You are on page 1of 13

SPE 84499

Predicting and Managing Sand Production: A New Strategy


Ian Palmer (BP*), Hans Vaziri (BP), Stephen Willson (BP), Zissis Moschovidis (PCM), John Cameron (PCM),
Ion Ispas (BP)
Copyright 2003, Society of Petroleum Engineers Inc.
This paper was prepared for presentation at the SPE Annual Technical Conference and
Exhibition held in Denver, Colorado, U.S.A., 5 8 October 2003.
This paper was selected for presentation by an SPE Program Committee following review of
information contained in an abstract submitted by the author(s). Contents of the paper, as
presented, have not been reviewed by the Society of Petroleum Engineers and are subject to
correction by the author(s). The material, as presented, does not necessarily reflect any
position of the Society of Petroleum Engineers, its officers, or members. Papers presented at
SPE meetings are subject to publication review by Editorial Committees of the Society of
Petroleum Engineers. Electronic reproduction, distribution, or storage of any part of this paper
for commercial purposes without the written consent of the Society of Petroleum Engineers is
prohibited. Permission to reproduce in print is restricted to an abstract of not more than 300
words; illustrations may not be copied. The abstract must contain conspicuous
acknowledgment of where and by whom the paper was presented. Write Librarian, SPE, P.O.
Box 833836, Richardson, TX 75083-3836, U.S.A., fax 01-972-952-9435.

Abstract
Sand prediction at BP has been developed by dividing it into
three parts: (1) onset, (2) transient sanding, (3) steady-state
sanding. For example, as drawdown is increased in a well in a
sand-prone formation, significant sanding begins at some
point (the onset), and this is followed by a transient sand burst,
which may last hours or days or months. The sanding
eventually declines to a background level (steady-state), in the
range 1100 pptb. We have made recent step-changes in (2)
and (3), and we now have a tool that can predict sanding
onset, and volumes during any stage of a wells production
history, or even injection history.
The onset of sanding is predicted using a stress-based
model. This model is conservative, based on a benchmarking
study of field applications. One application predicts sanding in
water injectors during shut-in, and recommends not using sand
control. Another application explains delayed sanding in an
HPHT gas reservoir, in terms of restraining forces due to
capillary cohesion (i.e., the damp sand effect at the beach).
The transient sanding model is a fully-coupled finite element
(FE) model. The model has successfully predicted sand
volumes in laboratory and field tests. With this model, we can
judge whether we can manage the produced sand, from both
production and injection wells. Finally, we have applied this to
predict whether a well will kill itself after a blowout, due to
sand in the wellbore increasing the hydrostatic pressure.
The steady-state model is an empirical model that is based
upon extensive tests of sanding from cores in the laboratory.
The model has been applied to predict sanding in an offshore
field, and this has led to the conclusion that sand rates can be
managed at surface, without sand control: a huge economic
advantage. Finally, we present a case history where we use all
three models to make an integrated prediction of onset,

now at Higgs Technologies, Houston

transient, and steady-state sanding, and find quite good


agreement with field observations. In summary, the new threefold strategy of sand prediction at BP has significantly
increased our capability to predict sanding in production or
injection wells. This quantum leap is invaluable to help decide
if we can manage the sand at surface (or if downhole sand
control is required); to decide if we can defer sand control
until a later date (possibly increasing production as well as
saving completion costs); to decide how much sand will be
produced if we increase the drawdown in a sand-prone well;
and even decide if we need sand control in water injectors.
Introduction
Sand prediction at BP has progressed in three stages: (1) onset,
(2) transient sanding, (3) steady-state sanding. As drawdown
is increased in a well in a sand-prone formation, significant
sanding begins at some point (the onset). Alternatively, the
trigger may be an increase in depletion. This is followed by a
transient sand burst, which may last hours or days or months.
The sanding eventually declines to a background level
(steady-state), in the range 1100 pptb. Figure 1 is a summary
of the three stages.

onset of sanding
quantum leaps in two areas
to complement prediction
of failure onset

transient
volume

steady state
sand rate

Figure 1. Strategy of sand prediction as a function of time, from


onset through transient to steady-state

Recent step-changes have been made in the transient and


steady-state stages, and we now have models that can predict
sanding onset, and volumes during any stage of a wells
production history, if it is an oil well. For a gas well, we can
predict the transient sand volume, but not the steady-state sand
rate. In this latter case, the fundamental methodology is

Note: when we speak of sand production, this may be equated


with solids production, as the modeling in this paper applies
equally well to carbonates or other formations.

The criterion for sanding is:2

CBHFP <

3 1 3 y
2 A

Pr

A
2 A

(1)

where CBHFP is the critical bottomhole flowing pressure, Pr


is the current average reservoir pressure, 1 and 3 are the total
principal major and minor stresses, A is a poroelastic constant
which is a function of the Poissons ratio and formation
compressibility, and y is the formation strength near the
opening:

y = 3.1 TWC

(2)

where TWC is strength as determined in the thick-walled


cylinder test. The factor 3.1 includes the scale transformation
from TWC laboratory sample (OD:ID = 3) to field (OD:ID =
infinity).
Note that 1 and 3 may depend linearly on the reservoir
pressure Pr . Therefore, Equation (1) should not be used with
constant 1 and 3 values for cases where reservoir depletion
effects are considered.
While there is some support for advancing the notion that
sand failure and sand production can be considered to be
essentially the same event, a number of field cases have shown
that failed sand can remain in a stable state until fluid and flow
conditions reach a certain level to produce the disaggregated
sand.
Figure 2 shows a typical CBHFP (oblique line sloping up
to the left). CBHFP is approximately 300 psi at the start of
field development. As the reservoir depletes, CBHFP gets
higher. The critical drawdown pressure (CDP) at the start of
production (~4,500 psi) is shown as the maximum drawdown
arrow. With depletion, this drawdown arrow becomes smaller
as it moves to the left. When the reservoir is depleted to
around 1,700 psi, any drawdown at all will lead to sanding.
6000

Max Depletion

5000

4000

3000

Sand Free Zone


2000

Max Drawdown

Onset of sanding
Model summary
The onset of sanding is predicted using a stress-based model
of shear failure around a perforation or an open hole wellbore.
Essentials of the BP sand onset model are:
Predicts shear failure around a perforation or an open hole
(but this may not predict when sand actually enters a well)
Predicts the onset of sand production in cased and
perforated and open hole completions using a combination
of empirical and analytical relationships.
The essential inputs to the model are thick-walled cylinder
tests (TWC) obtained from cores tested in the laboratory,
and unconfined compressive strength (UCS) predicted
from logs (gamma-ray, density and dipole sonic).
The TWC collapse strength corresponds to the point of
significant sanding (equivalent to development of many
shear bands that eventually coalesce).
Analysis is performed at the weakest point of the UCS log.

