DRAGAN D. GOVEDARICA SEPARATION OF OIL-IN-WATER …

Chemical Industry & Chemical Engineering Quarterly

Available on line at Association of the Chemical Engineers of Serbia AChE www.ache.org.rs/CICEQ

Chem. Ind. Chem. Eng. Q. 22 (3) 309−318 (2016) CI&CEQ

309

DRAGAN D. GOVEDARICA1

RADMILA M. ŠEĆEROV SOKOLOVIĆ1

OLGA M. GOVEDARICA1

DUNJA S. SOKOLOVIĆ2

SNEŽANA V. SINADINOVIĆ-FIŠER1

1Faculty of Technology, University of Novi Sad, Novi Sad, Serbia

2Faculty of Technical Sciences, University of Novi Sad, Novi Sad,

Serbia

SCIENTIFIC PAPER

UDC 66.067:544:66

DOI 10.2298/CICEQ150604045G

SEPARATION OF OIL-IN-WATER EMULSIONS BY FLOW THROUGH FIBER BEDS: A RESPONSE SURFACE APPROACH

Article Highlights • Polyethylene terephthalate fiber bed was used successfully for coalescence filtration • A novel approach for the estimation of the effluent oil concentration and the critical

velocity • A good correlation of experimental data was obtained with response surface regres-

sion model • Higher critical velocities were observed for high oil viscosity and high bed perme-

ability Abstract

Separation of oil-in-water emulsions using fiber bed coalescer was studied by response surface methodology. The bed was formed of polyethylene tereph-thalate fibers. The aim was to investigate the possibility of using response surface regression for the analysis and prediction of both the effluent oil con-centration and critical velocity in broad ranges of flow rates, bed permeabilities, and nature of dispersed oil phase. The developed response surface equations for a constant bed permeability are a responsive statistical method with the calculated multiple R higher than 90%. It was found that dispersed oil density, viscosity, neutralization number, and bed permeability influence significantly the oil removal efficiency of fiber bed coalescers. The region of highest critical velocity has been observed for a high viscosity and a high fiber bed perme-ability. The developed response surface models can be used to ensure high separation efficiency and improve coalescer performance when unexpected simultaneous changes in the bed permeability, hydrodynamic forces and dis-persed oil properties take place.

Keywords: liquid-liquid separation; oily water; bed coalescence; fiber material; critical velocity; response surface regression.

Emulsion separation is commonly required in the petroleum industry for treatment of wastewater and solvents, as well as for dehydration of crude oil and petroleum products such as diesel, gasoline, and kerosene [1-6]. A frequently applied method for breaking emulsions is fiber bed coalescence, which may operate stand-alone or integrated with other separation techniques [7,8]. Fiber bed coalescers are common separation equipment when it comes to polishing both of the two liquid phases involved in the

Correspondence: D.D. Govedarica, Faculty of Technology, Department of Oil and Petrochemical Engineering, University of Novi Sad, Bulevar cara Lazara 1, 21000 Novi Sad, Serbia. E-mail: [email protected] Paper received: 4 June, 2015 Paper revised: 12 November, 2015 Paper accepted: 13 November, 2015

emulsion. Coalescers are frequently located down-stream of the gravity settlers or decanters, to increase the quality of the fluid or meet environmental dis-charge limits. The coalescers are usually used for removal of small-size droplets smaller than 100 μm in the course of separation of water from petroleum products and treatment of quench water [1]. To avoid the uncertainty of the design procedure, each par-ticular application of the bed coalescer demands ext-ensive pilot testing over the range of expected work-ing conditions. Reliable mathematical models for pro-cess optimization are scarce because of the complex phenomena involved in the fiber bed coalescence: the diverse mechanisms of coalescence, the filter media structure, and the emulsion flow through the bed formed of fibers [9].

D.D. GOVEDARICA et al.: SEPARATION OF OIL-IN-WATER EMULSIONS… Chem. Ind. Chem. Eng. Q. 22 (3) 309−318 (2016)

310

Some mathematical models take into account the effect of hydrodynamic, attraction and repulsion forces on the coalescence efficiency [10]. Other models assume that the coalescence and breakage of droplets are dominantly governed by the fiber structure, topology of the fiber bed, and emulsion flow through the filter media [11].

