oxygen transfer prediction

Chemical Engineering Science 62 (2007) 71637171www.elsevier.com/locate/ces

Oxygen transfer prediction in aeration tanks using CFDYannick Fayollea, Arnaud Cockxb, Sylvie Gillota,, Michel Roustanb, Alain Hduita

aCemagref, Unit HBAN, Parc de Tourvoie, BP 44, 92163 Antony Cedex, FrancebLISPB, Complexe scientique de Rangueil, 31077 Toulouse Cedex 4, France

Received 20 April 2007; received in revised form 13 July 2007; accepted 12 August 2007Available online 4 September 2007

Abstract

In order to optimize aeration in the activated sludge processes, an experimentally validated numerical tool, based on computational uiddynamics and able to predict ow and oxygen transfer characteristics in aeration tanks equipped with ne bubble diffusers and axial slowspeed mixers, is proposed. For four different aeration tanks (1;1493;8191 and 29, 313 m3), this tool allows to precisely reproduce experimentalresults in terms of axial liquid velocities, local gas hold-ups. Predicted oxygen transfer coefcients are within 5% of experimental resultsfor different operating conditions (varying pumping ow rates of the mixers and air ow rates). The actual bubble size must be known withprecision in order to have a reliable estimation of the oxygen transfer coefcients. 2007 Elsevier Ltd. All rights reserved.

Keywords: Aeration; Multiphase ow; Mass transfer; Hydrodynamics; Bubble; Wastewater treatment

1. Introduction

In urban wastewater treatment, the main process to treat ni-trogen and organic components of wastewater is the activatedsludge one. Aeration in this process can represent up to 70%of the total energy expenditure of the plant. Optimizing aera-tion is therefore required to reduce operating costs, in additionto guaranty a reliable and efcient treatment in such installa-tions. Since the last 20 years, ne bubble aeration systems havebeen extensively implemented to perform aeration. They provedto give high oxygenation performances, and to be adaptive tooxygen requirements. For loop reactors commonly installed inEurope, ne bubble aeration is separated from mixing, car-ried out by axial slow speed mixers. Numerous on site resultspointed out the main parameters impacting oxygen transfer(Groves et al., 1992; Wagner and Ppel, 1998; Gillot et al.,2000; Mueller et al., 2002). However, oxygen transfer in deepaeration tanks remains difcult to predict (Gillot et al., 2005).This may be due to the parameters that are not taken intoaccount in the analysis of the experimental results (bubblesize, gas hold-up, local axial velocity). More precise informa-tion on the local hydrodynamics of aeration tanks is therefore

Corresponding author. Tel.: +33 1 40 96 60 66; fax: +33 1 40 96 61 99.E-mail address: [email protected] (S. Gillot).

0009-2509/$ - see front matter 2007 Elsevier Ltd. All rights reserved.doi:10.1016/j.ces.2007.08.082

necessary in order to better understand the impact of designand operating parameters on oxygen transfer.

In addition, computational uid dynamics (CFD) is more andmore used to optimize aeration systems (Simon, 2000; Cockxet al., 2001; Vermande et al., 2003, 2005). Previously intro-duced by Chatellier (1991) and Simon (2000), the modellingof monophasic ows in aeration tanks induced by axial slowspeed mixers was studied by Vermande et al. (2003). These au-thors showed that the modelling of axial slow speed mixers byimposing the characteristics of the ow induced by agitation,on a zone of the grid corresponding to the location of the mixer(xed values method), gives a good prediction of the meanexperimental axial liquid velocity.

In Cockx et al. (2001), ASTRID code was used to analysebubble interfacial area on a full scale aeration tank (15,000 m3).The structure of the ow was shown to be tri-dimensional andcomposed of large recirculation movements of liquid result-ing from mutual interaction between bubble plumes. However,this numerical study was not confronted to experimental re-sults which would have allowed validating the CFD results.Moreover, the size of the bubbles was empirically imposed andconsidered as constant in the entire tank. The impact of thishypothesis has therefore to be investigated.

