Sie sind auf Seite 1von 9

Wear 308 (2013) 1–9

Contents lists available at ScienceDirect

Wear
journal homepage: www.elsevier.com/locate/wear

A probability model for solid particle erosion in a straight pipe


Ri Zhang a, Haixiao Liu a,n, Chuntian Zhao b
a
School of Civil Engineering, Tianjin University, Tianjin 300072, China
b
CNOOC Research Institute, Beijing 100027, China

art ic l e i nf o a b s t r a c t

Article history: Erosion caused by solid particles entrained in pipelines can lead to serious economic loss and potential
Received 21 March 2013 safety hazards in the oil and gas industry. Long straight pipe is the most common form of pipeline. In the
Received in revised form present work, a probability model for particle erosion in a long straight pipe is proposed to predict the
18 September 2013
overall penetration rate caused by solid particles by analyzing the random movement of a single particle
Accepted 22 September 2013
in the turbulent flow field. The model works in two mechanisms. First, the distribution of the
Available online 14 October 2013
characteristic vortex length is deduced to describe the turbulent flow in a pipe based on the energy
Keywords: equilibrium among eddies with different scales and the distribution of fluctuation velocities. Second, the
Solid particle erosion particle motion in the radial direction is regarded as a Markov process which is only influenced by the
Impact wear
eddy randomly appearing in the fully developed turbulent flow. Based on the knowledge of the turbulent
Surface analysis
flow field and the motion of particles, a particle erosion formula is introduced into the probability model
Wear modeling
to calculate erosion in a straight pipe. By comparing with numerical simulations and experimental data,
the present model is well verified and demonstrates advantages in accuracy and applicability.
& 2013 Elsevier B.V. All rights reserved.

1. Introduction was noncorrosive and contained sands. Then a method to calculate


the limiting velocity accounting for different fittings and sand flow
Erosion is a severe problem during the production of oil and rates was proposed. The estimation on limiting velocity presented by
gas. Crude oil and natural gas extracted from the reservoirs usually Salama [3] was similar to above with the addition of the effect due to
contain solid particles such as sand which may cause considerable fluid density. The expression was just a deformation of his previous
erosion damage to the piping, fittings and other equipment. The prediction formula when specifying certain tolerable sand concen-
damage to these components reduces the operational reliability, tration and erosion rates. Jordan [4] used the concept about equiva-
increases the risk of failure, and may even result in significant lent stagnation length and predicted the limiting velocity by
financial loss to the industry and danger to the personnel and iterations for multiphase flow. The approach required some assump-
environment. Hence the ability to predict erosion accurately and tions about the particle production rate and the particle size.
effectively is of essential importance. Some investigators attempted to develop models to directly
As the earliest and most widely used guideline for establishing calculate the erosion rates. Salama and Venkatesh [5] presented a
production rates for erosion, the American Petroleum Institute (API) model for penetration rates for field ells and tees, which just
provided a maximum flow velocity below which only a tolerable applied to the single-phase flow initially, but after modifications it
amount of erosion occurred [1]. However the guideline just could be used to predict solid particle erosion in multiphase flow
accounted for the influence of the fluid density on erosion damage by taking a weighted average of the fluid density and viscosity.
and did not seem to be appropriate for situations involving sand Bourgoyne [6] proposed an equation for estimating the penetra-
production. Recognizing the limitations of API in the presence of tion rate in diverter systems based on the experimental study for
sand, some investigators modified the estimation on limiting velocity four different pipe fittings used at bends. The effects of flow
by considering more factors contributing to erosion to meet the velocity, liquid content and sand concentration were considered.
demand in practical cases. Svedeman and Arnold [2] suggested that Nevertheless, all of the approaches mentioned above are “rules of
the determination of limiting velocity should distinguish the differ- thumb” and valid only near the conditions for which experiments
ence in wear mechanisms such as erosion and corrosion. Solid were conducted. The distinctions among these prediction methods
particle erosion was the dominant wear cause when the flow stream are just the factors involved and the scopes of their applications.
There were still some mechanistic models focusing on the
whole process that particles pass through a complex flow field,
n
Corresponding author. Tel.: þ 86 2227401510. impinge on the inner walls of piping at certain impact angles and
E-mail address: liuhx@tju.edu.cn (H. Liu). speeds, and then remove the wall material. Shirazi, et al. [7]

0043-1648/$ - see front matter & 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.wear.2013.09.011
2 R. Zhang et al. / Wear 308 (2013) 1–9

Nomenclature v′r v′θ v′z fluctuation velocities of fluid in the radial, tangential
and axial directions, respectively (m/s)
A area (m2) vrf vzf velocities of fluid in the radial and axial directions,
CD drag force coefficient (dimensionless) respectively (m/s)
D pipe diameter (m) vrp vzp velocities of particle in the radial and axial directions,
dp particle size (m) respectively (m/s)
Er erosion ratio (dimensionless) vn friction velocity (m/s)
fr volume force in radial direction (N) W mass flow of sand (kg/s)
HV Vickers hardness number (GPa) z displacement in the axial direction (m)
K particle property factor (dimensionless)
k1 k2 k3 exponent factors (dimensionless) Greek letters
lp particle displacement (m)
n1n2 material hardness exponents (dimensionless) δ boundary layer thickness (m)
P probability (dimensionless) Δ absolute roughness (m)
Pf pressure (Pa) μf fluid viscosity (Pa s)
Pr penetration rate (m/s) ηδ attenuation coefficient (dimensionless)
Re Rep Reynolds numbers of pipe and particle, respectively ρf ρp ρt densities of fluid, particle and pipe material, respec-
(dimensionless) tively (kg/m3)
tλ eddy lifetime (s) λ length scale of eddy (m)
vλ characteristic eddy velocity (m/s) νf kinematic coefficient of viscosity (m2/s)
Vm mean velocity (m/s) ϕ friction coefficient (dimensionless)
V r V θ V z time-averaged velocities of fluid in the radial, tangen-
tial and axial directions, respectively (m/s)

