operating performance of a naturally driven rotational ... · fils these requirements is the...

Operating Performance of a Naturally Driven Rotational ParticleSeparator

By Eva Mondt, Erik van Kemenade*, and Rob Schook

DOI: 10.1002/ceat.200500395

The increasing amount of liquid, especially water, in the product stream of offshore gas wells, requires improvement of cur-rent separation methods. Nowadays, separation methods are mainly based on gravitational settling of the dispersed phases.In these separators low gas velocities are required to achieve a sufficient separating efficiency. As a result these devices arevoluminous, heavy, and expensive. As platforms are restricted to space and weight and the liquid amount is increasing, com-pact and efficient phase separation equipment is required to keep the exploitation of the wells profitable. A device which ful-fils these requirements is the naturally driven Rotational Particle Separator (RPS). In this study the operating characteristicsof such a separator was measured. For this purpose a full-scale prototype was built, which is capable to handle the volumeflow of one typical wellhead under high pressure (80 bar) and which separates droplets down to 2 micron. In order to validatethe operating characteristics of the prototype both hydrodynamic and separation performance measurements were per-formed. Overall, the performance of the prototype agrees well with expectations.

1 Introduction

With longer exploitation of offshore gas wells, the amountof liquid, especially water, in the product stream increases.This is mainly due to the fact that water is usually used tokeep the well under pressure. The increase in liquid contami-nants requires the improvement of current separation meth-ods. In commonly used separators gravity is used to separatethe dispersed phase. To allow for sufficient settling time, gasvelocities in the separators have to be low. As a result thesedevices are voluminous, heavy, and expensive and with in-creasing liquid amount the capacity is no longer sufficient.More and/or heavier separation devices are needed andsometimes even heavier and more expensive supportingstructures are required. In some cases, although considerablegas reserves are still present, the exploitation of a well has tobe stopped, as current gas treatment techniques are not eco-nomically viable. In order to make the exploitation of olderwells profitable, the offshore industry is searching for effi-cient and compact phase separation devices.

A device which fulfills these requirements is the naturallydriven Rotational Particle Separator (RPS) [1, 2]. The corecomponent of the RPS is the rotating filter element (seeFig. 1), which consists of a multitude of axially orientedchannels, which rotate as a whole around a common axis.Particles or droplets flowing in the fluid in a laminar motionare centrifuged to the outer walls of each individual channeland adhere to the collecting walls as a result of the centrifu-

gal force, van der Waals forces and/or forces due to surfacetension, while the purified fluid leaves the channels at theexit. As the radial distance over which the droplets have tomove to arrive at a collecting surface is small (typically1 mm), the RPS is capable of separating particles of smallsizes: e.g., solid and liquid particles entrained in gases withsizes down to 0.5 micron at relatively low angular speedsand small channel lengths [3].

Chem. Eng. Technol. 2006, 29, No. 3 © 2006 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim 375

I - IV

I

II

IIIIV

Ω

r

Figure 1. Left: filter element of the rotational particle separator consisting ofa multitude of axial channels. Right: while the channels rotate around a com-mon axis, particles entrained in the gas flowing through these channels arecentrifuged towards the channel wall.

Gas + liquid Gas

De-swirler &

bearing support

Swirler &

bearing support

Liquid Liquid

Axial cyclone /

pre-separator

Filter element

Post-separator

Figure 2. Schematic representation of the RPS-based separator, which isdesigned to purify natural gas under high pressure.

–

[*] Ir. E. Mondt, Eindhoven University of Technology, Department ofMechanical Engineering, Section Process Technology, PO Box 513, WH2.141, 5600 MB Eindhoven, The Netherlands; Dr. Ir. H. P. van Kemenade([email protected]), Eindhoven University of Technology, Depart-ment of Mechanical Engineering, Section Process Technology, PO Box513, WH 2.141, 5600 MB Eindhoven, The Netherlands; Dr. Ir. R. Schook,CDS Engineering, Business Park IJsseloord 2, Delta 101, 6825 MNArnhem, The Netherlands.

