This paper presents the results of a techno-economic investigation of a nanofiltration (NF) and reverse osmosis (RO) process for treating brackish water. Optimization experiments of six commercially available small scale RO and NF membranes were carried out using formulated artificial groundwater. A predictive model was developed by using response surface methodology (RSM) for optimization of input process parameters of brackish water membrane processes to simultaneously maximize water recovery and salt rejection while minimizing energy demand. A predictive model using multiple response optimization revealed that CSM RO and NF250 membranes showed the optimal efficiency with 20.24% and 18.98% water recovery, 90.22% and 70.64% salt rejection and 17.87 kWh/m3 and 9.35 kWh/m3 of specific energy consumption (SEC), respectively. Furthermore, confirmation of RSM predictions was carried out by an artificial neural network (ANN) model trained by RSM experimental data. Predicted values by both RSM and ANN modeling methodologies were compared and found within the acceptable range. Finally, a membrane validation experiment was carried out successfully at proposed optimal conditions, which proved the accuracy of the employed RSM and ANN models. Detailed analyses of the economic assessment showed that the recovery rate can play a major role in reducing the cost of a membrane system.

## INTRODUCTION

Pollution and overexploitation of groundwater are deteriorating its quality day-by-day in many parts of India (Chakraborti et al. 2010; Saha et al. 2013; Sharma & Gulati 2014). Paliwal (1974) reported high saline quality of groundwater in India with respect to electrical conductivity, influenced by a high groundwater table as a result of seepage from canals and poor drainage system.

Reverse osmosis (RO) membrane desalination systems are used all over the world for production of water by treating seawater and brackish water (Pangarkar et al. 2011). These systems prove to be an efficient and reliable solution to combat the water scarcity for our daily needs viz. drinking water and other different household applications (Pearce & Kumar 2003). However, the use of RO systems for treatment of brackish groundwater is very limited in India. Groundwater RO process also have some problems of inorganic fouling (Choi et al. 2009; Pangarkar et al. 2011) which can lead to limited recovery of water with time.

Nanofiltration (NF) is a membrane liquid separation technology which is positioned between RO and ultrafiltration. NF is a lower-pressure version of RO. Another feature of the NF membrane is that very high recovery is possible in some applications. Therefore, NF is an excellent balance of removal and selectivity.

The specific objectives of the present study were: (1) to optimize the performance of a laboratory-scale NF and RO membranes by maximizing water recovery, salt rejection and minimization of specific energy consumption (SEC); and (2) to provide detailed analyses of the economic assessment to investigate the factors that play a major role in reducing the cost of a membrane system.

Owing to the appreciable contribution of response surface methodology (RSM), this system is usually employed for experimental design (Razali et al. 2013). Furthermore, validation experiment of these results was conducted as per the optimized value of RSM prediction. Previous investigations highlighted the relevance of artificial neural network (ANN) models to approve the predictions recommended by RSM (Bhatti et al. 2011; Singh et al. 2012; Nourbakhsh et al. 2014). Finally, the validation of optimized process conditions suggested by the RSM was also carried out by the generated ANN model in MATLAB. The predictive models given by RSM and ANN were compared for their predictive and generalization abilities and the optimum conditions predicted by both the approaches were experimentally verified. Life cycle assessment of the full process was carried out to estimate per unit water production cost.

## METHODS

### System configuration and operational procedure

A laboratory membrane system was used to perform membrane experiments. The schematic diagram of the membrane system is in Figure 1. A small scale membrane system contains of a feed tank and high pressure pump along with membrane module which comprises of spiral wound NF or RO membranes (2 inch diameter and 12 inch length). For continuous evaluation of the system, permeate and concentrate streams were recycled in the feed tank. To regulate feed water temperature, automatic temperature control unit was used. Six thin film composite RO and NF (MWCO 100, 250 and 400 Da) membranes from four leading manufacturing companies (CSM, Dow, Vontron and Permionics) were used for experiments.

