cespart1_05.pdf

Seediscussions,stats,andauthorprofilesforthispublicationat:https://www.researchgate.net/publication/222648682

ReynoldsnumberscalingofflowinaRushtonturbinestirredtank.PartIMeanflow,circularjetandtipvortexscalingARTICLEinCHEMICALENGINEERINGSCIENCEJUNE2005ImpactFactor:2.34DOI:10.1016/j.ces.2004.12.039Source:OAI

CITATIONS40

READS185

5AUTHORS,INCLUDING:

DavidFHillOregonStateUniversity58PUBLICATIONS617CITATIONS

SEEPROFILE

SivaramakrishnanBalachandarUniversityofFlorida316PUBLICATIONS5,840CITATIONS

SEEPROFILE

RonaldJ.AdrianArizonaStateUniversity329PUBLICATIONS16,091CITATIONS

SEEPROFILE

ManYeongHaPusanNationalUniversity182PUBLICATIONS1,707CITATIONS

SEEPROFILE

Allin-textreferencesunderlinedinbluearelinkedtopublicationsonResearchGate,lettingyouaccessandreadthemimmediately.

Availablefrom:SivaramakrishnanBalachandarRetrievedon:28January2016

Chemical Engineering Science 60 (2005) 31693183www.elsevier.com/locate/ces

Reynolds number scaling of ow in a Rushton turbine stirred tank. PartIMean ow, circular jet and tip vortex scalingH.S.Yoona, D.F. Hillb, S. Balachandarc,, R.J. Adrianc, M.Y. Haa

aSchool of Mechanical Engineering, Pusan National University, San 30 Chang Jeon Dong, South KoreabDepartment of Civil & Environmental Engineering, Pennsylvania State University, 212 Sackett Building, University Park, PA, 16802, USA

cDepartment of Theoretical & Applied Mechanics, 216 Talbot Laboratory, University of Illinois at Urbana-Champaign, NC-262,104 South Wright Street, Urbana, IL, 61801 2983, USA

Received 4 November 2003; received in revised form 4 November 2004; accepted 23 December 2004Available online 5 March 2005

Abstract

We consider scaling of ow within a stirred tank with increasing Reynolds number. Experimental results obtained from two differenttanks of diameter 152.5 and 292.1mm, with a Rushton turbine operating at a wide range of rotational speeds stirring the uid, areconsidered. The Reynolds number ranges from 4000 to about 78,000. Phase-locked stereoscopic PIV measurements on three differentvertical planes close to the impeller give phase-averaged mean ow on a cylindrical surface around the impeller. The scaling of - andplane-averaged radial, circumferential and axial mean velocity components is rst explored. A theoretical model for the impeller-inducedow is used to extract the strength and size of the three dominant elements of the mean ow, namely the circumferential ow, the jetow and the pairs of tip vortices. The scaling of these parameters with Reynolds number for the two different tanks is then obtained.The plane-averaged mean velocity scales with the blade tip velocity above a Reynolds number of about 15,000. However, parametersassociated with the jet and tip vortices do not become Reynolds number independence until Re exceeds about 105. The results for the twotanks exhibit similar Reynolds number dependence, however, a perfect collapse is not observed, suggesting a sensitive dependence of themean ow to the ner details of the impeller. 2005 Published by Elsevier Ltd.

Keywords: Rushton turbine; Turbulence; Mixing; Scaling; Particle image velocimetry

1. Introduction

Scaling of ow and mixing within a stirred tank reactor isof signicant practical importance. For lack of a satisfactoryunderstanding, the design of production-scale stirred tanksoften evolves through a number of stages that iterate betweenlaboratory experiments and pilot plants, a time-consumingand expensive step-by-step scale-up process. Part of the dif-culty is that mean ow and turbulence quantities, such asrms uctuation and dissipation, often scale differently withincreasing tank size and operating speed. As a result thescaling of mixing within the tank can be complex.

Corresponding author. Tel.: +1 217 244 4371; fax: +1 217 244 9090.E-mail address: [email protected] (S. Balachandar).

0009-2509/$ - see front matter 2005 Published by Elsevier Ltd.doi:10.1016/j.ces.2004.12.039

The last three decades have seen several high qualityexperimental measurements of ow inside stirred tanks us-ing modern techniques such as laser doppler anemometry(LDA) and particle image velocimetry (PIV) (Desouza andPike, 1972; Vant Riet and Smith, 1975; van der Molen andvan Maanen, 1978; Kolar et al., 1984; Costes and Couderc,1988; Dong et al., 1994; Sturesson et al., 1995; Stoots andCalabrese, 1995; Ducoste et al., 1997; Kemoun et al., 1998;Lamberto et al., 1999; Montante et al., 1999; Mahouastet al., 1989; Schaffer et al., 1997; Derksen et al., 1999;Escudie and Line, 2003; Escudie et al., 2004). In particular,several of these experimental investigations have consideredow and mixing inside stirred tanks of varying size operat-ing over a range of speeds and, thus, have addressed, directlyor indirectly, the scaling of mean and turbulence quantities

3170 H.S. Yoon et al. / Chemical Engineering Science 60 (2005) 31693183

with Reynolds number (Desouza and Pike, 1972; Vant Rietand Smith, 1975; van der Molen and van Maanen, 1978;Kolar et al., 1984; Costes and Couderc, 1988; Dong et al.,1994; Sturesson et al., 1995; Stoots and Calabrese, 1995;Ducoste et al., 1997; Kemoun et al., 1998; Lamberto et al.,1999; Montante et al., 1999). These studies together cover awide range of Reynolds number from 10 to 1.2105, whereReynolds number is dened in terms of impeller blade-tipdiameter, D, and the number of blade rotations per second,N, as Re=ND2/ ( is the kinematic viscosity of the uid).However, the range of Re considered in each investigationis limited. At higher Reynolds numbers relevant to turbulentow, the overall time-averaged mean ow and rms turbu-lent uctuations appear to scale with the impeller blade-tipvelocity.Costes and Couderc (1988) have addressed the problem

of scaling of mean ow and turbulence using two differentsized tanks over three different Reynolds numbers. Theirresults suggest that velocity statistics such as, mean, rms,spectra and autocorrelation scale with the blade-tip velocity.However, higher order turbulence statistics, such as dissipa-tion, do not show complete collapse when scaled appropri-ately in terms of impeller diameter and rotation rate. Theresults of Stoots and Calabrese (1995) on deformation rateover a Reynolds number range of 29,20045,800, however,show a reasonable Reynolds number independence whennondimensionalized by the inverse time scale.van der Molen and van Maanen (1978) emphasized the