thought that the particle velocity would reduce when the particle a simplified alternate method was proposed though it was still
penetrated the fluid layer adherent to the pipe wall, so a concept difficult to handle. On the other hand, the solutions were only a
of stagnation length which represented the distance over which time-averaged flow field by applying numerical simulations. The
the particle slowed as it approached the target wall was proposed approach of Gosman and Ioannides [15] that incorporated the
to find the characteristic impact velocity of particles. A mechan- effect of the turbulence dispersion on the particle motion was
istic model was then developed based on this idea for predicting employed to obtain the transient features of the flow field.
erosion in elbows and tees. Afterwards, the researchers from Tulsa However, Zhang [16] proved that this approach was inaccurate.
University established a “Tulsa system” to analyze the erosion The present work aims at developing a probability model for
process occurring in various fittings based on the mechanistic particle erosion in a straight pipe. By describing the fluctuation
model. McLaury [8] proposed a generalized erosion prediction velocity in turbulence and tracking the corresponding response of
procedure that involved flow simulation, particle tracking and particles based on the probability analysis, and combing the two
erosion prediction. Wang [9] applied this approach to investigating steps, a direct correspondence between the distribution of parti-
the effects of elbow radius on erosion rates. Edwards [10] cles in the flow field and the impact points on the pipe wall can be
improved the numerical calculation process and implemented established. The effective mass of particles hitting the pipe wall
the generalized erosion prediction procedure in the commercial can then be expressed by calculating the probability that a single
software for Computational Fluid Dynamics (CFD). Mazumder, particle impinges on the target area. By introducing a particle
et al. [11] extended the application of the mechanistic model to erosion formula, the penetration rate for straight pipes can be
multiphase flow and analyzed the complex flow behaviors and predicted based on the probability analysis.
particle distribution that existed in different flow patterns. Chen,
et al. [12] introduced the stochastic rebound model to the
numerical method and investigated the relative erosion severity 2. Assumptions and inferences
in elbows and plug tees. Zhang, et al. [13] modified the particle
near-wall behavior and proposed a two-dimension mechanistic As shown in Fig. 1, the behavior of a single particle in turbulent
model based on numerical analysis. Besides, the model developed pipe flow can be described as: (1) vortexes of different scales appear
by Det Norske Veritas (DNV) [14] simply allowed actual calcula- randomly in turbulent flow and induce fluctuation velocities in every
tions of the trajectories of solid particles and emphasized the
effect of the particle impact angle. It provided detailed guidelines y
on straight pipes, wedded joints, reducers, elbows and blind tees.
Although solid particle erosion in a straight pipe is far less than
the erosion observed in elbows and tees, adequate attention
should be paid to straight pipes as the most common components
in piping systems. It is noted that the prediction method proposed
by DNV was an empirical formula which only demonstrated the m z
effects of flow velocity, pipe diameter and the mass flow of sand
r

on erosion. McLaury [8] pointed out the erosion that occurred in


straight pipes was due to the random impact of particles on the x
pipe wall caused by turbulent fluctuations. A comprehensive L z
model was then provided to calculate the erosion in straight pipes
using CFD. Because the method was complex and time consuming, Fig. 1. Particle track in turbulence.
R. Zhang et al. / Wear 308 (2013) 1–9 3

