INVESTIGATION OF CHANGES IN PERMEABILITY

DUE TO PORO-THERMO-ELASTIC EFFECTS IN

ENHANCED GEOTHERMAL RESERVOIRS

By
Nima Gholizadeh Doonechaly
Supervisor: Professor Sheik S. Rahman

A dissertation submitted to the University of New South Wales in partial fulfilment of

the requirements for the Degree of Doctor of Philosophy

In
Petroleum Engineering

School of Petroleum Engineering

The University of New South Wales
Sydney, NSW, AUSTRALIA

September 2013
iii
ii
iii
ACKNOWLEDGMENTS

I would like to express my sincerest gratitude to my advisor, Professor Sheik S. Rahman

for his consistent support and invaluable guidance throughout my PhD studies. It was a
great honour for me to study under his supervision. I also really appreciate his patience and
eagerness to supervise me. It would have been impossible to complete this work without his
helpful supervision.

I am grateful to Dr. Luiqi Wang for his valuable inputs at different stages of my thesis. I
also wish to acknowledge Nino Zajaczkowski for his immediate assistance in computing
resources. His helpful attitude is deeply appreciated. I want to thank all other staffs and
students in the petroleum engineering department, who have accompanied me during this
long journey.

I wish to acknowledge Magellan Petroleum Limited and Santos Petroleum Australia

Limited for providing valuable data, without which this research could not have been
completed.

I am also very thankful to University of New South Wales - Faculty of Engineering - for
providing me the full scholarship including the tuition fees and living allowance during my
study.

I am extremely grateful to my parents, Azizollah and Robabeh for their endless love and
support.

And my deepest thanks go to my wife, Sayedeh, for her continuous love, endless support
and patience over the years of my study.

iv
ABSTRACT
An innovative technique for estimating changes in stresses due to poro-thermo-elastic
effects during stimulation and circulation of cold water over a longer term in geothermal
reservoirs is presented. A hybrid of tectonic simulation and stochastic analysis of field data
is utilized to generate the subsurface fracture map of three formations. For this purpose, a
geological structure is reconstructed according to its past tectonic history. Then fractures
are stochastically simulated using Gaussian simulation. After generating the subsurface
fracture map, the fluid flow is simulated using finite element method in a poro-thermo-
elastic framework. Time-dependent heat transfer is modelled based on conductive heat
transfer within the reservoir rock and convective (including conduction) heat transfer in
discrete fractures. Changes of stress due to injection and circulation of cold fluid is studied
by using roughness induced shear displacement principle in a poro-thermo-elastic
environment. An analytical model for dilation of fracture surfaces based on the distributed
dislocation technique is used to estimate changes in fracture aperture. The roughness of fracture
surfaces is used in the calculation of residual fracture aperture.

The proposed methodology is applied to a section of Soultz geothermal reservoir at a

depth of 3650m and a number of numerical experiments were conducted to evaluate
potential for permeability enhancement. Results of this study agree well with the observed
field data. The results show that the average residual (retained) aperture is much lower and
the time required to reach maximum shear dilation events were greater than those predicted
by earlier studies including authors’ own study. In this study it was also observed that the
effective tensile normal stresses due to circulation of cold fluid tend to increase fracture
apertures within the zone of cooling. This increase in aperture and hence the permeability is
evident predominantly near wellbore region in the early circulation period. Over longer
period, a significant part of the reservoir, through which circulation is well established, is
subjected to larger thermal stresses. These thermal stresses consequently increased
permeability of the major interconnected fractures which led to significant changes in
pressure distribution (decrease in impedance) and hence, increase in the flow rate.

v
PUBLICATIONS PRODUCED FROM THIS WORK

1- Nima Gholizadeh Doonechaly, Sheik S. Rahman and Andrei Kotousov. A New

Approach to Hydraulic Stimulation of Geothermal Reservoirs by Roughness
Induced Fracture Opening. Accepted to be presented in The International
Conference for Effective and Sustainable Hydraulic Fracturing, 20-22 May 2013,
Brisbane, Australia.
2- Gholizadeh Doonechaly, N. and S. S. Rahman (2012). "3D hybrid tectono-
stochastic modelling of naturally fractured reservoir: Application of finite element
method and stochastic simulation technique." Tectonophysics 541–543(0): 43-56.
3- Nima Gholizadeh Doonechaly, Sheik S. Rahman and Andrei Kotousov. A Realistic
Assessment of Recoverable Thermal Energy from Australian Geothermal
Reservoirs: A Simulation Study, Proceedings of the 2012 Australian Geothermal
Energy Conference, Nov 2012.
4- Nima Gholizadeh Doonechaly, Sheik S Rahman and Andrei Kotousov An
Innovative Stimulation Technology for Permeability Enhancement in Enhanced
Geothermal System- fully coupled thermo-poroelastic numerical approach, GRC's
36th Annual Meeting, Sep 2012, Reno USA.
5- Nima Gholizadeh Doonechaly; Sheik Rahman; “Discrete Fracture Network
Simulation: Application of Tectonic Simulation and Conditional Global
Optimization Technique”, AAPG International Conference and Exhibition, Milan,
Italy, Oct 2011.
6- Joshua Koh, Nima Gholizadeh-Doonechaly and Sheik S Rahman, Luke Mortimer;
“Reservoir characterisation and numerical modelling to reduce project risk and
maximise the chance of success. An example of how to design a stimulation
program and assess fluid production from the Soultz geothermal field, France”,
Australian Geothermal Energy Conference, Nov 2010, Adelaide.

vi
Table of Content

COPYRIGHT STATEMENT ........................................................................................... ii

AUTHENTICITY STATEMENT .................................................................................... ii

ORIGINALITY STATEMENT ....................................................................................... iii

ACKNOWLEDGMENTS................................................................................................ iv

ABSTRACT .......................................................................................................................v

PUBLICATIONS PRODUCED FROM THIS WORK ................................................... vi

Table of Content .............................................................................................................. vii

List of Tables......................................................................................................................x

List of Figures .................................................................................................................. xi

Chapter 1 ............................................................................................................................1

1 Introduction ..............................................................................................................1

Chapter 2 ............................................................................................................................6

2 Subsurface generation of Fracture mapping .............................................................6

2.1 Fracture Properties ................................................................................................8

2.2 Fracture detection .................................................................................................9

2.2.1 Conventional well logs ..................................................................................9

2.2.2 Wellbore Image Logs ..................................................................................12

2.2.3 Outcrops ......................................................................................................13

2.2.4 Core Descriptions ........................................................................................13

2.2.5 Drilling Data ................................................................................................13

2.2.6 Seismic Data................................................................................................14

2.3 Post Processing of Data for Fracture Properties .................................................15

2.3.1 Core Description .........................................................................................18

vii
 2.3.2 Image Logs ..................................................................................................19

2.3.3 Caliper Log..................................................................................................20

2.3.4 Seismic ........................................................................................................21

2.3.5 Outcrops ......................................................................................................22

2.3.6 Tectonic Event Simulation ..........................................................................22

2.4 Field-Wide Generation of Discrete Fracture Map ..............................................31

2.4.1 Artificial Neural Networks ..........................................................................31

2.4.2 Stochastic Simulation ..................................................................................38

2.5 Closure ................................................................................................................49

Chapter 3 ..........................................................................................................................50

3 Fluid Flow Simulation and Heat Transfer ..............................................................50

3.1 Simulation of Fluid Flow in Fractured Media ....................................................51

3.1.1 Single Continuum Approach .......................................................................51

3.1.2 Dual Continuum Approach .........................................................................51

3.1.3 Discrete Fracture Network (DFN) Approach ..............................................53

3.1.4 Hybrid of Single Continuum and Discrete Fracture ...................................53

3.2 Domain Discretization Using the Hybrid Methodology .....................................54

3.2.1 REV Discretization for Permeability Tensor Calculation ...........................55

3.2.2 BEM Formulation .......................................................................................58

3.3 Reservoir Scale Fluid Flow Simulation ..............................................................61

3.4 Local Thermal Non-Equilibrium ........................................................................66

3.5 Discretization of the governing equations ..........................................................69

3.6 Results and discussions .......................................................................................71

3.6.1 Discrete Fracture Network ..........................................................................71

3.6.2 Pore pressure and Log10RMS Fluid Velocity..............................................71

viii
 3.6.3 Temperature ................................................................................................74

3.7 Closure ................................................................................................................75

Chapter 4 ..........................................................................................................................76

4 Stimulation .............................................................................................................76

4.1 Overview .............................................................................................................76

4.2 Current Approach ...............................................................................................90

4.3 Closure ................................................................................................................95

Chapter 5 ..........................................................................................................................96

5 Case Study ..............................................................................................................96

5.1 Model Setup ........................................................................................................96

5.2 Results and Discussions ....................................................................................100

5.3 Closure ..............................................................................................................118

Chapter 6 ........................................................................................................................119

6 Conclusions and recommendations ......................................................................119

6.1 Conclusions .......................................................................................................119

6.2 Scope for Future Research ................................................................................121

References ......................................................................................................................122

Appendix A: Probability of Occurrence of Fractures ....................................................134

Appendix B: Artificial Neural Network .........................................................................137

Appendix C: Finite Element Shape Function .................................................................141

ix
List of Tables
Table 2-1: Characteristic properties of different sets of fractures for Palm Valley
Reservoir (Tran 2004). ..........................................................................................................18
Table 2-2: Characteristic properties of the fractures sets obtained from the image logs
(Tran 2004) ...........................................................................................................................20
Table 2-3: Different fracture sets obtained from the caliper log ......................................21
Table 2-4: Fracture sets including their characteristic properties obtained from the
seismic interpretation(Tran 2004) .........................................................................................22
Table 2-5: fracture set with its own characteristic properties from the outcrop from Palm
Valley Reservoir ...................................................................................................................22
Table 2-6: Statistical Parameters for the Palm Valley Reservoir Obtained from the Field
Data ((Do Rozario 1991, Tran et al. 2006)) ..........................................................................42
Table 2-7: Characteristic statistical properties of the fractures obtained from the different
data sources, such as core sample descriptions, well logs etc. (Gentier et al. 2010). ...........46
Table 2-8: Statistical parameters for Patchwarra formation (Mildren et al. 2005) ..........47
Table 5-1: Characteristic statistical properties of the fractures obtained from the different
data sources, such as core descriptions, well logs etc. (Gentier et al. 2010) ........................97
Table 5-2: Stress and reservoir data for Soultz geothermal reservoir (Kolditz 1995,
Genter and Traineau 1996, Koh et al. 2010, Watanabe et al. 2010). Parameters are
considered to be constant during the cold fluid circulation. .................................................99
Table 5-3: Comparative study of the percentage increase in average fracture aperture
between the current study with that of Koh et al 2011. ......................................................103
Table 5-4: Reservoir parameters used for the sensitivity analysis (Baria et al. 1992,
Kolditz 1995, Sausse et al. 2006, Valley 2007, Watanabe et al. 2010) ..............................113

x
List of Figures
Fig. 2-1: A single disk shaped fracture characterized by centre point (A), Dip (  ),

Azimuth (  ) and radius(r) .....................................................................................................9

Fig. 2-2: Top P1 (horizon of interest) unit depth contour map of the Palm Valley
reservoir interpreted from seismic data (Roe 1995, Tran 2004). ..........................................15
Fig. 2-3: Box counting method for the determination of the fractal dimension value
(Tran 2004). ..........................................................................................................................17
Fig. 2-4: Plotting the data obtained from box counting method to determine the fractal
dimension (Tran 2004) ..........................................................................................................17
Fig. 2-5: Core Samples in Well A at Palm Valley Reservoir (Berry 1991, Tran 2004) ..19
Fig. 2-6: Wellbore Image taken from Well E, Palm Valley reservoir at the depth interval
of 2326m to 2333m (Rayner 1995, Tran 2004) ....................................................................20
Fig. 2-7: Caliper log from well A from Palm Valley reservoir at the depth interval of
1789m-1801m (Tran 2004) ...................................................................................................21
Fig. 2-8: Seismic interpretation and the detected fractures obtained from the Palm Valley
reservoir (Roe, 1995, (Tran 2004)) .......................................................................................22
Fig. 2-9: Flowchart for the calculation of velocity and pressure at each time step during
the tectonic Simulation (Gholizadeh Doonechaly and Rahman 2012) .................................27
Fig. 2-10: Unfolding, from (a) to (e) with time difference of 50,000 years and folding,
from (e) to (i) with the time difference of 50,000 years. A speed of 2 cm/yr is applied in X
direction from both sides in the numerical simulation. The finite element mesh contains 960
elements in the horizontal plane and 15 elements in the vertical direction thus summing up
to 14,400 total elements (Gholizadeh Doonechaly and Rahman 2012). ...............................30
Fig. 2-11: A schematic representation of the continuum map of the probability of
occurrence of the fracture in the Palm Valley reservoir for the top and bottom layer of the
formation with a thickness of 200m......................................................................................31
Fig. 2-12: A schematic representation of the flow chart for the neuro-stochastic
simulation of the field-wide discrete fracture map (Gholizadeh Doonechaly and Rahman
2012) .....................................................................................................................................32

xi
 Fig. 2-13: Location map of the drilled wells in the Palm Valley Reservoir (Berry et al.
1996). The Palm Valley area is shaded in the figure. ...........................................................33
Fig. 2-14: Fracture density and fractal dimension profile obtained by training the BPNN.35
Fig. 2-15: Neural network training (black) and validation (red) map for fracture density.
Optimal training reached after 48,720 epochs (iteration steps) after which the error
(difference between the simulated data and the actual field data) tends to increase because
of overtraining. At the optimal training point, all the obtained weight factors are recorded
and used to find the fracture density for other sections of the reservoir (Gholizadeh
Doonechaly and Rahman 2012). ...........................................................................................36
Fig. 2-16: Neural network training (black) and validation (red) map for fractal
dimension. Optimal training reached after 62,505 epochs (iteration steps) at which the
weight factors are recorded and used to calculate fractal dimension value for other parts of
the reservoir (Gholizadeh Doonechaly and Rahman 2012). .................................................36
Fig. 2-17: A cross section of the fractal dimension in P1 unit of horizon in the Palm
Valley reservoir (Gholizadeh Doonechaly and Rahman 2012). ...........................................37
Fig. 2-18: 3-D pixel based map of the fracture density at the depth of (a) 1000 m, (b)
1200, (c) 1400 m, (d) 1600 m, (e) 1800 m, (f) 2000m. The reservoir is divided into
sufficient number of blocks and fracture density value is calculated for each block which is
used for stochastic simulation of discrete fracture network (Gholizadeh Doonechaly and
Rahman 2012). ......................................................................................................................37
Fig. 2-19: Flowchart for Generating Discrete Fracture Network by using Sequential
Gaussian Simulation Technique (Gholizadeh Doonechaly and Rahman 2012) ...................39
Fig. 2-20: Circular statistical analysis: histogram and probability density function plot
for fracture azimuth. Statistical analysis on the circular azimuth data reveals a bimodal
wrapped normal distribution, with modes at approximately 6° and 172° (Tran 2004). .......41
Fig. 2-21: Circular statistical analysis: histogram and probability density function plot
for fracture dip showing one-modal wrapped normal distribution, whose mode is at 176°
(Tran 2004). ..........................................................................................................................42
Fig. 2-22: Discrete fracture network map of the Palm Valley reservoir (Gholizadeh
Doonechaly and Rahman 2012). ...........................................................................................43

xii
 Fig. 2-23: Validating results for fracture density and fractal dimension reproduced for
wellbore locations. The correlation values are close to 1 (0.8401 and 0.9386 for fracture
density and fractal dimension respectively) (Gholizadeh Doonechaly and Rahman 2012). 44
Fig. 2-24 Rose diagrams of fracture azimuth for Soultz reservoir (a) massive un-
fractured granite (N = 675 data), (b) moderately fractured granite (N = 554 data) and (c)
highly fractured granite (N = 1767 (Genter and Traineau 1996). .........................................45
Fig. 2-25: Histogram of fracture-dip directions for the Soultz-Sous-Forets geothermal
reservoir (Genter and Traineau 1996) ...................................................................................45
Fig. 2-26: Discrete fracture network map of Soultz-Sous-Forets geothermal reservoir ..46
Fig. 2-27: Statistical wrapped distribution of (a) Dip and (b) azimuth for Patchwarra
formation. (Mildren et al. 2005)............................................................................................47
Fig. 2-28: Discrete Fracture Network map of the Patchwarra formation.........................48
Fig. 3-1: Warren and Root (1963) used the sugar cube model for representing the
fractured rock in a dual continuum approach. In their model, the real fracture rock (a) is
represented by evenly spaced and infinitely distributed fractures which provide the main
flow conduit. .........................................................................................................................52
Fig. 3-2: Domain discretization by using the hybrid of the single continuum and discrete
fracture approach. (a) Fractures longer than 50 m are explicitly in the reservoir domain by
using the triangular elements. (b) After the discretization of the long fractures, the effect of
the short fractures (<50m) are taking into account by calculation of the permeability tensor
of the corresponding blocks which are cut by the fractures (Doonechaly et al. 2013). ........54
Fig. 3-3: Domain discretization based on different fracture lengths (Doonechaly et al.
2013). ....................................................................................................................................56
Fig. 3-4: intersecting region creates by two fractures. Ap is the intersection region
created by intersecting fractures. ..........................................................................................59
Fig. 3-5: Two medium treatment of the heat transfer in a fractured porous medium ......67
Fig. 3-6: A schematic representation of the solid/fluid interface. B is the length of the
interface and b is the fluid width (fracture aperture). ...........................................................68

xiii
 Fig. 3-7: Matrix/fracture interface in a discretised domain using triangular elements. As
is the area of the shaded element and Af is the area of a part of the fracture adjacent to the
corresponding matrix element...............................................................................................69
Fig. 3-8: (a) a schematic representation of the discrete fracture network of Soultz
geothermal reservoir at the depth of 3650 m (b) generated mesh of the domain using the
triangular elements (c) a sample area containing short fractures (d) calculated permeability
tensor of for the elements which are cut by short fractures . ................................................72
Fig. 3-9: A schematic representation of pore pressure distribution (top) and Log10RMS
fluid velocity (bottom) after (a) 1 and (b) 12 years of production for  H =78.9MPa,  h
=53.3MPa, Pinj=44.8MPa, Pprod=31MPa, Tinj=150oC . .........................................................73
Fig. 3-10: A schematic representation of the Reservoir temperature drawdown after (a) 1
year and (b) 12 years of production for  H =78.9MPa,  h =53.3MPa, Pinj=44.8MPa,
Pprod=31MPa, Tinj=150oC . ....................................................................................................74
Fig. 4-1: Mohr diagram describing the initiation of the shear dilation and normal fracture
surface separation (Murphy and Fehler 1986, Doonechaly et al. 2013). ..............................77
Fig. 4-2: roughness induced shear dilation caused by the fracture asperities. .................78
Fig. 4-3: Shear dilation vs. shear displacement for a hypothetical fracture specimen
(Piggott and Elsworth 1991). ................................................................................................82
Fig. 4-4: Experimental set up used by Olsson and Brown (1993). (a) Rotary shear
sample with radial flow injection (b) schematic representation of the aluminium core holder
(Olsson and Brown 1993). ....................................................................................................83
Fig. 4-5: A schematic representation of a fracture undergoing shear displacement and the
required parameters to characterize the dilation event..........................................................85
Fig. 4-6: Penny-shaped fracture undergoing propagation subject to maximum and
minimum horizontal stresses (Rahman et al. 2000). .............................................................89
Fig. 4-7: Fracture opening vs. shear slippage of fracture surfaces.  is the characteristic
height of the fracture surface asperities. ...............................................................................90
Fig. 4-8: A schematic representation of a fracture subject to in-situ stress boundary
conditions (Kotousov et al. 2011). ........................................................................................91

