Investigation of Flooding Re-Entrainment and Grade Efficiency in Axial Flo

INVESTIGATION OF FLOODING, RE-ENTRAINMENT ANDGRADE EFFICIENCY IN AXIAL FLOW CYCLONES

S. Y. NG, G. H. PRIESTMAN and R. W. K. ALLENDepartment of Chemical and Process Engineering, University of Sheffield, Sheffield, UK

Slotted wall axial flow cyclones with a conventional center-body swirler are found toexhibit significant flooding at low air velocities. This was observed to be due to sig-nificant liquid hold-up on the vanes which eliminated the swirl and led to a pulsating

frothing behaviour. At higher velocities, the liquid dispersed and separated through the wallslots. The onset of entrainment of liquid was determined visually. In the low velocity frothingregion liquid was entrained directly but coalesced and fell back down the vortex finder. Athigher gas velocities entrainment occurred due to several mechanisms.Design changes were introduced to improve performance. The center body was replaced by

a new design of vane swirler around the periphery of the tube inlet. This significantlyincreased the swirl in the tube. Adapting four of the vanes to operate as liquid drainage chan-nels eliminated the flooding phenomenon observed at low gas velocity and enabled the tubeslot area to be reduced. This, combined with a larger diameter vortex finder, significantlyincreased the fraction of flow through the vortex finder, with no extra overall pressuredrop. Liquid entrainment in the new design was reduced by positioning liquid slots at thetop of the tube and by the use of a skirt.Grade efficiency was determined for the improved cyclone design operating over a range of

gas flowrates. The cut size for outflow to the vortex finder was found to vary from about 7 mmat an air velocity of 4.5 ms21 to about 4.5 mm at 11 ms21.

Keywords: axial flow cyclone; flooding; re-entrainment; pressure drop; grade efficiency.

INTRODUCTION

The removal of fine liquid droplets from a gas stream, ordemisting, is an important stage in oilgas separation.Ideally, demisting equipment should be compact, able toexhibit high gas handling capacity with low pressureloss as well as having a low maintenance requirement. Itis also beneficial if the equipment can handle a broadrange of liquid to gas ratios and have a tolerance tochanges in inlet droplet size distribution and flow rate.Such characteristics suggest the application of an axialflow design of cyclone. A recent development in axialflow cyclones has been the incorporation of drainageslots into the barrel walls to allow immediate drainageof collected films and the avoidance of re-entrainment ofthe collected liquids. This type of cyclone is of consider-able interest to the oil and gas industry and is the subjectof this work.Cyclones have been a popular and valuable tool in the

separation industries for more than a century becausethere are no moving parts that can wear out or break.

Consisting of a cylindrical body, an inlet and exit port, acyclone separator is fairly simple, inexpensive both tobuild and to maintain as well as being safe in operation.Most research and industrial applications to date havefocused on reverse flow cyclones (Shepherd and Lapple,1939) for solidgas separation (Alexander, 1949; Ioziaand Leith, 1989; Stairmand, 1949; Hoffmann and Stein,2002). There has been very little work carried out usingcyclones, of whatever geometry, for gasliquid separationand the application of axial flow cyclones for either liquidor solid separation is not well understood. Axial flowcyclones do, however, offer the prospect of small size perunit gas throughput. This paper presents an experimentalstudy of the key potential operational limitations in axialflow cyclones, flooding and re-entrainment. Also importantis the cyclone pressure drop and grade efficiency, for whichdata is also presented. The significance of the data is per-haps best considered in the context of a typical cycloneapplication, as detailed below.

TYPICAL CYCLONE APPLICATION

Knitted meshes are currently the primary gas demistingmethod employed in oilgas production facilities,

Correspondence to: Dr S. Y. Ng, Department of Chemical and ProcessEngineering, University of Sheffield, Mappin Street, S1 3JD, Sheffield, UK.E-mail: [email protected]

884

02638762/06/$30.00+0.00# 2006 Institution of Chemical Engineers

www.icheme.org/cherd Trans IChemE, Part A, October 2006doi: 10.1205/cherd05063 Chemical Engineering Research and Design, 84(A10): 884894

Downloaded from http://www.elearnica.ir

however, their use is limited to low superficial gas vel-ocities and they have a tendency to flood under high load-ings. To counteract these deficiencies, cyclones can beincluded in a hybrid system which should increase the turn-down ratio whilst retaining the advantages of being cheap,reliable, easy to maintain and safe. As shown in Figure 1, aset of axial-flow cyclones is operated in parallel and incombination with primary and secondary meshes inside alarge vessel. A primary layer of mesh is provided immedi-ately following the gas inlet, which is intended to performthe required separation under conditions of low liquid orgas loading. At higher loading, this mesh can become inef-ficient or flooded and hence further downstream separationis provided by a combination of the cyclones and a second-ary mesh layer. The cyclones act to extend the range of thesecondary mesh. The clean gas flowing through the vortexfinders by-passes the secondary mesh layer, reducing thevelocity through it, and hence reducing its propensity toflood. Thus the flow split between vortex finder and meshand their relative flow areas are important design par-ameters. The work reported here focuses on a single tube,representative of one unit in the bank of cyclones, with atarget of about 70% of the supplied flow exiting throughthe vortex finder.