Full Paper

A schematic representation of the naturally driven RPS isgiven in Fig. 2. After passing the inlet part, which mainlyserves as a bearing support, the contaminated fluid (gas/liquid) enters a swirl generator where the fluid is brought ina rotational motion as it passes its static vanes. After theswirl generator, the mixture enters the pre-separation room,which serves as a kind of axial cyclone. In this separationroom large contaminants are swept out of the flow, due tothe centrifugal motion of the fluid. The contaminants leavethe separator through the pre-separator outlets, which aresituated before the filter element at the outside of theseparator. The filter element is brought into rotation by theangular momentum of the fluid. Due to the centrifugal force,the droplets in the fluid are driven towards the wall of thechannels where a liquid film is formed. This film breaks upat the end of the filter element and larger droplets areformed which can easily be separated in the post-separator.If the dispersed phase is the lighter medium, as is the casefor oil droplets dispersed in water, it is swept to the core ofthe post-separator [1]. In the case of liquid droplets in a gasflow, the droplets move towards the outer wall. The liquidseparated from the fluid flow leaves the separator throughthe post-separator outlets, which are positioned at the outerradius of the separator. The purified gas leaves the separatoraxially through the main outlet. Downstream of the post-separator a de-swirler is placed. The stator vanes convert alarge part of the rotational energy of the fluid into staticpressure. Without the de-swirler the rotational energy iscompletely dissipated in heat and thus wasted. Recoveringpart of the kinetic energy means that the separation processgenerates the waste stream at high pressure. This enables re-injection of the contaminants back into the gas reservoirfrom which they originally came.

Table 1. Desired operating characteristics for the RPS-based natural gas-water separator.

Value

dp,100% 2 lm

Do 0.6 m

L 0.36 m

Volume flow gas 0.65 m3s–1 (qg = 50 kg m–3)

Volume fraction contaminants 0.1–1 %

Pressure 80 bar

Temperature 340 K

Pressure drop < 1 bar

In this article the operating performance of an RPS-basednatural gas-water separator is presented. For this purpose aprototype was developed for separating water droplets fromnatural gas. Based on design relations [1] and design criteriabased on offshore process conditions, a full-scale prototypewas built, which is capable to handle the volume flow of onetypical wellhead under high pressure (80 bar) and which se-

parates droplets down to 2 micron (see Tab. 1). In order tovalidate the operating characteristics of the prototype mea-surements were performed.

2 Experimental Setup

The performance of the separator was measured in twotest facilities. In one test facility the separation performanceand the hydrodynamic performance of the prototype at low(atmospheric) pressure was measured. In the second test thefacility hydrodynamic performance of the separator was sim-ulated at elevated pressures.

A schematic overview of the low pressure test facility isshown in Fig. 3. In this test loop air is used as the workingfluid. The pressure and temperature inside the test loop arerespectively about 1.2 bar and 283 K. The measuring equip-ment is listed in Tab. 2. The pressure drop over the filter ele-ment is measured by using the four equally spaced pressureholes situated at the circumference of the housing upstreamas well as downstream of the filter element (see Fig. 4). Thefour holes were interconnected such that, before as well asbehind the element, the average pressure is measured. Be-sides one pressure hole is located upstream of the swirl gen-erator. By using this pressure hole the total pressure dropover the separator can be measured.

The angular speed of the filter element was measured witha Turck inductive sensor, type Bi1-EG05-AP6X. This sensordetects a small slot, constructed in the outer wall of the filterelement. The location of the inductive sensor is denoted inFig. 4.

The high pressure loop is filled with Sulfur Hexafluoride(SF6). SF6 is a high density gas, with a density of 5 to 6 timesthe density of natural gas. This means that at relatively lowpressures, high fluid density flows can be simulated. The

376 http://www.cet-journal.com © 2006 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim Chem. Eng. Technol. 2006, 29, No. 3

Figure 3. Schematic representation of the low pressure test facility.

Full Paper

actual volume flow of SF6 was moni-tored with an Instromet turbine gasflow meter, type Q75. The operatingrange of this flow meter is 0.05–1.8m3s–1 for SF6 at atmospheric condi-tions, with an accuracy of ± 2 %. Thefluid temperature was measured withan Invensys 1/A series TemperatureTransmitter, Model RTT20 (Pt100),with an operating range between 233and 358 K and an accuracy of ± 0.05K. The specifications of the absoluteand differential pressure and induc-tive sensors are given in Tab. 2. Mea-surements were performed at threedifferent absolute mean pressures:1.9 bar, 2.9 bar, and 3.6 bar, corre-sponding to fluid densities of respec-tively: 11.2, 17.5, and 22.2 kg m–3 [4].As only stationary measurementswere attempted, all process condi-tions were directly read from theLCD digital indicators during themeasurements in both loops.