Sareen et al. [12] and Grilc et al. [13,14] have pointed out the importance of fluid velocity to emul-sion flow as it controls the coalescence mechanisms and probability of droplet coalescence. Some res-earchers consider also the droplet size as a highly significant parameter for adjusting the coalescer oper-ating variables [15,16]. Fahim and Akbar [17] con-cluded that the coalescer separation efficiency is greatly dependent on fluid velocity, influent oil con-centration and bed length. However, some other authors have confined to a relatively narrow range of working conditions, and therefore the conclusions were contradictory [12,18,19].

Based on the available mathematical models for prediction of the coalescence efficiency, it can be concluded that occurrence of each coalescence mechanism has a certain probability [10,11,16,20-23]. The analysis by Ryan and Elimelech [10] suggested that some filtration mechanisms may be analyzed analogously to pseudo-first-order chemical reactions. Since the bed coalescence efficiency is dependent on numerous factors, it is expected that good correlation can be achieved with appropriate statistical models.

Factorial design of experiments and response surface methodology were used to analyze different separation processes for oily wastewaters, such as filtration [24], membrane processes [25-27], adsorp-tion [28] and resin bed coalescence [29]. The advent-age of using statistical methods for data correlation, such as response surface methodology, is in the red-uction of the number of pilot scale experiments.

Response surface methodology has been used by Kundu et al. [29] to identify significant interaction effects between experimental parameters for separ-ation of diesel oil/water emulsion using a resin bed coalescer. The granular bed was formed of spherical particles, and a second-order response surface model showed that the oil removal efficiency is a function of bed length, flow rate, and pH.

The usage of polyethylene terephthalate poly-mer fibers for bed coalescence has been a subject of numerous studies concerning the influences of super-ficial velocity, bed permeability and porosity, fiber sur-face roughness, and inflow direction on the phase separation efficiency of oil-in-water emulsions [18,30- -36]. However, there are no published results in which

response surface method was used to analyze the operation of fiber bed coalescer.

Therefore, the aim of this work was to inves-tigate the possibility of using response surface regres-sion for the analysis and prediction of separation effi-ciency of fiber bed coalescence in a broad range of working parameters. The goal was to develop reliable response surface models that would ensure high separation efficiency and improved coalescers per-formances when unexpected changes in the bed per-meability, hydrodynamic forces and dispersed oil pro-perties take place simultaneously.

EXPERIMENTAL

Experimental setup of the fibrous bed coalescer and operating procedure

The experiments were performed on a labor-atory-scale bed coalescer with the horizontal fluid flow orientation, whose design was described in detail in a previous paper [37]. Crude oil (A), naphtenic-base vacuum distillation fractions (A1 and A4), and petro-leum product with a high paraffinic content (P1) were used as the dispersed phase for bed coalescence experiments. The dispersed phases were mineral oils of different properties containing natural emulsifiers such as asphaltenes. Oil droplets were dispersed in tap water by adding measured amounts of oil to the supply tank. The oil-in-water model emulsions with constant oil concentration (500 mg/L) were prepared in a supply tank (80 L), by continuous agitating with a stainless steel impeller (650 rpm). In order to ensure the inlet mean droplet diameter of about 10 µm, each oily sample was continuously stirred 45 minutes prior to the experiment and onwards until the end of expe-riment. The mean inlet droplet size of the number distribution was dependent of the nature of the dis-persed oil phase and it was determined by an Elzone 280 PC particle counter and an Olympus BH.2 RFCA microscope:

• 9-10 µm (min. 0.8 µm, max. 31 µm) for oil A/water,

• 10-12 µm (min. 0.9 µm, max. 33 µm) for oil A4/water,

• 10-11 µm (min. 0.7 µm, max. 26 µm) for oil A1/water and

• 9-10 µm (min. 0.9 µm, max. 28 µm) for oil P1/water.

Hence, the influent droplet size did not vary significantly. The steady-state regime of bed coales-cence was achieved from the very beginning of the experiment by pre-oiling the polyethylene terephthal-ate fibers. A steady state was confirmed by moni-


311

toring the pressure drop, which did not change with time. The filter media were polyethylene terephthalate (polyester) fibers. In a coalescence experiment, the following parameters were kept constant: bed length (5 cm), bed permeability, and working temperature (20 °C). Each oily water sample was tested on five bed permeabilities. The oil-in-water emulsion was pumped through a membrane dosage pump at the fluid velocities ranging from 18 to 70 m/h. The sel-ected velocity was kept constant for 1 h. Composite samples of oil-in-water emulsion were collected at the sampling point downstream of the fiber bed after 45 min at 5-min intervals.