INITIAL CYCLONE DESIGN WITHCENTRE-BODY SWIRLER

The initial cyclone design used is detailed in Figure 2. Itwas adopted from the designs of Swanborn (1997) andCuypers and Stanbridge (2000) which also had slots as aliquid drainage route, but in addition had a recirculatingpath in a hollow swirling centre body so that a smallamount of the secondary gas flow was entrained backinto the main flow to increase the efficiency. There arealso some uniflow cyclone designs that use a number ofvanes shaped helicoidally around a body without the recir-culation zone (e.g., Oranje, 1990; Karlsson et al., 1993).The initial design of axial cyclone tested here was con-

structed from clear PVC with an inner diameter of52 mm and a total height of 842 mm. The cyclone itselfcould be considered in five zones, as indicated inFigure 2. Zone 1 was the inlet region (175 mm from tube

inlet to the vanes) where the air and water droplets wereintroduced. Zone 2 was the brass centre body with the swir-ling vanes. Six-bladed guide vanes at 308 between bladeplane and the axial axis were attached to a centre body.The 30-mm diameter centre body had an aerodynamichub at the front and a rounded end to aid swirl productionand minimise pressure drop. Immediately downstream ofthe blades, the liquid droplets were thrown to the cyclonewall by centrifugal force. This was the collection stage,Zone 3. Zone 4 was the stripping section where slotswere provided to allow the collected liquid to drain fromthe main cyclone tube. The slots were 5 mm wide and111 mm long and started 53 mm downstream of thevanes. Having passed through the slots, the liquid enteredZone 5, a disengagement space. In practice, in a multiplebank of cyclones, there would be no outer casing and dis-engagement would occur in gaps between the cyclones,above the vessels plenum plate to which the cyclonesinlets would be attached. The air and liquid streams inZone 5 are referred to as secondary flows. The secondarygas flows through layers of the second knitted mesh pad,which acts as a secondary collection device. The detailsof the secondary mesh can be found in Table 1. Liquid dro-plets that flow with the air should be trapped here, coalesceand drop back to the bottom of the disengagement space todrain. Therefore, it is a design objective that only a smallfraction, perhaps as low as 30%, of the total gas flowpasses with the liquid to the disengagement space, so asto reduce the load on the secondary mesh and hencereduce its propensity to flood. Ideally, only clean gas willleave the system through either the top of the secondarymesh or through the cyclone exit pipe or vortex finder.In the initial devise the vortex finder had an internaldiameter of 26 mm.

Figure 2. Initial tested geometry.

Figure 1. The overall demisting vessel.

Trans IChemE, Part A, Chemical Engineering Research and Design, 2006, 84(A10): 884894

FLOODING, RE-ENTRAINMENT AND GRADE EFFICIENCY IN AXIAL FLOW CYCLONES 885

FLOODING AND RE-ENTRAINMENTEXPERIMENTS

The cyclone was tested using an airwater system.Figure 3 illustrates the overall rig assembly for the floodingand re-entrainment tests. The primary knitted mesh was notincluded because of the uncertainty in then determining thecyclone liquid loading.Air was supplied by a centrifugal blower via calibrated

rotameters providing inlet velocities within the 52 mmcyclone tube of up to 20 m s21. The air flowrate was con-trolled using gate valves. The pipe connecting the inletair rotameters to the cyclone inlet was a 76 mm transparent,flexible ducting pipe. This type of pipe was also used toconnect the gas exit pipe or vortex finder to the air rota-meters to minimize the dynamic pressure at the outletand to enable traces of water in the gas outlet to bedetected. Water was supplied beneath the separator throughspray nozzles giving a solid spray cone with a typical dropsize between 1.5 and 2.1 mm. Different types of nozzlewere used in order to provide the flowrate required fordifferent liquid loading tests. The water flowrate was set,via the pressure regulator, to give the liquid loadingrequired. Liquid loading is defined as the ratio of waterto air in volume percent (L% v/v). An onoff valve wasplaced upstream of the pressure regulator. Most of the fit-tings were made of PVC Durapipe. The vortex finder wasconnected to a rotameter via a gate valve (VC). The gasthat came out through the liquid escape slots flowed outof the housing pipe via a 908 fitting and was also controlledby a gate valve (VA).

Flooding tests were carried out at various liquid loadingsand the pressure dropflowrate characteristics were deter-mined using inclined and vertical manometers and the cali-brated rotameters. The pressure drops were taken betweentapping points PTS to PTA and PTS to PTC. The static press-ures were also measured at PTA and PTC. In practice both thevortex finder flow and the secondary flow would be exitingto a common chamber and hence a common outlet pressure.During these experiments, this condition of common outletpressure was normally imposed on the operation by adjust-ment of the outlet gate valves. The onset of entrainmentand re-entrainment was detected by observation of the pre-sence of water drops in the vortex finder.The pressure drop across a cyclone is an important per-