The separation efficiency of theprototype was determined by inject-ing calcium hydroxide (Ca(OH)2),also called chalk hydrate or slakedlime (supplied by Carmeuse) into thelow pressure setup at the injectionpoint (IP in Fig. 3). The particle sizedistributions both up- and down-stream of the separator were deter-mined by sampling the flow isokine-tically with an eight stage Andersen

impactor type Mark II. The weight increase of the differentstages (see Tab. 3) of the impactor was determined with aMettler mechanical balance, type B5. This balance has ameasuring range of 0–0.2 kg, with a standard deviation of± 3 · 10–8 kg and an optical accuracy of ± 5 · 10–8 kg. Sincelime is hygroscopic, all collection media must be precondi-tioned prior to weighing, i.e., both before and after a sam-pling cycle. Preconditioning was done by placing the collec-tion media in an oven at 323 K for a period of 12 hours.

3 Measurement Results

3.1 Hydrodynamic Performance

The angular speed of the filter element can be predictedfrom the conservation of angular momentum, which is gen-erated in the swirl generator. At equilibrium the generatedangular momentum equals the drive momentum and themomentum losses in other parts of the separator. The otherparts that contribute the most to these losses are the pre-se-

Chem. Eng. Technol. 2006, 29, No. 3 © 2006 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim http://www.cet-journal.com 377

Figure 4. Location of the pressure holes and angular speed sensor on the prototype of the natural gas-waterseparator.

Table 2. Operating conditions of the measuring devices in the low pressure test loop.

Flow meter1

Abs. pres. sensor2

Diff. pres. sensor3

Inductive sensor4

Operating range 0.0042–0.55 m3s–1 0–21 bar55

0–0.2 bar 0–100 Hz

Temperature medium 255–473 K 227–394 K 227–394 K not applicable

Ambient temperature 233–358 K 233–358 K 233–358 K 248–343 K

Static pressure rating 100 bar 150 bar 150 bar not applicable

Accuracy Re < 20000:2 %Re > 20000:1 %

< 0.075 % 0.2 % ≤ 2 %

1 Invensys vortex flowmeter, type 83F2 Invensys I/A series absolute pressure sensor, type 1GP103 Invensys I/A series differential pressure sensor, type 1DP104 Turck inductive sensor, type Bi1-EG05-AP6X5 absolute or gauge units

Table 3. Range of the impactor stages for the Andersen Mark II impactor, ex-pressed in aerodynamic diameter, for the applied sampling flow.

Stage Aerodynamic diameter [lm]

pre-separator > 10

0 10–9

1 9–5.8

2 5.8–4.7

3 4.7–3.3

4 3.3–2.1

5 2.1–1.1

6 1.1–0.7

7 0.7–0.4

backup filter < 0.4

Full Paper

parator, the gap between the filter element and the statichousing and the bearings. This results in the following equa-tion1):

Gh sg = Gh fe + Gh gap + Gh pre + Gh bearing (1)

where Gh sg is the angular momentum generated in the swirlgenerator and Gh fe, Gh gap, Gh pre, and Gh bearing are respec-tively the loss of angular momentum in the filter element,the gap between the filter element and the static housing,the pre-separator and the bearings. Expressions for all com-ponents included in Eq. (1) are given in [1], except for theloss of angular momentum in the bearings, which dependson the type of bearing chosen. In the current design SKF ballbearings are used and their loss of angular momentum is cal-culated from the SKF catalogue [5].

