- Email: [email protected]
Ultrafiltration of orange press liquor: Optimization for permeate flux and fouling index by response surface methodology
Separation and Purification Technology 80 (2011) 1–10
Contents lists available at ScienceDirect
Separation and Purification Technology journal homepage: www.elsevier.com/locate/seppur
Ultrafiltration of orange press liquor: Optimization for permeate flux and fouling index by response surface methodology René Andrés Ruby Figueroa, Alfredo Cassano ⇑, Enrico Drioli Institute on Membrane Technology, ITM-CNR, c/o University of Calabria, via P. Bucci, cubo 17/C, I-87030 Rende (CS), Italy
a r t i c l e
i n f o
Article history: Received 3 February 2011 Received in revised form 31 March 2011 Accepted 31 March 2011 Available online 13 May 2011 Keywords: Ultrafiltration (UF) Orange press liquor Response surface methodology (RSM) Membrane fouling
a b s t r a c t A Box-Behnken design of response surface methodology (RSM) was used to analyze the performance of polysulphone ultrafiltration (UF) hollow fiber membranes in the clarification of orange press liquors. A regression model was developed for permeate flux, fouling index and blocking index as a function of transmembrane pressure (TMP), temperature and feed flow-rate. The experimental operating conditions were selected within the following ranges: TMP 0.2–1.4 bar, temperature 15–35 °C, and feed flow-rate 85–245 L/h. A total of 30 experiments was performed according to the total recycle configuration. Based on the lack-of-fit test, the analysis of variance (ANOVA) showed the regression model to be adequate. From the regression analysis, the permeate flux, fouling index and blocking index were expressed with quadratic equations of TMP, temperature and feed flow-rate. Quadratic terms of TMP, temperature and feed flow-rate showed a significant influence on the performance of the UF membrane. A strong interaction effect of temperature and feed flow-rate was observed on the permeate flux while interactions TMP-temperature and TMP-feed flow-rate were found to be less significant. In the case of fouling index, interactions TMP-temperature and TMP-feed flow-rate produced a significant effect. In order to maximize the permeate flux and minimize the fouling index, the desirability function approach was applied to analyze the regression model equations. The optimized operating variables were found to be 1.4 bar, 15 °C and 167.7 L/h for a maximum desirability of 0.76. Experimental data of permeate flux and fouling index, obtained in optimized operating conditions, resulted in a good agreement with the predicted values of the regression model. Ó 2011 Elsevier B.V. All rights reserved.
1. Introduction Membrane processes are today consolidated systems in various productive sectors, since the separation process is athermal and involves no phase change or chemical agents. The introduction of these technologies in the industrial transformation cycle of the fruit juices represents one of the technological answers to the problem of the production of juices with high quality, natural fresh taste and additive-free. Juice clarification, stabilization, depectinization and concentration are typical steps where membrane processes as microfiltration (MF), ultrafiltration (UF), nanofiltration (NF) and reverse osmosis (RO) have been successfully utilized [1]. UF has been also developed for the treatment of wastewater and sewage in order to remove particulate and macromolecular materials [2–4]. The critical issue in a membrane-based separation process is the decline in permeate flux during its operation. Membrane fouling in ⇑ Corresponding author. Tel.: +39 0984 492067; fax: +39 0984 402103. E-mail address: [email protected] (A. Cassano). 1383-5866/$ - see front matter Ó 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.seppur.2011.03.030
crossflow ultrafiltration is a key factor affecting the economic and technological viability of the processes, which essentially depends on the permeate fluxes obtained and their stability with time [4]. Consequently, the identification and quantification of the prevalent fouling mechanism and efforts to minimize its effects during a continuous filtration process are extremely important [5]. The term membrane fouling is often used to describe a long term flux decline caused by accumulation of certain materials at the membrane surface [6]. It may occur due to a concentration polarization layer development over the membrane surface, the formation of a cake layer and/or a blockage of the membrane pores. The pore blocking can be further characterized by complete, intermediate and standard pore blocking [5]. Several studies are reported in literature about the membrane fouling. Some examples include concentration polarization [7], cake filtration [8–9] and blocking models [10]. In all of these models, the particle size or the ratio of the particle size to the membrane pores is the key parameter for choosing a suitable membrane. Concentration polarization occurs only when the particle diffusion flux is dominant, while the cake filtration model is used when particles
2
R.A. Ruby Figueroa et al. / Separation and Purification Technology 80 (2011) 1–10
have accumulated on the membrane surface. When the particle size is smaller than or comparable to the membrane pores, particles have the opportunity to penetrate and block the pores. In this case, the membrane blocking model may be the appropriate method to relate the received filtrate volume, v, to filtration time, t [11]. Basically, researchers study the membrane separation performance by using the ‘‘one-factor-at-a-time’’ approach. However, the effect of operating variables such as transmembrane pressure, temperature, feed flow-rate and solute concentration on membrane fouling may not be independent of each other and it is necessary to consider their interactions. A possible solution is to use the response surface methodology (RSM) approach that is widely used to analyze the effects of multiple factors and their interactions [12]. RSM is a collection of statistical and mathematical techniques useful for developing, improving, and optimizing processes in which a response of interest is influenced by several variables and the objective is to optimize this response. RSM has important applications in the design, development and formulation of new products, as well as in the improvement of existing product design. It defines the effect of the independent variables, alone or in combination, on the process. In addition, in order to analyze the effects of the independent variables, this experimental methodology generates a mathematical model which describes the chemical or biochemical processes [13–15]. Citrus fruits are widely used in the food industry for the production of fresh and concentrated juice, citrus-based drinks and cans. This process is accompanied by the production of large amounts of by-products such as peels and seed residues which may account for up to 50% of the total fruit weight. Most of the waste residue from commercial juice extractors is shredded, limed, cured and pressed into press liquors and press cakes which are then processed independently. Press liquors are enriched in bioactive compounds, such as flavonoids and phenolic acids, recognized for their beneficial implications in human health due to their antioxidant activity and free radical scavenging ability [16]. They are usually concentrated in multiple effect heat evaporators to yield citrus molasses, while d-limonene is obtained from the condensate. The UF process can be considered a valid approach for separating and recovering valuable products from finely divided solid waste materials present in citrus press liquor. It separates the flow from the press liquor into a permeate having a total soluble solids content and an acidity level approximating those of the press liquor and a retentate containing the suspended solids such as proteins and fibers and high molecular weight carbohydrates such as pectins associated with cloud [17,18]. In this work the objective was to use the RSM approach for investigating the simple effect and the interaction of different operating conditions, such as transmembrane pressure (TMP), temperature and feed flow-rate, on permeate flux, membrane fouling and blocking index of UF polysulphone hollow fibers. From the regression model sets of operating variables were determined for optimizing the clarification of orange press liquors.
2. Theory Hermia [2,5,19–21] developed four empirical models for deadend filtration based on constant pressure filtration laws that correspond to four basic types of fouling: complete blocking, intermediate blocking, standard blocking and cake layer formation. The type of fouling considered depends on the value of the parameter i in the following equation:
i @2t @t ¼K 2 @v @v
ð1Þ
where the blocking index i and the resistance coefficient K are functions of the blocking models. The membrane blocking phenomena can be described by these parameters; for example, i = 2 is complete blocking, i = 1.5 is standard blocking, i = 1 is intermediate blocking, and i = 0 is cake filtration. The blocking index and the resistance coefficient can be calculated by plotting d2t/dv2 versus dt/dv or solving Eq. (1) [11]. In this work a non-linear regression, based on the permeate flux experimental data, was used to obtain the blocking index i for each experimental run. 3. Materials and methods 3.1. Feed solution Citrus press liquors, with a suspended solid content of 7.13 ± 1.41% (w/w), were supplied by Gioia Succhi Srl (Rosarno, Reggio Calabria, Italy). Liquors were left overnight at room temperature to let the majority of the cloud particles settle out. A partially clear liquor was recovered by decanting of the cloud layer. Liquors were then depectinized by adding 7 g/kg of pectinase (4 h at room temperature) from Aspergillus aculeatus (Sigma–Aldrich, Milan, Italy) and filtered with a nylon cloth. Final liquors (feed solution), depleted in suspended solids, had a pH of 3.6 and a total soluble solids content of 8.6 °Brix. They were stored at 17°C and defrosted at room temperature before use. 3.2. UF equipment and procedures UF experiments were performed by using a laboratory unit (Fig. 1) equipped with a hollow fiber polysulphone membrane module supplied by China Blue Star Membrane Technology Co., Ltd (Beijing, China) with an effective membrane area of 0.16 m2 and a nominal molecular weight cut-off (MWCO) of 100 kDa. A heat exchanger, placed into the feed tank, was used to keep the feed temperature constant. Experimental runs were performed according to the total recycle configuration in which permeate and retentate streams were both continuously recycled back to the feed tank to ensure a steady state in the volume and composition of the feed. The operating conditions varied according to the experimental design described in the following. The mass of permeate collected was measured with an accuracy of ±0.1 g every 10 min for periods of 180 min. 3.3. Cleaning procedure The initial hydraulic permeability of the membrane module was 191.22 L/m2hbar at 20 °C. After each run, the membrane module was submitted to a cleaning procedure by using an enzymatic detergent (Ultrasil 50, Henkel Chemicals Ltd., Dussendorf) at a concentration of 1.0% (w/w) and at a temperature of 40 °C for 60 min. After a rinsing with distilled water for 15–20 min, the membrane module was cleaned with a 0.2% (w/w) NaOH solution at 40 °C for 60 min. A final rinsing with distilled water for 20 min was performed. After the cleaning step the hydraulic permeability of the membrane module, in fixed conditions (temperature 20 °C, feed flow-rate 245 L/h), was measured. The recovery of the hydraulic permeability after the cleaning procedure was higher than 89.6%. 3.4. Fouling index The fouling index was calculated by comparing the hydraulic permeability before and after the treatment of orange press liquor according to the following equation:
R.A. Ruby Figueroa et al. / Separation and Purification Technology 80 (2011) 1–10
3
Fig. 1. Scheme of the UF experimental set-up.
