Cn

Aa

Cb

h

�����

a

ARRAA

KCNFM

1

(pcemCl

0h

Journal of Hazardous Materials 244– 245 (2013) 251– 258

Contents lists available at SciVerse ScienceDirect

Journal of Hazardous Materials

jou rn al h om epage: www.elsev ier .com/ loc ate / jhazmat

omputational fluid dynamics simulation of transport and retention ofanoparticle in saturated sand filters

shraf Aly Hassana, Zhen Lib, Endalkachew Sahle-Demessiea,∗, George A. Sorialb

U.S. Environmental Protection Agency, Office of Research and Development, National Risk Management Research Laboratory, 26 W. Martin Luther King Drive,incinnati, OH 45268, USASchool of Energy, Environmental, Biological, and Medical Engineering, Environmental Engineering Program, University of Cincinnati, Cincinnati, OH, USA

i g h l i g h t s

Breakthrough curves used to study fate of NPs in slow sand filters (SSF).CFD simulate transport, attachment/detachment of NPs in SSFs.CFD predicted spatial and temporal changes for transient concentrations of NPs.CFD predicts low concentrations and steady NP influx would not be retained by SSFs.Pulse input is retained with outlet concentration of 0.2% of the inlet.

r t i c l e i n f o

rticle history:eceived 28 August 2012eceived in revised form 31 October 2012ccepted 7 November 2012vailable online xxx

eywords:omputational fluid dynamics (CFD)anoparticleslow through porous media

a b s t r a c t

Experimental and computational investigation of the transport parameters of nanoparticles (NPs) flow-ing through porous media has been made. This work intends to develop a simulation applicable to thetransport and retention of NPs in saturated porous media for investigating the effect of process conditionsand operating parameters such, as ion strength, and filtration efficiency. Experimental data obtained fromtracer and nano-ceria, CeO2, breakthrough studies were used to characterize dispersion of nanoparticlewith the flow and their interaction with sand packed columns with different heights. Nanoparticle trans-port and concentration dynamics were solved using the Eulerian computational fluid dynamics (CFD)solver ANSYS/FLUENT® based on a scaled down flow model. A numerical study using the Navier–Stokesequation with second order interaction terms was used to simulate the process. Parameters were esti-

odeling transport and deposition mated by fitting tracer, experimental NP transport data, and interaction of NP with the sand media. Themodel considers different concentrations of steady state inflow of NPs and different amounts of spikeconcentrations. Results suggest that steady state flow of dispersant-coated NPs would not be retainedby a sand filter, while spike concentrations could be dampened effectively. Unlike analytical solutions,the CFD allows estimating flow profiles for structures with complex irregular geometry and unevenpacking.

. Introduction

Industrial and consumer market for engineered nanoparticlesNPs) have been growing over the past decade. The commercialroduction and utilization of the NPs is anticipated to dramati-ally increase in the next few decades [1]. Such wide use of nanonabled materials could lead to the release of NPs into the environ-

ent during manufacturing, transport, application, and end-of-life.

ommercial applications of NPs include sunscreens, paints, tires,ubricants, shampoos and drug delivery, medical use, and electronic

∗ Corresponding author. Tel.: +1 513 569 7739.E-mail address: sahle-demessie.endalkachew@epa.gov (E. Sahle-Demessie).

304-3894/$ – see front matter. Published by Elsevier B.V.ttp://dx.doi.org/10.1016/j.jhazmat.2012.11.021

Published by Elsevier B.V.

devices. Technological development and application are out-pacingresearch in health and environmental risk. In the last decade theresearch efforts has focused on the toxicity and human healtheffects of NPs [2,3], as well as their mobility and transport [4,5].Correct prediction of the transport and fate of colloidal parti-cles in saturated porous media is of practical interest for manyenvironmental applications, including transport of colloids andcolloid-associated pollutants in groundwater and natural filtrationof microorganisms in a deep-bed filtration in water and wastewatertreatment.

Once released into the environment, NP could be transportedto the ground water reserves. The mobility of NPs in groundwateraquifers and sand filters is a critical property affecting their fateand transport. The aggregation and mobility of NPs in porous

2 us Mat

mwFbrcu

accsums[ntBf

lrmpiiichtaaowtw[i[

cdoasmrwt

2

2

t

iacr

52 A.A. Hassan et al. / Journal of Hazardo

atrices depend on surface properties and particles size of NPs,ater chemistry such as pH, ionic strength, and ion valency [6,7].

ull and accurate assessment of the environmental risk posedy NPs requires effective quantitative analysis of their mobility,eactivity, transformation, ecotoxicity, and persistence. Laboratoryolumn studies and numerical simulation would help to betternderstand the different factors and scenarios.