The UCS log is calibrated to the measured TWC.


Sand production is assumed to occur once the maximum
value of the effective tangential stress around the
perforation exceeds the apparent UCS (i.e., the perforation
fails at the same cavity loading as occurs in the TWC test).
No consideration is given to sand transport by drag forces.
The BP model can account for different orientations of the
well or perforations.

CBHFP (psi)

established, but to date the necessary calibrating laboratory


data have yet to be collected. We have even been able to
apply the new strategy to predict onset and transient sanding
in an injection well, after such a well has been shut in.
However, the modeling has some limitations. We are not
yet able to obtain the profile of the transient sand event at the
surface (i.e., sand volume vs. time). For example, if the
drawdown is suddenly increased, we can compute the total
volume of sand influx into the well, but not the rate of influx
or the total time it takes. Clearly this depends on the rate of
sand uptake by the incoming fluid (e.g., factors like drag
force, velocity and viscosity, and whether the fluid is oil or
gas). The sanding profile at surface also depends upon the
velocity and slippage of sand particles relative to the fluid
flow up the well. Although this part of the model is still being
developed, we are able to present some crude modeling to
illustrate the results.
The main driver of this work has been to decide if a well
can be completed without sand control. If we can reliably
predict the volume and concentration of sand, we can decide if
sand can be handled at surface, and better design the facilities
for handling sand. The potential advantages of no sand control
are cheaper well installation, higher production rate, and
ability to shutoff water. Another driver is to be able to assess
whether an increase in drawdown can boost a wells flow rate,
without excessive sand produced. A third aspect is to be able
to evaluate whether a cased/perforated completion can be used
initially, and to defer a sand control option until later, with
cost and flow rate and environmental advantages.1 In
summary, there is often a substantial cost benefit both for
capital and operating expenditure if sand management can
be deployed successfully.
Below, we summarize the modeling used in each of the
three stages, so the overall strategy is clear, and we present
new applications. We conclude with some implications about
sand management.

SPE 84499

1000

0
0

1000

2000

3000

4000

5000

6000

Reservoir Pressure (psi)

Figure 2. Sand prediction using shear failure model for cased and
perforated vertical well

SPE 84499

predicted / observed CBHFP


4
lower limit: no sand, as
grains held together by
capillary forces

CBHFP ratio

conservative
predictions:
BP model predicts
sanding too early.
This is safety factor

2
1

over-optimistic prediction:
not consistent with BP model

0
-1

500

1000

1500

2000

2500

3000

3500

lower limit: no sand


still consistent with BP model

-2

1 / 7 cases BP
model is overoptimistic.not
good

log-UCS min (psi)

Figure 3: Comparison between CBHFP predicted and that


observed in the field

Two of these cases are open hole: Immortelle and Mahogany.


Note: two cases in Table 1 have a negative ratio of
predicted/observed CBHFP. This is because the predicted
CBHFP is negative, i.e., no sanding is expected, even if the
reservoir pressure is drawn all the way down to zero.
Our conclusions from Figure 3 are:
Five of the seven predictions have predicted/observed
CBHFP 1 meaning the BP model is conservative (i.e., the
field were able to go to lower BHFP than predicted,
without sand). Being conservative is acceptable, as there is
a safety factor in-built to the prediction.
Only one prediction is definitely too optimistic. This is a
field in North Sea, which was predicted to be safe
(negative CBHFP), but sanding occurred (we have not
found an explanation for this).
One other case has negative predicted/observed CBHFP,
but no sand has been observed yet, so we dont know if the
model is conservative or not.
Three out of seven of the predictions use 3 1 - 3 in
Equation (1). This means keeping 3 as the far-field stress
instead of resetting it to 1 (i.e., this does not double-dip
effect of stress concentration around wellbore). This is the
preferred method of application, but it makes the model
prediction even more conservative (i.e., CBHFP is larger).
The main conclusion is that the BP prediction is generally
conservative. That is, sanding occurs at lower BHFP than
predicted. And it can be very conservative, for example:
actual CBHFP = 4,500 vs. 16,000 psi predicted at
Tuscaloosa (largest discrepancy of the 5 conservative
wells)2
actual CBHFP = 3,500 vs. 7,000 psi predicted in a GOM
Shelf well (only one well)
The conservatism of the BP model, which predicts shear
failure of the formation, may be explained by:
residual cohesion holding sand grains together (i.e., due to
connate water, or interlocking grains)

stress reduction due to arching at the tip of the perforations


A comprehensive study of benchmarking for open hole
horizontal wells has not been done yet (the exception being
the two cases in Table 1). Nevertheless, recent predictions for
three open hole wells in the North Sea indicated no sand, and
for the first 6-12 months of well life this was the case.
The cased perforated results of Table 1, and the success of
BP sand prediction for open hole wells, means the BP onset
sanding prediction is viable, although conservative (in all but
one case). Furthermore, the predictions in four out of seven
cases from Figure 3 give:
CBHFP (pred) = (1 2) CBHFP (obs)
CBHFP (obs) = (0.5 1) CBHFP (pred)
and this can be useful as a quick rule-of-thumb.
Delayed Sanding
One class of problems where the BP sand prediction model
has not been well validated is in HPHT wells. As discussed
by Vaziri et al.,2 the shear-failure model is overly conservative
in predicting the onset of sanding. This is emphasized by
Figure 4, which shows, for five wells in Tuscaloosa, the
CBHFP predicted by the BP model, and by a service company
model (the two segmented lines agree fairly well). But the
field observations are shown by the blue squares, where
sanding has not yet occurred. Thus the model predicts sanding
much too early. This may be due to the fact that for a given
rock strength, failure is hastened in high stress systems. The
early failure of rock makes the difference between sand failure
and sand production more pronounced.
18000
Minimum Bottomhole Flowing Pressure (psi)

Benchmarking of onset model


Sanding onsets have been predicted in seven field cases,
which are summarized in Table 1. Sanding has been observed
in five out of seven cases. These are a variety of field cases:
four are gas reservoirs, the minimum UCS ranges from 250 to
3053 psi, and two of the cases can be classed as HPHT. We
can use this data set to compare sanding prediction with actual
observation (see Figure 3).

