Professional Documents
Culture Documents
A Numerical Investigation of Solid Particle Erosion Experienced Within Oilfield Control Valves
A Numerical Investigation of Solid Particle Erosion Experienced Within Oilfield Control Valves
Abstract
Sand particles, produced in addition to petroleum fluid, present a major erosional hazard to transport and process control equipment within
the w'roleum industry. Control valves are most susceptible to erosion due to their fundamental action, experiencing failure through loss of
control in a matter of days in extreme cases. This paper presents a computational fluid dynamics erosion model which allows erosion ra:es
within chokes to be predicted. Particle trajectories are computed through the momentum exchange from the turbulent flow field, while erosion
rates are predicted as a function of individual particle impact characteristics, given material types, particle shape and size. Techniques used to
define differing material types and the display of results as surface plots are provided. Such a process improves our understanding of the
erosion mechanisms induced, providing erosion limitation by design optimisation and material selection. © 1998 Elsevier Science S.A. All
rights reserved.
There are four stages to erosion modelling through the u ~ a solution to the flow field via an iterative method. The grid
of CFD: may contain upwards of I00.000 control cells, with a local
!. Predict the turbulent flow field within the choke. .solution provided for each of the mean flow properties at
2. Predict the trajectories of sand particles within the flow every cell centre.
field, including particle impacts. To provide "closure' to the time averaged Navier-Stokes
3. Predict erosion characteristics as a function of the particle equations, it is necessary to utilise a turbulence closure model.
impacts and material types. Two principal closure models exist commercially, the Dif-
4. Display of results as surface plots. ferential Stress model and the k-Epsilon model. The k-Epsi-
Ion model is limited by. its assumption ofisotropic turbulence.
the Differential Stress model includes anisotropic turbulence
2. !. The turbulent fluid continuum
via the inclusion of six differential equations for the Reynolds
stresses. The k-Epsilon model is numerically robust and well
The continuous pha~, oil or gas, is solved through the u ~
of the discretised, time-averaged Navier-Stokes equations. validated, however, it can lack prediction accuracy for com-
This descretisation produces a series o'f t3artial-differential plex flow. Such complexities include rapid variations in flow
equations which describe the mean flow properties through area, unconfined jets and backflow regions. The Differential
the conventional continuity and momentum equations. The stress model is computationally expensive, however, it pro-
mean flow properties include the turbulent kinetic energy, the rides improved prediction accuracy over the k-Epsiion model
dissipation of turbulence due to the viscous stresses, the due to the inclusion of the Reynolds stress, as such is utilised
velocity components of the flow field and the pressure as default closure model.
distribution. Fig. I pre~nts an example of the steady state, time-aver-
Each of the mean flow properties is solved through the u ~ aged flow through an evolution needle and .seat choke. Here,
of a finite volume grid which is body fitted to the flow domain. the mean flow Woperty shown is speed, or the vectorial root-
This grid allows convection, diffusion and source terms to be ~uare of the three-velocity components. The advantage of a
computed throughout the computational domain, providing CFD erosion model is that flow features which influence
3 . 0 9 2 | 6-*-02
J 2.5768E÷O2
2.O614E*02
| . 5 4 6 | 1~÷O~
1.0307E÷02
5. ! 5 3 6 1 ~ ÷ 0 !
O.O000E +o0 m/~
Fig, 1. Velocity contours l~f an evolution nccdtc and ~ a t choke. Fluid Gas. Pressure drop 6.8 MN/m-'. Inlet dian~ter 75 mm.
