The Designof targe Diameter Skim Tanks Using Computational Fluid Dynamics (CF'D) For Maximum Oil Removal 15frAnnualProduced WaterSeminm, January 2005 Hilton NASA ClearLake,Houston,Texas72058 Chang-MingLee, Ph.D. andTed Frankiewicz,ph.D. -NATCO Group Inc. 2950North Loop West,Suite750,Houston,TX77O92 Abstract A new designof internals of a large diameterskim tank is discussedin this paper. Both transient and steady-state CFD simulations were perfomred to assist the development of the design. Excellent flow distribution was achieved on the inlet diskibutor as the variancein massflow ratesout of the lateralsis lessthan + 0.5%. The retentiontime of producedwater was improved to almost 88.5% of the theoreticalvalue with installation of two largeperforatedplates. Flow path lines were also improvedwith a much reducedrecirculation pattern. This compareswith a residencetrme of 43.4Toof the theoreticalvalue for the casewithout the two lmge perforatedplates. The CFD model predictsthat oil droplets size larger than 100 microns are expectedto be fully skimmed off without assistancefrom gas flotation for producedwater at the design flow rate of 100,000 BBL/Day. Introduction The production of oil and gas is usually accompaniedby the extraction of assooiatedwater. After the initial bulk separationof the bulk produced fluids, the produced water still contains finely dispersedsolids and oil. In order to reduce the oontaminantcontentof dispersedoil in producedwater, a large diameterskim tank canbe used. Sepmationis basedon the differencebetweenthe specific gravity of oil and water andthe coalescenoe of small oil droplets. When the retentiontimo is sufficient, oil floats to the surfaceand can be separatedby an overflow. This techniqueis only suitable for removingdispersedoil with a sufficiently large particle size. Dissolvedmaterialssuchas benzeneandhealy metalscannotbe separatedusing this techaique. The largo diameter skim tank or its modified version, parallel plate interceptor (PPI) or comrgatedplate interceptor(CPf, is mostly usedas part of a set of techniques for the rerroval of dispersedoil. ln this paper, we discussthe developmentof a new configurationof intemals for a large diameterskim tank. The designof the intemalswas developedwith aid from a seriesof CFD (ComputationalFluid Dynamics) simulations. The objective of the CFD developeddesignis to provide a large diameterskim tank with optimized flow pattems that maximize liquid residencetime and improve oil/water seoaration, Tank Description The skim tank is desiped for a flow rate of 100,000BWPD, has a diameterof 80 feet (24.384m), and a height of 24 feet Q .3152 m). The drawing of the skim tank and the inlet distributor me shownin Figure 1. Tho inlet distributor oonsistsofa headorand eight parallel laterals. Becausegasbreakoutin the oiVwatersettling zonecould seriously degrade separation performance, the fluid inlet incorporates a tangential enrry* configuration to promote gas/liquid separation prior to liquid reaching the inlet distributor. The open-endof eachlateral is directedtowmd the wall of the skim tank and employs the tank wall as the equivalent of momentum breaker and primary flow doflootor. The skim tank has fwo large perforatedplates. One is located near the inlet distributor and another is located between the three water outlet nozzles and the oil skimming collectors. The large plates are an assemblyof several smaller pieces of perforatedsheetswith thicknessof l/2 nch. The two oil skimming collectorsaf,eset at different eievationsand they are suitablefor operatingthe skim tank at the high and low levels of 22.5 feet and 17 feet, respectively. There are alsothreebaffles surroundingthe outlet nozzles to prevent possible flow short circuiting. Among the three baffles, one solid plate is positionedpmallel to the lmge perforatedplate andthe other two perforated