direction; (2) the particle m moves along the axial direction of the the probability of the event that a single particle m strikes the
pipe while its trajectory will fluctuate erratically due to the influence target region Δz is
of vortexes; (3) when the particle in the near-wall region can possess
P ¼ PðA1Þ  PðA2Þ ð1Þ
enough momentum in the radical direction, it will impinge on the
target area Δz and cause damage to the inner face of the pipe. The The probabilities of the events A1 and A2 in Eq. (1) will be
probability model will estimate the overall erosion results induced by discussed in detail in the following section. Eq. (1) is important in
solid particles by calculating the probability P of the event that a single analyzing the overall erosion results caused by solid particles.
particle can impact on the target area Δz.
To provide a reasonable solution for engineering application,
several simplified assumptions and inferences are made to describe 3. Solid particle erosion in a straight pipe
the erosion process in a straight pipe.
When the particles pass through the flow field surrounded by
2.1. Similarity and axial symmetry the circle slice Δz, not all of them can reach the pipe wall. This
means that an effective mass flow of sand WP towards the pipe
The characteristics of the flow field and the distribution of wall must be used and not the actual mass flow of sand W along
particles are considered to be almost invariable in an arbitrary the axial direction. In fact, the reduction coefficient P is just the
cross section of fully developed pipe flow. The influence of gravity probability that a single particle impacts on the target zone Δz as
is ignored, and hence the flow field and the particle distribution mentioned in Eq. (1). A new description method about the
are of axial symmetry. Based on the similarity and symmetry, the turbulent flow flied is established to calculate the probability of
erosion degree at an arbitrary position in the straight pipe is the event A1, and the corresponding response of particles is
almost identical for fully developed pipe flow. In the following investigated to get the probability of the event A2. Hence the
study, a circle slice Δz perpendicular to the pipe axis is selected as overall erosion results caused by solid particles can be obtained by
the target area. calculating the probability of the event that a single particle
impacts on the pipe wall.
2.2. Particle velocity
3.1. Vortex distribution
The particles present in the fluid are dilute. This means that the
presence of solid particles does not influence the flow behavior The flow field in a straight pipe can be divided into the
and there are no interactions among particles. Generally, the turbulent core region and the boundary layer region, as shown
fluctuation velocity is far less than the time-averaged velocity in in Fig. 2. The boundary layer is a thin fluid layer next to the inner
the axial direction, hence the pulsation of particle in this direction face of the pipe. In this region the effects of viscosity are dominant
can be neglected. As discussed in Section 2.1, the target area Δz is a and the turbulence is small and can be neglected. The turbulent
slice where erosion damage is distributed uniformly, so the core occupies the most space of the flow field and contains many
movement of particles in the tangential direction is not under eddies with different scales. Besides the velocity along the axial
consideration herein. The particle velocity in the radial direction is direction, the velocity in the radial direction can also be observed,
essentially important to the particle distribution and impact which is caused by large or small eddies that appear randomly. As
damage, which will be discussed in detail later. discussed in Section 2.3, the turbulent energy is drawn continu-
ously from big eddies to small eddies. Kolmogorov [18] defined the
2.3. Energy transfer among eddies characteristic eddy velocity vλ and length scale of eddy λ in terms
of the power input to unit mass of the system, as
For isotropic turbulence, when the distribution of eddies has
come to equilibrium, it was considered by Davies [17] that the 4v2n V m
v3λ ¼ λ ð2Þ
same amount of energy must be continuously drawn from the D
medium-size eddies containing most of the turbulence energy to where, V m is the mean velocity; D is the pipe diameter; vn is the
the small-size eddies dissipating energy. In the present study, this friction velocity that can be obtained as [17]
theory is extended so that the turbulence energy is transferred
Vm
continuously among eddies with different scales. In addition, the vn ¼ ð3Þ
3:2 2:46InðΔ=DÞ
flow field in a straight pipe can be approximately treated as
isotropic turbulence except in the boundary zone. where Δ is the absolute roughness. In turbulent flow the vortexes
will appear randomly with different scales and force the vicinal
2.4. Markov process fluid to fluctuate, hence the fluctuation velocity of the fluid v′r is
equal to the characteristic eddy velocity vλ . From Eq. (2), the
The movement of particles in the radial direction is apparently expression can be obtained as
a stochastic process. It is supposed that the particle will sharply rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
3 4v V m λ
2
n
change its motion when trapped in a vortex. This means the v′r ¼ ð4Þ
stochastic process just rests with the current vortex. Then the D
movement of particles in the radial direction has the nature that
any future state is independent of the past state and depends only Boundary
layer
on the present state, which is defined as a Markov process.
Although the particle trajectory seems to be chaotic, the particle
must reach a key state to strike the pipe wall no matter what it has
experienced before. In the key state, two completely independent
events occur simultaneously: (1) A1, which is the event that a big
Turbulent
enough vortex emerges in the flow field near the pipe wall; (2) A2, core
which is the event that a particle can access this region. If PðA1Þ
and PðA2Þ denote the probabilities of A1 and A2, respectively, then Fig. 2. Flow field in a straight pipe.
4 R. Zhang et al. / Wear 308 (2013) 1–9

In Eq. (4), the length scale of the eddy λ is a random variable 3.2. Particle response
and the fluctuation velocity v′r is a function of λ. It is necessary to
analyze the distribution of λ and v′r to estimate the turbulent The particle follows a random walk in the radial direction
intensity of pipe flow. The Reynolds averaged Navier–Stokes caused by the surrounding fluid when trapped in a vortex.
equation in the radial direction is Suppose that the particle is so tiny that it cannot keep its former
motion state under the influence of the vortex, then the move-
∂V r ∂V 2r 1 ∂V r V θ ∂V r V z V r 2  V θ 2
þ þ þ þ ment of the particle in the radial direction is a Markov process.
∂t ∂r r ∂θ  ∂z r 
Due to the difference in density, the motion of the particle will lag
1 ∂P f V r 2 ∂V θ
¼fr 2
þ νf ∇ V r  2  2 behind the surrounding fluid. The governing equation of the
ρ ∂r r r ∂θ
! particle motion in vortex is
∂v′2r 1 ∂v′r v′θ ∂v′r v′z v′r 2 v′θ 2  
 þ þ þ ð5Þ 1 3 dvrp πdp
2   1 3 dvf
r
∂r r ∂θ ∂z r  
πdp ρp ¼ C D ρf vrf  vrp ðvrf  vrp Þ þ πdp ρf
6 dt 8   6 dt
where, νf is the kinematic coefficient of viscosity; V r and V θ are the !
1 dvrf dvrp
time-averaged velocities of fluid in the radial and tangential þ
3
πdp ρf  ð10Þ
12 dt dt
directions, respectively; v′r and v′θ are the fluctuation velocities
of fluid in the radial and tangential directions, respectively. As where, dp is the particle diameter; ρf and ρp are the densities of the
discussed in Section 2.1, the volume force f r and the partial fluid and the particle, respectively; vrf and vrp are the velocities of
differential ∂=∂θ can be ignored because of the similarity and axial the fluid and the particle in the radial direction, respectively; μf is
symmetry of the flow field. As assumed in Section 2.3, turbulence the fluid viscosity; C D is the drag force coefficient with the
in the turbulent core is isotropic, so v′r 2 ¼ v′θ 2 and ∂P f =∂r ¼ 0. In approximation as
addition, the partial differential ∂=∂t and the time-averaged 2 !0:687 3
velocities V r and V θ are all zero for fully developed steady pipe 24 24μf v′r ρf dp
CD ¼ ð1 þ 0:15Rep0:687
Þ r 4 1 þ 0:15 5
flow. Hence, Eq. (5) reduces to Rep jvf vrp jdp ρf μf
∂ð  v′r 2 Þ
¼0 ð6Þ The first term in the right hand of Eq. (10) describes the drag
∂r
force due to the relative motion of the particle in fluid. The second
Eq. (6) indicates that the mean square value of the turbulent term is the surface force acting on the particle resulting from the
fluctuation velocity in the radial direction does not change with pressure variation on the surface. The third term describes the
the radial coordinates, so the turbulent intensity is almost con- inertia of the added mass entrained with the relative motion of the
stant at any position of turbulent core zone for fully developed particle in the entraining fluid.
turbulent pipe flow. Although the solution to N–S equation is Note that the gravity is far less than the other items in Eq. (10)
time-averaged, it is approximately equal to the system average for and neglected for small particles, as discussed in Section 2.1. The
steady flow. The experimental data from Laufer [19] demonstrates effects of the Saffman force [21] are not considered since the
that Eq. (6) is roughly reasonable, as shown in Fig. 3. Thus an velocity gradient in the turbulent core is very small. Suppose that
approximation has been made as the particle motion in the radial direction is a Markov process as
v′2r  v2n ð7Þ stated earlier, and then the Basset force [22] can also be ignored.
As discussed in Section 2.2, the particle does not influence the
Batchelor [20] has confirmed that the fluctuation velocity in turbulence structure. Hence the radial gradient of fluid velocity is
isotropic turbulent flow obeys the Gaussian distribution. Based on almost zero. Eq. (10) is then simplified to
this theory, it is believed that the fluctuation velocity at an 8 dvr
arbitrary point in the turbulent core is normally distributed with >
< dt ¼ ða þ bλ
p 0:229
Þðvrf  vrp Þ
the same variance and mean value. From Eq. (7), the probability  0:687  2 0:229 ð11Þ
> 36μf 5:4μf ρf dp 4vn V m
:a¼ 2 ; b¼ μ
density function of v′r is expressed as ð2ρp þ ρf Þdp
2
ð2ρp þ ρf Þdp D
f