Finally, Vermande et al. (2005) recently investigated thebehaviour of two-phase ows in a full scale aeration tank

7164 Y. Fayolle et al. / Chemical Engineering Science 62 (2007) 71637171

Table 1Dimensions and equipment description of the studied aeration tanks (diffuser grid placement and mixers are represented on the schematic representation)

Tank Type Geometry anddiffuser grid placement

Volume (m3) Dimensions Mixers Air supply

Number ofdiffuser grids

Number of diffuseron each grid

1 Oblong 1 L = 3 m; l = 1 m;dw = 0.4 m

One axialmixer (D = 0.15 m)

1

2 Annular 1493 Dout = 20.25 m;Din=7.83 m; dw=5.45 m

One large bladeslow speed mixer(D = 2.5 m)

6 26

3 Oblong 8191 L = 50.65 m; l =27.5 m; dw=5.3 m

Six large bladesslow speed mixer(four of D = 2.5 mand two of D = 2.2 m)

18 and 10 88 and 44

4 Oblong 29,313 L = 90.63 m;l = 34.08 m;dw = 8.6 m

Eight large bladesslow speed mixers(D = 2.4 m)

24 87

Table 2Bubble Sauter diameter at P = 1.013 105 Pa introduced in each simulation (where Qdiff is the mean air ow rate per diffuser and dbs is the bubble Sauterdiameter)

Tank 1 2 3 4

Qdiff (Nm3 h1) 2.58 3.25 9.97 4.75 6.06UC (m s1) 0.10 0.30 0.10 0.30 0.00 0.42 0.15 0.23dbS (mm) 1.93 2.70 1.99 2.62 4.94 4.77 4.27 4.60

Table 3Number of cells of the grid for each one of studied aeration tanks

Tank 1 2 3 4

Volume (m3) 1 1493 8191 29,313Number of cells 29,147 13,266 128,704 452,440

(6200 m3). Unfortunately, no experimental validation of thecomputed results and no complete description of the modelsand of the closure relations implemented to simulate two-phaseow have been presented by the authors.

The objective of this paper is therefore to present an experi-mentally validated numerical tool able to predict ow and oxy-gen transfer characteristics in aeration tanks equipped with nebubble diffusers and axial slow speed mixers. Models used tothis aim are described and analysed. The impact of the bubblesize used to perform the simulations is also investigated.

2. Materials and methods

2.1. Experimental set-up

Four aeration tanks from pilot scale to full scale have beenanalysed. These tanks (129,313 m3) are closed loop reactorsin which aeration is separated from mixing, carried out by axialslow speed mixers. Air is injected at the bottom of the tankthrough EPDM (EthylenePropylene Diene Rubber) membranediffusers (ne bubbles) arranged in several separated grids. Themain characteristics of the basins are summarized in Table 1.

In the pilot plant (tank 1), experimental results (axial liquidvelocity, bubble size, local gas hold-up and oxygen transfercoefcient) were obtained by Simon (2000).

Axial liquid velocity, bubble size and oxygen transfer coef-cient measurements have been performed on each full scaleaeration tanks in this work. The in situ measurement methodshave been described in detail in a previous paper (Fayolle et al.,2006).

Y. Fayolle et al. / Chemical Engineering Science 62 (2007) 71637171 7165

Fig. 1. Streamlines coloured by velocity magnitude (m s1) (a) in tank 1without aeration (QT = 0 Nm3 h1; UC = 0.35 m s1) (b) in tank 4 withoutaeration (QT =0 Nm3 h1; UC =0.27 m s1) and (c) in tank 4 with aeration(QT = 12, 653 Nm3 h1; UC = 0.23 m s1).

2.2. Mathematical modelling

Flow and oxygen transfer computations were determinedwith the 6.2 version of the CFD code FLUENT. This code

uses the nite volume method for the discrimination of theNavierStokes equations. When turbulent ows are considered,such as in aeration tanks, the modelling of the bubbly ow isbased on the Eulerian two-uid model derived from Reynoldsaverage NavierStokes equations. This model is generally usedto simulate bubbly ow reactors with a large number of bubbles(Cockx et al., 1999; Vermande et al., 2005; Talvy et al., 2007).For example, in an aeration tank of 1 m3, a global gas hold-upof 1% represents around 700,000 bubbles. For industrial aera-tion tanks, it seems difcult to calculate the movement of eachinclusion by an EulerianLagrangian model. The EulerEulermodel is then selected in order to describe the average move-ment of inclusions. The development of this model is now clas-sical (Ishii, 1975). Only principal models used to simulate owand oxygen transfer characteristics are briey described in thefollowing.