Column studies provide quantitative description of transportnd retention of NPs based on the measurement of effluent streamoncentrations. Experimental studies coupled with mathematicalorrelations have been used to investigate the transport and depo-ition of Fullerene (C60) NPs in glass beads, quartz, or Ottawa sandnder varying flow conditions and electrolyte species [7,8]. Theobility of Fullerene and metal oxides NPs are affected by factors,

uch as water velocity, electrolyte species, and ionic concentration5,9]. According to clean-bed filtration theory, the main mecha-ism of particle’s transport includes interception on the particle byhe media, sedimentation caused by gravity, and diffusion due torownian motion. Brownian diffusion is the most significant factor

or deposition of nanoparticles [10].However, laboratory-scale experimental approaches have

imited number of points and time instances, and they have nar-ow range of problems and operating conditions. Mathematicalodels and simulations allow quantitative prediction of NP trans-

ort phenomena for all desired quantities, with high resolutionn space and time for the actual flow domain. Therefore, model-ng is a cheaper and faster alternative to provide us with deepernsight for NP transport through large scale systems. A modelan only be as good as the data that were used to build it andence the generation of quantitative, sufficiently rich time is impor-ant. Since mathematical modeling requires some simplifyingssumptions, the reliability of CFD simulations have to be evalu-ted [11]. Flow field takes into consideration the hydrodynamicsf fluid flow and NPs attachment and detachment interactionsith packing media. CFD has been used for modeling aerosol

ransport in a dilute gas flow [12]. Flow through membranesas modeled using CFD techniques [13] including ultrafiltration

14]. CFD has also been used successfully for design of reactorsncluding the complicated scenario of three phase fluidized beds15].

CFD simulation has long had the reputation of being too diffi-ult, slow, and expensive to be performed as part of the mainstreamesign process. In this work, we have shown that CFD simulationf NP transport in and around engineering systems is a powerfulnd faster technique for designing and operating studies. Two casetudies are presented on how the combination of computationalodeling tools and experimental methods at laboratory scale can

esult in a robust scale-up tool for NPs transport and retentionithin saturated, unconsolidated, porous media of industrial water

reatment systems.

. Theoretical background

.1. Transport of nonreactive and non-absorbing tracer

The advection–dispersion–diffusion model is used to describeransport of a conservative tracer through the mobile domain.

∂C

∂t= D

∂2C

∂x2− vp

∂C

∂x(1)

C is the concentration of NaCl in solution, t is time elapsed, xs the distance parallel to the flow, �p is the pore water velocity,nd D is the hydrodynamic dispersion. The initial and boundaryonditions used to solve Eq. (1) differ for each stage of the testuns.

erials 244– 245 (2013) 251– 258

For the deposition phase, the initial and boundary conditionsare:

C(x, t = 0) = 0, C(x = 0, t) = C0 and∂C

∂x(x = L, t) = 0 (2)

For the rinse phase, the initial and boundary conditions are:

C(x, t = t1) = C1, C(x = 0, t) = 0 and∂C

∂x(x = L, t) = 0 (3)

L is the total column length, t1 is the rinse phase starting time,and C1 is the concentration profile of non-interacting tracer, such asNaCl, in the bulk solution at the end of deposition phase. Equation(1) was numerically solved using the initial and boundary con-ditions of Eqs. (2 and 3). Nonlinear regression analysis was usedto find model parameter, the dispersion value that best fit theadvection–dispersion model to the tracer break through data.

2.2. Mathematical modeling of nanoparticle transport andretention

The transport and retention of particles in a water-saturatedporous medium can be described mathematically by consider-ing the processes of advection and hydrodynamic dispersionin addition to a sink/source term describing particle attach-ment and detachment on sand surface. Assuming conditions ofhomogeneity, laminar flow, and the absence of inter-particleinteractions, the transport and retention of NPs in a porousmedium can be described using the system of 1-dimensionaladvection–dispersion–sorption/desorption equations:

∂C

∂t= D

∂2C

∂x2− vp

∂C

∂x− �b

�w

∂S

∂t(4)

�b

�w

∂S

∂t= kattC

Smax − S

Smax− �b

�wkdetS (5)

where C is the concentration of NPs in solution, S is the NP con-centration associated with the solid collector phase, Smax is themaximum retention capacity of NPs within the bed, t is timeelapsed, x is the distance parallel to the flow, �b is the sand bulkdensity, D is the hydrodynamic dispersion coefficient, �w is the vol-umetric water content, and �p is pore water velocity. The kineticsof NPs attachment was expressed as the clean bed filtration the-ory [7,16] as a function of two coefficients, katt, which is the rateof nanoparticle attachment along the depth and kdet is the detach-ment rates. The boundary and initial conditions used to solve Eqs.(4 and 5) are:

C(x, t = 0) = 0; C(x = 0, t) = C0;∂c

∂x(x = L, t) = 0

and S(x, t = 0) = 0 (6)