Properties of dispersed oils

Four different kinds of dispersed oils with a wide range of physical and chemical properties were used (Supplementary material, available from the authors upon request). Density was determined according to ISO 3675. Kinematic viscosity was measured using glass capillary viscometers according to the standard ISO 3104. Neutralization number was determined by potentiometric titration (ISO 6619). The mean mole-cular weight was estimated according to standard method ASTM d 2502-67 from kinematic viscosity measurements. Interfacial tension and surface ten-sion measurements were done according the du Noüy ring method and stalagmometric method, res-pectively.

Properties of the bed

The bed was formed of waste polyethylene terephthalate fibers that remained after cutting out blocks of the filter for kitchen exhaust hood. The poly-ethylene terephthalate fibers were needle-punched and non-woven. The orientation of the fibers was random. The surface morphology and size of the fibers were characterized by scanning electron micro-scopy. The used fibers were smooth and staple with average length of 50 mm and average diameter of 60 μm. The fibers had a circular cross-section profile. Experiments were realized in a broad range of bed porosity (0.90-0.98) and solid surface (1.330-6.670 mm-1). Due to the compressibility of polyethylene terephthalate fibers, it was possible to vary the bed permeability over a wide range (0.18×10-9-5.389×10-9 m2). The bed permeability was calculated from the measured pressure drop across the bed for tap water, and the data complied with Darcy's law. Numerical values of the investigated bed permeabilities are available from the authors upon request.

Effluent oil concentration

Samples were stabilized and adjusted to pH 2 by adding HCl. Each emulsion sample was extracted with CCl4. The effluent oil concentration was deter-mined by FTIR spectrometry using a ThermoNicolet 5700 spectrometer. To determine the oil in water con-tent, the peaks with maxima at about 2930 (–CH2 groups), 2960 (–CH3 groups) and 3030 cm-1 (C–H aro-matic groups) were considered. The effluent oil con-centration was followed as a function of the fluid vel-ocity, which was increased until the effluent oil con-centration was higher than 15 mg/L.

Critical velocity (Vk)

The critical velocity in our experiments was defined as the fluid velocity when the effluent oil con-centration reaches 15 mg/L [38-40]. The value of the critical velocity was determined from the dependence of the effluent oil concentration on the superficial velocity of the emulsion.

Statistical analysis

The statistical analysis was based on 92 expe-rimental values of effluent oil concentration and 20 values of the critical velocity extracted from the raw experimental data. The fiber bed coalescence experi-ments were carried out over a wide range of dis-persed oil properties, bed permeability, and fluid vel-ocity. Because the concentration of the coalescer effluent may not be stable, each experiment was repeated twice. Experimental data were analyzed by response surface regression (RSR) and analysis of variance (ANOVA). Statistical analysis was performed using statistical package Statistica 12, StatSoft.

RESULTS AND DISCUSSION

Introduction

The experimental data correspond to a coal-escer whose bed permeability, hydrodynamic forces and dispersed oil properties varied simultaneously during the operation. Some changes in the bed permeability can be caused by deposition of some particulate matter, such as corrosion products, impu-rities, solid waste, surfactants, etc. Failure of the bed coalescer function can also be caused by ineffective pre-filtration of inflow emulsion and porosity reduction due to high holdup or fluctuations of dispersed phase. Emulsion flow to the bed coalescer can greatly vary depending on the operational modes of the equip-ment upstream the coalescer. Since such variations of operational conditions are common in the petro-leum industry, their influence on the efficiency of fiber


312

bed coalescence has to be examined. Therefore, the effects of dispersed oil properties, fluid velocity, and bed permeability on the effluent oil concentration were followed based on a statistical approach using RSR.

Dependence of the effluent oil concentration on the fluid velocity, bed permeability and dispersed oil properties

Experimental values for dispersed oil properties, bed permeability, fluid velocity and effluent oil con-centration are available from the authors upon request. Because the variables are not measured on the same scale, the natural logarithms of experimen-tal data values were fitted using an RSR model. The second-order response surface models are widely used because they are very flexible statistical methods for an approximation of the true response surface. There is considerable practical experience showing that these models work well in real separ-ation problems [24-29]. Hence, the experimental data in this work were fitted with response surface regres-sion using the following second-order model:

α α α

α

= =

<

= + + +

+

20

1 1ln ln ln

ln ln

n n

i i i ii ii i

n

ji i jj i

C X X

X X (1)