16000
14000
12000
10000
8000

Shear Fail Model-BP


Reality
Shear Fail Model- 3rd Party

6000
4000
2000
0
P-4/B3

P-8/C1

P-10/B8

P-11/C1

LL-1/B1

Well Name/Sand Unit

Figure 4: Comparison of the predicted critical BHFP against


observed minimum BHP

Do tensile failure models do any better? We illustrate this by


making predictions using the ARCO model3, under conditions
of turbulent flow. The field data indicate that very high flow
rates had prevailed throughout most of the production phase;
average rates in the order of 500 MSCFD/ft had been
recorded. In fact, the flow rate through some of the higher
permeability zones was probably several times higher than
this. Following Vaziri, et al.,2 the turbulent flow-corrected
results are shown in Figure 5. The reduction of CDP due to
turbulence is generally 25-50%, and we have used 25%. Using
50% would make the discrepancy with observation (blue dots
in Figure 5) worse. The interpretation is that the tensile failure

SPE 84499

12000

Tensile Fail Model


Reality
Regression Model

Drawdown Pressure (psi)

10000

8000

6000

4000

2000

0
P-4/B1

P-8/C1

P-10/B8

P-11/C1

LL-1/B1

LL-1/A4

Well Name/Sand Unit

Figure 5:
Comparison of the predicted critical drawdown
pressure against observed maximum drawdown pressure

For comparison, we include results of a fully-coupled


numerical model, ENHANS, in the form of a simplified
regression formula. This accounts for the sand failure
(disaggregation from intact to granular state) but does not
assume sand production until the disaggregated sand
undergoes a liquefied state, that is, a state of zero effective
stress among the sand particles. In this state, sand is assumed
to have reached a mobile state and is thus producible. Again
following Vaziri, et al.,2 we have plotted the results of the
regression formula in Figure 5 (red line).
The regression formula shows CDP values that all fall
above the maximum drawdown values observed, suggesting
that most of the wells could have been drawn down further
before sanding would occur. The prediction agrees with the
observation, in the sense that both indicate no sanding yet. The
regression formula can also calculate the volume of sand for
any drawdown level, in excess of that shown for the onset of
sanding, and hence allow an engineer to assess the risks
associated with adopting a more aggressive drawdown
strategy.
In summary: for this HPHT field case, the usual shearfailure models (e.g., BP) and the tensile failure models (e.g.,
ARCO), both predict sanding onset that is too early (i.e., both
models are conservative). The numerical-based model,
ENHANS, lies on the other side of the data, suggesting these
wells could have been drawn down even more before sanding.
Note that sanding did eventually occur in several of these
wells, coincident with water influx, and this is discussed
below under Water-induced sanding.
Transient sanding volumes
The transient sanding model is a fully-coupled FE model
called ENHANS. The model has successfully predicted sand
volumes in a large-block test, in a drawdown-induced transient
in the field that lasted a few months (and increased flow rate
by > 50%), and in a choked-back well that was opened up
until the well stabilized at a higher flow rate. With this model,
we can judge whether we can manage the sand, and we can
even predict the increase in well productivity, due to porosity

increases in the formation. Finally, we now have a means to


predict whether a well will kill itself after a blowout, due to a
sand in the wellbore increasing the hydrostatic pressure.
Essentials of the model are:2
Predicts tensile failure around a perforation or an open hole
(i.e., does predict when sand will enter a well)
Fully coupled stress and fluid flow finite element analysis.
It allows time-dependent simulation of applied boundary
conditions.
Any element that develops a liquefied state is assumed to
have failed. The element is smeared (not removed) and
given properties of a liquefied zone (e.g., high perm, loss
of cementation, reduced moduli).
The volume of sanding can be rigorously determined. As
such the model can be used not only to predict the onset of
sanding but also the severity of sanding.
The model computes the improvement in wellbore skin
with sanding, so can easily be converted to an increase in
well PI.
Key parameters are the real and projected cohesion.
The model has shown good agreement when compared
with volume of sanding in weak sands (laboratory and two
field cases).
Water-induced sanding
As described by Vaziri, et al.,4 failed and disaggregated sand
can be held together remarkably well by tiny capillary
cohesion forces, due only to connate water, when water
saturation is low (i.e., a sand-castle made with damp sand).
But when water saturation increases with water influx, this
capillary cohesion is extinguished, and sand will enter the well
and be produced (i.e., sand-castle collapses as tide comes in).
The numerical, fully-coupled FE model ENHANS has been
able to model this effect, and Figure 6 shows a typical result.
There is very little sand production before water influx. But
when this occurs, the sand production increases dramatically,
by a factor of ~5, if the capillary cohesion goes away
completely. In short, this explains why in many wells, sand
production accompanies water influx.
Influence of capillary adhesion after water production

Volume (ft^3 per 1.5 m of pay length)

model is also conservative, but not as much as the BP shear


failure model.

25

Base case properties:


Cip = 350, Cir = 15, Cpp = 145, Cpr = 7, Ctp = 10, Ctr = 1.5, Cppw = 135

20

15
Cprw = 0.3 psi
Cprw = 5 psi

10

water production

Cprw = 0.7 psi

0
0

-500

-1000

-1500

-2000

-2500

-3000

-3500

Drawdown (psi)

Figure 6: Influence of the capillary cohesion force within the


failed zone on the volume of sand after water production

SPE 84499

P1

P2

shutdown this will lead to cross flow, which may bring in


sand. The illustrative cross flow scenarios we consider are
those for which Pb - Pa = 100, 200, 300, 400, 500 psi. At
shutdown, these are the effective drawdowns that can bring
in sand.
Min Allowable Pressure in Zone A, to Avoid
Crossflow Sanding
min res pressure for zone A
(psi)

Illustrative modeling of water injection wells - onset


and transient sanding
In a weak reservoir that is to be water-flooded, an important
question is whether the water injectors need sand control. That
is, what is the likelihood of sand influx into a well during
injector shutdown? If water injectors can be completed
without sand control (i.e., just casing and perforations), the
advantages are substantial as the completion:
allows zonal isolation;
avoids inevitable screen or gravel-pack plugging, and
injectivity loss (it is very difficult to clean or replace a
screen economically);and
the total cost savings can be tens of millions of dollars,
or more

6000

normal Pa = 4900 + 100 psi frac pressure = 5000 psi

5000

no sand

4000
sand

2000
if UCS < 200 psi,
predict sanding
due to crossflow

0
0

result is
Pb Pa may
be up to
500 psi

Pa

Does this
produce sand
via crossflow
after
shutdown?