Figure 1

Schematic of NF/RO membrane experimental setup; 1 = high pressure pump, 2 = bypass valve, 3 = temperature probe, 4 = temperature controller, 5 = NF/RO module, 6 = concentrate stream, 7 = feed water tank, 8 = permeate stream, 9 and 10 = valves.

Figure 1

Schematic of NF/RO membrane experimental setup; 1 = high pressure pump, 2 = bypass valve, 3 = temperature probe, 4 = temperature controller, 5 = NF/RO module, 6 = concentrate stream, 7 = feed water tank, 8 = permeate stream, 9 and 10 = valves.

The membrane filtration unit was fed with artificial water which was formulated on the basis of major ionic elements of actual groundwater collected from Timarpur, Delhi (India) (Garg & Joshi 2014a).

### Central composite design experimental design and optimization by RSM

To minimize experimental cost and reduce the number of experiments, RSM is generally employed (Razali et al. 2013). RSM method using central composite design (CCD) was employed to investigate the mutual effect of various factors on the performance of membrane process. Design Expert software (Stat-Ease, Inc., Minneapolis, MN) was used for experimental design (Khayet et al. 2011). Experiments were carried out with different ranges of temperature, pressure, concentration and pH of feed water for different NF and RO membranes (Garg & Joshi 2014b).

After optimization of membranes by the RSM software validation experiments were performed employing the optimal values of the input parameters and the results verified against the predicted values of the RSM.

### ANN modelling

Neural network tool of MATLAB was used to train the ANN model by using the data generated from RSM experimental design. The designed ANN (4-5-3) has four neurons in the first (input) layer, five neurons in the second (hidden) layer, and three neurons in the third (output) layer (Garg & Joshi 2014a). The designed ANN model (4-5-3) with a log sigmoid transfer function (logsis) chosen for the input to hidden layer mapping, and a linear transfer function (purelin) for the hidden to output layer mapping was trained using the experimental data.

The training phase was completed when it yielded a negligible mean square error across all training experiments. Once ANN had been trained, it had a good predictive capability and ability to accurately describe the RS even without any knowledge of the physical and chemical background of the modeled system (Bezerra et al. 2008; Geyikçi et al. 2012).

Finally, the predictive models given by RSM and ANN were compared for their predictive and generalization abilities and the optimum conditions predicted by both the approaches were experimentally verified.

### Economic aspects of membrane system

Economic analysis of any device is essential for commercialization encouragement. Per cubic meter water production cost is affected by various factors including water recovery and feed flow rate. Detailed analysis of economic assessment was carried out.

## RESULTS AND DISCUSSION

### RSM analysis

The experimental data were analyzed using regression analysis and for the development of a predictive model. The predictive model revealed that CSM RO among RO membranes and NF250 among NF membranes showed optimum efficiency.

Figure 2(a) shows that the increase of the pressure from 0.59 to 0.79 MPa leads to an increase of the water recovery for CSM membrane. A similar trend was observed with increase in pressure from 0.49 to 1.08 MPa for NF250 membrane (Figure 3(a)). The water recovery increases because higher pressure allows enhanced flow of water through the membrane (Koyuncu et al. 2001; Sassi & Mujtaba 2010). The effect of concentration was found insignificant for water recovery by CSM membrane (Figure 2(b)). However, decreases in concentration from 1,500 to 3,500 mg/l led to a decrease of recovery for NF250 membrane (Figure 3(b)). Lower water recovery is because of higher salt concentration, causing the negative effect of concentration polarization and decreasing membrane water flux (Koyuncu et al. 2001).

Figure 2

Plots showing the effect of different input variables on recovery (a) and (b), rejection (c) and (d), and SEC (e) and (f) of CSM RO membranes.

Figure 2

Plots showing the effect of different input variables on recovery (a) and (b), rejection (c) and (d), and SEC (e) and (f) of CSM RO membranes.

