross sectional area of inlet channel (m2 ).
The next step is to do a simulation. The software for the simula- The geometries are made with Design Modeler, carried by An-
tion used is Ansys FLUENT. Before entering Ansys FLUENT, the sys. And for the FLUENT setup, the energy equation is on with
heat flux and inlet feed velocity magnitude are defined first. The a laminar viscous model. In addition, the boundary conditions
heat flux is converted from the power supply, as shown in Eq. (8). are set to be in 5 different zones, which are: feed bulk, module
While the inlet feed velocity is calculated from Eq. (9). wall, membrane surface, inlet feed, and outlet feed. Feed bulk
is the fluid domain parallel to the membrane. The feed module
𝐼 ×𝑉 wall is classified as wall type with assumptions of stationary wall
𝜙= (8) motion, no-slip shear condition, made from plastic, as the material
𝐴
property is shown in Table 3, and a wall thickness of 0.005 m.
where 𝜙 is heat flux W/m2 , 𝐼 is current (A); 𝑉 is voltage (V); and The membrane surface is also the same as the wall type condition
𝐴 is the cross-sectional area of the membrane (m2 ). with the module wall, but the wall thickness is 0.15 mm, made
from graphene-PVDF as the material property, which is calculated
𝑄 using the function of mixtures [29–31], as shown in Table 3. Ad-
𝑣= (9) ditionally, the specific heat flux is determined based on the input
𝐴

4 / PREPRINT FOR REVIEW Transactions of the ASME


variables analyzed using the response surface methodology (RSM). where 𝜖 is the membrane porosity, 𝜌 𝑚 is the density of mem-
The inlet feed has different velocity magnitudes depending on the brane material (kg/m3 ), and 𝜌 𝑝 is the density of polymer
input variables examined by RSM and a constant temperature of material (kg/m3 ).
18°C, this temperature is set to be the same as the outlet temper-
ature. While the fluid domain is assumed to be saline water not Tortuosity (𝜏) is defined as a divergence in pore structure
passing through the membrane pores. from a cylindrical form. As a result, the smaller the permeate
flux, the greater the tortuosity value. However, porosity and
tortuosity are correlated and can be different depending on the
Table 3 Material properties used in the Ansys simulation
material of the porous beds. In this case, the most suitable
Material Density Specific Thermal Viscosity
correlation was proposed by Wojciech S. et al. [34], as shown
Name Heat Conductivity in Eq. (14) below,
(𝑘𝑔/𝑚 3 ) (𝐽/𝑘𝑔𝐾) (𝑊/𝑚𝐾) (𝑘𝑔/𝑚𝑠) (2 − 𝜖) 2
Plastic [29] 950 1900 0.45 -
𝜏= (14)
𝜖
Graphene-PVDF [30–33] 1197 1499 783.24 -
Saline-water 1019 4068 0.6 0.001052 where 𝜏 is the membrane tortuosity and 𝜖 is the membrane
porosity.
Once the simulation is done, a numerical analysis is carried • Temperature Polarization Factor (TPF)
out. The output of the simulation using Ansys FLUENT is the To measure the TP’s degree, the Temperature Polarization
temperature distribution on the boundary wall, especially on the Factor (TPF) is used. As it is known, in conventional MD,
membrane surface wall and the bulk, which can be used to achieve when the hot feed is coming, the heat transferred through
the value of the performances JW , TPF, QSH , and GOR. the membrane makes the temperature drop in the surrounding
area to the cold side, such as the bulk side. This wastes
• Permeate Flux (JW ) the heat that would have been used to evaporate the saline
Throughout the operation of the system, the liquid that had water. This heat loss must be eliminated. Thus, TPF must be
permeated was collected and its volume was assessed at con- maximized to get better performance. TPF is calculated with
sistent intervals of time. The more permeate water that is Eq. (15) below.
collected in a short period of time, the better the performance (︃ 𝑚 )︃
will be. Thus, to get better performance, the permeate flux 𝑇𝑓
must be maximized. In accordance with Darcy’s law, the 𝑇 𝑃𝐹 (%) = 𝑏
× 100 (15)
formula for calculating JW is given in Eq. (10) below [11], 𝑇𝑓

where 𝑇𝑚 𝑓 is the membrane surface temperature at the feed


𝐽𝑊 = (𝐾 𝑚 ) × (𝑀) −0.5 × (𝑃𝑓 𝑚 − 𝑃𝑣 ) (10) side (K) and 𝑇𝑏 𝑓 is the temperature of the bulk feed.

