See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/254304370

The Estimation of a Formation Fracture Pressure Gradient by

Using Drilling Data and Artificial Neural Networks

Article  in  Energy Sources Part A Recovery Utilization and Environmental Effects · June 2012
DOI: 10.1080/15567036.2011.574191

CITATIONS READS

0 1,721

2 authors:

A. Mollakhorshidi Rasoul Arabjamaloei

Islamic Azad University University of Manitoba
4 PUBLICATIONS   3 CITATIONS    39 PUBLICATIONS   88 CITATIONS   

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Modeling Flow Regimes in Porous Media: Correcting Permeability Models by Combining Numerical Simulation and Experimentation View
project

Numerical and Experimental Modeling of Spontaneous Imbibition View project

All content following this page was uploaded by Rasoul Arabjamaloei on 13 January 2015.

The user has requested enhancement of the downloaded file.

This article was downloaded by: [University of Manitoba Libraries]
On: 25 February 2014, At: 23:47
Publisher: Taylor & Francis
Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered
office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK

Energy Sources, Part A: Recovery,

Utilization, and Environmental Effects
Publication details, including instructions for authors and
subscription information:
http://www.tandfonline.com/loi/ueso20

The Estimation of a Formation Fracture

Pressure Gradient by Using Drilling Data
and Artificial Neural Networks
a a
A. Mollakhorshidi & R. Arabjamaloei
a
Department of Petroleum Engineering , Islamic Azad University,
Omidiyeh Branch , Omidiyeh , Khuzestan , Iran
Published online: 07 Jun 2012.

To cite this article: A. Mollakhorshidi & R. Arabjamaloei (2012) The Estimation of a Formation
Fracture Pressure Gradient by Using Drilling Data and Artificial Neural Networks, Energy
Sources, Part A: Recovery, Utilization, and Environmental Effects, 34:15, 1384-1390, DOI:
10.1080/15567036.2011.574191

To link to this article: http://dx.doi.org/10.1080/15567036.2011.574191

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the
“Content”) contained in the publications on our platform. However, Taylor & Francis,
our agents, and our licensors make no representations or warranties whatsoever as to
the accuracy, completeness, or suitability for any purpose of the Content. Any opinions
and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content
should not be relied upon and should be independently verified with primary sources
of information. Taylor and Francis shall not be liable for any losses, actions, claims,
proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or
howsoever caused arising directly or indirectly in connection with, in relation to or arising
out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden. Terms &
Conditions of access and use can be found at http://www.tandfonline.com/page/terms-
and-conditions
 Energy Sources, Part A, 34:1384–1390, 2012
Copyright © Taylor & Francis Group, LLC
ISSN: 1556-7036 print/1556-7230 online
DOI: 10.1080/15567036.2011.574191

The Estimation of a Formation Fracture Pressure

Gradient by Using Drilling Data and Artificial
Neural Networks

A. MOLLAKHORSHIDI1 and R. ARABJAMALOEI1

1
Downloaded by [University of Manitoba Libraries] at 23:47 25 February 2014

Department of Petroleum Engineering, Islamic Azad University,

Omidiyeh Branch, Omidiyeh, Khuzestan, Iran

Abstract Fracture gradient of formation is a key to determine the casing setting

depth in drilling oil/gas wells. In addition, for projects, such as hydraulic fracturing
and enhanced oil recovery injection, knowing the fracture gradient of the injection
zone is necessary. Also, the pressure integrity of the exposed open hole dictates the
maximum allowed wellbore pressure. Several theoretical and operational methods for
predicting fracture pressures have been developed and refined. A Leack-off test, which
is the most reliable and common method for evaluating fracture pressure gradient, is
performed by too much cost and time and also this test cannot be performed at several
points. In the present article, a novel technique is presented to obtain an estimation of
fracture pressure gradient from drilling operation data reports. This method is based
on the effect of pore pressure and confining pressure on compressive strength of rock
and, consequently, on drilling speed. Artificial neural networks were implemented to
build a simulator for the rate of penetration and analyze the effect of hydrostatic
pressure of wellbore on the rate of penetration. The presented method was performed
on field data of an Iranian southern field and the results were satisfactorily close to
the actual measured fracture pressure by an average error of about 1%.