xiv
 Fig. 5-1: (a) a schematic representation of the discrete fracture network of Soultz
geothermal reservoir at the depth of 3650 m (b) generated mesh of the domain using the
triangular elements (c) a sample area containing short fractures (d) calculated permeability
tensor of for the elements which are cut by short fractures. .................................................98
Fig. 5-2: Percentage increase in average fracture aperture with stimulation time for
Pinj=65MPa,  min =53.3MPa, ..............................................................................................100
Fig. 5-3 : Cumulative location of the shear dilation events after (a) 20 weeks (b) 38
weeks (c) 50 weeks and (d) 60 weeks of stimulation for Pinj=65MPa ,  min =53.3MPa,  max
=78.9MPa. ...........................................................................................................................101
Fig. 5-4: Comparison of the percentage of average aperture increase between the current
study and the previous approach .........................................................................................103
Fig. 5-5: A comparison between the dilation events in this study (filled circles) with that
from the field test trial (unfilled).........................................................................................104
Fig. 5-6: Pore pressure distribution of the fractured reservoir at different stimulation
stages: after (a) 40 weeks and (b) 60 weeks with σH = 78.9MPa and σh = 53.3MPa, Pinj =
65MPa. ................................................................................................................................104
Fig. 5-7: x (top) and y (bottom) components of effective stress after: (a) 40 weeks and
(b) 60 weeks of stimulation for σH = 78. 9MPa and σh = 53.3MPa, Pinj = 65MPa. .........105
Fig. 5-8: Pore pressure distribution at different production stages: after (a) 3 years and
(b) 14 years of production for σH = 51.7MPa and σh = 44.8MPa, Pinj = 51.7MPa and
Pprod=31.7MPa. .................................................................................................................106
Fig. 5-9: Log10 RMS velocity profile at different production stages: after (a) 3 years and
(b) 14 years of production for σH = 51.7MPa and σh = 44.8MPa, Pinj = 51.7MPa and
Pprod=31.7MPa. .................................................................................................................106
Fig. 5-10: Production rate for pressure draw down of 20MPa over a production period of
14 years after 60 weeks stimulation for Pinj = 51.7MPa and Pprod=31.7MPa (Gholizadeh
Doonechaly et al. 2012). .....................................................................................................107
Fig. 5-11: Average matrix temperature and produced fluid temperature vs. time for 14
years of production .............................................................................................................108

xv
 Fig. 5-12: Reservoir temperature profile after (a) 1 year (b) 3 years, (c) 10 years and (d)
14 years of production for σH = 51.7MPa and σh = 44.8MPa, Pinj = 51.7MPa and
Pprod=31.7MPa.. ................................................................................................................109
Fig. 5-13: x (left) and y (right) component of effective stress with thermal stress after (a)
1 year (b) 3 years (c) 10 years and (d) 14 years of production for σH = 51.7MPa and σh =
44.8MPa, Pinj = 51.7MPa and Pprod=31.7MPa. ................................................................111
Fig. 5-14: x (left) and y (right) component of effective stress without thermal stress after
(a) 1 year (b) 3 years (c) 10 years and (d) 14 years of production for σH = 51.7MPa and σh
= 44.8MPa, Pinj = 51.7MPa and Pprod=31.7MPa. ............................................................112
Fig. 5-15: Sensitivity analysis of the effect of uncertainty in fracture density on the
average matrix temperature for a production period of 14 years for σH = 51.7MPa and σh =
44.8MPa, Pinj = 51.7MPa and Pprod=31.7MPa. ................................................................114
Fig. 5-16: Sensitivity analysis of the effect of uncertainty in fractal dimension on the
average matrix temperature for a production period of 14 years for σH = 51.7MPa and σh =
44.8MPa, Pinj = 51.7MPa and Pprod=31.7MPa. ...................................................................115
Fig. 5-17: Sensitivity analysis of the effect of uncertainty in fracture aperture on the
average matrix temperature for a production period of 14 years for σH = 51.7MPa and σh =
44.8MPa, Pinj = 51.7MPa and Pprod=31.7MPa. ................................................................116
Fig. 5-18: Sensitivity analysis of the effect of uncertainty in initial fracture aperture on
the percentage of increase in average fracture aperture for the stimulation period of 60
weeks for σH = 78.9MPa and σh = 53.3MPa, Pinj = 65MPa. ............................................117
Fig. 5-19: Sensitivity analysis of the effect of uncertainty in rock permeability on the
average matrix temperature for a production period of 14 years for σH = 51.7MPa and σh =
44.8MPa, Pinj = 51.7MPa and Pprod=31.7MPa. ................................................................117

xvi
Chapter 1

1 INTRODUCTION
A significant proportion of world’s geothermal systems are characterised by crystalline
rocks, high in-situ stresses and poorly interconnected fracture systems, steep thermal
gradients and lack of aquifer support for natural recharge (Narayan et al. 1998, MIT 2006).
These conditions lead to naturally high impedance between injector and producer wells,
thus making them commercially not viable. In order to overcome the fluid flow barrier
hydraulic stimulation has been employed to open, extend and interconnect the pre-existing
natural fractures (Taleghani 2009, Zimmermann et al. 2010, McClure and Horne 2013).
Commonly used hydraulic stimulation techniques include gel-proppant fracs, high-pressure
low-injection-rate water fracs (to induce shear dilation) or a combination of both (Hossain
2001). In hydraulic fracturing, sufficiently high pressure is used to cause shear failure. Upon
the cessation of fluid injection a fraction of the dilated aperture is retained by frictional
resistance of rough surfaces of natural fracture (Chen et al. 2000). Typically, to initiate shear
displacement of natural fractures, the maximum bottom-hole injection pressure is kept in
excess of the minimum formation stress at the injection point. Numerous attempts have been
made to maintain injection pressure at a much higher level than the minimum formation
pressure (Tester et al. 2007). However, there are two primary negative effects associated with
increasing the injection pressure in such a manner: Firstly, fractures at higher pressures can be
‘jacked’ open, creating few dominant flow paths, and secondly it can exceed the tensile
strength of the rock leading to fracture propagation which can extend beyond the circulation
region, enabling water to be lost, thus reducing the effective heat-transfer area (Murphy and
Fehler 1986, Brown 1987).

In a recent numerical study by Koh and Rahman et al (2011) it was observed that for a given
injection pressure, rock properties and stress conditions, the average change in aperture (shear
dilation) is a function of time. According to this study, there exists a threshold time below
which the change in average aperture is insignificant. After this time, the average aperture
increases rapidly (until it reaches a plateau). This means that for every set of reservoir
1
parameters, stress condition and injection schedule, an optimum level of shear dilation can be
achieved. Increase in average fracture aperture is also found to be directly related to the
orientation of fractures (Shaik et al. 2009, Shaik et al. 2011).

A subsequent poro-thermo-elastic reservoir study by Koh et al. (2010), Koh et al. (2011),
Siratovich (2011) and Gholizadeh Doonechaly et al. (2012) it was also shown that the long-
term circulation of low temperature fluid has a significant bearing on reservoir permeability.
Thermal contraction due to cooling reduces the effective stresses (normal to the fracture
surfaces), causing an increase in aperture (Siratovich et al. 2011). However, it is noteworthy to
mention that fluid-induced stresses (poro-elastic effect) dominates at early injection times while
thermally induced stress dominates at late injection times, due primarily to circulation of
injection fluid (Haupt 2000, Shaik et al. 2011). Similar results were observed by Ghassemi et
al. (2011) using a poro-thermo-elastic reservoir model featuring a single co-planar fracture.

Field observations and numerical studies have shown that permeability enhancement by
induced fluid pressure can be maximised by careful selection of a set of parameters including
injection time, injected volume and injection schedule (Rahman et al. 2002, Koh et al. 2011,
Pogacnik et al. 2012, Wang et al. 2012). Many authors have modelled the effects of different
parameters on the permeability change in a loose or decoupled form (Watanabe and Takahashi
1995, Hossain et al. 2002, Jing 2003, Ghassemi and Zhou 2011, Pengcheng and Carrigan 2012,
Wang and Ghassemi 2012). Others have presented coupled formulation of the abovementioned
problem, but on a small local scale or with simplified fracture geometry (Ghassemi et al. 2007).
Constitutive models were also derived by Bandis et al. (1983), Barton et al. (1985) and Willis-
Richards et al. (1996) to predict shear stress-displacement behaviour at any stress level of
interest, e.g. bottom-hole conditions. In another attempt Asadollahi and Tonon (2010) extended
Barton’s model (1985) to consider the peak shearing strength in the rock deformation. Weng et
al. (2010) also developed an elastic-viscoplastic model to consider the effect of shear
displacements in weak rocks (Weng et al. 2010).

Although numerous studies have made remarkable contributions to our knowledge and
provided a framework for future research, they are largely based on simplified empirical
relationships, including relationships used to calculate compressive normal surface traction,
residual aperture and shear displacement, which in some cases are derived from a simple
fracture experiment or a best guess. Recent studies, however, have shown that hydraulic
2
stimulation is highly dependent on the in-situ stress condition and characteristic rock and
fracture properties (Fu et al. 2011). Simple empirical relations can not be used to develop
stimulation strategies and often leads to erroneous predictions. Fluid flow and heat transfer in
crystalline geothermal reservoirs are dominated by the system of interconnected fractures
which provide the main flow conduits. Therefore, mapping of subsurface fracture and
modelling fluid flow and heat transfer need to be considered as an integral part of the
stimulation exercise in order to overcome its inherent limitations.

A fracture system in a geothermal reservoir is characterised by fracture density, fractal

dimension, fracture geometry and fracture aperture. There exist two distinct approaches to
generate discrete subsurface fracture map. They include mechanical and statistical
approaches (Gholizadeh Doonechaly and Rahman 2012). Mechanical approaches are used
to predict subsurface strain distribution to estimate intensity of brittle to brittle-ductile
deformations (Regenauer-Lieb 1998, Bourne et al. 2000, Bourne and Willemse 2001,
Maerten et al. 2002, Weinberg and Regenauer-Lieb 2010). Tectonic theories have also been
used extensively as a powerful geomechanical tool to generate discrete fracture networks
(Price 1966, Reddy et al. 1983, Cosgrove 1999, Jäger et al. 2008, Markner-Jäger 2008,
Gholizadeh Doonechaly and Rahman 2012). Statistical approaches, however, include a
wide range of techniques to generate the subsurface fracture map of the reservoir
(Gringarten 1998, Jensen et al. 2010). The main issue with stochastic approach, however, is
that a large amount of data is needed from a variety of sources (Gholizadeh Doonechaly
and Rahman 2012). A combination of geomechanical and geostatistical approaches have
also been used to simulate naturally fractured reservoirs (Rawnsley and Swaby 1996,
Gringarten 1998, Josnin et al. 2002). In these approaches, however, the relationship
between fracture spacing and its stochastic parameters, such as variograms are not
considered (Riley 2004). Also the spatial variation of fracture density and fractal dimension
are not integrated into the modelling procedure (Chilès 2005).

In order to examine displacement behaviour of fracture surfaces induced by pressure one

needs to know pressure profile away from the wellbore. Significant number of studies have
been carried out to study fluid flow in fractured reservoirs in a coupled framework (Borja
2006). Single porosity concept has been proposed for simulating coupled processes in fractured

3
domains (Arbogast et al. 1990). However, single porosity models cannot be effectively applied
to the large scale fractured reservoirs because of their limitations in estimation of the fracture
deformation (Bai 1999, Berryman 2002). To overcome such a limitation, dual porosity concept
has been proposed by different authors to simulate both coupled and uncoupled processes in
fractured reservoirs (Sonnenthal et al. 2005). Dual porosity approaches, however, are
applicable to uniform or idealised fracture pattern (Kim et al. 2012). Also most commonly used
heat transfer model in the abovementioned approaches was based on the thermal equilibrium at
the matrix fracture interface (Lu and Xiang 2012). There exists a thermal gradient at the
matrix/fluid interface which means that the temperature of the rock matrix and fluid are not
equal at their interface. Such a simplification causes erroneous predictions of the heat transfer
between the rock matrix and the circulating fluid (Gelet et al. 2012).

In all these studies following critical issues are not addressed:

- The effect of tectonic events on the generation of the subsurface fractures completely
ignored or considered as loosely coupled with the stochastic simulation techniques to
generate the subsurface fracture maps

- Explicit investigation of the effect of fractures on the fluid flow and heat transfer

- Thermal gradient at the rock matrix/fluid interface which introduces thermal stresses into
the system

- Effect of the surface roughness asperities in estimating the shear dilation of the fracture
surfaces

In this thesis, these shortcomings are addressed by firstly a hybrid of numerical simulation
of tectonic history of the formation and stochastic simulation of field data to generate the
subsurface discrete fracture map. Next fluid flow is simulated by combining grid-based
permeability tensor with flow through discrete fractures. Heat transfer is coupled with fluid
flow and is simulated by considering the thermal gradient and the resulting stress gradient in
rock/fluid interface. Finally the obtained pressure and temperature distribution is applied to
estimate the permeability changes within the reservoir by considering the roughness induced
shear dilation of the fracture surfaces caused by the induced fluid pressure.

In Chapter 2 a hybrid methodology is proposed to generate the subsurface fracture map of

the reservoir. For this purpose the tectonic history of the formation is simulated using Finite
4
Element Method. As a result of this simulation the filed wide distribution of the probability of
occurrence of fractures is obtained. Then the obtained probability values are integrated into the
stochastic simulation to randomly realise different fracture properties, such as dip, azimuth,
radius and centre point.

In Chapter 3 a combined methodology to simulate fluid flow and heat transfer in naturally
fractured reservoirs is presented. Fluid flow is simulated by a combination of single continuum
and discrete fracture approach in which fractures with lengths larger than 50m are discretely
meshed in the domain. Heat transfer between the fluid and the rock matrix are estimated by
Local Thermal Non-Equilibrium (LTNE) approach in which thermal gradient at the rock matrix
and fluid interface is considered.

In Chapter 4 a third methodology to estimate the shear dilation of the fractures surfaces
caused by induced fluid pressure is described. The analytical model for dilation of fracture
surfaces is described based on the distributed dislocation technique. The roughness of the
fracture surfaces is considered in the calculations.

In Chapter 5 these above mentioned methodologies are coupled and then applied to the
Soultz-Sous-Forets geothermal reservoir to estimate the effect of cold fluid circulation on long
term poro-thermo-elastic response of the reservoir. An uncertainty analysis is also carried out to
study the effect of variation of major parameters on the reservoir behaviour during the fluid
circulation period.

In Chapter 6 based on the observations made in chapters 2 to 5 conclusions are drawn and
recommendations made for further studies.

5
Chapter 2

2 SUBSURFACE GENERATION
OF FRACTURE MAPPING
Velocity and pressure data across the reservoir are critical to the study of fracture
stimulation behaviour by induced fluid pressure as well as heat transfer to the circulating
fluid from the matrix. Successful acquisition of velocity and pressure data again depends
primarily on availability of accurate information about subsurface fracture map. There exist
two distinct approaches to generate discrete subsurface fracture map. They include
mechanical and statistical approaches (Gholizadeh Doonechaly and Rahman 2012).
Generally, geomechanical approaches are used to predict subsurface strain distribution to
estimate intensity of brittle to brittle-ductile deformations (Bourne et al. 2000, Bourne and
Willemse 2001, Maerten et al. 2002). As an example, Weinberg and Regenauer-Lieb
(2010) used a numerical model to simulate rock creep and growth of micro-scale voids
which become interconnected and result in rock failure. The authors proposed that ductile
fractures are the dominant path for migration of magma (Regenauer-Lieb 1998, Regenauer-
Lieb et al. 2006, Weinberg and Regenauer-Lieb 2010). Bourne et al. (2000) used seismic
data to predict sub-seismic faults with the assumption that the dominant control on small
scale faults is the strain perturbation caused by large faults. Also tectonic theories have
been used extensively as a powerful geomechanical tool to generate discrete fracture
networks (Price 1966, Reddy et al. 1983, Cosgrove 1999, Jäger et al. 2008, Markner-Jäger
2008). This involves estimation of stress distribution during folding and/or unfolding of a
layer due to tectonic events (Ramsay 1983, Ramsay and Huber 1983). Folding and
unfolding have been modelled numerically by fluid mechanics theories with the assumption
that the rock is treated as nearly incompressible fluid (O'Driscoll 1962, Ramberg 1963, Lan
and Hudleston 1991, Schmalholz 2008). Schmalholz (2008) showed that 3-D folds can be
accurately restored to their original shape in a single extension event (reverse model) in a
reversible process. During the folding event, the stress distribution can be calculated for all
6
parts of the layer based on the deformation (strain rate) (Ramberg 1963). The stress
invariants obtained from the calculated stress tensor are used to estimate the probability of
occurrence of fractures. Therefore tectonic models cannot be effectively used to generate
subsurface fracture maps. A wide variety of geo-statistical techniques have been used to
characterise different properties of natural fractures (Gringarten 1998, Jensen et al. 2010).
Numerous authors have used stochastic simulation for petroleum reservoirs to generate 3-D
fracture map (He et al. 2001, Bahar 2003, Bogatkov 2008, Rafiee 2008). Sequential
Gaussian stochastic simulation is one of the many geo-statistical tools to generate
subsurface fractures (Sahimi 1995, Parney 2000, Tamagawa 2002, Tran 2006). In this
method fractures are treated as distinct objects with their own attributes, such as dip,
azimuth, radius etc. Each fracture is randomly realized based on the available probability
density functions and then checked with the boundary conditions. The main issue with
stochastic approach, however, is that a large amount of data is needed from a variety of
sources, such as conventional well logs, seismic attributes, core descriptions, borehole
images etc (Gholizadeh Doonechaly and Rahman 2012).