The attachment of NP flowing water to surface of solid parti-cles in the porous in media involves two processes: mass transportto the surface and NP-surface interactions. Two surface reactions,as shown in the schematic in Fig. 1, were considered to take intoconsideration the attachment and detachment phenomena:

NP + Skatt�kdet

NP − S + S′ (7)

where katt is the overall forward reaction rate of adsorption of NPand kdet is the detachment rate constant. NP–S, S′ are is solid formedin the surface used to balance the reaction, NP–S is a measure of NPsdeposited on the surface and S is the initial coating of the surface

(required as an initial condition) that would be exchanged by NP–S.The attachment rate coefficient katt depends on microscale flow andBrownian diffusion characteristics, as well as the surface propertiesof NP and sand grain.

A.A. Hassan et al. / Journal of Hazardous Mat

Ft

iE

S

wmp

k

m

A

A

2c

aAticzstdorau

bIbwfsatt

ig. 1. Schematic illustration of NP-sand surface interaction used in CFD simula-ions.

The maximum retention Smax appears in the CFD as site densityn kM/m2. The interaction of NP with packing media is described inq. (9):

ite density = Smax�sand�w

(Surface : Volume) MW(8)

here MW is molecular weight. The attachment and detach-ent kinetic rate are defined by input of an Arrhenius type

re-exponential factor and activation energy:

i = Aie|−E/RT (9)

where the pre-exponential factor, A, for attachment and detach-ent rates, katt and kdet can be shown as:

katt = katt

(Site density) (MW) �w(10)

kdet= kdetMW

�w(11)

.3. Description of geometry and selecting the boundaryonditions

A computational model and the corresponding structural grid ofn experimental packed column were created using the softwareNSYS and Fluent 13. The flow Eqs. (4) and (5) were solved for a

wo-dimensional double precision spatial. The grid consists of annlet, outlet and two symmetry of x and y axes. The inlet boundaryondition was set at velocity inlet with velocity components set toero except the component in the direction of flow. The outlet waset at pressure outlet with a gauge pressure of zero Pascal. To reducehe computational time, dimensions of the actual filter were scaledown by a factor of 10−3 in both x and y directions. The simulationutput was not affected by scaling, but the results needed to beescaled for analysis. The mass diffusivity needed to be reduced by

factor of 103 corresponding to spatial scaling. The solver used annsteady-state condition with an implicit formulation.

A porous media module was used to simulate the sand filtered that incorporates an empirically determined flow resistance.

t assumes the porosity is isotropic for the current calculations,ut the model can also handle media where the porosity variesith space and time. The interactions of nanoparticles to sand sur-

aces were performed on the surface of the porous media using a

urface-to-volume ratio equal to 6/dsand, the sand particles weressumed to be spherical in shape. For a reacting flow, the reac-ion time scales are much faster than convection and diffusionime scales; hence, the solution of the species transport equations

erials 244– 245 (2013) 251– 258 253

becomes numerically difficult. Such systems are termed “stiff” sys-tems. Stiff systems with laminar chemistry can be solved usingeither the pressure-based solver with the Stiff Chemistry Solveroption of Fluent enabled. In the first fractional step, the multiphasespecies transport and reaction equation was solved spatially withthe reaction term set to zero. In the second fractional step, the reac-tion term is integrated in every cell using a stiff ordinary differentialequation solver. Species used in the model were divided into twophases the bulk phase species, which include NPs and water, andthe species on the porous matrix phase.

The solution for the differential equation was obtained usingcoupled pressure velocity scheme. This coupling was performedto obtain an additional condition for pressure by reformatting thecontinuity equation. The algorithm solves the coupled momentumand pressure-based continuity equations together. The full cou-pling was achieved through an implicit discretization of pressuregradient terms in the momentum equations, and an implicit dis-cretization of the face mass flux. The spatial discretization was seton a least square cell based gradient. The momentum, NPs andenergy are solved using second order upwind, where quantities atcell faces were computed using a multidimensional linear recon-struction approach. The pressure was solved with the standardmethod using a second order implicit transient formulation.