Keywords drilling, fracture pressure, geomechanics, overbalanced pressure, neural

networks

1. Introduction
The subject of many discussions and technical papers in the last 20 years has been
the prediction of the wellbore pressure gradients that are required to induce or extend
fractures in subsurface formations. Encountered in several common types of operations in
the oil industry are problems associated with the prediction of formation fracture pressure
gradients.
When wells are being drilled in both new and old fields, lost circulation is often a
very troublesome and expensive problem. Complete loss of circulation has been disastrous
in some cases. Many times, such disasters could have been avoided if techniques for
calculating fracture pressure gradient had been employed in the well plans, and if casing
strings had been set, and mud weight plans had been followed accordingly. In areas
of abnormally pressured formations, the prediction of fracture gradients during the well

Address correspondence to Amir Mollakhorshidi, Department of Petroleum Engineering, Is-

lamic Azad University, Omidiyeh Branch, Omidiyeh, Khuzestan, Iran. E-mail: amir_mollakhorshidi
@yahoo.com

1384
 Formation Fracture Pressure Gradient 1385

planning stage is extremely important. In fact, it is as important as the prediction of

formation pressure, which has received a great deal of attention in recent years. There
are several published methods developed for determining fracture pressure gradients.
However, none of these methods appears to be general enough to be used with much
reliability in all areas (Eaton, 1969).
Harrison et al. (1954) presented techniques to determine the formation fracture
gradient based on stress analysis models that relate the fracture pressure to the pore
pressure, overburden pressure, and matrix stress ratio. Matthews and Kelly (1967) used the
matrix-stress-ratio concept to predict fracture pressure in normally pressured formations
along the U.S. Gulf coast. To improve Matthews and Kelly’s approach, Pennebaker (1968)
developed a similar correlation, by accounting for variable Gulf coast overburden. One
of the more widely used prediction techniques is Eaton’s (1969) correlation, based on
Downloaded by [University of Manitoba Libraries] at 23:47 25 February 2014

offshore Louisiana data in moderate water depths. Christman (1973) in his correlation
considered the effect of seawater on overburden stress in deep waters. Pilkington (1978)
averaged the matrix-stress ratios developed by the previous authors in south Texas,
Louisiana, and offshore California and presented the stress relationship curve to be used
for estimation of formation pressure gradient (FPG). Constant and Bourgoyne (1986)
proposed a correlation with constants that should be empirically adjusted. Their model
also covers Eaton’s correlation. Many researchers have used field test data to develop em-
pirical correlations (Brennan and Annis, 1984; Vuckovic, 1989; Breckels and van Eekelen,
1982). McPherson and Berry (1972) developed a correlation between elastic modulus for
a compressional wave and formation fracture pressure. Using measurements of interval
transit time by means of a sonic log and bulk density by means of a density log, the elastic
modulus and FPG is computed. Anderson et al. (1973) presented a method particularly for
the Gulf coast area that uses open hole log-porosity measurements and calibrated leak off
test data to estimate the fracture gradient in sandstone. Daines (1982) offered a procedure
for estimating fracture gradients in variable lithology and in those regions where little
information is available. His method was based on dividing the minimum in-situ hori-
zontal stress into two components: the elastic response to overburden and a superposed
tectonic stress. Zamora (1989) presented a prediction technique based on an empirical
family of overburden gradient and stress-ratio curves, which are dependent on geological
age and geographic setting. Holbrook et al. (1993) used porosity to develop an empirical
model for FPG and they found that results from the above equation were in agreement
with lower-bound fracture pressures evidenced in the North Sea and other areas.
This article presents a new approach for fracture gradient calculation based on drilling
data by using artificial neural network (ANN). The developed ANN model is capable of
predicting the fracture gradient as a function of drilling operation parameters. Considering
the fact that the fracture pressure at a point is calculated based on the data for that point,
this model is the most exact method based on input characteristics. The proposed fracture
gradient model is simple to apply and the results obtained from the model are reasonably
accurate.