where 𝑃𝑓 𝑚 is the partial pressure of water vapour at the feed • Specific Heating Energy (QSH )
side at the temperature of the membrane surface (Pa), 𝑃𝑣 is The ratio between energy consumption inputted and permeate
the permeate side’s downstream pressure (30,000 Pa), 𝑀 is water flux is identified as specific heating ratio (QSH ). The
the molar mass of water (18.01528 g/mol), and 𝐾 𝑚 is the performance will get better if QSH is made to a minimum so
coefficient of membrane mass transfer (smol1⁄2 kg-1⁄2 m), that that the input heating energy needed (including the electrical
is affected by the temperature of the feed as well as some energy) to collect permeate flux is as minimal as possible.
geometric properties of the membrane material as displayed The input energy. QSH is calculated with Eq. (16) below,
in Eq. (11) below, (︃ )︃
𝑄 𝑖𝑛
𝑄 𝑆𝐻 = (16)
)︃ −0.5 (︃ 𝐽𝑊
8 × 𝑅 × 𝑇 0.5
(︃ )︃ (︃ )︃
2×𝜖 ×𝑟 1
𝐾𝑚 = × × (11)
3×𝜏 𝑅 ×𝑇 𝜋 where 𝑄 𝑖𝑛 is the heating energy consumption (kW/m2 ) and
𝐽𝑊 is the permeate flux (L/m2 h). To calculate the heating
where 𝜖 is the porosity of the membrane (0.27), 𝑟 is the mean energy consumption (𝑄 𝑖𝑛 ), Eq. (17) is used which is shown
pore radius of the membrane (0.0025 mm), 𝜏 is pore tortuosity below,
of the membrane (11.087), 𝛿 is the thickness of the membrane 𝑣×𝐼
(0.1 mm), 𝑅 is the gas constant (8.31 J/mol.K), and 𝑇 is the 𝑄 𝑖𝑛 = (17)
𝐴 × 1000
average temperature of the membrane (K). While to find the
amount of water vapour pressure, 𝑃𝑓 𝑚, a formula proposed where 𝑣 is voltage of power source, 𝐼 is current of power
by Antoine is used as shown in Eq. (12) shown below, source (Ampere), and 𝐴 is the effective membrane area (m2 )
(︃ )︃ • Gain Output Ratio (GOR)
3816.44
𝑃𝑓 𝑚 = exp 23.1964 − 𝑚 (12) Calculating the ratio of the latent heat energy of evaporation to
𝑇 𝑓 − 46.13 the total heating energy input requires the usage of the GOR.
The perfect system indicates that the amount of latent heat
𝑚
where 𝑇 𝑓 is the temperature of membrane surface. energy required for water evaporation is larger than the entire
amount of heating energy that is fed into the system. Thus,
In addition, membrane porosity and tortuosity also contribute for better performance, the GOR value must be maximized.
to the VMD system performance. The porosity of the mem- The GOR can be calculated as per Eq. (18) given below,
brane has the greatest impact on the rate of mass transfer
across it. Membrane porosity ranges from 30 to 85 percent, 𝐽 × ℎ𝑣
𝐺𝑂𝑅 = 𝑊 (18)
according to El-Bourawi et al. [14]. Smolder-Franken’s equa- 𝑄 𝑖𝑛
tion may be used to calculate porosity (𝜖) shown in Eq. (13)
below, where 𝐽𝑊 is the permeate flux (L/m2 h), ℎ 𝑣 is the latent heat
𝜌𝑚 of evaporation for water (2256 kJ/kg), and 𝑄 𝑖𝑛 is the heating
𝜖 =1− (13) energy consumption (kW/m2 ).
𝜌𝑝

Journal of Thermal Science and Engineering Applications PREPRINT FOR REVIEW / 5


If all the previous steps are finished, then move into RSM anal-
ysis using Minitab. The regression model uses a second-order
function as shown in Eq. (3) and the last step is doing optimization
of the problem using the composite desirability function as stated
in Eq. (5) with the problem of the optimization is defined from
Eq. (4).

4 Results and Discussion (a) (b)


The 12 cases’ data results are obtained from simulation, and the
data analysis is assisted by using Minitab software with Response
Surface Methodology features. To get the optimal parameter for
each case, this research utilizes the desirability function as shown
by Eq. (5)-(7). Once they are obtained, they will be compared to
determine the best design for this self-heated VMD system. To
validate the analysis, the simulation using Ansys FLUENT of the
related factors and the experiment regarding the best design of all
are also conducted. Then, this chapter ends with a discussion of (c) (d)
the effective heating temperature of the optimal case
Fig. 4 Surface plot of voltage and flow rate to: (a) JW (b)
4.1 Results Data and Analysis. The results of Case 1, which TPF (c) QSH (d) GOR of Case 1
is the original design, are shown first. The summary of factor
and response results for Case 1 is shown in Table 4. To describe
5-slot horizontal shape design, gives the best result in JW as a
the relationship between the input (factors) and the response, an
result of response optimizer analysis. Case 6, a 3-slot vertical
equation regression is established. The regression equations for
shape design, gives the best TPF. While the best QSH and GOR
JW , TPF, QSH , and GOR, respectively, are shown in Eq. (19) with
are from Case 1. In the simulation validation, the most optimal
factors: A is flow rate and B is voltage. While the surface plot to
JW and TPF were obtained from Case 11. Case 1 still has the best
describe the results of the response is shown in Figure 4.
QSH and GOR, but Minitab’s validation of the response optimizer
shows some slight value fluctuation.
𝐽𝑊 = 2.1559 − 0.01324𝐴 + 0.00656𝐵 + 0.000434𝐴2 + Because Case 11 gives the best composite desirability value of
0.4796, which reflects the optimum parameter settings that produce
the most desirable quality characteristics, therefore, this study is
0.000051𝐵2 − 0.000268𝐴𝐵 going to show the results of Case 11. While all these parameters are
important, permeate flux (JW ) is often the primary focus because it
𝑇 𝑃𝐹 = 0.999082 − 0.000168𝐴 + 0.000552𝐵 + 0.000010𝐴2 directly affects the production rate and throughput of the process.
In Case 11, Eq. (20) shows the regression equation for each
+ 0.000019𝐵2 − 0.000022𝐴𝐵 response, where A is flow rate, B is voltage, and C is the length
of the slot. The width of the slot is not taken into account in this
𝑄 𝑆𝐻 = − 0.0282 + 0.001689𝐴 + 0.008019𝐵 − 0.000093𝐴2 regression equation since its value is kept at 2 mm considering the
available space as described in Table 1. The summary of factors
and response results in Case 11 using the BBD matrix is shown in
+ 0.000855𝐵2 + 0.000134𝐴𝐵 Table 6.