formance parameter, however, exact prediction is still notpossible. Additive losses in the cyclone can be consideredto be across the inlet region, the vanes, the separationspace and the gas exit. Most of the pressure drop modelshave been developed for predicting the pressure drop inreverse flow cyclones. The model with the closest resem-blance to the axial flow cyclone is that developed byRamachandran et al. (1994) for rotary flow cyclones. Theissue of the pressure distribution in cyclones is complicatedby the presence of swirl. Under normal situations, a lowerpressure drop means a lower energy requirement and is thepreferable operating regime. However, in the oil industry,there is a large amount of pressure from the wellhead todrive the fluids. The reason for keeping the pressure dropas low as possible is to avoid potential problems of floodingin the demisting vessel shown in Figure 1. If the pressuredrop is high, then a high vessel is required to prevent theliquid in the vessel collection sump from backing up theinternal drainage down-comer pipe and flooding the sec-ondary liquid region. Besides the influence of the cyclonegeometry, the pressure drop also depends on the physicalproperties of the fluids. The liquid density and viscositydoes not change significantly, but the gas density and vis-cosity changes as a function of temperature and pressure.However, in this work only the effect of the cyclone geo-metry on the system pressure drop was considered.

FLOODING RESULTS WITHCENTRE-BODY SWIRLER

The geometry of the axial flow cyclones studied wentthrough a broad range of changes during the course ofthis work (Ng, 2005), but only selected results are pre-sented here. Tests were performed with the initial designto gain an understanding of the experimental system andalso the importance and influence of the design parameters,such as slot dimensions and vortex finder size, on the flowsplit. The flow split is the ratio of the air leaving throughthe vortex finder to the air supplied and the dry flow splitfor the initial geometry was about 20%.From visual observation, a frothing zone of held-up liquid

was formed in the collection zone when the air flowrate waslow. This was because the strength of the vortex generatedwas not strong enough to throw the liquid to the wall. Theair flowrate was just enough to hold the liquid from fallingbackwards. This phenomenon was similar to weeping in asieve tray. Liquid was only expelled when the liquid levelbuilt up to the beginning of the slot rather than beingthrown to the wall due to centrifugal forces. Therefore,

Table 1. Details of the type of knitted mesh used in the system.

Position of mesh Type number Mesh depth (mm)

Initial layer of mesh 9036 100Secondary mesh 9030 50

9230 509030 509036 50

Figure 3. Overall rig assembly for flooding/re-entrainment test.


886 NG et al.

there was a pressure drop due to the height of the liquid inthe collection zone, which could be estimated from themeasured total pressure drop as follows.The total measured cyclone pressure drop can be regarded

as being the sum of three main components, the energyrequired to accelerate liquid within the system, any staticliquid hold-up and the losses associated with the air goingthrough the vortex finder. In the initial design, which hadthe centre body swirler, visual observation showed that theliquid tended to stick onto the swirler and the cyclonewall. Therefore, energy was required to re-accelerate theliquid within the system to the air velocity. This accelerationpressure loss was estimated using (1/2)rfUin

2 (the dynamicpressure loss), where rf is the fluid mixture density [rf rg(12 L) rWL], L is the liquid loading and Uin is takento be the linear air inlet velocity. Essentially this representsthe dynamic energy of the two phase mixture so taking all ofthis as the transferred energy loss gives an upper estimate asit assumes all the mixture had been initially brought to restby impacting the swirler and wall.Figure 4 shows the pressure drops for various dry tests

with the corresponding flow split stated at the end ofeach curve. The data were obtained by adjusting valveVA with valve VC fully opened, until the required flowsplit was achieved in a dry condition. This test was to simu-late the increase in air flow through the vortex finder withincreasing water loading resulting in the slots becomingincreasingly blocked. This is therefore an estimation ofthe pressure drop for the gas to flow through the vortexfinder when water is present and is referred to as theadjusted dry pressure drop, DPdry

a .Figure 5 plots pressure drops at different liquid loadings,

classified into various ranges. From the graph, it can beseen that even a small introduction of water into thesystem increases the pressure drop significantly. It isuseful to analyse liquid loadings individually. The resultsare similar for all the liquid loadings, so only the loadingsof 0.0600.065 v/v% and 0.1000.105 v/v% are discussed.For each range of liquid loading, the measured totalpressure drop is broken down into its main componentssuch that an estimate of the static liquid hold-up can bemade. Figure 6 shows the pressure drops across thesystem, expressed in mm H2O, at a liquid loading of

0.0600.065%. The solid line is the dry pressure dropwhen the common outlet pressure condition is met,giving 20% flow split. The dashed line represents theadjusted dry pressure drop (DPdry

a ) with the dry flow splitequal to the situation when the liquid loading is 0.0600.065%. This is approximately 21.5% and is only a minorincrease compared to the dry condition as little slot block-age has occured. This adjusted dry pressure drop can beobtained from Figure 4. The circles represent DPVF, thepressure drop associated with the losses through thevortex finder estimated here as DPdry

a , plus the estimatedenergy transfer to the liquid within the device, the dynamicenergy loss. In this case little water is present, so at low gasflows this sum is very similar to DPdry

a , though it increasesslightly at the higher gas flows. The crosses are themeasured total pressure drop of the system. The differencebetween the crosses and the circles gives the pressure dropestimated to be due to the static liquid hold-up in the collec-tion zone. At low air flowrates, it shows that significantpressure drop is caused by the liquid hold-up. However,when the air flowrate increases, the total pressure dropreduces to the dry condition. Note that the total pressuredrop can be lower than expected because the presence ofwater reduces the strength of the vortex giving a lower

Figure 4. Variation of dry tube pressure drop with flow split.