Table 1 Experimental range and levels of the independent variables for Box-Behnken design. Feed factors
Code
TMP (bar) Temperature (°C) Feed flow rate (L/h)
Fouling Index ¼
Variation levels
X1 X2 X3
1
L1p L0p
1
0
1
0.2 15 85
0.8 25 165
1.4 35 245
! 100
ð2Þ
where L1p and L0p are the hydraulic permeabilities measured after and before the treatment of the press liquor, respectively [22]. 3.5. Experimental design Based on the capability of the experimental set-up, operating variables were selected within the following ranges: temperature 15–35 °C, feed flow-rate 85–245 L/h and TMP 0.2–1.4 bar (Table 1). Experiments were performed according to the Box-Behnken design (Table 2) that is composed of 30 runs divided in two blocks, each one with three central points (runs 13, 14, 15 and 28, 29, 30). The correlation of the operating variables and the responses based on the Box-Behnken design are fitted to a quadratic polynomial equation [12] using the least-square method of the form:
Y k ¼ b0 þ
3 X i¼1
bi X i þ
3 X i¼1
in which it is assumed that random errors are identically distributed with a zero mean and a common unknown variance and they are independent of each other. The difference between ^) for the ith observation the observed and the fitted value (y e ¼ yi y^ is called the residual and it is an estimate of the corresponding ei . Our criterion for choosing the bi estimates is that they should minimize the sum of the squares of the residuals, which is often called the sum of squares of the errors [23].
bii X 2i þ
2 X 3 X i¼1
bij X i X j þ e
ð3Þ
ji1
where Yk is the response variable (Y1 for permeate flux, Y2 for fouling index and Y3 for blocking index), b0 the constant, e the residual (error or noise) term, bi the linear coefficients, bii the quadratic coefficients, bij the interaction coefficients and Xi is the dimensionless coded variables (X1 for TMP, X2 for temperature and X3 for feed flow-rate). The system of equations given above is solved by using the method of least squares (MLS), a multiple regression technique,
4. Results and discussion The effect of operating variables such as temperature, feed flowrate and transmembrane pressure, on the permeate flux, fouling index and blocking index was investigated by using a Box-Behnken response surface design model. The linear and quadratic effects of the factors studied and the interactions on the responses were obtained using analyses of variance (ANOVA). 4.1. Permeate flux The R-squared statistic indicates that the model as fitted explains 96.31% of the variability in the permeate flux. The lackof-fit test is designed to determine whether the selected model is adequate to describe the observed data, or whether a more complicated model should be used. The test is performed by comparing the variability of the current model residuals to the variability between observations at replicate settings of the factors. Since the P-value for lack-of-fit in the ANOVA is greater or equal to 0.05 (P = 0.0559, F-ratio 5.46), the model appears to be adequate for the observed data at the 95.0% confidence level. The Standard Error was equal to 0.687447. The Durbin–Watson (DW) statistic is a statistic tool used to detect the presence of autocorrelation in the residuals of a regression analysis. Since the P-value is greater than 5.0% (P = 0.9699), there is no indication of a serial autocorrelation in the residuals at the 5.0% significance level. Fig. 2 shows the linear, quadratic and interaction of each factor plotted in the form of Pareto chart where the effect is significant if
4
R.A. Ruby Figueroa et al. / Separation and Purification Technology 80 (2011) 1–10
Table 2 Experimental design and results of Box Behnken design. Run
Block
TMP (bar) X1
Temperature (°C) X2
Feed flow rate (L/h) X3
Permeate flux (Kg/m2h) Y1
Fouling index (%) Y2
Blocking index () Y3
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0
1 1 1 1 0 0 0 0 1 1 1 1 0 0 0 1 1 1 1 0 0 0 0 1 1 1 1 0 0 0
0 0 0 0 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 0 0 0
8.23 26.41 14.51 31.95 6.81 18.66 10.01 21.08 13.41 18.81 14.43 25.49 16.56 15.55 15.25 8.31 25.33 12.89 30.74 6.77 18.63 8.86 21.05 13.43 18.64 14.37 25.52 16.6 15.52 15.32
58.91 43.01 54.62 75.93 51.86 67.25 62.92 67.00 64.94 67.18 69.76 72.53 68.29 67.76 69.72 59.33 43.48 54.96 76.05 52.31 67.35 62.96 67.49 65.58 67.40 70.74 72.75 68.64 68.37 70.46
1.75 1.74 1.37 1.38 2.00 2.00 1.90 2.00 1.47 1.57 1.64 1.50 1.51 1.62 1.52 1.72 1.65 1.5 1.37 2.00 2.00 2.00 1.55 1.51 1.56 1.60 1.49 1.52 1.58 1.49
b1:TMP
+ -
b2:Temperature b22 b3:Feed Flow Rate b33 b23 b13 b11 b12
0
10
20 30 Standardized effect
40
50
Fig. 2. Standardized Pareto chart for permeate flux.
its corresponding bar crosses the vertical line at the P = 0.05 level. The linear coefficients of TMP (b1) were found to be the most significant effect to increase the permeate flux, followed by linear coefficient of temperature (b2), quadratic effect of temperature (b22) and then the linear effect of feed flow-rate (b3). On the other hand, the quadratic effect of feed flow rate (b33) produces a decrease in the permeate flux with a significant effect (P 6 0.05). The interaction factors as (b12), (b13) and the quadratic effect of TMP (b11) do not produce a significant effect (P P 0.05) in the permeate flux; therefore, for the next analyses these factors will not be included in the regression model equation of the permeate flux. These results indicate that the interaction of operating variables in the ultrafiltration of orange press liquor has to be considered. The quadratic regression equation describing the effect of the process variables on the permeate flux in terms of coded levels is the following:
Y 1 ¼ 15:79 þ 7:34125X 1 þ 3:41437X 2 þ 1:60313X 3 þ 4:01375X 22 þ 1:45X 2 X 3 1:79875X 23
ð4Þ
where Y1 is the predictive permeate flux for the ultrafiltration process. The positive linear coefficients of TMP (b1), temperature (b2) and feed flow-rate (b3) produce an increasing of the permeate flux when these factors are increased. Fig. 3 shows the effect of the investigated variables on the permeate flux; the 3D plot, obtained by using the Eq. (4), helps to predict the response at any combination of operating variables. The response surface of the permeate flux is plotted against two operating variables, while the third variable is kept constant (level 0 in Table 1). The TMP has a linear effect on the permeate flux for all the values investigated (Fig. 3a and c). The curvature in the 3D plot of the permeate flux arises due to the quadratic dependence on
5
R.A. Ruby Figueroa et al. / Separation and Purification Technology 80 (2011) 1–10
Permeate flux (kg/m2*h)
Permeate flux (kg/m2*h)
(a) 40 30 20 10 0 0
0.3
0.6 0.9 TMP (bar)
1.2
1.5
35 31 23 27 19 15 Temperature (°C)
(b) 26 23 20 17 14 11 15
19
23 27 Temperature (°C)
31
35
80
140
200
260
Feed flow rate (L/h)
Permeate flux (kg/m2*h)
(c) 25 20 15 10 5 0 0
0.3
0.6 0.9 TMP (bar)
1.2
1.5
80
140
200
260
Feed flow rate (L/h)
Fig. 3. 3D response surface with contour plot of permeate flux.
