Estimating Natural-Fracture Permeability From Mud-Loss Data

PROCEEDINGS, Thirty-Eighth Workshop on Geothermal Reservoir Engineering
Stanford University, Stanford, California, February 11-13, 2013

SGP-TR-198
ESTIMATING NATURAL-FRACTURE PERMEABILITY FROM MUD-LOSS DATA

Serhat Akin
Middle East Technical University
Petroleum & Natural Gas Eng Dept, Dumlupinar Blvd No:1
06800 Cankaya, Ankara, Turkey
e-mail: serhat@metu.edu.tr
ABSTRACT
Knowing locations, distributions and apertures of
fractures crossing a geothermal well is of vital
importance in order to minimize costs and increase
efficiency of drilling. Complete or small losses of
drilling fluid flowing from wellbore to the
surrounding formations have been used to identify
fracture zones in the past. An analytical model based
on transient radial mud-loss invasion from a borehole
into a fracture plane coupled with an artificial neural
network approach is developed to estimate naturalfracture permeability. The developed model is
compared to pressure transient test results obtained
for several wells located in a liquid dominated
geothermal reservoir in west Turkey. It has been
observed that the model fracture permeability values
are in accord with well test derived permeability
values.
INTRODUCTION
The idea of associating mud losses with fracture
permeability in a naturally fractured reservoir was
proposed in sixties (Drummond, 1964). Apart from a
few studies most studies are only qualitative (Dyke et
al, 1995). Most of the mud loss modeling studies
focuses on compressible Newtonian mud propagating
in a non-deformable fracture of constant aperture
with impermeable walls. Coupling the diffusivity
equation with a constant pressure difference
boundary condition and interpolating a tabulated
solution of the problem, Sanfillippo et al (1997)
obtained approximate analytical solutions between
time and mud (Newtonian) volume lost, which was
then converted to fracture hydraulic aperture. They
claimed that eventual mud loss stops were due to
fracture plugging by mud particles although this was
not accounted for in the model. Lietard et al (1999,
2002) developed type curves that describe mud loss
volume vs time. Type curve matching was then used
to obtain natural fractures hydraulic width. Majidi et
al (2008a, 2008b) developed mathematical solutions
using yield-power law fluids as opposed to Bingham

plastic fluids used by Lietard et al (1999, 2002).
Huang et al (2011) recently proposed a cubic
equation with input parameters given by the
overpressure ratio, maximum mud loss volume and
the well radius to obtain effective hydraulic fracture
aperture.
DRILLING
FRACTURES
MUD
INVASION
INTO
Lietard et al. (1999, 2002) developed a model based

on the radial flow of a Bingham plastic drilling fluid
through an infinite fracture. The drilling fluid flow
was described by the local pressure drop caused by
the laminar flow inside the fracture with a width of w.
Linear momentum equation relates the local pressure
gradient and average velocity using the following
equation,
(1)
Where and p are the yield stress and plastic
viscosity of the drilling fluid. The local velocity of
the mud in the fractures under radial conditions
around the well is given by vm.
( )
( )
(2)
Introducing dimensionless mud invasion radius (rd),
time (td) and dimensionless mud-invasion factor (d)
Lietard (2002) obtained the following differential
equation with initial conditions rd=1 when td=0.
(
(3)
Where
(4)
(5)
.
(6)
(7)
Solution of equation 3 is given by Civan and

Rasmussen (2002).
/1
( ( )0
[.
/]
/)
(8)
Where rdmax is the maximum dimensionless mud
invasion radius given by the following equation.
(9)
Lietard et al. (1999, 2002) expressed the cumulative
volume of mud loss Vm by substituting equations 4
and 5 leading to the definition of a parameter X and
equations 5 and 6 leading to the definition of a
parameter Y as shown below.
*, ( )+ (10)
(11)
(12)
A series of type curves relating X and Y can be
constructed by changing w and rdmax which then can
be used to matched field data to obtain w.
More recently Huang et al (2011) proposed a simpler
method based on Lietard et al.s solution. The
solution is based on the fact that the mud losses will
eventually stop because of the overpressure
eventually reaching the yield stress of the drilling
fluid. The ultimate invasion radius that depends on
the wellbore radius, the yield value of the drilling
fluid and the amount of overpressure can be written
as
(13)
The maximum mud-loss volume is then given by
( )
[
]
(14)
Substituting equation 14 into equation 13 gives a
cubic equation (15) in the fracture width (w) with
coefficients dependent on the well radius (rw), the
overpressure ratio (p/y), and the maximum mudloss volume (Vmmax). Solution of this equation for the
fracture aperture by discarding physically
meaningless roots is a simple and direct way of
obtaining the fracture aperture when compared to the
curve fitting method proposed by Lietard et al. (1999,
2002).
( )
( )
(15)
RESULTS & DISCUSSIONS