Figure 5. Variation of total pressure drop with inlet flowrate at differentliquid loadings.

Figure 6. Analysis of the pressure drop at liquid loading of 0.0600.065%.



energy loss than for the equivalent dry condition. Thisphenomenon was observed by some researchers with theintroduction of solids and their explanation for the reductionwas because of the increased friction due to the movementof particles, which attached themselves to the outer cyclonewall, lowering the tangential velocity (Gauthier et al.,1990; Yuu et al., 1978). The same effect has also beenobserved when water is added to water flow throughvortex throttles (Goh et al., 1994).Figure 7 gives the corresponding results for the higher

liquid loading in the range of 0.1000.105%, and thesame trends are apparent. At low air flowrate, the pressuredrop due to static liquid hold-up is again the dominatingfactor. The airflow is only enough to prevent the liquidfrom falling backwards. So, water level builds up in the col-lection zone and overflows through the slots, whereas athigh air flowrate liquid is centrifuged to the wall. Hence,the liquid hold-up reduces as the air flowrate increases.The significance of the loss due to the energy transferredto accelerate the water within the system is more prominentas the air flowrate increases. At this liquid loading the totalpressure drop increases as the air flowrate increases beyondthe frothing zone, as opposed to the lower liquid loading(0.0600.065 v/v%) where the total pressure drop revertedto the dry condition at this point. At higher liquid loading,the increased amount of liquid and hence increased bulkdensity increases the total pressure drop. Also, morewater is thrown to the wall and out from the slots, hencereducing the area of the slots and forcing more air toleave through the vortex finder. The flow split at thisliquid loading was around 2324%.In summary, a common trend in the pressure drop

flowrate characteristic was found for all the liquid loadingsexamined, as shown in Figure 8. At low air flowrates,below an inlet velocity of approximately 8 m s21, a rela-tively high liquid hold-up region was observed which rep-resented the main component of the system pressure drop.This is a deficiency of this design of swirler, because theairflow is not high enough to swirl and entrain the liquidbut is sufficient to prevent the liquid from falling backwardsor weeping. In this case, the slots tend to act as a weir, as ina sieve tray column. Liquid builds up in the collection zoneuntil it reaches and flows out through the slots. As the airflowrate increases, liquid starts to pulsate (frothing zone)

and drops are vigorously re-entrained into the vortexfinder. When the inlet velocity rises above about10 m s21, the centrifugal force throws the liquid outthrough the slots. At low air flowrates and low liquid load-ing (0.0600.065%), relatively little energy is transferredto accelerate the liquid. The energy transferred becomesmore significant at higher air and liquid flowrates(0.1000.105% and above). The pressure drop due to theincreased flow into the vortex finder increases with liquidloading, as more slot area is blocked.

RE-ENTRAINMENT WITH THE CENTRE-BODYSWIRLER

Extensive observations were also made of the onset andoccurrence of re-entrainment. Three entrainment mechan-isms could be identified, namely direct entrainment, re-entrainment due to liquid creep and re-entrainment due tostripping of the liquid film on the cyclone wall. These areillustrated on Figure 9, which plots the liquid loading (v/v%) versus linear air inlet velocity. It shows that at aninlet velocity lower than 8 m s21 re-entrainment occursirrespective of the liquid loading. This is due to the frothingphenomenon where liquid drops are pulsated into thevortex finder. Therefore, the main entrainment mechanismat low inlet velocities is dominated by direct entrainment.This is the region represented by the triangles. The

Figure 7. Analysis of the pressure drop at liquid loading of 0.1000.105%.

Figure 8. Typical dry and wet total pressure drop curves.

Figure 9. Onset of re-entrainment for the initial geometry.


888 NG et al.

amount of liquid going into the vortex finder increases asthe loading increases. The droplets entrained by this mech-anism do not, however, travel very far up the vortex finderbecause the occurrence of frothing weakens the gas swirl inthe vortex finder and, from visual observation, the dropletsthat are entrained from the liquid pulsation are relativelylarge, usually about 23 mm. These droplets rotate insidethe vortex finder wall and move up the vortex finder.They agglomerate and grow to about 45 mm in diameterand eventually become a ring of liquid. This ring continuesto move up the vortex finder with more droplets coalescingin the ring. Eventually, after traveling approximately 56 cm up the vortex finder, the liquid falls back down, andthe cycle repeats. This means that although these largerdroplets are found inside the vortex finder, they do not actu-ally leave it for low air flowrates up to a linear velocity of8 m s21, irrespective of the liquid loading used.As velocity increases above 8 m s21 and at the higher

liquid loadings, water droplets start to stick on the outerwall of the vortex finder, coalesce and then drain downthe tube until they reach the lip. There they form a film run-ning around the lip of the vortex finder. As the film buildsup it reaches a point where droplets are sheared off into thegas stream. At this point, the liquid takes one of two differ-ent courses, either it flies outwards towards the wall of thecyclone tube or it creeps into the vortex finder and up theinside wall. At higher velocities, the chances of the liquidbeing re-entrained due to stripping of liquid film on thecyclone wall also increase. Furthermore, some fine dropletsare entrained directly into the vortex finder, but this mech-anism is not dominant at high velocities. Thus, at high vel-ocities, re-entrainment is induced by a combination of thesethree mechanisms. This region is illustrated by the areawith the solid circles in Figure 9. Once on the inner wallof the vortex finder, the liquid, in the form of mobiledrops on the surface, tends to move around the inside ina swirling motion, following the gas. These drops are smal-ler than those entrained in the frothing zone. They areapproximately 12 mm in diameter from visual obser-vation and they reduce in size as the air flowrate increases,down to 0.5 mm at an inlet velocity of 14 m s21. These re-entrained droplets travel all the way up the vortex findertube and the rate increases as the inlet velocity increases.The open circles on Figure 9 represent the region whereno re-entrainment occurs. At low liquid loadings and highinlet velocities, the drops are thrown to the wall and exitthrough the slots due to the high centrifugal force.