𝐺𝑂𝑅 = 14.27 + 0.093𝐴 − 1.097𝐵 − 0.00318𝐴2 +


𝐽𝑊 = 2.221 − 0.00736𝐴 + 0.01034𝐵 − 0.00968𝐶
2
0.02051𝐵 − 0.00004𝐴𝐵
(19) + 0.000228𝐴2 + 0.000188𝐵2 + 0.000177𝐶 2

From Figure 4, it can be concluded that the greater the volt − 0.000383𝐴𝐵 + 0.000154𝐴𝐶 − 0.000252𝐵𝐶
value and the smaller the flow rate value, the more optimal the
performance of JW and TPF. As for QSH and GOR, the optimal 𝑇 𝑃𝐹 = 0.9955 − 0.00275𝐴 + 0.001225𝐵 + 0.00001𝐶
performance is obtained with a smaller volt value, while the flow
rate value has no influence on the performance. + 0.000029𝐴2 + 0.000025𝐵2 + 0.000005𝐶 2
In terms of optimization, based on the solutions from the re-
sponse optimizer, the optimal parameters that are expected to − 0.000046𝐴𝐵 − 0.000008𝐴𝐶 − 0.000016𝐵𝐶
achieve the most optimal performances in Case 1 for the feed flow (20)
rate and DC voltage, respectively, are 5 mL/min and 13.84 V, which 𝑄 𝑆𝐻 = 0.231 + 0.00056𝐴 − 0.01273𝐵 − 0.0107𝐶
for the obtained performances are 2.183 L/m2 h for JW , 100.83%
for TPF, 0.262 kWh/L for QSH , and 3.404 for GOR, with compos- − 0.000093𝐴2 + 0.00111𝐵2 + 0.000078𝐶 2
ite desirability of 0.3734. After trying to validate these response
values using simulation, it shows the results of 2.122 L/m2 h for + 0.000249𝐴𝐵 − 0.000034𝐴𝐶 + 0.00085𝐵𝐶
JW , 100.77% for TPF, 0.278 kWh/L for QSH , and 2.258 for GOR.
In Case 1, the response results from the response optimizer seem 𝐺𝑂𝑅 = 7.848 − 0.0054𝐴 − 0.4736𝐵 − 0.0501𝐶
to be slightly better than the simulation validation.
The same method of analysis is applied to all 12 cases. The + 0.000088𝐴2 + 0.007811𝐵2 + 0.000107𝐶 2
summary of the results of the comparison of the response optimizer
in Minitab and the simulation validation can be seen in Table 5. − 0.000047𝐴𝐵 + 0.000083𝐴𝐶 − 0.001364𝐵𝐶
If you look at it in general, the horizontal design provides better
performance than the vertical design. It shows that Case 11, the The surface plot to describe the results of the response is shown

6 / PREPRINT FOR REVIEW Transactions of the ASME


Table 4 Result data for Case 1

No. Run Flow Rate Volt Ampere Surface Temp. Bulk Temp. JW TPF QSH GOR
1 5 10 0.04 293.702 292.402 2.109 100.44% 0.161 3.893
2 25 10 0.04 292.350 291.521 2.101 100.28% 0.162 3.879
3 5 30 0.09 308.376 299.603 2.267 102.93% 1.011 0.620
4 25 30 0.09 299.252 293.652 2.152 101.91% 1.065 0.588
5 15 20 0.07 296.188 292.941 2.126 101.11% 0.519 1.207
6 15 20 0.07 296.188 292.941 2.126 101.11% 0.519 1.207
7 15 20 0.07 296.188 292.941 2.126 101.11% 0.519 1.207
8 0.86 20 0.07 312.614 307.202 2.345 101.76% 0.471 1.332
9 29.14 20 0.07 294.721 292.205 2.116 100.86% 0.522 1.202
10 15 5.86 0.02 291.671 291.335 2.098 100.12% 0.054 11.531
11 15 34.14 0.10 304.435 295.874 2.209 102.89% 1.317 0.476
12 15 20 0.07 296.188 292.941 2.126 101.11% 0.519 1.207
13 15 20 0.07 296.188 292.941 2.126 101.11% 0.519 1.207
14 15 20 0.07 296.188 292.941 2.126 101.11% 0.519 1.207

Table 5 Comparison of the analysis result of the response optimizer in Minitab for each design case