importance of blade-tip trailing vortices. They pointed outthat the time-averaged ow in the laboratory frame of ref-erence (as measured by a xed probe) averages out the in-uence of tip vortices as they sweep past the probe alongwith the blades. The resulting time-averaged mean owwas observed to scale well with the blade-tip velocity. Theeffect of tip vortex pairs was carefully isolated throughphase-average, and the tip vortex strength did not scale per-fectly with blade-tip velocity. Some dependence on tanksize was observed. Vant Riet and Smith (1975) also investi-gated the scaling of blade-tip vortices over a wide Reynoldsnumber range of 30090,000. The vortex trajectory was ob-served to be Reynolds number dependent at lower, transi-tional Reynolds numbers, but Reynolds number indepen-dent, within experimental uncertainty, above a Reynoldsnumber of 15,000. Other quantities such as vortex strength,when appropriately scaled, also tended towards Reynoldsnumber independence at higher Reynolds numbers.In spite of the above efforts the scaling of ow and mixing

within a stirred tank with increasing tank size and impellerspeed remains not fully understood. In this paper, we addressthe question of scaling with experiments performed in twodifferent tanks with impeller speed varying over Reynoldsnumbers ranging from 4000 to 80,000. The two tanks wereconstructed to be geometrically similar. Both employ a lidat the top of the tank in order to prevent free surface (Froudenumber) effects. The ow is entirely driven by the rotat-ing impeller and the operational speeds are such that Mach

and Rossby numbers are irrelevant. Reynolds number is ex-pected to be the only relevant parameter of the problem. Theresults for each tank, when appropriately nondimensional-ized by the blade tip radius and velocity, show Reynoldsnumber independence with increasing impeller speed at suf-ciently high Reynolds number. However, surprisingly theresults for the two different tanks do not exhibit a perfectcollapse, suggesting sensitive dependence on small differ-ences, especially in the geometric scaling of their impellers.Stereoscopic PIV measurements were made on three

different vertical planes within the tank. The instantaneousmeasurements are phase-locked with the blade position andensemble averaging over many such realizations yields thephase-averaged mean velocity. The measurement on thethree planes, which are located close to the impeller sweptvolume, are interpolated to obtain all three components ofthe phase-averaged mean velocity on a cylindrical plane ofconstant radius located just beyond the impeller tip. Thevelocity on this cylindrical plane is the impeller-inducedinow and it can be considered to dictate the ow overthe entire tank at large (Yoon et al., 2002). Thus, here weconsider the scaling of velocity measured over this plane asa proxy for scaling of ow over the entire stirred tank.The phase-averaged mean ow within the tank stirred

with a Rushton turbine can be considered to be made up ofthree different basic ow elements: circumferential ow, ajet ow and pairs of tip vortices associated with the impellerblades. This simple decomposition has been shown to be ef-fective in modeling the impeller induced ow, particularlyin the neighborhood of the impeller (see Yoon et al., 2001).Here we investigate the scaling of each of these elementsindividually with increasing tank size and impeller rotationrate. By looking at the scaling of parameters, such as thestrength and size of the jet and tip vortex pairs individu-ally, we hope to address the question of mean ow scalingin more detail. In particular, it is of interest to establish theminimum Reynolds number necessary for these parametersto be Reynolds number independent.Here we will also investigate the scaling of both vorticity

and dissipation for the two tanks and in particular, addresswhen they become Reynolds number independent. If thescaling of the mean ow were to be uniform over the entiretank, we expect the scaling of these higher order quantitiesto follow that of the mean ow. However, it can be antici-pated that the Reynolds number independence of the nondi-mensional vorticity and dissipation will be delayed to muchhigher Reynolds numbers, since these derivative quantitiesgive more importance to the smaller scales of motion. Thus,the results on the scaling of vorticity and dissipation can beused to interpret the scale-dependence of mean ow scal-ing. The present work is limited to investigation of only thescaling of velocity eld. Practical factors relevant to indus-trial mixing such as mixing time, heat and mass transfer, andmultiphase ow are not considered in this study. However,the scaling properties of the ow to be discussed here willlikely have an effect on these additional complex processes.

H.S. Yoon et al. / Chemical Engineering Science 60 (2005) 31693183 3171

Fig. 1. Overhead schematic of experimental apparatus and data acquisition system.

2. Experimental methodology

2.1. Apparatus

The experiments were an extension of more limited ex-periments rst reported by Hill et al. (2000). A schematicoverhead view of the set-up for the current stereoscopic PIVexperiments is shown in Fig. 1. To assess the inuence of ge-ometric scaling, two geometrically similar test sections wereconsidered. In both cases, an acrylic, unbafed1 circularcylinder was mounted within a slightly larger square tank.The diameters of the two circular tanks were T1= 152.5 cmand T2 = 292.1 cm. The cylinder was lled with de-ionizedwater to a depth equal to its diameter and the volume be-tween the cylindrical and the square tanks was also lled,in order to reduce the optical distortion due to the curvedsurface. A lid, with a small hole to accommodate the im-peller shaft, was placed on top of the liquid. The purpose of

1 The use of bafes is a practical means by which to impede therotation of the uid. However, one purpose of these experiments was toobtain detailed data for comparison to CFD results, so the simple bafe-free geometry was chosen.

the lid was to suppress the free-surface displacement, whichbecomes signicant otherwise at higher speeds of impellerrotation. The tolerances between the shaft, lid, and the cylin-drical tank were all very tight.The two tanks were geometrically similar. The lid on top

eliminates any free surface and associated Froude numbereffect. The ow within the tank is completely determinedby the impeller rotation and therefore in the present prob-lem the rotational and velocity scales are not independent(Rossby number becomes irrelevant). Furthermore, the ve-locities considered are quite low that Mach number effectcan be ignored. Thus dynamic similarity between the twotanks could be achieved by simply matching the Reynoldsnumbers. However, some subtle differences in the impellersof the two different arrangements must be noted. The Rush-ton impeller for the small tank was supplied by Lightnin,Inc., and the impeller for the large tank was custom madeby Proprmix, Inc. Fig. 1 also shows the detailed sketch ofthe impeller geometry with the corresponding measurementsfor the two tanks. While the diameter of the impeller, bladelength, blade height and diameter of the disk were geomet-rically scaled, the blade thickness, disk thickness, sleeve


