Professional Documents
Culture Documents
net/publication/285535915
CITATIONS READS
4 1,423
2 authors:
Some of the authors of this publication are also working on these related projects:
Special Issue "Integrated Approaches of the Water Use Cycle: Technical and Environmental Challenges for Sustainability" - Call for Papers, WATER MDPI View project
Sustainability Special Issue "The Sustainability Management of Waste Electrical and Electronic Equipment " View project
All content following this page was uploaded by Carmen Teodosiu on 07 February 2018.
______________________________________________________________________________________________
Abstract
Mathematical modeling of the ultrafiltration (UF) process is motivated by the complexity of retention phenomena that occur
during the UF in conjunction with the complexity of wastewater matrices submitted to ultrafiltration, as well as by the need to
predict the process behavior in extreme conditions, such as high pollution loads for UF-membrane cleaning cycles optimization or
for automated control of the process. Thus, series of laboratory-scale ultrafiltration experiments were performed with synthetic
wastewaters containing suspended solids and colloids onto two tubular polymeric membranes: one made of polysulfone (MWCO
20 kDa), the other made of modified polyethersulfone (MWCO 6 kDa). The experimental results have shown good removal
efficiencies for suspended solids and colloids for both the membranes (over 95% for the polysulfone membrane and 100% for the
other), as well as completely different hydrodynamic behaviors. The polysulfone membrane had produced high initial permeate
fluxes but was readily fouled by the suspended solids and colloids, while the polyethersulfone membrane had significantly lower
fluxes, but has shown a very good resistance to fouling. The mathematical model used for the verification of the experimental
results was based on the resistance-in-series model and considered, beside the intrinsic membrane resistance, the filtration
resistance due to superficial accumulation of the suspended solids and colloids. The model uses an exponential term for cake
filtration resistance growth and the experimental data verification was possible by calculating a statistical coefficient through non-
linear regression. The mathematical modeling was performed in two ways: for every experiment, as well as by using multiple data
sets. The results show that in both cases this model describes very well the UF process for the removal of suspended solids and
colloids from wastewaters, although the two membranes present different characteristics and filtration behaviors. The correlation
coefficients R were over 0.87, while the determination coefficients R2 had values of over 0.77. The good modeling results have
enabled the use of the model for the simulation of permeate fluxes in extreme conditions which usually are not replicable at
laboratory scale. The simulations were performed for very long UF times as well as for very high UF pressures.
∗
Author to whom all correspondence should be addressed: email: gb@ch.tuiasi.ro
Barjoveanu and Teodosiu/Environmental Engineering and Management Journal 9 (2010), 3, 361-371
and Teodosiu 2006; Barjoveanu, 2009). Furthermore, described in literature as the resistances in series
predicting ultrafiltration membranes behavior in model (Ahmad et al., 2006; Baker, 2004; Cheryan,
treating wastewaters is usually difficult because of the 1998; Katsikaris et al., 2005; Fane, 1998; Jacob and
complexity of phenomena that occur during this Jaffrin 2000; Jones et al., 1993; Strathmann et al.,
process, given the inherent process characteristics of 2006). The approach is based on a boundary layer
flux decline over time in ultrafiltration, as well as the theory which considers phenomena like:
inconsistency and variation of wastewater matrix solubilization – diffusion, sorption, pore and film
characteristics in real-scale operations. flow, electrical charging, to contribute to flux decline
In principle, the maximization of ultrafiltration in ultrafiltration.
process performances refers to the determination of The most common expression of this thin film
optimal process parameters (feed flow, theory in mass transport through membranes is the
transmembrane pressure, pollutants load, concentration-polarization model, which also is
ultrafiltration-membrane cleaning cycle duration) so extensively presented in literature (Beicha et al.,
as to obtain the highest volumes of high quality 2009; Judd and Jefferson 2003; Jonsson et al., 2006;
effluents to meet the discharging or recycling Kwon et al., 2008; Lee and Clark, 1998).
requirements at the lowest costs. This is limited by In a previous paper (Barjoveanu et al., 2006), a
the permeate flux decline which occurs in resistances in series – derived model was developed
ultrafiltration because of the membrane fouling, and successfully used to test experimental data
which consequently conducts to the need for cleaning obtained onto two polymeric membranes used for the
membranes which decreases the systems productivity ultrafiltration of petrochemical wastewaters as was
and can lead to secondary pollution due to the use of described by Teodosiu et al. (2000).