Flow Composite Response Optimizer Re-simulation Validation


Case Voltage Length Width
Rate Desirability JW TPF QSH GOR JW TPF QSH GOR
1 5 13.84 - - 0.3734 2.183 100.83 0.262 3.404 2.122 100.77 0.278 2.258
2 5 18.08 21.16 2 0.3838 2.145 101.30 0.445 1.526 2.142 101.32 0.468 1.339
3 5 18.08 22.12 4.34 0.3835 2.149 101.40 0.469 1.439 2.146 101.43 0.494 1.269
4 5 19.09 15 2 0.3959 2.153 101.47 0.497 1.348 2.149 101.46 0.524 1.195
5 5 18.28 20 2.38 0.4141 2.160 101.57 0.473 1.429 2.154 101.50 0.500 1.253
6 5 19.09 15 2 0.4167 2.159 101.69 0.510 1.320 2.152 101.58 0.539 1.163
7 5 17.68 20 2.12 0.4417 2.170 101.55 0.457 1.500 2.163 101.47 0.483 1.297
8 5 18.08 15 2.27 0.4183 2.156 101.61 0.485 1.413 2.149 101.46 0.513 1.222
9 5 17.68 20 2 0.4700 2.190 101.60 0.464 1.469 2.181 101.55 0.495 1.265
10 5 17.88 15 2.03 0.4328 2.158 101.61 0.485 1.433 2.154 101.46 0.509 1.230
11 5 17.88 20 2 0.4796 2.203 101.66 0.489 1.386 2.194 101.61 0.525 1.195
12 5 17.27 15 2.06 0.4365 2.158 101.54 0.470 1.475 2.152 101.46 0.496 1.263

Table 6 Result data for Case 11

Flow Surface Bulk


Run Volt Length Ampere JW TPF QSH GOR
Rate Temp. Temp.
1 5 20 20 0.07 314.962 303.344 2.396 103.83% 0.560 1.118
2 5 20 35 0.07 299.646 292.025 2.155 102.61% 0.737 0.850
3 25 20 20 0.07 304.898 298.514 2.215 102.14% 0.606 1.034
4 25 20 35 0.07 294.156 291.386 2.112 100.95% 0.752 0.833
5 15 10 20 0.04 302.496 296.245 2.185 102.11% 0.189 3.315
6 15 10 35 0.04 292.359 291.281 2.101 100.37% 0.233 2.694
7 15 30 20 0.09 316.390 300.692 2.430 105.22% 1.148 0.546
8 15 30 35 0.09 299.308 292.036 2.152 102.49% 1.533 0.409
9 5 10 27.5 0.04 293.927 291.632 2.111 100.79% 0.212 2.953
10 25 10 27.5 0.04 292.102 291.233 2.100 100.30% 0.213 2.938
11 5 30 27.5 0.09 309.896 294.406 2.293 105.26% 1.319 0.475
12 25 30 27.5 0.09 297.573 291.710 2.137 102.01% 1.415 0.443
13 15 20 27.5 0.07 295.495 291.642 2.121 101.32% 0.686 0.913

in Figure 5. Based on the analysis, it can be inferred that an for GOR. The composite desirability of the optimization values in
increase in voltage and a decrease in flow rate contribute to the Case 11 is 0.4796.
optimized performance of JW and TPF. As for QSH and GOR, the
optimal performance is obtained with a smaller volt value, while From the response surface regression, the standardized effect
the flow rate value has no influence on the performance. But for of the performance response is obtained. The purpose of its use
the length value, all the performance will get better if the length is is to evaluate the comparative significance of various aspects or
as small as possible variables inside a mathematical model. Typically, it is expressed
as a ratio of the estimated coefficient to its standard error. This
The response optimizer for Case 11 gives the following optimal ratio is often referred to as a t-value. A large absolute t-value (far
parameters and response results: 17.88 V for DC voltage, 5 mL/min from zero) suggests that the factor has a significant effect on the
for the feed flow rate, 2 mm and 20 mm for the slot’s width and response variable, while a small t-value suggests that the factor’s
length, respectively, which for the performances obtained are 2.203 effect is not statistically significant. In this work, the factors that
L/m2 h for JW , 101.66% for TPF, 0.489 kWh/L for QSH , and 1.386 are statistically significant at the 5% level (t-value = 3.18) are

Journal of Thermal Science and Engineering Applications PREPRINT FOR REVIEW / 7


(a) (b)

(c) (d)

Fig. 5 Surface plot of the factors to: (a) JW (b) TPF (c) QSH (d) GOR of Case 11

Fig. 6 Bar chart of standardized effect of Case 11

(a)
depicted in the bar chart in Figure 6. The voltage occupies the
most significant factor for all performance. The other significant
factors for JW and TPF are flow rate, the correlation between flow
rate and voltage, and length (for JW only). While for QSH and
GOR, there is the length, the correlation between voltage, and the
correlation between voltage and length. The correlation between
the same factors means that the impact of this factor is not linear
but quadratic. And the correlation between different factors means
that the impact of changing one variable depends on the value of
the other variable.
The temperature contour of Case 1 and Case 11 is shown in
Figure 7. It can be seen that Case 11 has a higher temperature than
Case 1, with average surface and bulk temperatures of 30.05°C
and 25.25°C, respectively. The highest temperature obtained from
the simulation reaches 45,28°C. Meanwhile, Case 1 only reached
22.43°C and 20.17°C, with the highest temperature being only
26.28°C.