186 A, Forder et aL / Wear 216 (19981 184-193
erosion can be captured, explaining certain erosion trends calculated by the closure models plus a contribution due to
observed in the field. effects of Iocalised turbulence. The contribution to the con-
At present, the turbulent flow properties can only be solved tinuum velocity is taken to be twice the local turbulent kinetic
for single phase flows, however, work is ongoing on the energy.
solution of multi-phase flows. Such work is extremely com- The contribution velocity is assumed to be made up of a
plex, being further complicated by the inclusion of suspended collection of randomly directed eddies. Where the interaction
solids. time of a certain eddy is taken to be the lower of either the
lifetime of the eddy or the trdnsit time required for the particle
2.2. Particle trajectories computation to pass through the eddy. The eddy time scale or lifetime tF
and the characteristic length scale IE of a numerical eddy are
To evaluate the particle trajectories and velocity history, it assumed to equal those of the energy containing eddies of the
is necessary to develop equations of motion for that particle. flow. The scales utilised hold the following forms:
A moving fluid exerts an aerodynamic force F on an
immersed body. Typically, F = D + L, where D is the drag tr.. = 1.5" "~C,~75-. (3)
E
force that acts in the opposite direction to the fluid motion
due to the momentum argument between phases.
= -~ - -g . (4)
CFX gives consideration to pressure gradient forces of the
surrounding fluid and gravitational effects. L is the lift force
Sato et al. [ I I concluded that the prediction performance
normal to the fluid motion, in this instance, we ignore the lift
provided by stochastic particle tracking is superior t¢ that
effects, such as Staffman and Magunus forces for computa-
provided by a deterministic method.
tional ease. Furthermore, virtual mass forces, Basset forces
and other cohesion forces are normally neglected in view of
2.3. Coefficient of restitution
the large differential density between phases.
The equations for the rate of change of panicle velocity
At impact, the reflected velocity of the particle is lower
within CFX are derived directly from Newton's second law.
than the incoming velocity due to energy transfer. Energy is
The velocities are obtained by simply integrating the force
dissipated as heat, noise and target material deformation. This
balance in a Lagrangian frame of reference. In the simplest
impact signature is described by the momentum based coef-
form, the equation of motion can be written as:
ficient of restitution, E .
The CFD code originally assumes that the impact signature
ntpat.~
itt = C D p ( U - u ) I U - u l - ~ + F. ( i) can be adequately described through consideration of the
normal velocity component alone and that the coefficient is
The main term on the right represents that which is the angle insensitive. However, Grant and Tabakoff [21 has
major aerodynamic force acting on the particle, the momen- shown that the coefficient is angle- dependent and that the
tum or drag force exerted by the continuous phase. This drag parallel velocity component, in addition to the perpendicular
force includes consideration of both skin and form effects. velocity component, is reduced at impact.
The total drag force is most conveniently expressed in To overcome the limitation of a single coefficient value, a
terms of the non-dimensional drag coefficient Ct,; defined by series of relationships to describe the rebound signature as a
following correlation: function of angle has been included, providing consideration
for both the parallel and perpendicular velocity components.
24 5.48
Ci) = Re~ + Re~'~7"-'-'--~+ (I.36. (2) Here, the coefficient has values of 0.3 to 1.0, depending on
target material type and the angle of impact; see Fig. 2. Work
It is assumed that no particle-particle interactions occur undertaken at Southampton University has concluded the fol-
due to the low particle concentrations experienced. The tra- lowing relationships for AISI 4130:
jectories of non-spherical particles can be modelled through
the application of shape factors, allowing non-uniform drag 0.9
distribution and bounce characteristics to be classified. ~ 0.8
The flow field solved by the turbulence closure models is
,,~ 0.6
time average, i.e., only the mean velocity flow field is resolved
• O.5 . ~
due to the computational limitations. However, during a par- _~o4 , ~ ,1
ticles motion, it interacts strongly with the turbulent eddies
of the flow. The effects of them eddies upon the particle are 0.2
/6
3, M i d particle e ~ i ~ v ; . ,9)
in previous erosion studies [3], universal models, based At high-impact angles, 30-90 °, an on-axis compression is
heavily on experimentally derived empirical constants, were generated under the impact site, inducing the defm'nmtion
used, These models take the form, W.--mr,kr(a)U~p. Such a erosion mechanism. At impact, stress concentrations are
model is adequate for simple erosion cases, where the particle established under the impact site, with the maximum shear
impact characteristics are relatively consistent. However, in stress occurring at the centre of the contact area at a depth of
the complex situation of a control choke, the particle impact approximately half the radius of the projected contact area.