There were also attempts to combine geomechanical and stochastic approaches to

simulate naturally fractured reservoirs. For example Gringarten (1998) as well as Rawnsley
and Swaby (1996) incorporated rule-based fracture propagation into stochastic simulation.
In these approaches, however, the relationship between fracture spacing and its stochastic
parameters, such as variograms are not considered (Riley 2004). In another attempt Cacas
et al. (1997) used a combination of deterministic analysis of the spatial variation of fracture
attributes at field scale and stochastic simulation at the local scale to simulate the fracture
network. They, however, did not consider spatial variation of fracture density and fractal
dimension which is considered important in simulation of joints and fractures (Chilès
2005).

In this study a hybrid of tectonic modelling and stochastic simulation techniques

(Gholizadeh Doonechaly and Rahman 2012) is used to simulate the discrete fracture
network in three different naturally fractured geothermal reservoirs. In the first part of this
study Finite Element Method (FEM) is used to reconstruct a layer by unfolding and folding
(reverse and forward model) in a single tectonic event. After reconstructing the original
7
shape of the layer, stress tensor values as well as the stress invariants are calculated for
each node. If the obtained stress tensor satisfies failure criterion, such as Mohr Coulomb,
stress invariants are used to calculate the rate of formation of fractures for each node
(Somerton et al. 1970, Wickham et al. 1982, Nelson and Laubach 1994, Fjar et al. 2008,
Jaeger et al. 2009). Rate of formation of fractures is used as one of the input parameters for
the next step which is the stochastic simulation. In stochastic simulation, the inter-
relationships between all available input data are established. Both continuum and discrete
measurements are integrated in a hierarchical ordering simulation. Also a neural network is
used to construct a continuum map of fracture density and fractal dimension based on the
available data from the reservoir. These maps are integrated in the discrete model. Finally,
the sequential Gaussian stochastic simulation technique is used to generate the discrete
fracture network (Tran 2004, Tran et al. 2006, Gholizadeh Doonechaly and Rahman 2012).

2.1 Fracture Properties

Every fracture is characterized by specific properties which distinguish from one

fracture to other. For this purpose different set of properties are defined in the literature
(Aguilera 1980, Nicholson 1994, Nelson 2001). In this study the fractures are treated as
penny shaped disks as shown in Fig. 2-1. The uniqueness of each fracture is defined based
on its orientation, centre point, radius and aperture. Orientation of any single plane in space
can be defined based on two values namely dip and azimuth (Groshong 1999). Dip (  in
Fig. 2-1) is defined as the angle between the fracture plane and the horizontal plane and it
changes from 0°-90°. Azimuth (  in Fig. 2-1) is the angle which the fracture intersection
line with the horizontal plane makes with the North direction. Azimuth value changes from
0°-180°. Since the fractures are considered as circular disk objects, their centre point is
defined as the centre of the disk (point A in Fig. 2-1). The radius of the fracture is defined
based on the radius of the disk (r in Fig. 2-1).

Fracture density is also an important property of a system of fracture network. Fracture

density is a measure of the cumulative length (area) of all the fractures in unit area
(volume) in a system of fracture network. Fractal dimension is also used to characterise

8
self-similar objects such as fractures. Fractal dimension is a measure of the irregularity in
the fracture pattern (Sahimi 1993, Watanabe and Takahashi 1995). Therefore, determining
the fractal dimension is critical in characterising reservoirs in which fractures provide the
main flow paths.

Z N

r
E
A

Fig. 2-1: A single disk shaped fracture characterized by centre point (A), Dip ( ), Azimuth ( ) and radius(r)

2.2 Fracture detection

There exists a wide range of data sources to detect fractures with their characteristic
properties in a variety of scales. Amongst them are core descriptions, conventional well
logs, seismic attributes, drilling data and wellbore images. In the following sections a brief
introduction is given for different data sources at wellbore scale and reservoir scale.

2.2.1 Conventional well logs

Well logs provide us with continuous information from the wellbore that can be used to
infer properties of fractures (Song et al. 2000, Martinez-Torres 2002). These logs are used
in almost every wellbore because of their availability and relatively low cost (Schatzinger
and Jordan 1999). Interpretation of conventional well logs for fractures is discussed below.

9
 Caliper Log

Caliper log represents the size of the bore hole. Fractures affect the caliper log
measurement by wellbore elongation or wellbore radius reduction along the direction of
fractures in vertical wellbore. Mud cake is formed especially when mud loss preventer
materials are used in which case wellbore diameter can be reduced. Both abovementioned
mechanisms, however, are not a guaranteed representative of the presence of fractures since
highly porous formations can also produce a thick mud filter cake. Also borehole breakouts
may happen during drilling in unconsolidated formations. Similar caliper responses
(wellbore elongation or wellbore radius reduction) are observed in shale and vugs (Tran
2004). Therefore, the caliper measurements are only indicative and need to be validated by
using other data sources (Wong et al. 2002, Ellis and Singer 2007, Serra 2007).

Spontaneous Potential (SP) Log

The SP log can be used for correlation, lithology indication, porosity, permeability and
formation water salinity. SP log is a measure of the potential difference between two
electrodes: One placed at the surface and the second electrode placed in the bore hole.
When the drilling mud invades the formation the contact between the mud filtrate and the
formation fluid produces an electric potential. At the presence of fractures, SP may show
irregular pattern or negative deflection due to the streaming of the ions from mud filtrate to
the formation fluid. It is, however, noteworthy to mention that such streaming may be
caused by the silt bed (Shanks et al. 1976). Also the SP log cannot be used in oil or air
based mud in which there is no conductive fluid in the wellbore (Asquith 1985, Darling
2005, Serra 2007).

Gamma Ray (GR) Log

GR log is a measure of formation radioactivity based on the natural Gamma ray

emission from the formations. Gamma ray log can also be used in cased holes since many
gamma rays can pass through the steel. Gamma ray log are used for three purposes,
namely: Correlation, evaluation of the shale content and mineral analysis. It is used mainly
for the quantitative measurement of the volume percentage of shale in the formation.

10
Presence of fractures may show the same pattern because of the precipitation of the
radioactive salt on the fracture surface or inside the fracture itself when the potassium
based mud is used. Carbonate precipitation in permeable zones such as fractures can also
cause high readings in GR value (Serra 2007, Satter et al. 2008, Fanchi 2010).

Density Log

Density log, estimates the bulk density of the formation which is composed of two
components (Kadkhodaie-Ilkhchi et al. 2009): Uncompensated density and a correction
value. Density log can be used in either oil-based or water based muds and the depth of
investigation of the density log is quite low. At presence of fracture the density log
measures lower value since the fluid filled fracture possesses a lower density compared
with the intact porous rock. Therefore negative peak in the density log and a corresponding
peak in the correction value may be a reason of presence of fractures (Gartner and Suau
1980, Mohebbi et al. 2007). Density log can also detect the porosity anomalies because of
the edges of the fractures which have been chipped away during the drilling operations
(Wong et al. 2002, Tran 2004, Serra 2007).

Neutron Log

Neutron log is a measure of the formation porosity and lithology. It can be used in open-
hole, cased hole, fluid filled and air-filled boreholes. In the presence of fractures, neutron
log is expected to show a similar behaviour as density log with showing a peak value.
However, thin layers of different lithology and change in the bore-hole diameter may cause
the same response in Neutron log. Also presence of vugs, gas bearing zones and drilling
mud invasion affects the Neutron Log readings (Johnson and Pile 2002, Teimoori et al.
2005, Serra 2007).

Sonic Interval Transit Time

Sonic interval transit time measures the time required for a sonic wave to travel a
specific length of the formation. The sonic wave is recorded when it reaches a threshold
level of the energy. If the energy of the arrived sonic wave does not reach the threshold
level it causes the cycle-skipping. Presence of fractures which are perpendicular to the tool

11
axis causes the cycle skipping in the log measurement. Besides fractures gas-cut, light
hydrocarbons and unconsolidated formations may also cause the cycle skipping in the log
measurement. Also high attenuation of the sonic waveform logs is an indicative of the
presence of fractures. Generally, high-received signal strength shows a more competent
rock which decreases the probability of presence of fractures (Tran 2004). The value of the
Poisson’s ratio can also be estimated based on the sonic wave’s (both compressive and
shear waves) velocity and transit time. High Poisson’s ratio is an indication of high
probability of presence of fractures. Information obtained from the sonic logs can be used
to estimate the fracture orientation as well. Also a combination of the sonic and neutron
logs can be used to generate a pseudo-acoustic curve (Song et al. 1998) and compared with
the seismic velocity detect the fractures (Luthi 2001, Tran 2004, Serra 2007).

Resistivity Log

Resistivity logs are a measure of the conductivity of the formation. The dual latero log
device measure the resistivity of the reservoir in three different radial depths: deep (LLD),
shallow (LLS) and invaded zone (MSFL). Because the fractures are filled with the drilling
mud, in particular water based muds, the resistivity log measurements show abnormally
high conductivity specifically in the invaded zones which also causes positive separation in
three induction logs namely SFL, ILM and ILD. On the other hand the effect of the fracture
properties, such as dip, aperture, type of invaded fluid and the length of fracture are needed
to be considered for fracture interpretation (Serra 1984, Da Prat 1990, FitzGerald et al.
1999, Assaad et al. 2004, Tran 2004, Serra 2007, Serra 2008).

2.2.2 Wellbore Image Logs

With the advancing of the technology more accurate imaging devices emerges and
imaging logs are widely applied. Amongst them are Borehole Televiewer (BHTV),
Formation Micro Imager (FMI) and Formation Micro Scanner (FMS). Both FMS and
BHTV are the most important image logs which provide high-resolution images of the
borehole. They are considered as the most effective logging techniques (Martinez-Torres
2002). It is, however, noteworthy to mention that the induced fractures during drilling
operations or image distortions may cause errors in the interpretation for fractures. Also
12
relatively high cost of the image logs limits their application (Williamson et al. 1999, Luthi
2001).

2.2.3 Outcrops

Outcrops play a very important role in understanding the spatial orientation of the
fractures, the lithology and the relationship between the regional structures and the
fractures (Song et al. 2000). Outcrops are also valuable source of information to determine
fracture density, aperture, spacing, stress regime etc. (Harstad et al. 1995, Lynn et al. 1995,
Olson et al. 1998). Underground fracture patterns can be estimated using the extrapolation
of the outcrop properties (Aguilera 1980, Nelson 2001), however, the effect of overburden
stress and weathering are needed to be considered when analysing the data Tran
2004(Wealthal et al. 2001, Ameen and London 2003, Lonergan 2007).

2.2.4 Core Descriptions

Core description provides a basis for the fracture identification and analysis. Amongst
the most important fracture properties which can be obtained from core samples are:
fracture aperture, fracture dip and fracture azimuth. Besides the fracture properties, the
response of the well logs to the fracture parameters can also be validated using the core
descriptions. Such an analysis can gives us a better understanding of how the lithology,
fracture spacing, filled material and fracture direction affects the well log response of the
formation (Song et al. 2000, Chiaramonte and University 2009).

2.2.5 Drilling Data

Drilling data can be used as a qualitative measurement of the presence of fractures

(Tran 2004). Increase of drilling exponent value, larger cutting size and lost circulation are
amongst the most important representatives of the fractures (Brown 1987, Bunge 2000). It
should also be considered that other phenomena, such as wellbore collapse may cause the
same response (Tran 2004).

13
2.2.6 Seismic Data

Seismic data are reliable data sources to have an understanding of the fractures in
regional and especially inter-well scale. In seismic surveys, an acoustic source and receiver
are placed on the surface. Acoustic waves are generated and they travel deep into the
ground. The velocity of the waves depends on the density and rock mechanical properties.
Wherever the formation is highly heterogeneous or a discontinuity exists, a portion of the
travelling wave is reflected. After the reflected waves reach the surface, they are captured
and recorded by the receivers as a sinusoidal waves based on the travel time. The obtained
seismic profiles are regional scale which can cover from tens to hundreds of feet depending
on the wavelength and amplitude of the signals (Tran 2004). Thickness, structure,
curvature, lithology, big faults and fractures and mechanical properties of the rock can be
obtained based on the seismic data (Ouenes et al. 2005). Seismic waves which travel into
the ground are categorized in two subclasses namely S-waves and P-waves.

P-waves show the fracture density and fracture azimuth since the velocity of the wave
decreases when moving perpendicular to the fractures (Kuwahara et al. 1991, Beckham
1996, Tran 2004). P-waves are also affected by the vertical and high angle fractures
(Nakagome et al. 1998).

S-waves are polarized into fast and slow directions when passing through the fluid filled
fractures and the time delay between the polarized waves can be used to determine the
fracture density (Iverson 1992, Li 1997, Tran 2004). Mapping of the polarisation can also
be used to detect the fracture orientation. S-waves are more affected by the horizontal and
low angle fractures. In addition to fracture density and orientation, S-waves are also
affected by the permeability and bulk density of the rock (Kouider El Ouahed et al. 2003).

Seismic signal has several attributes which most of them are affected by reservoir
heterogeneities, such as fractures. Seismic attributes are based on amplitude, phase and
velocity of seismic signal. For example Amplitude Variation with Offset (AVO) is used to
recognise anomalies in the seismic wave reflection due to fracture orientation. A 3-D AVO
provides fracture anisotropy, spacing and density. As another example Amplitude versus

14
Angle and Azimuth (AVAZ) analysis can trace open, filled with fluid or gas filled
fractures. Variation of amplitude along shot receivers offset of P-wave reveals fracture
orientation and fracture density. Also it is noteworthy to mention that only medium to large
scale fractures can be interpreted from the seismic attributes.(Gray and Head 2000, Smith
and McGarrity 2001, Shen et al. 2002, Neves et al. 2003).

2.3 Post Processing of Data for Fracture Properties

In this study three different reservoirs are considered: Palm Valley (a gas reservoir in
Amadeus basin), Patchwarra (a geothermal reservoir in Cooper basin) and Soultz-Sous-
Forets geothermal reservoir in France. Among the three reservoirs analysis of data from
palm valley reservoir will be used as examples. The Palm Valley gas reservoir is located in
the Amadeus basin in the Northern Territory. It is a nearly symmetrical East–west trending
anticline. The area considered in this study trends 10,000m in x direction, 3000m in y
direction with an average thickness of 1000m (Wells et al. 1970, Do Rozario 1991, Tran
2004, Gholizadeh Doonechaly and Rahman 2012). Field data, such as core descriptions,
well logs and borehole images were collected from five wells. Also 2D seismic survey data
including time, instantaneous phase and instantaneous frequency were also available for
interpretation for fracture characterization (Milne and Barr 1990, Tran 2004, Gholizadeh
Doonechaly and Rahman 2012). A schematic representation of the contour map of the P1
unit of horizon for the Palm Valley reservoir is shown in Fig. 2-2 (Roe 1995).

Fig. 2-2: Top P1 (horizon of interest) unit depth contour map of the Palm Valley reservoir interpreted from seismic
data (Roe 1995, Tran 2004).

15
 As discussed in previous section, two sources of data, wellbore scale data and field scale
data, will be analysed to characterise fracture properties at corresponding scales of these
reservoirs. Post-processing of wellbore scale data includes core description, image logs and
caliper logs. Field scale data includes outcrops, seismic attributes and numerical simulation
of tectonic events. Among the fracture properties theoretical background of post-processing
of fracture density and fractal dimension will be given next.

Fractal Dimension

Fractal dimension is used to evaluate the natural discontinuities, such as natural

fractures (Mandelbrot 1982, Willis-Richards et al. 1996). Fractal dimension is a function of
material strength, tectonic history and lithology of the formation. When there is a lack of
accuracy in the field data, fractal concept may establish a strong framework to analyse the
fracture properties especially fracture size. Also fractal geometry is a very powerful tool to
analyse the discontinuous objects, such as fractures (Mandelbrot 1983, Sahimi 1993, Lei
and Chen 1995, Watanabe and Takahashi 1995, Sahimi 2000).

Fractal geometry concept demonstrates the relationship between the numbers of

fractures with radius bigger than a specific value (NRad) and the nominated radius with
proportionality constant (C) and an exponent (D) which is called fractal dimension as
(Mandelbrot 1983, Sahimi 1993, Lei and Chen 1995, Watanabe and Takahashi 1995,
Sahimi 2000):
C
N Rad 
Rad D (2-1)

Different methodologies are proposed in the literature to estimate the fractal dimension
(D). In this study the box counting method is used to calculate the fractal dimension (Walsh
and Watterson 1993). In the box counting method the fractured system is divided into the
square blocks with a specific side length. The nominated side length is decreased
repeatedly. In each step the numbers of blocks that are cut by the fractures are recorded as
N (Eq. (2-1)) and the side length of the divided blocks is assumed as the Rad (Eq. (2-1)).
For each step ‘N’ and ‘Rad’ values are plotted in a Log-Log style. By plotting the data with
a linear regression, C (proportionality constant) and D (Fractal dimension) can be obtained.
16
A schematic representation of the box counting method and the plotting procedure are
shown in Fig. 2-3 and Fig. 2-4 respectively. As shown in Fig. 2-3, the domain is divided
into a number of blocks successively, and the numbers of the blocks which are intersected
with fractures are counted. Also Fig. 2-4 shows that the slope of the line plotted based on
the number of blocks cut by the fractures with a specific radius is called fractal dimension.

l
Fig. 2-3: Box counting method for the determination of the fractal dimension value ().

Slope=D

Fig. 2-4: Plotting the data obtained from box counting method to determine the fractal dimension (Tran
2004)

17
 Fracture Density

Fracture density is used as a very important parameter (representing the probability of

presence of fracture) in discrete fracture network modelling. In order to calculate fracture
density, the reservoir is divided into a number of blocks. In each block the area of all the
fractures intersecting that block are added together and then divided by the volume of the
corresponding block as:
N

 Area
FractureDensity  i 1 (2-2)
TotalVolume

where N is the total number of fractures that intersect with a representative elementary
volume.

2.3.1 Core Description

From Palm Valley reservoir 6 core descriptions from different depths from four wells
were received. An example photographs of core lay out from 1750m to 2000m depth cut
from the Palm Valley reservoir is shown in Fig. 2-5 (Berry 1991).

These photographs were analysed to determine fracture density, fractal dimension and
fracture orientation and the values are presented in Table 2-1. From the core descriptions it
was found that some of the fractures did not fully cut through the core samples (Tran
2004). Also at the depth of 5891ft two parallel fractures are detected with a dip value of 70°
and large aperture of 0.1mm (Tran 2004).