1 vrf ¼ v′r , the solution to Eq. (11) is


f v ðv′r Þ ¼ pffiffiffiffiffiffi e  ðv′r =2vn Þ
2 2
ð8Þ Considering that
2π vn 0:229
vrp ¼ v′r  v′r e  ða þ bλ Þt λ
ð12Þ
Adopting Eqs. (4) and (8), the probability density function of λ
can be directly obtained. Furthermore, an important expression of where t λ is the eddy lifetime which means the action time of the
the probability of the event A1 can be derived as eddy. The following expression can be obtained from Eq. (4) as
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2V 2m p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

3 2 2 2 2 λ λ2 D
λ  ð2=3Þ e  ð2V m λ =D vn Þ dλ
6
PðA1Þ ¼ f λ ðλÞdλ ¼ ð9Þ
3
2 2 tλ ¼ ¼ ð13Þ
729π D vn 3 v′r 4v2n V m

The particle displacement lp can be derived by integrating Eq.


1.5 (12) as
v′r 0:229
lp ¼ λ  ½1 e  ða þ bλ Þt λ  ð14Þ
1.0 a þ bλ0:229
v'r/v*

Although the single particle erratically moves in the flow field,


0.5 numerous particles will be evenly distributed at any cross section
for fully developed pipe flow. As a Markov process, whether a
particle can impact on the target area Δz depends only on the last
0.0 displacement caused by a near wall vortex regardless of the
0.0 0.2 0.4 0.6 0.8 1.0
previous movement. To reach the target area Δz after a random
2r/D
walk, the particle should be located in the annular zone of the pipe
Fig. 3. Experimental data from Laufer [19]. cross section, as shown in Fig. 4, where a mapping is created
R. Zhang et al. / Wear 308 (2013) 1–9 5

and hence it is reasonable to think that the threshold vortex length


λ just equals the boundary layer thickness δ. Adopting Eqs. (12)
and (17), the impact velocity in the radial direction vrpt can be
expressed as

vrpt ¼ ηδ vrp ð18Þ

The axial component of the impact velocity vzp is approximately


equal to the fluid velocity in the center of the vortex, that is
 
dlp lp lp þ 2δ
vzp  vzf ¼ V m þ 2:5vn In þ3:75vn ð19Þ
2 D

Adopting Eqs. (18) and (19), the impact velocity V p and the
lp impact angle θ can be expressed as
8 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
>
< V p ¼ vr2pt þ vp
z2

r ð20Þ
>
: θ ¼ arc tan vptz
v
p

Fig. 4. Particle distribution.


V p and θ can be used to calculate the erosion ratio Er , which is
defined as the mass loss of the pipe wall due to erosion divided by
between the target zone Δz and the circular zone dlp . The the mass of particles impacting on the wall. According to the study
probability of the event A2 is just the probability that the particle by Oka, et al. [23], the erosion ratio is given by
lies in the circle. For the uniform distribution, it represents the 8  k2  k3
area ratio, that is >
< Vp dp
Er ¼ KðH V Þk1 Vs s f 1 ðθÞρt
dp
ð21Þ
Aring πðD  2lp Þdlp 4ðD  2lp Þ dlp >
:
PðA2Þ ¼ ¼ ¼ dλ ð15Þ f 1 ðθÞ ¼ ð sin θÞn1 ½1 þ H V ð1  sin θÞn2
Apipe πD2 =4 D2 dλ