Residual Flux (kg/m2*h)
3.1 2.1 1.1 0.1 -0.9 -1.9 -2.9
0
10
20 Predicted Flux (kg/m2*h)
30
40
Fig. 4. Plot of residuals against predicted response of permeate flux in the UF process.
temperature and feed flow-rate (Fig. 3b and c). In addition, the temperature has a linear effect on the permeate flux after 25 °C (level 0). The feed flow-rate produces an increasing of the permeate flux from the level (1) to the level 0; after this point, an increasing in the feed flow-rate produces a decreasing of the permeate flux (Fig. 3b and c). In Fig. 4 the residual (errors) permeate flux was plotted against the predicted flux. The random and also the distribution of the
residuals over and below the centerline suggest that the model for the permeate flux is statistically significant. 4.2. Fouling index The R-squared statistic indicates that the model as fitted explains 90.47% of the variability in the fouling index. The P-value for lack-of-fit was equal to 0.0175 (F-ratio 10.5). This means that
6
R.A. Ruby Figueroa et al. / Separation and Purification Technology 80 (2011) 1–10
b12
+ -
b11 b2:Temperature b1:TMP b3:Feed Flow Rate b13 b22 b33 b23 0
10 15 Standardized effect
5
20
25
Fig. 5. Standardized Pareto chart for fouling index.
Fouling index (%)
(a) 79 69 59 49 39 0
0.3
0.6
0.9
1.2
1.5
35 31 27 23 19 15 Temperature (°C)
TMP (bar)
Fouling index (%)
(b) 77 73 69 65
260 200
61 15
19
23
27
31
35
80
140 Feed flow rate (L/h)
Temperature (°C)
Fouling index (%)
(c) 75 70 65 60 55 50
260 200
45
0
0.3
0.6
0.9
1.2
1.5
80
140 Feed flow rate (L/h)
TMP (bar) Fig. 6. 3D response surface with contour plot of fouling index.
the model as fitted is not completely adequate to represent the experimental data, and a more complex model including other factors should be used. The Standard Error was equal to 1.07706. Since the P-value of the Durbin–Watson statistic test is greater than 5.0% (P = 0.9700), there is no indication of serial autocorrelation in the residuals at the 5.0% significance level. Fig. 5 shows the linear, quadratic and interaction of each factor plotted in the form of Pareto charts where the effect is significant if its corresponding bar crosses the vertical line at the P = 0.05 level.
The interaction factor between TMP and temperature (b12) was found to be the most significant effect to increase the fouling index, followed by quadratic effect of TMP (b11) that produces a decreasing in the fouling index. The interaction factor between temperature and feed flow rate (b23) is the only factor that does not produce a significant effect (P P 0.05) in the fouling index. The quadratic regression equation describing the effect of the process variables on the fouling index in terms of coded levels is reported in the following:
7
R.A. Ruby Figueroa et al. / Separation and Purification Technology 80 (2011) 1–10
5.1
residual
3.1 1.1 -0.9 -2.9 -4.9 43
53
63 predicted
73
83
Fig. 7. Plot of residuals against predicted response of fouling index in the UF process.
b11
+ -
b33 b22 b2:Temperature b23 b1:TMP b13 b3:Feed Flow Rate b12
0
2
4
6 8 Standardized effect
10
12
Fig. 8. Standardized Pareto chart for blocking index.
Y 2 ¼ 68:8733 þ 3:10563X 1 þ 4:10438X 2 þ 2:6425X 3
8:52729X 21
þ 9:26875X 1 X 2 2:7275X 1 X 3
2:05979X 22
þ 2:04646X 23
ð5Þ
where Y2 is the predictive fouling index for the UF process. The positive linear coefficients of TMP (b1), temperature (b2) and feed flow-rate (b3) produce a significant (P 6 0.05) increasing of the fouling index when these factors are increased. The effect of different variables on the fouling index is shown in Fig. 6. The response surface of the fouling index is plotted against two operating variables while the third variable is kept constant (level 0 in Table 1). A strong interaction of TMP and temperature was observed from the warping of the 3D fouling index plot (Fig. 6a). The fouling index increased with temperature only at TMP higher than level 0; at lower level the fouling index decreased by increasing the temperature. The fouling index increased by increasing the feed flow-rate (Fig. 6b and c). The residual (errors) fouling index was plotted against the predicted flux (Fig. 7). The model for the fouling index is statistically significant as suggested by the random and the distribution of the residuals over and below the centerline. 4.3. Blocking index The R-squared statistic indicates that the model as fitted explains 76.36% of the variability in the blocking index. The P-value for lack-of-fit was equal to 0.2555 (F-ratio 1.44). This means that the model appears to be adequate for the observed data at the 95.0% confidence level. The Standard Error was equal to 0.106816. Since the P-value of the Durbin–Watson statistic test is
greater than 5.0% (P = 0.4672) there is no indication of serial autocorrelation in the residuals at the 5.0% significance level. In a membrane process the fouling mechanism is affected by modifications of operating conditions. Fig. 8 shows the linear, quadratic and interaction of each factor plotted in the form of Pareto charts. The linear effect of TMP (b1) and feed flow rate (b3) and the interaction effect for (b23), (b13) and (b12) do not produce a significant effect (P P 0.05) in the blocking index; consequently, these factors were not included in the regression model equations. The quadratic effect for TMP was found to be the most significant effect to increase the value of i (blocking index), followed by the quadratic effect of feed flow rate (b33) and the quadratic effect of temperature (b22). The quadratic regression equation describing the effect of the process variables on the blocking index in terms of coded levels is the following:
Y 3 ¼ 1:54 0:08375X 2 þ 0:204375X 21 0:184375X 22 þ 0:186875X 23
ð6Þ
Membrane pore blocking is theoretically influenced by three factors: the amount of particles simultaneously arriving at the membrane surface, the amount of particle accumulation, and the filtration rate. When more particles arrive at the membrane surface at the same time, they have less opportunity to migrate to the membrane pores due to the crowding out produced by neighboring particles. In contrast, particles have a greater chance to penetrate the pores if fewer particles have accumulated on the membrane surface. Furthermore, the drag force exerted on the particles and the resulting particle motion is significantly determined by the
8
R.A. Ruby Figueroa et al. / Separation and Purification Technology 80 (2011) 1–10
Blocking index (-)
(a)
2
1.8 1.6 1.4 1.2 0
0.3
0.6
0.9
1.2
1.5
15
19
23
27
31
35
Temperature (°C)
TMP (bar)
Blocking index (-)
(b)
2
1.8 1.6 1.4
260 200
1.2 15
19
23
27 Temperature (°C)
31
35
80
140 Feed flow rate (L/h)
(c) Blocking index (-)
2.3 2.1 1.9 1.7
260 200
1.5 0
0.3
0.6
0.9
1.2
1.5
80
140 Feed flow rate (L/h)
TMP (bar) Fig. 9. 3D response surface with contour plot of blocking index.
0.43
residual
0.23
0.03
-0.17
-0.37 1.3
1.5
1.7 predicted
1.9
2.1
Fig. 10. Plot of residuals against predicted response of blocking index in the UF process.
filtrate flow rate. A greater drag force generated by a higher filtration rate may cause the particles to migrate more easily to the membrane pores. However, these factors are not completely independent. The amount of particles simultaneously arriving at the membrane surface is proportional to the filtration rate, while particles accumulation is the integral sum of the instantaneous
particle amount [11]. Fig. 9b and c show the effect of the feed flow-rate on the i value. Increases in feed flow-rate from level 0 until level 1 produce an increasing on the i value; this means that at level 0 the principal mechanism is a combination of intermediate blocking and standard blocking; when the feed flow-rate is increased the blocking index assumes higher values.