3. Experimental

3.1. Experimental setup

Cerium oxide, CeO2, nanoparticles were used for the experi-mental studies. CeO2 is used as a polishing agent for glass mirrors,ophthalmic and precision optics, and in emission control systemsin automobile engines as a diesel fuel-borne catalyst to reduceparticulate matter emissions. The nanoparticles used for labora-tory experiments consisted of long-term stabilized CeO2, havingan average hydraulic diameter of 160 nm. Transport studies wereconducted using glass columns (45 cm length × 2.54 cm diameter)that were packed with sieved, acid washed, and oven calcined sil-ica sand (AGSCO Co., IL). The details of the sand and packing weredescribed in previously published work [6]. Steady-state upwardflow (20 mL/min, empty bed Re ∼ 0.8) of CeO2 NPs suspension(20 mg/L) was fed in to the column at a selected ionic strength forsix pore volumes of flow in the deposition phase. This was followedby a rinse phase during which particle free solution, which has thesame ionic strength as the NPs suspensions, was fed. A total of sixconditions were conducted at varying NaCl concentrations of 1, 2, 3,5, 10, and 100 mM and at a constant pH of 5.6. The Peclet number,which is the ratio of transport by advection to the rate of trans-port by diffusion, has been calculated. Longitudinal diffusion is notimportant since the Peclet number was calculated to be 41 (Pe > 6)suggesting advection is the predominant transport mechanism. AUV–vis spectroscopic cell installed in the effluent line was usedto measure the concentration of NPs in the effluent stream of thepacked column.

3.2. Tracer studies

Tracer studies were conducted to measure and quantify thehydrodynamic dispersion of an inert tracer (NaCl), which givesa macroscopic outcome of the various physical and chemicalphenomena that takes place within the pores. Change in the con-centration of inert tracer results in non-retarded fronts. Variation

in ionic strength, such as infiltration of sodium chloride solution(20 mM), was selected as inert. The solution was introduced to asand column saturated with DI water. The effluent stream concen-tration of NaCl was measured using a conductivity meter Oakton®

2 us Mat

Cpsisr

4

4

oslsibDRwTbwat0s

4

dstkbicsd

4

tScr

k

wb

�

w

vi

k

wpt

face charges forming a quasi stable suspension. Hence, it is assumedthe retention of NPs on the sand surface is a mono layer deposition.The increase in ionic strength decreased the magnitude of the zetapotential due to the compression of the electrical double layers and

ka

tt &

kd

et, h

-1 an

d S

ma

x, m

g g

-1

0.01

0.1

1

10

100Z

eta

po

ten

tia

l, m

v

-22

-20

-18

-16

-14

-12

-10

-8

-6

-4

katt

kdet

Smax

Zeta potential

54 A.A. Hassan et al. / Journal of Hazardo

ON 11 (Cole Parmer, IL, USA). Test studies were carried out usingacked columns of 300, 450, 600, and 1200 mm in length, and atix flow rates of 10, 20, 30, 50, 65, and 80 mL/min. An on-line mon-tor was used to measure the conductivity of the column effluenttream. The value of D increased with increase in flow rates, andemained constant with differing column heights.

. Results and discussion

.1. Results of tracer study

The coefficients of hydrodynamic dispersion, D, for Eq. (1) werebtained from the tracer study at different flow rates. The disper-ion coefficient has a linear relationship with the flow rate on aog–log scale with R2 of 0.96. As the flow rate increased at selectedtep from 10 to 65 mL/min the values of the dispersion coefficientncreased from 1.6 × 10−6 to 2.8 × 10−5 m2/S and the correlationetween the two can be represented in the dispersivity model

= 0.057V1.5, where V is the flow velocity, with correlation constant2 of 0.93. As the flow rate increased the Peclet number decreased,hich confirms that advection is more significant than dispersion.

he longitudinal coefficient was used to predict the shape of thereakthrough curve in the rinse phase. Moreover, the tracer dataas validated using different columns with the same porosity but

t different lengths. The D values calculated earlier fit the break-hrough curves at flow rate of 20 mL/min and total length of 0.3,.6 and 1.2 m. Breakthrough curves and model predictions are pre-ented in the Appendix A.

.2. Results of mathematical model

Utilization of longitudinal advection–dispersion–sorption/esorption kinetics to predict the breakthrough of CeO2 NPs inand columns was modeled successfully elsewhere [6]. Equa-ions (4) and (5) were solved simultaneously with attachmentinetics of CeO2 NPs and the initial and boundary conditions. Thereakthrough of the six different operating conditions described

n Appendix A are solved using the value of D of 3.2 × 10−6 m2/Sorresponding to 20 mL/min. NP Solid that was formed in theurface, was used to balance the reaction and measures the NPseposited on the surface.

.2.1. Parameter estimating strategiesSome estimation strategies were considered for the parame-

er matrix in multivariate regression models to solve katt, kdet, andmax of the transport and retention equations. The attachment rateoefficient can be related to the commonly used single-collectoremoval efficiency � using [10]:

att = 3 (1 − ε) v2dc

� (12)

here the experimental single collector removal efficiency, �, cane determined from NP break through curve:

= −23

dc

(1 − ε) Lln

(C

Co

)(13)

here ε is the bed porosity.Combining Eqs. (12) and (13) the first step was estimating the

alues of katt based on clean bed filtration theory using the follow-ng expression:

att = Uln

(C

)(14)

Lε Co

here U is Darcy velocity of flow, L is column length, and ε is the bedorosity. The value of katt is used to fit the value of Smax to the break-hrough curve in the deposition phase. After Smax was estimated,

erials 244– 245 (2013) 251– 258

both katt and Smax were used to predict kdet over the whole break-through range. Following this approach, the regression parameterswere correlated with log of ionic strength (IS) of the flowing wateras follows: katt = 15.1 log(IS), kdet = 0.3 log(IS), and Smax = 0.2 log(IS)with R2 of 0.84, 0.77 and 0.79, respectively. The increase of katt andkdet with increased ionic strength is logical while the same is nottrue for Smax, which is expected to be constant.

