Separation and Purification Technology 174 (2017) 44–56

Contents lists available at ScienceDirect

Separation and Purification Technology

journal homepage: www.elsevier.com/locate/seppur

Comprehensive bench- and pilot-scale investigation of NF for gold

mining effluent treatment: Membrane performance and fouling control
strategies
L.H. Andrade ⇑, A.O. Aguiar, W.L. Pires, L.B. Grossi, M.C.S. Amaral
Department of Sanitary and Environmental Engineering, Federal University of Minas Gerais, Av. Antônio Carlos, n° 6627, Pampulha, Belo Horizonte, Minas Gerais, Brazil

a r t i c l e i n f o a b s t r a c t

Article history: Despite being widely used, large differences in filtration-time and hydrodynamic conditions between
Received 21 May 2016 bench- and large-scale membrane units make the scaling up of bench-scale results an issue. On the other
Received in revised form 3 September 2016 hand, pilot tests allow obtaining more reliable information on hydrodynamic conditions, fouling rates,
Accepted 3 September 2016
and cleaning methods. Hence, the aim of this study was to evaluate nanofiltration (NF) membrane fouling
Available online 30 September 2016
control strategies in an integrated ultrafiltration (UF)-NF pilot plant applied to the treatment of gold min-
ing effluent. Increasing the pressure increased the permeate flux as well as the concentration polarization
Keywords:
and did not improve the membrane performance. The development of mass transfer condition by increas-
Nanofiltration
Pilot-plant
ing the feed cross-flow velocity effectively reduced the concentration polarization and fouling, allowing
Hydrodynamic conditions the achievement of higher permeate flux and removal efficiency. The addition of antiscalant was impor-
Antiscalant tant for the system studied, for reducing the flux decay up to 36%. The difference between some results in
Membrane cleaning bench- and pilot-scale underscores the importance of evaluating design parameters using pilot-scale
units. The cost of the proposed treatment was estimated at US$ 1.34/m3.
Ó 2016 Elsevier B.V. All rights reserved.

1. Introduction Membrane technology is attractive for treating mining wastew-

aters [4]. Among membrane processes, nanofiltration (NF) mem-
Gold mining is an industrial sector of significant economic brane can efficiently retain salts and metals from aqueous
importance. However, it is also responsible for many adverse medium [5,6]. Therefore, it shows a high potential to treat mining
impacts on water resources such as water cycle changes and dete- effluents. Although some authors have studied the NF of mining
rioration of the water quality of receiving water bodies [1]. The effluents [7–12], the literature is still small and does not reflect
effluents from ore processing have low pH and high concentrations the full potential of this process. It lacks advanced studies of
of heavy metals and metalloids, such as Cd, Cr, Hg, and As, since the application of membrane separation processes for mining
these elements are often associated with valuable components in effluents; especially, studies aimed at optimizing the operational
ores and concentrates [2,3]. conditions and ensuring a viable process in both technical and
Nowadays, many countries are facing a significant water scar- economic terms.
city problem. Over the last decade, the acceptance of water recla- Fouling is a major limitation of NF. Membrane fouling increases
mation and reuse has increased, which is an effective approach the chemical cleaning requirement and ultimately results in water
for reducing fresh water consumption and wastewater discharge. production loss, membrane integrity loss, poorer water quality and
Moreover, because of the high consumption of fresh water and shorter membrane lifetime; which consequently impose a heavy
high effluent generation, mining is one of the industrial sectors economic weight on the plant operation [13]. Fouling is related
with the highest potential for water reuse [4]. However, further to operational conditions and feed stream composition [14]. Both,
studies focused on the application of advanced treatment tech- fouling tendency and NF permeate quality, are strongly influenced
nologies on mining effluents, aimed at meeting the quality require- by operational conditions such as feed pre-treatment [15], feed pH
ments for reclaimed water, are needed. [16–19], concentration of solutes on the feed solution [16,20],
permeate recovery rate [21,22], temperature [22–25], and
⇑ Corresponding author. hydrodynamic conditions imposed by different transmembrane
E-mail address: lauraha@ymail.com (L.H. Andrade). pressures [19,22,26] and by different cross-flow velocities [27,28].

http://dx.doi.org/10.1016/j.seppur.2016.09.048
1383-5866/Ó 2016 Elsevier B.V. All rights reserved.
 L.H. Andrade et al. / Separation and Purification Technology 174 (2017) 44–56 45

Antiscalant dosage is often necessary in membrane separation influence it, and the monitoring of process performance and per-
process when the effluent has significant concentration of cations meate flux over long periods of time. So these pilot experiments
and anions of sparingly soluble salts because of the inherent provide more reliable information about fouling rates, cleaning
increase in concentrate concentration [29]. Some sparingly soluble methods and long-term membrane stability [42].
salts reported in the literature include calcium sulfate, -carbonate Consequently, the aim of this study was to evaluate different NF
and -fluoride; barium sulfate; and strontium sulfate [30]. Com- membrane fouling control strategies in an integrated UF-NF pilot
monly used antiscalants in membrane separation processes plant applied to the treatment of gold mining effluent. This study
include polyacrylic acid, polyacrylamide, polymaleic anhydride, focused on the factors that influence the hydrodynamic flow con-
and polyphosphates [31]. Antiscalant is used to convert the metal ditions (operational pressure and feed flow rate in the module),
involved in the formation of the salt to a complex, or to inhibit the the evaluation of antiscalant application, and the membrane clean-
crystal growth [32]. Many mechanisms have been proposed for ing efficiency by physical cleaning (forward flush) and chemical
antiscalants on crystallization. Rahardianto et al. [33] suggested cleaning.
that antiscalants can adsorb onto newly formed crystal facets,
causing defects and thereby retarding or halting further crystal
2. Materials and methods
growth or weakening the mechanical strength of the scale. It has
also been hypothesized that antiscalants increase the surface
2.1. Gold ore processing effluents
energy of crystal nuclei, effectively slowing the rate of crystal
nucleation [34,35] and/or that antiscalants cause a negative elec-
Two effluents from a gold mining factory in Brazil were studied,
trostatic potential on the antiscalant-modified scale surface, which
i.e., the effluent from the sulfuric acid production plant and the
can prevent not only the agglomeration of scale nuclei in the con-
water from the calcined dam (Table 1). The industrial process
centrate but also the precipitation of scale nuclei on the membrane
description and the effluent generation points are as described in
surface [36].
Andrade et al. [43].
Even with the optimization of operational conditions and antis-
Preliminary NF experiments have shown that membrane pro-
calant dosage, fouling can develop on the membrane surface or
cess performance was improved when applied to these two efflu-
inside the membrane pores. Therefore, membrane cleaning meth-
ents in combination instead of to a single effluent. Thus, both
ods aimed at fouling removal are mandatory for membrane separa-
effluents were mixed at a 1:1 ratio (corresponding to actual
tion processes [37]. Membrane cleaning methods are classified into
wastewater flow rates generated by the company, which are
physical, chemical and physico-chemical. Chemical cleaning is the
140 m3/h of effluent from the sulfuric acid production plant and
most studied membrane cleaning method, especially in NF and
140 m3/h of water from the calcined dam) before being transferred
reverse osmosis (RO) systems. Although physical cleaning methods
to the membrane treatment system. In this study, such a mixture
can be economically attractive [38], they have not been given
will hereinafter be referred to as ‘‘effluent from gold mining”.
much attention. As a result, more investigation is needed in order
to evaluate their effectiveness.
The most frequently used physical cleaning method for micro- 2.2. Description of the treatment process
filtration (MF) and ultrafiltration (UF) is backwashing. However,
the pressure necessary to perform backwashing on NF and RO The evaluated treatment route for the mixed effluent consisted
membranes with sufficient flux would be very high because of of three stages, namely, adjustment of pH to 5.0 with NaOH solu-
the high resistance of the membrane. As this could loosen the tion, ultrafiltration, and nanofiltration, as shown in Fig. 1. This pro-
membrane layer from its support layer, backwashing is not com- cess was evaluated in bench- and pilot-scale units.
monly used in these processes [39]. Therefore, other physical
cleaning methods should be used for NF and RO, which include, 2.3. Membranes
for example, hydrodynamic forward or reverse flush, air spurge
and automatic sponge ball cleaning [38]. Among these, forward Ultrafiltration (UF) was performed with a commercial sub-
flux is of special interest as no substantial changes in the NF/RO merged membrane module (ZeeWeed), PVDF-based polymer with
unit are needed for its application. During forward flush, the
Table 1
cross-flow velocity in the feed stream is maintained while the Characteristics of gold ore processing effluent.
transmembrane pressure is removed. This stops the permeate flux
and reduces the concentration polarization. Parameters Effluent from the Water from the Effluent from
sulfuric acid calcined dam gold mininga
The best operational conditions, antiscalant and cleaning pro- production plant
cesses must be carefully selected for each specific application, to
pH 1.82 8.23 2.08
achieve the best overall performance. This line of research has
Conductivity (lS/cm) 9420 3250 6615
received the attention of researchers throughout the world. How- Dissolved organic carbon 4.4 5.8 5.0
ever, these parameters are usually evaluated only through bench- (mg/L)
scale tests. Bench-scale tests use a flat sheet membrane cell with Total solids (mg/L) 9398 3159 8804
Suspended solids (mg/L) 25 14 15
a small area in the order of tens of cm2. Bench-scale units are prac-
Sulfate (mg/L) 6367 1944 3864
tical and inexpensive, and require only a small volume of effluent Chloride (mg/L) 31 291 168
for operation. While these studies are useful to improve our funda- Aluminum (mg/L) 121 0.1 67
mental knowledge of the membrane process, the scaling-up of Arsenic (mg/L) 657 2 322
their results remains an issue due to differences in filtration-time Calcium (mg/L) 382 541 550
Iron (mg/L) 75 0.2 34
and hydraulic conditions in bench-scale membrane cell and actual
Magnesium (mg/L) 462 29 247
spiral wound modules [40]. As a result, although some important Manganese (mg/L) 28 0.1 22
conclusions of the NF performance can be obtained from bench Potassium (mg/L) 49 52 44
scale tests, these results cannot be directly applied to long term Sodium (mg/L) 45 168 77
fouling of large-scale systems [41]. Moreover, some operational Zinc (mg/L) 115 0.1 64