9
R.A. Ruby Figueroa et al. / Separation and Purification Technology 80 (2011) 1–10 Table 3 Optimization results for the UF of citrus press liquor. Optimized coded level of variables
X1: TMP X2: Temperature X3: Feed flow-rate
Predicted responses
Permeate flux (kg/ m2h) Fouling index (%)
1 (1.4 bar) 1 (15 °C) 0.034 (167.72 L/ h) 23.73 48.02 0.76
Overall desirability
The response surface plots (Fig. 9) show a strong interaction between feed flow-rate, temperature and TMP. An increasing in TMP produces a decreasing in the value of the blocking index. Hwang et al. [11] observed that the pore blocking is of more significance under lower pressure. The same effect is shown in Fig. 9a and c when at minimum level (1) of TMP the values of the blocking index are closer to i = 2, where the main mechanism involved is the complete pore blocking (i = 2). This may be caused by fewer particles simultaneously arriving at the membrane surface. The model for the blocking index is statistically significant as suggested by the random and the distribution of the residuals over and below the centerline (Fig. 10).
desirability functions from the considered responses are then combined to obtain a total desirability function, D (0 6 D 6 1), defined by Derringer et al. [26] and Del Castillo et al. [27]. The aim of this optimization was to find operating conditions giving the maximum permeate flux and the minimum fouling index simultaneously. The desirability function approach was employed in this procedure. The minimum and maximum limits used for the permeate flux were 6.77 (dk = 0) and 31.95 (dk = 1) kg/m2h, respectively. In addition, the fouling index were 43.01 (dk = 1) and 76.05 (dk = 0)% for the minimum and maximum limit, respectively. For the maximum overall desirability of 0.76, the coded levels of the operating variables are TMP X1 = 1, temperature X2 = 1 and feed flow-rate X3 = 0.034 (Table 3). Fig. 11 shows the time course of permeate flux values experimentally determined in optimized operating variables (T = 15 °C, TMP = 1.4 bar, feed flow-rate = 167.7 L/h). The initial permeate flux of 35.5 kg/m2h was reduced of 26% after 71 min and then reached a steady-state value of 25.52 ± 0.40 kg/m2h. The fouling index evaluated after the treatment of press liquor was 42.35%. These results are in agreement with predicted values of the regression model (permeate flux 23.73 kg/m2h, fouling index 48.02%).
5. Conclusions 4.4. Optimization of multiple responses In RSM, the desirability function (DF) is widely used to determine a combination of variables to optimize multiple responses [24–25]. A DF method finds operating conditions that provide the ‘‘most desirable’’ responses. It is based on the conduction of experiments and fitting response models (yk) for all k responses, the definition of individual desirability functions for each response (dk), and the maximization of the overall desirability in comparison with controllable factors. For each response yk(xi), a desirability function dk(yk) assigns numbers between 0 and 1 to the possible values of yk, with dk(yk) = 0 representing a completely undesirable value of yk and dk(yk) = 1 representing a completely desirable or ideal response value. Depending on whether a particular response yi has to be maximized, minimized or assigned to a target value, different desirability functions dk(yk) can be used. The individual
The effect of operating variables such as transmembrane pressure (TMP), temperature and feed flow-rate on the performance of a UF membrane in the clarification of orange press liquor was studied with the response surface methodology. A Box-Behnken design was used for regression modeling and optimizing the UF operating variables. Optimization of multiple responses permitted to establish the operating conditions giving maximum permeate flux and minimum fouling index, simultaneously. For an overall desirability of 0.76, permeate fluxes of 23.7 kg/m2h and fouling index of 48.0% were estimated, respectively, in optimized operating variables (TMP = 1.4 bar, temperature = 15 °C and feed-flow rate = 167.7 L/ h). Compared with experiments performed at optimized operating conditions, the predicted steady-state permeate flux and fouling index showed differences of 7.1% and 11.7%, respectively.
40.0 37.5 35.0
2
Jp (kg/m h)
32.5 30.0 27.5 25.0 22.5 20.0 0
25
50
75
100
125
150
175
200
Operating time (min) Fig. 11. Time course of permeate flux in optimized operating variables (T = 15 °C; TMP = 1.4 bar; feed flow-rate = 167.72 L/h).
10
R.A. Ruby Figueroa et al. / Separation and Purification Technology 80 (2011) 1–10
Acknowledgment R.A. Ruby Figueroa gratefully acknowledges the Istituto Italo-latino Americano (IILA) which supported his stage at the ITM-CNR for the development of this work.
References [1] A. Cassano, L. Donato, E. Drioli, Ultrafiltration of kiwifruit juice: operating parameters, juice quality and membrane fouling, J. Food Eng. 79 (2007) 613–621. [2] M.C. Vincent Vela, S. Alvarez Blanco, J. Lora García, E. Bergantiños Rodríguez, Analysis of membrane pore blocking models adapted to crossflow ultrafiltration in the ultrafiltration of PEG, Chem. Eng. J. 149 (2009) 232–241. [3] H.K. Shon, S. Vigneswaran, I.S. Kim, J. Cho, H.H. Ngo, Effect of pretreatment on the fouling of membranes: application in biologically treated sewage effluent, J. Membr. Sci. 234 (2004) 111–120. [4] M. Cheryan, J.R. Alvarez, Food and Beverage industry applications, in: R.D. Noble, S.A. Stern (Eds.), Membrane Separation Technology–Principles and Applications, Elsevier, AmsterdamThe Netherlands, 1995, p. 341. [5] R.W. Field, D. Wu, J.A. Howell, B.B. Gupta, Critical flux concept for microfiltration fouling, J. Membr. Sci. 100 (1995) 259–272. [6] H. Strathmann, L. Giorno, E. Drioli, An introduction to membrane sciences and technology, Consiglio Nazionale delle Ricerche, Roma, 2006. [7] M. Cheryan, Ultrafiltration and Microfiltration Handbook, Technomic Publishing Co., Pennsylvania, USA, 1998. [8] W.M. Lu, C.C. Lai, K.J. Hwang, Constant pressure filtration of submicron particles, Sep. Technol. 5 (1995) 45–53. [9] C. Tien, B.V. Ramarao, On analysis of cake formation and growth in cake filtration, J. Chin. Inst. Chem. Eng. 37 (1) (2006) 81–94. [10] P.H. Hermans, H.L. Bredee, Principles of the mathematical treatment of constant-pressure filtration, J. Soc. Chem. Ind. 55T (1936) 1–11. [11] K.J. Hwang, C.Y. Liao, K.L. Tung, Analysis of particle fouling during microfiltration by use of blocking models, J. Membr. Sci. 287 (2007) 287–293. [12] M.N. Hyder, R.Y.M. Huang, P. Chen, Pervaporation dehydration of alcohol– water mixtures: optimization for permeate flux and selectivity by central composite rotatable design, J. Membr. Sci. 326 (2009) 343–353.
[13] M.F. Anjum, I. Tasadduq, K. Al-Sultan, Response surface methodology: a neutral network approach, Eur. J. Oper. Res. 101 (1997) 65–73. [14] R.H. Myers, D.C. Montgomery, Response Surface Methodology: Process and Product Optimization Using Designed Experiments, John Wiley Sons, Inc., New York, 1995. [15] M.C. Martí-Calatayud, M.C. Vincent-Vela, S. Álvarez-Blanco, J. Lora-García, E. Bergantiños-Rodríguez, Analysis and optimization of the influence of operating conditions in the ultrafiltration of macromolecules using a response surface methodological approach, Chem. Eng. J. 156 (2010) 337–346. [16] A. Bocco, M.E. Cuvelier, H. Richard, C. Berset, Antioxidant activity and phenolic composition of citrus peel and seed extract, J. Agric. Food Chem. 46 (1998) 2123–2129. [17] B.R. Breslau, E.A. Agranat, A.J. Testa, S. Messinger, R.A. Cross, Hollow Fiber Ultrafiltration. A Systems approach for Process Water and By-Product Recovery, in: Proc. of 79th National Meeting of the AIChE, Houston, Texas, March 1975. [18] R.J. Braddock, Ultrafiltration and reverse osmosis recovery of limonene from citrus processing waste streams, J. Food Sci. 47 (1982) 946–948. [19] J. Hermia, Constant pressure blocking filtration law: application to power law non-Newtonian fluids, Trans. Inst. Chem. Eng. 60 (1982) 183–187. [20] C. Duclos-Orsello, W. Li, C.C. Ho, A three mechanism model to describe fouling of microfiltration membranes, J. Membr. Sci. 280 (2006) 856–866. [21] M.C. Vincent Vela, S. Alvarez Blanco, J. Lora García, E. Bergantiños Rodríguez, Analysis of membrane pore blocking models applied to the ultrafiltration of PEG, Sep. Purif. Technol. 62 (2008) 489–498. [22] M. Mänttäri, M. Nyström, Membrane filtration for tertiary treatment of biologically treated effluents from the pulp and paper industry, Water Sci. Technol. 55 (2007) 99–107. [23] D. Basß, I.H. Boyaci, Modeling and optimization I: usability of response surface methodology, J. Food Eng. 78 (2007) 836–845. [24] S.J. Kalil, F. Maugeri, M.I. Rodrigues, Response surface analysis and simulation as a tool for bioprocess design and optimization, Process Biochem. 35 (2000) 539–550. [25] E.C. Harrington, The desirability function, Ind. Qual. Control 21 (1965) 494–498. [26] G. Derringer, R. Suich, Simultaneous optimization of several response variables, J. Qual. Technol. 12 (1980) 214–219. [27] E. Del Castillo, D.C. Montgomery, D.R. McCarville, Modified desirability function for multiple response optimization, J. Qual. Technol. 28 (1996) 337–345.