FLOODING AND RE-ENTRAINMENT WITH THEINLET SWIRL VANE DESIGN

In order to achieve better flow split and overcome thefrothing problem, the initial cyclone geometry was modi-fied. Attempts to avoid frothing required modificationswithin the device such as slot area, vortex tube diameterand the type of swirling device. The major change wasreplacing the centre body with a set of tangentially orientedswirling vanes. Subsequently, to encourage drainage, fourof the vane openings were used as additional drainagechannels in an attempt to eliminate the frothing regionand so avoid re-entrainment at low gas throughputs.Changes were also made to the liquid drainage slots anda larger, 34.5 mm vortex finder was introduced to achieve

a higher flow split. Liquid that creeps across the roof ofthe cyclone tube and down the vortex finder has a hightendency to leave with the clean gas. Therefore, a skirtwas placed around the vortex finder to reduce this liquidre-entrainment. Table 2 shows the evolutionary designchanges that took place.The new inlet was a 40-blade tangential vane design as

illustrated in Table 2 and detailed in Figure 10. It wasthen tested for flooding and re-entrainment using the orig-inal cyclone tube which contained 5 mm 111 mm slotsand the 26 mm i.d. vortex finder (geometry 2). The dryflow split for this design was around 14.3%, whichshowed that the new inlet design created stronger swirlthrowing most of the air out through the slots. This is sup-ported by the pressure dropflowrate characteristics shownin Figure 11. Since everything remained the same exceptfor the vortex inducer, the pressure rise indicated increasedswirl with the new inlet vanes.When liquid was introduced into the system, the frothing

phenomena still existed at low air flowrates and liquid wasstill re-entrained into the vortex finder due to liquid pulsa-tion. The frothing phenomenon was very serious and thiscaused difficulties in measuring the amount of air leavingthrough the vortex finder because the tube was seriouslyflooded. Most of the liquid was pulsated into the vortexfinder because the slots were now further away from thevanes; 273 mm compared to 53 mm for the initial design.The swirl could decay before it reached the slots, hencethe serious frothing. Figure 11 shows that the total wetpressure drop at low air inlet velocity was significantlyhigher than before, being about four times higher thanwhen the system is dry compared to about two timeshigher for the previous design. Again this is mainly dueto the increased distance from vanes to slots enablingmore liquid to build up. As the inlet velocity increased,the pressure drop reduced below the dry condition becauseof the liquid seriously weakening the swirl. At higher vel-ocity (above 13 m s21), re-entrainment still occurred, but itwas less serious. The liquid was swirling around thecyclone tube vigorously and out through the slots. Dropletsof size around 11.5 mm crept up the vortex finder tube.Overall, the results indicated that although the vortexstrength had been increased, the frothing zone still existedbecause liquid could not be drawn out fast enough and therewas also a need to increase the flow split. This led to thenext design, geometry 3.Geometry 3 was the same as geometry 2, but with four of

the swirl vane slots converted to act as drainage channelsand with narrower and shorter tube slots. Four slots of3 mm 20 mm were in the middle of the cyclone tubeand two slots of the same dimension at the top, as shownin Table 2. Slots were placed at the top end of the cyclonetube to drain the liquid that swirled beyond the first set ofslots. This significant reduction in slot area was in partpossible because of the effectiveness of the newly intro-duced vane drainage channels at the tube inlet. Figure 12shows the dry and wet pressure drop characteristics ofgeometry 3. With the reduced slot area, the dry flow splitobtained was 42.4%. Introducing the rear slots enabledtrapped liquid beyond the other slots to be drained.The new drainage vane slots prevented the flooding atlow velocities. No re-entrainment occurred for airvelocities between 5 and 20 m s21 for liquid loading up



Table 2. Summary of the geometries studied.


890 NG et al.

to 0.100 v/v%. However, the flow split was still consideredquite low, even with the additional liquid in the system,which increased it to approximately 50%.In order to increase the flow split, the area of the drai-

nage slots should be minimized or the area of the vortexfinder enlarged. Since further reducing the slot area couldincrease the possibility of liquid re-entrainment, the latteroption was implemented. The vortex finder internal diam-eter was increased from 26 mm to 34.5 mm. This increasedthe flow split to 56%, a substantial step towards the initialtargeted flow split of about 70%. Figure 13 shows thepressure dropflowrate characteristics of the initial geome-try and the new geometry 4. It can be clearly seen that thewet pressure drop of the geometry 4 is similar to the wetpressure drop of the initial centre-body geometry, but forgeometry 4 the flow split with the same liquid loading inthe system (0.1000.105 v/v%) was approximately 61%compared to only around 23% for the initial geometry.Thus the flow split has been increased by over 2.5 times,without the increasing the pressure drop. The wet pressuredrop is lower than the dry pressure drop because the pre-sence of water reduces the strength of the vortex. As theliquid loading increases, the increased bulk density