This nonlinear model includes the intercept, linear, quadratic, and interaction terms. The response is lnCi (effluent oil concentration), and lnXi and lnXj are the independent variables, i.e., factors. In this paper, the independent variables for the estimation of the effluent oil concentration were: bed permeability, viscosity, density, neutralization number, mean mole-cular weight, interfacial tension, surface tension, and

fluid velocity. The coefficient α0 is the intercept, whereas αi, αii and αji are the linear, quadratic and interaction regression coefficients, respectively. By applying the response surface regression analysis, the following equation was obtained:

μ μ ρμ

ρμ

ρ

= − + − −

− + + +

+ + + −− + ++ − −

+

20 0

2

20

0 0

0

ln 2610.85 1038.86ln 0.19(ln )

34.69ln 4.32(ln ) 380.44ln

950.3ln 3.96(ln ) 0.22ln ln154.12ln ln 2.56ln ln0.63ln ln 0.54ln ln

-141.26ln ln 2.75ln ln

i

b

b

C K K

V V KK K N

K V VV N V

(2)

The calculated multiple R was 76.72%, multiple R-squared was 58.86%, and adjusted R-squared was 51.38%. The deviations between the experimentally determined lnCi values and values predicted using response surface regression equation (2) are shown in Figure 1. Equation (2) can be used to predict the coalescer performances for a wide range of flow rates, bed permeabilities, and nature of the dispersed oil phase.

The relatively narrow 95% confidence interval was a result of the simultaneous effect of hydrodyn-amic forces on coalescence and occurrence of several bed coalescence mechanisms. In order to identify significant factors and evaluate those that have pertinent effect on the coalescence efficiency, ANOVA analysis was performed. This statistical method is commonly used for the hypothesis testing on the regression model and coefficients. The cal-culated Fisher variance ratio (F-value) and signific-ance probability value (p-value) of the RSR model were 7.86 and 0.00, respectively. The high F-value implies that most of the variation in the response can

Figure 1. Deviation between the calculated and experimental values of the effluent oil concentration for all investigated permeabilities.


313

be explained by Eq. (2). The low p-value (less than 0.05) indicates that the model terms are significant at the 95% confidence level.

In order to evaluate the effect of the linear, quadratic, and interaction terms, a Pareto chart of the calculated Student’s t-test (t-values) for the coef-ficients was constructed (Figure 2). Based on sigma-restricted parameterization, the Pareto chart illus-trates the most important effects of the factors. It should be noted that the higher the t-value of the effect, the higher the influence on the effluent oil con-centration response. Hence, a factor coefficient with a significance p-value less than or equal to 0.05 is con-sidered as statistically significant and a reference line is drawn [24,28,29]. Considering the accepted limit, it can be noted that any effect that falls beyond the reference line in Figure 2 is significant. The quadratic effects of fluid velocity and bed permeability had the greatest effect on the coalescence efficiency. Further-more, the interaction terms representing the effects of bed permeability, neutralization number, and density, as well as the linear terms of density, viscosity and bed permeability are significant. The quadratic effects of fluid velocity, bed permeability and viscosity cause the effluent oil concentration response surface to have a curved shape.

The 3D plot depicts better the interdependence of the effluent oil concentration, bed permeability and fluid velocity (Figure 3). It is clear that the region of the highest bed permeability and lower fluid velocity corresponds to the minimal effluent concentration. This region ensures the desired effluent concentration and separation efficiency. On the other hand, many research studies identified the region of low effluent oil concentration corresponding to the lowest bed per-meability [41]. How can we explain the observed trend

Figure 3. Three-dimensional plot representing the inter-

dependence of the effluent oil concentration, fluid velocity and bed permeability.

in the lower effluent oil concentration associated with the highest bed permeability? In our previous inves-tigations [36,38,42] we have shown that the low effluent oil concentration is dominantly influenced by bed permeability and coalescence mechanisms that occur in a fiber bed. It is plausible to suppose that the different bed coalescence mechanisms are predom-inant in low and high dense packed fiber bed. The highest bed permeability can provide the domination of the coalescence of the droplets into the surface of the saturated liquid i.e., a capillary-conducted fluid. On the contrary, the lowest bed permeability contri-butes to a domination of the coalescence of the drop-lets on the fibers surface. As can be seen, the effluent oil concentration increases with an increase in fluid

Figure 2. Pareto chart for standardized effects for effluent oil concentration for all investigated permeabilities.


314

velocity. The plot illustrates the region of high fluid velocities, corresponding to the highest standard devi-ation of the effluent oil concentration.