Contents lists available at ScienceDirect
Separation and Purification Technology journal homepage: www.elsevier.com/locate/seppur
Ultrafiltration of orange press liquor: Optimization for permeate flux and fouling index by response surface methodology René Andrés Ruby Figueroa, Alfredo Cassano ⇑, Enrico Drioli Institute on Membrane Technology, ITM-CNR, c/o University of Calabria, via P. Bucci, cubo 17/C, I-87030 Rende (CS), Italy
a r t i c l e
i n f o
Article history: Received 3 February 2011 Received in revised form 31 March 2011 Accepted 31 March 2011 Available online 13 May 2011 Keywords: Ultrafiltration (UF) Orange press liquor Response surface methodology (RSM) Membrane fouling
a b s t r a c t A Box-Behnken design of response surface methodology (RSM) was used to analyze the performance of polysulphone ultrafiltration (UF) hollow fiber membranes in the clarification of orange press liquors. A regression model was developed for permeate flux, fouling index and blocking index as a function of transmembrane pressure (TMP), temperature and feed flow-rate. The experimental operating conditions were selected within the following ranges: TMP 0.2–1.4 bar, temperature 15–35 °C, and feed flow-rate 85–245 L/h. A total of 30 experiments was performed according to the total recycle configuration. Based on the lack-of-fit test, the analysis of variance (ANOVA) showed the regression model to be adequate. From the regression analysis, the permeate flux, fouling index and blocking index were expressed with quadratic equations of TMP, temperature and feed flow-rate. Quadratic terms of TMP, temperature and feed flow-rate showed a significant influence on the performance of the UF membrane. A strong interaction effect of temperature and feed flow-rate was observed on the permeate flux while interactions TMP-temperature and TMP-feed flow-rate were found to be less significant. In the case of fouling index, interactions TMP-temperature and TMP-feed flow-rate produced a significant effect. In order to maximize the permeate flux and minimize the fouling index, the desirability function approach was applied to analyze the regression model equations. The optimized operating variables were found to be 1.4 bar, 15 °C and 167.7 L/h for a maximum desirability of 0.76. Experimental data of permeate flux and fouling index, obtained in optimized operating conditions, resulted in a good agreement with the predicted values of the regression model. Ó 2011 Elsevier B.V. All rights reserved.
1. Introduction Membrane processes are today consolidated systems in various productive sectors, since the separation process is athermal and involves no phase change or chemical agents. The introduction of these technologies in the industrial transformation cycle of the fruit juices represents one of the technological answers to the problem of the production of juices with high quality, natural fresh taste and additive-free. Juice clarification, stabilization, depectinization and concentration are typical steps where membrane processes as microfiltration (MF), ultrafiltration (UF), nanofiltration (NF) and reverse osmosis (RO) have been successfully utilized [1]. UF has been also developed for the treatment of wastewater and sewage in order to remove particulate and macromolecular materials [2–4]. The critical issue in a membrane-based separation process is the decline in permeate flux during its operation. Membrane fouling in ⇑ Corresponding author. Tel.: +39 0984 492067; fax: +39 0984 402103. E-mail address: [email protected] (A. Cassano). 1383-5866/$ - see front matter Ó 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.seppur.2011.03.030
crossflow ultrafiltration is a key factor affecting the economic and technological viability of the processes, which essentially depends on the permeate fluxes obtained and their stability with time [4]. Consequently, the identification and quantification of the prevalent fouling mechanism and efforts to minimize its effects during a continuous filtration process are extremely important [5]. The term membrane fouling is often used to describe a long term flux decline caused by accumulation of certain materials at the membrane surface [6]. It may occur due to a concentration polarization layer development over the membrane surface, the formation of a cake layer and/or a blockage of the membrane pores. The pore blocking can be further characterized by complete, intermediate and standard pore blocking [5]. Several studies are reported in literature about the membrane fouling. Some examples include concentration polarization [7], cake filtration [8–9] and blocking models [10]. In all of these models, the particle size or the ratio of the particle size to the membrane pores is the key parameter for choosing a suitable membrane. Concentration polarization occurs only when the particle diffusion flux is dominant, while the cake filtration model is used when particles
2
R.A. Ruby Figueroa et al. / Separation and Purification Technology 80 (2011) 1–10
have accumulated on the membrane surface. When the particle size is smaller than or comparable to the membrane pores, particles have the opportunity to penetrate and block the pores. In this case, the membrane blocking model may be the appropriate method to relate the received filtrate volume, v, to filtration time, t [11]. Basically, researchers study the membrane separation performance by using the ‘‘one-factor-at-a-time’’ approach. However, the effect of operating variables such as transmembrane pressure, temperature, feed flow-rate and solute concentration on membrane fouling may not be independent of each other and it is necessary to consider their interactions. A possible solution is to use the response surface methodology (RSM) approach that is widely used to analyze the effects of multiple factors and their interactions [12]. RSM is a collection of statistical and mathematical techniques useful for developing, improving, and optimizing processes in which a response of interest is influenced by several variables and the objective is to optimize this response. RSM has important applications in the design, development and formulation of new products, as well as in the improvement of existing product design. It defines the effect of the independent variables, alone or in combination, on the process. In addition, in order to analyze the effects of the independent variables, this experimental methodology generates a mathematical model which describes the chemical or biochemical processes [13–15]. Citrus fruits are widely used in the food industry for the production of fresh and concentrated juice, citrus-based drinks and cans. This process is accompanied by the production of large amounts of by-products such as peels and seed residues which may account for up to 50% of the total fruit weight. Most of the waste residue from commercial juice extractors is shredded, limed, cured and pressed into press liquors and press cakes which are then processed independently. Press liquors are enriched in bioactive compounds, such as flavonoids and phenolic acids, recognized for their beneficial implications in human health due to their antioxidant activity and free radical scavenging ability [16]. They are usually concentrated in multiple effect heat evaporators to yield citrus molasses, while d-limonene is obtained from the condensate. The UF process can be considered a valid approach for separating and recovering valuable products from finely divided solid waste materials present in citrus press liquor. It separates the flow from the press liquor into a permeate having a total soluble solids content and an acidity level approximating those of the press liquor and a retentate containing the suspended solids such as proteins and fibers and high molecular weight carbohydrates such as pectins associated with cloud [17,18]. In this work the objective was to use the RSM approach for investigating the simple effect and the interaction of different operating conditions, such as transmembrane pressure (TMP), temperature and feed flow-rate, on permeate flux, membrane fouling and blocking index of UF polysulphone hollow fibers. From the regression model sets of operating variables were determined for optimizing the clarification of orange press liquors.
2. Theory Hermia [2,5,19–21] developed four empirical models for deadend filtration based on constant pressure filtration laws that correspond to four basic types of fouling: complete blocking, intermediate blocking, standard blocking and cake layer formation. The type of fouling considered depends on the value of the parameter i in the following equation:
i @2t @t ¼K 2 @v @v
ð1Þ
where the blocking index i and the resistance coefficient K are functions of the blocking models. The membrane blocking phenomena can be described by these parameters; for example, i = 2 is complete blocking, i = 1.5 is standard blocking, i = 1 is intermediate blocking, and i = 0 is cake filtration. The blocking index and the resistance coefficient can be calculated by plotting d2t/dv2 versus dt/dv or solving Eq. (1) [11]. In this work a non-linear regression, based on the permeate flux experimental data, was used to obtain the blocking index i for each experimental run. 3. Materials and methods 3.1. Feed solution Citrus press liquors, with a suspended solid content of 7.13 ± 1.41% (w/w), were supplied by Gioia Succhi Srl (Rosarno, Reggio Calabria, Italy). Liquors were left overnight at room temperature to let the majority of the cloud particles settle out. A partially clear liquor was recovered by decanting of the cloud layer. Liquors were then depectinized by adding 7 g/kg of pectinase (4 h at room temperature) from Aspergillus aculeatus (Sigma–Aldrich, Milan, Italy) and filtered with a nylon cloth. Final liquors (feed solution), depleted in suspended solids, had a pH of 3.6 and a total soluble solids content of 8.6 °Brix. They were stored at 17°C and defrosted at room temperature before use. 3.2. UF equipment and procedures UF experiments were performed by using a laboratory unit (Fig. 1) equipped with a hollow fiber polysulphone membrane module supplied by China Blue Star Membrane Technology Co., Ltd (Beijing, China) with an effective membrane area of 0.16 m2 and a nominal molecular weight cut-off (MWCO) of 100 kDa. A heat exchanger, placed into the feed tank, was used to keep the feed temperature constant. Experimental runs were performed according to the total recycle configuration in which permeate and retentate streams were both continuously recycled back to the feed tank to ensure a steady state in the volume and composition of the feed. The operating conditions varied according to the experimental design described in the following. The mass of permeate collected was measured with an accuracy of ±0.1 g every 10 min for periods of 180 min. 3.3. Cleaning procedure The initial hydraulic permeability of the membrane module was 191.22 L/m2hbar at 20 °C. After each run, the membrane module was submitted to a cleaning procedure by using an enzymatic detergent (Ultrasil 50, Henkel Chemicals Ltd., Dussendorf) at a concentration of 1.0% (w/w) and at a temperature of 40 °C for 60 min. After a rinsing with distilled water for 15–20 min, the membrane module was cleaned with a 0.2% (w/w) NaOH solution at 40 °C for 60 min. A final rinsing with distilled water for 20 min was performed. After the cleaning step the hydraulic permeability of the membrane module, in fixed conditions (temperature 20 °C, feed flow-rate 245 L/h), was measured. The recovery of the hydraulic permeability after the cleaning procedure was higher than 89.6%. 3.4. Fouling index The fouling index was calculated by comparing the hydraulic permeability before and after the treatment of orange press liquor according to the following equation:
R.A. Ruby Figueroa et al. / Separation and Purification Technology 80 (2011) 1–10
3
Fig. 1. Scheme of the UF experimental set-up.