characteristics vary greatly. Impact angles vary from 0 to 90 °. Due to the local proximity of the stress concentration, it is
while the impact velocities may be a-; high as 140 m/s. The possible for the yield strength of the target material to be
tailoring of a single function to capture erosion over such a exceeded, The repeated collision of a large number of parti-
wide range is extremely difficult, requiring much experimen- cles induces plastically deformed surface layers, resulting in
tal work to ensure the integrity of this simplistic approach. erosion through delamination, the generation of micro-cracks
To overcome such difficulties, a more complex modelling a n d the coale~ence of voids to produce .surface fragmenta-
strategy has been incorporated within this study. The model tion. The deformation erosion model used is that first derived
incorporates contributions from two constituents of erosion by Bitter, holding the form:
generated upon the particle impact angle, i.e.. low angle cut-
ting and high angle deformation, a philosophy first generated Wd = rap( :#,,,sin ~ - - Dj, )2 (I0)
by Finnie [4l. The constituents simply operate in parallel. 2E,
with the summation of the two presented as the overall ero-
Here, Er is an empirical constant specifying the energy
sion. In this instance, the Hashish 151 model provides the required to remove a unit volume of the target material by
cutting contribution, while Bitter [ 61 pre~nts the deforma- deformation erosion. DK is the deformation characleristic
tion requirements. velocity. This has a similar analogy to that of the cutting
At low impact angles, typically 0 to 40 °, the panicles have characteristic velocity, denoting the erosive potential through
a tendency to cut the target material. As the particles strike
the degree of elastic and plastic deformation. Bitter p r o d
at the surface, the target material is subjected to shear over the following for 1020 steel:
an area equal to the vertical cross section of that part of the
particle that has penetrated into the surface. If the shearing
,,.,
strength of the target material is exceeded, surface destruction % E, j
will occur. The hardness of the target material influences the
severity of the cutting mechanism, until the target material Once again, AFF has modified the deformation character-
hardness equals that of the erodent, at which point cutting istic velocity for the consideration of tungsten carbide. A
erosion should theoretically cease. Hashish proposed the fol- further modification to the characteristic velocity is the inclu-
lowing correlation for cutting erosion, being an extension of sion of the particle roundness factor, improving particle-
Finnie: erosion correlations:
,oo 0.5
= [i-q;+ ,-¢]-'. (12)
XP,/ L G E, J
The velocity exponent, n, is assigned a value of between 2
Both culling and deformation erosion mechanisms are
and 3. C k , the cutting characteristic velocity, is a function that
assumed to coexist simultaneously. Thus, the total erosion
combines particle and target material characteristics, describ-
experienced at each particle impact is simply presented by:
ing the erosive potential of each impact through the onset of
plastic deformation. Hashish presented the following form w,=w~+w~. (13)
for 1020 steel:
188 A. Forder el al. / Wear 216 (1998) 184-193
A complication experienced within the choke is the use of Face name Material Parameters defined
differing materials. Here, the body may be an AISI 4130 steel,
while the control surfaces are commonly tungsten carbide. Default AIS! 4 1 3 0 steel o'y = 3 2 0 . 0 E + 06
The problem presented is two-fold. First, the erosion equa- E( = 207.0E + 09
r
/
,."
- ,-"
, . .
Fig. 3. Material specification. ~*¢dle and seat choke. Fig. 4. Surface grid o f a needle and seat choke.
A. Forder et al, / Wear 216 (1998) 184-193 189
This is a highly effective method of displaying large quan- through the high flow velocities induced and rapid variations
tities of information, allowing many particles, each of differ- in flow area and direct. Such flow phcmancna, although det-
ing size and shape, to be traced through the domain. Here, rimental with regards to erosion, are essential to the funda-
we may be considering upwards of five hundred particles, mental action of the choke. However, by careful consid-
ensuring a large sample size for good statistical confidence. eration of such effects and the implementation o f subtle
With such a large sample size, it is reasonable to assume that design details, tim overall erosion may be minimised while
more than one impact may be experienced at each raxle, in retaining the chokes dissipative nature.
such a situation, the total erosion is taken to be the summation The system under test in each case was a 3 in. (75 ram)
of the individual particle impacts, while the impact velocity control choke, being simplified through the apl~ication of a
and angle are averaged. symmetry plane. Each grid contmned approxinmlely 40.000
At present, the erosion rate predictions are based on the
nodes, with grid refinements in those areas of rapid variable
original surface g¢ongtry, i,¢., pre-eJ'osion scar. No provision
change. The choke openings were 25%, inducing a pressure
is given for updating the geometry to reflect the eroded vol-
drop of 68 bar when the suspension fluid was gas.
ume as a function of time. However, given that the location.