2.2.1. Two-phase hydrodynamicsThe two-phase ow mass conservation equations for each

phase are considered as follows:

kkt

+ .(kk vk) = 0. (1)

The equations of conservation of momentum are

kk vkt

+ .(kk vk vk)= (kPk) + (kk) + Mk + kk g, (2)

where Pk is the pressure of phase k and Mk = kk

F rep-resents the interfacial transfer of momentum due to pressureand viscous stress distribution and

F represents the sum of

the forces exerted on an inclusion. In the case of a gas/liquidreactor, the interfacial transfer of momentum is principally in-uenced by the drag force FD , the weight and the buoyantforce. The interfacial transfer of momentum related to drag isexpressed as follows:

FD = 12kApCD|Vr |Vr , (3)where Vr represents the relative velocity, CD is the drag coef-cient and Ap is the projected interfacial area. The drag co-efcient used in the present work was proposed by Clift et al.(1978) and is expressed asCD = 23

Eo, (4)

where Eo = (L G)gd2b/ is the Etvos number.This correlation is adapted to the modelling of drag coef-

cients for gas bubbles. The rise velocity is well representedwith this correlation for the range of studied bubble size (from3.0 to 5.5 mm). This correlation takes into account the ellip-soidal shape of bubbles (for a bubble size of 4.0 mm, the dragcoefcient is of 0.98 by the correlation of Clift et al. (1978)and of 0.44 by the correlation of Schiller and Naumann (1938),developed for rigid sphere).

The k dispersed model is used in order to model theturbulence. This model considers that bubblebubble and


0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0 0.25 0.5Distance from internal wall (m)

Tank 1-z=0.35 m (Without aeration) (Without aeration) (With aeration)

Tank 4-z=2.06 m Tank 4-z=2.06 m

Hor

izon

tal l

iqui

d ve

loci

ty (m

s-1 )

0

0.1

0.2

0.3

0.4

0.5

0 3.4 6.8 10.2 13.6 17Distance frominternal wall (m)

Hor

izon

tal l

iqui

d ve

loci

ty (m

s-1 )

-0.5

-0.4

-0.3

-0.2

-0.1

00 3.4 6.8 10.2 13.6 17

Distance from internal wall (m)

Hor

izon

tal l

iqui

d ve

loci

ty (m

s-1 )

Fig. 2. Axial liquid velocity (m s1) versus distance from internal wall (m) (, experimental; , CFD). z represents the distance from the bottom of the tank (m).

bubblewall collisions are negligible and the dominant processin the random motion of the secondary phases is the continu-ous phase turbulence. So, a k model and the Tchen-theorycorrelations (Hinze, 1975) are respectively applied to the con-tinuous and dispersed phase. The term vd , the drift velocity(Bel Fdhila and Simonin, 1992), is introduced in the expres-sion of the relative velocity Vr in order to express the interfaceturbulent momentum transfer. So, the relative velocity Vr ismodelled as follows:Vr = vG vL vd . (5)The dispersion of the gas phase transported by the turbulentuid motion must be taken into account in FLUENT by ac-tivating the drift velocity model. Drift velocity is essential towell simulate the gas hold-up (Talvy et al., 2007).

2.2.2. Two-phase mass transferTo reproduce the experimental set-up of oxygen transfer co-

efcient measurements, two scalars (quantity that is affected bythe ow without itself affecting the ow) must be introducedto represent oxygen concentration in each phase. The generaltransport equations of the concentration of these scalars in two-phase ows are expressed askckt

+ .(kck vk) = .(k( Jk + ckvk)) + Lk , (6)

where Lk represents the interfacial transfer of concentration, Jkis the ux due to the molecular diffusion and ckvk representsthe turbulent diffusion of the concentration.

The oxygen mass transfer between the gas bubble and theliquid is modelled as follows:

LL = kLa(CL CL), (7)where a is the interfacial area, kL is the local mass transfercoefcient and (CL CL) is the driven force of the oxygentransfer. The saturation concentration CL in water depends onlocal pressure according to Henrys law:

CL =yePGHe = mCG, (8)

where ye is the volume fraction of oxygen in air, PG is thepressure in the bubble, m is a coefcient linked to Henryscoefcient (He) and CG is the concentration of oxygen in thebubble. The interfacial area is dened as the ratio of the total airbubble surface and the volume of liquid. For spherical bubbleshaving a bubble Sauter diameter dbS , the interfacial area isexpressed as

a = 6dbS

G1 G . (9)

Finally, the local mass transfer coefcient is modelled usingthe classical penetration theory (Higbie, 1935):

kL = 2

DLVr

db, (10)

where DL is the diffusion coefcient of the oxygen in water(at 20 C).