A knowledgebase consisting of 162 different data
sets has been developed using the parameters
provided in Table 1. These parameters are then used
to obtain fracture apertures by solving the
aforementioned cubic equation. The model results
are used to train a feed forward artificial neural
network (ANN) model using error back propagation.
Then using this knowledgebase several different

ANNs were trained. The ANN is supplied with
mud-loss volume, number of events and overpressure
ratio values as input data and the average fracture
aperture is obtained. During the training process for
determining the weights, some data should be
withheld for later verification of network accuracy.
These data are often referred to as test or validation
data. Once the weights have been determined
through back propagation, the test data were used as
network inputs for determining the networks
accuracy in predicting unprocessed data sets. The
quality or goodness of training was judged based on
the closeness of the prediction of the remaining
testing data (i.e. 5% of the total fracture aperture
results using the input data that was not used for
training). This process was repeated for various
networks and the network with the highest accuracy
was used as the model. Rather than randomly
selecting the initial weight matrix, previously
generated successful matrices were used at the start.
This feature decreased the iterations approximately
30% and also guaranteed training of a good
network (Yilmaz et al, 2002).
Table 1:
Mud-loss volume, number of events and

overpressure ratio values used in
sensitivity analysis. .
Mud-loss Volume, bbl
10 - 80000
Number of Events
1 - 400
Overpressure ratio
364420 - 840000
Since the use of more than a single layer can lead to a
very large number of local minima and make the
training extremely difficult (Haykin; 1994), a single
hidden layer network was used. Several networks
with varying degree of complexity have been trained.
The best result was obtained with an ANN model
composed of 20 hidden nodes, with the momentum
and learning parameters of 0.3 and 0.5 respectively.
Figure 1 shows the average mean square error per
input for training and validation sets.
0.060
Mean square error
0.050
Training set
0.040
Validation set
0.030
0.020
0.010
0.000
0
100
200
300
Epoch
400
500
600
Figure 1: Average mean square error per input for

training and validation sets.
100
Mud Loss Rate, bbl/hr

200
300
400
500
600
950
1050
1150
Depth, m
Three wells drilled in Alaehir Graben in west

Anatolia, Turkey will be used to demonstrate the uses
of the proposed ANN methodology. First observed
mud losses will be used in Huang et al (2011) cubic
equation. Then the same data will be analyzed with
the ANN model. Two of these wells (Well-1 and
Well-2) showed limited mud loss rates at the
reservoir zone before a total loss was observed at the
fault zone that consisted of marble, calcschist and
calcschist-schist (Fig. 1 and Fig. 2) sequence. Well-3
mud loss rates (Fig. 3) were at the same order of
magnitude with the rates observed in the other wells.
However, total loss of mud was not observed in this
well. The mud loss volumes, number of events and
overpressure ratio observed in these wells are given
in Table 2.
1250
1350
1450
1550
1650
Figure 3: Mud loss rate observed in Well-2.