Sand concentralions of 1% by wt, were utilised with the
the eroded volume of each impact and the erosion timescales
are known; updating of the geometry to include the eroded panicle distribution at inlet assumed to be uniform. The size
volume is a realistic evolution of the code. distribution was 50-300 p.m; divided into five groups, a~ 50
g m intervals. The particles were classified as sub-angular,
4. !. Design ew, lution through the application of surface being deformed from the sphere by 20%.
plots Figs. 5 and 6 detail the panicle impact characteristics for
the standard production choke. This design has a relatively
An example is pre~nted which emphasises the merits of long acceleration length through tlm trim, with the emergent
the computational code as a cost effective design tool. A annular jet being constrained in the constricted outlet spool,
needle and seat choke design has been modified with the aim see Fig. l. This constriction encourages jet prolmga/ion
of erosion limitation. Erosion within chokes is generated downstream, with the resulting particle impact velocities
i 5.7.9319E+O!
2879E÷01
6.60991~ .0 !
.9¢5.~9E+O !
2. b 4 4 0 E ~-0 !
i 3220E¢-01
o.O0OOE÷O0 m/s
Fig. 5. Particle impact velocities, original choke design. Particle size range 50--300 p.m.
190 A. Forder et al, / Wear 216 (1998) 184-193
i |. 136!
9.4675EE - 0- 80 7
7 , 5 7401~'-O8
5.6805E-~8
3 787OE-08
1 rS9~5~--08
O.O000E+O0 m" * .~lday
Fig. 6. Localised erosion rate. original choke design. Particle size range 50.300 p.m, Concentration i% by wt.
being high. A peak impact velocity of 80 m/s is pledicted with jet propagation downstream being suppressed. The
immediate to the divergence in the outlet. expense of such modifications was an oversize choke and an
The impact distribution can be seen to be biased due to the increased risk of cavitation due to the pressure recovery
presence of the radial inlet. The inlet ~osition promotes a characteristics.
tendency for the particles to pass to the back of the body Fig. 7 details the particle impact velocity distribution.
cavity prior to acceleration across the trim, Such local particle Once again we observe a biasing in the outlet due to the radial
and flow phenomena produce Iocalised erosion effects or inlet. However, in this instance the peak impact velocity is
"hotspotting" in the outlet spool (Fig. 6). Such erosion pat- half that of the original design. Further, the angles of impact
terns have been observed in the field, being most undesirable in the outlet are low, the predicted impact angle is below 15°.
with regards to pressure containment. The downstream piping As a result of the reduced impact velocities and angle, the
is subject to aggressive and sustained erosion damage, which, peak erosion is r~duced by an order of magnitude from that
if undetected, can result in catastrophic failure to atmosphere. originally predicted ( Fig. 8). Such a reduction in erosion rate
An additional benefit of knowledge regarding the location of will significantly increasing the service life of the choke,
erosion hotspotting is the confident application of wall thick- allowing large cost savings to be realised.
ness measurements to monitor erosion damage.
It is most desirable for such erosive effects to be minimised.