Table 1 Experimental range and levels of the independent variables for Box-Behnken design. Feed factors
Code
TMP (bar) Temperature (°C) Feed flow rate (L/h)
Fouling Index ¼
Variation levels
X1 X2 X3
1
L1p L0p
1
0
1
0.2 15 85
0.8 25 165
1.4 35 245
! 100
ð2Þ
where L1p and L0p are the hydraulic permeabilities measured after and before the treatment of the press liquor, respectively [22]. 3.5. Experimental design Based on the capability of the experimental set-up, operating variables were selected within the following ranges: temperature 15–35 °C, feed flow-rate 85–245 L/h and TMP 0.2–1.4 bar (Table 1). Experiments were performed according to the Box-Behnken design (Table 2) that is composed of 30 runs divided in two blocks, each one with three central points (runs 13, 14, 15 and 28, 29, 30). The correlation of the operating variables and the responses based on the Box-Behnken design are fitted to a quadratic polynomial equation [12] using the least-square method of the form:
Y k ¼ b0 þ
3 X i¼1
bi X i þ
3 X i¼1
in which it is assumed that random errors are identically distributed with a zero mean and a common unknown variance and they are independent of each other. The difference between ^) for the ith observation the observed and the fitted value (y e ¼ yi y^ is called the residual and it is an estimate of the corresponding ei . Our criterion for choosing the bi estimates is that they should minimize the sum of the squares of the residuals, which is often called the sum of squares of the errors [23].
bii X 2i þ
2 X 3 X i¼1
bij X i X j þ e
ð3Þ
ji1
where Yk is the response variable (Y1 for permeate flux, Y2 for fouling index and Y3 for blocking index), b0 the constant, e the residual (error or noise) term, bi the linear coefficients, bii the quadratic coefficients, bij the interaction coefficients and Xi is the dimensionless coded variables (X1 for TMP, X2 for temperature and X3 for feed flow-rate). The system of equations given above is solved by using the method of least squares (MLS), a multiple regression technique,
4. Results and discussion The effect of operating variables such as temperature, feed flowrate and transmembrane pressure, on the permeate flux, fouling index and blocking index was investigated by using a Box-Behnken response surface design model. The linear and quadratic effects of the factors studied and the interactions on the responses were obtained using analyses of variance (ANOVA). 4.1. Permeate flux The R-squared statistic indicates that the model as fitted explains 96.31% of the variability in the permeate flux. The lackof-fit test is designed to determine whether the selected model is adequate to describe the observed data, or whether a more complicated model should be used. The test is performed by comparing the variability of the current model residuals to the variability between observations at replicate settings of the factors. Since the P-value for lack-of-fit in the ANOVA is greater or equal to 0.05 (P = 0.0559, F-ratio 5.46), the model appears to be adequate for the observed data at the 95.0% confidence level. The Standard Error was equal to 0.687447. The Durbin–Watson (DW) statistic is a statistic tool used to detect the presence of autocorrelation in the residuals of a regression analysis. Since the P-value is greater than 5.0% (P = 0.9699), there is no indication of a serial autocorrelation in the residuals at the 5.0% significance level. Fig. 2 shows the linear, quadratic and interaction of each factor plotted in the form of Pareto chart where the effect is significant if
4
R.A. Ruby Figueroa et al. / Separation and Purification Technology 80 (2011) 1–10
Table 2 Experimental design and results of Box Behnken design. Run
Block
TMP (bar) X1
Temperature (°C) X2
Feed flow rate (L/h) X3
Permeate flux (Kg/m2h) Y1
Fouling index (%) Y2
Blocking index () Y3
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0
1 1 1 1 0 0 0 0 1 1 1 1 0 0 0 1 1 1 1 0 0 0 0 1 1 1 1 0 0 0
0 0 0 0 1 1 1 1 1 1 1 1 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 0 0 0
8.23 26.41 14.51 31.95 6.81 18.66 10.01 21.08 13.41 18.81 14.43 25.49 16.56 15.55 15.25 8.31 25.33 12.89 30.74 6.77 18.63 8.86 21.05 13.43 18.64 14.37 25.52 16.6 15.52 15.32
58.91 43.01 54.62 75.93 51.86 67.25 62.92 67.00 64.94 67.18 69.76 72.53 68.29 67.76 69.72 59.33 43.48 54.96 76.05 52.31 67.35 62.96 67.49 65.58 67.40 70.74 72.75 68.64 68.37 70.46
1.75 1.74 1.37 1.38 2.00 2.00 1.90 2.00 1.47 1.57 1.64 1.50 1.51 1.62 1.52 1.72 1.65 1.5 1.37 2.00 2.00 2.00 1.55 1.51 1.56 1.60 1.49 1.52 1.58 1.49
b1:TMP
+ -
b2:Temperature b22 b3:Feed Flow Rate b33 b23 b13 b11 b12
0
10
20 30 Standardized effect
40
50
Fig. 2. Standardized Pareto chart for permeate flux.
its corresponding bar crosses the vertical line at the P = 0.05 level. The linear coefficients of TMP (b1) were found to be the most significant effect to increase the permeate flux, followed by linear coefficient of temperature (b2), quadratic effect of temperature (b22) and then the linear effect of feed flow-rate (b3). On the other hand, the quadratic effect of feed flow rate (b33) produces a decrease in the permeate flux with a significant effect (P 6 0.05). The interaction factors as (b12), (b13) and the quadratic effect of TMP (b11) do not produce a significant effect (P P 0.05) in the permeate flux; therefore, for the next analyses these factors will not be included in the regression model equation of the permeate flux. These results indicate that the interaction of operating variables in the ultrafiltration of orange press liquor has to be considered. The quadratic regression equation describing the effect of the process variables on the permeate flux in terms of coded levels is the following:
Y 1 ¼ 15:79 þ 7:34125X 1 þ 3:41437X 2 þ 1:60313X 3 þ 4:01375X 22 þ 1:45X 2 X 3 1:79875X 23
ð4Þ
where Y1 is the predictive permeate flux for the ultrafiltration process. The positive linear coefficients of TMP (b1), temperature (b2) and feed flow-rate (b3) produce an increasing of the permeate flux when these factors are increased. Fig. 3 shows the effect of the investigated variables on the permeate flux; the 3D plot, obtained by using the Eq. (4), helps to predict the response at any combination of operating variables. The response surface of the permeate flux is plotted against two operating variables, while the third variable is kept constant (level 0 in Table 1). The TMP has a linear effect on the permeate flux for all the values investigated (Fig. 3a and c). The curvature in the 3D plot of the permeate flux arises due to the quadratic dependence on
5
R.A. Ruby Figueroa et al. / Separation and Purification Technology 80 (2011) 1–10
Permeate flux (kg/m2*h)
Permeate flux (kg/m2*h)
(a) 40 30 20 10 0 0
0.3
0.6 0.9 TMP (bar)
1.2
1.5
35 31 23 27 19 15 Temperature (°C)
(b) 26 23 20 17 14 11 15
19
23 27 Temperature (°C)
31
35
80
140
200
260
Feed flow rate (L/h)
Permeate flux (kg/m2*h)
(c) 25 20 15 10 5 0 0
0.3
0.6 0.9 TMP (bar)
1.2
1.5
80
140
200
260
Feed flow rate (L/h)
Fig. 3. 3D response surface with contour plot of permeate flux.