Table 2-1: Characteristic properties of different sets of fractures for Palm Valley Reservoir.
Data Set Av. Dip Av. Azimuth Fracture Density Fractal Dimension
Set #1 79.34 29.57 0.212 1.4059
Set #2 90.35 57.32 0.32 1.8993
Set #3 25.5 259.7 0.45 1.8447

18
 Fig. 2-5: Example Core Samples at Palm Valley Reservoir from 1750m to 2000m Depth (Berry 1991,
Tran 2004)

2.3.2 Image Logs

Wellbore images were analysed from two wellbores of Palm Valley reservoir and
specialised software, such as Interactive Petrophysics (Senergy) is used to analyse the
images. Based on the analysis two sets of fractures have been detected from images and the
characteristic properties of these two fracture sets are listed in Table 2-2. Also an example
of the image log taken from the Palm Valley reservoir from the depth of 2326m to 2333m

19
is shown in Fig. 2-6. As shown in the example 11 fractures intersected the wellbore and
there exist three bedding layers along the imaged section (Tran 2004).

Fig. 2-6: Wellbore Image taken from Well E, Palm Valley reservoir at the depth interval of 2326m to 2333m (Rayner
1995, Tran 2004)

Table 2-2: Characteristic properties of the fractures sets obtained from the image logs (Tran 2004)
Data Set Av. Dip Av. Azimuth Fracture Density Fractal Dimension

Set #1 78 185 0.256 1.131

Set #2 22 3 0.0525 1.288

2.3.3 Caliper Log

A four-arm caliper log data from five wells of Palm Valley reservoir was received.
Three sets of fractures were detected based on the log data analysis. This fracture data was
validated by using other conventional logs, such as Gamma Ray, Spontaneous potential and
Resistivity logs. The system of fractures obtained from the caliper log and their orientation is
listed in Table 2-3. An example of the caliper log for the depth interval of 1789m to 1801m
is shown in Fig. 2-7.

20
 Table 2-3: Different fracture sets obtained from the caliper log
Data Set Av. Dip Av. Azimuth
Set #1 45 137
Set #2 85 118
Set #3 25 275

Fig. 2-7: Caliper log from well A from Palm Valley reservoir at the depth interval of 1789m-1801m (Tran 2004)

2.3.4 Seismic

As discussed in section 2.2.6 seismic attributes are interpreted by using commercial

software. Fracture properties in the form of coordinates of the centre of each fracture along
with their dip and azimuth were received. A typical seismic interpretation is shown in Fig.
2-8. Average dip, average azimuth, fracture density and fractal dimension were estimated
and presented in Table 2-4.

21
 Fig. 2-8: Seismic interpretation and the detected fractures obtained from the Palm Valley reservoir (Roe, 1995,
(Tran 2004))

Table 2-4: Fracture sets including their characteristic properties obtained from the seismic interpretation(Tran
2004)
Data Set Av. Dip Av. Azimuth Fracture Density Fractal Dimension

Set #1 85 169 0.157 1.623

Set #2 15 294 0.229 1.012
Set #3 69 12 0.14 1.36

2.3.5 Outcrops

Outcrop analysis from the Palm Valley reservoir was analysed (Tran 2004) and one major
fracture pattern was detected from the analysis. This fracture set with its own characteristic
properties is listed in Table 2-5.

Table 2-5: fracture set with its own characteristic properties from the outcrop from Palm Valley Reservoir (Tran 2004)
Data Set Av. Dip Av. Azimuth Fracture Density Fractal Dimension
Set #1 85 4 0.085 1.56

2.3.6 Tectonic Event Simulation

In this study a tectonic simulation of the folding and unfolding of the Palm Valley
structure is carried out to study the effect of tectonic events on the probability of generating
fractures. In this section a theoretical background of the numerical simulation of tectonic

22
event is described and then the Palm Valley is used as an example to demonstrate tectonic
event simulation procedure.

In tectonic simulation of folding of a layer is extended (reverse model) and then

compressed (forward model) to reconstruct the geological folded structure (Kaus and
Podladchikov 2001, Kocher and Mancktelow 2005). An efficient numerical algorithm
using Finite Element Method (FEM) is used to model the reverse and forward folding for
sufficient number of time steps. For this purpose mixed penalty finite element formulation
is used to model quasi-steady state incompressible viscous flow based on the Navier Stokes
flow equation (Cuvelier et al. 1986, Haupt 2000, Gholizadeh Doonechaly and Rahman
2012). Mixed penalty formulations are based on decomposing the stress tensor into its
deviatoric and hydrostatic components as (Yükselen and Erim 1996):

 kk
 ij  sij   ij
3 (2-3)

where sij is the deviatoric stress,  ij is Kronecker delta,  ij is Cauchy stress tensor and  kk

is the trace of the Cauchy stress (Lee 1993, Xing 2009, De Borst et al. 2012).

A constrained equation which correlates the divergence of displacement to the pressure is

used to make a link between the stress and strain rate tensor as:
p
T .u  0 (2-4)
k
Where k is the bulk modulus of the rock. Also the equilibrium of the momentum can be
described as:

  b  0 (2-5)

where b is the body force term. Combining Eq. (2-3) and Eq. (2-5) we have:

  p  b  0 (2-6)

23
 Equations (2-4) and (2-6) are discretised using relevant shape functions. It is also
important to note that only certain combinations of pressure and velocity shape function
interpolations can be used to reach the convergence conditions (Pastor et al. 1997). In this
study the Q2-P1 element with 27 velocity nodes and 4 discontinuous degrees of freedom
for pressure are used for the discretization purpose. Pressure nodes are on the corners of a
tetrahedral element which is located inside the brick type velocity element. This is the
simplest 3-D element which is second-order accurate and satisfies the Babuska-Brezzi
stability condition to avoid the pressure solution to be locked in the unwanted loop (Fortin
1981). For numerical modelling purpose, the folded layer is embedded between two heavier
layers with lower viscosity to simulate natural conditions. To eliminate the effect of
surrounding layers on the folding process, the thicknesses of the layers above and below
the reservoir are calculated as follow (Johnson 1970):
LD
H 2 (2-7)


Where, LD is the wavelength and H is the thickness of the matrix layer.

In the reverse model, the extension is continued until the folded layer reaches the initial
amplitude as (Biot 1961):
A0
 0.1
h (2-8)

Where, A0 is the initial amplitude and h is the thickness of the layer.

Predetermined velocities (6.31×10-10 m/s) are applied on the sides of the model along
the y direction as boundary condition. The time increment (1000 years in this study) in each
step is chosen such that the boundary strain rate remains constant (0.02 percent per
thousand years) during the tectonic simulation to satisfy the plate convergence rate of 2
cm/yr which is calculated as (Moore and Karig 1976):

1  e  0.02
t  Li (2-9)
2u

24
where Li is the wavelength of the layer and u is the boundary velocity. After discretization
one can obtain the following system of equations:

K Q 
 u   f 
Q T V       (2-10)
 p g
      

In which we have:

K   BT DBd  (2-11)

Q   BT mN p d  (2-12)

V   N Tp N p d  (2-13)

f   NuT  gd    NuT td  (2-14)

V
g Pn (2-15)


2 0 0 0 0 0
0 2 0 0 0 0 

0 0 2 0 0 0
D  (2-16)
0 0 0 1 0 0
0 0 0 0 1 0
 
0 0 0 0 0 1

Where, B is the discrete operator relating deviatoric strains to nodal displacements

(Zienkiewicz 2000), Nu and Np are velocity and pressure shape functions respectively, is
the viscosity of the material and γ is the penalty parameter that is used as an estimation of
the bulk modulus of the rock. Proper selection of the penalty parameter results in a
divergence of 10-16 for the velocity unknowns in the iterative solution procedure. Normally
the penalty number is chosen according to (Bathe 1996, Hughes 2000, Zienkiewicz 2000)
as:

  (107  108 ) (2-17)

25
 Discontinuous pressure shape functions allow the separation of the pressure at the
element level and the solution of the linear system of Eq. (2-10) is found in the following
consecutive steps:
V
g Pn (Eq. (2-15)

V
( K  QT Q)u  f  Q V 1 g (2-18)

Pn1  ( QTV 1u)  ( V 1g ) (2-19)
The linear system of equations (Eq. (2-10)) is solved using the Uzawa (Helmig et al.
2006) type iterative algorithm. First an initial guess is used as the pressure value for all
relevant pressure nodes. In the next step the “g” matrix is calculated as shown in Eq.
(2-15). Following that, Eq. (2-18) is solved for velocity unknowns. Finally, new pressure
values are calculated for each pressure nodes based on Eq. (2-19). Iteration is repeated until
the divergence of the velocity is reduced to a desired value (10-16 m/sec in this study).

After the convergence of the velocity solution is reached, the power law model is used
to adjust the viscosity value for each node. For this purpose the Oswald - de Waele model
as shown in Eq. (2-20) is used (Zienkiewicz 2000).
 ( m 1)
  0  (2-20)

Where, µ0 is the initial viscosity value and ̇ is the second invariant of the strain rate tensor
and ‘m’ is the flow behaviour index. For most of the non-Newtonian fluids, ‘m’ is smaller
than 1 and usually it varies from 0.125 to 0.25 for geological studies (Reddy et al. 1983,
Tsenn and Carter 1987). In this study m is taken as to 0.25. Then Uzawa iteration is
repeated with the new values of viscosities until the divergence of the viscosity reaches
1010. The iteration procedure used to find the final solution of the system of linear
equations is shown in Fig. 2-9. Finally the resultant velocity solution is used to move each
node to its new location by multiplying velocity by the time increment in each step. In each
time step during unfolding and folding, strain rate tensor is calculated as:

  Bu (2-21)

Where, ε is the strain rate tensor for each node and u is the nodal velocity.
26
 Generate an Initial Guess for
Pressure and Viscosity values for all
Nodes

Generating the stiffness matrix

(Eq.(2-10)) based on Viscosity Values

Calculate the Right Hand Side

Matrix of Eq. (2-10)

Solve Eq. (2-18) for Velocity Values

Solve Eq. (2-19) for

Divergence of Velocity
Pressure Values No
Solutions < 10-16

Yes

Correcting the Viscosity Values for all

nodes based on Oswald de Waele model

Divergence of Viscosity
No
Solutions < 1010

Yes

Final Velocity and Pressure Solution

Fig. 2-9: Flowchart for the calculation of velocity and pressure at each time step during the tectonic Simulation
(Gholizadeh Doonechaly and Rahman 2012)

27
Then stress tensor can be obtained as:

 xx  1   4 2 2 0 0 0   xx 
  1   2 4 2 0 0 0   yy 
 yy    
 zz  1  1  2 2 4 0 0 0    zz 
   p      (2-22)
 xy  0  3  0 0 0 3 0 0   xy 
 yz  0  0 0 0 0 3 0   yz 
       
 xz  0  0 0 0 0 0 3  xz 

where,  is the deviatoric stress and p is the pressure which is calculated as:

 xx   yy   zz
p (2-23)
3

Stress tensors are updated after each time step until the original folded layer is
reconstructed. Then stress values at each node are used in the next step as one of the most
important input parameters in stochastic simulation. There are also analytical solutions for
stress values in 3-D single layer folds (Fletcher 1991), however, such analytical solutions
are for small amplitude/limb folds (Schmalholz 2008).

In the forward model, in each time step, the resultant stress tensor value in each node is
checked with the Mohr-Coulomb fracture criteria to determine whether fracture occurs or
not. If the rock passed the failure criterion, the rate of formation of fractures is calculated
based on the generalization of the equilibrium thermodynamic as (Wickham et al. 1982)
(see Appendix A for the complete derivation):

dA 2 2 8K
f  ( I1  3I 2 )  I (2-24)
dt 3  1
Where, I1 and I2 are the first and second stress invariants, γf is the surface tension energy
and K is the material constant.

dA
In the above equation, if the is less than zero, no fracture exists and greater than zero,
dt
the probability of fracture exists.
28
 The developed tectonic event simulation model is used to estimate the probability of
occurrence of fractures for the Palm Valley gas reservoir in Australia. For the numerical
simulation of the tectonic event, the folded layer is extended and compressed (reverse and
forward model respectively as described in tectonic simulation part (section 2.3.5)) to
reconstruct the original geological structure. The finite element mesh used for this purpose
contains 960 elements in the horizontal plane and 15 elements in the vertical direction thus
summing up to 14,400 total elements.

According to the Eq. (2-7) the thickness of the matrix layer is 650m both above and
below the reservoir layer. The reservoir layer contains fine elements to increase the
accuracy in this region. Effect of the gravity force is also significant for wavelengths bigger
than 100m (as considered in this study). The specific gravity of 2.45 for the reservoir
(sandstone) and 2.75 for the layers (shale) above and below the reservoir are used in this
study. Also the initial viscosity of the reservoir layer is 6.75×1018 (Pa s), which represents
the estimated viscosity for sandstone (Griggs 1939, Birch et al. 1942, Borg and Handin
1966, Gholizadeh Doonechaly and Rahman 2012).

Boundary velocity is applied in such a way that the model's wavelength changes with a
strain rate of 0.02% per thousand years (6.34×10−13 s−1) (Moore and Karig 1976). Results
for numerical simulation of unfolding and folding of the Palm Valley reservoir are shown
in Fig. 2-10. After reconstructing the structure by using the tectonic event simulation model
Eq. (2-24) is used to generate the continuum map of the probability of fracture generation.
The obtained probability values are first scaled to the interval of 0-1 (0 for no fracture
probability and 1 for highest probability of occurrence of fracture).

A schematic representation of the continuum map of the probability of occurrence of

fractures for the Palm Valley reservoir is shown in Fig. 2-11.

29
 (a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)

(i)

Fig. 2-10: Unfolding, from (a) to (e) with time difference of 50,000 years and folding, from (e) to (i) with the time difference of 50,000 years. A
speed of 2 cm/yr is applied in X direction from both sides in the numerical simulation. The finite element mesh contains 960 elements in the
horizontal plane and 15 elements in the vertical direction thus summing up to 14,400 total elements (Gholizadeh Doonechaly and Rahman 2012).

30
Fig. 2-11: A schematic representation of the continuum map of the probability of occurrence of the fracture in the
Palm Valley reservoir for the top and bottom layer of the formation with a thickness of 200m.

2.4 Field-Wide Generation of Discrete Fracture Map

In the previous section both wellbore scale and field scale data was analysed and
corresponding fracture properties were generated. In this section these fracture properties
will be combined to generate the field-wide discrete fracture map by using neuro-stochastic
simulation. The process of combining individual sources of fracture properties into field
wide generation of discrete fracture map is described in Fig. 2-12.

2.4.1 Artificial Neural Networks

As shown in Fig. 2-12 continuum fracture density and fractal dimension map is required
as a priori in the sequential Gaussian stochastic simulation. However, the hard data to
calculate the fracture density and fractal dimension is not available in all parts of the
reservoir domain. In order to overcome such a limitation, neural network is used in to
establish a correlation between different data sources at several scales: wellbore scale (hard
data available) and field scale. At wellbore scales two wellbore images, four core

31
descriptions and five 4-arm caliper logs were available in a volume of 10km x 3km x
200m.

Generation of continuum map of

probability of occurrence of fractures
by using tectonic event simulation
model

Wellbore Data:
Conventional Well logs,
Generation of fracture density and Core descriptions
fractal dimension based on seismic Wellbore Images
interpretation

Neural Network Fractal

Dimension
Fracture Density

Continuum map of fracture density

and fractal dimension

Fracture
Sequential Gaussian Stochastic orientation
Simulation Fracture length
Fracture class
properties

Subsurface map of discrete

fracture network

Fig. 2-12: A schematic representation of the flow chart for the neuro-stochastic simulation of the field-wide
discrete fracture map (Gholizadeh Doonechaly and Rahman 2012)

32
These hard data used to train conventional logs (SP, GR, NP, DRHO, PEF, RT, MSFL,
LLS, LLD and RHOB) available in 10 wells by using artificial neural network and generate
continuum map of fracture density and fractal dimension at these well sections. These well
locations are shown in Fig. 2-13. In the next stage, these continuum maps of fracture
density and fractal dimension from 10 wells were used to train seismic derived fracture
density and fractal dimension as well as probability of occurrence of fractures derived from
simulation of tectonic events. At this stage 3-D continuum map of fracture density and
fractal dimension are generated. In this study Back Propagation Neural Networks (BPNN)
with Multiple Perceptrons is used to establish a correlation between the input and output
parameters (Gholizadeh Doonechaly and Rahman 2012). Back Propagation Neural
Network as well as its formulation details is described in Appendix B.

Fig. 2-13: Location map of the drilled wells in the Palm Valley Reservoir (Berry et al. 1996, Tran 2004). The Palm
Valley area is shaded in the figure.

As mentioned above, two stages of training (wellbore scale and field scale) are required
to generate the continuum map of fracture density and fractal dimension. These two stages
are discussed as follow.

Wellbore Scale
33
 In order to generate the continuum map of fracture density and fractal dimension at
wellbore scale, training data for the BPNN are collected from wellbores. These data include
conventional logs (SP, GR, NPHI, DRHO, PEF, RT, MSFL, LLS, LLD and RHOB),
fracture density and fractal dimension (obtained from images logs and core descriptions
(Tran 2004)). Three layer BPNN is used to establish a correlation between the input and
output data. The designed BPNN includes 10 input nodes (well log data), 12 hidden layer
nodes and 2 output nodes (fracture density and fractal dimension). Then the BPNN
establishes a complex correlation between the input and output nodes. The obtained
correlation is then used to estimate fracture density and fractal dimension at wellbore
locations where hard data is not available. A schematic representation of the continuum
map of fracture density and fractal dimension for the Well E Palm Valley reservoir is
shown in Fig. 2-14.

Field Scale

Field scale continuum map of fracture density and fractal dimension is generated with
the same procedure as the wellbore scale. In the field scale training, however, the input
nodes consist of probability of occurrence of fractures (obtained based on the tectonic event
simulation model) and seismic attributes (time, instantaneous frequency and instantaneous
phase) and the output nodes include fracture density and fractal dimension values obtained
from seismic data analysis. Fracture density and fractal dimension values are obtained from
wellbore scale training (previous section) (Tran 2004). Training patterns of Artificial
Neural Network (ANN) for fracture density and fractal dimension is shown in Fig. 2-15
and Fig. 2-16 respectively. From the figures it can be seen that there exists an optimum
training above which the error tends to increase. The optimum training for fracture density
and fractal dimension are 48,720 and 62,505 epochs respectively. The correlation that was
obtained through BPNN training is used to generate the 3-D continuum map of the fracture
density and fractal dimension. A continuum map of fractal dimension for the P1 unit
horizon for Palm Valley reservoir is shown in Fig. 2-17. Also the 3-D pixel based map of
the fracture density for the Palm Valley reservoir is represented in Fig. 2-18.

34
Fig. 2-14: Fracture density and fractal dimension profile obtained by training the BPNN.

35
 Number of Epochs
Fig. 2-15: Neural network training (black) and validation (red) map for fracture density. Optimal training reached
after 48,720 epochs (iteration steps) after which the error (difference between the simulated data and the actual
field data) tends to increase because of overtraining. At the optimal training point, all the obtained weight factors
are recorded and used to find the fracture density for other sections of the reservoir (Gholizadeh Doonechaly and
Rahman 2012).

