Journal of Petroleum seenceané Engineering 146 (2016) 735-145 Contents lists available at ScienceDitect ee SORE Journal of Petroleum Science and Engineering £ a ELSEVIER journal homepage: www elsevier.com/locate/petrol CFD simulation and experimental study of oil viscosity effect on multi- stage electrical submersible pump (ESP) performance Jianjun Zhu, Hattan Banjar, Zhenyan Xia, Hong-Quan Zhang McDougall ool of Petroleum Engineering, The Univesy of Tuba, S005 Tucker Dr, Tuba, OX 76104 United Sates Dew ARTICLE INFO ABSTRACT Tn this study, the olf viscosity elect on ESP performance is investigated by experimental study and computational fluid dynamics (CED) simulations, Oils with different viscosities are lowed through DNI7SOESP at varying low rates and rotary speeds. The temperature is maintained at different levels witha heat exchanger citculated by cold water. The pressure increase over the 3° stage 35 well s total 7 stages in ESP is measured with dierenil pressure transducers. The same geometries, Nid properties and flow characterises are implemented into CFD simulation. The three-dimensional (3D). steady-state Received in ceed fon ajune 2015, ‘Accepted 25 july 2015 este online 27 fly 2016 Keywords Reynolde-veraged Navier-Stokes (RANS) equations with standard SST (shear stress transport) tirbu- Ati it lence models are solved in ANSYS CFX by employing frozen-rofor technique. With high-quality struc- Viscosity eflect ‘tured hexahedral mes, the simulated pressure increments compared with corresponing experimental resulls. How structures inside ESP impeller and eiffuser channels ae analyzed. At pump best eliciency| Doint(BEP), the boosting pressure decreases 30-40% when oil viscosity increases from 10 cp to 100 ¢. ESP becomes ineffective when ol viscosity is higher than 200 cp. With oll viscosity increasing, pump H-Q performance curve becomes mote linear. CFD simulation reveals thatthe recirculation flow at impeller blade trailing edge is more prominent at lower liguid fw rates ‘© 2016 Elsevier BV. Al sights reserved. 1. Introduction Widely used in petroleum production nowadays, the electrical submersible pump (ESP) is good at boosting hydrostatic pressure and increasing production rate compared to non-totary artifical lift methods (Takacs, 2009), Usually, ESP is cascade-assembled and compactly stacked in the downhole ofan oil well. Its performance is affected greatly by downhole flow conditions such as multi- phase flow and hhigh-viscosity oll low. With the increase of oil viscosity, ESP boosting pressure becomes lower corresponding to the same flow rate, of the flow rate through ESP decreases cor- responding to the same boosting pressure. Compared to viscous ‘uid pipe flow, understanding and modeling viscosity effect on flow behavior inside twisted channels of ESP impeller are more challenging. Previous studies on prediction of ESP pressure increment for high viscosity oil mainly focused on experimental ‘measurements (Trevisan and Prado, 2011; Barrios et al. 2012; Banjar et al, 2013) and proposing empirical correction factors ot charts (Hydraulic Institute, 1948), Ippen (1945) conducted over 200 performance tests for oil viscosities up t© 10,000 SSU (Saybolt Second Universal) on four TCorrerponcing autor mal adden anjunhvoutleaede 2 hep féeoiorg)101016) peso! 201607033 {520-4105} 2016 Elsevier BY. Al rgteresere. variants of centrifugal pumps, The experimental results were summarized by plotting the ratio of oil head to water head, brake horse power (BHP) and efficiency against a Reynolds-type di- ‘mensionless number, based on which the general correction fac- tots for specific speeds from 800 to 2200 were proposed. liy- draulie Institute (1948) provided a typical empirical approach with correction factors to estimate conventional centrifugal pump boosting pressure for viscous liquid flow if the water performance were known, However, the accuracy of this approach was ques- tioned by Gilich (1998a, 1999) and Li (2002) since the experi- ments cartied out by Hydraulic Institute were within a narrow range of the pump specific speeds, Unreasonable errors were found if extrapolation was beyond that range. Turzo et al. (2000) generated equations to correct centrifugal pump boosting pressure for handling viscous fluids based on Hydraulic Institute charts. By i che original correction charts and performing a regtes- sion analysis, the curve-fitted expressions were obtained. Ste- panoit (1957) proposed similar Reynolds-rype number to correlate experimental data by using only one correction factor to get the new H-Q curves if the water performance were known. A more general model based on the evaluation of viscous dissipation for disk and hydraulic frictions to predict the boosting pressure of centrifugal pumps was proposed by Gilich. The friction losses on disk and in flow passage, as the author claimed, were dominating factors. impairing centrifugal pump’s ability to handle high Please cite this article as: Zhu. J, et al, CFD simulation and experimental study of oil viscosity effect on multi-stage electrical submersible pump (ESP) performance. J. Petrol. Sci, Eng. (2016), http: /éx.loi.org/ 10.1016)}petrol2016.0703326 12h eo oral of Peseun Science and Engineering 146 (201) 735-745, Nomenclature blade thickness, L mim best efficiency