The Pareto chart and 3D interdependence of the effluent oil concentration, bed permeability and fluid velocity indicate that the separation efficiency is sig-nificantly influenced by bed permeability. The change of polyethylene terephthalate bed permeability essen-tially influences: emulsion flow velocity, droplet deformation, coalescence of the adjacent droplets, coalescence of the droplets on the fibers surface, and coalescence of the droplets into the surface of the saturated oil [9,31,42].

Estimation of effluent oil concentration over constant bed permeability

Since the ANOVA analysis and RSR demons-trated that the effluent oil concentration is dominantly influenced by bed permeability, further analysis was conducted to verify the response surface models over a fixed bed geometry. Additionally, the analysis for a defined bed permeability is valuable for the inves-tigation of the working parameters for coalescers with integrated pre-filtration.

This approach enabled a more precise esti-mation of the effluent oil concentration over a cons-tant bed permeability using RSR model with signific-antly higher multiple R, multiple R-squared and F-value (Table 1).

For all investigated bed permeabilities, the cal-culated multiple R was higher than 90%. It should be also noted that the p-values of the RSR models were considerably lower than 0.05, indicating statistical sig-nificance of the presented models.

For the highest bed permeability, the calculated multiple R was 95.42%, multiple R-squared 91.06%, and F-value 14.00. The R-squared indicates that 91.06% of the variability within the range of values studied can be explained by the model. The lowest p-value indicates that for the set of the presented bed coalescence experiments, observed data are ade-quately fitted to the second-order polynomial model. The best fit of this model can be explained by the dominance of the coalescence of droplets into the surface of the saturated oil. Due to the high porosity, a significant amount of saturated liquid, i.e. capillary-conducted phase, is being formed. In such cases, the oil droplets coalescence on the surface of the capil-lary-conducted phase becomes the dominant mech-anism of coalescence. For a steady-state regime, Spielman and Goren observed 30 to 40% of the pores are occupied by the capillary-conducted phase [9]. The amount of saturated oil influences the distribution of certain coalescence mechanisms of all finely dis-persed oil droplets upstream of the fiber bed, and coalescence of the adjacent droplets in the pore space. Experimentally determined and calculated effluent oil concentrations over the highest bed per-meability are shown in Figure 4. The 95% confidence intervals of the RSR model for the highest polyethyl-ene terephthalate permeability include the bed coal-escence experiments carried out in the range of lower fluid velocities, which greatly decreased the intensity of the hydrodynamic forces (Figure 4, Table 1).

The ANOVA and Pareto analysis for the highest bed permeability showed that the quadratic term of fluid velocity and the linear term of oil density had the

Table 1. RSR equations and R, R2, F-value and p-value for estimation of effluent oil concentration over constant bed permeability

Bed permeability, m2 Second order RSR equation R R2 F-value p-value

K01 = 5.389×10-9 μ μ ρ

μρ

= + − − +

+ + − −− +

2

2

b

ln 6097.21 70.18ln 7.24(ln ) 898.27ln

9813.54ln 10.36ln 0.29ln ln1451.02ln ln 24.90ln ln

iC

V V VV N V

95.42 91.06 14.00 0.000093

K02 = 2.426×10-9 μ μ ρ

ρ

= + − − +

+ + − +

2

2b

ln 3573.69 84.18ln 10.77(ln ) 538.59ln

15723.28ln 5.13ln 2315.63ln ln 35.74ln lniC

V V V N V

92.66 85.87 5.32 0.020116

K03 = 1.128×10-9 μ μ ρ

μρ

= + − − +

+ + − −− +

2

2

b

ln 6300.764 40.963ln 3.368(ln ) 928.272ln


iC

V V VV N V

92.45 85.47 7.35 0.002467

K04 = 0.380×10-9 μ μ ρ

μρ

= + − − +

+ + + −− +

2

2

b

ln 7302.99 88.44ln 9.85(ln ) 1082.86ln


iC

V V VV N V

93.57 87.55 9.67 0.000521

K05 = 0.180×10-9 μ μ ρ

μρ

= + − − +

+ + −− +

2

2

b

ln 379.438 1.475ln 1.261(ln ) 38.419ln

1672.284ln 8.645ln +3.044ln ln255.845ln ln 3.844ln ln

iC

V V VV N V

90.90 82.64 4.76 0.020380


315

highest effect on the effluent oil concentration. Con-cerning the accepted limit of the p-value less than or equal to 0.05, it can be stated that the quadratic term of fluid velocity represents the influence of hydrodyn-amic forces on the effluent oil concentration, while the linear term of density represents the effects of the dispersed oil cross-sectional distribution and settling velocity in the coalescer housing [42]. The ANOVA and Pareto analysis revealed that the interaction term of neutralization number, linear term of viscosity, interaction term of density and fluid velocity, linear term of velocity, as well as the quadratic term of vis-cosity, had a high effect on the removal of dispersed oil droplets. This means that these terms are statist-ically significant for the estimation of the effluent oil concentration over constant bed permeability.

