Professional Documents
Culture Documents
net/publication/336888964
A Numerical Study on Erosion Model Selection and Effect of Pump Type and
Sand Characters in Electrical Submersible Pumps by Sandy Flow
Article in Journal of Energy Resources Technology, Transactions of the ASME · October 2019
DOI: 10.1115/1.4044941]
CITATIONS READS
7 580
6 authors, including:
Risa Rutter
Baker Hughes inc
14 PUBLICATIONS 55 CITATIONS
SEE PROFILE
Some of the authors of this publication are also working on these related projects:
All content following this page was uploaded by Haiwen Zhu on 25 November 2021.
Journal of Energy Resources Technology Copyright © 2019 by ASME DECEMBER 2019, Vol. 141 / 122004-1
Fig. 1 Components of ESP impeller and diffuser in meridional views: (a) impeller and (b) diffuser
From Levy and Chik’s study [8], the hardness of erodent particles K2 K3
VP dP
no longer influences erosion rate on AISI 1020 steel, Brinell hard- E90 = K(Hv) K1
(7)
ness of 119–235 kgf/mm2, when the solids’ Vickers hardness is V* d*
higher than 700 (SiO2). The hardness of produced particles is
usually close to that of SiO2, and the tested pump is made of type n1 = S1 (Hv)q1 (8)
1 Ni-resist iron, with Brinell hardness from 131 to 183 kgf/mm2.
Therefore, six models without considering the particle hardness n2 = S2 (Hv)q2 (9)
are selected for CFD simulation. The following equations are
changed to SI units for convenience.
Ahlert’s erosion model [9] was established by TUE/CR/C and K2 = 2.3(Hv)0.038 (10)
has been used in the ANSYS manual as a reference. The coefficients
are listed in Table 1, where BH is Brinell hardness in kg/mm. F(θ) = [sin(θ)]n1 {1 + Hv[1 − sin(θ)]}n2 (11)
ER = KVPn F(θ) (2) where V* and d* are the reference velocity and the reference particle
diameter, which is 104 m/s and 326 µm for SiO2-1, respectively, ρt is
the target material density (kg/m3), and Hv is Vicker’s hardness
aθ2 + bθ, θ ≤ θ0
F(θ) = (3) (GPa). The empirical coefficients are shown in Table 3.
x cos2 θ sin(wθ) + y sin2 θ + z, θ > θ0 Zhang derived his erosion equation from Ahlert’s study with a
polynomial impact angle function [12] and added a particle shape
The Haugen et al. erosion model [10] has a similar formula, but a coefficient Fs, which is 1.0 for sharp (angular), 0.53 for semi-
different particle impact angle function that is shown below: rounded, or 0.2 for fully rounded sand particles. The value of the
polynomial coefficient Ai is shown in Table 4.
ER = K × 10−9 VPn F(θ) (4)
ER = 2.17 × 10−7 (BH)−0.59 Fs VP 2.41 F(θ) (12)
8
θπ i
5
θπ i
F(θ) = Ai (5) F(θ) = Ai (13)
i=1
180 180
i=1
Table 1 Empirical coefficient in Ahlert et al.’s erosion model F(θ) = A[sin(θ)]n1 {1 + Hvn3 [1 − sin(θ)]}n2 (15)
Table 2 Empirical coefficient in Haugen et al.’s model Table 5 Empirical coefficients in Mansouri’s model
A1 A2 A3 A4 A5 A6 A7 A8 A N n1 n2 n3 C
9.37 −42.295 110.864 −175.804 170.137 −98.298 31.211 −4.17 0.6536 2.41 0.15 0.85 0.65 4.62 × 10−07
Numerical Methodology
Fig. 3 Pump head and efficiency
Simulations in this study were conducted using commercial CFD
software—ANSYS 17.2 FLUENT. High-quality meshes were gener-
ated using ICEM and TURBOGRID. The steady-state single-phase
sample collected during the test, the solid concentration before the
water flow was solved with κ-epsilon equations and scalable wall
pump intake varied from 1% to 1.2% by weight. The average
function model by employing the frozen-rotor technique in three
weight of impellers and diffusers and radius of skirts, balance rings,
ESPs (MTESP, TE2700, and DN1750) [32]. Particle trajectories
inter-stages, and hubs were measured and shown in Table 6. According
and erosion rates were calculated using the DPM model. Since
to Fig. 1, radius changes are in the secondary flow region (seals region,
the inlet mass flow profile deviates from the real profile, two
balance holes, and inter-stages), which was damaged due to abrasion
stages were simulated, and only the results of the second stage
(particles between two surfaces). Erosion in the primary flow
were used in order to eliminate the inlet effect. The monitoring
channel (created by impeller blades and diffuser vanes) is of interest
faces for pressure and mass flow rate were set at the impeller inlet
in this study. Therefore, the abrasion weight loss was roughly calcu-
and diffuser outlet of each stage to check the convergence. The
lated from the measured radius changes and subtracted from the total
default residual criterion is 10−3, which offers a good compromise
weight loss. According to geometry in ANSYS FLUENT, the surface
between the computational cost and numerical accuracy. The simu-
area of the primary flow channel in the diffuser is 0.028 m2, while
lated ESP head curves agree well with the pump catalog curves.