conditions can only be evaluated in pilot systems. Examples a

Mixture at a 1:1 ratio of effluent from the sulfuric acid production plant and of
include the hydrodynamic flow conditions and the factors that the water from the calcined dam.
46 L.H. Andrade et al. / Separation and Purification Technology 174 (2017) 44–56

Fig. 1. Scheme of the proposed treatment system for the mixed effluent.

average pore diameter of 0.04 lm. Nanofiltration (NF) was per- 5.0 with NaOH 10%. Then, a centrifugal pump with a maximum
formed with a Filmtec NF90 membrane, which is a thin film com- flow of 40 L/h sent the effluent to a submerged UF tank with a vol-
posite membrane comprised of three layers: (1) a polyester ume of 77 L. The submerged UF module has a membrane area of
support web, (2) a microporous polysulfone inter layer and (3) 0.9 m2. The UF permeate was connected to a vacuum tank with a
an ultrathin aromatic polyamide active layer. The molecular volume of 20 L. Vacuum was generated by a vacuum diaphragm
weight cut-off is 100 Da [44], the average membrane hydraulic pump with a maximum flow of 60 L/h. The pressure was measured
resistance is 5.8
1013/m, and NaCl (2000 mg/L) and MgSO4 by a manometer and adjusted by a needle-type valve. The perme-
(2000 mg/L) rejection are 85–95% and 97% respectively [45]. ate flux was monitored by measuring the time required to fill com-
pletely the vacuum tank. The UF concentrate was manually
2.4. Experimental setup removed daily.
Once the vacuum tank was full, the UF stopped and a solenoid
Tests were conducted in bench- and pilot-scale units. Both test valve opened to discharge the UF permeate to the NF supply tank.
conditions are described below. The NF supply tank had a volume of 450 L. A second diaphragm
pump equipped with a speed controller and a maximum flow of
2.4.1. Bench-scale integrated UF-NF 324 L/h was used for the NF system. A manometer and a needle-
Fig. 2 shows a schematic diagram of the UF-NF bench-scale type valve at the concentrate stream were used to adjust the oper-
units. The maximum operational pressure on the submerged UF ational conditions of this process. Permeate and concentrate fluxes
unit was 0.7 bar. The pressure was provided by a diaphragm pump were measured by a rotameter. This unit used a spiral wound NF
equipped with a speed controller and a maximum flow of 138 L/h. module (NF90-2540 - DowFilmtec) with a membrane area of
The pressure was measured by a manometer and adjusted by a 2.6 m2.
needle-type valve. The membrane filtration area of this unit was
0.047 m2. 2.5. Experimental procedure
The maximum operational pressure of the NF unit was 15 bar.
The pressure was provided by a rotary vane pump (Procon) 2.5.1. Bench-scale UF-NF experimental procedure
equipped with a speed controller and a maximum flow of 530 L/ Firstly, UF membrane was cleaned with a 200 ppm NaClO solu-
h. A needle-type valve was used to adjust the feed flow rate and tion in an ultrasound bath for 20 min. The membrane was then
the transmembrane pressure. The stainless steel filtration cell flushed with distilled water. The UF tests were carried out with
had a diameter of 9.8 cm and a filtration area of 75 cm2. A DOW 4 L of mixed effluent at pH 5.0. The pH was adjusted with NaOH
diamond-shaped feed spacer with 711 lm (28 mil) commonly 10%. The UF was carried out at a constant pressure of 0.3 bar. Per-
used in commercial spiral wound NF and RO modules was placed meate flux was measured every 15 min. Samples of the effluent
over the membrane to promote flow distribution. after pH adjustment and UF permeate were collected and analyzed
as described in Section 2.6.
2.4.2. Pilot-scale integrated UF-NF In all bench-scale NF experiments, the pressure and feed flow
Fig. 3 shows a schematic diagram of the pilot-scale integrated rate were set at 10 bar and 144 L/h, respectively. This gives a
UF-NF unit. The mixed effluent was sent to an effluent tank with cross-flow velocity of 1.9 m/s and a Reynolds number of 850
a volume of 250 L. The pH was manually adjusted in this tank to (Section 2.5.3).

Thermometer Manometer

Vacuum
Manometer Needle-type
Valve

Permeate
FI

Needle-type Rotameter
Valve
Diaphragm NF Membrane Cell
SC Pump
NF Supply Rotary-vane
Speed Controller Tank SC Pump
Speed Controller

UF membrane

Effluent Tank

Fig. 2. Schematic of the UF-NF bench-scale unit.

 L.H. Andrade et al. / Separation and Purification Technology 174 (2017) 44–56 47

Vacuum
Manometer

NaOH 10%

Mixed Effluent
Needle-type
Valve Vacuum Diaphragm
Vacuum Pump
Tank
Solenoid
Valve
UF Retentate
pH

UF membrane

Effluent Tank UF Tank

Rotameter NF Permeate
FI
Manometer
Thermometer

Centrifugal Pump Rotameter NF Retentate

FI
NF Membrane
NF Supply Cell
Tank Diaphragm
SC Pump
Speed Controller

Fig. 3. Schematic of the UF-NF pilot-scale unit.

Before each experiment, the NF membrane was cleaned by module at a flow rate of 144 L/h for 30 min and the water flux
soaking in citric acid solution at pH 2.5 followed by 0.1% NaOH was measured at 6 bar. If it was close to the flux of the virgin mem-
solution in an ultrasound bath for 20 min. After that, the mem- brane (which was around 42 L/h m2 at 6 bar), the module was con-
brane was flushed and compacted with distilled water. The perme- sidered clean and the next test was started. Otherwise, a chemical
ate flux was periodically measured by collecting the volume of cleaning procedure was performed. This cleaning consisted of
permeate produced in 60 s in a measuring cylinder. The feed tem- recirculating 0.2% HCl solution for 90 min. After cleaning, the
perature was monitored and maintained at approximately 25 °C. membrane was compacted with tap water for 60 min. The water
Permeate flux was normalized to 25 °C by means of a correction flux was again measured at 6 bar and compared to that of the vir-
factor calculated as water viscosity at the temperature of perme- gin membrane.
ation divided by water viscosity at 25 °C [46]. The feed temperature was monitored in all tests and the perme-
To evaluate the system behavior with increasing feed concen- ate flux was measured periodically. Permeate flux was normalized
tration and permeate recovery rate (RR), 10 L of ultrafiltered efflu- to 25 °C using the viscosity correction factor. Feed and permeate
ent at pH 5.0 was fed into the NF system. The NF unit was operated samples were collected and analyzed for conductivity, pH, sulfate,
in semi-batch mode, with continuous permeate removal and con- calcium, magnesium, and arsenic (as described in Section 2.6). A
centrate recycle to NF supply tank. NF was carried out until a per- summary of the operational conditions for each test is found in
meate recovery rate (RR) of 72% was achieved. Samples of the feed Table 2.
effluent and NF permeate were collected and analyzed for The commercial antiscalant agent Acumer 4300 (Dow Filmtec)
conductivity. at a dosage of 10 ppm was used for antiscalant evaluation. Accord-
To evaluate the long-term operation of the bench-scale unit, NF ing to the manufacturer, the product consists of maleic multipoly-
was carried out in semi-batch form. Initially, two liters of pre- mer and was developed to prevent both calcium carbonate and
treated effluent were added to the NF supply tank. The permeate calcium sulfate salt deposition in water systems [47].
was continuously collected and the concentrate returned to the Forward flush was evaluated as a physical cleaning procedure.
NF supply tank. Once the permeate RR reached 40% (maximum During forward flush, the system was depressurized while main-
recovery rate as defined previously in Andrade et al., [43]), every taining the same feed flow rate (equal 90 L/h). The physical clean-
time 0.2 L of permeate was collected, 0.5 L of pretreated effluent ing frequency assessed was two minutes per hour. Chemical
was added to the NF supply tank in order to maintain the RR near cleaning with 90-min, 0.2% HCl solution recirculation was evalu-
40%. This test was conducted for a total duration of 1500 min and a ated after continuous operation for intervals of 100 and 215 h.
volume of 18.5 L of effluent was used throughout the test. Water flux was measured at 6 bar before and after the cleanings.
Bench and pilot long-term NF operation (1500–1800 min) were
2.5.2. Pilot-scale UF-NF experimental procedure compared. Bench scale data were obtained using the conditions
Pilot-scale UF was continuously performed. The pH of the efflu- stated in Section 2.5.1. Pilot scale data were obtained at 6 bar, feed
ent was adjusted to 5.0 and subsequently, it was ultrafiltered at a flow rate of 90 L/h, without physical cleaning nor anticalant
pressure of 0.25 bar. Samples of the effluent after pH adjustment dosage. Permeate RR were 36–40% and 30–43% for bench and pilot
and UF permeate were collected periodically. systems, respectively.
Tests were carried out in the NF pilot-scale unit to evaluate the
effects of pressure, cross-flow velocity, concentration, antiscalant 2.5.3. Calculations
usage, and physical and chemical cleaning on membrane perfor- In bench-scale NF tests, the feed cross flow velocity (u0 ), m/s,
mance. Before each test, tap water was recirculated through the was calculated using Eq. (1) [48].
48 L.H. Andrade et al. / Separation and Purification Technology 174 (2017) 44–56