The pressure loss over the separator is mainly due to thepressure loss in the filter element and in the swirl generator.In both parts the pressure drop is caused by friction and bychanges in the direction of the fluid velocity. In the filter ele-ment the pressure drop due to friction at the channel walls,assuming circular channels with a diameter dc, is given by[6]:

DPfe fLc

dc n

12 qf u2

f (2)

where f is the friction factor, Lc the channel length, dc thechannel height, qf the fluid density, ufe the axial fluid velocityin the channels of the filter element, and n represents the ad-ditional pressure drop at the entrance region of the ductwhere the transition to a parabolic velocity profile takesplace. The channels of the filter element have a triangular orsinusoidal cross-section for which the channel width is muchlarger than the height. For a laminar flow through such chan-nel geometries Shah and London [7] have derived both f andn, see Tab. 4. In the expressions for f, Re is the Reynoldsnumber based on the maximum channel height and the aver-age fluid velocity in the channel. For turbulent flows throughchannels of non-circular cross sections, it is proved empiri-cally that the Moody diagram [8] can be used to define thefriction coefficient, provided that the hydraulic diameter ofthese channels is substituted for dc. In calculating the pres-sure loss over the filter element due to the friction, ufe istaken equal to the mean axial fluid velocity through thefilter element.

Table 4. Friction factor f and n-value for fully developed laminar flow indifferent channel geometries [7].

triangle sinus

f 48/Re 38.4/Re

n 2.971 2.271

Besides the friction, pressure loss also results from the de-velopment of a free vortex in the pre-separator to a solidbody rotation in the filter element. A relation for this pres-sure loss contribution is given in [1].

The pressure drop over the swirl generator is due to thefriction in the small annulus and the introduction of a tan-gential velocity component. The flow relations of a fluid withviscous effects in a duct of constant cross-sectional area givethe pressure drop due to friction in case a compressible flowis considered [9]. In case of an incompressible flow Eq. (2)gives the pressure drop due to friction.

The pressure drop caused by introducing a swirl compo-nent in the flow is calculated as:

DPsg 12

qf w2sg 1

2qf usg tan a 2 (3)

where wsg and usg are respectively the tangential and axialvelocity component at the exit of the swirl generator, and ais the blade angle of the vanes.

3.1.1 Angular Speed

In Fig. 5 the rotational speed n is depicted as a function ofthe flow rate through the separator for air at 1.2 bar. The sol-id line represents the model, given by Eq. (1), the circles,crosses and dots each represent one measurement series. Itcan be seen that the measurements are well reproducible.Comparing the measurement results with the theoreticalmodel shows good agreement, except for a slight deviationat low flow rates. For flow rates smaller than about 0.06 m3s–1

the filter element does not rotate. This is mainly due to thestatic bearing friction. The model also predicts the onset of


Figure 5. Rotational speed of the filter element as a function of flow rate forair at 1.2 bar.

–

1) List of symbols at the end of the paper.

Full Paper

the rotation well. The small bend in the model at small flowrates is due to the fact that the relation for the dynamic bear-ing friction at low angular speeds differs from the one athigher speeds [5].

In Fig. 6 the rotational speed at elevated fluid densities isdepicted as a function of the flow rate through the separator.The solid line represents the model, (see Eq. (1)), while thecrosses represent the measurement data. As the temperature

and absolute pressure in the test setup varied during onemeasurement cycle, the mean temperature and pressurevalues of one measurement cycle are used to calculate thetheoretical hydrodynamic performance of the separator. Thedensity and viscosity of SF6 are interpolated from the mea-sured data of Wilhelm [4].

For the two lowest fluid densities the models predict aslightly lower rotational speed than measured. Comparingthe three measurement results at elevated pressures showsthat the rotational speed of the filter element hardly de-pends on the fluid density. This is expected as both the gen-erated momentum and the loss of angular momentum in thefilter element, which has the highest contribution to the totaldissipated momentum, are linearly dependent on the gasdensity.

3.1.2 Pressure Drop

In Fig. 7 the pressure drop over the filter element for airas the working fluid is depicted. The crosses and dots eachrepresent one measurement series, while the lines representthe model. Losses are calculated for the two types of channelshapes: triangular and sinusoidal. The values of the frictionfactor, f and n are taken from Tab. 4. It follows that the pres-sure loss over the filter element is best represented by themodel, assuming sinusoidal channels. This is also expected,as the filter element channels of the prototype have a moreor less sinusoidal shape.