Pb
sealing fault or pinchout
causes Pb > Pa

Figure 7: Multizone injector with potential cross flow from zone B


to zone A

According to Santarelli, et al.,5 cross flow-induced sand can


fall into the rathole, or be carried into perforations of another
zone, where the sand can completely fill the perforationss (a
downhole TV failed to see plugged perforations in one case).
This perforation plugging by sand is the cause of injectivity
loss (not fines plugging). There is a simple field test for cross
flow sanding: if the injectivity index (II) decreases after each
shutdown then cross flow sanding is occurring. The II can be
measured using programs to calculate BHP from WHP.
There are two sources of sanding in an injector during
shutdown: (a) fluidization of the formation due to water
hammer (not evaluated here), (b) cross flow-induced sanding,
which itself has two origins:
pressure differential caused by diffusivity difference
between two pay zones.
pressure differential between two pay zones, due to sealing
fault or pinch-out (Figure 7).
From some typical reservoir modeling we have done, the
first source of cross flow sanding leads to typical
drawdowns that are small, <50 psi. However, the second
source leads to typical drawdowns that can be much larger,
100-500 psi. Thus the latter will dominate over the former.
To understand the sanding potential, one approach is to
calculate sand onset for cross flow scenarios caused by
potential sealing faults or pinch-outs, and to estimate sand
volumes for the same scenarios. This two-pronged approach
provides a stronger basis from which to decide whether
injectors need sand control. As Figure 7 reveals, a pressure
differential may exist within a multi-zone injector, and at

Pb = 4900 psi
+ 100 psi frac
pressure

3000

1000

Prediction
based on
BP model

200

the difference
between the 2
curves is in
the noise

Pb = 4900 psi
+ 500 psi seal
pressure

400

600

800

1000

1200

log-UCS min (psi)

Figure 8. Onset of cross flow sanding, for different UCSmin


values, and different Pa values. Pb lies in the range 5,000-5,400
psi. Cross flow is from zone B to zone A.

The BP model is used to predict cross flow sanding, based on


a typical log profile of UCS in zone B (calibrated to laboratory
measurements of TWC), and the results are shown in Figure 8.
The plot shows the minimum allowable pressure in zone A, to
avoid cross flow sanding from zone B to zone A. The results
are virtually the same whether Pb is elevated to 5000 or to
5400 psi. In short, if the minimum UCS ~500 psi, Pa would
have to fall to ~3,500 psi to get any cross flow sanding, even if
Pb were elevated to 5,400 psi by a sealing fault.
In the worst case scenario, Pb rises to 5400 psi (e.g., due to
a sealing fault) and Pa stays at about 5,000 psi (expected
situation). From Figure 8, the BP model predicts no cross flow
sanding unless min UCS <200 psi. Minimum UCS refers to
minimum UCS in zone B, as determined from a sonic log,
from where the cross flow enters the well. In a case that all
UCS (sonic log) > 500 psi, there should be no cross flowinduced sanding, and there would be quite a large margin of
safety.
This cross flow sanding prediction is based on a horizontal
perforation (i.e., the weakest perforation), and therefore should
be applicable to wells deviated from the vertical, if
perforations are not oriented. However if sanding did occur in
a highly deviated well, the sand may not fall into the rathole,
and this could exacerbate injectivity loss.
Predicting sand volumes during cross flow
The BP model only addresses the onset of significant sanding,
and it cannot predict the transient volume of sand caused by a
change in drawdown. But estimates of volumes would be
invaluable to quantify the risk of sanding, and below we
illustrate how we have done this using the transient sanding
model ENHANS. Morita et al.6 have done similar modeling of
sand production in injection wells.
In multi-zone injectors, the volume of sand induced by
injector shutdown depends on the drawdown between zone
B and zone A (see Figure 7). We assume 500 psi worst case,
since these injectors may lead to potential drawdowns of

SPE 84499

sand volume (bbl)

5
4

20
8 inch rathole fillup (ft)

100-500 psi between pay zone B and pay zone A. Then using
the ENHANS numerical model, we estimate a sand volume of
1.6 4 bbl (the uncertainty lies in the bilinear Coulomb plot of
strength vs. stress). This volume of sand would fill up 6-16 ft
of 8-inch rathole. We can classify this worst-case situation as
a P10 case. The volume ranges are plotted in Figures 9 and 10.

15
min rathole fillup

10

max rathole fillup

5
0
P10

min sand vol

max sand vol

1
0
P10

P50

P90

Figure 9: Summary of sand volume risk after cross flow:


water injector

The other component of cross flow sanding, due to


diffusivity differences between zones A and B in Figure 7,
leads to smaller sand volumes because typical drawdowns
are lower (<50 psi) and substantially smaller than the previous
values of 100-500 psi. Sand volumes are predicted in the range
0.5-1.0 bbl. We may regard this result as a P90 calculation of
sand volume, since it is likely to occur whether or not there is
any cross flow sanding due to a pressure differential between
zones A and B in a multi-zone injector. That is, single-zone
injectors will likely be subject to this component of transient
sanding, and so the P90 result applies to them. Obviously, the
risk of sanding for single-zone injectors is lower.
A P50 result is obtained by assuming a drawdown of
150 psi in the multi-zone injector case (recall the range of
pressure differential between zones A and B is 100-500 psi).
This leads to a prediction of sand volumes in the range 1.2-3.3
bbl.
We can summarize the risk of sand volumes in water
injectors by showing the P10, P50, and P90 results in Figures
9 and 10. Figure 9 shows the sand volume risk, while Figure
10 shows the rathole filling risk. Clearly, a deeper rathole
would be an advantage.
For gas injectors, the sand volumes in Figures 9 and 10
will be reduced by typically 5 times, that is, gas injectors are
much safer than water injectors because capillary cohesion can
hold disaggregated sand grains together.
Note: There are caveats in the calculations we have made of
sand volumes:
The volume estimate only applies to the first shutdown.
Later shutdowns are also expected to produce some sand,
but in diminishing volumes.
The calculations are for instantaneous shutdown. For
slower shutdown, sand volumes can be substantially less.
Nevertheless, these sand volume calculations illustrate how a
potentially powerful tool can help us decide whether to install
sand control in water injectors.

P50

P90

Figure 10. Summary of rathole filling risk after cross flow:


water injector

The profile of transient sanding versus time


If the drawdown in a well is suddenly increased, we can
compute the total volume of sand influx into the well, but not
the rate of influx, or the total time it takes. Clearly this
depends on the rate of sand uptake by the incoming fluid (e.g.,
factors like drag force, fluid velocity and viscosity, and
whether the fluid is oil or gas). The sanding profile at surface
also depends upon the velocity and slippage of sand particles
relative to the fluid flow up the well. Although this part of the
model is still being developed, we can do some crude
modeling to illustrate the results.
Assuming a single sudden increase in drawdown, there are
two steps:
Calculate the maximum percentage by volume of sand that
is picked up by the fluid as it enters the well (one way is a
drag force approach). This gives the time-profile of sand
entering the well. For example, the sand influx will be
constant, if flow rate is constant (and other fluid
parameters), and if only one sudden drawdown increase
has occurred that causes disaggregation of a large volume
of sand. That is, we assume the failure of the formation
takes much less time than the sand grain uptake by the
flowing fluid.
Calculate the dispersion of the sand as it moves with the
fluid up the well. This is complex, and involves slip of
sand grains through the fluid, which is a function of grain
size, fluid viscosity, velocity, well deviation, etc. If the
sand event at the sand face is small (i.e., an impulse), the
result at surface is a rapid onset of sanding followed by a
gradual decline (i.e., a transient decay response). If the
sanding event is large at the sand face, the surface signal
may be:
flat with time, if the dispersion is small;
gradually increasing with time, followed by a decline, if
dispersion is large.
Dispersion can be calculated by modeling the flow of fluid
and sand grains up the wellbore. To make things simple, we
assume here no dispersion, so that the time profile of the sand
at surface is the same as that at the sand face. Then the
problem reduces to the first bullet above. We first compute the
drag force, from the fluid properties and the sand size. Then
we use a correlation we have developed for uptake of sand (%
by volume) as a function of drag force. From the flow rate, we
can calculate how long the uptake of sand ensues, until the
total disaggregated sand volume at the sand face has influxed
into the well.