Table 2
Operational conditions of the pilot scale tests.

Test Operation Initial feed Feed flow rate (L/h) Operational Total filtration Antiscalant Physical
volume (L) pressure (bar) time (min) dosage cleaning
Effect of feed pressure Batcha 200 144 4, 6, 8 and 10 240 No No
Effect of feed cross-flow velocity Batcha 200 90, 96, 120, 144, 168, 384 6 240 No No
Concentration test (without antiscalant) Semi-batchb 180 90 6 360 No No
Concentration test (with antiscalant) Semi-batchb 180 90 6 360 Yes No
Physical cleaning Continuousc 75 90 6 3540 Yes Yes
Chemical cleaning Continuousc 75 90 6 – Yes No
a
The concentrate and permeate were returned to the NF supply tank.
b
The concentrate was returned to the NF supply tank and the permeate was continuously collected.
c
The concentrate and permeate were partially removed and partially returned to the supply tank to maintain the volume of effluent in the NF supply tank constant. The
partial recirculation was necessary since the maximum UF permeate flow rate was smaller than NF flow rate.

Q feedbench Asp ¼ 2  p  lm  dsp ð11Þ

u0 ¼ ð1Þ
4phr lntd
Thus, the Re can be calculated by Eq. (12) as,
where Q feedbench is the feed flow rate in m3/s, h is half the thickness
q  v  dh
of the feed spacer in m, and r lntd is the average logarithmic radius as Re ¼ ð12Þ
obtained by Eq. (2). l
R  r1 The mass transfer coefficient can be calculated using the dimen-
r lntd ¼   ð2Þ sionless Sherwood number (Sh) which, in spiral wound modules, is
ln rR1
associated with the dimensionless numbers of Reynolds and Sch-
midt (Sc) as shown by Eq. (13) [50].
where R is the stainless steel cell radius in m, and r 1 is the mem-
brane cell input channel radius in m. The Reynolds Number (Re) ki :dh
was then obtained from Eq. (3). Sh ¼ 0:14 Re0:64 Sc0:42 ¼ ð13Þ
D
qu0 dH
Re ¼ ð3Þ l
l Sc ¼
q  Di
ð14Þ
where q is the effluent density (kg/m ), l is the effluent viscosity
3
where ki is the mass transfer coefficient (m/s) and Di is the diffusiv-
(Pa s), and dH is the hydraulic diameter (m) (dH ¼ 4h [48]). The efflu-
ity (m2/s) of a certain component ‘i’ [51].
ent density and viscosity were considered to be equal to the water
In order to evaluate the salt precipitation on membrane surface
density and viscosity.
and consequent membrane fouling potential, the calcium sulfate
The Reynolds number and mass transfer coefficient were also
supersaturation index on the membrane surface (SIm ) was deter-
calculated for each pilot test. According to Van Gauwbergen and
mined for the feed concentration test. SIm was calculated according
Baeyens [49], cross-flow velocity of a spiral wound membrane
to Eq. (15) for different permeate RR [37,52]. Calcium sulfate was
module (v ), in m/s, can be calculated as a function of the feed flow
chosen as a representative salt precipitate as it has low solubility
rate (Q feedpilot ), in m3/s, and effective flow area of the feed channel
and high concentration on the mixed gold mining effluent (Table 1),
(Aeff , in m2) by Eq. (4):
and, thus, shows a higher precipitation potential.
Q feedpilot  0:5
v¼ ð4Þ
SIm ¼ C Ca2þm  C SO42m =K 0sp ð15Þ
Aeff
and Aeff can be obtained from Eq. (5): where C Ca2þm and C SO42m are the molar concentrations of Ca2+ and
Aeff ¼ B  dsp  e
0
ð5Þ SO2
4 on membrane/solution interface and K sp is the solubility pro-
duct constant for calcium sulfate, based on molar concentration.
where dsp is the thickness of the channel (m), which is assumed to
Molar concentrations on membrane surface (C im ) were deter-
be twice the average filament thickness (dF ) of the feed spacer; B is
mined by Eq. (16) [53].
the module width (m); and e is the porosity (Eqs. (6)–(8)).
 
V sp C im ¼ eJv =ki ðC ib  C ip Þ þ C ip ð16Þ
e¼1 ð6Þ
Vt
where C im , C ib and C ip are the concentrations of the specie ‘i’ on the
V sp ¼ 0:5  P 
2
dF  lm ð7Þ membrane/solution interface, bulk solution and permeate, respec-
tively; Jv is the permeate flux; and ki the mass transfer coefficient
2 of specie i, determined through Eq. (13).
V t ¼ lm  dsp ð8Þ
K 0sp can be calculated through the mass transfer constant based
where lm is the mesh size of the feed spacer (m). In order to calcu- on activity (K sp ) as shown in Eq. (17).
late Re, the hydraulic diameter (dH ) must be determined by Eq. (9)
K sp ¼ aCa2þm  aSO42m ¼ K 0sp  cCa2þm  cSO42m ð17Þ
4e
dH ¼   ð9Þ
2
þ ð1  eÞasp where cim and aim are the activity coefficient and the activity of spe-
dsp
cie ‘i’ on membrane surface. In solutions with ionic strength lower
where than 0.5 M, the activity coefficient can be calculated by Davies
equation [54]. Only calcium, magnesium, sulfate and arsenic were
Asp
asp ¼ ð10Þ considered for calculating the ionic strength of the solution, as they
V sp account for approximately 80% of the total ionic strength.
 L.H. Andrade et al. / Separation and Purification Technology 174 (2017) 44–56 49