The second strategy is to use Smax, and katt as regression param-eters to find the initial strength of the fitting of first part of thenanoparticles break through curves and estimate baseline R2. Thesevalues were used to estimate the value of kdet over the whole periodof break through and rinsing phases of the flow. The calculatedmodel parameters showed similar correlation with the change inthe ionic strength of the water: katt = 14 log (IS), kdet = 0.32 log(IS)and with R2 value of 0.86 and 0.80, respectively. The value Smax waspractically constant when complete or no breakthrough occurredat 1.1 ± 0.2 mg/g; however, when partial breakthrough occurredat conditions 3 and 4, Smax decreased significantly. The bind-ing coefficients of the sand are more sensitive to changes in pHand are expected to have constant retention capacity Smax. How-ever, a slight increase in the capacity for NP with increasing ofionic strength has been repeatedly [8]. Attempts were made torun regression analysis for katt while keeping Smax constant at1.1 mg/g for partial breakthrough conditions 3 and 4, but simula-tion results were not satisfactory. This might suggest the parameterSmax selected for the equation did not really simulate the true max-imum retention capacity. Therefore, the values of the parametersestimated using approach “A” are used later in the CFD model(Appendix A).

The estimation strategy where all the three parameters werefitted simultaneously under the general constraints for the entireexperimental period did not generate acceptable convergingresults. This may be due to the variation in the weight of theseparameters.

4.2.2. Implication of model parameters and DLVO theoryThe estimated values of model parameters katt, kdet, Smax, and

zeta potential on nanoparticles are plotted against the water ionicstrength for pH of 5.6 (Fig. 2). At these conditions (pH 5.6 and ionconcentration of less than 100 mM). CeO2 NPs have negative sur-

Ionic strength, mM

1 10 100

Fig. 2. The effects of ionic strength on the maximum retention capacity, attachmentand detachment coefficients, and zeta potential of NP.

s Mat

imi(w(pisataoa5msswis

A.A. Hassan et al. / Journal of Hazardou

ncreased particle aggregation. The zeta potential and the attach-ent rate constant katt increased almost linearly with the log of

onic strength until 10 mM, after which the values show no changeFig. 2). At low ionic strength a substantial repulsive energy barrieras observed according to Derjaguin–Landau–Verwey–Obserbeck

DLVO) theory indicating retention was kinetically controlled. Therimary energy barrier resulted in a low attachment rate, observed

n the value of katt, and full breakthrough conditions. As ionictrength increased, the value of the primary barrier decreasedllowing more attachment and increasing katt. At NaCl concentra-ions above10 mM, the energy barrier was minimal and particlettachment was more diffusion controlled, which is independentf ionic strength. The drop in the value of Smax was observed at

partial breakthrough for NaCl concentrations above of 3 and mM. This could be explained by the existence of a secondaryinimum, which facilitates the attachment of NPs to the sand

urface, but it did not offer the same retention capacity on theand. This explains the decrease in kdet at the same ionic strength

here attraction forces of the interaction potential were dominant

n the secondary minimum and where the particles are kineticallytable.

Fig. 3. Outlet breakthrough and cross-section of CeO2 concentration i

erials 244– 245 (2013) 251– 258 255

5. Case studies

The mathematical model developed and the parametersobtained above were used to simulate flows in real applications.In order to validate the method used in this study, a CFD modelwas used to simulate the experimental data obtained earlier usingthese parameters. The results agreed with the experimental dataand with the numerical solution described in earlier work [6].

Sand filters operating at water treatment facilities (WTF) are notdesigned to retain particles at nanosize range; the transport prop-erties of nanoparticles through large scale sand filters are not fullyunderstood. The facility in this case has 47 divided square shapedfilters with a depth of 0.61 m and each filter has surface area of133.8 m2. The flow rate is almost 20 mL/min/m2 (3 gal/min ft2) withan empty bed contact time (EBCT) of 6 min. Backwashing variesseasonally over the range of 30–50 h. These dimensions and oper-ating conditions were used to simulate the operation of the slowsand filters at the Greater Cincinnati Water Works facility, Cincin-

nati, OH. The objective is to investigate the model prediction forthe transport and deposition of manufactured NPs through actualslow gravitational sand filters and to evaluate the effectiveness of

n the bulk at different breakthrough times for a WTF sand filter.