2.2.3. Boundary conditionsThe boundary conditions relate to the injection of the gas,

the bubble size, the mixing modelling and the modelling of thefree surface.

Surfaces of gas injection correspond to the total area coveredby the modules of bubble diffusers. These surfaces do not takeinto account each one of the diffusers individually but the entiregrid. The injected air ow rate corresponds to the air owrate at the pressure of the diffuser submergence. The oxygenconcentration in the injected bubbles is expressed as

CG = ye.MO2 .PGRT

, (11)

where ye is the volume fraction of oxygen in air, MO2 is themolecular mass of the oxygen, R is the gas constant, T is thetemperature and PG is the pressure at the diffuser level.

The initial bubble size (just above the diffusers) is determinedfrom experimental results. Bubble sizes (at P =1.013105 Pa)introduced in each simulation are summarized in Table 2.

In order to reproduce the experimental data, the bubblesize is expressed as a function of the hydrostatic pressure


0

0.1

0.2

0.3

0.4

0.5

0.6

0

0.1

0.2

0.3

0.4

0.5

0.6

V =

1 m

3

QT =

0 Nm3

h-1

V =

149

3 m

3

QT =

0 Nm3

h-1

1 M

ixer

/ 0

Grid

V =

819

1 m

3

QT =

0 Nm3

h-1

6 M

ixer

s / 0

Grid

V =

293

13 m

3

QT =

0 Nm3

h-1

8 M

ixer

s / 0

Grid

V =

286

46 m

3

QT =

0 Nm3

h-1

8 M

ixer

s / 0

Grid

Uc

(m s-

1 )

Experimental CFD

V =

149

3 m

3

QT =

1555

Nm3

h-1

1 M

ixer

/ 6

Grid

s

V =

819

1 m

3

QT =

9606

Nm3

h-1

6 M

ixer

s / 2

8 G

rids

V =

819

1 m

3

QT =

9606

Nm3

h-1

4 M

ixer

s / 2

8 G

rids

V =

819

1 m

3

QT =

9606

Nm3

h-1

6 M

ixer

s / 1

8 G

rids

V =

293

13 m

3

QT =

135

00 N

m3

h-1

8 M

ixer

s / 2

0 G

rids

V =

286

46 m

3

QT =

126

53 N

m3

h-1

8 M

ixer

s / 2

4 G

rids

Uc

(cm s-

1 )

Experimental CFD

Fig. 3. Axial liquid velocities for each basin and whatever operating conditions(a) without aeration and (b) with aeration. (, experimental; , CFD).

(Fayolle et al., 2006). The coalescence and rupture phenomenaare not taken into account in the evolution because of the lowaverage gas hold-up in the aerated tank. The consumption ofoxygen contained in bubbles during their rising is consideredto have any effect on their size (Jupsin and Vasel, 2002).

Mixing is modelled by the xed value method validated byVermande et al. (2003). This method consists in simplifyingthe numerical resolution by ignoring the ow equations cross-ing the agitator and considering the mixers as a source of mo-mentum. Characteristics of the ow induced by agitation areimposed on a zone of the grid corresponding to the location of

the mixer. Moreover, no boundary condition concerning turbu-lence is imposed (Chatellier, 1991; Djebbar, 1996).

The free surface is modelled by a degasication condition forthe gas phase and a symmetry condition for the liquid phase.In this case, the surface is considered as mobile, strengthlessand horizontal. The oxygen transfer from the free surface isneglected in comparison to the bubble interfacial area. Indeed,the order of magnitude of bubble/liquid interfacial area (ab =310 m1) is more important than the free surface interfacialarea (aS = STVT = 1dw ), particularly for deep aeration tanks.

Finally, all simulations have been performed in clean water(L = 1000.0 kg m3; L = 1.0 103 kg m1 s1) and theow is assumed to be isothermal (the simulations take placewith water and air at T = 20 C).

2.2.4. Mesh gridThe ow structure in aerated tanks is three-dimensional

(Cockx et al., 2001). For each studied tanks, 3D grid mesheshave been created (using the pre-processor GAMBIT). Allmeshes are unstructured and composed of tetrahedrons. Thenumber of cells of resulting grids are summarized in Table 3.The grid has been rened near the wall for the pilot scale aera-tion tank to simulate precisely the ow characteristics. The sizeof the grid is sufcient to have a grid-independent solution.

3. Comparison of CFD and experimental results

Axial liquid velocity, gas hold-up proles and oxygen trans-fer coefcients obtained by CFD are compared with the exper-imental results.