point absolute velocity vector, LT, mis impeller diameter, Lm’ gravity acceleration vector, (12), mjs* channel height, (, mn sand equivalent roughness, 1 im dimensionless number hhydraulic head, L, m turbulent kinetic energy, 17), m/s" channel length, L, m rotary speed, 1/7. pm stage pressure increment, MI(LT2), Pa flow rate, UIT. m)s mass flow rate, MIT, kg/s radius, L, mm absolute values of real roughness, jim Reynolés number time, Ts velocity vector, LT, ms peripheral velocity vector, LT, m/s velocity fuetuation, Lj, mis relative velocity vector, IT, m/s blade number BeURSa Freon S goat RET Ose Greek symbols B tangential blade angle, deg 1 efficiency 2 rotary speed, 1/7 rads @ specific dissipation rate or turbulent frequency, 1/, s } flow coefficient y head coefficient Pp —— uid density, M/L, kgim? + external force exerting on bubbles, M(LT?), Pa tw wall shear stress, MI(LT2), Pa stress-strain tensor, MI(LT?), Pa . wall roughness, Lm Subscripts 1 inlet 2 outlet 6 blade eh centrifugal force Cor Coriolis force d diffuser t impeller ° initial or equivalent p relative to project area viscosity fuids. Compared with available data, Galich also pointed ‘out that friction losses were affected significantly by pump geo- ‘metrical features, fuid properties and thetmal conditions. Recent experimental studies conducted by Amaral etal, (2009) and So- Jano (2008) further revealed that the Hydraulic Institute charts and empirical correlations available in literature were unable to ive appropriate correction factors to predict ESP boosting pres- sure for viscous oil low. ‘With the continuous advance of computing technology, CFD is, becoming a powerful too! to study centrifugal pump performance under design and off-design conditions. Numerical simulations of centrifugal pump design optimization (Qi et al, 2012; Zhang et al. 2013), instantaneous pressure fluctuation (Gonzalez et al, 2002; Gonzalez and Santolatia, 2008), multiphase flow (Minemura and Uchiyama, 1993; Zhu and Zhang. 2014) and high viscosity fluid flow have been carried out. Shojaeefard et al. (2006, 2012) con- ducted both experimental study and numerical simulation on a centrifugal pump handling viscous fluids, The authors stated that a ‘good agreement between simulation and experimental data was obtained by solving the steady state RANS equations with SST k-ay turbulence model, Based on the same pump geometries, irino et al. (2013) and Stel et al. (2074) performed numerical in- vestigation of viscosity effects on single-stage and three-stage EsPs, respectively. Similar numerical methodologies were used in their work including SST turbulence model with transient rotor- stator technique, Both studies matched experiments well under a wide range of fluid viscosities. In addition, Stel et al. (2014, 2015) pointed out that CHD simulation with multistage ESP geometries agreed with experimental results better than that based on single- stage pump. Recently, an interesting phenomenon of sudden-ris- ing head effect when a centrifugal pump transports high-viscosity| fluids was studied by Li (2014). By implementing standard k-e turbulence model and non-equilibrium wall function into RANS ‘equations, the author confirmed that the sudden-rsing head effect was ascribed to the low transition from hydraulically rough re- ime to hydraulically smooth regime. ‘Although experimental studies and numerical simulation on centrifsgal pump transporting viscous fluids have been conducted, the accurate estimation of viscosity effect on ESP boosting pres- sure is still difficult due to pump complex geometries. In this study, the experiments and CFD simulations are carried out to investigate the effects of fluid viscosities on ESP boosting pressure Oils of varying viscosities are lowed through a DN1750 ESP with different flow rates and rotary speeds. The overall boosting pres- sure of all 7 stages in ESP and the pressure increment over the 3rd stage are measured by differential pressure transducers, The same geometries (6 blade impeller and 8 vane diffuser, 7 stage cascade assembly), uid properties and flow conditions ate incorporated into CFD simulations. The numerically simulated boosting pres- sures are compared with experimental results. The correspondent flow structures inside the impeller and diffuser channels are analyzed. 2. Experimental setup Fig. 1 below shows the layout of experimental facility in this study. The testing flow loop consists of an ESP. a control valve, a viscosity measurement section (pipe viscometer), a fluid injection section, motor and driver, lowmeters, pressurejtemperature sen- sors and cooling units etc The tested ESP is a seven-stage mixed flow type centrifugal pump. The pump is propelled by an AC electric motor (Baldor 18404489) which is controlled by a variable speed drive (VSD, Centrlit 22S0VT).11is connected to a thrust chamber which holds ump thrust force and allows shaft to rotate. As suggested by Croce (2014), the net positive suetion head required (NPSHR) for ESP operation is supplied by a compressor connected ta a port of Please cite this article asi Zhu, J. et al, CFD sinvulation and experimental study of oll viscosity effect on multi-stage electrical submersible pump (ESP) performance. J. Petrol. Sci. Eng, (2016), hp: |/dx doi. org/ 10.1016)) petrol 