The model assumes a fully developed flow in the channelsof the filter element. In reality this is not true, as the dy-namic entrance length, Lhy for the channels is rather large.The dynamic entrance length is defined as the required ductlength to achieve a maximal duct velocity of 99 % of that forfully developed flow when the entering fluid velocity profileis uniform [7]. For example, for sinusoids channels for whichthe width is much larger than the height, Shah and London[7] give a dimensionless entrance length L

hy (=Lhy/(Dh Re))of 0.0648. The hydraulic diameter of the channels is 1 · 10–3 m


Figure 6. Rotational speed of the filter element as a function of the SF6 flowrate. From left to right the density of SF6 is respectively 11.2, 17.5, and22.2 kg m–3.

Figure 7. Pressure drop over the filter element as a function of flow rate forair at 1.2 bar.

Full Paper

and the Reynolds number in the channels during the mea-surements is of the order 1 · 103. This means that the dynamicentrance length is about 0.0648 m, which is about one thirdof the total channel length (Lc = 0.18 m). Although themodel neglects the entrance length, there is a good agree-ment between the measurements and the model.

In Fig. 8 the total pressure loss over the separator is giventogether with the pressure loss over the filter element. Itfollows that the total pressure loss is well predicted by thetheoretical model. The pressure loss in the pre- and post-separator and the pressure drop in the pipe before the swirlgenerator are not included in the model, as they are negligi-ble compared to the other losses. It can also be seen that thepressure loss over the filter element is the main contributorto the total pressure loss over the separator. In the secondplace follows the pressure loss due to the generated swirl.

In Fig. 9 the pressure loss over the filter element and thetotal pressure loss over the prototype are given as a functionof the flow rate for the SF6 test loop. The lines represent thetheoretical calculations, while the crosses, dots and circlesrepresent the measurement data. Opposite to the measure-ments in the low pressure test setup during which the flow inthe channels of the filter element was laminar, the channelflow during the high pressure measurements was turbulent(the Reynolds number in the channels varied between 9 · 103

and 2 · 104). For calculating the pressure drop over the filterelement due to friction, the Blasius relation is used. A valueof two is assumed for the factor n, which represents the addi-tional pressure drop at the entrance region of the duct.

All measurement series show a gradual increase of thepressure loss with increasing flow rate. The filter elementhas again the highest contribution to the total pressure loss.The measurement data of the pressure loss over the filterelement are slightly below the predicted pressure loss. Themeasurement data for the total pressure loss on the otherhand are somewhat higher than predicted. This can beattributed to the fact that in the model some flow phenom-

ena in the swirl generator which contribute to an additionalpressure loss are neglected. One of these phenomena is thefact that the angle of incidence of the fluid at the inlet of thevanes differs from the angle of the vanes itself. Furthermore,there will be an additional pressure loss due to the accelera-tion of the fluid as it enters the vanes (decrease in cross-sec-tional area due to the thickness of the vanes). Besides, wakesmay occur at the trailing edges of the vanes, which cause an


Figure 8. Pressure drop over the filter element and the whole separator as afunction of flow rate for air at 1.2 bar.

Figure 9. Pressure loss over the filter element, the swirl generator and thetotal pressure loss over the prototype as a function of SF6 flow rate. From leftto right the density of SF6 is respectively 11.2, 17.5, and 22.2 kg m–3.

Full Paper

extra pressure loss.Furthermore, it can be seen that for increasing fluid den-

sity, the pressure losses increase. This is also predicted by themodel as all pressure loss contributions increase for increas-ing fluid density.

3.2 Separation Performance

The efficiency of the separator is calculated by the follow-ing procedure:– First, the change in weight for each stage of the impactor,

including the backup filter, is determined by measuring allcollection media prior to and after sampling.

– All weight changes are added up to obtain the total parti-culate weight collected by the impactor.

– The fractions of the total collected weight in each stage ofthe impactor are determined by dividing the weight col-lected on each stage by the total collected weight.

– Multiplying these fractions with the total particle concen-tration present in the gas stream gives the absolute con-centration of particles in each size class.

– The efficiency per size class follows from the difference inthe absolute particle concentration on each impactor platebetween the measurements up- and downstream of the se-parator.In an ideal situation the total particle concentration up-

stream the separator Cus measured with the impactor equalsthe injected particle concentration. In practice, however, theimpactor concentration was two to three times less than theinjected concentration due to the loss of lime from the injec-tion point to the impactor.