T

T/2

T

D

B

z

r

N

R

Fig. 2. Side and top view details of the tank and the impeller. Details ofthe three data acquisition planes and the cylindrical interpolation surface.

diameter were not perfectly scaled between the small andlarge tanks. The ramications of these seemingly minor dif-ferences will be discussed later.The Rushton turbine was mounted at mid-depth along the

axis of the cylinder. The shaft extends through the impellerto the bottom of the tank, where it was held in place by asmall nylon sleeve bearing. The bearing eliminated wobbleof the impeller. Detailed views of the experimental tankand impeller are shown in Fig. 2. Two different LightninLabmaster mixers drove the impellers: the rst with a speedrange of 501800 rpm and the second with a speed range of20550 rpm. Details, such as the mounting of the impeller,position of the top lid, etc., for the two tanks are maintainedgeometrically similar.The ow was seeded with 8-m particles from Pot-

ters Industries. New-Wave Nd:Yag lasers, operating at 50mJ/pulse, illuminated the ow with a light sheet ap-proximately 1mm thick. Two TSI PIVCAM 1030 CCDcameras (1K 1K resolution), equipped with Nikon AFMicroNikkor 60mm lenses, acquired images of the illumi-nated particles. The included angle between the angularlyoffset cameras was roughly 45. Scheimpug lens mountswere used to reduce the required depth of eld and to im-prove focusing over the eld of view. A TSI Laser PulseSynchronizer and a PC equipped with TSI Stereo Insightsoftware completed the image acquisition package.

2.2. Experimental procedure

Because of the lid, the only upper bounds on impellerspeed were those placed by the power and speed limits of

Table 1Rotational speed and Reynolds numbers of experimental data sets

Small tank Large tank

(rpm) Re =ND2/ (rpm) Re =ND2/100 4293 27 4229150 6440 41 6421367 15,756 100 15,662551 23,655 150 23,492734 31,512 200 31,3231102 47,311 300 46,9851469 63,066 400 62,646

500 78,308

the mixers. As a result, it was possible to cover Reynoldsnumbers ranging from 4000 to 80,000. Table 1 details theexperimental dimensions for both tanks.As indicated in Fig. 2, data were obtained on three vertical

planes located very close to the tips of the impeller blades.An optical encoder, focused on the impeller shaft, was usedto trigger acquisition such that the impeller blades were inthe position shown for each realization. For the small tankexperiments, these planes were located at distances of 23.22,25.13, and 27.67mm from the impeller shaft. For the largetank experiments, these planes were located at distances of45.12, 48.77, and 53.64mm. For each combination of im-peller speed and acquisition plane, an ensemble of 500 re-alizations was obtained.The raw images from the left and right cameras were

individually interrogated and then validated. Finally, thetwo two-dimensional vector elds were combined, usinga mapping determined by calibration, into a single three-component velocity vector eld on a planar surface. The in-dividual data sets were rst ensemble averaged to yield amean velocity eld.2 The mean velocity elds on the threevertical planes for a given rotational speed were then in-terpolated, using a second-order t, onto a 60 segment ofa cylindrical shell of radius R, as illustrated in Fig. 2. Fi-nally, the Cartesian velocity components were rotated intopolar cylindrical coordinates, yielding radial, circumferen-tial and axial velocity components, ur , u and uz, onthe curved cylindrical plane, i.e., for r = 1.06, 060and 0.6z0.6 (see Fig. 2 for coordinate denition).Here angle brackets indicate an average over the ensem-ble. Note that here and in what follows all lengths arenondimensionalized by the blade tip radius (D/2) and thevelocities have been nondimensionalized by the blade-tipvelocity (ND).

2 With regards to the present data, the term mean ow denotesphase-averaged ow obtained from the ensemble average of the phase-locked measurement. It is therefore dependent on all three coordinates.Any further space average of the mean ow will be explicitly stated inwhat follows.


3. Results

3.1. Mean ow scaling

The mean ow ensemble-averaged over all the realiza-tions (ur , u, uz) and interpolated onto the cylindricalsurface is shown in Fig. 3 for the small tank at the low-est rotation speed, corresponding to Re = 4300. In Fig. 3frame (a) shows the in-plane velocity vector plot and frame(b) shows the out-of-plane (radial) velocity contours. Onlya 60 sector is shown, and the view is limited to the top halfof the tank with the region below the center-plane obtainedby symmetry. The -dependence of the ow eld is due tothe presence of the tip vortices, whose impact is most evi-dent in the gure around =181 and z 0.10.1. The

Fig. 3. (a) The in-plane velocity vector plot and (b) the out-of-plane radialvelocity contours of the phase-averaged mean ow at Re= 4300 for thesmall tank. (c) The in-plane velocity vector plot and (d) the out-of-planeradial velocity contours at Re = 63,180 for the small tank.

Fig. 4. (a) The in-plane velocity vector plot and (b) the out-of-plane radialvelocity contours of the phase-averaged mean ow at Re= 4299 for thelarge tank. (c) The in-plane velocity vector plot and (d) the out-of-planeradial velocity contours at Re = 62, 646 for the large tank.

tip vortex pair cuts the cylindrical plane at an angle, and itssense of rotation is such that it enhances the radial velocitydue to the jet at the midplane (z = 0), but opposes the jetsufciently away from the midplane. In fact, the negativeradial velocity and its localized peak around = 18 andz 0.2 is a clear signature of the tip vortex. This obser-vation is quite consistent with the recent measurements ofEscudie et al. (2004). Fig. 3 (frames c and d) shows the meanow for the small tank corresponding to a higher Reynoldsnumber of 63,180. Although the in-plane and radial veloc-ity components are qualitatively similar to the low Reynoldsnumber case, measurable quantitative differences can be ob-served, indicating a Reynolds number effect. For example,the peak (positive and negative) radial velocities are some-what lower at the higher Reynolds number than at the lowerRe. Fig. 4 shows the mean ow results for the large tank at