Here, the design methodology utilised was the reduction of 5. Model validation
both particle impact angles and velocities. To achieve this
goal, the trim area was profiled, such that no rapid variations Two choke manufacturers consented to provide a number
in flow direction occurred. Such profiling simply involved of chokes for erosion testing within facilities at BP; allowing
increasing the taper length of the needle, minimising the the turbulent flow field, particle trajectories and erosion
length of the constricted outlet spool and reducing the diver- model to be validated with carefully controlled experimental
gence angle of the outlet expansion. These modifications data. Erosion was assessed under various operating condi-
ensured that the particle acceleration length was minimised, tions, such as valve opening, pressure drop, flow rate and trim
A. Forder et aL / Wear 216 ¢1998~ 184--193 191
4.OOO4E+O1
I 3 ~l3 3"/E+O i
i 2.66701=_..01
2 . 0 0 0 2 I~.+0 I
1 3743.~1E+0 t
6.6674E+OO
O OOOOE+OO m/s
Fig. 7. Panicle impact velocities, evolulion choke design. Particle size range 50-.300 I~m. Com:enlralion !% by wt.
material. The aim of this work was to ensure that the erosion slants to b¢ tailored to provide the best fit for the erosion data.
location, in addition to its intensity, could be captured by the Fig. 9 shows the data fit for 6% cobalt binder tungsten car-
CFD erosion model. Initial indications show that the corre- bide. The average error over the entire range of angles was
lations are good with regards to both location and erosion reduced to 0.05% by careful +selection of parameters and
intensity for all trim materials tested. Table 2 presents a sum- refinements to the erosion model.
mary of the valves and trim materials tested.
Furthermore, to full valve tests, flat sample testing has been
undertaken at Southampton University to evaluate the empir- 6. C o n c l m
ical erosion constant used in the erosion models. See Table I
for details of n and Er. This evaluation process involved ( I ) A state-of-the-art computational fluid dynamics ero-
jetting a water sand mixture onto flat samples at known oper- sion model has been developed for the analysis of erosional
ating conditions, such as flow rate and sand concentration. trends in complex geometry such as chokes.
The flow rate was 50 I/rain, peak jet velocities of 38 m/s, (2) Large quantities of data can be processed by the ero-
sand concentrations of 2% by wt., and sand size distribution sion model through the application of highly visual and
of 220-250 p,m. Them operating conditions were replicated informative surface plotting techniques, ensuring good sta-
within the CFD erosion model, allowing the empirical con- tistical confidence.
Table 2
Choke lypes and trim materials tested
Manufacture and choke type Cameron CC30 equal c~ plug aqd cage ABB Kenl Introl series 7,$ equal % plug and cage
Trim material Three of 17.4 PH stainless steel, one of 6c~ coball tungsten Two of Stellke 6, one of 6% cobalt zungsten carbide, one of
carbide PSZ ceramic
192 A. Forder et aL / Wear 216 (1998) 184-193
i 3.3
2 7 954~-Io1~-O8
1E-OI~
2 2353E-O8
I 6765E-O8
1.1 i T"/'E-OB
5.SBg3E-O9
00000E+OOm* * 3 / d a y
Fig. 8. l..ocali~d erosion rate. evolution choke design. Panicle size range 50--300 ~rn. Concentration 1% by wt.
J- 4O
20
0
0 30 45 60 75 9O
desirable.
(8) Good correlation between experimental and predicted
data can be achieved.
PlIII ~ (deg| 7. N o m e n c l a t u r e
Fig. 9. Comparison of treasured and predicted data tot 6'/, cobalt binder
tungsten c',u'bide. A~, Area of particle m~
CL Cutting characteristic velocity m/s
(3) The surface plots facilitate rapid design optimisation Dk Deformation characteristic velocity m/s
of the choke via a visual and numerical interface with the C,, Eddy viscosity constant (value 0.09) -
user. Et Deformation erosion factor J/m ~
(4) The erosion model allows differing material types to Ep Young's modulus for particle N/m-"
be specitied within the computational domain. A database of E, Young's modulus for target N/m-"
material constants is provided within the embedded k Fluid turbulent kinetic energy mZ/s 2
FORTRAN. m~, Mass of particle kg
(5) The coefficient of restitution has been modilied to n Characteristic velocity exponent (value
include consideration of both impact angle and material type. 2-3)
(6) The model provides useful information regarding the qr Poisson's ratio for particle
mean flow properties within the choke and the relationship q, Poisson's ratio lor target
A. Forder et aL / Wear 216 (1998) 184-193 193