3.1. Axial liquid velocity

3.1.1. Axial liquid velocity prolesThe axial liquid velocity proles have been measured in

a non-aerated experimental section using a mono-directionalow-meter.

Fig. 1 represents simulated streamlines respectively in tank1 without aeration, and in tank 4 without and with aeration.The red rectangle represents the section of experimental axialvelocity measurement.

The ow in aeration tanks is mainly axial without aeration(Fig. 1(a) and (b)). However, with aeration, the streamlinesare deformed by the gas injection (Fig. 1(c)). Zones of re-circulation due to the drag of the liquid by the gas phase (spiralows) are observed in the aerated zone. Spiral ows are alsoobserved near the internal wall downstream from the bends.

Table 4Measurement section area of axial velocity, number of measurement pointsand the ratio between the number of measurement points and measurementsection area

Tank 1 2 3 4

Measurement section area (m2) 0.2 33.8 70.2 142.2Number of measurement points 20 35 30 48Number of points/m2 ofmeasurement section

100.0 1.0 0.4 0.3


0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0 0.1 0.2 0.3 0.4 0.5Distance from internal wall (m)






Loca

l gas

hol

d-up

(-)

Loca

l gas

hol

d-up

(-)

Loca

l gas

hol

d-up

(-)

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

Loca

l gas

hol

d-up

(-)

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

Loca

l gas

hol

d-up

(-)

Loca

l gas

hol

d-up

(-)

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.09

0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.09

0

0.01

0.02

0.03

0.04

0.050.06

0.07

0.08

0.09

QT = 2.4 m3 h-1

UC = 0 m s-1QT = 3.0 m3 h-1

UC = 0 m s-1

QT = 2.4 m3 h-1

UC = 0 m s-1

QT = 2.4 m3 h-1

UC = 0 m s-1

QT = 3.0 m3 h-1

UC = 0.3 m s-1

QT = 3.0 m3 h-1

UC = 0.1 m s-1

Fig. 4. Transverse gas hold-up prole for different air ow rates (QT) and mean axial liquid velocity (UC). experimental data / - numerical data.

Fig. 2 presents the evolution of local axial liquid velocitiesversus the distance from the internal wall for tanks 1 (1 m3)and 4 (29, 313 m3). Velocity proles without (tanks 1 and 4)and with aeration (tank 4) are presented for a distance fromthe bottom of the tank (z) of 0.35 m (tank 1) and 2.06 m (tank4). Velocity measurement sections are located away from themixers.

For pilot and full scale basins, axial liquid velocity prolesmeasured without and with aeration are well predicted by themodel.

For tank 1 without aeration, the experimental axial liquidvelocity increases from 0.18 to 0.51 m s1 with the distance

from the internal wall. The experimental heterogeneity of thelocal axial velocity is well predicted by the model.

For tank 4 without aeration, the axial liquid velocity variesfrom 0.22 to 0.42 m s1 with the distance from the internal wall.The axial liquid velocity prole is relatively homogenous onthe measurement section and is well reproduced by simulation.

For tank 4 with aeration, the axial liquid velocity variesfrom 0.28 to 0.03 m s1 with the distance from the in-ternal wall. With aeration, the axial liquid velocity prole isstrongly impacted when the ow crosses the bubble plume (themeasurement section is placed downstream from this plume).The numerical simulation also allows to determine the order of


0

0.5

1

1.5

2

2.5

3

V =

1 m

3

QT =

2.58 N

m3 h-

1

UC

= 10

cm

s-1

V =

1 m

3

QT =

2.58 N

m3 h-

1

UC

= 30

cm

s-1

V =

1 m

3

QT =

3.25 N

m3 h-

1

UC

=10

cm s-

1

V =

1 m

3

QT =

3.25 N

m3 h-

1

UC

= 30

cm

s-1

V =

149

3 m

3

QT =

1555

Nm3

h-

1

0 M

ixer

/ 6

Grid

s

V =

149

3 m

3

QT =

1555

Nm3

h-

1

1 M

ixer

/ 6

Grid

s

V =

819

1 m

3 QT

= 96

06 N

m3 h-

1

6 M

ixer

s / 1

8 G

rids

V =

286

46 m

3

QT =

126

53 N

m3

h-1

8 M

ixer

s / 2

4 G

rids

k La

20x10

3 (s-

1 )

Experimental CFD

Fig. 5. Oxygen transfer coefcient for each basin and operating conditions (, experimental; , CFD).