SPE 84499

Figure 11. Sand volume and failure zones, assuming creation of a


cavity-like geometry

total loss of cementation. The dark blue line shows the radius
of the zone experiencing sand production (tensile failure), the
so-called cavity zone.
The yellow line in Figure 11 shows cumulative sand
increasing with drawdown, and reaching ~13 bbl at the end.
We can now predict whether the well would have killed itself.
We do this by first calculating the drag force acting on a
typical sand grain,7 and then comparing it with the results of
Figure 12.
80
70

Upper Limit
60
Carrying Capacity (%)

Application to well blowout


We consider here an illustrative vertical gas well, cased and
perforated, with pay zone at approximately 10,000 ft depth,
and producing ~ 10 MMcfd. Suppose the original reservoir
pressure was over-pressured, and approximately 7,500 psi.
In a typical situation, flow rate is maintained by gradually
increasing the drawdown as the reservoir depletes.
The finite element code ENHANS was used to predict the
likelihood of sanding under a scenario in which the drawdown
reached 650 psi when the reservoir had depleted by 3,200 psi
(i.e., total drawdown defined as depletion + drawdown was
3,850 psi). One of the key parameters in the modeling is the
in-situ sand cementation. This is actually the cohesion due to
the sum of mechanical cement and capillary cohesion due to
connate water. We assumed the capillary cohesion to be 4 psi
and the mechanical bonding to be 16 psi. Using a lower total
cementation would result in sanding at much lower total
drawdown. For reference, a total cementation of 20 psi is
equivalent to a UCS of approx 60 psi (this is very weak sand).
In the simulation, we modeled both depletion and
drawdown. The scheme used to vary these with time was
arbitrary, but based on limited parametric runs we did not see
an appreciable influence of this. At the time of sanding, it is
reasonable to assume that all the mechanical cementation has
been destroyed due to a depletion of >3,000 psi. And
evidently the capillary cohesion due to connate water is not
sufficient to hold the sand grains together.
Figure 11 shows the sanding prediction. Another key
parameter that affects the sand volume is the state of the sand
behind the casing after the sand event. We assume that sand
production leads to a cavity (which could be a zone of
disaggregated sand with much higher porosity than the
formation sand). This results in a stable condition. The much
higher permeability in this zone creates a sharp drop in the
pressure gradient, which leads to stability when the well is put
back on line (i.e., no more sand).
In Figure 11, the X-axis represents the total drawdown
(depletion + drawdown). The left hand Y-axis shows the
drawdown (light blue line). The yellow line shows the volume
of sand produced in bbls. Sanding is predicted at a drawdown
of around 610 psi. At this point the mechanical cementation
started to break down and was totally destroyed by the time
the drawdown reached about 650 psi (note that there is still 4
psi of capillary cohesion). The sanding level escalates rapidly
as cementation erodes away and culminates in about 12 bbls
of sand. A brief shutdown is modeled after the drawdown has
reached 650 psi, and during this time, sand continues to be
produced and goes up from about 12 to 13 bbls. After this, the
well was beaned back up to a drawdown of 650 psi, and then
the drawdown was increased more rapidly, up to 970 psi (see
blue line in Figure 11). During this period, sanding increases
but only by a small amount. This shows that if a cavity-like
sand state develops around the well, sanding will not be retriggered. For completeness, the pink line represents the
shear-failed or plastic zone: this is the zone that has been
impacted by the

50

y = 14.373x - 0.2952

40
30
20
10
0
0.00

0.50

1.00

1.50

2.00

2.50

3.00

3.50

4.00

Drag Force

Figure 12. Correlation between sand-carrying capacity of fluid,


and drag force on a sand grain

For the well of Figure 11, producing ~10 MMcfd, the drag
force is high, and the sand-carrying capacity is at the upper
limit of the plot in Figure 12 (i.e., 60% of sand by volume in
the well). About 12 bbl of sand would occupy a sand column
of only 426 ft in this 10,000 ft well. The extra hydrostatic
pressure due to this column of sand is only 316 psi. When
added to the gas column in the rest of the well (727 psi), we
get a total hydrostatic pressure of 1,043 psi. This could not kill
the well, since the BHP just inside the well is 3,700 psi
(depleted reservoir pressure minus drawdown). We estimate
that >100 bbl of sand would be needed to kill the well. Note
however, we have not investigated possible well self-kill by
bridging of sand in tubing, or by flow surging which can
increase friction pressure and thereby increase BHP.
Note that if the sand-carrying capacity were 30% instead
of 60%, the height of sand column would be twice as high, but

SPE 84499

the hydrostatic pressure due to the sand column would be


almost the same (this is true only if both sand columns do not
exceed the well height).
The important point is that we now have a tool to calculate
transient sand volume during a well blowout, and to estimate
whether the well will kill itself by hydrostatic
pressure increase.

10000.00

5
Formation Permeability (mD)

4.5

Formation Permeability (mD)

Sand Production Rate (lbs/ft)


1000.00

4
3.5

100.00

3
2.5

10.00

2
1.5

1.00

1
0.5

0.10
9750

0
9775

9800

9825

9850

9875

9900

9925

9950

9975

Sand Production Rate Rate Per Half-Foot Interval


(lbs/day)

Steady-state sanding
The steady-state model is an empirical model that is based
upon extensive tests of sanding from cores in the laboratory
(Willson, et al.8). A non-dimensionalized approach is used to
combine and interpret laboratory sand production
experimental data. This includes a Loading Factor concept
that allows the derived sanding rate model to be consistent
with existing models for predicting the onset of sand
production. A Reynolds Number concept includes fluid
flow effects, and is well documented from perforation cleanup research. Lastly, an empirical water boost factor, which
accounts for the effects of water production, is corroborated
by field evidence.