Number of Epochs
Fig. 2-16: Neural network training (black) and validation (red) map for fractal dimension. Optimal training
reached after 62,505 epochs (iteration steps) at which the weight factors are recorded and used to calculate fractal
dimension value for other parts of the reservoir (Gholizadeh Doonechaly and Rahman 2012).

36
Fig. 2-17: A cross section of the fractal dimension in P1 unit of horizon in the Palm Valley reservoir (Gholizadeh
Doonechaly and Rahman 2012).

(a) (b) (c)

(d) (e) (f)

Fig. 2-18: 3-D pixel based map of the fracture density at the depth of (a) 1000 m, (b) 1200, (c) 1400 m, (d) 1600 m,
(e) 1800 m, (f) 2000m. The reservoir is divided into sufficient number of blocks and fracture density value is
calculated for each block which is used for stochastic simulation of discrete fracture network (Gholizadeh
Doonechaly and Rahman 2012).

37
2.4.2 Stochastic Simulation

After determining the field wide distribution of the fracture density and fractal
dimension, the stochastic simulation is used to estimate different fracture attributes, such as
dip, azimuth, centre point and radius in a probabilistic framework. Fractures are considered
as distinct objects with their own attributes, such as dip, azimuth, centre point and radius.
The fracture properties and probability of distribution of each fracture attribute are
determined based on the available field data. After determining the statistical properties of
different fracture attributes, the reservoir is divided into a specific number of grid blocks.
Each block is then populated with fractures by using sequential Gaussian stochastic
simulation in which each fracture attribute is generated based on the random realization of
its corresponding probability density function. Fractures are generated consecutively (one
by one). After generating individual discrete fracture of a block fracture density and fractal
dimension are calculated. If the fracture density and fractal dimension correspond to the
previous set values for this block the next step is to update the fracture density and fractal
dimension map. If the value of fracture density or fractal dimension is exceeded or fall
short of the set values then fractures are removed or added. The random realisation of the
fractures continues until the total fracture density and fractal dimension of the
corresponding block (obtained from the 3-D continuum map of the fracture density and
fractal dimension) is met. Then the algorithm moves to the next block and all blocks of the
reservoir are filled with fractures successively. A schematic representation of the algorithm
for generating the subsurface fracture map by using sequential Gaussian stochastic
simulation is shown in Fig. 2-19. The fractures are represented by circular disks, each of
which is known by its unique centre point, radius, dip and azimuth. In order to randomly
realise such fracture attributes by sequential Gaussian simulation the statistical analysis of
the fracture orientation is needed to be performed. For this purpose they are initially
converted to circular data with range from zero to 360° (Fisher et al. 1993). Natural
fractures tend to distribute in more than one dominant direction because of the complex
stress distribution of the reservoir layer due the tectonic events (Fisher et al. 1993). In this
study the bi- and uni-modal wrapped normal distributions are used to characterize the
fracture azimuth and dip respectively.

38
 Field Data
 Wellbore Data
o Well logs
o Borehole Image
o Core Samples
o Dipmeter
 Field Scale Data
o Probability of occurrence of Fracture
o Seismic

Statistical Analysis and Fracture Density and FNN

Spatial Distribution Fractal Dimension
from Wellbore Data

Sequential Gaussian Continuum Map of

Model Boundary Stochastic fracture Density and
Conditions Simulation Fractal Dimension

Random Sampling of
Fracture Attributes:
Dip, Azimuth, Radius
Record the Generated
and Centre Point
Fracture and Update
Continuum Fracture
Density Map
Fracture Satisfies
No
Boundary

Yes

Total Fracture Density of

No
the Reservoir Reached

Yes

Discrete Fracture Network

Fig. 2-19: Flowchart for Generating Discrete Fracture Network by using Sequential Gaussian Simulation Technique
(Gholizadeh Doonechaly and Rahman 2012)

39
 Wrapped normal distribution is achieved by wrapping the normal distribution of a
variable around the unit radius. Multi-modal distribution is a linear combination of uni-
modal ones. The resultant distribution WN(θ) is a linear combination of two wrapped
normal distributions, WN(μ1,σ1) and WN(μ2,σ2) as (Fisher 1995, Tran 2007, Xu and
Dowd 2010):

WN ( )  n WN (1 , 1 )  (1  n) WN (2 ,  2 ) (2-25)

Where, a bimodal wrapped normal distribution contains two parts with proportion factors n
and (1−n), corresponding to two sets of means, resultant lengths, and standard deviations:
μ, R and σ respectively as:

1 
(1  2 Ri j cos[ j ( j  i )])
2
WN ( i ,  i )  (2-26)
2 j 1

where 0≤θ≤2π, 0≤R≤1 and i=1, 2. WN (θ) produces a sample Gaussian field for the random
realization purpose. Random realization is also needed to be used for the radius of the
fracture. Fractal geometry concept is used to stochastically simulate the fracture radius. For
this purpose a minimum and maximum radii for the fracture generation are found based on
the obtained field data. Based on these two values and a modification of Eq. (2-1) the
following equation is obtained (Tran 2004).

N RRmin
max D
 C ( Rmin D
 Rmax ) (2-27)

where ΔN is the number of fractures with the radii between the maximum and minimum
defined values. If a fraction of the total number of the fractures considered as p and Rp be
the maximum radius in that fraction, then Eq. (2-27) can be written as follows (Tran, 2004).

N RRpmin  pC ( Rmin
D D
 Rmax ) (2-28)

Also based on Eq. (2-27) it can be easily shown that:

40
 N RRpmin  C ( Rmin
D
 Rp D ) (2-29)

Based on Equations (2-27) and (2-28), the radius of the fracture is calculated as:
1

D
Rp  [(1  p)  Rmin D
 p  Rmax ] D (2-30)

Where, D is the fractal dimension. In a random realization, a random value is assigned to

the p parameter and then the resultant radius is used as one of the generated fracture
properties. Also fracture centre point, should be generated stochastically. For this purpose,
a spatial random distribution of the fracture centre points is made based on the given field
data. Then a random realization is carried out utilizing the Poisson process to generate a
value as the centre point of the fracture. The abovementioned procedure is used to generate
the discrete fracture map of Palm Valley reservoir. For this purpose the reservoir is divided
into 2000 blocks. In each block fractures are generated by random realization scheme as
discussed earlier.

Statistical analysis of the fracture dip and azimuth data results in bi-modal wrapped
normal distribution properties for fracture azimuth as shown in Fig. 2-20. Also the uni-
modal distribution of fracture dip values are identified and presented in Fig. 2-21. Other
circular statistical parameters are represented in Table 2-6.

Fig. 2-20: Circular statistical analysis: histogram and probability density function plot for fracture azimuth.
Statistical analysis on the circular azimuth data reveals a bimodal wrapped normal distribution, with modes at
approximately 6° and 172° (Tran 2004).
41
Fig. 2-21: Circular statistical analysis: histogram and probability density function plot for fracture dip showing
one-modal wrapped normal distribution, whose mode is at 176° (Tran 2004).

Table 2-6: Statistical Parameters for the Palm Valley Reservoir Obtained from the Field Data ((Do Rozario
1991, Tran et al. 2006))
Statistics Azimuth 1 Azimuth 2 Dip 1

Weight factor 0.25 0.75 1

Circular mean 6.16 3.11 3.14

Circular StdDev 0.37 0.91 0.71

Skewness -0.038 -0.038 0.000

Kurtosis -0.336 -0.336 4.025

Fracture azimuth is found to have two distinct trends according to the wrapped normal
distribution, namely 3° and 86°. The dip values, however, are more distributed around one
value (88°) and therefore uni-modal distribution is used to analyse them. After the
statistical analysis of the fracture orientation, the sequential Gaussian stochastic simulation
is used to generate the discrete fracture network map of Palm Valley reservoir. A schematic
representation of the discrete fracture network for Palm Valley reservoir is shown in Fig.
2-22.

42
 Fig. 2-22: Discrete fracture network map of the Palm Valley reservoir (Gholizadeh Doonechaly and Rahman
2012).

The stochastic simulation results cannot be directly compared with the target reservoir
fractures since it is impractical to obtain the actual fracture system of the whole reservoir. It
is however, noteworthy to mention that fracture density and fractal dimension used to
characterize fracture properties can be reproduced in a greater accuracy as described by
different authors (Clark et al. 1999, Babadagli 2001, Tran 2004). In this thesis additional
numerical experiment is conducted to test the reproducibility of these two important
properties, fracture density and fractal dimension. In this experiment fracture density and
fractal dimension are calculated at wellbore locations which are presented in their
normalized form in the x-axis of Fig. 2-23 as target data. These target data together with
other field data, such as seismic, outcrops etc., is then used to predict field wide distribution
of fracture density and fractal dimension. After this step the fracture densities and fractal
dimension at wellbore locations are removed and reproduced based on the field wide
distribution of the nominated data (Gholizadeh Doonechaly and Rahman 2012). These
predicted values are first normalized and then presented in the Y axis of Fig. 2-23.

43
Fig. 2-23: Validating results for fracture density and fractal dimension reproduced for wellbore locations. The
correlation values are close to 1 (0.8401 and 0.9386 for fracture density and fractal dimension respectively)
(Gholizadeh Doonechaly and Rahman 2012).

From the results of this analysis it can be seen that the majority of the data are
distributed along the “line of best fit (y=x)” with the correlation values of 0.8401 and
0.9386 for fracture density and fractal dimension respectively.

As part of this study two other numerical experiments were also carried out by using the
same procedure as mentioned above to generate the discrete fracture network map of
Soultz-Sous-Forets geothermal reservoir in France (Gholizadeh Doonechaly et al. 2012)
and Patchwarra formation in Australia (Gholizadeh Doonechaly et al. 2012). The filed data
which are used for these simulations as well as the generated discrete fracture network map
of those nominated reservoirs are presented as follow.

Soultz-Sous-Forets Geothermal Reservoir

Soultz-sous-Forets is located in the Upper Rhine Graben (Sausse 2002) and north of
Strasbourg (Jupe et al. 1995). Soultz reservoir is near the centre of large heat-flow
anomalies which is extended over an area of 150 km by 20 km across the French-German
border.

Different sources of field data are used to obtain the dip and azimuth of the fractures in
locations where hard data is available. As part of statistical analysis bi-modal wrapped

44
normal distribution properties for fracture azimuth is shown in Fig. 2-24 and the
distribution of the dip values is represented in Fig. 2-25. Fracture dip and azimuth is found
to have two distinct trends according to the wrapped normal distribution and because of
that bi-modal distribution is used. Also circular statistical parameters for the fracture data
obtained from different data sources are listed in Table 2-7.

(a) (b) (c)

Fig. 2-24 Rose diagrams of fracture azimuth for Soultz reservoir (a) massive un-fractured granite (N = 675 data),
(b) moderately fractured granite (N = 554 data) and (c) highly fractured granite (N = 1767 (Genter and Traineau
1996).

Fig. 2-25: Histogram of fracture-dip directions for the Soultz-Sous-Forets geothermal reservoir (Genter and
Traineau 1996)

45
 Table 2-7: Characteristic statistical properties of the fractures obtained from the different data
sources, such as core sample descriptions, well logs etc. (Gentier et al. 2010).

Fracture set F1 F2 F3 F4 F5
Distribution Law Normal Normal Normal Normal Uniform
Azimuth
Mean 2 162 42 129 0
Half-Width 16 19 6 6 180
Distribution Law Normal Normal Normal Normal Normal
Mean 70 70 74 68 70
Dip

Half-Width 7 7 3 3 9
Dip Direction NW NE NW SW -
Fracture No. 1.3E-7 3E-9 1.76E-8 3.3E-8 1E-8
Radius (m) 187 150 95 112 100
Transmissivity (m2\s) 6E-6 6E-6 4E-6 2E-6 5E-7

Then the discrete fracture network map of the Soultz geothermal reservoir is generated
based on the field data by using the developed hybrid tectono-stochastic model as shown in
Fig. 2-26.

Fig. 2-26: Discrete fracture network map of Soultz-Sous-Forets geothermal reservoir

46
 Patchwarra Formation

As another attempt the developed tectono-stochastic technique is used to generate the

subsurface discrete fracture network map of Patchwarra formation in Australia. Patchwarra
formation is located in Cooper Basin. The characteristic data for this reservoir is presented
in Table 2-8. Also the azimuth and dip distribution for the Patchwarra formation is shown
in Fig. 2-27 (a) and (b). The discrete fracture network map of the Patchwarra formation is
also presented in Fig. 2-28.

Table 2-8: Statistical parameters for Patchwarra formation (Mildren et al. 2005)
Statistics Azimuth 1 Azimuth 2 Dip 1

Circular mean 2.35 2.041 3.0702

Circular StdDev 1.52 1.02 0.472

Skewness 0.256 -1.526 0.325

Kurtosis -1.678 3.024 0.256

(a) (b)
Fig. 2-27: Statistical wrapped distribution of (a) Dip and (b) azimuth for Patchwarra formation. (Mildren et al.
2005)

47
Fig. 2-28: Discrete Fracture Network map of the Patchwarra formation

48
2.5 Closure

The model developed in this work presents the discrete fracture network simulation
using a combination of tectonic event simulation, artificial neural networks and sequential
Gaussian stochastic simulation. Although stochastic simulation minimizes the scale errors
the accuracy depends on the amount of data sources available from the field which are
often scarce. For example in the Palm Valley reservoir available hard data is only available
from 10 wells and limited 2-D seismic attributes for P1 unit of horizon. This problem was
overcome by integrating a 3-D spatial map of the probability of occurrence of fracture for
all parts of the reservoir which was obtained by reconstructing the reservoir structure
during unfolding and folding events. Such a model enables us to have a field-wide estimate
of the probability of occurrence of fractures in all parts of the reservoir. This hybrid
approach has allowed us to develop a comprehensive discrete fracture network map.

49
Chapter 3

3 FLUID FLOW SIMULATION

AND HEAT TRANSFER
In this chapter a numerical model is presented to simulate the fluid flow and heat
transfer in naturally fractured reservoirs. Finite Element Method (FEM) is used as a
powerful numerical technique for the simulation purpose. Fluid flow is simulated using a
hybrid technique which combines single continuum and discrete fracture approaches. Also
the local thermal non-equilibrium approach is used to simulate heat transfer by conduction
in the rock matrix and convection (including conduction) in the fluid.

Fluid flow and heat transfer models are integrated within a fully coupled poro-thermo-
elastic framework. Developed model is used to evaluate the long term response of the
geothermal reservoirs subject to specific boundary conditions and injection/production
plans.

FEM is the most popular numerical technique used for the simulation purposes
specifically in rock mechanics (Jing 2003). The initial concepts of the FEM and domain
discretization were first established in 1943 (Courant 1943) followed by an approach
introduced by Clough (1960) to use the FEM for the structural analysis. Then Argyris and
Kelsey (1960) proposed the duality of the force and displacement and estimation of virtual
work. Later Bathe (1996), Zienkiewicz (2000) and Stephansson (2007) presented numerous
applications of FEM technique to solve a wide range of engineering problems. When using
the FEM, first the whole domain is divided into finite number of elements which is known
as discretization. Based on the complexity of the problem and the stability issues of the
solution, different types of elements with different accuracies are used.

50
3.1 Simulation of Fluid Flow in Fractured Media

Three distinct approaches are used in the literature to simulate the fluid flow in naturally
fractured reservoirs namely: single continuum, dual continuum and discrete fracture
approach. In single continuum, the fractured block is represented by an equivalent
homogeneous system using a specific permeability tensor. In dual continuum approach the
whole domain is divided into two major sub-sections: fractures and matrix. In discrete
fracture approach, fractures are explicitly discretised in the domain. These approaches are
briefly discussed below followed by the proposed methodology which is used in this study.

3.1.1 Single Continuum Approach

In single continuum approach the reservoir is divided into a number of blocks and the
fracture properties in each block are averaged. An example of single continuum approach is
calculation of permeability tensor for individual grid blocks. Such a representation does not
require discretization of individual fractures over the entire region which reduces
computational effort. Use of permeability tensor in reservoir flow modelling is first
proposed by Lough (Lough et al. 1997). They divided the reservoir into small number of
blocks. Then they discretised fractures in each block by using Boundary Element Method
(BEM). After determining the velocity profile inside the block, the equivalent permeability
tensor of the corresponding block is calculated and is used for further reservoir scale flow
simulations. The approach proposed by Lough et al. (1997) can be applied for regular
patterns of simple fracture geometries. This approach was further improved by Teimoori et
al. (2005) to simulate fluid flow in arbitrary oriented complex fracture geometry.

Despite the improved computational efficiency of the single continuum approach, it can
not adequately address the flow behaviour of fractures because of the volume averaging
mechanism (Tarahhom et al. 2009).

3.1.2 Dual Continuum Approach

In this technique, the whole domain of the fractured porous media is divided into two
major parts: fractures and matrix. Dual continuum approach was first proposed by Warren
and Root (Warren and Root 1963) for fluid flow modelling in fractured petroleum
51
reservoirs. They simplified a fractured reservoir as a series of the sugar cubes as shown in
Fig. 3-1 and the fractures are evenly spaced (Carlson 2003). Based on the assumptions of
Warren and Root (1963) fractures provide the main flow conduit and the matrix acts as a
source/sink to the fractures. Also the fluid transfer between the fracture and the matrix is
defined based on the specific transfer functions (Beckner et al. 1987).

(a) (b)
Fig. 3-1: Warren and Root (1963) used the sugar cube model for representing the fractured rock in a dual
continuum approach. In their model, the real fracture rock (a) is represented by evenly spaced and infinitely
distributed fractures which provide the main flow conduit.

Application of numerical methods in dual continuum approaches was first proposed by

Kazemi (Kazemi et al. 1992) to simulate the fluid flow in large scales. Following the work
done by Kazemi et al. 1992 a significant number of studies have been carried out in
analytical and numerical frameworks to analyse different aspects of the dual continuum
models (Choi et al. 1997, Pride and Berryman 2003, Gong et al. 2008). In dual continuum
approach all the existing fractures are interconnected and the rock segments (sugar cubes in
Warren’s (1963) model) are surrounded with fracture surfaces (Carlson 2003). Therefore
the isolated fractures can not be part of the model. Also the matrix-fracture transfer
functions are highly affected by the size and shape of the matrix blocks (Torsæter and
Silseth 1985). In order to solve such a limitation a sensitivity analysis is needed to be
carried out based on the different element shapes to select the proper matrix block shape
(Beer and Watson 1992).

52
3.1.3 Discrete Fracture Network (DFN) Approach

DFN approaches have been proposed to consider the fluid flow and transport inside the
fractures explicitly. There are two main reasons to use DFN in fluid flow modelling. First is
to incorporate the real geometry of the fractures and secondly, the transmissivity of
individual fractures and its effect on the fluid flow.