that of the impeller is 0.015 m2 [30]. Then, the area-weighted-averaged
erosion rates on the diffuser and impeller were calculated to be 1.83 ×
10−6 kg/m2 s and 9.19 × 10−7 kg/m2 s. Geometry and Meshing. The 4-in. MTESP is the mixed-type
centrifugal pump with a specific speed of Ns = 2000. There are
six blades in the impeller and eight vanes in the diffuser. The
Sand Properties. Commercial hydraulic sand particles from a
pump is designed to resist solid particle wear within the liquid
local company were used in the test to mimic sand produced from
flow range of 0.002–0.006 m3/s and a rotational speed of
proppant backflow. The sand was replaced every 2 h in the 64-h
3500 rpm. TE2700 is a 5-in. radial-type pump, while DN1750 is
erosion test to keep the sand properties [31]. Scanning electron
another 4-in. mixed-type ESP. The detailed pump information can
microscopy (SEM) analysis and advanced first-trimester screening
be found in Table 7.
(AFS) test before and after the 2 h of recirculation are shown in
The optimum grid number was obtained by comparing pressure
Fig. 4. As can be seen, solids were damaged in the test, but the
increment and hydraulic efficiency results from one-stage
sharpness remained to unity. The AFS number was 106 at the begin-
MTESP. As shown in Fig. 7(a), both parameters become stable
ning and increased to 195 in the end. Therefore, the averaged par-
when the grid number reaches 1.5 × 106. Therefore, the mesh of
ticle size was assumed to be 100 µm. Stages were coated to
MTESP has 1.863 × 106 grids per stage as shown in Fig. 6(b).
investigate the erosion pattern inside the pump.
Thereafter, DN1750 and TE2700 were meshed in the same way
with the quality above 0.3 and near-wall grid refinement. The
Paint-Removal Photo. In order to capture the detailed erosion mesh with a grid number around 1.5 × 106 per stage was generated
pattern and detect the most severe erosion location, the coated as shown in Figs. 6(a) and 6(c).
Impeller weight (kg) 3.15 × 10−01 3.14 × 10−01 3.13 × 10−01 3.12 × 10−01 3.09 × 10−01 6.26 × 10−03 9.19 × 10−07
Impeller balance ring OD (m) 6.27 × 10−02 6.26 × 10−02 6.26 × 10−02 6.25 × 10−02 6.24 × 10−02 4.78 × 10−04
Impeller skirt ring OD (m) 5.23 × 10−02 5.22 × 10−02 5.22 × 10−02 5.21 × 10−02 5.20 × 10−02 4.22 × 10−04
Impeller-outside shroud OD (m) 7.88 × 10−02 7.88 × 10−02 7.88 × 10−02 7.88 × 10−02 7.88 × 10−02 1.18 × 10−04
Impeller hub ID (m) 1.75 × 10−02 1.75 × 10−02 1.75 × 10−02 1.75 × 10−02 1.75 × 10−02 4.01 × 10−05
Impeller hub OD (m) 2.22 × 10−02 2.22 × 10−02 2.22 × 10−02 2.21 × 10−02 2.21 × 10−02 1.38 × 10−04
Diffuser weight (kg) 1.12 × 10+00 1.11 × 10+00 1.11 × 10+00 1.11 × 10+00 1.10 × 10+00 1.36 × 10−02 1.83 × 10−06
Diffuser balance ring ID (m) 6.30 × 10−02 6.31 × 10−02 6.32 × 10−02 6.32 × 10−02 6.33 × 10−02 5.21 × 10−04
Diffuser skirt ring ID (m) 5.27 × 10−02 5.28 × 10−02 5.28 × 10−02 5.29 × 10−02 5.29 × 10−02 3.47 × 10−04
Diffuser-outside shroud ID (m) 7.98 × 10−02 7.99 × 10−02 7.99 × 10−02 7.99 × 10−02 7.99 × 10−02 1.64 × 10−04
Diffuser skirt depth (m) 1.25 × 10−02 1.25 × 10−02 1.25 × 10−02 1.25 × 10−02 1.25 × 10−02 1.34 × 10−06
Diffuser bore ID (m) 2.25 × 10−02 2.25 × 10−02 2.26 × 10−02 2.26 × 10−02 2.28 × 10−02 2.08 × 10−03
Fig. 5 Paint-removal view: (a) impeller original, (b) impeller 2 h water, (c) impeller 2 h slurry, (d) impeller 6 h slurry, (e) diffuser
original, ( f ) diffuser 2 h water, (g) diffuser 2 h slurry, and (h) diffuser 6 h slurry
Then, a three-stage geometry of MTESP was created to investi- separately. The calculated head curves of the second stage of
gate inlet and outlet effects [17]. The boosting pressure accuracy of both pumps in Fig. 7(b) show a good agreement with catalog
each stage was compared with catalog curves in Fig. 7(b). As can data, which verifies the validity of the adopted methodology in
be seen, the calculated head deviates from catalog data at the first this study.
