Gas flow in rotary kilns

Particuology 8(2010)613-616Contents lists available at Science DirectPARIICLOLOGYparticuologyELSEVIERjournalhomepagewww.elsevier.com/locate/particGas flow in rotary kilnsPatrick R. Davies, Michael J S Norton, D lan Wilson, John F. Davidson, David M. ScottDepartment of Chemical Engineering and Biotechnology, University of Cambridge, Pembroke Street, Cambridge, CB2 3RA, UKARTICLE INFOA BSTRACTArticle historyThis paper describes an experimental investigation on the flow characteristics within a rotating cylinderReceived 4 May 2010ontaining a rolling bed of sand. The axis of the cylinder was horizontal and there was no axial bulkAccepted 10 July 2010flow of particles. The velocity field of the gas flowing through the cylinder was measured by hot-wireanemometry. The measurements indicate that the velocity field is asymmetric with respect to a diameterKeywordsperpendicular to the granular bed. CFD calculations confirm this finding. The gas velocity profiles areelocity profilecrucial in determining heat transfer from gas to solido 2010 Chinese Society of Particuology and Institute of Process Engineering, Chinese Academy ofSciences. Published by Elsevier B V. All rights reserved1. IntroductionBrimacombe, Watkinson, 1983). In order of increasing speed ofrotation these are: (i)slipping, where the bed is approximately staRotary kilns are widely used in the chemical, mining and tionary in a transverse plane with the kiln slipping underneath it,metallurgical industries for the processing of granular materi(ii) slumping (or avalanching), where the granular material fallsals(boateng, 2008 ). Applications include drying, calcining, iron down the bed surface in discrete avalanches, (iii) rolling, whereore reduction, pyrolysis and titanium dioxide production. Recent the part of the bed closest to the cylinder wall rotates in rigid bodyinterest has focussed on using rotary kilns for pyrolysis of wastes rotation, and there is continuous flow of granular material downarias, Roustan, Ichat, 2005 ) A typical kiln comprises a cylin- the free surface, (iv)cascading, where centrifugal effects cause theder of up to 6 m in diameter with length /diameter ratio between free surface to become curved, (v)cataracting, where the rotational0 and 40, depending on the required residence time and with an speed is high enough that particles are projected into the freeboardclination to the horizontal of a few degrees. Granular material is region, and(vi)centrifuging, where the rotational speed is highfed at the raised end forming a bed of granular material along the enough that the particles form an annulus adjacent to the rapidlyylinder, and removed at the lower end. Rotation of the cylinder rotating cylinder Industrial rotary kilns typically operate in thecauses the axial motion of particles, and mixes the bedrolling regime(iii)Gas flow is ty pically counter-current to the bed movement, andWithin a rotary kiln heat can be transferred by conduction,the gas is heated to supply the requisite energy for processing convection and radiation, from gas to wall, gas to bed and wallthe material. The gas can be heated by way of a direct flame or to bed. The mechanism whereby the gas heats, by direct contact,by passing through an external furnace. Operational temperatures the exposed wall of the kiln, which then rotates to be under theare application specific, but typically range from 300 K for drying bed and thereby heats it, is sometimes referred to as regeneraand calcining to 1200 K for TiO2 production and 1800 K for cement tive heat transfer. The simplest models of rotary kiln operationmanufacture(Green Perry, 2007)treat temperatures within the kiln as depending only on axial posi-Both physical processes and chemical reactions can occur inside tion, so the gas and the bed are well-mixed over a cross-section,a kiln. For example in a lime kiln, granular material proceeds and lumped heat transfer coefficients have been used to describethrough a drying section, then a pre-heating section, and finally the each heat transfer mechanism. The gas-to-bed and gas-to-wall heatcalcining and burning sections where chemical reaction occurstransfer coefficients have often been estimated using duct flowThe transverse motion of the particle bed falls into six distinct correlations such as the Dittus-Boelter correlation, which may beregimes depending on the hold-up and rotation speed(Henein, modified to account for surface roughness, as appropriate. Hoever, there are indications that the heat transfer coefficient from gasto bed may be somewhat higher, perhaps by an order of magnitude.than that given by the Dittus-Boelter correlation