The concentration polarization factor (CP) was also calculated Rphysicrev ¼ Rtotal  Rmembrane  Rphysicirrev ð25Þ
by Eq. (18).
Based on water flux after chemical cleaning (J chemclean Þ, the
C im irreversible fouling resistance can be found (Eq. (26)).
CP ¼ ð18Þ
C ib P
Rirrev ¼  Rmembrane ð26Þ
The resistance in series concept was used to evaluate the chem- J chemclean
l
ical cleaning. The total resistance (Rtotal ) was divided into mem-
Finally, chemically reversible fouling resistance was calculated
brane resistance (Rmembrane ), physically reversible fouling
using Eq. (27).
resistance (Rphysicrev Þ and physically irreversible fouling resistance
(Rphysicirrev ). Rphysicirrev in turn was divided into chemically reversi- Rchemrev ¼ Rphysicirrev  Rirrev ð27Þ
ble fouling (Rchemrev Þ and irreversible fouling (Rirrev ) resistances.
Physically reversible fouling resistance comprises of the concentra-
2.6. Analytical methods
tion polarization resistance and the resistance from fouling remov-
able by physical cleaning. On the other hand, chemically reversible
Samples of the effluent, UF permeate, NF permeate and NF con-
fouling resistance is the one removable by cleaning procedure with
centrate were periodically collected, both in bench- and pilot-scale
HCl solution. Finally, irreversible fouling resistance comprises of
tests. For each sample, the following parameters were measured:
the resistances which cannot be removed by physical nor chemical
conductivity (Hanna conductivity meter HI 9835), pH (pH meter
cleanings (Eqs. (19) and (20)).
Qualxtron QX 1500), sulfate, calcium, magnesium (ion chro-
Rtotal ¼ Rmembrane þ Rphysicrev þ Rphysicirrev ð19Þ matograph Dionex ICS-1000 equipped with AS-22 and ICS 12-A
columns), and total arsenic [56].
Rphysicirrev ¼ Rchemrev þ Rirrev ð20Þ The retention efficiencies of the UF and NF (EUF and ENF ) were
calculated according to Eqs. (28) and (29).
Based on the virgin membrane water permeability, the intrinsic
membrane resistance to filtration (Rmembrane ) was calculated accord- EUF ¼ C effluent  C pUF =C effluent
100% ð28Þ
ing to Eq. (21):
ENF ¼ C pUF  C pNF =C pUF
100% ð29Þ
P
Rmembrane ¼ ð21Þ
J v irgin
l where C effluent , C pUF and C pNF are the conductivities, ion concentra-
tions or arsenic concentrations in the effluent after pH adjustment,
where J v irgin is the water flux at 25 °C in m3/s m2 measured before UF permeate and NF permeate, respectively. As shown in Eq. (28),
the beginning of the tests, P is the operational feed pressure (Pa) the EUF comprises of the retention efficiency of the UF, excluding
and l is the permeate (water) viscosity in N s/m2. the removal by precipitation caused by the pH adjustment.
Eq. (22) was used to calculate the total resistance:
P  r Dp 2.7. Economic aspects
Rtotal ¼ ð22Þ
l
Jeffluent
A cost estimate of the UF-NF membrane system to treat the
where Jeffluent is the permeate flux in m3/h m2 obtained right before mixed gold mining effluent was conducted. The variables consid-
the cleaning procedure, r is the reflection coefficient, estimated by ered in this estimate were the system capital expenditure (CapEx),
the intrinsic membrane rejection [55], and Dp is the difference in membrane replacement costs, power consumption costs, system
the osmotic pressure of the solution at the membrane surface and maintenance, chemical products for membrane cleaning, alkalizing
permeate. Thus, P  rDp is the effective pressure, in Pa. The osmotic agent for pH adjustment, personnel costs, and unit depreciation.
pressure at membrane surface was estimated to be similar to that in All calculations were based on a designed system capacity (Q des )
concentrate stream. Although this simplification is not strictly cor- of 280 m3/h, which is the gold mining effluent flow rate mentioned
rected due to concentration polarization phenomenon, it resulted in by the mining company. The considered dollar quotation was R$
differences of at most 8% in Dp values and was then considered 4.00/US$ 1.00.
adequate. The UF-NF CapEx was US$ 2,450,000. The replacement costs of
The osmotic pressure difference between the NF concentrate the NF and UF membranes were US$ 50 and US$ 75 dollars per
and the permeate was estimated by the Van’t Hoff Equation, square meter, respectively. These prices were provided by a large
described in Eq. (23): commercial membrane supplier. The required membrane area
Dp ¼ RT RDC ð23Þ was calculated considering stabilized pilot-scale permeate flux
(10 L/h m2 for NF and 22 L/h m2 for UF). The membrane lifetime
where R is the universal gas constant (L Pa/K mol), T is the perme- was considered to be 5 years [57]. Also according to the membrane
ation temperature (K), and RDC is the sum of the difference in molar supplier, the system maintenance charges were equivalent to 5% of
concentrations of dissolved species present in concentrate and per- the initial investment.
meate. Once a real industrial effluent was used in this study, the To estimate the CapEx per cubic meter of effluent, the capital
quantification of all the dissolved species was impractical. Thus, cost was annualized using the amortization factor (A/P) as
only the species presented in greatest concentration in the effluent presented in Eq. (30) [58].
(sulfate, calcium, magnesium and arsenic, Table 1) were considered DL
for calculation of RDC and r. ic ð1 þ ic Þ
A=P ¼ DL
ð30Þ
The physically irreversible fouling resistance can be calculated ð1 þ ic Þ 1
by the water flux after 30 min of water recirculation (J physicclean ,
where ic is the investment rate, considered equal to 12% in Brazil,
in m3/s m2) (Eq. (24)).
and DL is the design life of the plant, taken as 15 years. The capital
P cost per cubic meter of effluent (C cap=m3 ) was obtained from Eq. (31):
Rphysicirrev ¼  Rmembrane ð24Þ
J physicclean
l
On the other hand, the physically reversible fouling resistance CapEx  A=P
C cap=m3 ¼ ð31Þ
can be calculated by Eq. (25). Q des
50 L.H. Andrade et al. / Separation and Purification Technology 174 (2017) 44–56

The NF power consumption, estimated by ROSA 9.1 software, Table 3

was 0.61 kW h/m3 while the UF power consumption was Concentration polarization factor calculated for each operating pressure.

0.22 kW h/m3 [59]. The electricity tariff was US$ 0.04/kW h, which Pressure (bar) 4 6 8 10
is the actual tariff paid by this mining factory. CP Ca2þ 1.87 2.02 2.33 2.62
The alkalizing agent used for pH adjustment was NaOH. The CP SO2 1.67 1.78 2.03 2.27
4
volume of NaOH solution used to adjust the effluent pH was mea-
sured and used to determine the amount of neutralizing agent
required (0.96 kgNaOH/m3effuent). The price of NaOH was US$ 1.00/ were obtained at lower pressure, which causes a dilution of the
kg. The cost of cleaning agents was calculated by considering one solute in the permeate. On the other hand, the higher concentra-
chemical cleaning with HCl 0.2% solution every two weeks and HCl tion polarization causes an increase in the solute concentration
35% solution at the rate of US$ 0.38/L. at the membrane surface and the amount of solute being trans-
Personnel costs included payroll for hiring four technicians/op- ferred to the permeate increases.
erators and 13 annual salaries plus 100% corresponding labor costs Hence, increasing the pressure should be avoided as it leads to
(payroll taxes and benefits). higher energy expenditure. Moreover, no benefits were observed
either in terms of concentration polarization control or in terms
3. Results and discussion of solute retention.

3.1. Evaluation of NF at different feed pressures 3.2. Evaluation of NF at different cross-flow velocities

Fig. 4 shows the permeate flux and the ratio of permeate flux to Fig. 6 shows the ratio of permeate flux to initial permeate flux
initial permeate flux (J/J0) for the NF at different feed pressures. for the NF at different feed cross-flow velocities.
It can be seen from Fig. 4 that the higher the pressure, the
greater is the initial flow, but the faster is the flux decay and the
lower is J/J0 after 4 h of filtration. This is because the driving force
is enhanced at higher pressures and both solvent and solutes are
convected towards the membrane surface even more, leading to
augmented concentration polarization, fouling and sharp flux
decline [19,28]. The concentration polarization factor (CP) was cal-
culated for calcium and sulfate ions for each condition considering
the average permeate flux observed between 0 and 15 min filtra-
tion (Table 3). The growth in CP caused by increase in pressure
was significant, which resulted in higher filtration resistance. Con-
sequently, at the end of the experiment, the largest flows were
observed for lower pressures.
With regard to the retention of ions, it is observed that, contrary
to the findings reported by other authors [22], retention decreases
with increasing pressure (Fig. 5). This can be explained by two fac- Fig. 5. Retention efficiencies of conductivity, sulfate, calcium, and arsenic at
tors. On one hand, it was observed that higher final solvent fluxes different NF pressures.

Fig. 4. Permeate flux and ratio of permeate flux to initial flux (J/J0) for different NF pressures. Operating conditions: pilot-scale plant, feed flow rate 144 L/h, without
antiscalant.
 L.H. Andrade et al. / Separation and Purification Technology 174 (2017) 44–56 51