chemical reagents (Barjoveanu and Teodosiu, 2009). The objective of this paper is to present the
Modeling of the ultrafiltration processes is results of testing of the same mathematical model of
usually difficult because the phenomena that occur the ultrafiltration process onto experimental data
during the separation and membrane cleaning cycles coming from two other polymeric membranes tested
and the complex interactions between the membrane with synthetic wastewaters containing suspended
and the wastewater lead to problems in finding solids and colloids. The modeling and simulation are
mathematical equations that accurately describe this also motivated by the need to predict the process
process. behavior in extreme conditions: for very long UF
Mathematical modeling of wastewater times as well as for very high UF pressures.
ultrafiltration is particularly justified by the
following: 2. Model development
• Most industrial wastewaters, even with
appropriate pre-treatment, contain a large The model used in this paper is based on the
number of pollutants (organic and inorganic resistance in series model described by Eq. 1, and
compounds, suspended solids, colloids or considers, beside the intrinsic membrane resistance
microorganisms) in diverse concentrations that RM, the resistance given by the fouling layer (cake)
can lead to a very rapid pore blocking at the developed at the membrane surface – RC, as well as
membrane surface or to membrane degradation the resistance generated by the surface pore blocking
(Nuortila-Jokinen et al., 2003; Pizzichini et al., - RB.
2005; Teodosiu et al., 1999);
• Ultrafiltration is a feasible wastewater treatment ∆P (1)
J (t ) =
option from the technical and economical points η ⋅ (R M + ∆ R M (t ) + RC (t ) + R B )
of view as long as the productivity of the UF
step justifies the investment and operational where: ∆P – transmembrane pressure, (Pa/m); ∆RM(t)
costs and in this respect mathematical modeling – increase of resistance due to irreversible fouling of
of the UF process is an essential tool to support the membrane pores (m-1); RC(t) – resistance due to
the optimal operation and control of the process. the cake deposition on membrane surface (m-1); RB –
There are two major approaches for modeling resistance due to pore blocking (m-1);
mass transport phenomena in ultrafiltration (Judd and To verify the model against experimental data,
Jefferson, 2003). The most common approach starts an expression proposed by Bhatacharyya et al. (1979)
from Darcy’s law for mass transport through porous for the cake development in time was used.
media according to which the permeate flux depends According to this, the cake resistance exponentially
on the driving force of the process as well as on a grows to a maximum Rcmax, as expressed in Eq. 2.
phenomenological coefficient which describes For using this model, the maximum cake
membrane permeability (Strathmann et al., 2006). resistance was calculated for the lowest permeate
In this approach, flux decline is described by fluxes, being impossible to determine it for zero flux,
adding to the intrinsic membrane resistance different as this would have raised high risks of irreversible
hydraulic resistance coefficients to quantify the membrane fouling and possible membrane
effects of filtration phenomena in order to generate a destruction.
relation between the permeat flux and pressure
(Toyomoto and Higuchi, 1992), which is extensively
362
Modeling and simulation of an ultrafiltration process for the removal of suspended solids and colloids from wastewater
(
Rc = Rc max 1 − e − k ⋅t ) (2) of suspended solids and colloids in real wastewaters.
The presence of suspended solids and colloids in the
wastewater was measured with a turbidity analyzer
where: Rcmax - maximum cake resistance, (m-1), k –
(Hanna Instruments, model 93703-11). The particle
statistically determined coefficient, t – ultrafiltration
size distribution of the bentonite used to produce
time, (s).
synthetic wastewater was determined on a Shimadzu
In these conditions, the resistance in series
Laser Diffraction Particle Size Analyzer, SALD –
model used for the modeling of experimental data is
7001 model and it is presented in Fig. 1.
described by Eq. 3:
363
Barjoveanu and Teodosiu/Environmental Engineering and Management Journal 9 (2010), 3, 361-371
The experimental results have shown very modifies at each iteration the parameters vector of the
good removal efficiencies for the suspended solids result function (eq.3) so as the sum of the squared
and colloids: over 95% for the polysulfone membrane differences between the experimental and the
and 100% for the modified poly-ether-sulfone generated values to be minimal (Kelley, 1999). The
membrane. Furthermore, the two membranes have result of this process was analyzed by calculating
presented completely different hydrodynamic specific dispersion analysis parameters like the χ2
behaviours. While the polysulfone membrane reduced test, the correlation coefficient (R), the
(PS20000) has produced high initial permeate fluxes determination coefficient (R2) and the dispersion (σ).
and it was readily fouled by the suspended solids and
colloids (Figs. 3 – 10), the polyetersulfone membrane 4. Results and discussions
(EM006) had produced significantly lower fluxes, but
it has presented a very good resistance to surface 4.1. Modeling of the PS2000 membrane fluxes
fouling (Figs. 11 – 22) (Barjoveanu et al., 2008).