10000

Depth (feet TVD.SS)

Figure 13. Predicted Distribution of Sand Production For Well B/1


For Specified Producing Conditions

Then we have:
Sanding rate = f (Loading factor, Reynolds number, Water
boost factor)
When applied to field examples from sand producing wells,
the new model is seen to perform well when compared with
the measured data.
The model has been applied to predict continuous sanding
in an offshore field,8 and has led to the conclusion that sand
rates can be managed at surface, without sand control. This is
a huge economic advantage. Figure 13 shows the predicted
sand influx versus depth for well B/1 for the following
specified producing conditions: 29,690 bpd gross liquid
production; 77% water-cut; 592 psi drawdown and 265 psi
depletion. The overall predicted sand production for the entire
perforated interval is 119 lbs/day, equivalent to a sanding rate
of 4 pptb. Also shown in Figure 13 is the formation
permeability distribution. This correlates well with porosity
and inversely with formation strength (high permeability, low
strength). The figure shows that a high permeability streak
from 9927 ft to 9930 ft TVD.SS is predicted to produce 12 lbs
of sand per day, approximately 10% of the overall predicted
total. Therefore, if sand production rate and erosional

constraints were of concern in this well, then it may be


prudent to omit perforating this 3-foot long interval.
Integrated sand prediction
Background
An example from the field, described previously by Vaziri, et
al.2, is reexamined here to demonstrate how all three phases of
sand prediction can be integrated (onset, transient, and steadystate) in a deviated well. The well is referred to as A/3 of Field
A for the purpose of this paper.
This unconsolidated sand reservoir produced sand from
early times: well A/3 has been online since September 1999.
In early 2001 the well was producing at 9,000-10,000 bopd
(no water), when a decision was made to try to increase the
flow rate by either gravel packing the well, or just opening the
choke. Near the end of April 2001, the choke was opened, and
a prolonged transient sand burst occurred, lasting about 3
months (see Figure 15). The flow rate increased to ~15,000
bopd, which is a prime example of how to increase a wells
productivity by deliberately producing sand (see also Palmer,
et al.9).
A comprehensive data gathering and logging program was
performed for Field A, as also described previously by
Willson, et al.8. The in-situ stresses were estimated from logs
(e.g. sonic logs, density log, etc) and were calibrated to
measured stresses from leak-off tests. The following is a
summary of all the data needed to make the various sand
predictions:
Sanding event data (pptb, choke size, flow rates, BHFP)
UCS log and TWC log (estimated from logs and calibrated
to measurements in the laboratory)
Cohesion log (estimated from logs)
In situ stresses (overburden, horizontal stresses, reservoir
pressure)
Drawdown pressure changes with time
Permeability log
Fluid viscosity
Completion information (shots per foot and perforation
diameter
Onset of sanding prediction
The in situ strength of the reservoir was estimated from sonic
logs and given in terms of UCS (unconfined compressive
strength). Clean sands (GR < 40 API units) have lower
strength than shaly intervals, and the weakest interval has
UCS ~ 2000 psi.
Figure 14 shows the results of our onset prediction, for the
deviated A/3 well, producing at a rate of 15,000 bopd through
a cased-and-perforated completion (CHC). These results
apply to the transient event that began in late April 2001 as
shown in Figure 15. The CBHFP (Critical Bottom Hole
Flowing Pressure) is the lower limit of the BHP to avoid sand
production.

SPE 84499

9
A/3 Sand Cleanout April-July 01
Av-Sand (pptb)

5000

360

4000

320

3000

280

2000

240

1000

200
160

0
-1000

120

-2000

80

-3000

40

-4000
15300

0
15400

15500

15600

15700

15800

15900

16000

16100

16200

MD (ft)

Figure 14. Sanding onset prediction for deviated A/3 well

We notice in Figure 14 that the BHP of the well is lower than


the CBHFP at a depth 15604.76 ft MD (indicated by the
horizontal arrow), and therefore the formation there is
expected to fail next to the wellbore. This is viewed as an
indication of loose sand, which could be produced to the
surface (i.e., sand production). This agrees with actual sand
production, which occurred during the transient event (Figure
15). Several other intervals are also very close to failure,
where the actual BHP is only slightly higher than the CBHFP.
This plot shows that the well is prone to produce sand,
especially as the reservoir is depleted further.
Before the transient event, flow rate was lower: 9,000-10,000
bopd. Drawdown was lower by 140 psi, and actual BHP was
higher by 140 psi. Figure 14 predicts that at the weakest point
(15604.76 ft MD) there would still be sanding. Thus the
prediction of onset of sanding agrees with that observed, both
before and during the transient sanding event of April 2001.
Transient sand volume prediction
Figure 15 shows the sanding event in the field due to a change
in choke size designed to increase oil production from ~10600
B/D to 15,000 B/D, initiated on April 17, 2001. The sand rate
is measured by a strap-on sand detector.
As discussed by Vaziri, et al.,2 the BHP changes in steps, as
portrayed in Figure 16, and these have been computed from
WHP values using a model for oil/gas flowing up a wellbore.
The corresponding drawdown changes can be found by adding
in depletion changes (0.8 psi/day). It turns out the maximum
drawdown increase of 103 psi occurs on 3 June, at the time of
maximum in Figure 16.
To predict the volume of transient sand, the parametric
transient equation was applied to every foot of the perforated
interval, using the cohesion derived from the wireline logs.2
Although we did not have any cores to perform direct strength
measurements in this well, several

Sand rate [pptb], Choke


[1/64"]

GR.GAPI

Shutdowns

BHP (PSIG)

Oil Rate(STB/D)

Av-WHP(psi)

160

16,000

140

14,000

120

12,000

100

10,000

80

8,000

60

6,000

40

4,000

20

2,000

0
14-Apr-01

4-May-01

24-May-01

13-Jun-01

3-Jul-01

0
23-Jul-01

Date

Figure 15. Sand rate in pptb (second from top) and oil flowrate
(top curve) during transient sanding event

uniaxial strength and TWC (thick wall cylinder ) tests were


available in the field and were used to calibrate the strength
obtained from logs. The clean sand intervals (GR <40),
occupying 312 ft of sand formation (compared with 647 ft of
perforations), were assumed to produce the sand. Since sand
had been produced before this transient event, we assumed the
perforations had been washed out; i.e., the transient sand
Change in BHP versus time
160

change in BHP (psi)

BH Wellbore Press. (psi)

GR (GAPI)

CBHFP (psi)

Current Reser. Press.(psi)

[psi or STB/D]

CBHFP, Reservoir Pressure, BHWP and GR vs. Depth Well A/3


CBHFP (psi)

140
120
100
80
60
40
20
0
4/9/2001 4/19/2001 4/29/2001 5/9/2001 5/19/2001 5/29/2001 6/8/2001 6/18/2001

Figure 16. Change in BHP with time during transient sand event

volume is predicted for an open hole well, not a


cased/perforated well. Note that the sand volume is predicted
in ft3 in the model, and converted to lbs using 120 lb/ft3,
corresponding to a porosity of 30%.
The drawdown profile derived from Figure 16 is based on
nine individual calculations in Figure 15, with straight-line
interpolations between these nine values. The straight-line
interpolations were then broken up into many small steps, and
for each small drawdown step, a corresponding cumulative
sand volume was predicted. This gives a profile of cumulative
sand vs. time at the formation face. If we assume that the sand
is carried into the well and up to the surface very quickly
compared with the time between drawdown steps, this is also
the profile of cumulative sand vs. time at the surface. This is
probably a good approximation, since the transient sand event
lasts a relatively long time (~ 3 months). The inferred profile
of cumulative sand vs. time at the surface is presented in
Figure 17. This is the prediction.
In Figure 17 is also shown the observed cumulative sand
vs. time, obtained by integrating under the sand rate curve in
Figure 15 (see also Vaziri, et al.2). The prediction agrees rather
well with that observed, given the many uncertainties.