Fig. 8. Nanofiltration permeate flux after 4 h filtration and corresponding required

pump power.
Fig. 6. Ratio of permeate flux to initial permeate flux for the NF of gold mining
effluent at different cross-flow velocities. Operating conditions: pilot-scale plant,
operating pressure 6 bar, without antiscalant. and Wiesner [58], the feed pump power requirement (Ef , in W)
can be estimated by Eq. (32):
Table 4 P
Q feed
Ef ¼ ð32Þ
Flow conditions tested.
g
Feed flow Cross-flow Reynolds Effective SO2
4 mass Effective Ca2+ mass
rate (L/h) velocity number transfer coefficient transfer coefficient
where Q feed is the feed flow rate (m3/s), P is the pressure (Pa), and g
(m/s) 5
(
10 m/s) (
105 m/s) is the pump efficiency, estimated as 70%. Fig. 8 shows the nanofil-
90 0.008 8 1.08 0.86 tration permeate flux after 4 h of filtration and the required feed
96 0.009 8 1.13 0.90 pump potency calculated by Eq. (32) for a single-vessel NF unit
120 0.011 11 1.30 1.04 (considering a 2.6 m2 module), for each cross-flow velocity tested.
140 0.013 13 1.46 1.16 It can be seen that the cross-flow velocity of 15
103 m/s
168 0.015 15 1.61 1.28
384 0.034 34 2.74 2.18
(feed flow rate of 168 L/h) can be regarded as a threshold cross-
flow velocity. From this velocity on, little variation was observed
on the permeate flux, and it was very close to the water permeabil-
By increasing the feed flow rate, the cross-flow velocity grows ity (42 L/h m2 at 6 bar). This shows that fouling can be effectively
and, consequently, the Reynolds number increases. This leads to controlled through the improvement of hydrodynamic and mass
improved mass transfer conditions (Table 4), which facilitates the transfer conditions. It is observed that the required feed pump
back diffusion of solutes retained by the membrane to the bulk power is a linear function of the feed flow rate and cross-flow
solution and reduces the concentration polarization. Consequently, velocity. Thus, cross-flow velocity above 15
103 m/s increase
the permeate flux increases for two reasons: reduction of the the energy expenditure without providing benefits in terms of per-
membrane fouling and reduction of the osmotic pressure on the meate flux, and therefore, are not recommended.
membrane surface, thus increasing the effective filtration pressure.
At higher feed flow rates, since the back transport from the 3.3. Concentration test with and without antiscalant
membrane surface to the solution bulk is increased, fewer solutes
enter the permeate channels [60]. This can be seen in Fig. 7, which Sulfate and calcium concentrations in the effluent were close to
shows an increase in the retention efficiencies of solutes at higher 3700 and 560 mg/L respectively. The calcium sulfate supersatura-
feed flow rates. tion index, obtained from Eq. (15), was 1.52, which means that
On the other hand, higher feed flow rates require pumps with the effluent was already saturated. The supersaturation index is
higher power, which consumes more energy. According to Sethi important as it can be related to the variation in the Gibbs free
energy (DG) through Eq. (33). For SIm > 1, the DG is negative and
the precipitation process is thermodynamically spontaneous.
DG ¼ RTlnðSIm Þ ð33Þ
However, although precipitation may be expected when SI > 1, a
significantly higher value of supersaturation index must be reached
to result in scaling because of kinetics and metastability effects [37].
The presence of other cations in the effluent, such as Mg2+, can also
hinder the calcium sulfate precipitation when the SI is slightly lar-

Table 5
CaSO4 supersaturation index at the membrane surface (SIm ) for different permeate
recovery rates during effluent nanofiltration without antiscalant. Operating condi-
tions: pilot-scale plant, feed flow rate 90 L/h, operating pressure 6 bar, without
antiscalant.

RR 10% 20% 30% 40% 50% 60% 70% 80% 90%

Fig. 7. Retention efficiencies of conductivity, sulfate, calcium, and arsenic at SIm 2.86 2.96 2.96 3.29 3.27 3.45 4.07 5.38 10.68
different cross-flow velocities.
52 L.H. Andrade et al. / Separation and Purification Technology 174 (2017) 44–56

enhancement or threshold effect, which reduces the precipitation

of low solubility inorganic salts; crystal modification, which
deforms the growing inorganic salt crystals to produce small, irreg-
ular, readily fractured crystals that do not adhere well to surfaces;
and dispersing activity, which prevents the precipitated crystals or
other inorganic particles from agglomerating and depositing on
surfaces [47]. Al-Roomi e Hussain [31] concluded that maleic anhy-
dride based polymers can adsorb on both, seeds (inhibiting the
nucleation phase) and microcrystals. In the second case, the poly-
mer may convey a higher negative charge to the crystals, which
slows the agglomeration, settling and deposition of crystals. The
presence of antiscalant agents can also interfere with the regular
crystal structure, thereby increasing the internal stress of crystals
and preventing the deposition of microcrystals.
Therefore, the overall effect of the antiscalant dosage is to
reduce the formation of precipitates that can deposit at the mem-
brane surface. At permeate recovery rates below 40%, nanofiltra-
tion performance with and without antiscalant was almost
identical, which suggests that the deposition of CaSO4 or other salts
at the membrane surface was low. With an increase in the RR, the
supersaturation index of calcium sulfate became higher than 3.0
(Table 5) and, consequently, fouling was expected to occur when
antiscalant was not used. This explains the higher permeate flux
observed with antiscalant dosage.
The permeate conductivity with and without antiscalant dosage
tends to rise with an increase in the RR and the concentration of
Fig. 9. Permeate conductivity and ratio of permeate flux to initial permeate flux for
different permeate recovery rates, with and without antiscalant dosage. Operating
ions in the concentrate stream. The increase of ionic concentration
conditions: pilot-scale plant, feed flow rate 144 L/h, operating pressure 6 bar. in the concentrate solution causes a growth in the solute concen-
tration at the membrane surface, and, as the intrinsic retention
of the membrane remains unchanged, the solute flux to the perme-
ger than 1, since the availability of sulfate ions for gypsum incipient ate increases. The permeate conductivity follows a pattern similar
nuclei formation is reduced because of the competition from MgSO4 to that of the permeate flux. Up to an RR of approximately 40%, the
complexation [61]. According to Dydo et al. [62], scaling can be pre- permeate conductivity, with and without antiscalant dosage, were
vented if SICaSO4 < 1.62–2.02. similar. At RR higher than 40%, the values for NF with antiscalant
On the other hand, once the nanofiltration begins, the formation were lower. This can be related to the deposition of precipitates
of the boundary layer and concentration polarization starts. This at the membrane surface of NF without antiscalant. The formation
increases the concentration of solutes on the membrane surface of such deposits reduces the solute diffusion from the vicinity of
and the SIm (Table 5). It can be seen that, even for a permeate the membrane surface back to the bulk solution, resulting in the
recovery rate as low as 10%, the solution SIm was above 2.8. More- so-called cake-enhanced concentration polarization effect [63].
over, as the RR increases, the ionic concentrations on the concen- This causes the concentration of ions in the membrane-solution
trate increases, consequently increasing the SIm . This indicates interface to be higher, promoting a higher solutes concentration
the importance of antiscalant dosage to minimize membrane in the permeate.
fouling. Therefore, it can be seen that, for the effluent studied, antis-
Fig. 9 shows the NF permeate conductivity and the ratio of per- calant is an important asset to minimize fouling, as it increases
meate flux to initial permeate flux for different RR values, with and the permeate flux by 13–36% for an RR of 45–90%. Moreover, the
without antiscalant dosage. It is noted that, until an RR of approx- antiscalant dosage decreases the solute flux and improves the final
imately 35% is reached, permeate flux with and without antiscalant permeate quality.
dosage are similar. After that, the permeate flux without antis- Fig. 10 shows the permeate flux and conductivity as a function
calant is always lower, indicating higher membrane fouling. of permeate recovery rate in bench- and pilot-scale tests without
According to the supplier, the antiscalant Acumer 4300 (Dow antiscalant dosage. The fouling tendency, observed by the decrease
Filmtec) acts through at least three mechanisms: solubility of permeate flux (Fig. 10a), was considerably higher in the bench-

Fig. 10. (a) Permeate flux, and (b) permeate conductivity; for the concentration tests at bench- and pilot-scale without antiscalant.
 L.H. Andrade et al. / Separation and Purification Technology 174 (2017) 44–56 53