where dlp =dλ is the derivative of the particle displacement lp with where, K is the particle property factor; k1 , k2 and k3 are exponent
respect to the eddy length scale λ. The following expression can be factors; H V is the Vickers hardness number; n1 and n2 are material
s
obtained from Eq. (14) as hardness exponents; V s and dp are the standard impact velocity
0:229 0:667 0:229 0:667 and the standard particle diameter, respectively; ρt is the density
dlp 1 e  cða þ bλ Þλ 0:229bð1  e  cða þ bλ Þλ Þ
¼ 1 0:229 0:667
þ of pipe materials. The parameter values are listed in Table 1. As
dλ 3cða þbλ Þλ cða þbλ0:229 Þ2 λ0:438 discussed in Section 2.1, the mass loss of the pipe will be uniformly
0:229 0:667
ð0:667a þ 0:896bλ0:229 Þe  cða þ bλ Þλ distributed over the annular target zone Δz. The width of target
 ð16Þ
a þ bλ0:229 zone Δz corresponds to the thickness of circle area dlp , and can be
p ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi expressed as
where c ¼ 3 D=ð4v2n V m Þ. Adopting Eqs. (1), (9) and (15), given the
mass flow of sand W, the effective mass of the sand that impacts  z
dz dðvzp t λ Þ dvp dt λ z
on the wall forced by the vortex with scale λ is WP in unit time. Δz ¼ Δλ ¼ dλ ¼ tλ þ vp dλ ð22Þ
dλ dλ dλ dλ

3.3. Process of particle impact where the differential relations are


8 z
> dv
< dλp ¼ l2:5v n dlp
Once the particle experiences the last vortex and leaves the p þ 2δ dλ
qffiffiffiffiffiffiffiffiffiffiffiffi :
turbulent core, it will enter the boundary layer where the impact >
: dt ¼ 3 27v2D  ð1=3Þ
2V λ
λ

velocity will decay further. Suppose that the velocity in the axial n m

direction almost does not change because the penetration time is


Let P r denote the penetration rate defined as the thickness loss
quite short, then the velocity varies mainly in the radial direction.
per unit area and per unit time. Given the vortex length scale λ, the
Eq. (10) can be used again to describe the particle movement in
differential equation of penetration rates caused by numerous
the boundary layer. The attenuation coefficient ηδ in the radial
particles is
direction can be expressed as
δ WPðA1ÞPðA2ÞEr
ηδ ¼ 1  ða þ bc0:687 vr0:687 Þ ð17Þ dP r ¼ ð23Þ
vrp p πρt DΔz

where δ is the boundary layer thickness and can be expressed as It should be pointed out that PðA1Þ, PðA2Þ, Er and Δz are
[8] functions that only depend on the eddy length scale λ, and PðA1Þ,
40νf PðA2Þ and Δz all contain the infinitesimal item dλ. By canceling the
δ¼ infinitesimal, a differential equation can be derived. Hence the
vn
If the initial particle velocity vrp takes a quite small value, the Table 1
attenuation coefficient ηδ will turn negative inevitably. This makes Parameters for calculating the erosion ratio [23].
Eq. (17) have no physical meaning. In fact the particle needs to
reach a threshold initial velocity to penetrate the boundary layer. V s (m/s) s
dp (μm) K k1 k2 k3

In other words, a threshold vortex length scale λ should be


104 326 50  0.12 2:3ðHV Þ0:038 0.19
determined, since the value of vrp monotonously increases with
n1 n2 s1 s1 q1 q2
increasing value of λ. The particle deceleration in the boundary s1 ðH V Þq1 s2 ðHV Þq2 0.71 2.4 0.14  0.94
layer is a reverse process of the particle acceleration in a vortex,
6 R. Zhang et al. / Wear 308 (2013) 1–9

penetration rate caused by vortexes of all scales is Table 3


Z D=2 Parameters for calculating the penetration rate [24].
WPðA1ÞPðA2ÞEr
Pr ¼ ð24Þ Δ (mm) ρf (kg/m3) ρp (kg/m3) ρt (kg/m3) μf (Pa s)
δ πρt DΔz W(kg/s) HV (GPa)

The penetration rate can be calculated by substituting Eqs. (9), 0.05 983.6 2650 7850 4.67  10  4 8.68  10  5 1.4
(15), (21) and (22) in Eq. (24). The lower limit of integral δ
indicates that the threshold vortex length scale is equal to the
boundary layer thickness, as discussed before, while the upper
limit of integral D=2 indicates that the largest eddy length scale Table 4
accounts to half of the pipe diameter. The bigger eddy absorbing Comparison of penetration rates between the probability model and numerical
particles will lead to greater impact velocity and cause more simulations (mm/year).

severe erosion in pipe. The appearance probability of large Case 1 2 3 4 5


vortexes is so small that it is almost impossible to find this scale
of vortex in the flow field. Hence, Eq. (24) demonstrates a balance Numerical 5.59  10 3
1.74  103
5.22  10 4
2.58  10 4
1.59  10  4
between the damage induced by a certain scale eddy and the Probability 9.74  10  4 3.42  10  4 1.10  10  4 5.38  10  5 3.02  10  5
Case 6 7 8 9 10
occurrence possibility of this eddy.
Numerical 1.11  10  4 5.22  10  4 1.20  10  3 2.05  10  3 3.15  10  3
Probability 3.02  10  5 1.10  10  4 2.58  10  4 4.68  10  4 7.53  10  4
Case 11 12 13 14 15
4. Validation of the erosion model Numerical 1.04  10  5 2.68  10  4 4.40  10  4 5.22  10  4 5.82  10  4
Probability 7.48  10  5 1.16  10  4 1.25  10  4 1.10  10  4 1.02  10  4

In this section, the developed probability model is examined by


comparing with the numerical simulations and physical experi-
ments. The numerical simulations performed by McLaury, et al.
[24] show a variety of cases in which the pipe diameter, velocity,
and particle diameter vary. The physical experiments contain three
sets of laboratory experiments [8,25,26] and one field experiment
[27] to explore the accuracy of the probability model.