Estimation of the critical velocity

In the petroleum industry, the goals of coalescer design are the acceleration of emulsion separation and equipment size minimization. A bed coalescer designed to operate efficiently at high fluid velocities would be much more effective in oily wastewater treating facilities. In this work, the effects of nature of dispersed oil phase and bed permeability were anal-yzed using response surface methodology. Experi-mental values of dispersed oil properties, bed per-meability, and the critical velocity available from the authors upon request.

The logarithmic experimental data were fitted with RSR using the following second-order model:

α α α α= = <

= + + + 2k 0

1 1ln ln ln ln

n n n

i i ii i ji i ji i j i

V X X X X (3)

The response surface model includes the inter-cept, linear, quadratic and interaction terms. The independent variables (factors) were ln Xi and ln Xj, and the response was the critical velocity term ln Vk. The effluent oil concentration was estimated for the following independent variables: density, viscosity, neutralization number, mean molecular weight, interfacial tension, surface tension and bed perme-ability, and the following RSR equation was deve-loped:

μ μ ρμ ρ

= − − +

+ − − ++ + −

2k 0 0

2

0 0 0 b

ln 1137.218 331.66ln 0.005(ln )

16.071ln 1.684(ln ) 171.685ln0.05ln ln 48.653ln ln 0.802ln ln

V K K

K K K N (4)

The calculated multiple R was 92.50%, multiple R-squared 85.55%, and adjusted R-squared 75.05%. As already mentioned, the high calculated Fisher vari-ance (F-value) of 8.1428 implies that most of the variation in the critical velocity can be explained by Eq. (4). The Model F-value of 8.1428 implied that the model was significant. The model p-value was very low (0.001119), proving the significance of the model. Equation (4) gives the possibility to estimate the values of the critical velocity for a given set of factors, and it may be useful for the estimation of the upper limit of the critical velocity for a successful operation.

In order to confirm the normal distribution of the data, a normal probability plot of the residuals was analyzed (Figure 5). Since the points on the normal probability plot fall approximately on a straight line, the regular residuals were normally distributed [28].

The ANOVA and Pareto analysis illustrated that the linear terms of density and viscosity, as well as the quadratic term of viscosity, had the highest effect on the critical velocity. It was found that these terms

Figure 4. Relationship between the calculated and experimental values of the effluent oil concentration over K01 for all investigated oils.


316

have a more significant effect on the coalescence compared to the other parameters. The dispersed oil viscosity influences the flow of unstable emulsions through the bed and dominance of the competitive coalescence mechanisms.

As can be seen from the interdependence of the critical velocity, bed permeability and viscosity (Figure 6), the region of highest critical velocity can be observed for high viscosity and high fiber bed per-meability. In this region, the critical velocity is above 50 m/h.

Figure 6. Three-dimensional plot representing the interdep-

endence of the critical velocity, viscosity, and bed permeability.

CONCLUSION

In this work, the effects of dispersed oil pro-perties, fluid velocity, and bed permeability on the

effluent oil concentration were followed based on a statistical approach using RSR, and several quadratic nonlinear models were developed. Based on the RSR and sigma-restricted parameterization, the effluent oil concentration was found to be a function of fluid vel-ocity, bed permeability, viscosity, density and neutral-ization number. Since the RSR and ANOVA analysis demonstrated that the effluent oil concentration is dominantly influenced by bed permeability, further analysis was conducted to verify response surface models over constant bed geometry. The response surface models over constant bed permeability were a responsive statistical method, with the calculated multiple R higher than 90%.

An RSR equation was developed for predicting the critical velocity. It was found that dispersed oil density, viscosity, neutralization number and bed per-meability influence significantly the oil removal effi-ciency of fiber bed coalescers. The region of highest critical velocity can be observed for high viscosity and high fiber bed permeability.

The presented mathematical models can be used to predict the coalescer performances under a great range of flow rates, bed permeabilities, and nature of dispersed oil phase.