baffIesareboth perpendicularto the solid baffle. Model Simplification The constructionof the three-dimensionalgeometrywas basedon the dimensions of the skim tank as provided by the field operatorand somesimplification was required for modeling pulpose. The three-dimensionalmodel with a description of intemal componentsof the skim tank is shownin Figure 2. Tlrc perforatedplatesaremodeledas "porous-jump" boundariesof zero-thickness- the mathematicalequivalent of physical fritted plates - in order to avoid extremely complex meshing of the region around individual orifices. The oil skimming troughs are simplified without the V-notch slot detailsandtheir inclination. In addition,all of the intemal piping is ignored. During the fust stageof the project, the bulk part of the tank volume positioned betweenthe two large perforatedplates was excluded. As a result, more details and tighter meshingcould be appliedin the inlet region. This perrnittedthe study of the fluid flow distribution through the inlet devioewith the computingtime still constrainedto a manageablescalefor rururing dynamic simulations. In the later stageof the project, the bulk volume of the tank is taken into accountbut inlet boundariesare further simplified andassumeequalflow rate out of eachlateral in order to studythe flow patteminsidethe skim tank. The simplified tank model is shown in Figuro 3 and meshing details of the inlet region areshownin Figure 4. Two LargePerforatedBafnes Figxe2. 3-D Model of the entire skim tank with description of intemal components. t_ Figure 3. Simplified skim tank model wio the bulk volume between the large perforated plates. 4 L - -: ? F r b0! pt '-= = r irt) ?ig'. v A - l l r t r Y a) - E J Meshine of the Flow Domain For CFD, the computationalapproachdiffers from analytical (or theoretical) solutionsin that the CFD softwmeonly solvesequationsat specificpoints ratherthan for the entire flow field. Properly choosing these points may become quite diffioult especially for a very complex geometry and sometimesit may require hundreds of thousandsor millions of points. Before thesespecific points can be generated,the user must first determine the size of the region to be modeled -- this is the so-callec "computational domain". The computational meshing process breaks up the computationaldomain in space,then the calculations(solving the firndamontalequations for mass, momentum and energy balance) are carriod out at each node point simultaneouslyor sequentially,dependingon the selectedsolver. The quality of the gdd is extremely important and can strongly influence the solution- sometimesalso detemriningwhetheror not a valid or a convergedsolution can be aohievedat all. The most important qualifioation about a computationalmeshingis that it must define enoughpoints to captureeverythingof interestthat is happeningin the computationaldomain without becoming so extensive that umeasonablecomputation times are requhed. If there are too few points included, someof the critical infomration aboutthe flow regimemay be completelylost. The types of meshingthat arrangethe spatial points in a roughly rectangular(or hexahedral)affangementme referred to as "structured" meshing. Another method of meshgeneration,which involves a triangular (or tetrahedral)layout of the computational points, is called "unstructured"meshing. Sometimes,different meshescanbe adaptedin fifferent regions decomposedfrom the whole flow field to take advantageof specific characteristicsof the different meshitrg systems. Such a combination is so-ca1led '.hybrid" meshing(also including prism/wedgeandpyramid elements). In this study, a hybrid meshingwas constructedin the first stageof the study for the simplified skim tar* model without the bulk volume betweenthe large perforated plates. The hybrid meshingwas comprisodof 1,519,160hybrid cells,andthe total node countwas367,855with 