2016.07.03J.2hu el Journal of Petroleum Scene and Eninerg 16 (2018) 725-745 ny Nip the fluid injection section located upstream of ESP intake. Thus, no booster pump is needed in the experiments. The gas entrainment is avoided by reducing the contact area between air and liquid using a floating cylindrical piece in the main vertical injection pipe. The flow rate is controlled by a pneumatic choke valve (Cy- clonic TDC2050) downstream of ESP outlet, The flow rate is ‘measured by one of the two flowmeters for high range (Micro Motion DS300, 07000 Ibjmin, +-0.15%) and low range (Micro Motion CMFI00, 0-100 Ibjmin, 0.10%). All pipe sections are thermally insulated. The flow loop temperature is maintained at required levels by a heat exchanger (Mueller AT20PH) circulated with cold water Pressure tansducers (Rosemont 3051 SD, 0-200 psi, +0.25%), and resistance temperature detectors (RTD, Omega PRII, ~200 to {600 °C, 0,00385)*C) are mounted at the prescribed locations (see Fig. 1). The stage 3 is used to measure ESP single stage pressure increment (from 3rd impeller inlet to 3rd diffuser outlet). Differ- ential pressure and temperature transmitters are also installed at the stage 1 inlet and the stage 7 outlet to measure the ESP overall boosting pressure. All signal cables from instruments, including pressure and temperature transmitters, control valves, VSD, etc. are wited to National Instrument (Nt) Fieldpoint /0 modules en- closed by a water-proof box. Ethernet cable is used to commu- nicate between NI modules and data acquisition computer, where the analog signals are collected and processed by data acquisition (DAQ) system based on Labview 140 virtual graphic programing language. 3. Numerical methodology For 3D numerical simulations, the steady-state RANS equations with standard SST turbulence models are solved in ANSYS CFX 15 by employing frozen-rotor technique. The simulated geometry includes 7 pump stages, exactly the same configuration used in experiments, Each stage comprises of a channel wise-sliced impeller and diffuser, on which the structured hexahedral grids are generated with Turbogrid 15, 31, Geometry and meshing There are 6 blades and 8 vanes in impeller and diffuser for each, stage. The major geometrical specifications are listed in Table 1. At best efficiency point (BEP), the operation parameters are: rotary speed N=3500 rpm, mass flow rate Q=3.3 kgis (1795 bpd), hy- ddraulic head H=5.? m (6.11 psi), and efficiency 7 =68.5%. Fig. 2 displays the DN1750 ESP 3D model, including impeller blades (Fig. (a), diffuser blades (Fig. 2(b)), and the entire single- stage assembly (Fig. (0), Since the flow felds inside centrifugal pump are axisymmetric (Zhu and Zhang, 2014), a single channel «an be used to save computational cost and improve numerical efficiency (Caridad et al, 2008). Thus, the computational domains of impeller and diffuser can be stream isely sliced into 1/6 and 1/8 as shown in Fig. (b) and (©). Similar configurations have been ‘numerically implemented on a three stage radial-type ESP to study ump two-phase performance under gas-liquid flow conditions (Zbu and Zhang, 2014, 2015), In this study, the ESP flow domain CGeometial specications of simulated DN1750 E52 compens Deseipaon ‘vaue impeller ‘Number of des) 6 Inlet ris (rm). a9 (Channel length mm) 8 Inlet blade ansle Pde 203 (ute ade ange (iy det) tea Please cite this article as: Zhu. J, et al, CFD simulation and experimental study of oil viscosity effect on multi-stage electrical submersible pump (ESP) performance. J. Petrol. Sci, Eng. (2016), http: /éx.loiorg/10.1016)}petrol.2016.07.0331 2h e b/oural of Peoeum Science and Engineering 146 (208) 735-745 @ &) ©) Fig. 2. eometie of single stage DNS. 2} impeller ages, (b) ser blades) em comprises of 7 stages which are assembled in series (see Fix. 3(a). ‘The single-stage pressure increment is obtained by calculating the pressure difference between the inlet and outlet of stage 3, while the simulated overall boosting pressure of ESP is obtained by subtracting total pressure atthe inlet of stage 1 from total pressure at the outlet of stage 7, ‘The grid quality of calculation domain is critical for CFD st mulation as it affects not only computational accuracy but also convergence efficiency when solving the fully discretized Navier- Stokes (NS) equations by numerical algorithms. Due to complex pump geometry, the generation of 3D unstructured mesh of tet- rahedrons is easier compared to structured mesh comprising of hexahedtons. itis a trade-off when selecting the proper mesh type to perform CFD simulation, The unstructured mesh mostly com- posed by non-orthogonal grids jeopardizes the regularity of data structure, which in turn compromises algorithmic accuracy and reliability. Therefore, the high-quality structured mesh generated with ANSYS Turbogrid 15 is adopted in this study. The structured hexahedral grids for simplified multistage geometry are shown in Fig. 3, including single flow passage of impeller (Fig, 3(b)) and itfuser (Fig. 3(¢). Each domain is meshed with hexahedrons with the refinement near blade surfaces. The frozen-rotor technique is used to calculate interactions between impeller and. diffuser within each stage, Impeller domains are set to rotating frame of reference, and diffuser domains are set to stationary one. This 3p assembly. approach is categorized as steady state simulation, which does not account for instantaneous effects in the flow field such as pump startup. 