Residual Flux (kg/m2*h)
3.1 2.1 1.1 0.1 -0.9 -1.9 -2.9
0
10
20 Predicted Flux (kg/m2*h)
30
40
Fig. 4. Plot of residuals against predicted response of permeate flux in the UF process.
temperature and feed flow-rate (Fig. 3b and c). In addition, the temperature has a linear effect on the permeate flux after 25 °C (level 0). The feed flow-rate produces an increasing of the permeate flux from the level (1) to the level 0; after this point, an increasing in the feed flow-rate produces a decreasing of the permeate flux (Fig. 3b and c). In Fig. 4 the residual (errors) permeate flux was plotted against the predicted flux. The random and also the distribution of the
residuals over and below the centerline suggest that the model for the permeate flux is statistically significant. 4.2. Fouling index The R-squared statistic indicates that the model as fitted explains 90.47% of the variability in the fouling index. The P-value for lack-of-fit was equal to 0.0175 (F-ratio 10.5). This means that
6
R.A. Ruby Figueroa et al. / Separation and Purification Technology 80 (2011) 1–10
b12
+ -
b11 b2:Temperature b1:TMP b3:Feed Flow Rate b13 b22 b33 b23 0
10 15 Standardized effect
5
20
25
Fig. 5. Standardized Pareto chart for fouling index.
Fouling index (%)
(a) 79 69 59 49 39 0
0.3
0.6
0.9
1.2
1.5
35 31 27 23 19 15 Temperature (°C)
TMP (bar)
Fouling index (%)
(b) 77 73 69 65
260 200
61 15
19
23
27
31
35
80
140 Feed flow rate (L/h)
Temperature (°C)
Fouling index (%)
(c) 75 70 65 60 55 50
260 200
45
0
0.3
0.6
0.9
1.2
1.5
80
140 Feed flow rate (L/h)
TMP (bar) Fig. 6. 3D response surface with contour plot of fouling index.
the model as fitted is not completely adequate to represent the experimental data, and a more complex model including other factors should be used. The Standard Error was equal to 1.07706. Since the P-value of the Durbin–Watson statistic test is greater than 5.0% (P = 0.9700), there is no indication of serial autocorrelation in the residuals at the 5.0% significance level. Fig. 5 shows the linear, quadratic and interaction of each factor plotted in the form of Pareto charts where the effect is significant if its corresponding bar crosses the vertical line at the P = 0.05 level.
The interaction factor between TMP and temperature (b12) was found to be the most significant effect to increase the fouling index, followed by quadratic effect of TMP (b11) that produces a decreasing in the fouling index. The interaction factor between temperature and feed flow rate (b23) is the only factor that does not produce a significant effect (P P 0.05) in the fouling index. The quadratic regression equation describing the effect of the process variables on the fouling index in terms of coded levels is reported in the following:
7
R.A. Ruby Figueroa et al. / Separation and Purification Technology 80 (2011) 1–10
5.1
residual
3.1 1.1 -0.9 -2.9 -4.9 43
53
63 predicted
73
83
Fig. 7. Plot of residuals against predicted response of fouling index in the UF process.
b11
+ -
b33 b22 b2:Temperature b23 b1:TMP b13 b3:Feed Flow Rate b12
0
2
4
6 8 Standardized effect
10
12
Fig. 8. Standardized Pareto chart for blocking index.
Y 2 ¼ 68:8733 þ 3:10563X 1 þ 4:10438X 2 þ 2:6425X 3
8:52729X 21
þ 9:26875X 1 X 2 2:7275X 1 X 3
2:05979X 22
þ 2:04646X 23
ð5Þ
where Y2 is the predictive fouling index for the UF process. The positive linear coefficients of TMP (b1), temperature (b2) and feed flow-rate (b3) produce a significant (P 6 0.05) increasing of the fouling index when these factors are increased. The effect of different variables on the fouling index is shown in Fig. 6. The response surface of the fouling index is plotted against two operating variables while the third variable is kept constant (level 0 in Table 1). A strong interaction of TMP and temperature was observed from the warping of the 3D fouling index plot (Fig. 6a). The fouling index increased with temperature only at TMP higher than level 0; at lower level the fouling index decreased by increasing the temperature. The fouling index increased by increasing the feed flow-rate (Fig. 6b and c). The residual (errors) fouling index was plotted against the predicted flux (Fig. 7). The model for the fouling index is statistically significant as suggested by the random and the distribution of the residuals over and below the centerline. 4.3. Blocking index The R-squared statistic indicates that the model as fitted explains 76.36% of the variability in the blocking index. The P-value for lack-of-fit was equal to 0.2555 (F-ratio 1.44). This means that the model appears to be adequate for the observed data at the 95.0% confidence level. The Standard Error was equal to 0.106816. Since the P-value of the Durbin–Watson statistic test is
greater than 5.0% (P = 0.4672) there is no indication of serial autocorrelation in the residuals at the 5.0% significance level. In a membrane process the fouling mechanism is affected by modifications of operating conditions. Fig. 8 shows the linear, quadratic and interaction of each factor plotted in the form of Pareto charts. The linear effect of TMP (b1) and feed flow rate (b3) and the interaction effect for (b23), (b13) and (b12) do not produce a significant effect (P P 0.05) in the blocking index; consequently, these factors were not included in the regression model equations. The quadratic effect for TMP was found to be the most significant effect to increase the value of i (blocking index), followed by the quadratic effect of feed flow rate (b33) and the quadratic effect of temperature (b22). The quadratic regression equation describing the effect of the process variables on the blocking index in terms of coded levels is the following:
Y 3 ¼ 1:54 0:08375X 2 þ 0:204375X 21 0:184375X 22 þ 0:186875X 23
ð6Þ
Membrane pore blocking is theoretically influenced by three factors: the amount of particles simultaneously arriving at the membrane surface, the amount of particle accumulation, and the filtration rate. When more particles arrive at the membrane surface at the same time, they have less opportunity to migrate to the membrane pores due to the crowding out produced by neighboring particles. In contrast, particles have a greater chance to penetrate the pores if fewer particles have accumulated on the membrane surface. Furthermore, the drag force exerted on the particles and the resulting particle motion is significantly determined by the
8
R.A. Ruby Figueroa et al. / Separation and Purification Technology 80 (2011) 1–10
Blocking index (-)
(a)
2
1.8 1.6 1.4 1.2 0
0.3
0.6
0.9
1.2
1.5
15
19
23
27
31
35
Temperature (°C)
TMP (bar)
Blocking index (-)
(b)
2
1.8 1.6 1.4
260 200
1.2 15
19
23
27 Temperature (°C)
31
35
80
140 Feed flow rate (L/h)
(c) Blocking index (-)
2.3 2.1 1.9 1.7
260 200
1.5 0
0.3
0.6
0.9
1.2
1.5
80
140 Feed flow rate (L/h)
TMP (bar) Fig. 9. 3D response surface with contour plot of blocking index.
0.43
residual
0.23
0.03
-0.17
-0.37 1.3
1.5
1.7 predicted
1.9
2.1
Fig. 10. Plot of residuals against predicted response of blocking index in the UF process.
filtrate flow rate. A greater drag force generated by a higher filtration rate may cause the particles to migrate more easily to the membrane pores. However, these factors are not completely independent. The amount of particles simultaneously arriving at the membrane surface is proportional to the filtration rate, while particles accumulation is the integral sum of the instantaneous
particle amount [11]. Fig. 9b and c show the effect of the feed flow-rate on the i value. Increases in feed flow-rate from level 0 until level 1 produce an increasing on the i value; this means that at level 0 the principal mechanism is a combination of intermediate blocking and standard blocking; when the feed flow-rate is increased the blocking index assumes higher values.