scale unit. This difference was attributed to three factors. First, the
pressure applied at bench scale test was higher, which leads to
higher permeate flux. It caused increased concentration polariza-
tion and fouling. Secondly, the hydrodynamic condition of the
bench unit (Re = 850) is theoretically favored over the pilot unit
(Re = 8). However, the membrane cells traditionally used in labora-
tory experiments have a very bad flow distribution, causing short
circuiting of the feed stream and recirculation areas where the flow
velocity is very low [64]. Finally, due to the low filtration area of
the bench-scale membrane cell, more than 57 h of filtration was
required to concentrate 10 L of effluent to an RR of 70%, which
caused an excessive recirculation of the concentrate to the supply
tank. On the other hand, in the pilot scale unit, less than 6 h were
required to concentrate 180 L of effluent to a RR of 90%. Since the
precipitation of salts (especially CaSO4) and fouling formation are
time-dependent processes [35], extending the bench-scale test Fig. 11. NF permeate flux with and without forward flush physical cleaning.
duration may have caused an overestimation of the fouling forma- Operating conditions: pilot-scale plant, feed flow rate 90 L/h, operating pressure
tion. According to Gorzalski and Coronell [65], the bench-scale 6 bar, with antiscalant.
fouling tests produced foulant layers that were different in compo-
the productivities in both cases were almost the same. With phys-
sition from those generated at full-scale units. The dissimilarities
ical cleaning, as the nanofiltration was carried out for 58 min per
between bench- and full-scale results were likely caused by the
hour, with no permeate generation in the remaining 2 min, the sys-
various challenges in reproducing full-scale conditions at bench-
tem productivity with physical cleaning was 10.9 L/h m2.
scale units, particularly the use of a one-pass filtration scheme.
With forward flux, as the recirculation of the feed stream is
However, up to an RR of approximately 45%, no difference was
maintained even in the absence of permeate flux, the convective
observed in the permeate conductivity between bench- and pilot-
flux of the solutes towards the membrane stops, and the foulants
scale tests (Fig. 10b). At RR higher than 50%, the permeate conduc-
return to the solution bulk by diffusion [66]. As a result, physical
tivity of the pilot-scale membrane was increasingly better than
cleaning can contribute to slowing the fouling process by removing
that of the bench scale. This can also be attributed to the overesti-
a part of the accumulated material at the membrane surface and
mation of the membrane fouling in the bench scale tests and the
delaying the consolidation of the boundary layer. However, once
consequent cake-enhanced concentration polarization effect.
the salts begin to precipitate, deposit and adsorb on the membrane,
These results corroborate the importance of evaluating the nanofil-
physical cleaning by forward flush alone is unable to remove them.
tration in a pilot scale unit in order to collect data before scaling
This would explain why the initial permeate flux decay with phys-
up.
ical cleaning was lower than without physical cleaning, but the
It must be emphasized that the permeate conductivity
steady flux was similar in both conditions.
remained high and almost constant up to an RR of 40% for both
In conclusion, for long-term continuous operation, physical
the cases, bench- and pilot-scale. From this RR onwards there
cleaning has little impact on the fouling control and the mainte-
was a significant increase in the permeate conductivity. Therefore,
nance of the membrane permeability. Similar results were found
in order to maintain a high permeate quality and generate a trea-
by Bonné et al. [67] who evaluated periodical air flush (15 s air
ted effluent fitted for industrial reuse, the pilot-test indicated a
and 45 s water flush), and by Kramer et al. [68] who evaluated
maximum RR of 40%, which validated the results obtained in the
5 min of forward flush every 24 h of filtration. Finally, the use of
bench-scale unit [43].
these methods usually results in a more complex system control
and equipment design [38], and reduces the system productivity.
3.4. Evaluation of physical cleaning
Therefore, it was not considered an effective fouling control strat-
egy for the mining effluent.
In order to evaluate the effectiveness of physical cleaning in
controlling membrane fouling, nanofiltration permeate flux was
monitored during operation with and without periodic forward 3.5. Evaluation of chemical cleaning
flush (forward flush was carried out for 2 min, every 58 min of
nanofiltration). The dashed line in Fig. 11 indicates the chemical Chemical cleaning was applied after 100 and 215 h of continu-
cleaning process after the operation, without physical cleaning, ous NF. The resistances to filtration measured after continuous NF
and subsequently, the operation with periodic forward flush. are shown in Fig. 12.
From Fig. 11, it is observed that the flux pattern followed a typ-
ical flux decline kinetic: at the beginning, there is a remarkable
decrease of flow followed by a less-sharp progressive decay, which
can be assumed to tend to zero [50]. Physical cleaning reduced the
rate of the first stage flux decay. In the initial 210 min of operation,
the average variation rate of the permeate flux without physical
cleaning was 0.034 L/h m2 min; while with physical cleaning, this
variation was 45% lower, i.e. 0.018 L/h m2 min. However, after the
initial transient period, semi-steady state was achieved and the
permeate flux remained almost constant in both cases due to the
presence of forced convection by the concentrate stream in the
axial direction [28]. The stabilized fluxes of the two operational
conditions were similar, at 10.8 and 11.3 L/h m2 for the processes
with and without physical cleaning, respectively. However,
although the stabilized flux with physical cleaning was 4% higher, Fig. 12. Resistances to filtration after 100 and 215-h continuous NF.
54 L.H. Andrade et al. / Separation and Purification Technology 174 (2017) 44–56

The increase in filtration time from 100 to 215 h increased the

total resistance by almost 30%. This growth is mainly due to
increased strongly attached fouling, which cannot be removed by
physical methods but only by chemical cleaning. On the other
hand, the physically reversible fouling resistance was substantially
similar in both cases (variation of only 5%). This resistance is
mainly related to concentration polarization. Therefore it was con-
cluded that the development of the polarized layer occurred
rapidly, and thereafter there was no increase in its influence on
the filtration process. Irreversible fouling, although minor contrib-
utor to the total resistance, experienced an increment with increas-
ing filtration time. Thus, in order not to cause irreversible loss of
permeability over time, the best chemical cleaning frequency must
be assessed.

3.6. Bench- and pilot-scale integrated UF-NF performance

Table 6 presents the UF average retention and permeate flux in

both bench- and pilot-scale tests. The UF had low retention effi-
ciencies (below 7%) for all evaluated parameters. This result is con-
sistent with the effluent characterization since most of the
contaminants are dissolved inorganic solids, which are not effi-
ciently removed by the UF membrane. The small observed reten-
tion of sulfate, calcium, magnesium and arsenic is due to
electrostatic interactions between the solutes and the membrane
surface and pores [69] or ion adsorption on membrane or cake Fig. 13. Permeate flux and permeability measured for bench and pilot NF.

[70]. Moreover, the retention efficiencies in bench- and pilot-

scales were similar. Since the same membrane was used in both
The membranes have similar initial permeability. Due to the
tests and the operational conditions do not significantly influence
higher pressure applied, the permeate flux at the bench scale test
ion retention by UF [69], this similarity was expected. In relation
was higher. However, the greater convective flow of solutes
to fouling, the bench scale UF showed better performance. Differ-
towards membrane causes higher concentration polarization and
ences in hydrodynamic conditions and total filtration time may
fouling, leading to a more pronounced permeability decay. Differ-
have caused this.
ence in hydrodynamic conditions, discussed previouly, may also
Permeate flux and permeability measured in 1500–1800 min-
have caused this discrepancy.
nanofiltration are shown in Fig. 13. Bench and pilot scale tests were
Regarding solutes retention (Table 7), it is possible to note that
conducted at 10 and 6 bar. Permeability (K effluent , in L/h m2 bar) was
pilot sulfate retention is slightly higher and calcium, magnesium
calculated according to Eq. (34).
and arsenic retentions are lower than the bench scale’s. This may
Jeffluent be due to differences in concentrations levels of the raw effluents,
K effluent ¼ ð34Þ
DP  rDp since different samples were used in each test.

Table 6
Average physico-chemical characteristics and operational parameters of the UF of mixed gold mining effluent.

Sample Sulfate Calcium Magnesium Arsenic Permeate flux Pressure Permeability

Bench-scalea Gold mining effluent at pH 5.0 (mg/L) 3694 532 468 524 – – –
UF permeate (mg/L) 3582 511 437 516 – – –
UF removal efficiency 3% 4% 7% 1% – – –
Filtration parameters – – – – 71 L/h m2 0.30 bar 238 L/h m2 bar
Pilot-scaleb Gold mining effluent at pH 5.0 (mg/L) 4190 607 341 596 – – –
UF permeate (mg/L) 4114 579 325 571 – – –
UF removal efficiency 2% 5% 5% 4% – – –
Filtration parameters – – – – 22 L/h m2 0.25 bar 89 L/h m2 bar
a
Average of 5 samples.
b
Average of 14 samples.

Table 7
Physico-chemical parameters of bench and pilot scale NF feed and permeate.

Scale Sample Conductivity (lS/cm) Sulfate (mg/L) Calcium (mg/L) Magnesium (mg/L) Arsenic (mg/L)
Bench scalea NF feed 3970 2553 426 110 506
NF permeate 238 126 24 6 161
Retention 94% 95% 94% 95% 68%
Pilot scalea NF feed 4649 4040 605 315 581
NF permeate 265 83 64 53 239
Retention 94% 98% 89% 83% 59%
a
Average of 17 samples.
 L.H. Andrade et al. / Separation and Purification Technology 174 (2017) 44–56 55