3.2. Governing equations t sito 2 ef sean equ sk hypothesis that treats fluid medium and motion infinitely differ- teal contin ape we fad ow dona Teta theconerston eu of ey an be oe. The mas coca ese Bev(a)=o ® where p. 7 are liquid density and velocity vector. The source in carte esos ter hee Menaaenenateh agen etch 47) 3 a where # is the stress-strain tensor given in Bq. (2). is the gravity acceleration vector, Sis external forces. For fluid flow in centrifugal + v (pitt) =—wP 4 VA) v8 +5 Please cite this article as; Zhu, J. et al, CFD simulation and experimental study of ofl viscosity effect on multi-stage electrical submersible pump (ESP) performance. J. Petrol. Sci. Eng, (2016), hp: |/dx doi. org/ 10.1016)) petrol 2016.07.034.2hu eck Journal of Petroleum Se ence and Engineering 146 (2018) 735-745 9 (b) Fig. 5. Grid generation (2 entire seve-stac. ump, S—Sevr+ Sr Scr ad Sug Fepresent the Coriolis force and Centrifugal force effects, For stationary frame of reference Seae=Sg™=0. FOF a rotating frame with constant angular velocity ().Seq~ ~ 20H V andSee= ~ pfx (Gx), where fd and 7 are angular velocity vector and position vector, respectively = vit « (vit))+(2-3a)var “The RANS equations are used in CED solver, which statistically average the turbulence fluctuations by decoupling instantaneous velocity vectors. The additional Reynolds stress terms in. RANS equations are modeled by two-equation turbulence model. Several turbulence models are available in literature including standard k- ¢ (Launder and Spalding. 1974), RNG (renormalization group) k-e (Yakhot etal, 1992), standard ko (Wilcox, 1998), BSL (baseline) k- ‘@ (Menter, 1994) and SST k-e (Mente, 1994), among, others. AS recommended by ANSYS (2015), the SST k-a two-equation tut- bulence modet is applied here due to its ability of handling se- paration flow and resolving flow very close to walls. ® 33, Numerical schemes and boundary conditions ‘Two types of interfaces are used in CFD simulation. First. the xeneral connection interface model is employed in each pair of impeller and diffuser interfaces, which is able to apply a fran change and connect non-matching grids. Totally, 13 pairs of gen- eral connection interfaces are generated. 7 of them are within stages, and the remaining 6 pairs are located at inter-stages. Sec- fond, periodic interfaces based on circumferential periodicity are set within each domain of impeller or difuser, resulting in 14 pairs of periodic interfaces, ‘The frozen-rotor algorithm is used to simulate interactions across the interfaces of impellers and diffusers. This model treats © sembly (6) impel channel (6) fer channel each component of computational domain with an individual frame of reference, while it keeps the relative orientation of these components actoss the interface fied. It requires the least amount of computational effort compared to other interface models. However, the frozen-rotor model is unable to capture transient effects atthe frame change interface due to its steady state nature. In our simulation, the axisymmetric property of ESP geometries is used by assuming periodic flow characteristics if pump working condition is stable. The streamwise-designed blades and vanes inside ESP provide additional compensation that further weakens interactions across impellerdiffuser interfaces. Thus. the frozen- rotor algorithm is used as it offers an acceptable compromise be- tween computational effort and! numerical efficiency Due to the simplified geometries of impeller and diffuser, the grids at interface are non-conformal and mismatching with dif- ferent pitch angles. In consideration of this, the GGI (general grid Interface) mesh connections are employed, which permit non- ‘matching of grids on either side of the two connected surfaces (ANSYS. 2015) Boundary conditions are specified according to the cotre- sponding experimental configurations from ESP inlet to outlet. For wetted walls, the no-slip velocity condition is imposed. As pointed ‘out by Li (2014), the wall roughness is estimated by equivalent sand-grain roughness hy. The dimensionless numberh is defined as: hy = hy ® ‘The range hf<5 corresponds to hydraulically smooth regime, ‘Schf<70 corresponds to transition regime, and hf>70 for hy- draulically rough regime, The standard wall function holds only within hydraulically smooth regime. Thus, special attention needs toe paid to near wall treatment in turbulence model with low-Re Please cite this article as: Zhu, J, et al, CFD simulation and) submersible pump (ESP) performance. J Petrol. Sci, Eng. (2016), ‘experimental study of oil viscosity effect on multi-stage electrical |tp:{/éxdoi.org/101016/j petrol 2016.07.03mo 12h /Joral of Peseu Sciece and Enineeing 146 (201) 735-745, flow and rough walls. In this study, the near-wall treatment is automatic wall functions for omega-based turbulence models. It automatically switches from standard wall-functions to 2 low-Re near wall formulation as the mesh is refined. We apply a simple correlation of hs with arithmetic average of absolute values of real roughness Ra by hg=6 Ra (Li, 2014). For a cast wallRa=125— 50 