Re = 4229 and 62,646. They are, again, qualitatively com-parable to those shown in Fig. 3. Some differences can beobserved, in particular, the signature of the tip vortex movescloser to the blade (smaller values of ), suggesting somedifference in the tip vortex location and orientation. Again,a comparison of Fig. 4b with 4d reveals that the intensity ofradial velocity appears to decrease with increasing angularvelocity of the blade. However, for about the same Reynoldsnumber the radial velocity for the large tank is somewhatstronger than the small tank.The normalized mean radial, circumferential and axial ve-

locities averaged over , ur , u, uz, (overbar indicates-average) are shown in Fig. 5 for the small tank. Upon -average the effect of tip vortex pairs gets averaged, and thecontribution to the mean ow from the jet stands out, beingclearly evident as the rapid increase in the radial velocity asthe midplane is approached. The weak negative peak in theradial velocity around z = 0.2 is due to the tip vortex pair.Such negative radial velocity above and below the blade hasbeen observed in other measurements as well (Escudie andLine, 2003). However, here due to circumferential averagingthe negative peak is much weaker than observed in (Escudieand Line (2003). The asymptotic negative radial velocity ap-proached for large z is due to the broad return ow backtowards the impeller region. The jet has a strong circumfer-ential component oriented in the direction of blade rotation(note that is measured in the direction opposite to bladerotation). At this radial location (r = 1.06) the radial com-ponent of the jet is larger than the circumferential compo-nent. As can be seen in Fig. 5b, as z increases the circum-ferential velocity approaches a constant value, suggesting abackground circumferential ow, which is only weakly de-pendent on z. The magnitude of the radial velocity steadilydecreases with increasing Reynolds number, while the mag-nitude of circumferential velocity increases with Re. How-ever, a tendency towards Reynolds number independencefor these normalized velocity proles is observed with in-creasing Reynolds number. The axial velocity component ismuch weaker in magnitude, and it arises from the jet-inducedentrainment and also from the large tank-wide circulation.Although the variation with increasing Re is not monotonic,a tendency towards Reynolds number independence can beobserved here, as well.The above results for the small tank compare favorably

with those for the large tank shown in Fig. 6. The jet ampli-tude for the large tank appears to be slightly stronger thanthat for the small tank. Apart from this difference, the com-parisons of all three components of velocity in Figs. 6 and5 are quite good, especially in the high Reynolds numberrange. This observation, along with the differences observedearlier between Figs. 3 and 4, suggests that the tip vortexpairs depend on tank size. This may be due to the criticaldependence of tip vortices on the impeller geometry and thesmall difference in the impellers between the two tanks maybe contributing to this small, but noticeable, difference inthe tip vortex signatures.

-0.1 0 0.1 0.2 0.3 0.4 0.5

0

0.1

0.2

0.3

0.4

0.5

0.6

z

Re = 4300Re = 6450Re = 15785Re = 23700Re = 31570Re = 47400Re = 63180

-0.7 -0.6 -0.5 -0.4

0

0.1

0.2

0.3

0.4

0.5

0.6

zRe = 4300Re = 6450Re = 15785Re = 23700Re = 31570Re = 47400Re = 63180

-0.15 -0.1 -0.05 0

0

0.1

0.2

0.3

0.4

0.5

0.6

z

Re = 4300Re = 6450Re = 15785Re = 23700Re = 31570Re = 47400Re = 63180

(a)

(b)

(c)

Fig. 5. The normalized phase-averaged mean (a) radial, (b) circumferen-tial and (c) axial velocities averaged over the circumferential direction(ur , u, uz); overbar indicated -average) as a function of z for thedifferent Reynolds numbers for the small tank.


-0.1 0 0.1 0.2 0.3 0.4 0.5

0

0.1

0.2

0.3

0.4

0.5

z

Re = 4229Re = 6421Re = 15662Re = 23492Re = 31323Re = 46985Re = 62646Re = 78308

-0.7 -0.6 -0.5 -0.4

0

0.1

0.2

0.3

0.4

0.5

z


-0.15 -0.1 -0.05 0

0

0.1

0.2

0.3

0.4

0.5

z


(a)

(c)

Fig. 6. The same as Fig. 5 for the large tank.

Fig. 7a shows the normalized mean radial, circumferen-tial and axial velocities averaged over the entire z plane,ur , u, uz, (the double overbar indicates a planar

15000 30000 45000 60000 75000Re

-0.6

-0.5

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

Planar averaged velocities

Smalltank

Largetank

15000 30000 45000 60000 75000Re

-1

-0.8

-0.6

-0.4

-0.2

0

0.2

0.4

0.6

0.8

1Maximum

velocities

uruuz

uruuz

Smalltank

Largetank

(a)

(b)

Fig. 7. (a) The normalized phase-averaged mean radial, circumferentialand axial velocities averaged over the entire .z plane (ur , u, uz)(the double overbar indicates a planar average) as a function of Reynoldsnumbers for the two different tank sizes. (b) The maximum (over the .zplane) phase-averaged mean radial, circumferential and axial velocities asa function of Reynolds number for the two different tanks.

average) as a function of Re for the two different tank sizes.Above a Reynolds number of about 15,000 the results forall three components of velocity show very little variation.The average velocities in the two different tanks are inreasonable agreement as Re increases, except for the radialcomponent of velocity. The scaling of the maximum valueof ur , u and uz, maximum over the z plane, isshown in Fig. 7b as a function of Re for the two differenttank sizes. Even the axial component of velocity shows aweak dependence on tank size in its asymptotic value ap-propriate at large Re. These differences in the results for the


two tanks must have their origin in the subtle departuresfrom a perfect geometric scaling of the two impellers.Here, it should be pointed out that in some of the ear-

lier efforts (van der Molen and van Maanen, 1978; Ducosteet al., 1997) the comparison of different tank sizes is per-formed while maintaining the volume averaged mean dissi-pation rate to be the same. Since dissipation scales asN3D2,it can be expected to go as Re3 in a xed tank with in-creasing blade tip velocity, or as D4 with increasing tanksize for a xed Re. Fixed dissipation rate implies that withincreasing tank size one must consider increasing blade tipvelocity such that Re D4/3. The good degree of Reynoldsnumber independence exhibited in both the tanks in Fig. 7suggests that the lack of perfect collapse between the twotanks will persist even when compared on the basis of con-stant volume averaged mean dissipation rate.It is interesting to note that for the large tank at the low-

est Re considered the maximum circumferential velocity ex-ceeds the blade tip velocity (possibly due to the tip vor-tex inuence). With increasing Reynolds number, however,the magnitude of peak circumferential velocity is reducedbelow the blade tip velocity. While the peak radial veloc-ity remains comparable to the peak circumferential veloc-ity, in terms of the surface-averaged mean, the radial veloc-ity is an order of magnitude weaker than the circumferen-tial component. This is to be expected, since a radial returnow is required to satisfy continuity, and as can be seen inFigs. 5 and 6, even a short distance away from the impeller(for |z|> 0.2) the radial velocity is negative.