increases the total pressure drop, the total pressure dropat 0.200 v/v% being higher than for 0.100 v/v% at thesame inlet flowrate.In a flooding/re-entrainment test, geometry 4 performed

very well in collecting and draining the liquid that enteredthe system. At a liquid loading of 0.1000.105 v/v%, allthe liquid left through the slots until an air flowrate of1700 l min21 was reached. Water was then seen swirlingaround the lip of the vortex finder and drops of less than5 mm diameter were seen swirling up the vortex findervery quickly. However, the number of drops moving upwas insignificant compared to re-entrainment in the pre-vious designs. This was because the droplets were re-entrained at intervals. They did not constantly swirl upthe vortex finder. It was suspected that droplets werestripped off the collected film on the cyclone wall anddeposited onto the wall and lip of the vortex finder ordirectly into the vortex finder, which covered a largerarea in the cyclone tube compared to all the previousdesigns. With this design a higher flow split was obtainedwith a lower pressure drop compared to the previous geo-metry because with a larger vortex finder, the tangentialvelocity inside the cyclone tube would be lower than theprevious designs, producing a lower vortex strength (Hoek-stra et al., 1999). When the liquid built up at the lip until itssurface tension could not hold it together, drops went up the

Figure 13. Comparison of the dry and wet pressure drop characteristics ofthe initial geometry and the geometry 4.

Figure 10. Design details of vane swirler.

Figure 12. Comparison of the dry and wet pressure drop characteristics ofthe initial geometry and geometry 3.

Figure 11. Comparison of the dry and wet pressure drop characteristics ofthe initial geometry and geometry 2.



vortex finder. Therefore, a skirt was put around the vortexfinder (geometry 5) to eliminate re-entrainment with every-thing else remained unchanged as for geometry 4.Figure 14 plots the onset of re-entrainment of geometry 4

(without a skirt) and geometry 5 (with a skirt). It is seenthat below the linear air inlet velocity of 14 m s21, re-entrain-ment does not occur. It is of interest to make sure that nore-entrainment occurs over an inlet velocity range of 520 m s21 below a liquid loading of 0.1000.105 v/v%.With the addition of the anti-creep skirt, the range of there-entrainment region was only slightly reduced from above14 m s21 to approximately 15.5 m s21. Above 15.5 m s21,there was no significant improvement in eliminating re-entrainment. This suggested that most of the liquid wasbeing thrown directly into the vortex finder or onto theouter wall of the vortex finder rather than creeping over theroof of the cyclone tube and into the vortex finder. To deter-mine the overall effectiveness of the cyclone also requiresinformation on its grade efficiency, which is considered next.

GRADE EFFICIENCY TESTS

Grade efficiency measurements required the introductionof a smaller quantity of much finer droplets, of diameter inthe range 0.420 microns. The fine droplets were producedusing a nebuliser (Model 1004 AquaTower2TM by GlenmoreHealth Limited). A water aerosol stabilised with 13% gly-cerol, C3H5(OH)3, was used in order to minimize drop sizechanges due to evaporation. The drops of glycerolwaterapproximate an ideal spherical particle system, enablingmore accurate measurement of droplet size distribution. Anoptical particle analyser (Polytec HC-15), which has alsobeen used by other researchers (Buttner and Ebert, 1989;Mitchell et al., 1989), was used to count and measure thedrops. Measurement in the Polytec HC-15 is performed onindividual drops as they move through a small measuringvolume having an effective diameter of 7 mm, defined opti-cally in the flow field by the lenses of two optical systems setat right angles. One system illuminates the measuringvolume and the other detects the scattered light. The lightscattered by the particles is collected and amplified by thephotomultiplier and transformed into electrical pulses,which are a measure of the individual particle size. Gradeefficiencies can be determined by comparing inlet andoutlet measurements. Prior to use in the rig, the Polytec

was calibrated in a special rig using commercially availablelatex particles supplied by LGC Promochem in the sizerange 15 microns.The experimental arrangement for the grade efficiency

tests was very similar to that used in the flooding tests,but with the nebulizer drop generator and an additional fra-mework built to support the Polytec. The separation effi-ciency tests were all carried out using the most successfulgeometry from the flooding/re-entrainment tests, that iswith the inlet swirl vanes and the larger vortex finder, geo-metry 5. Since only one Polytec was available for both inletand outlets measurements, a platform that could be raisedand lowered was required. To give a uniform drop size dis-tribution, the nebulizer was positioned 88 cm, or abouteight pipe diameters, below the inlet sample probe. Thesampling probe was made from 9.8 mm internal diameterstainless steel, with a large radius 908 bend. The samplerate was controlled by varying sampling suction pressure,and a rotameter was used downstream of the Polytec tomeasure the sample flowrate. Drop size measurementswere made at the inlet, the top of the vortex finder and inthe annulus. With no flow straightener in the vortexfinder the resulting high swirl in the gas exit preventedaccurate measurement of the size distribution. Therefore,a flow straightener was introduced, made of a bundle ofthin wall tubes (plastic drinking straws) each of 6 mmdiameter and 120 mm long. The secondary mesh was notinstalled in the annulus flow for the results given.Air flowrates of 600, 1000 and 1400 l min21 were used.