Figure 3

Plots showing the effect of different input variables on recovery (a) and (b), rejection (c) and (d), and SEC (e) and (f) of NF250 membranes.

Figure 3

Plots showing the effect of different input variables on recovery (a) and (b), rejection (c) and (d), and SEC (e) and (f) of NF250 membranes.

At high and low values of temperature, salt rejection did not show any significant change with pressure for either CSM. This is because higher diffusion rate of solute through the membrane is possible as the solubility increases with temperature (Arora et al. 2004; Gedam et al. 2012). The effect of pressure was found insignificant for salt rejection by CSM membrane (Figure 2(c)). Salt rejection increases with the pressure for NF250 membrane (Figure 3(c)). Salt rejection decreases with feed water concentration ranging from 1,500 to 3,500 mg/l for both CSM RO and NF250 membranes (Figures 2(d) and 3(d)). This is because at high feed salinity, salt passage increases (Bartels et al. 2005).

It was observed that the increase of the pressure diminishes SEC for both CSM and NF250 membranes (Figures 2(e) and 3(e)). Increasing concentration results in enhancement of the SEC for both CSM RO and NF250 membranes (Figures 2(f) and 3(f)).

### Multiple response optimization

Membrane optimization was carried out by the RSM through regression analysis to achieve maximum recovery, highest salt rejection and lowest SEC. Predicted numerical optimization of input parameters was obtained and is presented in Table 1.

Table 1

Experimental validation of RSM/ANN predictions

RO membrane
NF membrane
ParametersCSMDowVontronNF100NF250NF400
Optimized input parameters
Temperature (°C) 31.92 32 30 30 30 29.58
Pressure (MPa) 0.79 1.43 1.178 0.86 1.08 0.59
Conc. (mg/l) 1,500 1,500 1,500.8 1,500.1 1,505 1,501.1
pH 6.53 6.14 6.73 7.15
RSM predictions
Recovery (%) 19.25 16.1 12.75 12.35 18.98 14.76
Rejection (%) 89.2 82.53 89.66 73.43 70.64 51.38
SEC (KWh/m317.6 18.99 19.02 13.22 9.35 12.11
ANN predictions
Recovery (%) 19.51 16.16 12.48 12.28 18.59 14.86
Rejection (%) 88.92 82.95 88.44 73.12 71.4 51.46
SEC (KWh/m316.60 18.8 18.79 13.39 9.43 11.63
Validation experiment at optimized conditions
Recovery (%) 19.69 16.87 13.54 12.16 18.37 15.3
Rejection (%) 89.98 83.64 90.77 74.72 71.04 53.74
SEC (KWh/m316.53 20.2 20.12 13.16 9.07 11.56
RO membrane
NF membrane
ParametersCSMDowVontronNF100NF250NF400
Optimized input parameters
Temperature (°C) 31.92 32 30 30 30 29.58
Pressure (MPa) 0.79 1.43 1.178 0.86 1.08 0.59
Conc. (mg/l) 1,500 1,500 1,500.8 1,500.1 1,505 1,501.1
pH 6.53 6.14 6.73 7.15
RSM predictions
Recovery (%) 19.25 16.1 12.75 12.35 18.98 14.76
Rejection (%) 89.2 82.53 89.66 73.43 70.64 51.38
SEC (KWh/m317.6 18.99 19.02 13.22 9.35 12.11
ANN predictions
Recovery (%) 19.51 16.16 12.48 12.28 18.59 14.86
Rejection (%) 88.92 82.95 88.44 73.12 71.4 51.46
SEC (KWh/m316.60 18.8 18.79 13.39 9.43 11.63
Validation experiment at optimized conditions
Recovery (%) 19.69 16.87 13.54 12.16 18.37 15.3
Rejection (%) 89.98 83.64 90.77 74.72 71.04 53.74
SEC (KWh/m316.53 20.2 20.12 13.16 9.07 11.56