Between the tests, the polysulfone membrane Some modeling results of the experimental
(PS2000) was cleaned by backwashing with permeate data obtained onto the polysulfone ultrafiltration
at 2.5 bar for 5 minutes, while the poly-ether-sulfone membrane are presented in the Figs. (3-10).
membrane (EM006) was cleaned by forward flushing
also with permeate, as backwashing would have R = 0,942; R2 = 0,889
damaged the membrane. 3
2.5
3.4. Modeling procedure
F lu x , m 3 /m 2 s , 1 0 -5
1.5
Mathematical modeling consists in comparing
the experimentally determined fluxes with the 1
364
Modeling and simulation of an ultrafiltration process for the removal of suspended solids and colloids from wastewater
F lu x , m 3 /m 2 s , 10-5
7
correlation coefficients R had values between 0.786 6
1.5 Fig.7. Experimental and calculated flux values for the T40-
1 test , p=2 bar, dead-end
1
0.5
R = 0,960; R2 = 0,923
0 8
F lu x, m 3 /m 2 s , 1 0 -5
Experimental Calculated 5
4
2 0
10 R = 0,938; R = 0,880 0 200 400 Time, s 600 800 1000
9
8
Experimental Calculated
F lu x, m 3 /m 2 s, 10-5
7
6
5
4 Fig.8. Experimental and calculated flux values for the T40-
3 2 test , p=2 bar, dead-end
2
1
0 2
10 R = 0,896; R = 0,804
0 100 200 300 400 500 600
Time, s 9
8
Experimental Calculated
F lu x , m 3 /m 2 s , 1 0 -5
7
6
Fig. 5.Experimental and calculated flux values for the T30- 5
4 test, p=2 bar, dead-end 4
3
2
R = 0,954; R = 0,912
2 1
10
9 0
8 0 100 200 300 400 500
Flux, m 3 /m 2 s, 10-5
7 Time, s
6
Experimental Calculated
5
4
3
2
Fig. 9. Experimental and calculated flux values for the T50-
1 3 test, p=2 bar, dead-end
0
0 100 200 300 400 500 600 700 800 900 1000
By analyzing the data presented in Figs. 3 - 10
Time, s
and in Table 2, one may observe that the accuracy of
Experimental Calculated the model decreases from the tests performed with
synthetic wastewaters of 20FTU (R between 0.843
Fig.6. Experimental and calculated flux values for the T30- and 0.977), to the tests performed at 50 FTU (R
5 test , p=2 bar, dead-end between 0.845 to 0.934).
365
Barjoveanu and Teodosiu/Environmental Engineering and Management Journal 9 (2010), 3, 361-371
2
R = 0,934 R = 0,872
This supports the idea that at the beginning of
10
9
the ultrafiltration tests, along with the cake filtration
8
retention mechanism, other phenomena occur
7 (especially pore blocking) leading to the dramatic flux
-5
Flux, m /m s, 10
5
confronting the data with pore-blocking filtration
3
4
3
models (Chang et al., 2002; Hermia 1982) has not
2 produced reliable results.
1 The calculated k coefficient presented values
0
0 100 200 300 400 500 600 700
situated between 0.0004 and 0.00201, with an average
Time, s of 0.001, along with a slight tendency of rising
Experimental Calculated together with the bentonite concentration in the
synthetic wastewater. Comparing these results with
Fig. 10. Experimental and calculated flux values for the the ones in the previous paper (Barjoveanu et al.,
T50-6 test, p=2 bar, dead-end 2006), in which k had an average value of 0.0005, we
may advance the hypothesis that this coefficient
Also, it is important to notice that the represents a measure of the influence on the flux
differences between the calculated fluxes and the ones decline of the characteristics and concentrations of the
of experimental data occur mostly in the first period pollutants that are present in the wastewaters.
of the ultrafiltration tests; in each case, the real flux
decline being faster than the one calculated with
model in Eq. 3.