Acknowledgment

The work was supported by the Ministry of Education, Science and Technological Development of the Republic of Serbia, Grant number 172022.

NOMENCLATURE

K0 = bed permeability, m2 V = fluid velocity, m/h

Figure 5. Normal probability plot of the residuals of the RSR.


317

Vk = critical velocity for the effluent oil concentration of 15 mg/l, m/h Ci = effluent oil concentration, mg/ l Nb = neutralization number of dispersed oil, mg KOH/l M = mean molecular weight of dispersed oil, kg/kmol

Greek letters

μ = dispersed oil viscosity, mPas ρ = dispersed oil density, kg/m3 γ = surface tension, mN/m σ = interfacial tension, mN/m

Abbreviations

RSR, response surface regression ANOVA, analysis of variance

REFERENCES

[1] G. Viswanadam, G. G. Chase, Sep. Purif. Technol. 104 (2013) 81–88

[2] M. Shaohua, Y. Kang, S. Cui, J. Dispersion Sci. Technol. 35 (2014) 103–110

[3] M. Stewart, K. Arnold, Emulsions and Oil Treating Equip-ment: Selection, Sizing and Troubleshooting, Elsevier, New York, 2009, pp. 158–168

[4] J. Li, Y. Gu, Sep. Purif. Technol. 42 (2005) 1–13

[5] Q. Yao, M. Jia, X. Wang, J. Men, Fiber. Adv. Mater. Res. 233–235 (2011) 652–657

[6] A.Q. Clark, A.G. Smith, S.E. Taylor, Ind. Eng. Chem. Res. 50 (2011) 5749–5765

[7] A. Motta, C. Borges, K. Esquerre, A. Kiperstok, J. Membr. Sci. 469 (2014) 371–378

[8] M. Cheryan, N. Rajagopalan, J. Membr. Sci. 151 (1998) 13–28

[9] L.A. Spielman, S.L. Goren, Ind. Eng. Chem. 62 (1970) 10–24

[10] J.N. Ryan, M. Elimelech, Colloids Surfaces, A 107 (1996) 1–56

[11] W. Orciuch, I. Makowski, A. Moskal, L. Gradon, L. Chem. Eng. Sci. 68 (2012) 227–235

[12] S.S. Sareen, P.M. Rose, R.C. Gudesen, R.C. Kintner, AICHE J. 12 (1966) 1045–1050

[13] V. Grilc, J. Golob, R. Modic, Chem. Eng. Res. Des. 62 (1984) 48–52

[14] V. Grilc, L. Golob, R. Modic, Chem. Eng. Res. Des. 64 (1986) 67–70

[15] L.A. Spielman, Separation of Finely Dispersed Liquid- –Liquid Suspensions by Flow Through Fibrous Media. Ph.D. Thesis, University of California, Berkeley, CA, 1968

[16] D.F. Sherony, R.C. Kintner, Can. J. Chem. Eng. 49 (1971) 314–320

[17] M.A. Fahim, A.M. Akbar, J. Environ. Sci. Health 19 (1984) 299–319

[18] J.N. Chieu, R.S. Schechter, E.F. Gloyna, J. Environ. Eng. Div. (Am. Soc. Civ. Eng.) 103 (1977) 163–175

[19] E.J. Clayfield, A.G. Dixon, J. Colloid Interface Sci. 104 (1985) 512–519

[20] H. Speth, A. Pfennig, M. Chatterjee, H. Franken, Sep. Purif. Technol. 29 (2002) 113–119

[21] C. Shin, G.G. Chase, AIChE J. 50 (2004) 343–350

[22] E. Dahlquist, F. Setterwall, Sep. Technol. 3 (1993) 198–211

[23] D.F. Sherony, R.C. Kintner, Can. J. Chem. Eng. 49 (1971) 314–320

[24] A. Ur Rahmah, M.A. Abdullah, Desalination 266 (2011) 51–55

[25] A. Fouladitajar, F.Z. Ashtiani, B. Dabir, H. Razaei, B. Valizadeh, Environ. Sci. Pollut. Res. 22 (2015) 2311–2327

[26] M. Khayet, M.N. Abu Seman, N. Hilal, J. Membr. Sci. 349 (2010) 113–122

[27] M. Khayet, C. Cojocaru, M. Essalhi, J. Membr. Sci 368 (2011) 202–214

[28] A. Srinivasan, T. Viraraghavan, J. Hazard. Mater. 175 (2010) 695–702

[29] P. Kundu, I.M. Mishra, Sep. Purif. Technol. 118 (2013) 519–529

[30] S. Agarwal, V. Von Arnim, T. Stegmaier, H. Planck, A. Agarwal, Ind. Eng. Chem. Res. 52 (2013) 13164–13170