DFN approach considers the existence of fractures explicitly. So detailed information of

the fracture properties, such as orientation, size and location are needed for the flow
simulation purpose. Since the volume averaging techniques, such as single continuum
approach are not accurate enough for the flow simulations in moderate scale fractures, the
DFN approach can be used as a powerful approach for the simulation purposes. As an
example of the application of the DFN technique, the GeoSys FEM model was developed
to simulate the poro-thermo-elastic effects in discrete fractures with rough walls embedded
in the rock matrix by using supercomputing facilities. Main disadvantage of the DFN
model, however, is the need for an extensive computation.

3.1.4 Hybrid of Single Continuum and Discrete Fracture

As mentioned above, different approaches have been used in the literature to incorporate
the fractures into the flow modelling. Each technique has its own drawbacks and benefits.
In this study a hybrid methodology combining the single continuum and discrete fracture
networks model is used to increase the efficiency of the fluid flow simulation. In the
proposed methodology a threshold value is defined for the fracture length. Those fractures
which are smaller than the threshold value are used to generate the grid based permeability
tensor using an appropriate numerical technique (Fahad et al. 2011, Fahad 2013), which is
discussed in the following section, and the fluid flow in simulation is carried out by using
the single continuum approach in the nominated blocks. Those fractures which are longer
than the threshold value are explicitly discretised in the domain using appropriate elements
and the fluid flow in them is modelled using the discrete fracture approach. Such an
approach provides a more accurate and realistic framework to consider the effect of long
fractures on the fluid flow in fractured medium.

53
3.2 Domain Discretization Using the Hybrid Methodology

In this study the medium and long fractures (l  50m) are discretised using triangular
elements and the contribution of flow by fractures (l < 50m) are taken into account by
calculating permeability tensor for each discretised element. A schematic representation of
the domain discretization for a fractured reservoir domain is shown in Fig. 3-2 (a) and (b).

(a) (b)

Fig. 3-2: Domain discretization by using the hybrid of the single continuum and discrete fracture approach. (a)
Fractures longer than 50 m are explicitly in the reservoir domain by using the triangular elements. (b) After the
discretization of the long fractures, the effect of the short fractures (<50m) are taking into account by calculation
of the permeability tensor of the corresponding blocks which are cut by the fractures (Doonechaly et al. 2013).

Permeability tensor for each block is expressed as:

 k xx k xy 
K 
 k yx k yy  (3-1)

Permeability tensors are calculated by simulating fluid flow in individual fractures in

each element. The concept of permeability tensor was first introduced by Snow (1969) by
considering a set of parallel fractures in a Representative Elementary Volume (REV) with
zero matrix permeability. Long et al. (1982) developed a methodology for calculation of
permeability tensor for arbitrary oriented fractures using superposition technique. In this
54
study the authors have considered interconnected fractures with fracture surface as infinite
plate without roughness. Baumgartner and Moore et al. (1995) estimated permeability
tensor by assuming fractures as a planar sink/source term. Rasmussen and Yeh et al. (1987)
extended the approach and studied the effect of vertical fracture/matrix permeability ratio
on the permeability tensor. In a separate study, Lough et al. (1998) used a numerical
technique, Boundary Element Method (BEM), to calculate the permeability tensor of the
REV containing medium sized fractures considering fractures as a sink/source term.
Following this work Chen et al. (1999) presented an analytical model to calculate the
permeability tensor of the blocks containing infinite parallel fracture sets. Lee et al. (2001)
improved the efficiency of their previous approach by considering the effect of the short
fractures using the analytical method proposed by Oda (1985). Teimoori et al. (2005)
presented the first comprehensive methodology to calculate the permeability tensor for
arbitrary oriented fractures in different length scales. In this study permeability tensor was
determined by discretising the solution domain into different subdomains depending on the
length of the fractures using BEM. Short fractures are considered as matrix
inhomogeneities to improve the permeability of the matrix. However medium and long
fractures are discretised explicitly in the domain and fluid flow is simulated using BEM.
Fahad and Rahman et al. (2011) extended Teimoori’s (2005) work by increasing the
efficiency of the BEM to be used for greater number of fractures. The authors presented for
the first time effective permeability tensor calculation for the fractured REV by using the
BEM. The effective permeability model was validated using laboratory derived data.

3.2.1 REV Discretization for Permeability Tensor Calculation

To calculate the effective permeability tensor, the fractured REV is divided into three
distinct regions which are matrix (region 1), fracture (region 2) and region around the
fractures (region 3) as shown in Fig. 3-3.

55
 Fig. 3-3: Domain discretization based on different fracture lengths (Doonechaly et al. 2013).

Flow inside the fractures (region 2) is modelled using the cubic law. With the
assumption of smooth fracture surfaces, cubic law can accurately simulate the flow inside
the fractures (Long et al. 1982, Rasmussen and Civan 2003). Cubic law describes the flow
through fractures as (Long et al. 1982):

b3
Q P
12 (3-2)

where b is the fracture aperture, µ is the dynamic fluid viscosity, P is the pressure and Q is
the flow rate. Also boundary conditions for the fractures are as follow:

V  V   Q (3-3)

pi  pav i=1,…,N
(3-4)

where V+ and V- represent the normal velocities at the opposite fracture surfaces and Q is
the sink/source term. Also Eq. (3-4) implies that the fluid pressure at all the fracture
boundary nodes is assumed to be equal to the average fracture pressure. In matrix regions
close to the fractures (region 3), the Darcy equation (Eq. (3-5)) is coupled with the mass
conservation equation to consider the effect of the fracture on the flow of fluid in the region
close to the fractures. Size of this region depends on the size of the fracture.

56
 V   K P (3-5)

K 2 P  Q  0 (3-6)

where “Q” represents the sink/source term between matrix (fracture) and fracture (matrix).
Also fluid flow simulation in the matrix (region 1) is described as follow:

V f   K f Pf
(3-7)

 p
(k f )  Q f  q ff  0
L L (3-8)

where kf is the fracture permeability, Q source/sink term and q is the fluid flow contribution
by the fracture intersection. Also the boundary conditions around the medium size fractures
are defined as:

p fi  pav (3-9)

vi  vi  Q (3-10)

where pfi is the pressure at the ith node along the fracture, pav is the average pressure along
the fracture surface, vi+ and vi- are the fluid velocity at the opposite nodes along the
fractures and Q is the net flow interaction between the matrix and the fracture. Also for the
short fractures which are considered as part of the matrix porosity, the Laplace equation is
solved using the following boundary conditions:

pmi  p fi (3-11)

vmi  v fi (3-12)

where pmi is the matrix pressure and pfi is the fracture pressure at the matrix/fracture
interface a v is the normal fluid velocity at the ith fracture node along the fracture surface.
Since the pressure on the matrix fracture interface is unknown, periodic boundary condition
is applied in an iterative scheme to calculate the pressure values.

57
3.2.2 BEM Formulation

In the following subsections, the formulation of the BEM is presented for the calculation
of the permeability tensor for the reservoir elements which are cut by the fractures (< 50m).

Short Fractures

The boundary integral equation for an arbitrary point in the sub-domain containing the
small fractures is written as follow:

c( ) p( )   G( x,  )v( x)d    F ( x,  ) p( x)d    Q( x)G( x,  )dA( x)

   (3-13)

where G is the fundamental solution, v is the normal velocity,  is the boundary of the
domain,  is the entire domain, F is the flux corresponding to the fundamental solution
and Q is the Dirac delta function (Beer and Watson 1992). Since the short fractures are
considered as part of the matrix porosity, therefore the source/sink term is neglected and
Eq. (3-13) can be written as:

c( ) p( )   G( x,  )v( x)d    F ( x,  ) p( x)d 

  (3-14)

For short fracture discretization, the fracture and domain boundaries are discretised by
using constant type BEM. The discretised form of Eq. (3-14) is written as below:
N N
ci ( ) p( )    F ( x,  ) p( x)d     G( x,  )v( x)d 
j 1  j 1 
(3-15)

where F is a function to relate the ith node to the jth element and N is the total number of
elements. For medium and long fractures, Eq. (3-13) can be written in the following form:
N N M
ci ( ) p( )    F ( x,  ) p( x)d     G( x,  )v( x)d     Q( x)G( x,  )d 
j 1  j 1  i 1 
(3-16)

Medium Fractures

Eq. (3-13) for medium and long fractures can be written as:

58
 N N M
c( ) p( )    F ( x,  ) p( x)d     G( x,  )v( x)d     Q( x)G( x,  )d 
j 1  j 1  i 1 
(3-17)

Where, F and G are arbitrary functions which relate a specific node in the domain to the
corresponding element. Eq. (3-16) can also be written in compact form as follow:

[ A]  p  [ B]  v  b
(3-18)

where A and B are the integrals of F and G functions respectively, b represents the second
term on the right hand side of Eq. (3-18) which is the sink/source term from matrix to
fracture.

Fracture Intersection

In arbitrary oriented discrete fracture networks, a certain number of fractures may

intersect at a specific point (2D representation). The resulting connecting points may have
properties significantly different from their adjacent locations.

Integrating the continuity equation over the intersecting region, Ap (see Fig. 3-4),
caused by intersection of different fractures results in:

 vd   0
 (3-19)

Where, v is the fluid velocity in the direction of the outward normal perpendicular to the
boundary of the intersection region.

Fig. 3-4: intersecting region creates by two fractures. Ap is the intersection region created by intersecting fractures.

Assuming that the major flow contribution in the intersecting region is caused by the
59
fractures and a constant pressure in the intersection region, Eq. (3-19) can be rewritten in
discretised form as follow:

w
i 1
f ,i  v f , n ,i  0
(3-20)

Where, N is the total number of fractures arriving the intersection point, w is the aperture of
the ith fracture and v is the fluid velocity along the direction of the ith fracture normal to the
intersecting region.

Boundary Conditions

Periodic boundary conditions are used at the block boundaries for the fluid flow
simulations. The periodic boundary conditions are defined based on the constant pressure
difference in the x and y direction as (Watanabe et al. 2010):

p( x,0)  p( x,1)  J x
(3-21)

p(0, y)  p(1, y)  J y
(3-22)

v( x,0)  nx  v( x,1)  ny

(3-23)

v(0, y)  ny  v(1, y)  nx
(3-24)
where J is the pressure gradient in the corresponding Cartesian coordinate, n is the unit
normal vector to the boundary of the domain at a specific x and y and v is the velocity of
the fluid. After applying the abovementioned boundary conditions and solving the
Laplace’s equation, the velocity of the fluid is calculated through the REV. then the
average velocity through the entire REV in both x and y directions are calculated as follow:

v x    (v  n)dx (3-25)


60
 v y    (v  n)dy (3-26)


After calculating the average fluid velocity through eh REV, the permeability tensor
components is obtained using the following system of equations:

vx  (kxx J1  k xy J 2 ) (3-27)

vy  (k yx J1  k yy J 2 ) (3-28)

The above system of linear equations are solved two times (for J1=0 and J2=0) assuming the
pressure gradients only in x and y directions respectively.

3.3 Reservoir Scale Fluid Flow Simulation

Fluid flow in long fractures (l>50m) is coupled with discretised element based permeability
tensor in poro-thermo-elastic environment. There exist numerous approaches to model flow
and heat transfer in coupled poro-thermo-elastic environment. However, the main challenge
in coupling thermo-hydro-mechanical (THM) processes is the difference between the
characteristic time and spatial scales of each of the three processes, namely deformation,
pressure and temperature changes (Stephansson et al. 2004). For example, temperature
changes are subject to large time scale. On the other hand, the rock deformation happens in
short time scales because the geomechanical responses of the rocks has a similar
propagation speed as the speed of the elastic waves (Buzzi et al. 2007). Also the pore
pressure is affected by both small scale (local pore space heterogeneities) and large scales
(fractures, faults, joints, etc.).

Different numerical techniques have been used to model poro-thermo-elastic phenomena

in fractured porous media. To have a detailed understanding of the complex geomechanical
aspects of the fractured rocks and the induced perturbation, such as thermal drawdown
caused by the cold fluid injection in geothermal reservoirs an appropriate numerical
61
technique should be used which is capable of (a) adequately applying the boundary and
initial conditions and (b) accurate representation of the system geometry. In order to take
the aforementioned issues into account, FEM is used in the current study.

In FEM approach, material anisotropy can be introduced into the formulation by

assigning different values of a specific property to different elements. After discretising the
domain, the value of the objective function is estimated using an appropriate interpolating
function over the whole domain of the corresponding element. The interpolating functions
are usually polynomials degree of which varies depending on the number of nodes in the
element. After selecting an appropriate interpolating function, relevant Partial Differential
Equation (PDE) is formulated for the corresponding element. After forming the PDE it is
integrated over each element in such a way that the objective function is replaced with a
trial function in a polynomial form. Trial functions enable estimation of the objective
function in every location inside the element through a weighted sum of its values at each
node of the element. After forming the corresponding PDE for each element and
assembling all of them in a global system, the global system of equations are solved using
an appropriate iterative solver. A direct solver can also be used as an alternative.

In the following sub-sections, the governing constitutive equations for conservation of

mass, momentum and energy equations are presented. Also the local thermal non-
equilibrium concept which is used in this study is described. Then the discretization of the
governing equations in matrix form is shown. After that an efficient solution strategy used
to solve the system of linear equations is presented.

Governing Equations

In this section the governing equations for momentum, mass and energy conservations
are presented.

Momentum Conservation

The momentum balance equation for the linear elastic deformations can be written as:

62
     g  0 (3-29)

where,  is the density of the porous medium which can be written as Eq. (3-30) and g is
the gravity acceleration.

  l  (1   ) s (3-30)

Also Eq. (3-29) can be written in terms of effective stress as follow:

 (   pI )   g  0 (3-31)

Where,   is the effective stress, p is the pore pressure and I is the identity matrix. The
constitutive equation for the stress-strain relationship in a non-isothermal environment can
be written as shown in Eq.(3-32) and with few modifications can be written as in Eq.
(3-33).

d   D(d  d T ) (3-32)

   C (  T T  I ) (3-33)

Where, D is the operator matrix (Zienkiewicz 2000),  T is the thermal strain,  is the
total strain,  T is the thermal expansion coefficient, T is the temperature difference and C
is the fourth order material tensor as shown in Eq. (3-34).

C   ij kl  2G ik jl (3-34)

In Eq. (3-34),  is the Kronecker delta function, G is the shear modulus of elasticity, and
 is the Lame coefficient. Also the constitutive equation for the total strain-displacement
relationship is defined as follow:

63
 1
  (u  (u )T ) (3-35)
2

Mass Conservation

The mass balance equation for the fluid phase in a deformable non-isothermal porous
medium can be written as follow:

p u T
Ss    q    ( )  T Q (3-36)
t t t

where S id the specific storage which is defined as shown in Eq. (3-37), p is the pore fluid
pressure, T is the temperature,  T is the thermal expansion coefficient as shown in Eq.
(3-38), q is the fluid flux and Q is the sink/source term.

1 
S ( )( ) (3-37)
Ks Kl

T  (1   )   s    l (3-38)

Also the fluid flux term (q) in Eq. (3-36) can be described by using Darcy’s flow equation
as follow:

k
q (p   g )
 (3-39)

Introducing Eqs. (3-39) and (3-37) into Eq. (3-36) results in the following equation as the
general mass conservation equation for a fully saturated (single phase) deforming porous
medium in a non-isothermal environment:

 1   p u T  k 
( )  ( )     ( )  T  T   (p   g )   0 (3-40)
 Ks Kl  t t t  

In Eq. (3-40)  T is the thermal expansion coefficient of the medium and can be obtained
using the following equation:
64
 T  (1   )T solid
 Tliquid (3-41)

Where,  is the porosity of the medium, Tsolid is the thermal expansion coefficient of the

solid phase and Tliquid is the thermal expansion coefficient of the liquid phase.

Energy Conservation

The energy balance equation for a saturated porous medium can be written as:

T
(  c p )eff  .qT  QT (3-42)
t

Where, (  c p )eff is the effective heat storage of the porous medium which is defined in Eq.

(3-43), q is the heat flux and QT is the heat sink/source term.

(  c p )eff   (c p  )liquid  (1   )(c p  ) solid (3-43)

Also, both conduction and convection heat transfers are considered in this study based on
which the heat flux term in Eq. (3-42) can be written as:

qT  eff T  (c p  )liquid v  T (3-44)

where eff is the effective heat conductivity of the porous medium which can be defined as

Eq. (3-45) and v is the velocity of the fluid. Also the first term on the right hand side of Eq.
(3-44) is the conduction term and the second term is the convective heat transfer term.

eff  liquid  (1   )solid (3-45)

By introducing the Darcy’s law into Eq. (3-44) the general energy conservation equation
for a fully saturated single phase deformable porous medium in a non-isothermal
framework can be written as:

T  k 
(  c p )eff  (c p  )liquid   (p   g )  T  T (eff T )  Q (3-46)
t   

65
3.4 Local Thermal Non-Equilibrium

Classical macroscopic heat transfer within the porous medium is often treated using the
single equation of heat transfer which implied the condition of local thermal equilibrium
between the solid and the fluid phases at their interface (Birch et al. 1942). However, in
many transient heat transfer conditions, such as convection dominated systems, the rate of
heat transfer between the solid and liquid phases is not fast enough that two phases (solid
and liquid) can reach an equal temperature at their interface (Griggs 1939). Such a
difference between the temperatures at the solid/liquid interface region causes a gradient in
pore pressure and stress values at the interfaces. In such a treatment, local-thermal
equilibrium assumption is no longer valid in the calculations. However, both liquid and the
solid phases are considered as continuous and the heat transfer between the two phases is
described using an appropriate heat transfer coefficient. Also the heat transfer coupling
between the two phases are carried out using either analysing the microstructures or by
using empirical equations. In empirical approaches, two separate equations are used to
describe the heat transfer mechanism in each phases where both equations have a
modelling parameter called interfacial convective heat transfer coefficient (Piggott and
Elsworth 1991). A two equation model has been proposed by Fourie and Du Plessis (2003)
to consider the effect of local thermal non-equilibrium on the conduction heat transfer
between the solid and liquid phases. Local thermal non-equilibrium concept (LTNE) has
been extensively studied in both theory and applications by different authors (Borg and
Handin 1966). As an example, the effect of local thermal non-equilibrium on free surface
flows has been investigated by Alazmi and Vafai (2004). A schematic representation of the
local thermal non-equilibrium in a two phase, solid/liquid system is shown in Fig. 3-5. As
can be seen in the figure, the fluid and the solid (matrix) are in contact with each other and
they have different temperatures at the matrix/fluid interface. Because of the existence of
such a temperature difference, two distinct equations need to be applied for describing the
heat transfer between corresponding phases which are discussed below (Kaviany 2008,
Vafai 2009).