Each measurement was repeated many times, at differentpoints in the flow cross-section and on different days, toensure consistency of results. The grade efficiency foreach drop size, hd, was calculated from Nod and Nid, thenormalized number of drops of diameter d in each outletand inlet respectively:

hd 1Nod

N id

Three velocities are important in measuring the drop sizedistribution (DSD), the local velocity in the flow field beingsampled, Vl, the sampling velocity, Vs, and the velocitygoing through the measuring volume in the Polytec,referred to as the measured velocity, Vm. Isokineticsampling (Vl Vs) is often recommended to ensure thatthe sample is a true representation of the local flow in thepipe. Initial tests using isokinetic sampling gave a relativelylow drop count at the highest velocity of 6 m s21. To inves-tigate this, the sampling velocity was varied (6, 3, 1.5 and1.0 m s21) but the sampled volume kept constant by vary-ing the sample time. The measured inlet DSDs for oper-ation at 1000 SLM are shown in Figure 15. Each curvewas the arithmetic average of three measurements of fourruns each. Again a difference is apparent at the highestsampling velocity of 6 m s21. The 9.8 mm samplingprobe was larger than the 7 mm measuring volume, sothat Vm was approximately twice Vs. A conical nozzlewhich reduced the probe area to 6 mm was fitted. Thisboth reduced Vm to about 0.75 Vs, but also reduced thearea blockage of the probe. Figure 16 plots measuredDSDs for a flowrate of 600 SLM with and without thenozzle, again with a fixed sample volume. There is seento be little difference between any of the distributions.

Figure 14. Onset of re-entrainment for geometry 4 and geometry 5.


892 NG et al.

The indication is that, provided the measured velocity wasno more than about 6 m s21, the sampling velocity hadlittle effect on the DSD. This suggests that isokineticsampling was not necessary to obtain a representativeDSD. Rhodes and Laussmann (1992) found that in regionswhere dense suspensions are to be sampled the solids fluxwas virtually independent of the sampling velocity pro-vided that the gas velocity in the sampling lines was suffi-cient to prevent blockage. The mean solids flux used intheir work was 30 kg m22 s21 with particle density of2456 kg m23. This is very similar to the results observed

in this study with liquid flux of 40 kg m22 s21 with liquiddensity of about 1034 kg m23.Compared to solids, liquid drops have a tendency to stick

to the wall and coalesce, so a further factor which becomesimportant at higher sampling velocities is potential separ-ation of the liquid droplets by deposition in the probe.This phenomenon was investigated using CFD (Ng, 2005)and found to account for some of the reduced drop countat the higher sampling velocities.It was apparent that to sample at the inlet and the outlets,

over the full flow range, non-isokinetic sampling with thecone reducing the sample probe area was the preferredtechnique to obtain a representative DSD. Tests weredone to determine the grade efficiency of the cyclone oper-ating at the three air flowrates, with at least 24 measure-ments being taken at each location for each flowrate. Theresultant grade efficiency curves are plotted in Figure 17.From Figure 17 the separation efficiency is seen to

improve with increasing gas velocity, as would be expected.Also the separation for outflow through the vortex finder isslightly better than for the flow out of the wall slots into thedisengagement space. It should be noted that outflowthrough both the vortex finder and the slots can be regardedas overflow streams, with the underflow being the coa-lesced liquid which is collected and drained from thebottom of the annulus or directly from the swirl vanes, asindicated in Table 2. The better separation in the vortexflow reflects the higher swirl experienced by this stream.

Figure 15. Comparison of the normalized inlet drop size distribution atvarious sampling velocities for an air flowrate of 1000 l min21.

Figure 16. Comparison of inlet drop size distribution at flowrate of600 l min21 with and without cone attached to the probe at (a) 6 m s21

(b) 3 m s21 measured velocity (the sampled velocity is about 1.4 timeshigher with cone and about half without).

Figure 17. Grade efficiency curves of the (a) vortex finder and (b) disen-gagement space at air flowrates of 600, 1000 and 1400 l min21.



Some further droplet removal from the slot outflow wouldbe expected in the secondary mesh, which was not used inthese tests. Overall, the cut size is seen to vary from about9 mm to 4.5 mm microns, which can be considered as satis-factory for demisting.

CONCLUSIONS

Experiments have been performed to determine the flood-ing and entrainment characteristics of axial flow cyclonesapplied as gasliquid demisting separators. An initialdesign with a conventional centre body type swirler wasbuilt and tested with air and water as operating fluids. Atlow gas velocities, below about 8 m s21, significant floodingwas observed due to liquid being held-up on the vanes. Anestimate of this liquid hold-up effect was obtained fromanalysis of measured pressure drops. This high liquid hold-up effectively eliminated the swirl, preventing normalcyclone operation and leading to a pulsating frothing beha-viour. As gas velocity increased above 8 m s21, the liquidbecame dispersed and was successfully separated throughwall slots due to the centrifugal forces. The onset of entrain-ment of liquid into the cyclone vortex finder was determinedvisually. In the low velocity frothing region, significantliquid was entrained directly but coalesced and fell backdown the vortex finder. At higher gas velocities, entrainmentoccurred due to several mechanisms.Several design changes were introduced in an attempt to