Table 3. Modeling results of the PS20000 flux values based on multiple data sets
Test series k Error, 10-4 Reduced χ2, 10-11 σ, 10-8 R R2 No. of exp. points
T20 0.00087 0.232 84.5 1.09 0.935 0.874 1288
T30 0.00181 1.10 3.49 2.00 0.940 0.883 574
T40 0.00109 0.671 7.90 5.17 0.912 0.831 659
T50 0.00128 1.03 0.168 0.117 0.878 0.770 658
366
Modeling and simulation of an ultrafiltration process for the removal of suspended solids and colloids from wastewater
Flux, m 3 /m 2 s, 10-6
4
– 14 FTU with dimensions varying between 7-8 µm,
3
and the values of k ranged between 0.0002 and
0.0008, while in this situation the major fraction 2
Flu x, m 3 /m 2 s, 10-6
“global” values of k are presented in table 3, where a 4
0
2 0 5 10 15Time, min 20 25 30 35
5
T10 - 2.0 k=0.0024, R =0,951, R = 0,905
4.5 Experimental Calculated
3.5
3 T20-2.6 test, EM006, P = 2.6 bar
2.5
2
2
T40 - 2.6, k = 0,00312, R =0,881, R = 0,777
1.5 7
1 6
0.5
Flux, m 3 /m 2 s, 10 -6
5
0
0 10 20 30 40 50 60 70 4
Time, min
3
Experimental Calculated
2
1
Fig. 11. Experimental and calculated flux values for the 0
T10-2.0 test, EM006, P = 2 bar 0 5 10 15 20 25 30 35
Time, min
Experimental Calculated
By studying Figs. 11 to 22 and the Table 4
given below, it is clear that the resistances in series
Fig. 15. Experimental and calculated flux values for the
model describes fairly well the experimental data. T40-2.6 test , EM006, P = 2.6 bar
367
Barjoveanu and Teodosiu/Environmental Engineering and Management Journal 9 (2010), 3, 361-371
2 2
T40 - 2.0, k = 0,00193, R =0,624, R = 0,389 T75-2.2, k = 0,00426, R =0,742, R = 0,551
6 6
5 5
F lu x, m 3 /m 2 s, 1 0 -6
F lu x, m 3 /m 2 s , 10 -6
4
4
3
3
2
2
1
1
0
0 0 5 10 15 20 25 30 35
Time, min
0 5 10 15 20 25 30 35
Time, min
Experimental Calculated
Experimental Calculated
F lu x, m 3 /m 2 s, 10 -6
3.5
4
3
F lu x , m 3 /m 2 s , 1 0 -6
3.5
2.5
3 2
2.5 1.5
2 1
1.5 0.5
1 0
0.5 0 5 10 15 20 25 30 35
Time, min
0
Experimental Calculated
0 5 10 15 20 25 30 35
Time, min
Experimental Calculated Fig. 20. Experimental and calculated flux values for the
T75-2.0 test, EM006, P = 2.0 bar
Fig. 17. Experimental and calculated flux values for the
T50-2.0 test, EM006, P = 2 bar 2
T115-2.2, k = 0,00263, R =0,924, R = 0,854
5
4.5
2 4
T50 - 2.2, k = 0,00214, R =0,857, R = 0,735
F lu x, m 3 /m 2 s, 10 -6
6 3.5
3
5 2.5
2
F lu x , m 3 /m 2 s , 1 0 -6
4 1.5
1
3 0.5
0
2
0 5 10 15 20 25 30 35
Time, min
1
Experimental Calculated
0
0 5 10 15 20 25 30 35 Fig. 21. Experimental and calculated flux values for the
Time, min
Experimental Calculated
T115-2.2 test, EM006, P = 2.2 bar
2
T115-2.4, k = 0,00226, R =0,905, R = 0,815
Fig. 18. Experimental and calculated flux values for the 6
T50-2.2 test, EM006, P = 2.2 bar
5
F lu x , m 3 /m 2 s , 1 0 -6
368
Modeling and simulation of an ultrafiltration process for the removal of suspended solids and colloids from wastewater
Another aspect that has to be mentioned is that resistances in series model, it was possible to use this
the correlation and determination coefficients are model and the values of k presented in Tables 3 and 5
affected by the number of experimental points to conduct simulations of flux decline for the two
considered for the approximation, and in the case of membranes in extreme conditions that would not be
the EM006 tests the number of experimental points possible in the laboratory: for very high operational
obtained during one ultrafiltration test was pressures, and for very long UF tests durations,
significantly lower than for the PS20000 membrane. In general, ultrafiltration tests on the
For these reasons, just as for the PS20000 data polysulfone membrane lasted for 9 to 20 minutes,
another modeling round was performed considering given the acute tendency of rapid fouling of this
simultaneously multiple data sets. The results membrane, while the tests performed on the EM006
presented in Table 5. evidence that conducting the membrane lasted for 30 or 60 minutes. Simulating
modeling onto multiple data sets has lead to an very long ultrafiltration times can be helpful for
improvement of the correlation and determination optimizing real-scale applications. In Figs. 23 and 24
coefficients, their values being of over 0,89 for the are presented two of these simulations performed for
correlation coefficient and of over 0.807 for the the two membranes respectively.