Table 5Oxygen transfer coefcient as a function of the input bubble Sauter diameter

Bubble Sauter diameter at P = 1013 hPa (mm) 4.29 4.77 5.25kLa20 103 (s1) 2.19 1.91 1.71Variation/db = 4.77 mm (%) 15 11

magnitude of local axial liquid velocities. However, differencesbetween experimental and CFD results on full scale aerationtank are more important than those observed on pilot scale tank.This may be due to the placement of the measurement probewhich is not known with precision on full scale aeration tank.A measurement error related to the position of the probe shouldbe added to the sampling error which is indicated in Fig. 2.

3.1.2. Mean axial liquid velocityThe good prediction of the local axial liquid velocities in-

duces a reliable determination of the mean axial liquid circu-lation velocity, which is deduced by integration of the axialcomponents of the velocity vector on the measurement surface.The experimental mean axial liquid velocity is calculated asthe arithmetic average of the velocities measured on the ex-perimental section. The results without and with aeration aresummarized in Fig. 3. An axial velocity measurement error of10% has been previously determined on full scale tank (Fayolleet al., 2006) and is indicated in Fig. 3.

Without aeration, the mean horizontal velocity is well pre-dicted (differences between experimental and numerical dataare about 10%) for each studied tank and whatever the operatingconditions. With aeration, the difference between experimentaland numerical data is more important. This difference couldbe mainly due to the difcult estimation of the pumping owrate of the mixers. The inuence of the aeration on this owrate is indeed neglected. The observed difference could also bedue to a number of experimental points which was certainly

insufcient to determine the mean axial velocity. A previousstudy (Fayolle, 2006) showed that one measurement point persquare meter of section was necessary to determine preciselythis value. The number of measurement points for tanks 3 and4 was lower than this recommendation (Table 4).

3.2. Gas hold-up

For the pilot scale aeration tank, the computed gas hold-upproles are compared to experimental local measurements us-ing an optical probe (Simon, 2000). Simulations are performedfor two different air ow rates (QT = 2.4 and 3.0 m3 h1) andthree different axial liquid velocities (UC=0, 0.1 and 0.3 m s1)corresponding to the experimental operating conditions. The re-sults in terms of transversal gas hold-up proles (at z= 0.18 mfrom the diffusers) are summarized in Fig. 4.

A good agreement between experimental and numerical gashold-up proles is obtained whatever the operating conditions,in the pilot scale aeration tank. The differences observed be-tween numerical and experimental gas hold-up proles on thepilot aeration tank probably result from:

the numerical gas injection that is numerically realized on thetotal surface of the diffusers and not only on the experimentalperforated zones of the diffusers;

the experimental position of the probe which is not clearlydened (in the centre of the bubble plume, Simon, 2000).

3.3. Oxygen transfer

When the oxygen transfer is simulated, the dissolved oxy-gen concentration in the liquid phase is plotted versus time atseveral locations corresponding to the experimental placementof the dissolved oxygen probes. The apparent oxygen transfer


coefcient (kLa) is determined by tting the curve with thefollowing exponential function, based on the perfectly mixedreactor hypothesis (NF-EN-12255-15, 2004):CL = CS (CS C0) ekLa20t , (12)where CL is the dissolved oxygen (DO) concentration in water,C0 is the initial DO concentration in water, CS is the DO con-centration in water at the saturation and kLa20 is the oxygentransfer coefcient at T = 20 C.

The oxygen transfer coefcients are deduced from experi-mental and simulated evolution of the oxygen concentration inthe liquid versus time. The obtained coefcients are reportedin Fig. 5.

For both pilot and full scale tanks, the oxygen transfercoefcient is well predicted by simulation whatever the operat-ing conditions (difference between experimental and computeddata lower than 5%). The simulation of two-phase ows, cou-pled with oxygen transport for each phase, allows to determineprecisely the oxygen transfer coefcient of full scale aerationtanks by taking into account the complex and competitiveeffects of the pressure.

3.4. Inuence of the bubble size on the computed oxygentransfer coefcient

The bubble size is the dominating parameter in oxygen trans-fer phenomena, as, together with the gas hold-up, it governs theinterfacial area. During CFD simulation, the global gas hold-up is computed whereas the bubble size is an input parameter.A sensitivity analysis of the CFD results to the input bubblesize was therefore realized on tank 2. The input bubble Sauterdiameter was modied by 10% from the experimental value.The computed global oxygen transfer coefcients are shown inTable 5.