stage, but the difference disappears at the second stage. In addi-
tion, the area-weighted-average erosion rate of the first stage is Boundary Conditions and Numerical Scheme. In this study,
almost 50% lower than that of the second and third stages in the boundary conditions of the parametric studies were set as Q =
Figs. 7(c) and 7(d ). Due to the comparable simulation results at 3100 bpd, N = 3500 rpm, and the outlet pressure 220 psig, which
the two stages downstream, the two-stage geometry was selected are similar to the test conditions. Sand properties were set according
in order to save the computational cost. Using the same metho- to correspondent test samples. The frozen-rotor method was used in
dology, the two-stage computational domains of TE2700 and the simulation, which applied an individual frame of reference to
DN1750 were generated by ICEM and TURBOGRID software each component in the computational domains. By using this
Pump Qbep (m3/s) Head (m) RMP Sand (wt%) Blade Vane OD (m) Ns Pump type
−3
MTESP 5.7 × 10 7.38 3500 1 6 8 0.1016 2000 Mixed
TE2700 5.5 × 10−3 16.17 3500 1 5 9 0.127 1100 Radial
DN1750 3.3 × 10−3 5.62 3500 1 6 8 0.1016 1900 Mixed
Fig. 7 Mesh validation: (a) pressure increment and hydraulic efficiency versus grid number, (b) head increment
error analysis, (c) three-stage MTESP area-weighted-average erosion rate in the impeller, and (d) three-stage
MTESP area-weighted-average erosion rate in the diffuser
method, the relative orientation can be kept at the interface. This Sand particles with a density of 2637 kg/m3 were used to
method requires the least amount of computational effort compared compare erosion models in three pumps. The Rosin-rambler
to other interface models [33,34]. In the cell zone, the properties of method with a mean diameter of 100 µm, the maximum diameter
the diffuser cell were kept at the default, while the frame motion was of 200 µm, the minimum diameter of 50 µm, a spread parameter
selected for impeller cells. The rotation-axis direction was set to of 3.5, and the number of diameters of 10 was used. The Discrete
global z-axis, and the rotational velocity was 3500 rpm. The mesh Random Walk Model and Random Eddy Lifetime Model were
interfaces were created at the connections between rotating and sta- enabled to analyze the particle random movement. Parametric
tionary parts. studies on particle diameter and density were simulated with the
Model Ahlert et al. (1994) Haugen et al. (1995) Oka et al. (2005) Zhang et al. (2007) Mansouri (2014) DNV (2015)
−8 −9 −9 −8 −8
K 7.3 × 10 2 × 10 6.23 × 10 1.02 × 10 2.17 × 10 2 × 10−9
n 1.73 2.6 2.35 2.41 2.41 2.6
dup g(ρp − ρ)
= FD (u − up ) + +F (19)
dt ρp
Fig. 8 Impact angle function of six erosion models The particle trajectories are identical with the same boundary con-
ditions, fluid properties, and sand properties. Therefore, the
impact angle and the velocity are also identical with the same
inlet mass flow rate. Different erosion rates and patterns result
Fig. 9 MTESP impeller erosion contour: (a) Haugen (1995), (b) DNV (2015), (c) Zhang (2007), (d) Oka (2004),
(e) Mansouri (2014), and ( f ) Ahlert (1994)
Fig. 10 MTESP diffuser erosion contour: (a) Haugen (1995), (b) DNV (2015), (c) Zhang (2007), (d) Oka (2004), (e) Mansouri (2014),
and (f ) Ahlert (1994)
Fig. 12 Area-weighted-averaged erosion rate comparison: (a) diffuser and (b) impeller
validated by Figs. 9 and 10. Therefore, the proper erosion model is Although these models were derived from the air impingement
selected based on the weight loss test. erosion test, some models surprisingly perform well in the ESP
As mentioned in the test result section, the area-weighted-averaged under liquid–sand flow. As seen in Fig. 13, the Oka erosion model
erosion rates on impeller and diffuser were 9.19 × 10−7 kg/m2s and offers the best agreement against experimental measurement on the
1.83 × 10−6 kg/m2s. Figures 12 and 13 show the comparison of impeller with a 37% error, while the Haugen et al. model predicts
area-weighted-averaged erosion rate among six erosion models. best on the diffuser with −14% error. The average error of impeller