It can be predicted by the above analysis that CSM membrane (among RO membranes) showed the best performance with 19.25% water recovery, 89.2% salt rejection and 17.6 kWh/m3 of SEC at 31.92 °C temperature, 0.79 MPa pressure, 1,500 mg/l feed salt concentration and 6.53 pH (Table 1). Also, NF250 showed the best performance (among NF membranes) at 30 °C temperature, 1.08 MPa feed pressure, 1,500 mg/l concentration and pH 7.15 with 18.98% water recovery, 70.64% salt rejection and 9.35 kWh/m3 of SEC.

Model predictions validated by confirmation run at these optimal process conditions are in agreement with the predicted responses. Less than 6% error for each response showed the reliability of the CCD optimization process. These results demonstrate an improvement in the individual RO and NF membranes performance at optimized input parameters along with the best performed membrane compared with the other RO and NF membranes.

### ANN predictions

ANN model was used to validate the RSM predicted optimized process conditions. ANN predicted 19.51% and 18.59% of water recovery, 88.92% and 71.4% of total dissolved solids (TDS) rejection and 16.60 kWh/m3 and 9.43 kWh/m3 of SEC for CSM and NF250 membranes, respectively, at optimal process conditions (Table 1). The predicted values of both RSM and ANN model are nearer to experimental values.

### Economic assessment

For estimation of water production cost, common technical assumptions, specifications and design parameters were considered for RO and NF membrane systems (Table 2). The following items were included:

• (1) Capital cost expenditure.

• (2) Operation and maintenance cost.

• (3) Other costs and operating expenditures.

Table 2

Design parameters of membrane system, calculation of capital and per unit water production cost (US$) ParametersNFRO Design parameters Recovery (%) 19.92 20.05 Feed flow rate (LPH) 435 171 Hours of operation/day Water production (liter/day) 519.91 205.71 Annual product volume (m3/year), p 189.77 75.09 Membrane life (years) Interest rate (%) RO plant availability (%), f 90 90 NF/RO membrane cost (US$) 80 80
Capital cost (US$) High pressure pump (with 1 HP motor) 240 240 High pressure connecting pipes 80 80 Membrane housing 40 40 NF/RO membranes 80 80 Feed/permeate tank 32 32 Temperature control unit 32 32 Membrane system installation cost 80 80 Total cost of membrane system (US$) 504 504
Membrane replacement cost per year, A40 40
Annual amortized capital cost, A35.76 35.76
O&M annual cost of membrane system, A3 = A2 × 0.2 7.15 7.15
Annual operating cost C = (A1 + A2 + A3) 82.91 82.91
Unit production cost = C/(f × p), US$/m3 0.49 1.23 ParametersNFRO Design parameters Recovery (%) 19.92 20.05 Feed flow rate (LPH) 435 171 Hours of operation/day Water production (liter/day) 519.91 205.71 Annual product volume (m3/year), p 189.77 75.09 Membrane life (years) Interest rate (%) RO plant availability (%), f 90 90 NF/RO membrane cost (US$) 80 80
Capital cost (US$) High pressure pump (with 1 HP motor) 240 240 High pressure connecting pipes 80 80 Membrane housing 40 40 NF/RO membranes 80 80 Feed/permeate tank 32 32 Temperature control unit 32 32 Membrane system installation cost 80 80 Total cost of membrane system (US$) 504 504
Membrane replacement cost per year, A40 40
Annual amortized capital cost, A35.76 35.76
O&M annual cost of membrane system, A3 = A2 × 0.2 7.15 7.15
Annual operating cost C = (A1 + A2 + A3) 82.91 82.91

## ACKNOWLEDGEMENT

This work was carried out with the financial support provided by the Ministry of Drinking Water and Sanitation, New Delhi (India).