4.2 Experimental Setup and Validation. To verify the analy-


sis of the RSM, further experiments were carried out. The modules
of Case 1 and Case 11, as the most optimal designs obtained from
various considerations, were fabricated. Since this is only experi-
(b)
mental validity, which aims to prove that the simulation and RSM
analysis so far are reliable, the design table of experiments in this
Fig. 7 The temperature contour graph of the fluid domain
chapter does not refer to CCD or BBD.
geometry of: (a) Case 1 (b) Case 11
Before doing the experiment, the graphene-PVDF membrane

8 / PREPRINT FOR REVIEW Transactions of the ASME


Fig. 8 Schematic of experimental setup

oven at 50°C for 2 hours.


After the membrane is ready, continue to the experimental setup
which is schematized in Figure 8 with the following explanations:
A peristaltic pump with flow rates of 5, 15, and 25 mL/min pumps
3.5 wt% NaCl solutions kept in a feed tank at room temperature.
Additionally, two cell-body modules are ready for this experiment
by positioning membranes on either side of the two modules and
attaching a flat temperature probe to the surface of each membrane.
To measure the bulk temperature, another temperature probe is
placed on the outlet feed side. A cell-body module is for the feed
side, while the other one is for the permeate side. A DC power
source is connected to each end of the membrane, delivering power
directly under voltages of 10, 20, and 30 V. In the permeate module,
a vacuum pressure of 30 mbar is applied to suck the permeated
vapour by using a vacuum pump and bringing it into a cold trap
for condensation. While the inlet side of the permeate module can
be covered or closed. Then the permeate water is obtained and
Fig. 9 Effects of various flow rates on permeate flux in pre-
measured on a weighing balance.
vious work and present verification simulation
The simulation results of the present work were verified with
various experimental data presented by Subrahmanya et al. [18].
Figure 9 presents the comparison between the predicted model and
the experimental data of 3.5 wt% NaCl solutions using graphene to
used as the joule heater needs to be fabricated first. While the PVDF wt% ratios of 2.5:1 and 2:1, as represented by experiments
fabrication steps were carried out based on the reference paper and simulations 1 and 2, respectively, with membrane porosity of
done by Subrahmanya et al. [18], with the following explanation: 60%±2% and tortuosity of 3.22 ± 0.08. The experiments were
10 gr of PVDF powder is mixed with NMP 90 gr by stirring it done at 15 V of voltage and 1.0–2.5 mL/min of flow rate.
for 48 hours, becoming a PVDF 10 wt% solution. That PVDF 10 This figure has an average percentage error of 8% for permeate
wt% solution was then mixed with graphene 2.5 wt% to PVDF by flux under different flow rates. This may have happened because
stirring it using a homogenizer for 2 hours at 6,000 rpm. If there of an uneven distribution of permeability in membrane character-
are air bubbles inside, ultrasonication can remove them. Usually, istics such as porosity and tortuosity. The measurement error also
it takes an hour. Then, prepare a non-woven PET support layer by contributes, which leads to an unstable value of the deliverable
soaking it first with PDMS solution within 20 minutes, followed by voltage. While there may be some discrepancies, the simulation
casting the PVDF solution made before on it using a stainless-steel provides valuable insights and predictions that are useful for un-
knife with a thickness of 0.1 mm. It will be a PVDF membrane. derstanding the system. However, when compared to the findings
Once it is done, transfer it directly into a coagulation bath contain- of the experiments that Subrahmanya et al. [18] presented, these
ing distilled water. Keep it in it for 24 hours, and then dry it in the findings show a good degree of concordance.

Journal of Thermal Science and Engineering Applications PREPRINT FOR REVIEW / 9


Fig. 10 The experimental validation and re-simulation results of Case 1 and Case 11

Fig. 11 Effective heating temperature in the optimal result Fig. 12 Effective heating area of a different run scenario in
of Case 11 Case 11