Brimacombe andCorresponding author. Tel. +44 1223 334 791: fax: +44 1223 334 796E-mail address: diw1 1@cam ac uk(D L. Wilson).Watkinson (197Current address: Exxon Mobil Research and Engineering Europe Ltd, Fawleycorrelation to中国煤化工 in the Dittus-BoelterRefinery, Southampton, Hants, So45 1TX, UKestimated the gCNMHGorder of magnitude1674-2001/s-see front matter o 2010 Chinese Society of Particuology and Institute of Process Engineering, Chinese Academy of Sciences. Published by Elsevier B V. All rights reserveddoi:10.1016/ 1. partic.2010.07.014614P.R. Davies et al. Particuology 8(2010)613-616Two explanations were proposed:(i)the movement of particles atthe bed surface increases heat transfer, and (ii)the area for gasto-bed heat transfer is underestimated as surface roughness wasignored. Tscheng(1978)and Tscheng and Watkinson(1979)esti-mated heat transfer coefficients from their experimental data andfound the gas-to-bed heat transfer coefficient to be an order ofmagnitude higher than the gas-to-wall heat transfer coefficient. EThe use of duct flow correlations requires the gas flow characteristics within the kiln to be similar to those in a duct. However, the 2wall conditions are not the same, in that in a rotary kiln the kiln wall sis rotating, and the surface of the bed is in motion, either continuous 3 1.5t6or avalanching, and this might have an effect on boundary layersnear the kiln wall and bed surface and thus on heat transfer characteristics. Indeed, modelling of a rotary coke calcining kiln by Bui,Simard, Charette, Kocaefe, and Perron(1995)shows a transversegas flow field and gas temperature fields which are asymmetricvith respect to a line bisecting the granular bed and perpendicular to it. Modelling of flow and heat transfer in a rotary lime kilnby Georgallis, Nowak, Salcudean, and Gartshore(2005), using 3Dmodelling of the gas flow, shows a temperature field displayingx, y(cm)similar asymmetry. The current paper describes an experimentalinvestigation of the flow characteristics within a rotating cylinder Fig. 1. Independently measured x and y profiles of axial air flow velocity in a cylinder was horizontal and there was no axial bulk flow of particles. 1//th power law fit for turbulent flow(solid locus). Solid circles -traverse in horicontaining granular material. For simplicity, the axis of the cylinThe velocity field of the gas flowing through the cylinder was measured by hot-wire anemometry. The measurements indicate thatthe velocity field is asymmetric with respect to a diameter perpen was assumed that there were no significant exit effects, an assump-dicular to the granular bed. CFD calculations are described which tion supported by CFD calculations described belowconfirm this findingThe voltage readings from the Cta were calibrated by measuring centreline velocities in empty cylindrical pipes of diameters0. 105 m and 0. 15 m. The maximum velocity, umax, in a fully developed turbulent flow is 1. 22u mean, where umean is the mean velocityover a croSs-section found from measurement of the volumetio The gas velocity profile was measured in a horizontal cylindrical flow rate using an orifice plate. a range of flow rates and pipe sizesPerspex shell of length 2 m and internal diameter 0.105 m contain- gave a calibration curve for each probe tip, and a fitted polynoing a bed of sand with mean diameter 0.49 mm and angle of repose mial gave gas velocity as a function of voltage. The calibration was32. The cylinder was operated in semi-batch mode with no flow checked by performing traverses across an empty cylindrical pipeof sand in or out. The bed depth was 0.025 m, which was constant and comparing the results to the 1/7th power law relationship.along the cylinder. At each end, the sand was retained by a flatat which showed good agreement(see Fig. 1)plate. The entry plate rotated with the cylinder and contained aMeasurements of air velocities were taken at approximatelycentral hole, 0.04m diameter, for the air. At the discharge end, the 10 Hz, and average readings were calculated from traces lastingsand was retained by a dam, formed by the end plate, which just approximately 30scovered the bed segment at the end of the kiln the underside ofAs the Cta works by maintaining the wire at constant tema high-pile carpet was glued to the end plate, which was a semi- perature, the voltages measured are sensitive to air temperaturecircular Perspex sheet cut to the shape of the cross-section of the (which was always ambient) so errors arising from temperaturesand bed, but a little bigger. This plate was attached to a metal fluctuations during runs were assumed small. However, day to dayclamp unit, positioned