3,194,845faces.As describedpreviouslyandshownin Figure4, the surfacemeshingof the inlet headerandthe lateralsillushatesthe detailsof the tighter meshingsoherneimplementednear the region of the inlet distributots. In the later stage of the study, a fully structuredmeshingwas createdincluding the bulk volume of the tank betweenthe two large perforatedplates and there are a total of 695,620hexahedralcells with 730,041nodesand2,122,171faces.The reductionof total cell countwas dueto the relativoly loosemeshingschemeand a firther simplification of the inlot boundmies. Material Properties The CFD sotwme used in this study is capable of simulating multiple phase flows of a computationaldomain with both structuredor hybrid meshing. The fluids used in this study were producedwater and its associatedoi1 content of7l0 ms/liter. Underthe operatingconditions,35'c and 101.3kpa, the densityof the producedw-ateris I I 04.3kg/m' andits viscosityis 0.7225cp,-whereas, the densityand viscosityof rheoil phaseat operatingconditionsare 814.9kg/m3and 3.54 cp, respectively. Simulation Models In this study for skim tank design, two diflerent CFD simulation models were employed.The following is the brief descriptionof the two simulationmodels. VOF (Volumo of Fluid) Model: It is a multiphasemodel accqrting simultaneous different fluid flows. However,the phasesme not interpenetrating.The volume fraction of eachphasewithin a computationalcell is determinedand usedto volume-averagethe propertiesof the individual fluids. No particle sizes can be specified for the incoming phasesand it is usually done with transient calculationswith small time steps. This model was used in the first stage study focusing on the flow rate through the inlet distribution device. Steady-State(Water PhaseOnly) Model: The Single-PhaseSteady-Statemodel is usedto developthe flow regime below the water surface. Therewas a virtual barrier at the surface in order to generaterealistic flow profiles for a totally water-filled tank model. Oncethe flow regimehas beencalculatedin the steady-statemodel, particlescan be injected at ths inlet (or any other location of the operator'schoice) to visualize their movementwifhin ths flow field. The propertiesof the particles (size, density, etc.) can be chosento simulateany of the phasesbeing studied. Using this technique,it is possible to deteffrine partiole sizes likely to be carried out with the oppositephaseas well as retentiontimes of eachphasewithin the skim tank. Results and Discussions Flow Ratesof Inlet Distributor: The mass flow rates in the inlet distributor through each lateral at different simulationtime from 100 to 120 secondsare listed in Table 1 and the correspondingplot is shown in Figure 5. The aboveflow rateswere calculatedduring the first stagestudy using the VOF Model with transient simulation and a total simulation time of 120 seconds.Note that the variancein massflow ratesout of the lateralsis lessthan + 0.5%. This constitutesexcellentflow distribution from the inlet distributor. Table1: MassFlowRatesthroughInlet DistributorLaterals Time Time '110sec FlowRatio Time 00 sec FlowRatio oA lo 120sec FlowRatio yo % 25.234 't2.55% 25A41 12.550 25.390 12.54% Arm-82 24.401 12.14% 24.72'1 24.755 12.24o/o Arm-83 24.684 12.28% 24"973 12.20o/o '12.32% 25,019 12"36V" Arm-84 26.301 13.08% 26.912 12.98% 26.141 12.s1% Arm-F1 25.280 12.58% 25.490 12.58o/o 25.442 12.5tvo Arm-F2 24.g16 12.10Yo 24.632 12.15% 24.690 12.20% Arm-F3 24.525 12.20% 24.8'14 12.24% 24.860 12.280 Nm+4 26.278 1i.O7% 26.291 12.97% 26.122 12.90% Summation 2O1.O19 100.00e/o 202.674 100"00% 202.448 100.00% lfl |'l a-u t "-, E Ete l4 IU Sloulation Tlne {sec} Figure 5. Massflow ratesout of ind:ividualinlet distributorlaterals. Skim Tank Retention Timet oil and water separationis fundamentallya graity separationprocess. Almost any production facility can