0
10

20
30
40
50
950
1150
Mud-loss volume, number of events and

overpressure ratio values observed during
the drilling.
Well
Drill bit diameter,
inch
Mud-loss Volume,
bbl
Number of Events
Overpressure ratio
Calculated
aperture, m
Calculated
permeability, mD
Well Test Derived
permeability, mD
ANN aperture, m
1
81/2
2
81/2
3
81/2
1666
1147
1264.9
36
641777.8
3.71 x10-4
22
684000
3.7 x10-4
138
684000
2.07 x10-4
745.12
1100.23
83.37
517.21
968.35
255
3.87x10-4
4.45x10-
1.68x10-4
ANN permeability,
mD
846.26
100

200
300
44.5
400
1550
1750
1286.62
1350
Depth, m
Table 2:
500
1950
2150

First average mud loss volumes were calculated by
dividing the total mud loss volume to the number of
events monitored during drilling the reservoir section.
This value is used together with the overpressure
ratio in equation 15 and the fracture aperture values
presented in Table 2 are obtained by solving the
cubic equation using MS Excels nonlinear solver.
Fracture aperture values are then converted to
permeabilities using the cubic law (Golf Racht,
1982). In doing so several assumptions are made
(Norbeck et al, 2012):
950
1050
Depth, m
1150
1.
2.
1250
1350
3.
4.
1450
1550
1650
5.
All natural fractures are finite with a constant

aperture,
The fractures are transverse to the wellbore and
have circular geometry,
Cubic law relationship is valid,
Matrix permeability is significantly lower than
fracture permeability,
Fractures are not charged during the process
Using the Huang et al cubic equation average mud

loss was calculated by dividing the mud loss volume
to the number of events. Substituting these values
together with overpressure ratio the average fracture
aperture was calculated as 3.71x10-4 m. Since
conventional petrophysical logs as well as fracture
detection logs and core plugs are not available for
any of the wells used for this study to quantify and
prove the existence of fractures, the results obtained

from cubic equation approach has been compared to
the results of pressure buildup analyses. Using cubic
law (Golf-Racht, 1982) the corresponding
permeability value was calculated as 745.12 mD,
which compared favorably with the well test derived
permeability of 517.21 mD. Typically, pressure
buildup test volume is much larger than the volume
sampled based on the mud losses observed during
drilling. Thats why we expect a similar order of
magnitude in permeability but the average values
may differ. Similar observations were obtained for
the other wells.
That is to say the derived
permeability was in the same order of magnitude but
the value was somewhat different.
When ANN results were compared to model and well
test derived permeability values, it was observed that
the permeability values were somewhat higher than
both the model and well test derived permeabilities
for complete mud loss wells. On the other hand the
permeability was somewhat smaller in well #3. One
reason for the mismatch could be the small number
training data. When only a relative small data set is
available for the application of a neural network a
number of drawbacks occur. Firstly, one has to split
the already small data set into a training set and a
testing set. Secondly, overtraining is more likely to
occur with small datasets, because the degrees of
freedom in the network rapidly increase with the
number of neurons. For a good model, the number of
data pairs should exceed the number of weights in the
neural network. One way to tackle this problem
could be the use of orthogonal functions such as
Fourier transforms to transform the input data before
they are used to train the network (Silvert and
Baptist, 1998).
CONCLUSIONS
An analytical model based on transient radial mudloss invasion from a borehole into a fracture plane
coupled with an artificial neural network approach is
developed to estimate average natural-fracture
aperture and the corresponding average fracture
permeability.
The developed ANN model is
compared to pressure transient test results obtained
for several wells located in a liquid dominated
geothermal reservoir in Alaehir graben, west
Turkey. It has been observed that the model fracture
permeability values are in accord with well test
derived permeability values that shows that it is
possible to use the model during drilling to predict
permeability of natural fractures.
REFERENCES
Civan, F., Rasmussen, M.L., (2002), Further
Discussion of Fracture Width Logging While
Drilling and Drilling Mud/Loss-CirculationMaterial Selection Guidelines in Naturally
Fractured Reservoirs, SPE Drill & Compl 17
(4), 249-250, SPE 81590-DS.
Drummond, J.M. (1964), An Appraisal of Fracture
Porosity, Bull. Can. Pet. Geol., 12, 226.
Dyke, C.G., Wu, B., Milton Taylor, D. (1995),
Advances in Characterising Natural Fracture
Permeability From Mud Log Data, SPEFE,
Sept, 160.
Golf-Racht, Van, T. D., (1982), Fundamentals of
Fractured Reservoir Engineering. Elsevier
Publishing Company, 710 pp.
Haykin, S.S., (1994) Neural Networks - A
Comprehensive Foundation. Prentice-Hall
International, London, 842 pp.
Huang, J., Griffiths D.V., Wong, S.W., (2011),
Characterizing Natural-Fracture Permeability
From Mud-Loss Data SPE Journal, March,
111-114.
Lietard, O., Unwin, T., Guillot, D., Hodder, M.H.,
(1999) Fracture Width Logging While Drilling
and Drilling Mud/Loss-Circulation-Material
Selection Guidelines in Naturally Fractured
Reservoirs SPE Drill & Compl 14 (3), 168-177.
Lietard, O., Unwin, T., Guillot, D., Hodder, M.H.,
(2002) Fracture Width Logging While Drilling
and Drilling Mud/Loss-Circulation-Material
Selection Guidelines in Naturally Fractured
Reservoirs SPE Drill & Compl 17 (4), 237-246.
Majidi, R., Miska, S.Z., Yu, M., Thompson, L.G.,
(2008a) Quantitative Analysis of Mud Losses in
Naturally Fractured Reservoirs: The effect of
Rheology, SPE 114130 presented at the SPE
Western Regional and Pacific Section AAPG
Joint Meeting, Bakersfield, 31 March 2 April.
Majidi, R., Miska, S.Z., Yu, M., Thompson, L.G.,
Zhang, J. (2008b) Drilling Fluid Losses in
Naturally Fractured Formations, SPE 114630
presented at the SPE ATCE, Denver, 21-24 Sept.
Norbeck, J., Fonseca, E., Griffiths, D.V., and Wong,
S-W. 2012. Natural Fracture Identification and
Characterization While Drilling Underbalanced.
Paper SPE 154864-MS presented at the
Americas
Unconventional
Resources
Conference, 5-7 June, Pittsburgh, Pennsylvania,
USA. doi: 10.2118/154864-MS.
Sanfillippo, F., Brignoli, M., Santarelli, F.J., Bezzola,