so that the carpet pile pressed into the end ambient laboratory temperatures could vary by 2-3.C, anticipatedflange of the kiln, forming a close brush seal retaining the sand in to be a source of mismatch between data recorded on different daysthe cylinder, while enabling access to the inner cross-sections of To account for this effect, overlap points were recorded betweenthe kiln. Detailed descriptions of the apparatus are given by Davies data sets to allow them to be combined(2009)and Norton (2009)a key assumption in measuring velocity profiles is that the flowThe air flow rate was 0.0162 m/s at room temperature corre- is essentially axial, because the probe measures the magnitudeonding to a Reynolds number of 13800, based on the mean air of the velocity and not the direction. To test this assumption ofvelocity in the freeboard and the hydraulic diameter of its cross- essentially axial velocities measurements were taken in a rotatingsection. The inner wall of the cylinder was lined with grade P80 cylinder without through air flow.It was observed that even withinsandpaper(average particle size 0. 197 mm) to prevent slip as thecylinder rotated, ensuring the bed was in the rolling regime. The the anemometer to resolve from zero. It is therefore reasonable tocylinder was supported by 6 pairs of rubber rollers along its length assume that the measured velocities are effectively the magnitudesand rotation was achieved via a belt connecting the kiln to a motorThe typical rotation speed was 7. 2 rpm. Air was delivered to thelet via a flexible hose after passing through an orifice plate which 3. ResultsV凵中国煤化工enabled the measurement of the volumetric flow rate of the airLocal gas velocity measurements were made at a location 0. 1 mCNMHGVelocity profilFig 2 shows thfrom the cylinder exit using a one-dimensional probe, which mea- measurements in a cylinder which was not rotating Measurementssured air-speed, connected to a Dantec miniCTA anemometer It were taken on a 1 x 1 cm grid the numbers in the squares are veloc-P.R. Davies et al/ Particuology 8(2010)613-616615toTIn2TT4T5 6ST aTg TT IT i an indication of the variation of turbulence intensity over a cross-section, indicating higher turbulence intensity near the lower partof the active surface than in its upper part.2.22.12.02.02.24. CFD modelling222.42.52.52.42.323252.72728272.62.1The objective was to create a simple model to simulate the airflow in the cylinder with sufficient accuracy to make212527272829282724parison with the experimental results. Such model calculations areuseful in understanding the nature of gas flow, for example, entry232627282829282.72421and exit lengths, and could be developed to include heat transfer2326272727292726232The package used was ANSYS-CFX. The geometry was constructed and meshed, the appropriate fluid definitions and252.72.72.62.82.62.31boundary conditions applied and then the equations solved. Thenature of CFD modelling required assumptions and simplifications.252.72.6242320as follows:242.5232.1(i)The system is isothermal, taken to be at temperature 293 K192.22ii)The k-E turbulence model is used (Launder Spalding, 1974)iii) No-slip is assumed for the gas at all moving surfaces, i.e. theall and bed surface(iv)To attempt to replicate any exit effects, air leaving the cylinder flows into a large body of air at 293 K.Fig. 2. Axial velocity (in m/s) measured over a cross-section 0. 1 m from the air(v) Surface roughness is estimated based on the particle sizes ofutlet end of the cylinder. Maximum bed depth, measured normal to the bed surfacethe sandpaper and the bed The bed roughness is increased0.025m. Air flow rate 0.0162 m /s No rotationfor modelling in order to try to account for lower density atthe surface caused by surface movement. The roughness usedes in m/s. The profile in Fig. 2, without rotation, appears to bewas 0.2 mm for the wall and 1. 0 mm for the bed surfacepproximately symmetric with respect to a diameter perpendic(vi) The walls were treated as rotatingular to the granular bed surface. Fig 3 shows measurements in a (vii) The air directly adjacent to the bed was defined to have aylinder rotating at 7.2 rpm. In this case, measurements were takertransverse velocity parallel to the bed surface of 0.2 m/s, toon a 0.5 x 0.5 cm grid The data show asymmetry with respect to asimulate the bed motion; the value of 0. 2 m/s was suggesteddiameter perpendicular to the granular bed surface, in contrast toby experiment. Internally the bed was treated as a solid stathe case with no rotation the velocities are lower near the lowerbody, because modelling of the internal circulation of gralpart of the active bed surface than in the upper part of the activear material in the bed is not necessary to investigate the airflow. Motion of air within the interstices was ignoredFig 4 shows the standard deviation of the voltage measurein a cylinder rotating at 7. 