gain significant benefits &om inclusion of a gravity separator in the primary stage of producedwater treaknent. However, the primary limitation of successlul gravity separation is the size of the vessel required to prwide adequate retsntion timo for fi;ll heatrnentand the efficient use of availabletank volume. In tlFD simulations, produced water retention times were detennined by injecting water "particles" into the model and tracking their time to exit the outlet nozzles. Thi results are shown in Figure 6 and Figure 7, in histogram plots and cumulative curves as a percentageof total particles exiting, for skim tank intemals without the two larse perforatedplatesand after the installation of the two lmge perforatedplates. A comparison of the theoretioal (volumetric displacement)retention times for produced water with the median retention times determinedby cFD analysis also is shown in the table on the right-hand-sidein both Figures 6 and 7. The theoretical retentiontime of the skim tank is 17,406secondsbasedon a liquid level of 225 feet witl a total producedwater flow rate of 100,000BBL/Day. As can be seenfrom the tables, water retentiontime for the skim tank without the two perforatedplateswas abofi 43.4yo of the theoreticalvalue. This is a clear indication that short circuiting occursinside the skim tank. On the other hand, tle retention time of producedwater was improved to almost88.5% of the theoreticalvalue after installation of the two large perforatedplates. Furthermore,minimum value of the retention time for the latter was 7434 secondsas comparedto 634 secondsfor the former case. These results suggesteda significant improvementof effectively utilizing the bulk part of the tank volume when implementing the designand installation of the two largeperforatedplates. Fluid Flow Patterns: As traced by the flow pathJines plots of the CFD software,the fluid flow paths, coloredby retentiontime, inside the skim tank without and with the two large perforated plates are shown in Figures 8(a) and 8(b), respectively. In Figure 8(a), ttrere me lmge swirling pattemsthat are characteristicsof severerecirculationand the presenceof shortcircuiting flow paths inside the skim tank. In contrast,the path lines are improved in Figure 8(b) with a much reduced recirculation pattern. In separatesimulations, the calculationsshowed similar flow patterns with a produced water flow rate of 30,000 BBL/Day and the skim tank liquid level set at l7 feet insteadof the oiginal 22.5 feat. The plots of flow path lines for thoselow flow simulationsareillustrated in Figures9 (a) and9(b). Carry-under of OiI dropletsin Water Phase: In order to evaluatethe sepmationperformanceof the skim tank with desigaed internals,the entrainmentof oil dropletsin the water phasewas studiedby the injection of various sizes of oil "particles" into the water phaseat the jnlet and then determ:ining their potential for carry-underby following the particle hacks generatedby the CFD simulations. It should be noted that current cFD modeling is unablo to simulato the coalescenceof oil droplets over time. Therefore,the sepmationperfomranceillustrated hereis for the worst casescenariowherethe reverseemulsionis fully stabilized. The CFD model used in thesedeterminationswas the single-phasesteady-state model with a virtual barrier placedat the surf,aceas previously described. The sideview of the tank with particle tracking for 50 Fm, 75 pm, and 100 pm oil dropletsareshownin Figures 10' 1l' and 12, respectively. Examination of these figures showsthat there is substantialenhainrnentofthe 50 pm oil droplets,only slight entuainmentofthe 75 pn oil droplets, and no entrainment of the 100 pm droplets. In summary, with the implernentationof designedinternals,the oil dropletsof size larger than 100microns are expectedto be readily slcimmedoff without exha assistancefrom gasflotation