3.2. Decomposition of mean ow

The theoretical model presented in Yoon et al. (2001),considered the impeller-induced phase-averaged mean owas a superposition of a circumferential ow, a circular jet anda pair of tip vortices associated with each impeller blade.Their superposition for the mean ow, in a frame of referencerotating with the blade, can be expressed as

u(r, , z)= uc(r, z)+ ujet(r, z)+ uvort(r, , z), (1)where uc is a purely circumferential ow. Its strongest vari-ation is along the radial direction, and it is only weakly de-pendent on z. In the laboratory frame of reference the time-averaged ow is axisymmetric as the effect of the impellerblades gets averaged out and remains invariant to . The ef-fect of the impeller blades and the associated tip vortices canbe better accounted for in the mean ow in a frame of refer-ence that rotates with the blades. In the rotating frame, theblades remain xed in position and the time-averaged meanow is periodic over the 60 sector between the blades. Thus,the above superposition in the rotating frame of referenceallows for the description of a complex impeller-inducedow, with all three (radial, circumferential and axial) compo-nents of velocity strongly dependent on all three coordinatedirections.

In the case of a Rushton turbine the jet ow slowlychanges direction from a circumferential direction to amore radial direction with increasing radial distance fromthe axis (Kolar et al., 1982, 1984; Kresta and Wood, 1991).Thin shear layer theory can be applied and an approxi-mate self-similar solution can be obtained for the jet (seeYoon et al., 2001). At any radial distance, r, from theaxis of the tank the self-similar jet is directed at an anglejet = cos1(

r2 a2/r) to the radial direction and the jet

velocity along this direction is given by

ujet = Ar1/2(r2 a2)1/4

{1 tanh2

(z

r2 a2)}

. (2)

The jet ow is fully characterized by three parameters: Aameasure of the jet momentum, 1/a measure of the jetthickness and athe virtual origin.The tip vortices are generated from the roll-up of the shear

layers as ow accelerates around the rotating impellers (vander Molen and van Maanen, 1978; Stoots and Calabrese,1995; Yianneskis et al., 1987). The tip vortices are at theirpeak strength close to the impellers and weaken as they ex-tend radially out into the tank. The backbone of the tip vor-tex pair is curved backwards in the circumferential directionopposite to the direction of rotation. The tip vortices remaincoherent, and as shown in the recent large eddy simulations(Yoon et al., 2002), at any instant in time they extend over180 along the circumferential direction. While their headremains anchored behind the rotating blades, their tail oscil-lates in position over time as they extend into the tank. Upontime averaging, a tip vortex pairs extent into the tank issomewhat reduced to about 60 along the circumferential di-rection.Yoon et al. (2001) observed that the strength and sizeof the average tip vortex pair changes along its backbone.The full details of the tip vortex pair along its entire back-

bone cannot be obtained from the present PIV measure-ments. But, velocity elds such as those shown in Figs. 3and 4 can be used to characterize the tip vortex pair as it cutsthrough the cylindrical plane r 1.06. On this plane thelocation of tip vortex pair is characterized by v , the anglebetween the vortex center and the nearest blade and zv ,the axial location of the vortex centers from the midplane.The vortex backbone cuts the cylindrical plane, r 1.06, atan oblique angle and thus the local orientation of the vortexpair is additionally characterized by the angle, v , betweenthe backbone and the local radial direction. To good approx-imation the vortex pair can be taken to be viscous vortices.Therefore the azimuthal velocity (different from circumfer-ential or component of velocity) in the neighborhood ofany of the vortex on a plane normal to the vortex backbonecan be expressed as

uazim = [1 exp(2)], (3)

where is the normal (shortest) distance from the vor-tex backbone to the point at which the velocity is evalu-ated. The other two parameters that characterize the vortex


Table 2The vortex parameters: zv , the axial location of vortex center from themidplane, v , the angle between the vortex center and the nearest blade,v , the angle between the vortex backbone and the local radial directionat the point where it intersect the cylindrical plane

Re =ND2/ zv v () v ()4300 0.08 18 56.36450 0.08 19 56.43

15,785 0.085 19 52.2323,700 0.085 24 55.631,570 0.087 21 57.8747,400 0.087 21 58.6363,180 0.087 21 59.35

4229 0.087 12 48.416421 0.091 12 51.94

15,662 0.092 15 57.123,492 0.093 16 57.1531,323 0.094 16 57.546,985 0.096 15 59.2762,646 0.09 13 58.478,308 0.09 16 59.61

are: , the strength (or circulation) of the vortex and, 1/,

the vortex core diameter. The above ve parameters com-pletely characterize the effect of the tip vortices, at least inthe neighborhood of the impeller.

3.3. Tip vortex scaling

The inuence of circumferential ow and circular jet canbe subtracted from the experimental measurements to iso-late the contribution from the tip vortex pairs. Following aprocedure similar to that inYoon et al. (2001) we extract theparameters associated with the tip vortex pair that yield thebest match to the experimental measurements. Table 2 showsthe location of the tip vortex centers on the cylindrical plane(r 1.06) in terms of their axial position (zv)with respectto the midplane and circumferential (v) position measuredwith respect to the nearest blade in the direction opposite tothe direction of impeller rotation. In nondimensional terms,the axial location of the vortex center about the midplaneremains nearly independent of Reynolds number and showsa slight increase with the size of the tank. The nondimen-sional total height of the blade is 0.4; thus the vortices arelocated roughly halfway between the center plane and thetop or bottom edge of the blade. The circumferential locationof the vortex center also does not show a strong dependenceon Reynolds number, however, with increasing tank size thevortex center is observed to move closer to the blade.Vant Riet and Smith (1975) observed the curvature of the