66
 Tf Ts

Fig. 3-5: Two medium treatment of the heat transfer in a fractured porous medium
In this study, the basic two equation model for local thermal non-equilibrium heat
transfer is used in a numerical poro-thermo-elastic framework to consider the effect of
LTNE on estimating potential recoverable energy from geothermal reservoirs. For this
purpose the Continuous-Solid model (Gholizadeh Doonechaly et al. 2012)Nelson 2001 is
used. The two equation model for the solid/liquid phases is expressed as:

T  As 
 v T f   hsf   (Ts  T f )   T ( f T ) (3-47)
t f    (cp ) f 

T  As 
  hsf   (T f  Ts )   T (sT ) (3-48)
t s  (1   )  (  c p ) s 

Where, As is the specific surface area,  is the porosity of the porous medium, T f and Ts

are the fluid and the solid temperature at the fluid/solid interface,  is the heat conductivity
and  c p is the heat storage of the corresponding medium. As shown in Equations (3-47)

and (3-48) the heat transfer is assumed by conduction in the solid phase and by conduction
and convection in the liquid phase. The axial conduction in both phases is also included in
the calculation by using the thermal conductivities of both solid and liquid phases (  ).

As shown in Equations (3-47) and (3-48) hsf is the heat transfer coefficient between the

solid and the liquid phases. Heat transfer coefficient can be formulated by using the
summation of the thermal resistances of the adjacent phases (solid and liquid) (Bejan,
1993) and is expressed as:

67
 1 1 1
  (3-49)
U  A hs  As h f  Af

where U is the total heat transfer coefficient, A is the area and h is the thermal
conductivity of each phases. Gelet et al. (2012) rearranged Eq. (3-49) to calculate interface
heat transfer coefficient for the solid/fracture geometry as shown in Fig. 3-6 as:

1 2b B
  (3-50)
hsf 2n f   f 2ns   s

Liquid

Solid B

Fig. 3-6: A schematic representation of the solid/fluid interface. B is the length of the interface and b is the fluid
width (fracture aperture).

where ni is the volume fraction of the ith phase, b is the fracture aperture, B is the length of
the common interface between the solid and liquid and  i is the thermal conductivity of
the ith phase. ni in Eq. (3-50) can be calculated as:

B2
ns  (3-51)
( B  2b) 2

4b( B  b)
nf  (3-52)
( B  2b) 2

However, the abovementioned formulation corresponds to the square matrix blocks

with the side length B. In this study the triangular block are used to increase the flexibility
in discretization fractures explicitly in the system as shown Fig. 3-7.
68
 A A
s f

Fig. 3-7: Matrix/fracture interface in a discretised domain using triangular elements. A s is the area of the shaded
element and Af is the area of a part of the fracture adjacent to the corresponding matrix element.

Therefore the volume fraction of the solid and the liquid phases for the triangular
geometry as shown in Fig. 3-7 can be calculated as:

As
ns  (3-53)
As  Af

Af
nf  (3-54)
As  Af

Also, Eq. (3-50) shows that the heat transfer coefficient is inversely related to the
interface length and the fracture aperture which may change due to the applied boundary
conditions and cold fluid circulation. Such a non-linearity needs the value of the heat
transfer coefficient to be calculated in an iterative framework.

3.5 Discretization of the governing equations

Weighted residual method and the Green’s theorem are applied to discretise the
abovementioned mass, momentum and energy conservations equations (Bathe 1996). As
mentioned before, the finite element method is used in this study for the numerical
simulation purpose. Therefore the state variables namely: displacement, pore pressure and
temperature are defined using proper shape functions as:
69
 u  Nu u (3-55)

p  Np p (3-56)

T  NT T (3-57)

where N is the corresponding shape function and u , p and T are the nodal values of the
corresponding state variable. By applying the Galerkin’s method and replacing the
weighting functions by the corresponding variables’ shape functions, the weak form of the
conservation equations can be written as follow (Ghassemi and Tarasovs 2006, Ghassemi
et al. 2007, Watanabe et al. 2010, Ghassemi and Zhou 2011, Koh et al. 2011, Gholizadeh
Doonechaly et al. 2012, Gholizadeh Doonechaly et al. 2012, Watanabe 2012):

p u T

 wS s
t
d    wT   d    w

t 
  wT  qH d  
t 
(3-58)
 w(qH  n)d    wQH d   0
 qH 

p b T
 wb S
d
m s
t
d    w m d    w
d
t d
t
d    wT  (bh qH )d 
d
(3-59)
  wbh (qH  n)d    wqH d  

 wqH d   0


 qH d d

T
 wc 

p
t
d    wc p  qH Td    wT  (T )d 
 
(3-60)
  w(T  n)d    wT QT d   0
Tq 

T
 wb c  d    wclp  l bh qH Td    wT  (bm  l T )d 
l l
m p
d
t d d
(3-61)
  w(bm  l T  n)d    wqT d  

 qT d   0


Tq d d

 w  (    pI )d    wT   gd    wT  td   w
s T T
 td d 
  t d
(3-62)
  wT  td d   0

70
where, w is the test function,  is the model domain,  is the domain boundary, t is the
traction vector and d is the fracture plane.

3.6 Results and discussions

The fluid flow and heat transfer model developed as part of this study is used to study the
long term response of Soultz-Sous-Forets geothermal reservoir due to the cold fluid
circulation. The results of this study are presented and discussed below.

3.6.1 Discrete Fracture Network

The fracture network of Soultz geothermal reservoir at the depth of 3650m is shown in
Fig. 3-8 (a). Also the corresponding discretised mesh of the domain (triangular elements) is
represented in Fig. 3-8 (b). As mentioned in the previous sections, short fractures (< 50 m)
are considered in calculating the permeability tensors. A schematic representation of the
short fractures in Soultz reservoir as well as the calculated permeability tensors are shown
in Fig. 3-8 (c) and (d). After generating the discrete fracture network map of Soultz
reservoir (Fig. 3-8 (a)) the fluid flow and heat transfer is simulated using the developed
model (as discussed in this chapter, (3.5)) and the long term response of the reservoir is
assessed for 1 and 12 years of production.

3.6.2 Pore pressure and Log10RMS Fluid Velocity

Results for pore pressure distribution and Log10RMS fluid velocity for Soultz
geothermal reservoir after 1 and 12 years of production are shown in Fig. 3-9 (a) and (b)
respectively. As can be seen from these figures after 1 year of production fluid mobilization
has taken place only through major interconnected fracture system. After 12 years of
production, however, the injection pressure significantly advances towards the production
well. By this time the fluid sweeps through the major parts of the reservoir.

71
 (a) (b)

(c) (d)
Fig. 3-8: (a) a schematic representation of the discrete fracture network of Soultz geothermal reservoir at the
depth of 3650 m (b) generated mesh of the domain using the triangular elements (c) a sample area containing short
fractures (d) calculated permeability tensor of for the elements which are cut by short fractures .

72
 (a) (b)
Fig. 3-9: A schematic representation of pore pressure distribution (top) and Log10RMS fluid velocity (bottom)
after (a) 1 and (b) 12 years of production for =78.9MPa, =53.3MPa, Pinj=44.8MPa, Pprod=31MPa,
o
Tinj=150 C .

73
3.6.3 Temperature

Temperature drawdown results (see Fig. 3-10 (a) and (b)) show that at early stage of
production (1 year) the heat transfer between rock and fluid has taken place near wellbore
region. With the pass of time the fluid sweeps over a large part of the reservoir which
increases heat transfer from the rock to fluid and after 12 years of production the average
matrix temperature drops from 200 to 170°C which is quite low (drop) compared to
previous study (Koh et al. 2011). This is due to the fact that in previous study formulation
of heat transfer is based on local thermal equilibrium. After 12 years of production the
temperature of the rock matrix throughout the reservoir is significantly reduced as
significant part of the reservoir is swept by the circulating fluid.

(a) (b)
Fig. 3-10: A schematic representation of the Reservoir temperature drawdown after (a) 1 year and (b) 12 years of
production for =78.9MPa, =53.3MPa, Pinj=44.8MPa, Pprod=31MPa, Tinj=150oC .

74
3.7 Closure

In this chapter, different methodologies which have been used in the literature to
simulate the fluid flow in the fractured reservoirs are briefly explained. Advantages and
disadvantages of each approach are also mentioned.

FEM is used in this study as a powerful technique to simulate the fluid flow and heat
transfer in fractured porous medium. Also the heat transfer in the fractured region is treated
as with two distinct equations using the local thermal non-equilibrium approach to consider
the difference between the heat transfer rates in two medium (solid/liquid).

Finally the discretised form of the mass, momentum and energy conservation equations
are presented to simulate the fluid flow in a poro-thermo-elastic framework.

Results of this study show that fractures provide the main flow paths in which
conduction as well as convection heat transfer mechanism occur during heat transfer
between the fluid and the rock matrix.

75
Chapter 4

4 STIMULATION
4.1 Overview

Fluid pressure is applied to displace fracture surfaces relative to each other in order to
enhance the fracture permeability. When the fluid pressure is applied the effective and
shear stresses change and they can be expressed as (Detournay and Cheng 1993):

 eff   t  p (4-1)

  c   n tan( ) (4-2)

where  t is the total stress,  eff is the effective stress,  is the friction angle, c is the

cohesive strength and  n is the normal stress.

As the fluid injection continues the shear stress increases and reaches a threshold value.
At this time the shear strength of the rock can no longer resist shear displacement of the
fracture surfaces and thus the shear dilation occurs. Rock failure by shear displacement is
known as Mode II failure. In Mode II, the surface asperities of the rock slide over each
other which cause normal separation (dilation) of the fracture surfaces. Interlocking of
asperities irreversibly increases the permeability of the rock depending on material strength
properties (Murphy and Fehler 1986). As the effective closure stress reaches zero,
separation of the fracture surfaces takes place perpendicular to the fracture walls. If the
injection is continually increased the effective stress becomes negative and at some stage
the tensile strength of the rock is overcome which leads to tensile failure of the rock
(known as Mode I). Mechanical representation of the shear displacement and the normal
separation of the fracture surfaces can be described based on a specific failure criterion,
such as Mohr-Coulomb (see Fig. 4-1 ). As shown in figure, when the minimum principal
stress (closure stress) reaches the origin (O) the normal separation (Mode I) occurs.
However, the shear dilation happens much earlier, when the Mohr circle encounters the
76
failure envelope (CD). As the effective stress decreases, Mohr-circle moves towards the
origin. At point E (when the Mohr-circle merely touches the failure line) the shear
displacement happens (Murphy and Fehler 1986, Mogi 2007, Fjar et al. 2008, Sivakugan et
al. 2012).

O A B
Increase in Fluid pressure
required to start shear dilation

Increase in Fluid pressure required to start normal

separation of the fracture surfaces

Fig. 4-1: Mohr diagram describing the initiation of the shear dilation and normal fracture surface separation (Murphy
and Fehler 1986, Doonechaly et al. 2013).

Since the shear dilation is caused by the slippage of the asperities over each other a
maximum dilation can be achieved when large asperities ride on top of each other. A
schematic representation of the roughness induced fracture opening caused by the shear
displacement is shown in Fig. 4-2. This maximum displacement is called characteristic
height of the fracture (Kotousov et al. 2011). Based on experimental studies the
characteristic height is measured to be of the order of a fraction of a millimetre (Heidinger
et al. 2006). Fracture aperture that can be created by conventional hydraulic fracturing,
however, is in the order of tens of millimetres (Blumenthal et al. 2007). Reservoir rocks
with rough surfaces and high shear strength are highly desirable as they cause maximum
shear displacement. Shear dilation by induced fluid pressure was first detected in the

77
laboratory experiments in 1970s. One of the early attempts was made by Lockner and
Byerlee (1977) and the results showed a significant permeability increase-

Fig. 4-2: roughness induced shear dilation caused by the fracture asperities.

by shear displacement. This observation was confirmed by Hast (1979) and Solberg (1980)
both in the laboratory and in the field. Another field example is the geothermal reservoir at
Fenton Hill, New Mexico. During the field trial a large volume of water (21,000 m 3) was
injected into the granite rock at the depth of 3.5 km (Richards et al. 1994) and the
information on shear displacement (micro-earthquakes) was obtained using geophones and
seismometers. Since then, shear dilation has been comprehensively studied in geotechnical
and mining engineering. Investigation of permeability enhancement by shear dilation in
petroleum reservoirs, however, began much later (Rahman et al. 2000).

In order to understand factors controlling the shear displacement process it is essential

to model the rock failure mechanism by including parameters, such as surface asperities,
rock material properties, fracture geometry, in-situ stress condition etc. Also noteworthy is
that the injected fluid pressure may cause significant changes to the stresses around the
fractures. Therefore it is important to couple displacement with fluid flow and heat
transfers.

Several attempts have been made in the literature to characterize the shear dilation of
the fracture surfaces (Hossain et al. 2002, Rahman et al. 2002). Contribution to our existing
knowledge on shear displacement by induced fluid pressure can be summarized as follows.

One of the most comprehensive attempts to characterize the shear dilation caused by
the fracture surface asperities was developed by Barton et al. (1985). In their model, the

78
rock behaviour is described by characterization of fracture surfaces, fracture aperture and
modelling of the effect of normal and shear displacement on fracture aperture. Their
simulation results are also compared with experimental data. According to Barton et al.
(1985) shear dilation caused by the slippage of the fracture surfaces is estimated as (Barton
and Choubey 1977):

1 JCSn
d n0 (mob)  JRCn (mob) log( ) (4-3)
2  n

where d is the amount of dilation and  n is the normal stress on the fracture surfaces.  n is
defined using an empirical equation (Bandis, 1980) as:

VJ
 n  (4-4)
a  b  VJ

where VJ is the fracture closure,  n is the normal closure stress and ‘a’ and ‘b’ are
constants. A number of parameters are needed to be known as a priori in their model, such
as Joint compression strength (JCS), joint roughness coefficient (JRC), Unconfined
compression strength (  c ), Residual friction angle ( r ), Conducting aperture ( e ) and
Mechanical aperture ( E ) (Barton et al. 1985). The methodologies for determination of
these parameters as summarized below.

The JRC can be calculated as:

  r
JRC 
JCS (4-5)
log( )

 no

 is the
where  is the tilt angle when the sliding between the fracture surfaces happens,  no

effective normal stress during shear displacement and r is the residual friction angle. The
residual friction angle may be lower than the basic friction angle because of the weathering
effects which can be calculated according to Barton and Choubey (1977) based on the
Schmidt hammer test results as:

79
 r
r  (b  20)  20 (4-6)
R

where b is the basic friction angle, R is the Schmidt rebound on dry un-weathered sawn
surface, and r is the Schmidt rebound on wet joint surface. The JRC varies between 0 to
about 15 where zero corresponds to the residual non-dilatant joint surface.

Mechanical aperture can be calculated using an empirical equation proposed by Barton

and Bakhtar (1983) as:

JRC 
Am  (0.2 c  0.1) (4-7)
5 JCS

c
where Am is the mechanical aperture, is a measure of the fracture surface alteration
JCS
and weathering effects. In case of no alteration or weathering,  c is equal to the JCS which
implies that the mechanical aperture is a function of the JRC only. As an alternative, the
mechanical aperture can be estimated experimentally by using pumping test in which the
equivalent smooth wall aperture can be obtained in terms of conductivity obtained from the
flow tests as:

e  (12k )0.5 (4-8)

where k is the conductivity and e is the conductivity aperture. Conductivity aperture can be
obtain by using statistical interpretation of the borehole flow tests (Snow 1968).

By calculating the shear dilation value at each stage of the shear displacement the
conductivity aperture of the fracture can be calculated using Eq. (4-8) and the fluid flow
simulation is carried out using the updated permeability values.

Main advantages of the model proposed by Barton et al. (1985) are: Considering the
effect of sample size, using simple data obtained from laboratory for representing fracture
surface roughness and coupling between the shear stress, displacement, dilation and
conductivity. The main problem in implementing this model, however, is under-estimation
of the value of hydraulic aperture (Buzzi et al. 2007).

80
 Following Barton et al. (1985) Piggott and Elsworth (1991) proposed a methodology
based on fracture surface topography to estimate shear displacement under different
stresses. Such a topographical model uses empirical correlations to describe the fracture
surface roughness both qualitatively and quantitatively. Fracture surface roughness is also
characterized by using the self-affine fractal model (Voss 1985, Rayner 1995). A self-affine
fractal surface is defined based on the semi-variogram of the fracture surface elevation as:

 z   h (4-9)

where h is the lag distance. Also  and  are the constants describing the surface
roughness.

The fractal dimension of the rough surface can also be calculated as:


D  2 (4-10)
2

Also the aperture distribution of a fracture subjected to shear ( x ) and normal ( z )

displacement is calculated as follow:

bx  zxx  z x z (4-11)

where z x the fracture surface elevation and bx is is the aperture distribution.

Based on the aperture distribution, the arithmetic mean and variance of the aperture
distribution can be used to generate an appropriate aperture distribution model, such as
normal distribution.

After defining the aperture distribution, a mechanical model is utilized to describe the
stress-displacement relationship. For this purpose Piggott and Elsworth (1991) used the
Gangi’s (1978) bed of nails model. In a bed of nails model, the force between the fracture
surfaces is calculated as:

EA
Fi   bi H (bi ) (4-12)
L

81
Where, E is the modulus of elasticity, L is the length, A is the cross sectional area, bi is the

aperture and H (bi ) is the Heaviside step function ( H (bi ) is 1 and 0 for positive and
negative apertures respectively). Also negative aperture is a representation of an elastic
contact between the fracture surfaces which causes elastic deformation.

Eq. (4-12), gives the transmitted force between the fracture surfaces at location “i”. Then,
average normal stress acting between the fracture surfaces is calculated by summation of
the forces at all available locations along the fracture length and dividing by the fracture
surface as (Piggott and Elsworth 1991):
0
E
S zz    b  pb db
L 
(4-13)

Where, pb is the aperture distribution, E is the Young’s modulus of elasticity, L is the

fracture length and b is the incremental fracture length. The dilation of the fracture is then
calculated as:
d  z  z0 (4-14)

Where, z0 is the initial dilation corresponding to the in-situ stress boundary condition and
z is the calculated dilation based on the stress value by using Eq. (4-13). An example of
the shear dilation versus shear displacement calculated by Piggott and Elsworth (1991) for
a hypothetical fracture specimen is shown in Fig. 4-3.

Fig. 4-3: Shear dilation vs. shear displacement for a hypothetical fracture specimen (Piggott and Elsworth 1991).

82
 By using surface topographical model it is possible to characterize the fracture surface
roughness and estimate shear dilation under different test conditions, however, Piggott and
Elsworth (1991) used empirical equation to stress-displacement relationship based on the
fracture surface roughness.

Olsson and Brown (1993) developed an experimental procedure to study the effect of
normal stress and shear dilation on fluid flow properties of a naturally fractured core
sample. They have used a servo-controlled axial/torsion load frame to test the fluid flow
and mechanical behaviour of the fracture surface during normal stress, slip and shear
dilation (Berry et al. 1996). A schematic representation of the experimental set up designed
by Olsson and Brown (1993) is shown in Fig. 4-4 (a) and (b).