4.1. Comparison with numerical simulations

The numerical analysis carried out by McLaury, et al. [24]


captured the behavior that particles traveled through the flow
field and impacted on the pipe wall. Three main sections were
comprised in the analysis, i.e., a flow model, a particle tracking
model, and an erosion ratio model. A two-equation k ε turbu-
lence model was used to solve the flow field, and a Lagrange Fig. 5. Predicted penetration rates versus pipe diameter.

method and an erosion formula proposed by E/CRC [8] were


respectively implemented in the following steps. 15 cases were 4.1.1. Influence of the pipe diameter
designed to examine the effects of various parameters on erosion The effects of the pipe diameter on penetration rates from
in straight pipes, in which Cases 1–5, Cases 6–10 and Cases 11–15 Cases 1 to 5 are presented in Fig. 5, which indicates that the results
were used to investigate the influences of the pipe diameter, the of numerical simulations are about 4 or 5 times greater than those
flow velocity and the sand size, respectively. The analytical cases of the probability model, yet both methods predict a decrease in
are listed in Table 2. Parameters for calculating the penetration penetration rates as the pipe diameter increases. Three factors
rate and the erosion ratio are listed in Tables 3 and 1, respectively. may explain the variation trend predicted by both methods. First,
The comparison of penetration rates between the probability from Eq. (3) the turbulence will be more violent for the pipe with
model and numerical simulations are presented in Table 4 and smaller diameter. Therefore the particle will get more momentum
Figs. 5–7. to impact on the pipe wall and cause more severe erosion. Second,
the particles distributed in the near-wall region are rarer from
Table 2 Eq. (13) for large diameter pipes. This means that the particles
Analytical cases [24].
have less opportunity to reach the wall. Third, the boundary layer
Case D(m) V m (m/s) dp (μm) is thinner in the pipe with smaller diameter, where the particle
will pass more easily as shown in Eq. (17).
1 0.0254 6.0960 300
2 0.0508
3 0.1016
4 0.1524
4.1.2. Influence of the flow velocity
5 0.2032 The effects of the flow velocity on penetration rates from Cases
6 0.1016 3.048 300 6 to 10 are presented in Fig. 6. The erosion predicted by the
7 6.096 numerical method is also higher than that by the probability
8 9.144
model. From both methods, penetration rates show an increase
9 12.192
10 15.240 with increase of the flow velocity. Higher flow velocity provides
11 0.1016 6.0960 50 the particle more momentum both in radial and axial directions
12 100 and results in higher penetration rates. But known from Eq. (22),
13 200 higher flow velocity also widens the target zone Δz. Then an equal
14 300
15 400
amount of erosion will be spread over a larger area and the
penetration rate will decrease. Note that the penetration rates are
R. Zhang et al. / Wear 308 (2013) 1–9 7

proportional to the power of the particle impact velocity while as known from Eq. (17). Third, the large particles will lead to more
linearly changing with the damage area. Hence the impact velocity severe damage to material at the same condition, as known from
is still the dominant factor of the penetration rate. Eq. (21). Anyway, the variation trend of penetration rates is
determined by the combination of the three factors.
4.1.3. Influence of the sand size
The effects of the particle size on penetration rates are 4.2. Comparison with experiments
complicated, as shown in Fig. 7. If the particle size is larger than
100 μm as in Cases 12 to 15, the penetration rates predicted by the The present model is also examined by physical experiments
probability model are less than those by numerical simulations, available from the literature, which were respectively performed
yet for the particles of 50 μm as in Case 11, the value calculated by Okayama [27], Postlethwaite and Nesic [25], McLaury [8] and
with the probability model is far greater than that with numerical Wood, et al. [26], and contained three sets of laboratory experi-
simulations. Compared to the monotonic increase predicted by ments and one field experiment. The erosion was investigated by
McLaury, et al., the penetration rates calculated with the prob- Postlethwaite and Nesic [25] in a complex flow geometry that
ability model show a different variation trend. As the particle size included a sudden expansion, a sudden constriction, and a groove
increases, the penetration rates firstly increase and then vary in a in a stainless steel pipe carrying dilute sand slurries in the absence
very narrow range. The particle size influences the penetration of corrosion. The resulting small pipe section was so long that it
rate in three ways. First, the turbulent fluctuations have great could be considered as a straight pipe. In the present study, the
effects on the small particles, as indicated by Eqs. (12) and (14). erosion observed in the downstream section of the sudden
Hence the small particles have a great chance to get enough constriction is picked up as Cases 1 and 2 in Table 5. The facilities
momentum to reach the pipe wall. Second, the large particles can used by McLaury [8] were similar to the above fittings. A choke
pass through the boundary layer more easily due to greater inertia, geometry constructed of aluminum was investigated and the
erosion in the constricted region is selected as Case 3 in Table 5.
The experiment performed by Wood, et al. [26] consisted of an
upstream straight pipe section followed by a bend section. The
erosion measured in the straight pipe is taken as Case 4. Cases 5 to
10 are from the field tests conducted by Okayama [27], in
which the wear in a dredge pipeline of a large steel pipe and a
polyurethane lined pipe was investigated under operational
conditions.