vortex trajectory and as a result the circumferential locationof the vortex, v , to decrease initially with Reynolds number.However, for Re greater than about 5000 they observed thevortex trajectory to be nearly Reynolds number independentwithin the uncertainties of the experimental measurement.The Reynolds number range for present investigation is in

general larger than 5000. Nevertheless, the slight increasein v with increasing Reynolds number seen in Table 2,opposes the trend observed by Vant Riet and Smith (1975).Also shown in Table 2 is the angle, v , the vortex back-

bone makes with the radial direction at the point where it in-tersects the cylindrical plane (r 1.06). For both the smalland large tanks the angle, v , shows a slight increase withincreasing Reynolds number, indicating a more circumfer-ential orientation for the tip vortex trajectory at higher Re.This result is consistent with the observed slight increasein v with Reynolds number. At lower Reynolds numbersthe tip vortex backbone is directed closer to the radial di-rection in the large tank than in the small tank. However,as Re increases to larger values, for both tanks v appearsto asymptote to an angle of about 60. Within the impellerswept volume (for r < 1) on the leeward side of the rotatingblades the tip vortex pairs lie parallel to the blade and havea near radial orientation. Within a short distance away fromthe blade tip the tip vortices sharply curve backwards. Ascan be seen from Table 2, for all cases considered v is no-ticeably higher than 45, indicating a more circumferentiallocal orientation for the vortex backbone.Fig. 8a shows a plot of the nondimensional vortex

strength as a function of Re for the two different tanksconsidered (nondimensionalized by ND2/2). The nondi-mensional vortex strength decreases with increasing Re,with the rate of decrease higher at lower Re. A tendencytowards Reynolds number independence can be inferred,although even for the highest Reynolds number consideredthis asymptotic state is not yet attained for the large tank.From the gure it appears that for both tanks an asymptoticvalue of about 0.017 may be appropriate for Re greaterthan about 105. The above scaling indicates that the relativeimportance of the vortex pair is higher at lower Reynoldsnumber and levels off at higher Reynolds numbers. The ob-served dependence of nondimensional vortex strength withtank size may be due to the subtle differences in the actualimpellers employed for the small and large tanks. Interest-ingly the measurements of van der Molen and van Maanen(1978) also shows a dependence of tip vortices on tanksize.The measure of vortex size, 1/

, is shown in Fig. 8b. For

the small tank the nondimensional size decreases slightlyfrom a value of about 0.085 at lower Re to an asymptoticvalue of about 0.07 with increasing Re. The relative sizeof the vortex is somewhat small for the large tank. Withincreasing Re the asymptotic value of 1/

for the large

tank is about 0.055. The large tank is twice bigger than thesmall tank, and therefore in dimensional terms the vortexsize for the large tank is only about factor 1.5 bigger thanthat for the small tank.

3.4. Scaling of the circular jet

Fig. 9a shows the nondimensional jet momentum pa-rameter, A (nondimensionalized by ND2/2) for varying


15000 30000 45000 60000 75000Re

0.01

0.02

0.03

0.04

0.05

,Vortex strength

Large TankSmall Tank

15000 30000 45000 60000 75000Re

0.05

0.06

0.07

0.08

0.09

0.1


1/ ,Vortex size

(a)

(b)

Fig. 8. (a) The scaling of nondimensional vortex strength, and (b) thenondimensional vortex size, 1/

, as a function of Reynolds numbers

for the two different tank sizes.

Reynolds number for the two different tanks. For the smalltank an asymptotic constant value of about 0.165 is reachedfor Re greater than about 45,000. For the large tank a com-plete Reynolds number independence is not observed evenfor the largest Reynolds numbers considered. However, atendency towards Reynolds number independence may de-velop at Re exceeding 75,000. The strength of the jet innondimensional terms is observed to be somewhat strongerfor the large tank and that is consistent with the differencesbetween the two tanks seen in Figs. 5 and 6.Fig. 9b shows that the inverse jet thickness, , for the two

different tanks exhibits better collapse than the jet momen-tum. The nondimensional jet thickness decreases with in-creasing Reynolds number at lower Reynolds numbers, andabove a Reynolds number of about 30,000, the jet thick-ness appears to level off and become nearly independentof Re. The asymptotic value of nondimensional inverse jet

15000 30000 45000 60000 75000Re

0.16

0.18

0.2

0.22

0.24

0.26

0.28

0.3

0.32

A, Je

t mom

entum


15000 30000 45000 60000 75000Re

4

4.5

5

5.5

6

6.5

, Inverse jet th

ickness


15000 30000 45000 60000 75000Re

0.3

0.4

0.5

0.6

0.7

a, Virtual o

rigin of je

t


(a)

(b)

(c)

Fig. 9. (a) The scaling of the nondimensional jet momentum, A; (b) theinverse jet thickness, ; and (c) the virtual origin, a, as a function ofReynolds number for the two different tanks.


thickness for the two tanks is, asym 6. Note that the im-peller radius is the length scale used in the nondimension-alization and hence, in dimensional terms, the jet width isabout one-sixth (0.166) of the impeller radius. This can becompared with the height of the impeller blade, which isapproximately 0.4 times the impeller radius.Fig. 9c shows the variation of virtual origin, a, with re-

spect to Re for the two different tanks. A similar trend isexhibited for both tanks with the virtual origin progressivelyincreasing with increasing Reynolds number. The rate of in-crease is, however, somewhat larger for the large tank. Forboth tanks a clear asymptotic behavior is not reached evenfor Reynolds number as large as 75,000. The approach ofthe virtual origin closer to blade tip (r = 1) suggests that asReynolds number increases the importance of circumferen-tial component of jet velocity increases at the expense of theradial component, which is again consistent with the veloc-ity proles shown in Figs. 5 and 6. However, as pointed outearlier, at the radial location of velocity measurement theradial component of the jet is somewhat stronger than thecircumferential component over the entire Reynolds numberrange considered.