C. (1997), Characterization of Conductive
Fractures While Drilling, SPE 38177 presented
at the 1997 SPE European Formation Damage
Conf, The Hague, Netherlands, 2-3 June.
Silvert, W., Baptist, M., (1998) Can Neural
Networks be used in Data-Poor Situations? Int.
Workshop on Applications of Artificial Neural
Networks to Ecological Modelling, Toulouse,
France, 14-17 Dec. 1998.
Yilmaz, S., Demircioglu, C., Akin, S., (2002)
Application of Artificial Neural Networks to
Optimum Bit Selection Computers &
Geosciences, 28 (2), 261-269.

Estimating Natural-Fracture Permeability From Mud-Loss Data

Uploaded by

Copyright:

Available Formats

Estimating Natural-Fracture Permeability From Mud-Loss Data

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Estimating Natural-Fracture Permeability From Mud-Loss Data

Uploaded by

Copyright:

Available Formats

PROCEEDINGS, Thirty-Eighth Workshop on Geothermal Reservoir Engineering

Stanford University, Stanford, California, February 11-13, 2013

ESTIMATING NATURAL-FRACTURE PERMEABILITY FROM MUD-LOSS DATA

using yield-power law fluids as opposed to Bingham

Lietard et al. (1999, 2002) developed a model based

Solution of equation 3 is given by Civan and

RESULTS & DISCUSSIONS

Then using this knowledgebase several different

Mud-loss volume, number of events and

Mean square error

Figure 1: Average mean square error per input for

Mud Loss Rate, bbl/hr

Three wells drilled in Alaehir Graben in west

Figure 3: Mud loss rate observed in Well-2.

Mud Loss Rate, bbl/hr

Mud-loss volume, number of events and

Mud Loss Rate, bbl/hr

Figure 4: Mud loss rate observed in Well-3.

Figure 2: Mud loss rate observed in Well-1.

All natural fractures are finite with a constant

Using the Huang et al cubic equation average mud

prove the existence of fractures, the results obtained

Sanfillippo, F., Brignoli, M., Santarelli, F.J., Bezzola,

You might also like