4.2.1. Laboratory experiments


The comparison of penetration rates between the present
model and laboratory experiments is listed in Table 6. A summary
of the results listed in Table 6 is illustrated in Fig. 8, where the line
denotes the points whose x-coordinate is equal to y-coordinate.
Fig. 6. Predicted penetration rates versus flow velocity.
The closer to the line a point is, the better the agreement between
the measured and predicted data. For Cases 1 to 4 from laboratory
experiments, good agreement between the present model and
laboratory experiments can be observed over a broad range of
penetration rates. In Case 4, the sand size is slightly big and results
in an uneven distribution of erosion in the vertical direction due to
gravity. Hence the average value is selected to test the accuracy of
the probability model.

4.2.2. Field tests


Okayama [27] provided the friction coefficient ϕ instead of the
absolute roughness Δ in the research. Then the friction velocity
can be calculated using the following expression:
rffiffiffi
ϕ
vn ¼ V m ð25Þ
Fig. 7. Predicted penetration rates versus sand size. 8

Table 5
Experimental cases [8,25–27]

Case D (mm) V m (m/s) dp (μm) W (kg/s) ρf (kg/m3) ρt (kg/m3) μf (Pa s) H V (GPa) Δ (mm) ϕ

1 21.2 13.3 430 0.249 995.6 7850 7.97  10  4 1.8 0.05 –


2 21.2 13.3 430 0.622 995.6 7850 7.97  10  4 1.8
3 6.48 19.8 155 0.013 1000 2700 10.6  10  4 0.24
4 77.8 3 1000 3.789 1000 7850 9  10  4 1.8
5 750 6.5 150 914.7 1000 7850 10.6  10  4 1.8 – 0.0125
6 1300 5.87
7 1300 6.58
8 1300 5.67
9 750 5.2 3500 732.3 1000 7850 10.6  10  4 1.8 – 0.0125
10 1300 5.87
8 R. Zhang et al. / Wear 308 (2013) 1–9

Table 6 radial direction is regarded as a Markov process which is only


Comparison of penetration rates between the probability model and experiments influenced by the eddy randomly appearing in the fully developed
(mm/year).
turbulent flow. After introducing a particle erosion formula, the
Case 1 2 3 4 5 probability model can calculate the overall penetration rate caused
by solid particles in the turbulent flow field by analyzing the
Experiments 7.63 18.10 104.61 0.71 7.47 random movement of a single particle.
Probability 10.44 26.07 93.20 0.66 14.37 The predictions from the probability model are generally lower
Case 6 7 8 9 10
Experiments 1.82 3.82 2.42 178.7 51.68
than those from numerical simulations performed by McLaury,
Probability 3.47 3.03 3.62 6.24 1.60 et al. [24]. The probability model predicts an increase in penetra-
tion rates as the pipe diameter decreases and the flow velocity
increases, similar to the trends obtained from numerical simula-
tions. While the penetration rates do not monotonously change
with the particle size. This is different to the results obtained from
numerical simulations. As the particle size increases, the penetra-
tion rates firstly increase and then vary in a very narrow range. The
comparison with the experimental data demonstrates that the

probability model can predict erosion well for particle sizes


ranging from 50 to 1000 μm and is applicable in a very broad
range. It should be noted that the particle shape would also affect
the penetration rate. However, the present probability model
ignores the effect of particle shape due to the complexity of the
problem, which has not been well known by researchers. In the
future study, further work may be tried to improve the expressions
of the drag force coefficient CD and the erosion ratio to account for
the effect of particle shape.
Although the probability model is developed for solving prac-
tical problems in engineering, the concept that the erosion
damage caused by a single particle can be considered as a random

event makes the procedure have more scientific meaning. In fact
Fig. 8. Comparison of penetration rates between the probability model and the probability model also contains three parts including the flow
experiments. modeling, the particle tracking and the erosion prediction, just like
the method adopted by McLaury [8]. While the probability model
As illustrated in Fig. 8, good agreement between the probability extracts the dominant factors and ignores the secondary factors
model and field tests can also be observed for Cases 5 to 8, in that influence the erosion results based on reasonable assump-
which the particles are smaller, yet an evident discrepancy arises tions and inferences. This makes the new model simpler and more
for Cases 9 and 10 (dp ¼3500 μm). effective in solving practical problems.
The hypothesis that the influence of gravity can be ignored, as
stated in Section 2.1, is reasonable for particle sizes less than
1000 μm. In this case, the influence of turbulence is so dominant
that the gravity almost contributes nothing to the particle motion Acknowledgments
and distribution. But for large particles, to which the gravity
cannot be neglected, the particle distribution in the flow field is Financial support from the National Natural Science Foundation
not uniform and actually the particle concentration is higher at the of China (Grant no. 51279129) is greatly acknowledged.
bottom of pipe. The contact mode between the particles and pipe The authors would also like to thank Mr. Simiao Wang and
wall is a plowing action due to gravity, rather than the impact- Mr. Mingyang Liu for their generous help with the work.
bouncing process for small and medium size particles. Further-
more, the particle motion in the radial direction is regarded as a
Markov process as assumed in Section 2.4. This assumption is only References
rational for small particles since large ones possess greater
momentum and therefore their states cannot be easily changed. [1] API, API RP 14E Recommended Practice for Design and Installation of Offshore
These mean that the probability model is inappropriate to predict Production Platform Piping System, 3rd edition, American Petroleum Institute
(1981) 1981; 22.
erosion caused by large particles, as in Cases 9 and 10. However in [2] S.J. Svedeman, A.E. Arnold, Criteria for sizing multiphase flowlines for erosive/
the oil and gas industry, the particles entrained in pipelines corrosive service, SPE Prod. Oper. 9 (1994) 74–80.
normally range from 50 to 500 μm [28]. Hence the probability [3] M.M. Salama, An alternative to API 14E erosional velocity limits for sand laden
fluids, J. Energy Resour. Technol. 122 (2000) 71–77.
model can be applicable in a very broad range. [4] K.G. Jordan, Erosion in multiphase production of oil and gas, in: Proceedings of
NACE International Corrosion 1998, San Diego, CA, 1998, Paper no. 98058.
[5] M.M. Salama, E.S. Venkatesh, Evaluation of API RP 14E erosional velocity
limitations for offshore gas wells, in: Proceedings of Offshore Technology
5. Conclusions Conference, Houston, TX, 1983, OTC 4485.
[6] A.T. Bourgoyne, Experimental study of erosion in diverter systems due to sand
In the present work, a probability model is presented in detail production, in: Proceedings of SPE/IADC Drilling Conference, New Orleans, LA,
1989, SPE/IADC 18716.
to predict the particle erosion in long straight pipes. Two mechan-
[7] S.A. Shirazi, B.S. McLaury, J.R. Shadley, E.F. Rybicki, Generalization of the API RP
isms lay the foundation of the probability model. First, the 14E guideline for erosive services, J. Pet. Technol. 47 (1995) 693–698.
distribution of the characteristic vortex length is deduced to [8] B.S. McLaury, Predicting Solid Particle Erosion Resulting From Turbulent
describe the turbulent flow in a pipe based on the energy Fluctuation in oil Field Geometries, Department of Mechanical Engineering,
The University of Tulsa, 1996. (Ph.D. dissertation).
equilibrium among eddies with different scales and the distribu- [9] J. Wang, Modeling Flow, Erosion and Mass Transfer in Elbows, Department of
tion of fluctuation velocities. Second, the particle motion in the Mechanical Engineering, The University of Tulsa, 1997. (Ph.D. dissertation).
R. Zhang et al. / Wear 308 (2013) 1–9 9