The biggest challenge to doing experimental validation is the nu- membrane and the ratio of permeate flow, denoted as 𝑃𝑛𝑜𝑟 𝑚 , as ex-
merical analysis part, which in this research is to find the membrane pressed in Eq. (21). Consequently, two ideal temperatures, denoted
characteristic since there is no direct practical means to properly as T*, were identified at 29°C and 35°C in the most favourable sce-
quantify the membrane tortuosity and other membrane properties nario of Case 11, as depicted in Figure 11 below.
when utilizing the tortuosity factor (𝜏). The change in mass trans-
fer resistance was the major reason for the current tortuosity in 𝑇𝑒 𝑓 𝑓 = | 𝐴𝑛𝑜𝑟 𝑚 − 𝑃𝑛𝑜𝑟 𝑚 | (21)
these investigations. Figure 10 shows the experimental results of
Case 1 and Case 11 designs, respectively, with the porosity and For additional information, the area percentage of the effective
tortuosity ranges of 27±5 and 11.6±2.8. When it’s re-simulated heating area for the most optimal case, Case 11, and its optimal
again, the results look quite acceptable, with the same trend. The data obtained from the response optimizer are also computed and
average percentage error of the experiment and the re-simulation shown in Figure 12. The variable data for every run is taken from
is about 11%. It is assumed that this problem happened because of Table 6, and the optimal data is from Table 5. It is calculated from
uncontrollable conditions, such as a temperature range of 18-22°C the cumulative heating area of each temperature. As a result, the
which had a large enough effect on its performance. Another cause temperature of 29°C has an area percentage of 43.3%, while 35°C
is that the membrane had an uneven distribution of permeability has 27.5%.
and different materials contained in PVDF powder. However, from
this experimental validation, it is proven that a 5-slot horizontal
design case has better performance as analyzed by the Response 5 Conclusion
Surface Methodology. And it’s true that the permeate flux value is The primary objective of this research is to analyze and assess
directly proportional to volts and inversely proportional to the flow simulation data in conjunction with experimental data. The aim
rate value. This is in accordance with the surface plots in Figure 4 is to identify the optimal settings for the self-heated VMD sys-
and 5. tem, with an emphasis on achieving optimal performance metrics
such as maximizing JW , TPF, and GOR, while simultaneously de-
4.3 Effective Heating Temperature. The concept of an ef- creasing QSH . The selection of the matrix design of the factors
fective heating area refers to the membrane area that exhibits a is conducted using the Central Composite Design for the original
temperature higher than the threshold temperature T*, resulting in design and the Box-Behnken Design for the remaining design. The
the generation of a pure water flow that is considered effective. cell body is variously designed by adding 1–6 slots in a vertical
In order to ascertain the optimal heating temperature, denoted as shape and 1–5 slots in a horizontal shape. The optimization of data
𝑇𝑒 𝑓 𝑓 , it is necessary to identify the minimal disparity between the was analyzed using Response Surface Methodology from Minitab
ratio of the effective heating area, represented by 𝐴𝑛𝑜𝑟 𝑚 , of the software.

10 / PREPRINT FOR REVIEW Transactions of the ASME


Based on the results discussed previously, the following conclu- 𝑥 = independent variable
sion could be drawn: 𝑦 = objective function in Response Surface Methodology
(1) From the analysis, the horizontal shape design gives better
performance than the vertical one. Acronyms
(2) In all cases, optimal performances are obtained from the 5- BBD = box-behnken Design
slot horizontal shape, Case 11, which has the factors of 5 CCD = central composite design
mL/min for the feed flow rate, 17.88 V for the DC voltage, DC = direct current
20 mm for the slot’s length, and 2 mm for the slot’s width. DCMD = direct contact membrane distillation
The performances obtained from Minitab and re-simulation DOE = design of experiment
analysis, respectively, are 2.203 and 2.194 L/m2 h for JW , GOR = gain output ratio
101.66% and 101.61% for TPF, 0.489 and 0.525 kWh/L for MD = membrane distillation
QSH , and 1.386 and 1.195 for GOR. PVDF = Polyvinylidene fluoride
(3) The most significant factor for all the performances is the RSM = response surface methodology
voltage, followed by the length (except TPF). Then, in JW TP = temperature polarization
and TPF, there is flow rate and the correlation between flow TPF = temperature polarization factor
rate and voltage. In QSH and GOR, there is a correlation VMD = vacuum membrane distillation
between voltage and another between voltage and length.
(4) The effective temperatures obtained are 29°C and 35°C, with Greek Letters
an area percentage of 43.3% and 27.5%, respectively.
(5) The performance will be at its maximum if the flow rate and 𝛽 = estimation of parameter coefficient
the width are small. This is in line with the conclusions of 𝛿 = membrane thickness
Pangarkar et al. [35]. While the length and voltage depend 𝜖 = membrane posrosity
on the desired performance. 𝜀 = controllable input variable
(6) By using RSM, optimal parameters can be achieved. This 𝜌 𝑚 = density of membrane material
can be proven by the results of the experiment, which show 𝜌 𝑝 = density of polymer material
that Case 11 achieves better than Case 1. 𝜏 = pore tortuosity
(7) Permeate flux measurement error results in an average per- 𝜙 = heat flux
centage error of 8-11%, potentially due to uneven distribu-
tion of permeability in membrane characteristics which leads Superscripts and Subscripts
to an unstable value of the deliverable voltage. Surrounding L = lower
environmental conditions such as temperature that cannot be U = upper
regulated also contribute to errors.