10

SPE 84499

yet reached its true steady-state, and that agreement will


improve over time. But this is not supported by a verbal
report made in October 2001, that sand production was in the
range of 7-20 pptb. We should not take the latter range too
seriously, without examining the sand detector records, to see
whether there may have been other new transient sanding
events that occurred.
Note that for a cased and perforated completion, our model
predicts 5.19 pptb (or 78 lbs/day), which matches better the
data of Figure 18. However, we do not think that this rate is
applicable for extended periods of sand production, because
the perforations will wash out.
In summary, in this one example, we have shown how the
BP suite of sanding models can predict the onset of sanding,
the volume of sand in a transient event, and the steady-state
level of sanding after the transient event. This new integrated
strategy is a powerful tool for helping to decide on the need
for, and timing of, sand control installations.
100000
steady-state observed
for 8 pptb

80000
cum sand (lbs)

The prediction at 100 days is 47.6 bbl of cumulative sand,


compared with 50.6 bbl observed.
The initial portion of the match is not very good, perhaps
because sand is not carried out of the well as quickly as
assumed in the model (we assumed an exit that was very
rapid). In addition, at late times, the sand prediction levels off
because drawdowns are no longer increasing (i.e., they are
lower than the maximum drawdown on 3 June). According to
the ENHANS model, this situation will not result in extra
sanding. However, it seems possible that delays when the sand
exits the well could also explain this discrepancy. Overall
though, it is clear that the prediction matches remarkably well
the actual transient sanding, especially given the many
uncertainties in the prediction.
The prediction in Figure 17 is cumulative sand vs. time,
and this could be differentiated to find a profile of sand rate
vs. time. The best fit to the predicted cumulative sand vs. time
is a polynomial, as shown in Figure 17. If we differentiate this,
we obtain a sand rate curve that falls below the first large peak
in Figure 15 (pink curve), but then agrees pretty well with the
observed sand rate out to 70 days, before falling off too
rapidly. This is an area of future development: to extend the
ENHANS model by adding a model for sand transport from
the sand face to wellhead (including sand pickup at the sand
face, and slippage of sand grains through the carrying fluid in
the wellbore).

60000
transient predicted

40000

steady-state predicted
(2.13 pptb)

20000
40000

transient observed

cum sand volume (lbs)

polynomial best fit

steady-state
observed
for 1 pptb

0
0

30000

100

200

300

400

500

Days
prediction

Figure 18. Observed profile of cumulative sand versus time,


including steady-state prediction

20000
observation

10000
y = -0.016x3 - 2.0381x2 + 593.16x + 6887.7
R2 = 0.9216
0
0

20

40

60

80

100

Days

Figure 17: Observed profile of cumulative sand versus time at


surface, compared with that predicted by transient sanding model
ENHANS

Steady-state sanding prediction


Finally, to assess the rate of steady-state sand production, the
BP model was applied to the production conditions after the
transient sand event of Figure 15. Since the well has been
producing sand for a significant time, it is expected that the
perforations will be washed-out and the completion will
behave as an open hole completion. The model has been
discussed in detail elsewhere.8
The steady-state sanding rate predicted by the model is 2.13
pptb (or 32 lbs/day). This is compared with observation in
Figure 18, where the observed sand rate after the transient
event lies between 1 and 8 pptb, as judged by the interval after
3 July 01 in Figure 15. Although the observed range does
bracket the prediction, it appears the prediction is on the low
side. However, it is possible that the observed range has not

How this helps in sand management:


In the sand management scenario, the biggest risk and
challenge is being able to reliably estimate the amount and
concentration of the produced sand. This is important for
sizing facilities sand handling capabilities, as well as ensuring
that erosion limits for chokes and surface pipework are not
exceeded. From a HSE perspective, this is especially critical
in high rate gas wells, as well as in high rate oil wells,
particularly where gas-oil ratios are high. From an operating
cost perspective, the consequences of severe sand production
and choke erosion could be very costly in subsea wells,
especially in deepwater. On the positive side, the cased and
perforated completion option usually employed with sand
management does permit avoidance of producing from notably
sanding prone intervals through selective or optimized
perforating. Cased and perforated completions also maintain
access to the producing interval to shut-off water or to
recomplete in other secondary producing horizons. This has
allowed significant increases in reserves recovery in a number
of fields worldwide.
The alternative to sand management is sand exclusion.
When properly implemented, downhole sand control will
exclude the bulk of the formation sand from being produced.
(It is noted, however, that some fines, smaller than the filter
media apertures, may still be produced to surface even for

SPE 84499

successfully installed sand control; this is particularly true of


transient fine sand production). The downside of this option is
typically a significant increase in up-front well completion
cost, and oftentimes, a lower well productivity than a
comparable cased and perforated completion. Occasionally,
sand control completions may also fail during the well life,
either mechanically, so permitting the influx of formation
sand, or suffer degrading inflow performance due to plugging.
The ability to easily intervene in sand control completions to
shut-off water is often difficult, as the preferred completion
option typically open-hole gravel packs, screen completions
and frac-packs may allow the water to by-pass the treated
interval.
Therefore, there is often a significant cost benefit both
for capital and operating expenditure if sand management
can be successfully implemented. However, to reliably do this
in a new project development it is necessary to be able to
produce a credible prediction of the rate at which the sand
might be produced.
These cost implications were the principal motivation for
developing the sand prediction tools described in this paper.
In totality, they comprise over 20 man-years of R&D effort
expended over the past 15 years.
In particular, the
development and application of these technologies has
undergone a significant advance following the mergers
between BP, Amoco and ARCO. This is because each
company was actively focusing on different aspects of the
sanding problem.
These inter-related and complementary efforts have now
evolved into a powerful suite of tools that enable realistic
predictions to be made of sanding potential, rates, and
volumes throughout the life of a well.
Conclusions
In summary, the new three-fold strategy of sand prediction at
BP has increased tremendously our ability to predict sanding
in production or injection wells. This quantum leap is
invaluable to help decide if we can manage the sand at surface
(or if downhole sand control is required); to decide if we can
defer sand control until a later date (possibly increasing
production as well as saving completion costs); to decide how
much sand will be produced if we increase the drawdown in a
sand-prone well; and even decide if we need sand control in
water injectors.