The total particle concentration downstream of the se-parator can be determined with an absolute filter. Measure-ments performed at the University of Twente, however,showed that the total particle concentration downstream ofan RPS-based separator is approximately equal to the totalparticle concentration on the impactor plates when measur-ing downstream of the separator. In these experiments alsocalcium hydroxide and the same Anderson impactor wereused. Equal concentrations can be expected, as the impactorworks as a kind of absolute filter. Therefore, in processingthe measurement results the concentration downstream ofthe separator Cds is assumed equal to the total impactor con-centration of the downstream measurements.

When the total injected concentration and the total impac-tor concentration downstream of the separator are known,the total efficiency Etotal follows from:

Etotal 1 Cds

Cus

100% (4)

In Fig. 10 the results of the measurements are shown to-gether with two analytical predictions of the filter separationefficiency. The circles represent the mean separation effi-ciencies per size class obtained from the measurements,while the vertical lines through the circles represent the stan-

dard deviation. In the figure eight measurement points aredepicted. These represent all collection plates of the impac-tor, except the pre-separator and the back-up filter. The pre-separator is not accounted for as it only serves to removecoarse particles from the flow. The back-up filter is notregarded because, after the measurements upstream of theseparator, no weight increase of the back-up filter could bemeasured. The filter efficiency is given as a function of thedimensionless particle diameter x. This diameter representsthe mean particle diameter per size class divided by the di-ameter of the smallest particle which is separated with100 % probability in the filter element [3].

Analytical expressions for the separation efficiency of thefilter element were derived by Brouwers [10]. In Fig. 10 thedotted line represents the theoretical separation efficiencyfor triangularly shaped channels, which have a flow distribu-tion linearly proportional to the radial distance r, while thesolid line represents the efficiency for triangular channels inwhich the axial flow distribution is constant over the chan-nels, uf 1. In both models a parabolic velocity profile inthe channels is assumed (in reality this is not valid, as thedynamic entrance length, Lhy for the channels is ratherlarge). Theoretical expressions for triangular channels areused, as they are easier to derive than expressions for sinu-soidally shaped channels. Besides, the theoretical results fortriangularly shaped channels are a good approximation tothose of sinusoidally shaped channels [10]. Furthermore, inpractice the channel shape is quite irregular and also thesinusoidal shape is only an approximation of the actual chan-nel shape.

It can be seen that the standard deviation increases forsmaller particle sizes. This is expected as the measurementerrors are larger in these size ranges, due to a smaller weightdifference between the impactor plates prior to and after themeasurements. For particle sizes around x = 1 and larger,the agreement between theory and experiments is reason-able. However, for particle sizes smaller than x = 1 a large


Figure 10. Separation efficiency as a function of particle size. The solid linerepresents the theoretical efficiency prediction for a constant axial flow distri-bution, the dotted line represents the theoretical efficiency prediction for anaxial flow distribution, which is linearly proportional to r and the circles rep-resent the measurement data (vertical bar denotes the standard deviation).

Full Paper

difference between the models and the experiments is ob-served. Previously performed measurements with an electri-cal driven RPS, using the same Anderson cascade impactorto measure efficiency, showed similar results [11], see Fig.11. It can, therefore, be concluded that the naturally drivenRPS has the same particle separation efficiency as an exter-nally driven RPS.

In the results presented above, the particle losses frominjection to the sample point are not accounted for. If theinjected particle concentration is divided by a factor two theresult given in Fig. 12 follows. With this modification a betteragreement between the theoretical and the mean measure-ment values is obtained for values of x smaller than one. Thestandard deviations, however, have increased as a result ofthe adjusted inlet concentration.

From the measurement results also the total mean separa-tion efficiency can be determined, as was given by Eq. (4).In case the particle concentration upstream of the filter ele-ment Cus is taken equal to the injected particle concentra-

tion, the mean total efficiency is 86.9 % with a standard de-viation of 1.7 %. If the particle concentration upstream ofthe filter element is taken half of the injected concentration,the mean total efficiency is 73.7 % with a standard deviationof 3.4 %.