2 us Materials 244– 245 (2013) 251– 258

ta

vwrtr

5N

wosr[tcaatcb

cstw

5

sTNtrwahctabstbCt

wEm0vctEtotantn

Time, h

0.0 5.6 11.1 16.7 22.2 27.8 33.3 38.9

Co

ncen

trati

on

, p

pm

0

1

2

3

4

5

6

7

8

9

10Co = 10 ppm

Co = 9 ppm

Co = 8 ppm

Co = 7 ppm

Co = 6 ppm

Co = 5 ppm

Co = 4 ppm

Co = 3 ppm

Co = 2 ppm

Co = 1 ppm

Co = 0 ppm

Fig. 4. Concentration of NP in the outlet effluent at different inlet concentration.

Time, h

0.00 0.14 0.28 0.42 0.56

Co

nc

en

tra

tio

n, p

pm

0.00

0.02

0.04

0.06

0.08

0.10

0.12

Co = 10 ppm

Co = 20 ppm

Co = 30 ppm

Co = 40 ppm

Co = 50 ppm

Fig. 5. Outlet response for different pulse inlet concentrations of NP in continuousfeed flow.

56 A.A. Hassan et al. / Journal of Hazardo

he filter for a steady state feed of NP and for cases where there was single spike.

For both models the initial conditions are kept similar. The inletelocity was set at 3.3 × 10−4 m/s. The NP mol fraction at the inletas adjusted to 1.2 × 10−6, where the rest was water. The reaction

ates were adjusted according to Fig. SM 2 approach “A” condi-ion 1, where are 0.7, 0.08 h−1 and 0.1 mg/g for katt, kdet, and Smax,espectively.

.1. Simulating filter performance for steady state feed of CeO2Ps to the WTF

The computational domains were generated for filter mediaith different depth using mesh size of 748 cells with a minimum

rthogonal quality of 0.99. Cell sizes were selected to optimize theimulation time, the number of iterations needed, and the accu-acy of the solution. The meshes were generated using Gambit17]. The flow fields are fairly refined to facilitate accurate particleracking and discritized control volume aspect ratio in the generalomputational domain. Structural meshes were used to take thedvantage of flow directionality for increasing the convergence ratend minimize error in computational approximation [14]. In ordero assess the grid independence, of simulation on the mesh cellount was doubled and tripled with no significant changes in thereakthrough results.

The model assumed a dynamic pressure of 5.4 × 10−5 Pa and aonstant Reynolds number of 0.18, based on empty bed velocity,uggesting laminar flow. The molecular Prandtl Number is 1.58 andhe Courant Number is 0.39. The mass imbalance within the modelas negligible in the range of 10−20.

.1.1. Steady state flowThe performance of the sand filter for handling a steady con-

tant concentration of CeO2 NP in the feed stream was assessed.he simulation was run for a continuous feed of water with CeO2Ps concentration at 10 ppm for a period of 2760 s for the deposi-

ion phase followed by flow with water and no NP simulating theinse phase. Fig. 3 shows the breakthrough of the WTF sand filterith the profile of CeO2 NPs concentration in the column bulk phase

t selected time steps. In part a, the breakthrough of the particlesas not occurred yet, but the upward propagation of the CeO2 NPsoncentration front through the bed can be seen. The concentra-ion front of CeO2 NPs moved upwards resulting in a breakthrought the outlet, part b. After this, the outlet CeO2 NPs concentrationecame stable and the concentration within the bulk became con-tant all over the bed (Part c). After stopping the CeO2 NPs in feed,he ceria free water front was formed and propagated through theed (Part d,). This front kept moving upwards causing the effluenteO2 concentration to drop (Part e). The concentration continuedo drop and the CeO2 NPs front shrank in parts f and g.

The simulation can evaluate different scenarios including waterith high ionic strength, which may not be plausible at a WTF.

quations (4–6) were solved simultaneously with the estimatedodel parameters for different NPs concentration ranging from

to10 ppm. The CeO2 NP showed no breakthrough at one poreolume of flow with no retention (Fig. 4). The column retentionapacity at the flow condition was negligible with small concen-rations of NPs. The breakthrough occurred at 6 min, which is theBCT. It took on average 24 h to reach 95% of the initial concen-ration at the exit. Hence, a steady inflow of NP could reduce theverall capacity and life time of the filter. The CeO2 NP’s concen-ration on the sand bed (CeO2–S) was negligible due to the low

ttachment coefficient under these conditions. CeO2–S values didot change regardless of the loading conditions. This fact suggestshat if there are CeO2 NPs particles in the feed stream they wouldot be removed by the conventional sand filter. Fig. 6. Mesh and boundary conditions of a generic spherical shared sand filter bed.

s Mat

5

ofreto

A.A. Hassan et al. / Journal of Hazardou

.1.2. Pulse input of NP into continuous feed flow of a sand filterThe filter response to pulse input of the NPs to a continu-

us flow of water with initial concentrations of 10–50 ppm in theeed stream was simulated. The boundary condition in Eq. (6) was

eplaced by a Dirac-delta function to ensure the immediate damp-ning and to avoid singularity. Fig. 5 shows the sand filter was ableo redistribute the sharp influent peak. The maximum effluent peakbserved was 0.21 ± 0.01% of the influent. This scenario is more

Fig. 7. Outlet breakthrough and cross-section of NP concentration in the bulk

erials 244– 245 (2013) 251– 258 257

likely to occur if a WTF receives an impulse of NP contaminatedwater, such as after accidental spills.