[31] S. Bansal, V. Von Arnim, T. Stegmaier, H. Planck, J. Hazard. Mater. 190 (2011) 45–50

[32] S. Agarwal, V. Von Arnim, T. Stegmaier, H. Planck, A. Agarwal, Sep. Purif. Technol. 107 (2013) 19–25

[33] P.S. Kulkarni, S.U. Patel, G.G. Chase, Sep. Purif. Technol. 85 (2012) 157–164

[34] E.J. Clayfield, A.G. Dixon, A.W. Foulds, R.J.L. Miller, J. Colloid Interface Sci. 104 (1985) 500–511

[35] D.D. Govedarica, R.M. Šećerov Sokolović, D.S. Sokolo-vić, S. M. Sokolović, Sep. Purif. Technol. 104 (2013) 268– –275

[36] R.M. Šećerov Sokolović, D.D. Govedarica, D.S. Soko-lović, Ind. Eng. Chem. Res. 53 (2014) 2484–2490

[37] R.M. Šećerov Sokolović, S.M. Sokolović, B.D. Đoković, Ind. Eng. Chem. Res. 36 (1997) 4949–4953

[38] R.M. Šećerov Sokolović, T.J. Vulić, S.M. Sokolović, R.P. Marinković Nedučin, Ind. Eng. Chem. Res. 42 (2003) 3098–3102

[39] R.M. Šećerov Sokolović, T.J. Vulić, S.M. Sokolović, Sep. Purif. Technol. 56 (2007) 79–84

[40] R.M. Šećerov Sokolović, T.J. Vulić, S.M. Sokolović, Ind. Eng. Chem. Res. 45 (2006) 3891–3895

[41] A. Krasinski, Filtr. Sep. 51 (2014) 22–28

D.D. Govedarica, R.M. Šećerov Sokolović, D.S. Sokolović, S.M. Sokolović, Ind. Eng. Chem. Res. 51 (2012) 16085–16091.


318

DRAGAN D. GOVEDARICA1

RADMILA M. ŠEĆEROV SOKOLOVIĆ1

OLGA M. GOVEDARICA1

DUNJA S. SOKOLOVIĆ2

SNEŽANA V. SINADINOVIĆ-FIŠER1

1Tehnološki fakultet Novi Sad, Univerzitet u Novom Sadu, Bulevar

cara Lazara 1, 21000 Novi Sad, Srbija 2Fakultet tehničkih nauka, Univerzitet u Novom Sadu, Trg Dositeja Obradovića

6, 21000 Novi Sad, Srbija

NAUČNI RAD

SEPACIJA EMULZIJE ULJE U VODI KOALESCENCIJOM U VLAKNASTOM SLOJU: METODOLOGIJA ODZIVNE POVRŠINE

U ovom radu ispitivane su mogućnosti primene metodologije odzivne površine na sepa-raciju ulja iz emulzije tipa ulje u vodi. Kao filtarski materijal korišćena su vlakna polietilen tereftalata. Cilj je bio da se analizira i proceni izlazna koncentracija dispergovane faze i kritična brzina u širokom opsegu radne brzine, permeabilnosti sloja vlakana i prirode dispergovane uljne faze. Razvijene su regresione jednačine metodologijom odzivne površine pri konstantnoj permeabilnosti sloja koje su pouzdane i imaju koeficijent kore-lacije R viši od 90%. Utvrđeno je da gustina, viskoznost, i neutralizacioni broj ulja kao i permeabilnost sloja značajno utiču na efikasnost separacije dispergovanog ulja. U regionu visoke viskoznosti ulja i visoke permeabilnosti sloja nalazi se maksimalna vred-nost kritične brzine. Modeli razvijeni medologijom odzivne površine mogu se primeniti i na koalescentnu filtraciju da se prognozira efikasnost separacije, kao i da se optimizuje efikasnost separacije pri simultanim promenama permeabilnosti sloja, intenziteta hidro-dinamičkih sila i prirode dispergovane uljne faze.

Ključne reči: separacija tečno-tečno, zauljena voda, koalescencija u vlaknastom sloju, vlakna, kritična brzina, metodologija odzivne površine.

Documents

DRAGAN D. GOVEDARICA SEPARATION OF OIL-IN-WATER …