Fig. 14 Impeller and diffuser erosion contour using Oka et al.’s model: (a) MTESP impeller, (b) TE2700 impeller, (c) DN1750 impel-
ler, (d) MTESP diffuser, (e) TE2700 diffuser, and ( f ) DN1750 diffuser
0 deg to 5 deg for two mixed pumps (MTESP and DN1750) and reduce the erosion on the impeller-outside shroud surface and
0 deg–10 deg for the radial pump (TE2700). The maximum increase that on the impeller-inside shroud surface. The difference
impact angle is around 14 deg for two mixed pumps and 25 deg can be caused by the significant mass change of a single particle.
for the radial pump. Similarly, in Figs. 15(b) and 15(c), the In a mixed-type ESP impeller primary flow channel, the drag
impact velocity and hit times of TE2700 are much higher. As a force from fluid helps protect the impeller-inside shroud. For
result, the erosion rate is much higher on TE2700 than the others large particles, the drag force is no longer strong enough to elimi-
as shown in Fig. 15. nate the inertia of the solid. In the diffuser, the outside shroud
surface is always eroded more from solids and erosion rate increases
and becomes more concentrated with larger particles. The facet
Particle Diameter Effect. Tilly [39] proposed that the erosion maximum erosion rates and area-weighted-average erosion rates
rate increases with the particle size before a more flat and stable shown in Figs. 17(a) and 17(b) illustrate that the larger particles
relationship is reached. Three-phase erosion tests conducted by cause more wear when the sand diameter is smaller than 300 µm.
Kesana et al. [40] also show that the erosion rate does not always Above 300 µm, the changes depend on the selected surfaces.
increase with the particle diameter in an elbow. With the larger
size, the grains possess more inertia to resist the flow direction
changes and higher impact velocities when colliding the target sur- Particle Density Effect. Normally, the particle density is
faces. On the contrary, the particle number decreases with the closely related to its hardness. The density will affect the trajec-
increase in the particle diameter. The erosion phenomenon tory, while the effect of particle hardness is summarized as an
becomes more complicated with complex geometry, i.e., ESPs empirical factor in most erosion equations. However, particle hard-
[41]. Therefore, the particle size was changed from 50 µm to ness may take less effect in downhole situations as mentioned in
1000 µm to investigate its effect. the Introduction section [8]. Therefore, only particle density was
Comparing erosion patterns as displayed in Fig. 16, it is convinc- analyzed. Although normally produced sand particles only have
ing that erosion rate increases on the impeller-outside shroud a density from 1500 kg/m3 to 3000 kg/m3, density of minerals
surface with the particle size increases from 50 µm to 200 µm. can vary from 3000 kg/m3 to 6000 kg/m3 or higher, and carbide
However, increasing the particle diameter further to 1000 µm can debris from drilling and production equipment can have an even
higher density up to 15,000 kg/m3. Therefore, particles with affect the erosion pattern. Therefore, the eroded surfaces are compa-
density from 1500 kg/m3 to 6000 kg/m3 were simulated in this rable in Fig. 18. Both the maximum and average erosion rates,
study in order to obtain the trend of density effect on particle tra- shown in Figs. 19(a) and 19(b), increase with the density when it
jectory and erosion rate. is relatively small. The relation becomes flat and stable when the
As shown in Fig. 18, the highest erosion is on the outside shroud density is higher than 3000 kg/m3. Therefore, the density of com-
of both impeller and diffuser with simulated particle densities. With monly produced sand particles has a strong influence on erosion
higher density, the enlarged particle inertia is not big enough to inside ESP.
Fig. 17 Particle diameter effect: (a) facet maximum erosion rate and (b) area-weighted-average erosion rate
Fig. 19 Particle density effect: (a) facet maximum erosion rate and (b) area-weighted-average erosion rate