5.1.3. Simulating CeO2 NPs transport through spherical shaped

sand filter

Nanoparticles in sunscreens, cosmetics, and personal-care prod-ucts could end up accumulating in closed-loop systems such as out-door swimming pools. It would be useful to test the effectiveness

at different breakthrough times in a generic spherical shaped sand bed.

2 us Mat

ofpNwswtfidqbaeors

wTwii

tdtw1itcofhwponstAwpu

6

matcartdesribi

w

[

[

[

[

[

[

58 A.A. Hassan et al. / Journal of Hazardo

f the sand filter in removing washed NP from pool water. The per-ormance of a typical forced flow through above ground swimmingool sand filter was simulated for transport and retention of CeO2Ps. The spherical shaped filter has a diameter of 1 m and is filledith sand up to two third of its depth. Influent water is spread on the

and surface through nozzles. Continuous steady inflow of wateras assumed over the surface. The outlet is a round disk located at

he bottom of the filter with clearance of some distance that is stilllled with sand. The tank was modeled in a 2-D axi-symmetric con-ition; with mesh size of 2132 cells having a minimum orthogonaluality of value of 0.70. The mesh structure with the appropriateoundary conditions is shown in Fig. 6. The materials, interactions,nd hydrodynamic conditions were similar to the pervious casearlier in Section 5.1. CeO2 NPs discharge was applied for a periodf 350 s for the deposition phase then removed for the remainingun period simulating the rinse phase. Grid independent study hashown no effect when the grid cell count was doubled and tripled.

The model of the pool has a dynamic pressure of 3.6 × 10−4 Pa,hich is one order of magnitude higher than the rectangular model.

he Reynolds number varied as well from almost zero to of 0.72,hich is still in the laminar range. The molecular Prandtl Number

s 1.58 and the Courant Number ranged from zero to 0.67. The massmbalance within the model was negligible in the range of 10−20.

The breakthrough profile observed at the outlet of the pool fil-er was different than the WTF sand filter simulation due to theifference in Geometry. The filter geometry resulted in a differentransport pattern of the NPs through the round section next to theall, resulting in different concentrations profiles than the earlier

-dimension model. The breakthrough curve, as shown in Fig. 7,s very steep, which gives an advantage of extra time for the par-icles to interact with the filter sand. The NPs wave front, whichould be seen in Part a, is moving forward as the breakthrough isccurring at the outlet in Part b. When the filter is washed with NPsree water a similar wave propagated from the inlet as seen in Part. The problem with this type of tank appears to have dead zonesith little mixing. Backwashing of the tank is needed to removearticles form dead zone regions. Similar NPs movement could bebserved in Fig. SM 3 but the attached CeO2–S is insignificant inumber as described in the earlier model. Further dissolved con-tituent (tracer) analysis and compartmental modeling is neededo delineate the hydrodynamics of flow in the dead zones of tank.lthough, correlation of the model with the water ionic strengthas used for this simulation, the water chemistry of a swimmingool may be significantly different than those of the test conditionsed for the preliminary experiments.

. Conclusions

CFD was used to simulate transport, attachment, and detach-ent of NPs in laboratory sand columns. The experimental tracer

nd nanoparticle studies were used to characterize porous struc-ures and to develop models that would predict breakthroughurves of NPs through sand filter in a WTF and in a commonly usedbove ground forced flow sand filter used for swimming pools. Theesults obtained by the tracer runs generated consistent values ofhe hydrodynamic dispersion coefficient, which could be used atifferent column lengths. The CFD model successfully predicted thexperimental columns. Model predictions at lower ionic strengthuggested that constant levels of NPs in the influent of a filter feedesulted in a breakthrough of the sand filtration regardless of thenfluent concentration. On the other hand, a pulse input of NP could

e retained effectively causing an outlet concentration of 0.2% of the

nlet concentration.CFD simulations were able to shed light on the mobility of NPs

ithin the saturated porous bed. The model predicted spatial and

[

[

erials 244– 245 (2013) 251– 258

temporal changes for transient concentrations of NPs in bulk andretained on sand. This information is important especially if thesand filter requires improved design for the removal of engineeredNPs. CFD gives the power to calculate bulk and surface concen-trations even in complex shapes that cannot be solved otherwise.Simulations of forced flow swimming pool sand filter has shownthat NPs could accumulate in the sand bed at dead zone areas, whichwould require improvement in the design of filtration system. CFDhas been shown to be an easier, faster, and less expensive way topredict the transport and retention of engineered NPs inside fullscale engineering systems.