Acknowledgment References
[1] Karagiannis, I. C. and Soldatos, P. G., 2008, “Water desalination cost literature:
This paper was supported by Ministry of Science and Technol- review and assessment,” Desalination, 223(1), pp. 448–456.
ogy (MOST), Taiwan (grant numbers MOST 110-2823-8-011-001, [2] Giwa, A., Dufour, V., Marzooqi, F. A., Kaabi, M. A., and Hasan, S., 2017,
110-2221-E-002-022 and 108-2221-E-011-129-MY3) and High- “Brine management methods: Recent innovations and current status,” Desalina-
Speed 3D Printing Research Center, which is a Featured Areas tion, 407, pp. 1–23.
Research Center in Higher Education Sprout Project of Ministry [3] Jones, E., Qadir, M., van Vliet, M. T., Smakhtin, V., and mu Kang, S., 2019,
“The state of desalination and brine production: A global outlook,” Science of
of Education (MOE), Taiwan. The Total Environment, 657, pp. 1343–1356.
[4] Camacho, L. M., Dumée, L., Zhang, J., de Li, J., Duke, M., Gomez, J., and
Gray, S., 2013, “Advances in Membrane Distillation for Water Desalination and
Nomenclature Purification Applications,” Water, 5(1), pp. 94–196.
𝑏 [5] Amy, G., Ghaffour, N., Li, Z., Francis, L., Linares, R. V., Missimer, T., and Lat-
𝑇 𝑓 = temperature of bulk feed temann, S., 2017, “Membrane-based seawater desalination: Present and future
𝑚 prospects,” Desalination, 401, pp. 16–21.
𝑇𝑓 = temperature of membrane surface in the feed side [6] Charisiadis, C., 2016, “An Introduction to Membrane Dis-
𝐴𝑛𝑜𝑟 𝑚 = ratio of the effective heating area tillation,” SlideShare, https://www.slideshare.net/ccharisiadis/
ℎ𝑣 = latent heat of evaporation for water an-introduction-to-membrane-distillation
𝐽𝑊 = permeate water flux [7] Li, Z., Rana, D., Matsuura, T., and Lan, C. Q., 2019, “The performance
of polyvinylidene fluoride - polytetrafluoroethylene nanocomposite distillation
𝐾𝑚 = membrane mass transfer coefficient membranes: An experimental and numerical study,” Separation and Purification
𝑃𝑓 𝑚 = water vapor partial pressure in the feed side Technology, 226, pp. 192–208.
𝑃𝑛𝑜𝑟 𝑚 = the ratio of permeate flux [8] Lee, A., Elam, J. W., and Darling, S. B., 2016, “Membrane materials
𝑃𝑣 = the downstream pressure in the permeate side for water purification: design, development, and application,” Envronmental
Sciance:Water Research Technology, 2, pp. 17–42.
𝑄 𝑖𝑛 = heating energy consumption [9] Werber, J. R., Osuji, C. O., and Elimelech, M., 2016, “Materials for next-
𝑄 𝑆𝐻 = specific heating energy generation desalination and water purification membranes,” Nature Reviews
𝑇𝑒 𝑓 𝑓 = optimal heating temperature Materials, 1(5), p. 16018.
𝐴 = cross-sectional area of membrane [10] Tibi, F., Park, S.-J., and Kim, J., 2021, “Improvement of Membrane Distilla-
𝐷 = overall/composite desirability tion Using PVDF Membrane Incorporated with TiO2 Modified by Silane and
Optimization of Fabricating Conditions,” Membranes, 11(2), p. 95.
𝑑 = individual desirability [11] Abu-Zeid, M. A. E.-R., Zhang, Y., Dong, H., Zhang, L., Chen, H.-L., and Hou,
𝑓 = function in Response Surface Methodology L., 2015, “A comprehensive review of vacuum membrane distillation technique,”
𝐼 = current Desalination, 356, pp. 1–14, State-of-the-Art Reviews in Desalination.
𝑘 = number of input variable [12] Ameen, N. A. M., Ibrahim, S. S., Alsalhy, Q. F., and Figoli, A., 2020,
“Highly Saline Water Desalination Using Direct Contact Membrane Distilla-
𝑟 = mean pore radius of membrane tion (DCMD): Experimental and Simulation Study,” Water, 12(6).
𝑅 = gas constant (8.31 J/mol.K [13] Khayet, M., Cojocaru, C., and García-Payo, C., 2007, “Application of Response
𝑇 = average temperature of membrane Surface Methodology and Experimental Design in Direct Contact Membrane
𝑇∗ = ideal temperature Distillation,” Industrial & Engineering Chemistry Research, 46(17), pp. 5673–
5685.
𝑣 = feed flow velocity [14] El-Bourawi, M., Ding, Z., Ma, R., and Khayet, M., 2006, “A framework for bet-
𝑉 = voltage ter understanding membrane distillation separation process,” Journal of Mem-
𝑤 = weight of desirability function brane Science, 285(1), pp. 4–29.