improve performance. The center body swirler was replacedby a set of 40 swirl vanes around the periphery of the tubeinlet. These significantly increased the swirl in the tube.Four of the vanes were adapted to operate as liquid drainchannels. These successfully eliminated the low gas velocityflooding phenomenon and enabled the tube slot area to besignificantly reduced. This, combined with the use of alarger diameter vortex finder, more than doubled the flowsplit of gas exiting through the vortex finder, with no extraoverall pressure drop. Liquid entrainment in the newdesign was reduced in part by the positioning liquid slotsat the top of the tube, and in part by the use of a skirtaround the vortex finder. The onset of entrainment wasthus raised to a gas velocity of about 15.5 m s21, and the sub-sequent rate of entrainment was also significantly reduced.The improved cyclone geometry was tested to determine

its grade efficiency. The rig was modified to supply muchsmaller drops, and the drop size distribution measured atthe inlet and outlets. Significant work was undertaken toexamine the effect of sampling velocity on drop sizemeasurement. The measured velocity was found to bemost critical, and needed to be below a critical value.The measured grade efficiencies for outflow through thevortex finder had a cut size which reduced from about7 microns at a gas velocity of 4.5 m s21 to about 4.5microns at about 11 m s21. For outflow through the slotsthe corresponding cut sizes were only slightly higher at 9microns and 5 microns, respectively.

NOMENCLATURE

h grade efficiencyN normalized number of dropV velocityDP pressure drop

L liquid loadingPT pressure tappingr density

Subscriptsd droplet diameterod droplet of diameter, d, at the outletid droplet of diameter, d, at the inletin inlets samplingl localm measuredf fluidg gasW waterS supplyA annularC centralVF vortex finder

REFERENCES

Alexander, R.M., 1949, Fundamentals of cyclone design and operation,Proceedings of Australian Institute of Mining Metallic Processes, 152:203228.

Buttner, H. and Ebert, F., 1989, Particle collection in small laboratorycyclones as a basis for the design of large scale cyclones, ChemicalEngineering Technology, 12: 295301.

Cuypers, C. and Stanbridge, D.I., 2000, Device for treating a gas/liquidmixture, Patent Cooperation Treaty, The Netherlands, WO 00/25931.

Gauthier, T.A., Briens, C.L., Bergougnou, M.A. and Galtier, P., 1990,Uniflow cyclone efficiency study, Powder Technol, 62(3): 217225.

Goh, Y.R., Priestman, G.H. and Tippetts, J.R., 1994, Some multiphasephenomena in fluidic elements, 4th International Symposium on FluidControl and Measurement, Flucome94.

Hoekstra, A.J., Derksen, J.J. and Van der Akker, H.E.A., 1999, An exper-imental and numerical study of turbulent swirling flow in gas cyclones,Chem Eng Sci, 54(1314): 20552065.

Hoffmann, A.C. and Stein, L.E., 2002, Gas Cyclones and Swirl TubesPrinciples, Design and Operation (Springer-Verlag, Berlin, Germany).

Iozia, D.L. and Leith, D., 1989, Effect of cyclone dimensions on gas flowpattern and collection efficiency, Aerosol Science and Technology,10(3): 491500.

Karlsson, R.H., Tinoco, H. and Henriksson, M., 1993, Separator, PatentCooperation Treaty, Sweden, WO9305339.

Mitchell, J.P., Nichols, A.L. and Van Santen, A., 1989, The characteris-ation of water-droplet aerosols by polytec optical particle analysers,Particle and Particle System Characterisation, 6(3): 119123.

Ng, S.Y., 2005, Gas-liquid separation using axial flow cyclones,PhD thesis, University of Sheffield, UK.

Oranje, L., 1990, Device for separating liquids and/or solids from a gasstream, European Patent Application, The Netherlands, 0 436 973 A2.

Ramachandran, G., Raynor, P.C. and Leith, D., 1994, Collection efficiencyand pressure drop for a rotary-flow cyclone, Filtration and Separation,31: 631636.

Rhodes, M.J. and Laussmann, P., 1992, A simple non-isokinetic samplingprobe for dense suspensions, Powder Technol, 70: 141151.

Shepherd, C.B. and Lapple, C.E., 1939, Flow pattern and pressure drop incyclone dust collectors, Ind Eng Chem, 31(8): 972984.

Stairmand, C.J., 1949, Pressure drop in cyclone separator, Engineering,168: 409412.

Swanborn, R.A., 1997, Method and device for treating of a gas/liquidadmixture, Patent Cooperation Treaty, The Netherlands, WO9749477.

Yuu, S., Jotaki, T., Tomita, Y. and Yoshida, K., 1978, The reduction ofpressure drop due to dust loading in a conventional cyclone, ChemEng Sci, 33: 15731580.

ACKNOWLEDGEMENTS

This work was supported by Engineering and Physical SciencesResearch Council, United Kingdom.

The manuscript was received 6 December 2005 and accepted forpublication after revision 31 May 2006.


894 NG et al.

Documents

Investigation of Flooding Re-Entrainment and Grade Efficiency in Axial Flo