ym. Us 30 jm, one can obtain fy=300 ym, which is close to 250 ym, a sand equivalent roughness of the natural surface of cast ion recommended by Patankar and Spalding (1872) ‘The total pressure with zero gradient flow direction and tur- Dulence intensity is set at 1st stage inlet. A mass flow rate scaling down to 1/8 of inner cross-sectional area is imposed at 7th stage outlet. This is a more robust configuration of boundary conditions for numerical convergence as recommended in ANSYS (2015), especially for part-load CFD simulation (Stel et al, 2015), Al simulation cases are conducted with the same numerical schemes, For discretization of advection terms and turbulence ‘equations in space, the high resolution scheme is used, whieh is a second order algorithm in ANSYS CFX-solver. For steady-state si- mulation, a false time step as a means of under-relaxing governing ‘equations is applied, which requires a relatively large time scale due to robust and fully implicit CFX-solver. A fixed physical timescale of 1/(22) is used with maximum 500 outer loop iterations to achieve convergence. The convergence criterion Is satisfied if RMS (root mean square) residual drops below 10~*, 4, Results and discussion In this section, the numerically sinmulated ESP boosting pres- sures are presented and compared with experiment results under different flow conditions. At frst, CFD simulations is compared with experimental results for water flow to validate numerical methodology. Then, experimental conditions of viscous oil flows are incorporated into numerical simulations as inputs. The outputs from CFD-post include pump pressure increment, streamline pressure and velocity fields etc. Four rotary speeds and four oil viscosities are used to conduct experiments and numerical simu- lations, namely, 3500, 3000, 2500, 2000 rpm and 56, 98, 180, 220 op. 44. Mesh and turbulence model validation ‘The mesh quality depends on the dimensionless distance (y*) at the first grid point near the wall, According to boundary layer ‘theory, the viscous sublayer exists in the near-wall region, Within viscous sublayer, the dimensionless velocity (u~) isa logarithmic function of y~ away from the wall. This is also referred to as ' —" dss oe 300 ey yt | mi i |" — | Fig. 4 Mesh validation and wal fnetion check Se E Howat be standard wall-function, which holds for y* < 100. Thus, the first atid layer should be sufficiently fine to meet wall function requirement: ‘The mesh number is counted on single-stage simplified fuid domains of impeller and diffuser. As shown in Fig. 4 the simulated stage pressure increment becomes constant when the grid num- ber reaches about 0.2 milion, where the average value of y" on blade surface is below 20. Therefore, the grids used for simplified impeller and diffuser sections consist of 158,976 and 124295 elements, respectively. Total grids for the complete seven-stage computational domain cantain 1,982,904 elements, which are sufficient to guarantee grid independence, Fig, 5 shows the effect of turbulence models on sinmulated stage pressure increment and comparison with corresponding. experi- ‘mental results under water flow. The selection of turbulence ‘model is a delicate task for CFD simulation, which is also a com- promise of computational effort and numerical accuracy. As it can be seen in Fig. 5, there is no prominent variance among different turbulence model predictions. However, the aforementioned SST turbulence model is used due to its advantages in capturing shear flow spreading and separation at low Reynolés number flow. 42, Comparison with catalog curves For comparison of simulated ESP boosting pressure with ex: perimental data, the dimensionless variables: flow coefficient, head coefficient, and hyéraulic efficiency are defined by Eqs. (5)= a Flow coefficient: 90009200000055 oo ses wl o sms : Please cite this article asi Zhu, J. et al, CFD sinvulation and experimental study of oll viscosity effect on multi-stage electrical submersible pump (ESP) performance. J. Petrol. Sci. Eng, (2016), hp: |/dx doi. org/ 10.1016)) petrol 2016.07.031.2hu el Journal of Petroleum Scene and Eninerg 16 (2018) 725-745 4 8s 85 roo oso ge Pix ge z He x z Es Bs é 5 234 ou al 3 E17 PP SCoce iv ° a ° S00 3000 2000 a 5001000 —«a500—~—«2000 Flow rate ps Flow rate od) (@) (by 0 <0 1s Pressure increment (ps Pressure increment (p » as ° ° = BO Hm © Head coeticient: vet DP 6 Hydraulic efficiency our Ta a Where Tis shaft torque, and & is ESP rotary speed. ig G shows the comparisons of single-phase simulation results of head coefficient (y), pump efficiency (7) a8 function of flow coefcints (6) with the catalog curves within the pump operation range. A good agreement can be seen for y-versus i. However, the simulation results for are slightly higher than the catalog curve. indicating that CFD simulation over predicts ESP single-phase ef- ficiency. This may be due to the neglect of leakage fow through the radial clearance between impeller and éifluser, which causes additional boosting pressure loss in reality. Meanwhile, the smooth wall assumption also contributes to the deviation by un- devestimating wall shea stresses. 