Acknowledgments

This research was partially supported by an appointment ofALH to the U.S. EPA Postdoctoral Program at the ORD/NationalRisk Management Research Laboratory, administered by Oak RidgeAssociated Universities through a contract with U.S. EPA. Finan-cial support provided from U.S. Environmental Protection Agencyunder contract No. PR-C108-1170. We are grateful for the contin-uous helpful assistance from Ansys Inc. technical support; HossamMetwallying and John Ibrahim.

Appendix A. Supplementary data

Supplementary data associated with this article can be found,in the online version, at http://dx.doi.org/10.1016/j.jhazmat.2012.11.021.

References

[1] A.D. Maynard, R.J. Aitken, T. Butz, V. Colvin, K. Donaldson, G. Oberdorster, M.A.Philbert, J. Ryan, A. Seaton, V. Stone, S.S. Tinkle, L. Tran, N.J. Walker, D.B. Warheit,Safe handling of nanotechnology, Nature 444 (2006) 267–269.

[2] J. Fabrega, S.R. Fawcett, J.C. Renshaw, J.R. Lead, Silver nanoparticle impact onbacterial growth: effect of pH, concentration, and organic matter, Environ. Sci.Technol. 43 (2009) 7285–7290.

[3] B. Gaiser, T. Fernandes, M. Jepson, J. Lead, C. Tyler, V. Stone, Assessing exposure,uptake and toxicity of silver and cerium dioxide nanoparticles from contami-nated environments, Environ. Health 8 (2009) S2.

[4] S.A. Bradford, S.R. Yates, M. Bettahar, J. Simunek, Physical factors affecting thetransport and fate of colloids in saturated porous media, Water Resour. Res. 38(2002) 1327.

[5] H.F. Lecoanet, J.-Y. Bottero, M.R. Wiesner, Laboratory assessment of the mobilityof nanomaterials in porous media, Environ. Sci. Technol. 38 (2004) 5164–5169.

[6] Z. Li, E. Sahle-Demessie, A. Aly Hassan, G.A. Sorial, Transport and depositionof CeO2 nanoparticles in water-saturated porous media, Water Res. 45 (2012)4409–4418.

[7] Y. Wang, Y. Li, J.D. Fortner, J.B. Hughes, L.M. Abriola, K.D. Pennell, Transportand retention of nanoscale C60 aggregates in water-saturated porous media,Environ. Sci. Technol. 42 (2008) 3588–3594.

[8] Y. Li, Y. Wang, K.D. Pennell, L.M. Abriola, Investigation of the transport anddeposition of fullerene (C60) nanoparticles in quartz sands under varying flowconditions, Environ. Sci. Technol. 42 (2008) 7174–7180.

[9] H.F. Lecoanet, M.R. Wiesner, Velocity effects on fullerene and oxide nanoparti-cle deposition in porous media, Environ. Sci. Technol. 38 (2004) 4377–4382.

10] K.-M. Yao, M.T. Habibian, C.R. O’Melia, Water and waste water filtration. Con-cepts and applications, Environ. Sci. Technol. 5 (1971) 1105–1112.

11] J.H. Ferziger, M. Peric, Computational Method for Fluid Dynamics, 3rd edition,Springer, New York, 2001.

12] P. Nerisson, O. Simonin, L. Ricciardi, A. Douce, J. Fazileabasse, Improved CFDtransport and boundary conditions models for low-inertia particles, Comput.Fluids 40 (2011) 79–91.

13] R. Ghidossi, D. Veyret, P. Moulin, Computational fluid dynamics applied tomembranes: state of the art and opportunities, Chem. Eng. Process. ProcessIntens. 45 (2006) 437–454.

14] M.A. Monfared, N. Kasiri, A. Salahi, T. Mohammadi, Modeling ultrafiltration ofgelatin–water suspension by computational fluid dynamics, Chem. Eng. Res.Des. 90 (2012) 1098–1104.

15] K. Sivaguru, K.M.M.S. Begum, N. Anantharaman, Hydrodynamic studies onthree-phase fluidized bed using CFD analysis, Chem. Eng. J. 155 (2009)

207–214.

16] J.E. Saiers, G.M. Hornberger, L. Liang, First- and second-order kineticsapproaches for modeling the transport of colloidal particles in porous media,Water Resour. Res. 30 (1994) 2499–2506.

17] FLUENT-Inc, GAMBIT: user’s guide, Fluent Incorporated, 1999.