[10] J.K. Edwards, Development, Validation, and Application of a Three-Dimen- [20] G.K. Batchelor, The Theory of Homogeneous Turbulence, Cambridge Univ.
sional, CFD-Based Erosion Prediction Procedure, Department of Mechanical Press, Cambridge (1953) 1953; 169–170.
Engineering, The University of Tulsa, 2000. (Ph.D. dissertation). [21] P.G. Saffman, The lift on a small sphere in a slow shear flow, J. Fluid Mech. 22
[11] Q.H. Mazumder, S.A. Shirazi, B.S. McLaury, J.R. Shadley, E.F. Rybicki, Develop- (1965) 385–400.
ment and validation of a mechanistic model to predict solid particle erosion in [22] J.O. Hinze, Turbulence, 2nd edition, McGraw-Hill, Netherlands (1975) 1975;
multiphase flow, Wear 259 (2005) 203–207. 463–464.
[12] X. Chen, B.S. McLaury, S.A. Shirazi, A comprehensive procedure to estimate [23] Y.I. Oka, K. Okamura, T. Yoshida, Practical estimation of erosion damage caused
erosion in elbows for gas/liquid/sand multiphase flow, J. Energy Resour. by solid particle impact. Part 1: effects of impact parameters on a predictive
Technol. 128 (2005) 70–78. equation, Wear 259 (2005) 95–101.
[13] Y. Zhang, B.S..McLaury, S.A..Shirazi, E.F..Rybicki, A two-dimensional mechan- [24] B.S. McLaury, J. Wang, S.A. Shirazi, J.R. Shadley, E.F. Rybicki, Solid particle
istic model for sand erosion prediction including particle impact character- erosion in long radius elbows and straight pipes, in: Proceedings of SPE
istics, in: Proceedings of NACE International Corrosion 2010 Conference and
Annual Technical Conference and Exhibition, San Antonio, TX, 1997, SPE
Exhibition, San Antonio, TX, 2010, Paperno.10378.
38842.
[14] DNV, Recommended Practice RP O501 Erosive Wear in Piping Systems,
[25] J. Postlethwaite, S. Nesic, Erosion in disturbed liquid/particle pipe flow: Effects
Revision 4.2-2007, Det Norske Veritas, 2011.
of flow geometry and particle surface roughness, Corrosion (Houston, TX, US)
[15] A.D. Gosman, E. Ioannides, Aspects of computer simulation of liquid-fueled
combustors, J. Energy 7 (1983) 482–490. 49 (1993) 850–857.
[16] Y. Zhang, Application and Improvement of Computational Fluid Dynamics in [26] R.J.K. Wood, T.F. Jones, J. Ganeshalingam, N.J. Miles, Comparison of predicted
Solid Particle Erosion Modeling, Department of Mechanical Engineering, The and experimental erosion estimates in slurry ducts, Wear 256 (2004)
University of Tulsa, 2006. (Ph.D. dissertation). 937–947.
[17] J.T. Davies, Turbulence Phenomena, Academic Press, New York, 1972. [27] Y. Okayama, A study of the wear of a steel dredge pipeline and wear resistance
[18] A.N. Kolmogorov, Dissipation of energy in the locally isotropic turbulence, in: of a polyurethane lined pipe, Dredg. Port Constr. 13 (1986) 49–55.
Proceedings of Mathematical Physical Sciences, London, 1991, Paper no. 1890. [28] N.A. Barton, Erosion in Elbows Hydrocarbon Production Systems: Review
[19] J. Laufer, The Structure of Turbulent Flow in Fully Development Pipe Flow, Documents, Scottish Enterprise Technology Park, East Kilbrede, 2003.
National Advisory Committee for Aeronautics, Washington, DC, 1954. (NACA (Research Report 115).
Report 1174).

Das könnte Ihnen auch gefallen