2 rpm, using a 1 x 1 cm grid. This gives0123456789101112】重理事事12222222a817068646572798.1665.84.64.84.85.0648.122s242745.64.23.53.43.64.64.66.47.0523.64.55.3505.3545.577986.950383259717481716528272726262.62626252.5圆团Basa24242411.3725.54.35.35.97.68.7922222424242279625.4667.910.1222422a22242c2c222242422ss2d2d2s222中国煤化工utlet end of the cylinder. Maximum bed depth, measured normal to the bed surface, measured normal to the bed surface, 0.025 m Air flow rate 0.0162 m/s rotatio0.025m. Air flow rate 0.0162 m /s Rotation 7.2 rpm7.2 rpmP.R. Davies et al. Particuology 8(2010)613-616(crossspeednot reported. The results presented here indicate that the bed-gasficant on the scale of the apparatpatterns persist at industrial scale requires further work.5. ConclusionsThe gas velocity profile has been measured to be asymmetricith respect to a diameter perpendicular to the granular bed, witha region of slower flows near the foot of the rolling bed This experimental finding has been confirmed by a calculation using CFD. Theasymmetry is contrary to a common assumption in modelling that1.95 of plug flow. The implications for the modelling of heat transferarrant further investigationAcknowledgmentsWe wish to acknowledge Dr Mark Williamson for assistance onthe use of ANSYS-CFX, and Dr David Martin of the Department ofEngineering, University of Cambridge, for the loan of the miniCTA75This work was part of Chemical Engineering Tripos Part IIBFig. 5. Axial velocity contours computed over a cross-section 0. 1 m from the air- Research Projects(Davies, 2009: Norton, 2009)utlet end of the cylinder. Maximum bed depth measured normal to the bed surface.0.025 m. Air flow rate 0.0162 m/s. Rotation 7.2 rpm.References(vii)It was assumed that the mesh was sufficiently fine to provide Boateng. AA(2008). Rotary kilns: Transport phenomena and transport processesuseful results. Although a detailed meshing sensitivity analy- Brimacombe, ]. K, Watkinson, A P.(1978). Heat transfer in a direct-fired rotarysis was not undertaken, consideration was given to checking kiln. L. Pilot plant and experimentation. Metallurgical Transactions B: Processthat there was sufficient mesh resolution for the boundaryMetallurgy, 9, 201-208Bui,R T, Simard, G, Charette, A, Kocaefe, Y, Perron, J(1995). Mathematical mod-The computations show that jet formation at the inlet can be Davies, P. R (2009). Convective heat transfer in rotary kilns. MEng Project Report. UKen,and under the conditions of the experiment, the entry lengthis approximately 0.8 m. This supports the conclusions of Lam(2007) Georgallis, M, Nowak, P, Salcudean, M,& Gartshore, I S(2005). Modelling therotary lime kiln. The Canadiaixit effects at a location 0. 10 m from the cylinder exit. It thus seems Hen New York: McG-raw Hl(2007). Perry's chemical engineers"handbook(8th ed.otion in rdischarge end where the air velocity measurements were takenig. 5 shows computed axial velocity contours over a crossdissertation. UK: Department of Chemical Engineering and Biotsection 0.10 m from the cylinder exit. As with the experimentalversity of Cambridgemeasurements of the velocity profile, flow asymmetry is evident, Launder, B. E.& Spalding. D. B (1974). The numerical computation of turbuvith a region of slower flows near the foot of the rolling bedThe experimental and computational results clearly show that Marias, F, Roustan, H,&Pichatkiln for the pyrolysisle assumption that the air flow pattern in the cylinder is similarg Science,60,4609-4622to that in a duct is not valid Estimates of the gas-wall and gas-bedtransfer in rotary kilns. MEng Project Report.UK: Department of Chemicaleering and Biotechnology, University of Camheat transfer coefficients based on pipe flow correlations are likelyto underpredict these parametersTscheng, S H (1978). Convective heat transfer in a rotary kiln. Unpublished doctoralComplex gas flow patterns were also reported in simulations bydissertation. Vancouver, Canada: University of British Columbia.Yang, Rakhorst, Reuter, and Voncken(1999)and georgallis et al.The Canadian Journal of(2005), but these studies included combustion and involved more Yang, Y, Rakhorst, J,ReutVoncken, J. H L(1999, December ). Analysiscomplicated entry geometries so it is impossible to discern whichator. In 2nd internationalconference on CFD in theand process industries CSIRO, Melbourne, Auseffect was dominant. Comparison with experimental flow data was中国煤化工CNMHG