## REFERENCES

REFERENCES
Arora
M.
Maheshwari
R. C.
Jain
S. K.
Gupta
A.
2004
.
Desalination
170
(
2
),
105
112
.
Bartels
C.
Franks
R.
Rybar
S.
Schierach
M.
Wilf
M.
2005
.
Desalination
184
(
1–3
),
185
195
.
Bezerra
M. A.
Santelli
R. E.
Oliveira
E. P.
Villar
L. S.
Escaleira
L. A.
2008
.
Talanta
76
(
5
),
965
977
.
Bhatti
M. S.
Reddy
A. S.
Kalia
R. K.
Thukral
A. K.
2011
.
Desalination
269
(
1–3
),
157
162
.
Chakraborti
D.
Das
B.
Murrill
M. T.
2010
.
Environmental Science & Technology
45
(
1
),
27
33
.
Choi
Y. H.
Kweon
J. H.
Kim
D. I.
Lee
S.
2009
.
Desalination
247
(
1–3
),
137
147
.
Garg
M. C.
Joshi
H.
2014a
.
Environmental Technology
35
(
23
),
2988
2999
.
Garg
M. C.
Joshi
H.
2014b
.
Desalination
353
,
57
74
.
Gedam
V. V.
Patil
J. L.
Kagne
S.
Sirsam
R. S.
Labhasetwar
P.
2012
.
Journal of Membrane Science & Technology
2
(
3
),
2
5
.
Geyikçi
F.
Kılıç
E.
Çoruh
S.
Elevli
S.
2012
.
Chemical Engineering Journal
183
,
53
59
.
Khayet
M.
Cojocaru
C.
Essalhi
M.
2011
.
Journal of Membrane Science
368
(
1–2
),
202
214
.
Koyuncu
I.
Yazgan
M.
Topacik
D.
Sarikaya
H. Z.
2001
Evaluation of the low pressure RO and NF membranes for an alternative treatment of Buyukcekmece Lake
.
Water Science and Technology: Water Supply
1
(
1
),
107
115
.
Nourbakhsh
H.
Emam-Djomeh
Z.
Omid
M.
Mirsaeedghazi
H.
Moini
S.
2014
.
Computers and Electronics in Agriculture
102
,
1
9
.
Paliwal
K. V.
1974
Quality of water resources in India
. In:
Proceedings of the Indian National Science Academy, Part A: Physical Sciences
, pp.
331
341
.
Pangarkar
B. L.
Sane
M. G.
M.
2011
.
ISRN Materials Science
2011
,
9
.
Pearce
W.
Kumar
M.
2003
Evaluation of new generation brackish water reverse osmosis membrane
. In:
AWWA Membrane Technology Conference
,
American Water Works Association
,
Atlanta, Georgia
, pp.
1
8
.
Razali
N. F.
A. W.
Hilal
N.
Leo
C. P.
Alam
J.
2013
.
Desalination
311
,
182
191
.
Saha
D.
Dwivedi
S. N.
Singh
R.
2013
Aquifer system response to intensive pumping in urban areas of the Gangetic plains, India: the case study of Patna
.
Environmental Earth Sciences
71
,
1721
1735
.
Sassi
K. M.
Mujtaba
I. M.
2010
Simulation and optimization of full scale reverse osmosis desalination plant
. In:
Computer Aided Chemical Engineering
(
Pierucci
S.
Ferraris
G. B.
, eds).
Elsevier
, pp.
895
900
.
Sharma
R. K.
Gulati
S.
2014
1.3 – Water quality issues and solutions in India
. In:
Comprehensive Water Quality and Purification
(
Satinder
A.
ed.
, ed.).
Elsevier
,
Waltham
, pp.
21
39
.
Shatat
M.
Worall
M.
Riffat
S.
2013
.
Sustainable Cities and Society
9
(
0
),
67
80
.
Singh
K.
Singh
A.
Gupta
S.
Rai
P.
2012
.
Environmental Science and Pollution Research
19
(
6
),
2063
2078
.