3.5. Scaling of the circumferential ow

At the plane of interpolated experimental mean ow data(at r 1.06) the jet is neither completely radial nor purelycircumferential. The jet ow has a circumferential compo-nent, which for increasing Reynolds number can be inferredto be slightly larger for the large tank than for the small tank(owing to the somewhat larger value of a for the large tank).This contribution alone does not fully account for the mea-sured -velocity and the circumferential impeller-inducedow, uc, is dened to account for the balance (note that ucis directed purely in the circumferential direction). Figs. 10aand b show the circumferential ow, uc, as a function of axiallocation at r 1.06 for the varying Reynolds numbers forthe two different tanks. Only the region close to the impeller(z< 0.5), where the measurement is made, is shown. Forcomparison the blade extends from 0.2


Mean velocities

z

-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 0 0.1 0.20

0.1

0.2

0.3

0.4

0.5

0.6



(a)

(b)

Fig. 12. (a) The square-root of surface-averaged mean square vorticity,r , plotted as a function of Re for the two different tanks, (b) the

corresponding trend of the surface maximum radial vorticity.

ation of all components of the mean velocity gradient andthus vorticity and dissipation elds associated with the meanow. Here we will investigate the scaling of both vorticityand dissipation for the two tanks. The phase-averaged meanradial vorticity can be dened in terms of the mean ow as

r = 1r

uz

uz

. (4)

The square-root of surface-averaged mean square vorticity,r , as a function of Re for the two different tanks is

shown in Fig. 12a and the corresponding trend of the surfacemaximum radial vorticity is shown in Fig. 12b. Comparedto the mean ow the approach to Reynolds number inde-pendence is delayed for the case of mean vorticity, whichis dictated more by the smaller scales of motion. For thesmall tank an approximate Reynolds number independent


123455 61

(degree)

z

0 10 20 30 40 50 600

0.1

0.2

0.3

0.4

0.5

0.6

9 100.087 60.065432 10.0

50.0

80.0

40.030.020.0

1 5.0

9 100.087 60.065432 10.0

50.0

80.0

40.030.020.0

1 5.0

9 100.087 60.065432 10.0

50.0

80.0

40.030.020.0

1 5.0

9 100.087 60.065432 10.0

50.0

80.0

40.030.020.0

1 5.0

12341

(degree)

z

0 10 20 30 40 50 600

0.1

0.2

0.3

0.4

0.5

0.6

123456789

(degree)

z

0 10 20 30 40 50 600

0.1

0.2

0.3

0.4

0.5

123456

(degree)

z

0 10 20 30 40 50 600

0.1

0.2

0.3

0.4

0.5

(a)

(b)

(c)

(d)

Fig. 13. Contours of dissipation associated with the mean velocity plottedon the .z plane at R 1.06 for the small tank at (a) Re = 4300, (b)Re = 63,180 and for the large tank at (c) Re=4229 and (d) Re = 78,308.

behavior can be observed above Re= 45,000 and for thelarge tank even a higher Reynolds number needs to bereached before attaining full Reynolds number indepen-dence. The difference between the two tanks decreases withincreasing Reynolds number, however, a complete collapseis not observed.Nondimensional dissipation associated with the phase-

averaged mean ow can be dened as

m = 2 uixjuixj

, (5)

where all the nine components of the gradient of the meanvelocity eld can be obtained from the measurements on thethree vertical plane. The resulting dissipation can then be

15000 30000 45000 60000 75000Re

0

1

2

3

4

5

6

7

8

9

10

/


15000 30000 45000 60000 75000Re

0

20

40

60

80

100

120

Maxim

um

/


(a)

(b)

Fig. 14. (a) The surface-averaged dissipation of mean ow plotted as afunction of Re for the two different tanks, (b) the corresponding trend ofthe surface maximum.

interpolated onto the cylindrical surface at r 1.06. A con-tour plot of m for the small tank is shown in Figs. 13a andb for the two Reynolds numbers Re=4330 and 63,180. Thedistribution is similar to that of vorticity and the primarycontribution to dissipation from the mean ow appears to beassociated with the tip vortex pair. Again the distribution isqualitatively similar for both the Reynolds numbers, how-ever, the magnitude of dissipation decreases with increasingReynolds number. The corresponding dissipation elds forthe large tank for Re=4229 and 78,308 are shown in frames(c) and (d). As with the vorticity eld, the dissipation of themean ow is higher for the large tank.It must be cautioned that, owing to the quadratic depen-

dence of dissipation on velocity, m does not correspond tothe mean dissipation one would obtain from an ensemble av-erage of dissipation computed at individual instances. In thepresent set of experiments, since the individual realizations


on the different vertical planes are temporally uncorrelated,it is not possible to compute the instantaneous dissipationaccurately. m simply represents dissipation associated withthe phase-averaged mean ow and its magnitude can be ex-pected to be much smaller than mean dissipation, since theuctuating small scales play the dominant role in dissipativeprocess. Also there is evidence to show that dissipation in-creases within the impeller-induced ow from r=1 to aboutr=1.4 (Escudie and Line, 2003). The surface-averaged dis-sipation as a function of Re for the two different tanks isshown in Fig. 14a and the corresponding trend of the sur-face maximum m is shown in Fig. 14b. The approach toReynolds number independence is similar to that for meanvorticity observed in Fig. 12.

4. Conclusions

Experimental measurements of ow induced by a Rush-ton turbine in an unbafed stirred tank have been performedover a wide range of operating speeds. Two different tanksizes were used with water as the working uid to covera Reynolds number range of 400080,000. Phase-lockedstereoscopic PIV measurements were made on three differ-ent vertical planes near the impeller to obtain all three com-ponents of the impeller-induced ow. Instantaneous realiza-tions were averaged to obtain the phase-averaged velocity ona 60 sector of a cylindrical plane of nondimensional radiallocation, r 1.06, just beyond the blade tip radius. Data onthis cylindrical surface dictates the ow in the interior of thetank. Hence, the Reynolds number scaling of this velocitydata can serve as a proxy for the Reynolds number scalingof the entire ow. The phase-averaged velocity is dependenton both and z, from which - and surface-averaged meanvelocities are deduced.All three components of the mean velocity are observed

to scale with the blade tip velocity beyond a Reynolds num-ber of about 15,000, in agreement with the ndings of pre-vious researchers. The phase-averaged mean ow was de-composed into circumferential, jet and tip vortex elementsand the parameters associated with the tip vortex pairs andthe circular jet were extracted from the experimental mea-surements. It is observed that the nondimensional jet andtip vortex strength appears to become Reynolds number in-dependent only at large Re. The thickness of the jet ap-pears to scale with the dimensions of the impeller above amodest Reynolds number of about 20,000; however the di-ameter of the tip vortex does not appear to scale similarly.The virtual origin of the circular jet, however, shows onlya slow tendency towards Reynolds number independence.In the case of tip vortex pairs, the location of their cen-ters on the plane of mean ow measurement (r 1.06)remains independent of Re and the tip vortex orientation(with respect to the radial direction), quickly approachesan asymptotic value of about 60 with increasing Reynoldsnumber.