43. Comparison with experimental data ‘ig. 7 shows the comparison between experimental results and. ‘numerical simulations for ESP overall pressure increment over 7 stages. The experimental measurements were taken with the éifferental pressure transducer spanning from the inlet of the Ist oa 800 Fow rate (Opa) @ 1200 16007000 impeller to the outlet of the 7th diffuser, The corresponding nu- ‘merically simulated pressure increment is calculated as + Hf ee ) 2 ® where Pag is static pressure, C is absolute velocity given by C= W + 0 according to velocity triangle. W and 0 are relative and peripheral velocities, respectively. Subscripts I, 2 indicates inlet dnd outet. in Fig. 7, the solid curves represent experimental messure- iments of ESP overall pressure increment. while the symbols re- present CFD simlation results. A good agreement canbe found in the comparison for water case, However, numerical simulation over-predits ESP overall boosting pressure for viscous els with an értor around 15%. The prediction of ESP overall pressure increment by CRD simulation at lower rotary speeds (2000 an 2500 rpm) is better than at higher rotary speeds (3000 and 3500 rpm). The trends of pressure increment versus low rates are wel captured by numerical simulations. As oil viscosity increases, the overall pressure increment decreases. At higher oil viscosity and lower flow rate, the trend becomes more linear. This is due t the fow regime change inside ESP from turbulent t lamina. in Fig, 8. The horizontal and vertical coordinates stand for ex- perimental and CFD simulation results, respectively, As can be Seen, numerical simulation predicts ESP single-stage pressure in- crenment with an error below ISX compared with experimental results. The deviation may be partially due to pump rusty and ‘worn conditions after years of wage Please cite this article as: Zhu. J, et al, CFD simulation and experimental study of oil viscosity effect on multi-stage electrical submersible pump (ESP) performance. J. Petrol. Sci, Eng. (2016), http: /éx.loiorg/10.1016)}petrol.2016.07.0332 $9 $9 Be Be (a) (b) qe go Be 6 Become | & 3 ol o © (@d) 2 3 g3 Bones ° Experimental ests (ps) © ig. 8. Comparison of CE simulated ESP pesformances of tage 3 wth experimental resus for diferent lui vise, a} water, (B56 [98.180 «9, (6) 220 44, Analysis of low and pressure flelds Fig, 9 shows the streamline plots under different flow condi- tions inside the 3rd stage. The recirculation that contributes to the pump hydraulic oss to a great extent can be seen near the trailing edges of impeller blades. Two hydraulic factors affect the re- circulation flow from the comparison in Fig. 9. First, due to high viscosity, the flow regime shifts from turbulent flow to laminar flow, causing changes of flow recirculation inside impeller (Fis. 9 (by, (A), (D) Second, under off-design operation conditions, the fluid velocities at the outlet of impeller deviate from blade angle increasingly. Tis in turn leads to additional departure of stream- lines from the designed flow path. Therefore. further recirculation and pressure potential dissipation is induced (Fig, 9(a) vs. (b), (ows (@). For the simulated cases shown in Fig 9, the vortices exist near the pressure sides of diffuser vane due to highly twisted vane geometry. The vortex shape inside diffuser channel is affected by several [actors including viscosities, rotary speeds and liquid flow rates, However, this vortex and recirculation contribute little to ESP pressure increment. ESP boosts pressure by converting kinetic energy to pressure Please cite this article asi Zhu, J. et al, CFD sinvulation and experimental study of oll viscosity effect on multi-stage electrical submersible pump (ESP) performance. J. Petrol. Sci. Eng, (2016), hp: |/dx doi. org/ 10.1016)) petrol 2016.07.034.2 eal Journal of Petroleum Scene and Engineering 146 (2018) 735-745 m8 Case 1: 3500 xpm, 0.40er = Water Case 2: 2000 rpm, 0.25Qser 56 0p 20ep ‘ig. 9. Stcamline comparizon under dierent few conditions a half span of sage 3 potential. The impeller-difuser interaction guides fluid away from impeller, which brings in additional energy dissipation and local hhydraulic pressure loss, resulting in abrupt pressure reduction. Fig, 10 shows the averaged total pressure along streamwise loca- tion. Fg, 10(a) presents the total pressure calculated forthe entire 7 stages at 3500 rpm and 0.4Qyep, and Fig. 10(b) shows the total pressure taken across stage 3 at 2000 rpm, 0.25Qsee, The stream- Wise location is the dimensionless distance from the inlet to the futlet. I ranges from 0 to 1 forthe fist stage, 1-2 for the second fone, and so on. As compared in Fig, 10(a), ESP overall boosting pressure suffers from degradation increasingly as the fluid be- comes more viscous. From inlet of stage 1 to the outlet of stage 7, the fluid pressure is boosted within the impeller of each stage. However, the total pressure does not change much in diffusers. Comparison of ESP single-stage pressure increments for di ferent fluid viscosities is presented in Fig. 10{b). The impeller fuser interaction causes a pressure drop which is captured by ‘numerical simulation. The pressure drop appears to be the same (about 5 psi) for different fluid viscosities. Inside impeller, the pressure in suction section is flat corresponding to streamwise location from 2.0 ~ 22 in Fig. 10(b) Based on above discussions, ESP boosting pressure is affected, by pump geometry and fluid properties. In order to improve pump hhydraulic efficiency, the blade angles at impeller outlet and if- fuser inlet need to be aligned around operation flow rate so that the pressure drop due to impeller-diffuser interaction can be ‘minimized (Wu et al, 2015). As Fig. 10(b) shows, the suction sec- tion of ESP impeller contributes little to pressure boosting, Thus. the suction section can be made as short as possible. Fluid visc- osity has a significant effect on ESP boosting pressure and eff- ciency due to the increased wall shear stresses. 