2. Artificial Neural Networks

An ANN is a system based on the operation of biological neural networks. ANNs are
composed of elements called neurons. These elements are inspired by biological nervous
systems. As in the nature, artificial neuron is a process unit, and the connections between
neurons specify the causal relationship between the connected neurons. To characterize
a specific ANN, it is necessary to determine the number of neurons, how they are
 1386 A. Mollakhorshidi and R. Arabjamaloei

interconnected, and the processing that takes place throughout the network. Functionally,
there are three types of layers of neurons: input layers, hidden layers, and output layers.
The learning rule determines how the parameters (of neurons) should be updated to
minimize mapping error.
The basic learning rule of adaptive network is the steepest descent method, which
uses gradient vector to reach minimum error.

3. Geology of South Pars Field

The South Pars Field is located in the interior platform of the Arabian Shelf structural
province. The gas-bearing horizons of South Pars Field belong to the Kangan and
Dalan formations of the Triassic and Permian age, respectively. The Dalan formation
Downloaded by [University of Manitoba Libraries] at 23:47 25 February 2014

is separated from the Kangan formation by an unconformity caused by a local emersion

cycle. The Kangan formation is directly overlain by the Aghar Shale Member of the
Dashtak formation, which acts as Cap Rock for the productive horizons.
In the Persian Gulf region, the Kangan and Dalan formation are also known as Khuff
formation, in Permo/Triassic boundaries. The Kangan-Dalan formation consists of cycles
composed of dolomites interbedded with limestones and anhydrite that are thought to
have been deposited in a shallow open marine to supratidal setting.

4. Theory
The basis of this study is the effect of pore pressure on rock strength under triaxial
stresses. The effect of this pressure has been studied in many studies and it’s effect is
completely known. Figure 1 shows the result of the process of increasing pore pressure
in a rock sample under triaxial stress.
With an initial application of pore pressure, fissures and some pores begin to close,
producing an inelastic, concave-upward stress-strain section. This is followed by linear
relationships between axial stress and axial strain. Starting from point B, new cracks begin
to form inside the most critically stressed portions of the rock. After point C, cracks that
form propagate to the edges of the rock and a system of intersecting, coalescing cracks is

Figure 1. The effect of pore pressure on a rock under triaxial stresses. (color figure available
online)
 Formation Fracture Pressure Gradient 1387

developed and eventually leads to rupture of the rock at point D. From here, the strength
of the rock decreases fast. This point is evaluated by a leak-off test on the field and is
called fracture initiation pressure.

5. Methodology
In this study, the first step is to develop a model for estimation of rate of penetration
(ROP) by using ANN. In the process of modeling ROP by ANN, the effects of all crucial
parameters, including inclination and azimuth, of well trajectory are taken into account.
In fact, the ratio of weight on bit over bit size (WOB/d), rotation per minute of drill
string (RPM), type of bit, total nozzle, or flow area (TFA), mud circulation rate and
stand pipe pressure, mud weight, dull bit condition, true vertical depth of drilling (TVD),
Downloaded by [University of Manitoba Libraries] at 23:47 25 February 2014