A decrease in the bubble size of 10% induces an increase inthe oxygen transfer coefcient by 15%. Inversely, increasingthe bubble diameter by 10% induces a decrease in the oxygentransfer coefcient (11%). The evolution of kLa20 with dbsis not linear because an increase in the bubble size has twomain impacts: (i) an increase in the local interfacial area and(ii) an increase in the local mass transfer coefcient. Theseresults conrm the necessity to measure in situ the air bubblediameter. Adequate tools to estimate and model the bubble sizeat the diffuser level are required to overcome this issue.

4. Conclusions

A numerical tool to simulate ow characteristics and oxy-gen transfer in aeration tanks has been presented. The proposedprotocol includes the main numerical models to be used, espe-cially concerning the drag force and the drift velocity. This toolhas been validated by comparison of measured and computeddata in terms of axial liquid velocity, local gas hold-up and oxy-gen transfer coefcient for four tanks of different sizes (pilotand full scale). The axial liquid velocities and the gas hold-up proles are well predicted by the model. Moreover, the ap-parent oxygen transfer coefcient is precisely reproduced with

an accuracy of 5%. The proposed protocol allows taking intoaccount the complex and competitive effects of the pressure,principally on the bubble size along their rise. This validatedmodel will be used to predict and to optimize oxygenation ca-pacities of full scale tanks. However, the predicted accuracy isdependant of the inlet bubble size value.

Notation

a bubble interfacial area, m1Ap projected interfacial area, m2ck local instantaneous scalar concentration in phase

k, kg m3ck local instantaneous turbulent component of scalar

concentration in phase k, kg m3C0 initial dissolved oxygen (DO) concentration in

water, kg m3CD drag coefcient, dimensionlessCG concentration of oxygen in the bubble, mg l1CL DO concentration in water, kg m3CL instantaneous DO concentration at the gas-liquid

interface, kg m3CS DO concentration at the saturation, kg m3db bubble diameter, mdbS bubble Sauter diameter, mdw water depth, mDL diffusion coefcient of the oxygen in the water,

m2 s1Eo Etvos number, dimensionlessFD drag force, kg m s2g gravitational acceleration constant, kg m2He Henrys coefcient, Pa m3 kg1Jk ux due to molecular diffusion, mol s1 m2kL local mass transfer coefcient, m s1kLa20 oxygen transfer coefcient in standard conditions

(P = 1.013 105 Pa and T = 20C), s1l width of the basin, mL length of the basin, mLk interfacial transfer of concentration between the

two phases, kg m3 s1m coefcient linked to Henrys coefcient, dimen-

sionlessMk local instantaneous interfacial momentum trans-

fer, kg m2 s2MO2 molecular mass of O2, kg mol1PG pressure in bubble, PaPk local statistical averaged phase pressure,

kg m1 s2Qdiff mean air ow rate per diffuser, m3 h1QT total air ow rate, m3 h1R gas constant, (R = 8.314 J mol1 K1)T temperature, CUC mean axial liquid velocity, m s1vd drift velocity, m s1vk local instantaneous phase velocity of

phase k, m s1


vk local instantaneous turbulent component of phasevelocity,m s1

Vr relative velocity, m s1ye volume fraction of oxygen in air, dimensionlessz distance from the bottom of the tank, m

Greek letters

k retention of phase k, dimensionless dissipation rate of the turbulent kinetic energy,

m2 s3L water viscosity, kg m1 s1k density of phase k, kg m3 supercial tension between air and liquid, kg s2k viscous stress tensor in phase k, kg m1 s2

Acknowledgements

The authors are grateful to the persons who were involvedin the experimental measurements on full scale aerationtanks.

References

Bel Fdhila, R., Simonin, O., 1992. Eulerian prediction of turbulent bubblyow downstream of a sudden pipe expansion. In: Workshop on Two-phaseFlow Predictions, 30 March2 April, Erlangen.

Chatellier, P., 1991. Simulation de lhydrodynamique des chenaux doxydationpar lutilisation des quations de NavierStokes associes au modle Ke:evaluation de la vitesse de circulation. Thse de doctorat, INSA, Toulouse,168pp. + annexes.

Clift, R., Grace, J.R., Weber, M.E., 1978. Bubbles, Drops and Particles.Academic Press, New York.

Cockx, A., Do-Quang, Z., Lin, A., Roustan, M., 1999. Use of computationaluid dynamics for simulating hydrodynamics and mass transfer inindustrial ozonation towers. Chemical Engineering Science 54, 50855090.