5. Conclusions In this paper, the oll viscosity effect on ESP boosting pressure is, investigated through experimental testing. and CFD simulations. Please cite is article as: Zhu, J, et al, CFD simulation and experimental study of oll viscosity effect on multistage electrical submersible pump (ESP) performance. J. Petrol. Sci, Eng. (2016), http: /éx.loiorg/10.1016)}petrol.2016.07.033ma 12h eb Jornal of Peseun Science and Engineering 146 (2016) 735-745, ‘Sane Sage? Stages Saget Stages Sage Si pt Totaloressire of? age (st (@) Case 1: 3500 pm, 0405 Inpeter Dire a 5 ; a Intersting of Sate x impeller ditser #5609 (©) Case 2: 2000 pm, 0.250 pep Fig. 10, Tol pressure averaged along streannvise locaton, (2 oa pressure of 1 Sages, (0) ttl presse of stage 3. Based on the analyses, several conclusions can be drawn: 1) Using SST turbulence model, the simulated ESP pressure in- crement under water flow matches well with experimental results, which validates the numerical methodology. ) Experimental tests show that ESP boosting pressure suffers from deterioration with the increase of fluid viscosity. At higher ‘uid viscosities, the H-Q curves become more linea, indicating ow regime transition from turbulent flow to laminar flow, 3) ESP pressure increment under viscous fluid flow is over pre- dicted by CFD simulation about 15%. The linear decline trend on HQ plots due to high liquid viscosity can be captured by nu- ‘merical simulation, ) Recirculation flow at the trailing edges of impeller blades occurs if duid viscosity increases or liquid flow rate decreases further, 5) The experimental and numerical approaches are proven to be effective to study viscosity eflects on ESP boosting pressure, A better understanding of recirculation, shear flow spreading and separation can help develop a more accurate mechanistic model for predicting ESP performance under viscous fluid flow. Acknowledgments ‘The authors appreciate the technical and financial support of the Tulsa University Attifical Lift Projects (TUALP) member companies. Appendix A. Error ana The experimental error originates from instrument measure- ment errors, including pressure transducers, flowmeters, and temperature sensors etc. For differential pressure measurement, the instrument accuracy is + 0.25%. Thus, the uncertainty for differential pressure measurement is around + 025%, ‘CED simulation errors contain modeling error and numerical feror, Modeling errors, originating from mathematical re- presentation of physical problem, are usually negligible in CFD simulation error analysis compared with numerical errors (Stern et al, 2001), which are due to numerical solution of the mathe- ‘matical equations. In this study, the Richardson extrapolation (RE) is employed to analyze numerical error caused by coarse grids (Wilson et al, 2001), Suppose coarse, medium and fine grids corresponding to nu- ‘merical solutions: 5y, 5; and 5, respectively. Variances between medium-fine «;=5;- 5, and coarse-medium o)~5,- 8) are used to define the convergence ratio Ro ta ay R corresponds to three different convergence conditions. 0=R=1 is monotonic convergence condition. Simulation un- cercainties can be analyzed by generalized RE. R-<0 is oscillatory convergence condition, The oscillation maximum/minimum boundary is adopted to quantify numerical errors. R> 1 corres- ponds to divergence, whose errors and uncertainties cannot be estimated, Based on RE, the numerical solution with the first n terms of, series expansion can be expressed as: s cto aS + Saws ya's we) where Asis the thickness of gi layer, pis the order of accuracy, is series coefficient, For simplicity, n=1 is analyzed in this Study, Thus, can be estimated by. In(enfen) TH wy 4) Here, ris the ratio of gr layer thickness, To account for effects of higher-order terms and provide a quantitative metric to de- termine proximity of the solutions to the asymptotic range, the ‘multiplication factor Cis introduced, as where pox is an estimate for the limiting order of accuracy. C ap- proaches 1 as spacing size goes to zero and asymptotic range is reached. Thus, the numerical uncertainty due to grid size can be obtained from: eds as) From Fig. 4, an estimation of aumerical error by Eq. (A-6) based fon grid number of 62.755, 143,440 and 201,833 is below 3X. As ‘mentioned above, the grids used in this study for performing CFD simulation contain 283.272 elements, which ate sulficient to en- ssure numerical accuracy, Please cite this article as: Zi, J. et al, CFD sinvulation and experimental study of oll viscosity effect on multi-stage electrical submersible pump (ESP) performance. J. Petrol. Sci. Eng, (2016), htt: |/dx doi org! 10.1016)) petrol 2016.07.034.2 eal Journal of Petroleum Scene and Engineering 18 (2018) 735-745 5 References ‘Amaral. Btevam,V. Panga, FA, 2008. On the influence of wisest on ESP peiornance SPE Pied. Opes 242) 305-10 -ANSYS, 2015, ANSYS Academie Resear Release 15, Help Sistem, FX Dots reatatian. ANSYS Ine Banja. Gamboa. Zhang, H-2. 