Table 8 The total cost of treatment was US$ 1.34/m3, which is smaller
Mining effluent treatment costs by UF and NF. than the actual wastewater treatment currently installed in the
Item Cost (US$/m3) gold mining factory. It was observed that the highest fraction of
Membrane replacement 0.117 the cost was associated with pH adjustment. Capital cost amortiza-
Capital cost amortization 0.147 tion was the second highest fraction.
Alkalizing agent 0.960
Cleaning agent 0.004
Energy 0.034 Acknowledgements
Maintenance 0.050
Labor 0.027 The authors would like to thank the Coordination of Improve-
Total 1.338 ment of Higher Education Personnel (CAPES) and Foundation for
Research Support of Minas Gerais (FAPEMIG) for the scholarships
and financial resources provided.
3.7. Economic aspects
References
Table 8 shows the cost of treatment of the mining effluent by UF
and NF in US$/m3 of treated effluent. The biggest portion of the [1] F.L. Lobo, M.P. Costa, E.M. Novo, Time-series analysis of Landsat-MSS/TM/OLI
cost of treatment is associated with pH adjustment (alkalizing images over Amazonian waters impacted by gold mining activities, Remote
Sens. Environ. 157 (2015) 170–184.
agent). However, as the effluent has an acidic pH, neutralization [2] B.K.C. Chan, A.W.L. Dudeney, Reverse osmosis removal of arsenic residues from
is necessary even for conventional treatments, like precipitation bioleaching of refractory gold concentrates, Miner. Eng. 21 (2008) 272–278.
and sedimentation, and effluent discharge in water bodies. Capital [3] J.E. Langsch, M. Costa, L. Moore, P. Morais, A. Bellezza, S. Falcão, New
technology for arsenic removal from mining effluents, J. Mater. Res. Technol.
cost amortization is the second largest portion of the total cost. 1 (2012) 178–181.
This reflects the high cost of the initial purchase of equipment, [4] F. Awadalla, A. Kumar, Opportunities for membrane technologies in the
and of the NF membrane replacement. Although the cost with treatment of mining and mineral process streams and effluents⁄, Sep. Sci.
Technol. 29 (1994) 1231–1249.
the cleaning agent is almost negligible, the high cost of membrane
[5] B. Al-Rashdi, D. Johnson, N. Hilal, Removal of heavy metal ions by
replacement reinforces the importance of establishing the best nanofiltration, Desalination 315 (2013) 2–17.
chemical cleaning conditions of operation, aimed at increasing [6] N. Pages, A. Yaroshchuk, O. Gibert, J.L. Cortina, Rejection of trace ionic solutes
in nanofiltration: influence of aqueous phase composition, Chem. Eng. Sci. 104
the membrane lifetime.
(2013) 1107–1115.
The total cost of treatment was US$ 1.34/m3, which can be con- [7] C. Sierra, J.R.Á. Saiz, J.L.R. Gallego, Nanofiltration of acid mine drainage in an
sidered low. The cost of the actual wastewater treatment currently abandoned mercury mining area, Water Air Soil Pollut. 224 (2013) 1–12.
installed in the gold mining factory is US$ 1.60/m3. The current [8] H. Al-Zoubi, A. Rieger, P. Steinberger, W. Pelz, R. Haseneder, G. Härtel,
Nanofiltration of acid mine drainage, Desalination Water Treat. 21 (2010) 148–
cost of fresh water is US$ 0.18/m3. Since the permeate is fit for 161.
industrial reuse, the treatment of this effluent with UF-NF would [9] S. Al-Thyabat, H. Al-Zoubi, Purification of phosphate beneficiation wastewater:
reduce the water consumption by approximately 932,000 m3 per separation of phosphate from Eshydia Mine (Jordan) by column-DAF flotation
process, Int. J. Miner. Process. 110 (2012) 18–24.
year (considering a UF and NF permeate recovery rate of 95% and [10] B.C. Ricci, C.D. Ferreira, A.O. Aguiar, M.C. Amaral, Integration of nanofiltration
40%, respectively), representing savings of more than US$ and reverse osmosis for metal separation and sulfuric acid recovery from gold
165,000/year. mining effluent, Sep. Purif. Technol. 154 (2015) 11–21.
[11] K. Häyrynen, E. Pongrácz, V. Väisänen, N. Pap, M. Mänttäri, J. Langwaldt, R.L.
Keiski, Concentration of ammonium and nitrate from mine water by reverse
osmosis and nanofiltration, Desalination 240 (2009) 280–289.
[12] M. Hoyer, D. Zabelt, R. Steudtner, V. Brendler, R. Haseneder, J.-U. Repke,
4. Conclusions Influence of speciation during membrane treatment of uranium contaminated
water, Sep. Purif. Technol. 132 (2014) 413–421.
[13] M. Wilf, The Guidebook to Membrane Technology for Wastewater
Changes in flow conditions caused by the increase in pressure Reclamation, Balaban Desalinations Publications, Hopkinton, 2010.
were not favorable to the system. The increase of permeate flux [14] M. Beyer, B. Lohrengel, L.D. Nghiem, Membrane fouling and chemical cleaning
increases the concentration of polarization and enhances the decay in water recycling applications, Desalination 250 (2010) 977–981.
[15] I. Petrinic, J. Korenak, D. Povodnik, C. Hélix-Nielsen, A feasibility study of
of the permeate flux. On the other hand, the improvement in mass
ultrafiltration/reverse osmosis (UF/RO)-based wastewater treatment and
transfer by increasing the cross-flow velocity can effectively reduce reuse in the metal finishing industry, J. Clean. Prod. 101 (2015) 292–300.
the concentration polarization and fouling tendency. The cross- [16] F.-F. Chang, W.-J. Liu, X.-M. Wang, Comparison of polyamide nanofiltration and
flow velocity threshold was identified as 15
103 m/s; after low-pressure reverse osmosis membranes on As (III) rejection under various
operational conditions, Desalination 334 (2014) 10–16.
which, there is an increase in energy expenditure, without any [17] G. Capar, L. Yilmaz, U. Yetis, Reclamation of acid dye bath wastewater: effect of
benefits in terms of membrane fouling tendency. pH on nanofiltration performance, J. Membr. Sci. 281 (2006) 560–569.
Antiscalant dosage reduced the permeate flux decay for recov- [18] Z. Wang, G. Liu, Z. Fan, X. Yang, J. Wang, S. Wang, Experimental study on
treatment of electroplating wastewater by nanofiltration, J. Membr. Sci. 305
ery rates higher than 40%. Since the antiscalant works by minimiz- (2007) 185–195.
ing the nucleation, growth or deposition of the precipitated [19] Y. Kaya, Z. Gönder, I. Vergili, H. Barlas, The effect of transmembrane pressure
crystals at the membrane surface, one can conclude that the pre- and pH on treatment of paper machine process waters by using a two-step
nanofiltration process: flux decline analysis, Desalination 250 (2010) 150–157.
cipitation of this effluent becomes imperative after an RR of 40%. [20] A.J. Semião, A.I. Schäfer, Estrogenic micropollutant adsorption dynamics onto
Physical cleaning by forward flush was not effective for this sys- nanofiltration membranes, J. Membr. Sci. 381 (2011) 132–141.
tem. The increase in continuous filtration time from 100 to 215 h [21] N.V. Patil, X. Feng, J.J. Sewalt, R.M. Boom, A.E. Janssen, Separation of an inulin
mixture using cascaded nanofiltration, Sep. Purif. Technol. 146 (2015) 261–
without chemical cleaning increased the total resistance by almost 267.
30%, which was mainly due to increase in chemically reversible [22] Y. Kaya, H. Barlas, S. Arayici, Nanofiltration of Cleaning-in-Place (CIP)
fouling. wastewater in a detergent plant: effects of pH, temperature and
transmembrane pressure on flux behavior, Sep. Purif. Technol. 65 (2009)
The importance of evaluating these operational parameters in a
117–129.
pilot-scale unit must be highlighted. We observed significant dif- [23] H.Q. Dang, W.E. Price, L.D. Nghiem, The effects of feed solution temperature on
ferences in the results of the permeate flux and retention efficien- pore size and trace organic contaminant rejection by the nanofiltration
cies from the bench- and pilot-scale tests. These differences were membrane NF270, Sep. Purif. Technol. 125 (2014) 43–51.
[24] N.B. Amar, H. Saidani, J. Palmeri, A. Deratani, Effect of temperature on the
attributed to longer concentrate recirculation and different flow rejection of neutral and charged solutes by Desal 5 DK nanofiltration
conditions. membrane, Desalination 246 (2009) 294–303.
56 L.H. Andrade et al. / Separation and Purification Technology 174 (2017) 44–56