The standard BP prediction of sanding onset is generally


conservative. That is, sanding occurs at lower BHFP than
predicted. In four out of seven cases
CBHFP (obs) = (0.5 1) CBHFP (pred)
and this could be used as a quick rule-of-thumb.
In one HPHT field case the BP shear-failure model is very
conservative. The ARCO tensile-failure model is also
conservative, but not as much as the BP model. On the
other hand, the ENHANS coupled model actually
suggests these wells could have been drawn down even
more before sanding (disaggregated sand grains held
together by capillary cohesion due to connate water in
pendular rings).

11

In this same field case, sand finally comes in with water


influx, and this is modeled by the removal of capillary
cohesion forces.
We have illustrated a tool to calculate transient sand
volume during a well blowout, and to estimate whether
the well will kill itself by hydrostatic pressure increase.
To understand the sanding potential in water injector
wells, one approach is to calculate the onset of significant
sanding for various cross flow scenarios caused by
potential sealing faults or pinch-outs, and to estimate
transient sand volumes for the same scenarios. We have
illustrated the risk of transient sanding by showing P10,
P50, and P90 results for sand volumes produced, and for
rathole filling with sand. Clearly, a deeper rathole would
be an advantage.
For gas injectors, the sand volumes will be reduced by
typically 5 times, i.e., gas injectors are much safer than
water injectors because capillary cohesion can hold
disaggregated sand grains together.
The new steady-state model has been applied to predict
continuous sanding in an offshore field, and has led to the
conclusion that sand rates can be managed at surface,
without sand control. This is a huge economic advantage.
In one illustrative example, we have shown how the BP
suite of sanding models can predict the onset of sanding,
the volume of sand in a transient event, and the steadystate level of sanding after the transient event. The
prediction of sanding onset agrees with that observed,
both before and during a large transient sanding event in
April 2001. During the transient sanding event, the
predicted profile of cumulative sand vs. time at the
surface agrees rather well with that observed, given the
many uncertainties. After the transient event, the range of
sand rates observed does bracket the prediction, but it
appears the prediction is on the low side.

Development of the sand prediction tools described in this


paper comprised over 20 man-years of R&D effort expended
over the past 15 years. In particular, the development and
application of these technologies has undergone a significant
advance following the mergers between BP, Amoco and
ARCO. This is because each company was actively focusing
on different aspects of the sanding problem. These interrelated and complementary efforts have now evolved into a
powerful suite of tools that enable realistic predictions to be
made of sanding potential, rates, and volumes throughout the
life of a well.
Aknowledgments:
We thank Marie Morkved, Yogi Patel, and Mike Kutas for
assistance with field data. We thank Xu Li for contributing to
the calculations behind Figure 12. We are grateful to BP for
permission to publish this paper.

12

Nomenclature:
A
a poroelastic constant (a function of the Poissons
ratio and formation compressibility)
BHFP
bottomhole flowing pressure
BHP
bottomhole pressure
CDP
critical drawdown pressure at the start of
production
CBHFP critical bottomhole flowing pressure
CHC
cased hole completion
Pr
current average reservoir pressure
TWC
thick-walled cylinder strength
UCS
unconfined compressive strength
WHP
well head pressure
Greek
1 and 3 total principal major and minor stresses
y
the formation strength near the opening

References:
1. Ispas, I., Bray, R.A., Palmer, I.D., and Higgs, N.G.:
Prediction and Evaluation of Sanding and Casing
Deformation in a GOM Shelf Well, SPE/ISRM 78236
presented at the SPE/ISRM Rock Mechanics Conference,
Irving, Texas, October 20-23, 2002.
2. Vaziri, H., Xiao, Y., and Palmer, I.D.: Assessment of
several sand prediction models with particular reference to
HPHT wells, SPE/ISRM 78235 presented at the,
SPE/ISRM Rock Mechanics Conference, Irving, TX,
October 20-23, 2002.
3. Weingarten, J.S. and Perkins, T.K.: Prediction of Sand
Production in Gas Wells: Methods and Gulf of Mexico
Case Studies, SPE 24797 presented at the 67th Annual
SPE Technical Conference and Exhibition, Washington
DC, October 4-7, 1992.
4. Vaziri, H., Barree, R., Xiao, Y., Palmer, I., and Kutas, M.:
What is the Magic of Water in Producing Sand?, SPE
77683 presented at the SPE Annual Technical Conference
and Exhibition., San Antonio, TX, September 29 October
2, 2002.
5. Santarelli, F.J., Skomedal, E., Markestad, P., Berge H.I.,
Nasvig, H.: Sand Production on Water Injectors: Just
How Bad Can It Get?, SPE 47329 presented at the
SPE/ISRM Eurock 98, Trondheim, Norway, July 8-10,
1998.
6. Morita, N., Davis, E., and Whitebay, L. Guidelines for
Solving Sand Problems in Water Injection Wells, SPE
39436, pres. at SPE Intl. Symp. Formation Damage
Control, Lafayette, LA, Feb 18-19, 1998.
7. Tronvoll, J., Santarelli, F.J., Sanfilippo, F., and Dusseault,
M.B.: Sand Production Management for Seabed
Separation Systems, Report prep. for CoSWaSS
Consortium, Dec 1998.
8. Willson, S.M., Moschovidis, Z.A., Cameron, J.R., and
Palmer, I.D.: New Model for Predicting the Rate of Sand
Production, SPE/ISRM 78168, presented at the,
SPE/ISRM Rock Mechanics Conference, Irving, TX,
October 20-23, 2002.

SPE 84499

9. Palmer, I.D., McLennan, J.D., and Vaziri, H.H. CavityLike Completions in Weak Sands, SPE 58719, presented
at the International Symposium on Formation Damage
Control, Lafayette, Louisiana, February 23-24, 1999.

SPE 84499

13

Table 1: Field cases used to benchmark model for


onset of sanding

field

gas/oil

field 1
field 2
field 3
field 4
field 5
field 6
field 7

gas
gas
oil
oil
gas
gas
oil

TVD
22000
15000
9400
10050
13150
17000
8315

initial
Po
15800
15000
4100
4910
5750
14500
3400

UCSmin 3Sv-Sh? HPHT?


2000 yes
3053 yes
2000 no
250 no
1000 no
800 no
1900 yes

yes
yes
no
no
no
no
no

sand
observed?
no
yes
no
yes
yes
yes
yes

CBHFP ratio
pred/observed
3.72 lower limit
1.94
-1.45 lower limit
1.83
0.93
1.33
-0.456