Cockx, A., Do-Quang, Z., Audic, J.M., Lin, A., Roustan, M., 2001. Globaland local mass transfer coefcients in waste water treatment process bycomputational uid dynamics. Chemical Engineering and Processing 40,187194.

Djebbar, R., 1996. Etude hydrodynamique des racteurs de type chenauxdoxydation par la mcanique des uides numrique. Post Doc, INSA,Toulouse.

Fayolle, Y., 2006. Modlisation de lhydrodynamique et du transfert doxygnedans les chenaux daration. Thse de doctorat, Ecole doctorale Transfert,Dynamique des Fluides, Energtique et Procds, INSA de Toulouse,295pp. + annexes.

Fayolle, Y., Gillot, S., Cockx, A., Roustan, M., Hduit, A., 2006. In situlocal parameter measurements for CFD modelling to optimize aeration. In:79th Annual Technical Exhibition and Conference WEFTEC06, October2125, Dallas, USA, 12pp.

Gillot, S., Capela, S., Hduit, A., 2000. Effect of horizontal ow on oxygentransfer in clean water with surfactants. Water Research 34 (2), 678683.

Gillot, S., Capela-Marsal, S., Carrand, G., Wouters-Wasiak, K., Baptiste,P., Hduit, A., 2005. Fine bubble aeration with EPDM membranes:conclusions from 15 years of practice. In: IWA Specialized Conferenceon Nutrient Management in Wastewater Treatment Processes and RecycleStreams, Krakow, Poland, pp. 607616.

Groves, K.P., Daigger, G.T., Simpkin, T.J., Redmon, D.T., Ewing, L., 1992.Evaluation of oxygen transfer and alpha-factor on a variety of diffusedaeration systems. Water Environment Research 64 (5), 691698.

Higbie, R., 1935. The rate of absorption of a pure gas into a still liquidduring short periods of exposure. Am. Inst. Chem. Eng. 365389.

Hinze, J.O., 1975. Turbulence. Mcgraw-Hill, New York.Ishii, M., 1975. Thermo-uid Dynamic Theory of Two-phase Flow. Eyrolles,

Paris.Jupsin, H., Vasel, J.-L., 2002. Dtermination du Cs dans des racteurs ars par

insufation. 4mes Journes Francophones sur les Reacteurs GazLiquideet GazLiquideSolide, p. 8.

Mueller, J.A., Boyle, W.C., Ppel, H.J., 2002. Aeration: principle and practice.In: Water Quality Management Library Series, 353pp.

NF-EN-12255-15 (2004). Wastewater treatment plantsPart 15:measurements of the oxygen transfer in clean water in aeration tanks ofactivated sludge plants.

Schiller, L., Naumann, Z., 1938. A drag coefcient correlation. Z Ver DeutschIng. 77318.

Simon, S., 2000. Etude dun chenal doxydation par des approches globaleset localeshydrodynamique et transfert de matire. Thse de doctorat,INSA, Toulouse, 188pp. + annexes.

Talvy, S., Cockx, A., Line, A., 2007. Modeling hydrodynamics of gasliquidairlift reactor. A.I.Ch.E. Journal 53 (2), 335353.

Vermande, S., Essemiani, K., Meinhold, J., De Traversay, C., Fonade, C.,2003. Trouble shooting of agitation in an oxidation ditch: applicability ofhydraulic modeling. In: 76th Annual Technical Exhibition and ConferenceWEFTEC03, October 1115, Los Angeles, USA.

Vermande, S., Chaumaz, M., Marsal, S., Dumoulin, L., Essemiani, K.,Meinhold, J., 2005. Modlisation numrique dun bassin grandeprofondeur. Rcent progrs en Gnie des Procds 92, 8pp.

Wagner, M.R., Ppel, H.J., 1998. Oxygen transfer and aerationefciencyinuence of diffuser submergence, diffuser density, and blowertype. Water Science and Technology 38 (3), 16.

Oxygen transfer prediction in aeration tanks using CFDIntroductionMaterials and methodsExperimental set-upMathematical modellingTwo-phase hydrodynamicsTwo-phase mass transferBoundary conditionsMesh grid

Comparison of CFD and experimental resultsAxial liquid velocityAxial liquid velocity profilesMean axial liquid velocity

Gas hold-upOxygen transferInfluence of the bubble size on the computed oxygen transfer coefficient

ConclusionsAcknowledgementsReferences

oxygen transfer prediction

Documents