inside the skim tank at a producedwater flow rate of 100,000BBL/Day. In a later study with a lower producedwater flow rate of 30,000BBL/Day and skim tank liquid lovel set at 17 feet, the minimum oil droplet sizethat was readily skimmedoffwas reducedto about50 microns. Conclusions Computationalfluid dynamics simulations were perfomred on a lmge diameter skim tank and the results indicate that a satisfactoryperformanceof the tank can be achievedwith the proposedintemal configuration under the specified producedwater flow rate provided that the incoming oil droplets are not too small and that bulk gas evolution in the inlet region is eliminated or minimized. With the installation of two large perforatedplates,the "Median" value of the estimatedretentiontime from particle traoking is 88.5% of the calculatedtheoretical retention timo. This suggestsa very effectiveutilization of the bulk part ofthe skim tank volume. The results from transient CFD simulations demonstratethat an excellent flow distribution throughthe inlet distributor is achievedwith the variancein massflow rates out of eachlateral being less than + 0.5%. From the particle tracking results,the CFD model predictsthat oil dropletslargerthan 100microns are expectedto be fully skimmed off without assistancefrom gas flotation at a produoed water flow rate of 100,000 BBL/Day. 10 s [ f r n | . t u 3 | . r c f , t F q E F i X f t r i E o i + ( f i d F .-) r t i h ll ; - afl F'1 : e J ! J = a ltr ,f; R t ,a fr clt l > q # B ! = q t ab c E l r r r t r 5 t C r € c c i \ | E J d E t 'F o c 5 E = . E = i c = o pa * F = o ] 1 ooooo + 000s6 T 000c6 | $ ooooo | + 0001.8 I i[ ooots t"t| :1,000r0 I q; 0008r / \r oo (E d 'lt c, ,F' .E !- o lL o o(\I o | $ oooer ooozt. ]$ t_ :;00089 s a o o P o ,P I + oooss | + 000es I j: 0000s a, 1 g: 000ts {t # 000ts E F It ooors I l: 0008t .9 l€:000sf I B: 0002t Eo I E: ooooc o t! EO F o .F E .E I E:ooogc € oooec ls 0000e CD I a 00012 E E €: € ooosl H_ 0009r !a L o .9 I J c cl F 5 ll- ffi ooorz ooolz :=- 0002! H # 0006 0009 ffi ooot Efo a . r € L n E e s OJ :A \ r r c a ! } = E t {,i 6 S f [ b ? * E r 9 o 5 : E E E( J f E i ! $ Gi 5 - / C . E l ( E E s $ H $ E 00066 00096 00006 00006 00028 000t8 000r8 0008r. 00091. 00021. 00069 00099 000e9a 00009 () o 00029(, 00019 .E 000t9F a Fl IL (JI lut = = H 2l (!l 6 r l'fl r c E = ( sl o ; f * o , 6 dt 5 F E - F i E I r f i S 8 ; j + E i f r q I o o .E oE (, tg o rF L. o o- 000st; N ooozl I(, 0006ec 00098 00000 00008 = E IE ct OOOLZ o .9 ooovz 00012 0008t 00091 0002I 0006 0009 000e 0 ,:< E (E 5 lt- Icuenberg s (tn /^. T H, (a) T E-X (b) Figure 8. Fluid flow path lines insidethe skim tank (a) Withoutlargeperforatedplales. (b) With two largeperforatedplates. t3 E ;--9 > o e d ! '5 . q) : o a0 ' d ; trO -i -5Z ;6C/] -_l t r ( ) E E 3C 3 gS 3 C 3 3S 3 8 3 3 3 33 3 3 33 3 3 3 3B 3SSHe8 h S P 3 f i8 + E N s 8 S$ f iE E E f 6l 8 8 3 ct(\rN6i q : d > q ) oi .:,'f:il a t f d Y o . v o - i i /.-l\ c I Eo I w v ^ 2.40s+04 2.34R+04 e.e5F+04 2.1Se+04 2.fl7e+04 1 'I.98e+04 .Bge+04 1.804F04 1.71e+04 1.62e+04 1.558+04 1.44e+04 1.S5FFS4 '1.26s|.04 ffi F*Ettiq,r \/ia:vi, l.l 78+04 1.08s+04 S.g0e+03 9.00e+09 8. 10e+03 7.20RF03 6.90R+03 5.408+03 4.508+03 3.60e+09 2.f08F03 1.808+09 9.008-F02 0.00e.+00 Particle Size: 50 nlicron Figure10. Particlehackingofoil dropletsof50 micronrurderdesignedflow rate. 2.34e+04 2.2Se+04 2.1S8F04 2 . 1 1e + 0 4 2.02Fr-04 l.Sge+04 1.84e+04 '1.76e+04 i-;fii illii: rr' E{:W 1.67e'F04 l.5BFlo4 1.49e+04 1.40e+04 1.S28+04 1.23e+04 1.14e+04 1.058+04 $.66e+03 S.78F|09 7.U0e+09 7.0?s+03 w g*l 6.14RF03 5.27s103 4.ggB+09 3.51B|03 e.6gFFi3 '1 .76e-'-03 T | L7ge+02 0.00s+00 r ,- ParticleSize: 75 micron Figure 1I . Particletracking of oi1dropletsof 75 micron underdesignedflow rate. lf 3-SFa+09 3.gBs+09 2.8P8+09 2.25FF0,9 1.SSB+0S 1.1S+$0S 5.S48+02 l-, Y----Y 0.008+0S ParticleSize= 100micron - 16