The distribution of dissipation illustrates the importanceof the tip vortex contribution to the mean ow. The spa-tial compactness of the tip vortex pairs contributes to highergradients and thus to gradient dependent quantities such asdissipation. Reynolds number independence for the nondi-mensional dissipation is observed above a Reynolds numberof about 45,000 and the difference between the small andlarge tanks persist even at the higher Reynolds numbers.The most signicant nding of the study is the extreme

sensitivity of the ow within the tank to small geometric de-tails of the impeller. The small and large tanks used in thisstudy were geometrically similar. The Rushton turbine wasalso geometrically scaled from the small to the large tank;however, small differences exist between the two. The bladeand disk thickness were not perfectly scaled and similarlysmall differences existed in blade length. If the two tankswere geometrically similar in every respect then Re is theonly controlling parameter (since free surface was avoidedwith a lid, Froude number is not relevant in the present case).The persistent difference in the mean velocity that exists be-tween the small and the large tanks even at asymptoticallylarge Reynolds numbers (see Fig. 7), can only be attributedto the small deviations in the scaling of the impeller. Theseresults suggest a sensitive dependence of even the mean owto the geometric details of the impeller. The main conclu-sion to be drawn is that scale-up will not work well unlessall aspects of the turbine blade are scaled perfectly. In mostindustrial scale-up from small laboratory to large productionsized tanks it is conceivable that a perfect geometric scalingis often not realized. Based on present results, this will cer-tainly be a contributing factor to lack of a consistent scalingbehavior.

Acknowledgements

This work was supported by a grant from National Sci-ence Foundation (CTS-9910543) and a gift from the DowChemical Company.

References

Costes, J., Couderc, J.P., 1988. Study by laser Doppler anemometry ofthe turbulent ow induced by a Rushton turbine in a stirred tank:inuence of the size of the units I: mean ow and turbulence. ChemicalEngineering Science 43, 2765.

Derksen, J.J., Doelmann, M.S., Van Den Akker, H.E.A., 1999. Three-dimensional LDA measurements in the impeller region of a turbulentstirred tank. Experiments in Fluids 27, 522.

Desouza, A., Pike, R.W., 1972. Fluid dynamics and ow patterns instirred tanks with a turbine impeller. Canadian Journal of ChemicalEngineering 50, 15.

Dong, L., Johansen, S.T., Engh, T.A., 1994. Flow induced by an impellerin an unbafed tankI. Experimental. Chemical Engineering Science49, 549.

Ducoste, J.J., Clark, M.M., Weetman, R.J., 1997. Turbulence inocculators: effect of tank size and impeller type. A.I.Ch.E. Journal43, 328.


Escudie, R., Line, A., 2003. Experimental analysis of hydrodynamics ina radially agitated tank. A.I.Ch.E. Journal 49, 585.

Escudie, R., Bouyer, D., Line, A., 2004. Characterization of trailingvortices generated by a Rushton turbine. A.I.Ch.E. Journal 50, 75.

Hill, D.F., Sharp, K.V., Adrian, R.J., 2000. Stereoscopic particle imagevelocimetry measurements of the ow around a rushton turbine.Experiments in Fluids 29, 478.

Kemoun, A., Lusseyran, F., Mallet, J., Mahousat, M., 1998. Experimentalscanning for simplifying the model of a stirred-tank ow. Experimentsin Fluids 25, 23.

Kolar, V., Filip, P., Curev, A.G., 1982. The swirling radial jet. AppliedScientic Research 39, 329335.

Kolar, V., Filip, P., Curev, A.G., 1984. Hydrodynamics of a radiallydischarging impeller stream in agitated vessels. Chemical EngineeringCommunications 27, 313.

Kresta, S.M., Wood, P.E., 1991. Prediction of the three-dimensionalturbulent ow in stirred tanks. A.I.Ch.E. Journal 37, 448460.

Lamberto, D.J., Alvarez, M.M., Muzzio, F.J., 1999. Experimental andcomputational investigation of the laminar ow structure in a stirredtank. Chemical Engineering Science 54, 919.

Mahouast, M., Cognet, G., David, R., 1989. Two component LDVmeasurements in a stirred tank. A.I.Ch.E. Journal 35, 1770.

Montante, G., Lee, K.C., Brucato, A., Yianneskis, M., 1999. Anexperimental study of double-to-single-loop transition in stirred vessels.Canadian Journal of Chemical Engineering 77, 649.

Schaffer, M., Hofken, M., Durst, F., 1997. Detailed LDV measurementsfor the visualization of the ow eld within a stirred tank reactorequipped with a Rushton turbine. Transactions I. Chemical Engineering75 (part A), 729.

Stoots, C.M., Calabrese, R.V., 1995. Mean velocity eld relative to aRushton turbine blade. A.I.Ch.E. Journal 41, 1.

Sturesson, C., Theliander, H., Rasmuson, A., 1995. An experimental (LDA)and numerical study of the turbulent ow behavior in the near walland bottom regions in an axially stirred vessel. A.I.Ch.E. SymposiumSeries 91, 102.

van der Molen, K., van Maanen, H.R.E., 1978. Laser-Dopplermeasurements of the turbulent ow in stirred vessels to establish scalingrules. Chemical Engineering Science 33, 1161.

Vant Riet, K., Smith, J.M., 1975. The trailing vortex system produced byRushton turbine agitators. Chemical Engineering Science 30, 1093.

Yianneskis, M., Popiolek, Z., Whitelaw, J.H., 1987. An experimental studyof the steady and unsteady ow characteristics of stirred reactors.Journal of Fluid Mechanics 175, 537.

Yoon, H.S., Sharp, K.V., Hill, D.F., Adrian, R.J., Balachandar, S., Ha, M.Y.,Kar, K., 2001. Integrated experimental and computational approach tosimulation of ow in a stirred tank. Chemical Engineering Science 56,6635.

Yoon, H.S., Balachandar, S., Ha, M.Y., Kar, K., 2002. Large eddysimulation of ow in a stirred tank. Journal of Fluids Engineering 125,486.

cespart1_05.pdf

Documents

Transcript of cespart1_05.pdf