[25] M. Nilsson, G. Trägårdh, K. Östergren, The influence of pH, salt and [48] V. Minnikanti, S. DasGupta, S. De, Prediction of mass transfer coefficient with
temperature on nanofiltration performance, J. Membr. Sci. 312 (2008) 97–106. suction for turbulent flow in cross flow ultrafiltration, J. Membr. Sci. 157
[26] N.-E. Belkhouche, M.A. Didi, S. Taha, N.B. Farès, Zinc rejection from leachate (1999) 227–239.
solutions of industrial solid waste—effects of pressure and concentration on [49] D. Van Gauwbergen, J. Baeyens, Macroscopic fluid flow conditions in spiral-
nanofiltration membrane performance, Desalination 239 (2009) 58–65. wound membrane elements, Desalination 110 (1997) 287–299.
[27] C.-V. Gherasim, P. Mikulášek, Influence of operating variables on the removal [50] C.M. Salgado, L. Palacio, P. Prádanos, A. Hernández, C. González-Huerta, S.
of heavy metal ions from aqueous solutions by nanofiltration, Desalination Pérez-Magariño, Comparative study of red grape must nanofiltration:
343 (2014) 67–74. laboratory and pilot plant scales, Food Bioprod. Process. 94 (2015) 610–620.
[28] R. Mukherjee, M. Mondal, A. Sinha, S. Sarkar, S. De, Application of [51] D.-X. Wang, M. Su, Z.-Y. Yu, X.-L. Wang, M. Ando, T. Shintani, Separation
nanofiltration membrane for treatment of chloride rich steel plant effluent, J. performance of a nanofiltration membrane influenced by species and
Environ. Chem. Eng. 4 (2016) 1–9. concentration of ions, Desalination 175 (2005) 219–225.
[29] A. Plottu-Pecheux, B. Houssais, C. Democrate, D. Gatel, C. Parron, J. Cavard, [52] Q. Huang, W. Ma, A model of estimating scaling potential in reverse osmosis
Comparison of three antiscalants, as applied to the treatment of water from and nanofiltration systems, Desalination 288 (2012) 40–46.
the River Oise, Desalination 145 (2002) 273–280. [53] Y. Song, X. Gao, C. Gao, Evaluation of scaling potential in a pilot-scale NF–
[30] A. Antony, J.H. Low, S. Gray, A.E. Childress, P. Le-Clech, G. Leslie, Scale SWRO integrated seawater desalination system, J. Membr. Sci. 443 (2013)
formation and control in high pressure membrane water treatment systems: a 201–209.
review, J. Membr. Sci. 383 (2011) 1–16. [54] P.L. McCarty, C.N. Sawyer, G.F. Parkin, Chemistry for Environmental
[31] Y.M. Al-Roomi, K.F. Hussain, Potential kinetic model for scaling and scale Engineering and Science, McGraw-Hill Higher Education, 2003.
inhibition mechanism, Desalination 154 (2015) 11–21. [55] S. Mattaraj, C. Jarusutthirak, R. Jiraratananon, A combined osmotic pressure
[32] J.A. van Paassen, J.C. Kruithof, S.M. Bakker, F.S. Kegel, Integrated multi- and cake filtration model for crossflow nanofiltration of natural organic
objective membrane systems for surface water treatment: pre-treatment of matter, J. Membr. Sci. 322 (2008) 475–483.
nanofiltration by riverbank filtration and conventional ground water [56] R. Dhar, Y. Zheng, J. Rubenstone, A. Van Geen, A rapid colorimetric method for
treatment, Desalination 118 (1998) 239–248. measuring arsenic concentrations in groundwater, Anal. Chim. Acta 526
[33] A. Rahardianto, B.C. McCool, Y. Cohen, Reverse osmosis desalting of inland (2004) 203–209.
brackish water of high gypsum scaling propensity: kinetics and mitigation of [57] R. Liikanen, J. Yli-Kuivila, J. Tenhunen, R. Laukkanen, Cost and environmental
membrane mineral scaling, Environ. Sci. Technol. 42 (2008) 4292–4297. impact of nanofiltration in treating chemically pre-treated surface water,
[34] H.-J. Oh, Y.-K. Choung, S. Lee, J.-S. Choi, T.-M. Hwang, J.H. Kim, Scale formation Desalination 201 (2006) 58–70.
in reverse osmosis desalination: model development, Desalination 238 (2009) [58] S. Sethi, M.R. Wiesner, Cost modeling and estimation of crossflow membrane
333–346. filtration processes, Environ. Eng. Sci. 17 (2000) 61–79.
[35] E. Lyster, M.-M. Kim, J. Au, Y. Cohen, A method for evaluating antiscalant [59] C.M. Chew, M.K. Aroua, M.A. Hussain, W.M.Z.W. Ismail, Evaluation of
retardation of crystal nucleation and growth on RO membranes, J. Membr. Sci. ultrafiltration and conventional water treatment systems for sustainable
364 (2010) 122–131. development: an industrial scale case study, J. Clean. Prod. 112 (2016) 3152–
[36] P.-Q. Yuan, N. Kong, Z.-M. Cheng, R. Semiat, Electrostatic potential on anti- 3163.
scalants modified CaCO3 (104) surface: a molecular simulation study, [60] Y. Roy, M.H. Sharqawy, J.H. Lienhard, Modeling of flat-sheet and spiral-wound
Desalination 238 (2009) 246–256. nanofiltration configurations and its application in seawater nanofiltration, J.
[37] A.I. Schäfer, A.G. Fane, T.D. Waite, Nanofiltration: Principles and Applications, Membr. Sci. 493 (2015) 360–372.
Elsevier, 2005. [61] Y.A. Le Gouellec, M. Elimelech, Calcium sulfate (gypsum) scaling in
[38] A. Al-Amoudi, R.W. Lovitt, Fouling strategies and the cleaning system of NF nanofiltration of agricultural drainage water, J. Membr. Sci. 205 (2002) 279–
membranes and factors affecting cleaning efficiency, J. Membr. Sci. 303 (2007) 291.
4–28. [62] P. Dydo, M. Turek, J. Ciba, K. Wandachowicz, J. Misztal, The nucleation kinetic
[39] B. Rietman, Cleaning spiral wound membrane modules with a two phase aspects of gypsum nanofiltration membrane scaling, Desalination 164 (2004)
solution, in: Department of Water Management, Delft University of 41–52.
Technology, Delft, 2013, p. 85. [63] D. Vogel, A. Simon, A.A. Alturki, B. Bilitewski, W.E. Price, L.D. Nghiem, Effects of
[40] C. Ayache, M. Pidou, J. Croué, J. Labanowski, Y. Poussade, A. Tazi-Pain, J. Keller, fouling and scaling on the retention of trace organic contaminants by a
W. Gernjak, Impact of effluent organic matter on low-pressure membrane nanofiltration membrane: the role of cake-enhanced concentration
fouling in tertiary treatment, Water Res. 47 (2013) 2633–2642. polarisation, Sep. Purif. Technol. 73 (2010) 256–263.
[41] G.R. Shetty, S. Chellam, Predicting membrane fouling during municipal [64] F. Cortés-Juan, B. Balannec, T. Renouard, CFD-assisted design improvement of a
drinking water nanofiltration using artificial neural networks, J. Membr. Sci. bench-scale nanofiltration cell, Sep. Purif. Technol. 82 (2011) 177–184.
217 (2003) 69–86. [65] A.S. Gorzalski, O. Coronell, Fouling of nanofiltration membranes in full-and
[42] M. Frank, J. Westerink, A. Schokker, Recycling of industrial waste water by bench-scale systems treating groundwater containing silica, J. Membr. Sci. 468
using a two-step nanofiltration process for the removal of colour, Desalination (2014) 349–359.
145 (2002) 69–74. [66] S. Bhattacharya, S.-T. Hwang, Concentration polarization, separation factor,
[43] L.H. Andrade, A.O. Aguiar, W.L. Pires, G.A. Miranda, L.P.T. Teixeira, G.C.C. and Peclet number in membrane processes, J. Membr. Sci. 132 (1997) 73–90.
Almeida, M.C.S. Amaral, Nanofiltration and Reverse Osmosis applied to gold [67] P. Bonné, P. Hiemstra, J. Van der Hoek, J. Hofman, Is direct nanofiltration with
mining effluent treatment and reuse, in, Brazilian Journal of Chemical air fluxh an alternative for household water production for Amsterdam?,
Engineering, 2016. Desalination 152 (2003) 263–269
[44] S. Zulaikha, W. Lau, A. Ismail, J. Jaafar, Treatment of restaurant wastewater [68] F.C. Kramer, R. Shang, S.G. Heijman, S.M. Scherrenberg, J.B. van Lier, L.C.
using ultrafiltration and nanofiltration membranes, J. Water Process Eng. 2 Rietveld, Direct water reclamation from sewage using ceramic tight ultra-and
(2014) 58–62. nanofiltration, Sep. Purif. Technol. 147 (2015) 329–336.
[45] DOWFilmtecTM, Product Information: FILMTEC NF90-400 Nanofiltration [69] E. Alventosa-deLara, S. Barredo-Damas, E. Zuriaga-Agustí, M.I. Alcaina-
Element, in: Form n. 609-00345-0312. Miranda, M.I. Iborra-Clar, Ultrafiltration ceramic membrane performance
[46] A. Drak, K. Glucina, M. Busch, D. Hasson, J.-M. Laîne, R. Semiat, Laboratory during the treatment of model solutions containing dye and salt, Sep. Purif.
technique for predicting the scaling propensity of RO feed waters, Desalination Technol. 129 (2014) 96–105.
132 (2000) 233–242. [70] B. Tanhaei, M.P. Chenar, N. Saghatoleslami, M. Hesampour, M. Kallioinen, M.
[47] DOWFilmtecTM, ACUMERTM 4300 Polymer Technical Data Sheet, in: 713-00006- Sillanpää, M. Mänttäri, Removal of nickel ions from aqueous solution by
0712-EN, 2012. micellar-enhanced ultrafiltration, using mixed anionic–non-ionic surfactants,
Sep. Purif. Technol. 138 (2014) 169–176.