9
R.A. Ruby Figueroa et al. / Separation and Purification Technology 80 (2011) 1–10 Table 3 Optimization results for the UF of citrus press liquor. Optimized coded level of variables
X1: TMP X2: Temperature X3: Feed flow-rate
Predicted responses
Permeate flux (kg/ m2h) Fouling index (%)
1 (1.4 bar) 1 (15 °C) 0.034 (167.72 L/ h) 23.73 48.02 0.76
Overall desirability
The response surface plots (Fig. 9) show a strong interaction between feed flow-rate, temperature and TMP. An increasing in TMP produces a decreasing in the value of the blocking index. Hwang et al. [11] observed that the pore blocking is of more significance under lower pressure. The same effect is shown in Fig. 9a and c when at minimum level (1) of TMP the values of the blocking index are closer to i = 2, where the main mechanism involved is the complete pore blocking (i = 2). This may be caused by fewer particles simultaneously arriving at the membrane surface. The model for the blocking index is statistically significant as suggested by the random and the distribution of the residuals over and below the centerline (Fig. 10).
desirability functions from the considered responses are then combined to obtain a total desirability function, D (0 6 D 6 1), defined by Derringer et al. [26] and Del Castillo et al. [27]. The aim of this optimization was to find operating conditions giving the maximum permeate flux and the minimum fouling index simultaneously. The desirability function approach was employed in this procedure. The minimum and maximum limits used for the permeate flux were 6.77 (dk = 0) and 31.95 (dk = 1) kg/m2h, respectively. In addition, the fouling index were 43.01 (dk = 1) and 76.05 (dk = 0)% for the minimum and maximum limit, respectively. For the maximum overall desirability of 0.76, the coded levels of the operating variables are TMP X1 = 1, temperature X2 = 1 and feed flow-rate X3 = 0.034 (Table 3). Fig. 11 shows the time course of permeate flux values experimentally determined in optimized operating variables (T = 15 °C, TMP = 1.4 bar, feed flow-rate = 167.7 L/h). The initial permeate flux of 35.5 kg/m2h was reduced of 26% after 71 min and then reached a steady-state value of 25.52 ± 0.40 kg/m2h. The fouling index evaluated after the treatment of press liquor was 42.35%. These results are in agreement with predicted values of the regression model (permeate flux 23.73 kg/m2h, fouling index 48.02%).
5. Conclusions 4.4. Optimization of multiple responses In RSM, the desirability function (DF) is widely used to determine a combination of variables to optimize multiple responses [24–25]. A DF method finds operating conditions that provide the ‘‘most desirable’’ responses. It is based on the conduction of experiments and fitting response models (yk) for all k responses, the definition of individual desirability functions for each response (dk), and the maximization of the overall desirability in comparison with controllable factors. For each response yk(xi), a desirability function dk(yk) assigns numbers between 0 and 1 to the possible values of yk, with dk(yk) = 0 representing a completely undesirable value of yk and dk(yk) = 1 representing a completely desirable or ideal response value. Depending on whether a particular response yi has to be maximized, minimized or assigned to a target value, different desirability functions dk(yk) can be used. The individual
The effect of operating variables such as transmembrane pressure (TMP), temperature and feed flow-rate on the performance of a UF membrane in the clarification of orange press liquor was studied with the response surface methodology. A Box-Behnken design was used for regression modeling and optimizing the UF operating variables. Optimization of multiple responses permitted to establish the operating conditions giving maximum permeate flux and minimum fouling index, simultaneously. For an overall desirability of 0.76, permeate fluxes of 23.7 kg/m2h and fouling index of 48.0% were estimated, respectively, in optimized operating variables (TMP = 1.4 bar, temperature = 15 °C and feed-flow rate = 167.7 L/ h). Compared with experiments performed at optimized operating conditions, the predicted steady-state permeate flux and fouling index showed differences of 7.1% and 11.7%, respectively.
40.0 37.5 35.0
2
Jp (kg/m h)
32.5 30.0 27.5 25.0 22.5 20.0 0
25
50
75
100
125
150
175
200
Operating time (min) Fig. 11. Time course of permeate flux in optimized operating variables (T = 15 °C; TMP = 1.4 bar; feed flow-rate = 167.72 L/h).
10
R.A. Ruby Figueroa et al. / Separation and Purification Technology 80 (2011) 1–10
Acknowledgment R.A. Ruby Figueroa gratefully acknowledges the Istituto Italo-latino Americano (IILA) which supported his stage at the ITM-CNR for the development of this work.
References [1] A. Cassano, L. Donato, E. Drioli, Ultrafiltration of kiwifruit juice: operating parameters, juice quality and membrane fouling, J. Food Eng. 79 (2007) 613–621. [2] M.C. Vincent Vela, S. Alvarez Blanco, J. Lora García, E. Bergantiños Rodríguez, Analysis of membrane pore blocking models adapted to crossflow ultrafiltration in the ultrafiltration of PEG, Chem. Eng. J. 149 (2009) 232–241. [3] H.K. Shon, S. Vigneswaran, I.S. Kim, J. Cho, H.H. Ngo, Effect of pretreatment on the fouling of membranes: application in biologically treated sewage effluent, J. Membr. Sci. 234 (2004) 111–120. [4] M. Cheryan, J.R. Alvarez, Food and Beverage industry applications, in: R.D. Noble, S.A. Stern (Eds.), Membrane Separation Technology–Principles and Applications, Elsevier, AmsterdamThe Netherlands, 1995, p. 341. [5] R.W. Field, D. Wu, J.A. Howell, B.B. Gupta, Critical flux concept for microfiltration fouling, J. Membr. Sci. 100 (1995) 259–272. [6] H. Strathmann, L. Giorno, E. Drioli, An introduction to membrane sciences and technology, Consiglio Nazionale delle Ricerche, Roma, 2006. [7] M. Cheryan, Ultrafiltration and Microfiltration Handbook, Technomic Publishing Co., Pennsylvania, USA, 1998. [8] W.M. Lu, C.C. Lai, K.J. Hwang, Constant pressure filtration of submicron particles, Sep. Technol. 5 (1995) 45–53. [9] C. Tien, B.V. Ramarao, On analysis of cake formation and growth in cake filtration, J. Chin. Inst. Chem. Eng. 37 (1) (2006) 81–94. [10] P.H. Hermans, H.L. Bredee, Principles of the mathematical treatment of constant-pressure filtration, J. Soc. Chem. Ind. 55T (1936) 1–11. [11] K.J. Hwang, C.Y. Liao, K.L. Tung, Analysis of particle fouling during microfiltration by use of blocking models, J. Membr. Sci. 287 (2007) 287–293. [12] M.N. Hyder, R.Y.M. Huang, P. Chen, Pervaporation dehydration of alcohol– water mixtures: optimization for permeate flux and selectivity by central composite rotatable design, J. Membr. Sci. 326 (2009) 343–353.
[13] M.F. Anjum, I. Tasadduq, K. Al-Sultan, Response surface methodology: a neutral network approach, Eur. J. Oper. Res. 101 (1997) 65–73. [14] R.H. Myers, D.C. Montgomery, Response Surface Methodology: Process and Product Optimization Using Designed Experiments, John Wiley Sons, Inc., New York, 1995. [15] M.C. Martí-Calatayud, M.C. Vincent-Vela, S. Álvarez-Blanco, J. Lora-García, E. Bergantiños-Rodríguez, Analysis and optimization of the influence of operating conditions in the ultrafiltration of macromolecules using a response surface methodological approach, Chem. Eng. J. 156 (2010) 337–346. [16] A. Bocco, M.E. Cuvelier, H. Richard, C. Berset, Antioxidant activity and phenolic composition of citrus peel and seed extract, J. Agric. Food Chem. 46 (1998) 2123–2129. [17] B.R. Breslau, E.A. Agranat, A.J. Testa, S. Messinger, R.A. Cross, Hollow Fiber Ultrafiltration. A Systems approach for Process Water and By-Product Recovery, in: Proc. of 79th National Meeting of the AIChE, Houston, Texas, March 1975. [18] R.J. Braddock, Ultrafiltration and reverse osmosis recovery of limonene from citrus processing waste streams, J. Food Sci. 47 (1982) 946–948. [19] J. Hermia, Constant pressure blocking filtration law: application to power law non-Newtonian fluids, Trans. Inst. Chem. Eng. 60 (1982) 183–187. [20] C. Duclos-Orsello, W. Li, C.C. Ho, A three mechanism model to describe fouling of microfiltration membranes, J. Membr. Sci. 280 (2006) 856–866. [21] M.C. Vincent Vela, S. Alvarez Blanco, J. Lora García, E. Bergantiños Rodríguez, Analysis of membrane pore blocking models applied to the ultrafiltration of PEG, Sep. Purif. Technol. 62 (2008) 489–498. [22] M. Mänttäri, M. Nyström, Membrane filtration for tertiary treatment of biologically treated effluents from the pulp and paper industry, Water Sci. Technol. 55 (2007) 99–107. [23] D. Basß, I.H. Boyaci, Modeling and optimization I: usability of response surface methodology, J. Food Eng. 78 (2007) 836–845. [24] S.J. Kalil, F. Maugeri, M.I. Rodrigues, Response surface analysis and simulation as a tool for bioprocess design and optimization, Process Biochem. 35 (2000) 539–550. [25] E.C. Harrington, The desirability function, Ind. Qual. Control 21 (1965) 494–498. [26] G. Derringer, R. Suich, Simultaneous optimization of several response variables, J. Qual. Technol. 12 (1980) 214–219. [27] E. Del Castillo, D.C. Montgomery, D.R. McCarville, Modified desirability function for multiple response optimization, J. Qual. Technol. 28 (1996) 337–345.