azimuth and inclination of the well are chosen from drilling operation parameters while
unconfined compressive strength and three principal stresses in the geological region of
interest are used to include the formation properties. For simplicity and compatibility of
inputs data to the ANN model, type of bit is introduced by the International Association
of Drilling Contractors’ Code (IADC) and the ratio of bit working-hours over the first
digit of IADC of drilling bit is set as an equivalent of dull bit condition to be used as
one of the ANN input parameters. The selected inputs cover almost all of the effective
parameters on the function of rate of penetration. Table 1 represents a sample of data
for developing a ROP model. For training the ANN model, geomechanical data, logging
results, drilling reports, and mud logging results of nine directional wells drilled in the
south pars oil field located at the southwest of Iran were gathered. Among 3,000 sets
of obtained data, which contain 15 input parameters and one output parameter (ROP),
70% was used for training, 15% was used for validating, and the remaining 15% was
applied for testing the model. To obtain a higher performance of ANN modeling, before
importing inputs and outputs data into the ANN model, they were scaled to fall into the
range of [ 1 1]. In developing the model, among several tested structures for ANN a
three-layer network with a structure of {50 1} of neurons showed the best performance
for ROP prediction. By readjusting the weights, the final developed ROP model, which
results in the lowest mean square error, is achieved. Figure 2 shows the testing cross plot
of the developed network for estimation of ROP.

Table 1
Sample of data set for developing a ROP model using ANN

IADC TVD Azimuth Angle Pressure RPM WOB ROP

115 210 164 1.45 973.3 80 2.72 15

115 211 171 1.44 1,219.6 78 7.21 89.49
115 212 178 1.43 1,209.5 78 6.57 47.47

Working
Sigma3 Sigma2 Sigma1 TFA hours Bit size Mw

3.368 4.33 6.586 1.110039 31.17 23.5 1.05

3.41 4.36 6.610 1.110039 31.50 23.5 1.05
3.451 4.386 6.635 1.110039 31.82 23.5 1.05
 1388 A. Mollakhorshidi and R. Arabjamaloei
Downloaded by [University of Manitoba Libraries] at 23:47 25 February 2014

Figure 2. The testing cross plot of the developed network for estimation of ROP. (color figure
available online)

One of the principal parameters that affect ROP is rock strength. It is known that
during the drilling, the formation rock is broken under the bit by triaxial loading. The
well pressure acts as pore pressure. As explained in section 3, point D of the graph is
the leak off point and the point in which the increase in pore pressure results decrease
in rock strength and, therefore, increase of rate of penetration. In this study, it is aimed
to obtain this point by observing the effect of well pressure on the strength of the rock
and observing the effect of the strength of the rock on rate of penetration.
For this purpose, many points were chosen in a drilled well and for these points the
effect of hydrostatic pressure on ROP by keeping other parameters at a constant value
were analyzed. The hydrostatic pressure was increased by increasing stand pipe pressure
value and mud weight value up to the maximum of these values in the data sets that
were used for training. Figure 3 represents the effect of pressure on ROP at the depth of
1,900 m. As Figure 3 shows, ROP is decreasing as ECD increases up to the ECD of 1.75
and then starts to increase from this point. The same figure was generated for the four

Figure 3. The effect of increase of ECD on ROP. (color figure available online)
 Formation Fracture Pressure Gradient 1389
Downloaded by [University of Manitoba Libraries] at 23:47 25 February 2014

Figure 4. Comparison of measured and estimated formation fracture equivalent ECD. (color figure
available online)

other depth points of 1,500, 2,100, 2,400, and 2,600 m. In all of the curves, ROP starts
to increase from a point and this shows that the rock strength is decreasing from this
point and it means that this point is equivalent to the point D of the graph in Figure 2.
In the available data, the maximum stand pipe pressure and maximum mud weight that
were used in the training data set is 3,200 psi and 11.6. Therefore, these values cannot
make up an ECD to cause a critical point in ROP-ECD curve for the depths of more
than about 2,700.
The real leak-off tests have been performed for many points of the south pars oil
field and the leak-off pressures are known. For the five points that critical ECD were
obtained, the actual leak-off tests were used to evaluate leak-off pressure and equivalent
ECD. The results were compared and it was observed that the critical ECD is about 1%
less than leak off ECD for all measured points (Figure 4).