4 Discussion

The hydrodynamic operating performance of an RPS-based natural gas-water separator was measured at atmo-spheric conditions and at elevated pressures. During thesemeasurements the angular speed of the filter element andthe pressure drop over the separator were recorded for vary-ing fluid volume flow. At atmospheric conditions also theseparation performance of the separator was determined byinjecting solid lime particles upstream the separator andconsequently measuring the particle distribution both up-and downstream of the separator by a low pressure impac-tor. All measurement results are compared with theoreticalpredictions. Overall it can be concluded that the characteris-tics of the prototype are well predicted by these models.

The efficiency of the prototype is determined for laminarflow conditions in the channels of the filter element. Underfield test conditions, however, the flow in the filter elementis turbulent. Due to turbulence the radial migration of thedroplets towards the channel walls may be disturbed. To de-termine the influence of the turbulent channel flow on thefilter efficiency, measurements should be performed at high-er flow rates and higher angular speeds. These measure-ments could be performed with the lime particles as wasdone in this study. However, it would be better to use drop-lets. The reason is that not only the influence of turbulenceon the radial migration of the particles in the channels of thefilter element may influence the separation efficiency, butalso the behavior of the liquid film in the channels of the fil-ter element and the outer wall of the post-separator may beinfluenced by the turbulent flow conditions.

Received: December 12, 2005

Symbols used

Cds [kg m–3] particle concentration downstream ofseparator

Cus [kg m–3] particle concentration upstream ofseparator

dc [m] channel heightdp [m] droplet or phase diameterdp, 100% [m] diameter of droplets or phase

collected with 100 % probabilityDo [m] outer diameterDh [m] hydraulic diameterE [–] efficiencyf [–] friction factorGh [N m] angular momentum


Figure 11. Separation efficiency as a function of particle size. Comparison ofcurrent measurements (circles) with previously performed measurements (as-terisks and triangles).

Figure 12. Separation efficiency as a function of particle size. Injected particleconcentration is divided by a factor 2 to account for particle losses during in-jection. The solid line represents the theoretical efficiency prediction for aconstant axial flow distribution, the dotted line represents the theoretical effi-ciency prediction for an axial flow distribution, which is linearly proportionalto r and the circles represent the measurement data (vertical bars denote thestandard deviation).

Full Paper

L [m] lengthLc [m] channel lengthLhy [m] dynamic entrance lengthL

hy [–] dimensionless entrance lengthn [rev min–1] rotational speedr [m] radiusRe [–] Reynolds numberufe [m s–1] axial fluid velocity in filter elementusg [m s–1] axial fluid velocity in swirl generatorwsg [m s–1] tangential fluid velocity in swirl

generatorx [m] dimensionless particle diametera [–] blade angleDP [Pa] pressure dropqf [kg m–3] fluid densityqg [kg m–3] gas densityn [–] pressure loss factor at entrance

channel

References

[1] H. P. van Kemenade, E. Mondt, A. J. A. M. Hendriks, P. H. J. Verbeek,Chem. Eng. Technol. 2003, 26 (11), 1176. DOI: 10.1002/ceat.200301803

[2] E. Mondt, Compact Centrifugal Separator of Dispersed Phases, PhD the-sis, Eindhoven University of Technology, 2005.

[3] J. J. H. Brouwers, Exp. Thermal Fluid Sci. 2002, 26, 325.[4] J. Wilhelm, D. Seibt, E. Bich, E. Vogel, E. Hassel, J. Chem. Eng. Data

2005, 50, 896.[5] SKF Hoofdcatalogus, Veenendaal: Svenska Kullagerfabriken Nederland

1990.[6] H. Schlichting, Boundary-Layer Theory, McGraw-Hill Inc, New York

1979.[7] R. K. Shah, A. L. London, Laminar Flow Forced Convection in Ducts:

A Source Book for Compact Heat Exchanger Analytical Data, AcademicPress, London 1978.

[8] R. B. Bird, W. E. Stewart, E. N. Lightfoot, Transport Phenomena, JohnWiley & Sons, New York 1960.

[9] J. A. Owczarek, Fundamentals of Gas Dynamics, International Text-book Company, Scranton, PA 1964.

[10] J. J. H. Brouwers, Powder Technol. 1997, 92, 89.[11] J. J. H. Brouwers, Chem. Eng. Technol. 1996, 19 (1), 1. DOI: 10.1002/

ceat.270190102

______________________


Full Paper

operating performance of a naturally driven rotational ... · fils these requirements is the...

Documents