determination coefficient respectively. The values Simulating pressure conditions different from
calculated for k are consistent with the individual test the ones realizable on the laboratory scale setup is
modeling, but still no correlation with the bentonite very important because it allows the study of the
concentration can be drawn. membranes behavior during real-scale applications. In
Figs. 25 and 26 are presented simulations of flux
4.3. Simulations for different operating conditions decline over time at different pressures.
It has to be mentioned that these simulations
Because the experimental data generated in can be performed only for the pressure range for
completely different conditions (onto two membranes which the clean water fluxes variation is linear, so as
with different material and operational characteristics, to exist a linear dependence between the permeate
at different pressures and initial pollutant flux and the intrinsic membranes resistance.
concentrations) were described fairly well by the
Table 5. Modeling results considering multiple data sets from the EM006 membrane
369
Barjoveanu and Teodosiu/Environmental Engineering and Management Journal 9 (2010), 3, 361-371
5. Conclusions
T20-6, PS20000
8
The mathematical modeling of the
7
ultrafiltration process is a very useful instrument that
6
allows the identification and verification of the
Flux, m 3/m 2 s, 10-5
5
complex phenomena that cause permeate flux decline
4
3
during ultrafiltration. Finding and using a reliable
2
mathematical model to accurately describe the
1
complex phenomena that occur at membranes level
0
during ultrafiltration offer the possibility of
0 0.5 1 1.5 2 2.5 conducting simulations of the process that can be
Time, hours
subsequently used to test the process in extreme
Experimental Simulated conditions and to determine optimal process
parameters without the risk of permanently damaging
Fig. 23. 2.3 hour s UF test on the PS20000 membrane at p = the membranes.
2 bar For this study, the experimental data were
obtained by testing 2 polymeric tubular ultrafiltration
T20-2.0, EM006 membranes with very different material and
5 hydrodynamic characteristics. A semi-empirical
4.5 model of the flux decline during ultrafiltration was
4
tested for the evaluation of data obtained during
Flux, m 3 /m 2 s,10-9
3.5
3 ultrafiltration experiments with synthetic wastewaters
2.5
2 containing suspended solids and colloids. The model
1.5 is based on the well-known resistances in series
1
0.5 model and considers as main phenomenon
0 responsible for flux decline the cake formation at the
0 2 4 6 8 10 12 14
Time, hours
membrane surface, together with the cake filtration
Experimental Simulated
mechanism. The model describes the flux decline as a
cause of exponential growth of the cake resistance
subdued by the depositing of solid material at the
Fig. 24. 12 hours UF test on the EM006 membrane at p = 2
bar
membrane surface.
The verification of the experimental data was
performed using a non-linear algorithm by calculating
PS20000, T50
45.00
statistically a coefficient (k), which confers the
40.00
35.00
empiric characteristic of the model. Mathematical
modeling has lead to good results in terms of
F lux, m 3 /m 2 s, 10-5
30.00
25.00 approximation quality, given the completely different
20.00 experimental conditions: membranes characteristics
15.00
10.00
(fouling tendencies and intrinsic membrane
5.00 resistance), synthetic wastewater loadings, and
0.00 operational parameters (pressure).
0 100 200 300 400 500 600 700
The good modeling results have enabled the
Time, s
use of the model for the simulation of permeate fluxes
2 bar 4 bar 6 bar 10 bar in extreme conditions which usually are not replicable
at laboratory scale. The simulations were performed
Fig.25. UF tests simulations on the PS20000 membrane at p for very long UF times as well as for high UF
= 2, 4, 6 and 10 bar pressures.
This study is part of a complex research in
25.00 EM006, T50 which a comparison between several processes for
20.00
advanced wastewater treatment (such as membrane
processes, wet hydrogen peroxide oxidations,
Flux, m 3 /m 2 s, 10-6
15.00
adsorption on surfactant modified zeolites) is
10.00 performed based on a unitary experimental approach
5.00
and similar wastewater matrices.
0.00 Acknowledgement
0 5 10 15 20 25 30 35
Time, s
This paper was elaborated with the support of IDEI
project (code 368, contract no. 739/2009): Combined
2 bar 4 bar 6 bar 10 bar
advanced wastewater treatment processes for persistent
organic pollutant removal and enhancement of wastewater
Fig. 26. UF tests simulations on the EM006 membrane at p biodegradability, financed by the National Council of
= 2, 4, 6 and 10 bar Scientific Research (CNCSIS), Romania.
370
Modeling and simulation of an ultrafiltration process for the removal of suspended solids and colloids from wastewater
371