6. Discussion of the Results

As the results show, and based on what was expected, drilling data can be used to develop
a model for estimation of rate of penetration and analyze the effect of various parameters.
Based on the studies on the effect of pore pressure on rock strength and correspondingly
on rate of penetration, it was completely expected that by analyzing the effect of high
hydrostatic pressure values on ROP, a similar result as the effect of pore pressure on
rock strength would be obtained. Since ANN models are not expected to work well for
extrapolation, therefore, in the procedure of studying the effect of pressure on ROP the
 1390 A. Mollakhorshidi and R. Arabjamaloei

applied values for pressure should be in the range of available data. This means that it is
not possible to calculate fracture pressure for the entire depth interval. The results show
that the obtained value for fracture pressure is about 1% lower than field tests, since the
theory and the test results both propose the applicability of the presented method, which
shows reliability of the method.

7. Conclusions and Recommendations

In this article, a new method was introduced for estimation of formation fracture pressure
gradient by using artificial neural networks. The presented method can be used as a real
time method while drilling. Based on the theories behind the developed method, it was
expected that the method would be precise. The results of testing the method showed
Downloaded by [University of Manitoba Libraries] at 23:47 25 February 2014

that its precision is excellent. Based on the theory and test results, it is claimed that
the developed method in this article is a reliable and accurate method for estimation of
formation fracture pressure gradient.

Acknowledgments
The authors would like to acknowledge Islamic Azad University-Omidieh Branch for
scientific and financial support of this research.

References
Anderson, R. A., Ingram, D. S., and Zanier, A. M. 1973. Determining fracture pressure gradients
from well logs. J. Pet. Technol. 25:1259–1268.
Breckels, I. M., and van Eekelen, H. A. M. 1982. Relationship between horizontal stress and depth
in sedimentary basins. J. Pet. Technol. 34:2191–2199.
Brennan, R. M., and Annis, M. R. 1984. A new fracture gradient prediction technique that shows
good results in Gulf of Mexico abnormal pressure. SPE Paper 13210. SPE Annual Technical
Conference and Exhibition, Houston, TX, September 16–19.
Christman, S. A. 1973. Offshore fracture gradients. J. Pet. Technol. August:910–914.
Constant, W. D., and Bourgoyne, Jr., A. T. 1986. Method predicts fracture gradient for abnormally
pressured formations. Pet. Eng. Intl. January:38–46.
Daines, S. R. 1982. Prediction of fracture pressures for wildcat wells. J. Pet. Technol. 34:863–872.
Eaton, B. A. 1969. Fracture gradient prediction and its application in oilfield operations. J. Pet.
Technol. 1353–1360.
Harrison, E., Kieschnick, W. F., and McGuire, W. J. 1954. The mechanics of fracture induction
and extension. Pet. Trans. AIME 201:252–263.
Holbrook, P. W., Maggiori, D. A., and Hensley, R. 1993. Real-time pore pressure and fracture
gradient evaluation in all sedimentary lithologies. SPE Paper 26791. SPE Offshore European
Conference, Aberdeen, MD, September 7–10.
Matthews, W. R., and Kelly, J. 1967. How to predict formation pressure and fracture gradient from
electric and sonic logs. Oil & Gas J. February:96–106.
McPherson, L. A., and Berry, L. N. 1972. Prediction of fracture gradients. Log Analyst 13:12–19.
Pennebaker, E. S. 1968. An engineering interpretation of seismic data. SPE Paper 2165. SPE
Annual Fall Meeting, Houston, TX, September 29–October 2.
Pilkington, P. E. 1978. Fracture gradient estimates in tertiary basins. Pet. Eng. Intl. May:138–148.
Vuckovic, V. 1989. Prediction of fracture gradients offshore Australia. SPE Paper 19468. Asia-
Pacific Conference, Sydney, Australia, September 13–15.
Zamora, M. 1989. New method predicts gradient fracture. Pet. Eng. Intl. September:38–47.

View publication stats