2013, Expesimentl study of iu viscosity fect on ewphase cage performance of elec submersible pumps. SPE ‘Annual Techical Conference and Exibition, SPE 155374, Scie of Ftcleu, Engieets Bates, Lj Scat. 1, Sheth KK, 2012 FSP technology maturation: subsea ‘oorting syem with high GOR and veo fn: SPE Anil Teh Confecence and txhibiten PF 15986, society of Ferleur Fines. Caridad. J, uae M_RenyeyFTeante A Aullen, 0.2008, Chtacterization fa cesta pnp ingen wep aw ons Pee ‘roce , 2014 Sad of Oil Wter Raw and Emulsion Formation n Electric Submersible Purps, MS Thesis, the Univers of Tusa, Tulsa, OK Gonzalez | Santora. c. 2005, Unsteady few structure ae gba vals in 2 ceatiugn pup. Huis ag 128 (3) 937-346 Gontales, | Femancez | Bano, €. Sanlu, C. 2002 Numesel slain of ‘he dante effets duet mpellervlute eration in ena pump us Eng. 124(2), 348-355 ‘oleh 1. 1998. Pumping highly Wscous Duds wit ceenfugal pumps Part 1 "Worl Pumps 19283 (399, 30-34 Colic, 1 1998, Pumping highly wscous Mus with cetnugal pumps = Part 2 Woe Pups 1889 (395) 39-12 Hyeratite nite 1946, Tentative Standards of Hau Inte: Chat fr the Determination of Pup Performance When Handing Viscous Lig The n= Ippen AT 104. Te nduence af Visco 0 Centetugal Pump Pesformance, sue 159 of iz Engineering Laboratory Report, ASME Pape: No. A-5-57 Launder BE Spaling, DB 1978 The name computation a belt flows ‘Comput Methods AppL Mech Eng 32} 268-288 1, Wc 2002 experimental investigation of psfcrance of commercial en- fal ol pump. Word Punpe 2002 429) 26-2, LU, We Hote Mechs fr onset of suddea sing head etc a cenutugal ump when Bending vscous os} Fis Eng 136 (7) 78501 enter £8, 1988 Twa-equatan edd -isastytrbuleee model fr engineering "ppations. AAA] 32 (3) 1558-1605 aera K Uj, T1995. Thee-dinensional clean of water swe hase low i cenrfugl pump inp based a8 bubiy How tedel) ule Fatanar SV. Spalding DB. 1972, cleltonprocedire fr best and mass "Caster in thre ehmensonal parable ows ne) Heat Mas Tran 15 (10) (9 X. Turneust N.Ghasipor F, 2012 Advanced eee submersible pup “sign tal for getheralppictions. GRC Trans 2012 (36) 5-546 ‘Shojaetard, MH, Boyahch FA. Phghagh MB, 2006 Experimental stay and areeedimensionl pumercl Tow simulation in acenfugal pump when handling viscous fide UST ft. J Rg Se 7 (2) 55-60, Shajaeetaray MH, Tahar. M, Ehghaghi MB. alahian, MAA, Bela M. 2012 Nurmi stud) of the eet of some geomet carats of centuga trump impeller tha puny «viscous fd Comput Pus 2012 (60), 61-70, sinner, Marlen RE, 2073 Neri ay ofthe infence of wc ‘on the performance ofan elec submeruble pump. ns ASME 2013 Fads Enginerng Dviion Summes Meeting American Society af Mechanical Enginees. Selane EA. 2008, Viscous Etfects onthe Performance of Electro Submersible Pumps (ESP) MS Theva the Universi of Tea Tl, O ‘el H, Sin, 1 Ponce), Chiva 5. Morsles REM 20, Numerical investiga ‘uo othe Now ina multstageelecine submersible pump) Pet 8. Eng 136, a138 ‘Ste H, Sin, T, Prohmann PR. Ponce, . chia. Morales, RE. 20:4, CFD in- ‘esugauon ofthe eet of vscasty ona teeta led submelle pump In ASME 2014 th Joint US-turogean Fide Engineering Division Summer Meeting collected with the ASME 2014124 Internal Conference om Ninochannets (Micrechanvel, and Minicannel) American Scie of Me- hanel Engmneers. seerm Filton. Ri Coleman, HW, Paterson. EG, 2001, Comprenensive ap- isi to veidiation and vabation of CD snilains—Care methodology End proces } iis Eg. 125 (795-802, Seepano Al. 1957 Centura an Aa How Pumps Theory, Design and Ap lean, second dion. joa Wey & Sons, ‘atte C2608 flees! Scbmerse Ppr Marsal: De ‘Maintenance. Gulf Peftssonal Fubishing. MA ‘eeisan Pago, Me 2011 Experimental seston f the viscous effect 2 "orsphaseiow patterns and hydraulic performance of elect sbmerile puumpe ] can Pee Techn 30 (4 4-32. ‘turte, 2 Takacs, C. Zig], 2000 Equa comec ental pump aves for “eesti Gas) 98 22} 97-6 ‘wieex. DC. 1988 Turbulence modeling for CFD, 2nd eition, DCW Indus In ta cabada A ‘widen RV Ste, Coleman, RW, Paterson EG, 2001. Compreensie a> roach overifestion and vahation of CFD simlatons—Par :sppieston for FANS sumulason of» czgojontaner sup. lds Eng. 123 (4, 802-810, ‘wl. Yan F Chen X, Wu Yang 5, 2015 filet of waling-edge modeation a raedsiw pomp. ids Eng. 13710), 101208, VabbotV Oreag SA. thangam, 5. Gash. 13. Spezia. CG. 1982, Development ‘of turbulence models for shat lows by a doube expansion tecnigue Piss. Fis AP ym (71510-1520 ‘hang, Sh W. Xu, Cao, X. Wang C. Ua W.Ma.D, 2015.4 new propose elu guide vane ft compact mulistage centugal psp ni) Rotating Macy 2003. 683-73 ‘hu Zhang. 4-0, 204. CFD simulation of EP pectrmance and bubble size ‘malo under gusy condos. SPE Annual Tehseal Coeretee ad Exhibition sPe 170727 Society of Petroleum Engineers, ‘hu. J Zhang H-0, 2015 "oceling of tas Bubble Se In Sect Submersible inn tS ai hme Sinn (te Medion oe pron, Please cite this article as: Zhu, al, CFD simulation and experimental study of oil viscosity effect on multistage electrical submersible pump (ESP) performance. J. Petrol. Sci, Eng. (2016), http: /éx.loiorg/10.1016)}petrol.2016.07.033