Journal of Thermal Science and Engineering Applications PREPRINT FOR REVIEW / 11


[15] Chang, H., Wang, G.-B., Chen, Y.-H., Li, C.-C., and Chang, C.-L., 2010, croscopic Workpiece Inspection and Response Surface Modeling,” Ph.D. disser-
“Modeling and optimization of a solar driven membrane distillation desalination tation, National Taiwan University of Science and Technology, Taipei, Taiwan.
system,” Renewable Energy, 35(12), pp. 2714–2722. [25] Myers, R., Montgomery, D., and Anderson-Cook, C., 2009, Response Surface
[16] Dudchenko, A. V., Chen, C., Cardenas, A., Rolf, J., and Jassby, D., 2017, Methodology: Process and Product Optimization Using Designed Experiments,
“Frequency-dependent stability of CNT Joule heaters in ionizable media and Wiley Series in Probability and Statistics, Wiley.
desalination processes,” Nature Nanotechnology, 12(6), pp. 557–563. [26] Derringer, G. and Suich, R., 1980, “Simultaneous Optimization of Several Re-
[17] Anvari, A., Kekre, K. M., Azimi Yancheshme, A., Yao, Y., and Ronen, A., 2019, sponse Variables,” Journal of Quality Technology, 12(4), pp. 214–219.
“Membrane distillation of high salinity water by induction heated thermally [27] Zhang, J., Lin, P. T., and Jaluria, Y., 2013, “Design and Optimization of Mul-
conducting membranes,” Journal of Membrane Science, 589, p. 117253. tiple Microchannel Heat Transfer Systems,” Journal of Thermal Science and
[18] T. M., S., Lin, P. T., Chiao, Y.-H., Widakdo, J., Chuang, C.-H., Rahmadhanty, Engineering Applications, 6(1), p. 011004.
S. F., Yoshikawa, S., and Hung, W.-S., 2021, “High performance self-heated [28] Consultants, P., “DESIGN OF EXPERIMENT,” PT PROQUAMAN KONSUL-
membrane distillation system for energy efficient desalination process,” Journal TAN, https://pqm.co.id/area-of-expertise/design-of-experiment/
of Materials Chemistry A, 9, pp. 7868–7880. [29] Plastics, P., 2010, “Thermal Properties of Plastic Materials,” Pro-
fessional Plastics, https://www.professionalplastics.com/professionalplastics/
[19] Boubakri, A., Hafiane, A., and Bouguecha, S. A. T., 2014, “Application of
ThermalPropertiesofPlasticMaterials.pdf
response surface methodology for modeling and optimization of membrane dis-
[30] FLUOROTHERM, “PVDF Properties,” FLUOROTHERM, New Jersey,
tillation desalination process,” Journal of Industrial and Engineering Chemistry,
USA, https://www.fluorotherm.com/technical-information/materials-overview/
20(5), pp. 3163–3169.
pvdf-properties/
[20] Ali, A., Tsai, J.-H., Tung, K.-L., Drioli, E., and Macedonio, F., 2018, “Designing
[31] Properties, M., “Graphene – Material Table – Applications
and optimization of continuous direct contact membrane distillation process,”
– Price,” Material Properties, https://material-properties.org/
Desalination, 426, pp. 97–107.
graphene-properties-application-price/
[21] Eykens, L., De Sitter, K., Dotremont, C., Pinoy, L., and Van der Bruggen, B., [32] GmbH, D., “Thermal Conductivity of N-Methyl-2-pyrrolidone,” DDBST
2016, “How To Optimize the Membrane Properties for Membrane Distillation: GmbH, http://www.ddbst.com/en/EED/PCP/TCN_C284.php
A Review,” Industrial & Engineering Chemistry Research, 55(35), pp. 9333– [33] Sang, M., Shin, J., Kim, K., and Yu, K. J., 2019, “Electronic and Thermal
9343. Properties of Graphene and Recent Advances in Graphene Based Electronics
[22] Mohammadi, T. and Safavi, M. A., 2009, “Application of Taguchi method in Applications,” Nanomaterials, 9(3).
optimization of desalination by vacuum membrane distillation,” Desalination, [34] Sobieski, W. and Lipiński, S., 2017, “The analysis of the relations between
249(1), pp. 83–89. porosity and tortuosity in granular beds,” Technical Sciences, 20, pp. 75–85.
[23] Yang, C., Peng, X., Zhao, Y., Wang, X., Fu, J., Liu, K., Li, Y., and Li, P., 2020, [35] Pangarkar, B., Thorat, P., Parjane, S., and Abhang, R., 2010, “Performance
“Prediction model to analyze the performance of VMD desalination process,” evaluation of vacuum membrane distillation for desalination by using a flat sheet
Computers Chemical Engineering, 132, p. 106619. membrane,” Desalination and Water Treatment - DESALIN WATER TREAT,
[24] Izdihar, U. F., 2021, “Optimization of Turning Process Parameters Based in Mi- 21, pp. 328–334.

12 / PREPRINT FOR REVIEW Transactions of the ASME


List of Figures
1 The self-heated VMD module assembly . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
2 Flow chart of methodology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
3 The new design: (a) vertical (b) horizontal shape . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
4 Surface plot of voltage and flow rate to: (a) JW (b) TPF (c) QSH (d) GOR of Case 1 . . . . . . . . . . . . . . . . . . . 6
5 Surface plot of the factors to: (a) JW (b) TPF (c) QSH (d) GOR of Case 11 . . . . . . . . . . . . . . . . . . . . . . . . . 8
6 Bar chart of standardized effect of Case 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
7 The temperature contour graph of the fluid domain geometry of: (a) Case 1 (b) Case 11 . . . . . . . . . . . . . . . . . . 8
8 Schematic of experimental setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
9 Effects of various flow rates on permeate flux in previous work and present verification simulation . . . . . . . . . . . . 9
10 The experimental validation and re-simulation results of Case 1 and Case 11 . . . . . . . . . . . . . . . . . . . . . . . . 10
11 Effective heating temperature in the optimal result of Case 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
12 Effective heating area of a different run scenario in Case 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

List of Tables
1 The factors and values . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
2 Design of experiment of CCD and BBD . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3 Material properties used in the Ansys simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
4 Result data for Case 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
5 Comparison of the analysis result of the response optimizer in Minitab for each design case . . . . . . . . . . . . . . . . 7
6 Result data for Case 11 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

Journal of Thermal Science and Engineering Applications PREPRINT FOR REVIEW / 13

You might also like