(a) (b)

Fig. 4-4: Experimental set up used by Olsson and Brown (1993). (a) Rotary shear sample with radial flow injection
(b) schematic representation of the aluminium core holder (Olsson and Brown 1993).

They also used a “microscopic theory of surface contacts” (Berry et al. 1996) to
determine the stress dependent properties measured in the laboratory. Such a microscopic
model, describes the elastic properties of the fracture surface based on the surface
topography. By using this approach parameters related to surface roughness, such as radius
of curvature, aperture distribution and roughness elevation can be determined based on the
statistical analysis of the fracture surface roughness.

Based on the experimental results a relationship between the shear displacement and
roughness induced shear dilation was estimated. In their experiments dilation did not

83
happen until 0.2 mm of shear displacement. Above this value the shear strength of the rock
is overcome. They also determined the wavelength of surface roughness of fracture and
established a threshold value below which the fracture surfaces were closely matched. They
have related this threshold value to the characteristic slip shear test. They predicted that the
fracture surfaces can slip atop each other without any dilation before the large scale
asperities interact with each other. Also Olsson and Brown (1993) predicted that if the
fracture surfaces are closely matched, fluid injection inside the fracture may cause
significant and permanent permeability increase of the system. In summary based on this
study one can determine parameters, such as surface roughness, mismatched wavelength
and surface toughness (in the form of spring constant) and used in the numerical model to
estimate shear dilation and permeability enhancement for a given rock parameter and
reservoir condition. Willis-Richards et al. (1996) proposed an improved methodology to
determine the change in fracture aperture based on the amount of shear displacement
between the fracture surfaces and the stress boundary condition. They have used the
hyperbolic joint closure model to describe the normal compliance of the fractures (Lockner
et al. 1977, Hast 1979). Also the effect of fluid pressure on the fracture surface separations
is characterized by using normal fracture stiffness. A schematic representation of a fracture
undergoing shear displacement and required parameters for shear dilation characterization
are shown in Fig. 4-5. In their model, the fracture aperture when the fracture surfaces are in
contact is defined as:

a0
a  as  ares
9  (4-15)
1
 nref

Where, a0 is the initial fracture aperture,  nref is the effective normal stress at the fracture

surfaces required to cause 90% reduction in fracture aperture, as is aperture change caused

by consequent shear displacement and ares is the residual aperture at high effective stress
which is usually assumed to be zero for simplicity. Also fractures are assumed as
ellipsoidal objects with minor and major axis of R and 2R respectively. The volume of such
a shape is calculated as (Willis-Richards et al. 1996):

84
 Us

=dilation angle
a0 =initial aperture
as =stimulated aperture
Us =shear displacement
Us =

Fig. 4-5: A schematic representation of a fracture undergoing shear displacement and the required parameters to
characterize the dilation event.

85
 8 
V   R3 (4-16)
3 E

Where,  is the normal stress exerted on the fracture surface and E is the modulus of
elasticity of the rock. After calculating the volume of the fracture, the average aperture can
be calculated by dividing the fracture volume by the fracture surface area as:

8 
a R (4-17)
3 E

Then the change in average fracture aperture respect to the change in stress (average
normal compliance) can be calculated by taking the derivative of Eq. (4-17) as:

da 8 R
 (4-18)
d 3 E

Also the peak shear stress needed to start the shear displacement is defined as:

    tan(basic  dileff ) (4-19)

where  is the threshold shear stress to start the shear displacement,   is the effective
stress exerted on the fracture surfaces, basic is dependent on material properties of the

fracture surface and dileff is the effective shear dilation angle at a specific normal stress. dileff
depends on the fracture surface asperities and the stress boundary condition and can be
calculated as:

dil
dileff 
9  (4-20)
1
 nref

When the shear stress exceeds the threshold value as shown in Eq. (4-19), the fracture
surfaces start to move. However, the amount of such a displacement depends on the shear
stiffness of the fracture and the amount of the excess shear stress above the threshold value.
Fracture shear stiffness can be calculated as:

86
 G
Ks   (4-21)
R

Where,  is a constant close to unity, G is the shear modulus and R is the short radius of
the ellipsoidal fracture.

After calculating dileff by using Eq. (4-20), the change in fracture aperture can be calculated
as a multiplication of the displacement and the effective shear dilation angle as:

as  U tan(dil
eff
) (4-22)

By combining Equations (4-15), (4-20) and (4-22) the aperture of a fracture under shear
dilation can be calculated as:

a0  U tan(dil )
a
9  (4-23)
1
 nref

The formulation proposed by Willis-Richards et al. (1996) can be very helpful in

understanding the behaviour of a fractured system under shear stimulation, however, the
derived equations are not “entirely self-consistent” especially when jacking happens
(Willis-Richards et al. 1996). The derived equations are a simple set of correlations to have
an understanding of the behaviour of a system which consists of a large number of fractures
in wide range of scales. Also the fracture propagation was not included in the modelling
procedures because of the applied stress boundary conditions and inherent assumptions. In
real in-situ stress boundary conditions, however, a significant number of fractures may
undergo propagation before reaching the shear displacement threshold (Willis-Richards et
al. 1996)Tran 2004.

Hossain et al (2002) and Rahman et al. (2002) extended the work of Willis-Richards et
al. (1996) by considering the effect of fracture propagation during shear dilation event.
They used an analytical approach (Rahman et al. 2000) to incorporate the fracture
propagation into the modelling procedure. The applied methodology is briefly discussed
below.

First the direction of the probable propagation of the fracture is calculated as:
87
   0 if K II  0 (4-24)

 K  K 2  8K 2 
  2 tan  I 1 I II
 if K II  0 (4-25)
 4 K II 

Where  is the probable propagation angle (considered as anti-clockwise), K I and K II are

mode I and II fracture stress intensity factors respectively and can be calculated by using
Equations (4-26) and (4-27) as follow:

K I  C l  p f  1 sin 2    3 cos 2    (4-26)

1
K II  C l  ( 1   3 )  sin(2 ) (4-27)
2

2
Where, C is proposed to be equal to (Solberg et al. 1980),  is the angle between the

fracture plane and the maximum horizontal stress,  1 and  3 are the maximum and
minimum horizontal stresses respectively.

Equation (4-25) gives two values of  because of the  sign, however, the
corresponding  value which yield bigger maximum tensile strength is selected for further
calculations. The maximum tensile strengths corresponding to both  values (obtained
from Eq. (4-25)) are calculated as:

1    2   3 
  cos( 2 )   K I cos ( 2 )   2 K II sin( )  (4-28)
max
2r    

The calculated propagation angle (  ) is then used to check whether the fracture
propagation happens or not by using the following equation.

  2 3 
 cos( )( K I cos  K II sin  )   K Ic (4-29)
2 2 2 

Where, K Ic is the mode I fracture toughness. In case of any fracture propagation, the new
fracture half-length is calculated as:

88
 li 1  li2  r 2  2li r cos  (4-30)

where li is the initial fracture half length,  is the propagation angle and r is the pre-
specified incremental length.

The fracture propagation is incrementally modelled until no further propagation

happens and the final li 1 will be the resulting fracture’s half-length. Therefore an iterative
procedure is applied to obtain the amount of fracture propagation. A schematic
representation of a penny-shaped fracture undergoing propagation which is characterized
by Rahman et al. (2002) model is shown in Fig. 4-6.

Fig. 4-6: Penny-shaped fracture undergoing propagation subject to maximum and minimum horizontal stresses
(Rahman et al. 2000).

As mentioned, all the previous attempts focused on fractures with limited sizes subject
to uniform stress boundary conditions. It is, however, important to consider fractures
subject to varying boundary conditions, such as stress and rock mechanical properties along
the surface of the fracture. This can yield a more accurate estimation of the aperture
distribution. In this study, an analytical methodology based on distributed dislocation
technique proposed by Kotousov et al (2011) is used to estimate the aperture distribution
caused by the shear dilation in a fracture subject to varying stress boundary conditions.

89
4.2 Current Approach

Two major assumptions are used in this study to characterize the shear dilation of the
fracture surfaces. These two assumptions are discussed below.

Shear Slippage

The shear slippage between the fracture surfaces is described by using Coulomb friction
law which explains the friction stress during the shear slippage based on the normal stress
exerted on the sliding planes as:

n 0  f n (4-31)

Where,  0 is the threshold shear stress value to initiate the shear slippage between the

fracture surfaces,  n is the normal stress exerted on the fracture surfaces and f is the
friction factor. Friction factor is dependent on the material properties, fracture geometry
and surface asperities of the fracture (Kotousov et al. 2011).

Coupling Between the Shear Slippage and Shear Dilation

In this study the coupling between the shear displacement and the fracture aperture is
described by a step function as shown in Fig. 4-7.

Fracture Opening

Shear
Displacement
Fig. 4-7: Fracture opening vs. shear slippage of fracture surfaces. is the characteristic height of the fracture
surface asperities.

90
 Fracture displacement normal to the fracture plane is simulated by using virtual springs
distributed along the fracture length. The springs are characterized by a spring constant
which can be calculated numerically, experimentally or analytically (Kotousov et al. 2011).
The spring deformations are then modelled in an elastic framework which results in the
following system of equations describing the stress between the fracture surfaces:

 n  kE (   y ) for  y   (4-32)

 n  0 for  y   (4-33)

Where,  is the characteristic height of the fracture as shown in Fig. 4-7 and k is the spring
constant. Also noteworthy is that that Eq. (4-33) implies the complete separation of the
fracture surfaces in which no contact exists between the fracture asperities. The spring
constant is also calculated based on the bed of nails model proposed by Gangi (1978) as:

b
kE (4-34)
L

Where, E id the Young modulus, L is the fracture length and b is a constant less than unity.

Boundary conditions

The complete set of boundary conditions for a fracture as shown in Fig. 4-8 are listed
below (Kotousov et al. 2011):

Fig. 4-8: A schematic representation of a fracture subject to in-situ stress boundary conditions
(Kotousov et al. 2011).

91
  y    2 2
for x  y   (4-35)

 y  kE (   y )  p c x a (4-36)
for

 y  p x c (4-37)
for

 n   0  f  kE (   y ) for c  x  a (4-38)

 n  0 for x  c (4-39)

ux  0 for x  a (4-40)

u y  0 for x  a (4-41)

where   is the far-field stress boundary condition, p is the fluid pressure inside the
fracture, k is the spring constant,  is the characteristic height of the fracture, f is the
friction factor, E is the Young’s Modulus of elasticity, 2ais the fracture length, 2c is the
length of the fracture which the fracture surfaces are not in contact and  0 is the threshold
shear stress to start the shear displacement.

System of Integral Equations

The perturbation of the fracture is assumed to be caused by the distribution of dislocations

along the fracture length (Gangi 1978). Appropriate dislocation densities are used to solve
the following system of integral equations (Kotousov et al. 2011):

E  y ( )
a
a x   d    p for x  c

(4-42)
4

E a  y ( )
a x   d    p  kE (   y ) for


4 (4-43)
c x a

E  x ( )
a

4 
a x  
d    for x  c (4-44)

92
 E a  x ( )
4 a x 
d      0  f  kE (   y ) for
(4-45)
c x a

where  is the appropriate dislocation density which can be calculated by introducing a

new function as (Zhang et al. 2005):

 ( )
 ( )  (4-46)
1  2

where  is normalized distance respect to the centre of the fracture as:

x
 (4-47)
a

Also the fracture aperture at each location along the fracture length can be calculated as:
a
 ( x)    y ( )d (4-48)
x

Solving the resulted system of Equations (4-42)-(4-48) is carried out by using the
Gauss-Chebyshev quadrature method (Rasmussen et al. 1987) which results in the
following linear system of equations:

1 N  ( si ) 4
  (   p) for  (t j )  
N i 1 t j  si E
(4-49)

1 N  ( si ) 4  ka i  j

N i 1 t j  si
 
E
(  p  kE  )  4  (si )
N i 1
(4-50)
for  (t j )  

Where, si and t j are calculated as shown in Equations (4-51) and (4-52) respectively.

2i  1
si  cos( ) for i=1,…,N (4-51)
2N

j
t j  cos( ) for j=1,…,N-1 (4-52)
N

93
 After calculating the  in the above solution procedure, the fracture aperture at
different locations along the fracture length is calculated as:
i j
1
 (t j ) 
N
 (s )
i 1
i (4-53)

After calculating the fracture aperture along the fracture surface, the Mode I and II
stress intensity factors are calculated as (Rasmussen et al. 1987):

k (E )2
KI  (4-54)
(   p)  a  (   p) a  2k (E ) 2

2 0 E  fk (E )2
K II   
a  (4-55)
(   p)  a  (   p) a  2k (E ) 2

Such an analytical approach enables us to have a more accurate estimate of the aperture
along the fracture length with an extremely lower computational effort compared with the
other approaches. After calculating the fracture aperture distribution due to the applied
boundary condition, the change in the fracture permeability is calculated and the fluid flow
simulation will be updated with the new values.

94
4.3 Closure

In this chapter, an overview of the previous approaches used to model shear induced
fracture opening is carried out and the methodology which has been used in this study is
discussed. Also the proposed methodology gives the fracture aperture distribution along the
fracture surface for fractures with different geometries subjected to varying boundary
conditions with a considerably low computational cost compared with other proposed
approaches.

95
 Chapter 5

5 CASE STUDY
The stimulation technique described in Chapter 4 is used to study the response of
Soultz-Sous-Forets geothermal reservoir due to the pressurization by injection fluid. The
tectono-stochastic model presented in Chapter 2 is used to generate subsurface discrete
fracture network map of Soultz-Sous-Forets geothermal reservoir. The reservoir was
pressurized by injected fluid pressure for 60 weeks. During the pressurization period the
shear dilation events and change in permeability was monitored. After this period a
circulation test was carried out for 14 years and the changes in permeability due to poro-
thermo-elasticity was monitored. Also data, such as the produced fluid temperature and
matrix temperature drawdown was monitored during the same period.

5.1 Model Setup

In the current poro-thermo-elastic numerical model, flow from the injection well is assumed
to be planar and is approximated through the open-hole interval (3500m to 3800m). A square
region of the Soultz geothermal reservoir is chosen with a side length of 500m at the depth
of 3650m which is midway through the depth-averaged open-hole section (see Fig. 5-1 (a)).
The injection well is placed at the bottom left corner of the domain (see Fig. 5-1 (a)).
Fractures are spatially distributed according to their statistical properties obtained from the
field data (see Table 5-1). The corresponding discrete fracture network and the discretised
mesh of the domain (triangular elements) are represented in Fig. 5-1 (a) and (b)
respectively. The reservoir is divided into 32,000 grid blocks. The model is meshed using
triangular elements. In order to improve the stability of the numerical solution linear triangles
are used for pressure and temperature and quadratic triangular element for the displacements.
As mentioned previously (section 3.2), short fractures (< 50 m) are considered in
calculating the permeability tensors. A schematic representation of the short fractures in

96
Soultz reservoir as well as the calculated permeability tensors are shown in Fig. 5-1 (c) and
(d).

Table 5-1: Characteristic statistical properties of the fractures obtained from the different data
sources, such as core descriptions, well logs etc. (Gentier et al. 2010)

Fracture set F1 F2 F3 F4 F5
Distribution Law Normal Normal Normal Normal Uniform
Azimuth

Mean 2 162 42 129 0

Half-Width 16 19 6 6 180
Distribution Law Normal Normal Normal Normal Normal
Mean 70 70 74 68 70
Dip

Half-Width 7 7 3 3 9
Dip Direction NW NE NW SW -
Fracture No. 1.3E-7 3E-9 1.76E-8 3.3E-8 1E-8
Radius (m) 187 150 95 112 100
Transmissivity (m2\s) 6E-6 6E-6 4E-6 2E-6 5E-7

The reservoir is pressurized by injecting fluid through the injection well (GPK2) at a set
pressure of 65MPa. In order to increase the injectivity, a planar fracture of half-length of 50m is
placed at the injection well. The pressurization was carried out over a period of 60 weeks
(Gholizadeh Doonechaly et al. 2012). During the pressurization, the change in fracture aperture
for each individual natural fracture and the resulting permeability tensor are calculated.
Maximum and minimum horizontal principal stresses are acting along the x- and y-axis
respectively. Stress boundary conditions as well as the reservoir parameters used for this study
are presented in Table 5-2. No flow boundary condition is applied at the boundaries of the
reservoir (Gholizadeh Doonechaly et al. 2012).

Overall, two major computational parts are involved in this section. A poro-thermo-elastic
finite element based model is used to characterize the induced stresses during the stimulation
period. After obtaining the in-situ stress distribution, the change in fracture aperture is
calculated using the methodology described in Chapter 4. The results are discussed below.

97
 Injection Well
(GPK2)

(a) (b)

(d) (c)

Fig. 5-1: (a) a schematic representation of the discrete fracture network of Soultz geothermal reservoir at the
depth of 3650 m (b) generated mesh of the domain using the triangular elements (c) a sample area containing short
fractures (d) calculated permeability tensor of for the elements which are cut by short fractures.

98
 Table 5-2: Stress and reservoir data for Soultz geothermal reservoir (Kolditz 1995, Genter and Traineau 1996,
Koh et al. 2010, Watanabe et al. 2010). Parameters are considered to be constant during the cold fluid circulation.

Rock Properties
Young’s modulus (GPa) 40
Poisson’s ratio 0.25
Density (kg/m3) 2700
Fracture basic friction angle (deg) 40
Shear dilation angle (deg) 2.8
90% closure stress (MPa) 20
In situ mean permeability (m2) 5.2 x 10-17
Fracture properties
Fractal Dimension, D 1.2
Fracture density (m2/m3) 0.12
Smallest fracture radius (m) 15
Largest fracture radius (m) 250
Stress data
Minimum horizontal stress (MPa) 53.3
Maximum horizontal stress (MPa) 78.9
Fluid properties
Density (kg/m3) 1000
Viscosity (Pa s) 3 x 10-4
Hydrostatic fluid pressure (MPa) 34.5
Injector pressure, stimulation (MPa) 65
Injector pressure, production (MPa) 51.7
Producer pressure, stimulation (MPa) N/A
Producer pressure, production (MPa) 31.7
Other reservoir data
Well radius (m) 0.1
Number of injection wells 1
Number of production wells 2
Reservoir depth (m) 4430
Specific heat (rock) (Jkg-1K-1) 1.098x103
Specific heat (fluid) (Jkg-1K-1) 4.05x103
Thermal Conductivity (rock) (Wm-1s-1) 2.58
Thermal Conductivity (fluid) (Wm-1s-1) 0.68

99
5.2 Results and Discussions

Results of shear dilation are presented as average percentage increase in fracture

aperture with time (see Fig. 5-2).

Fig. 5-2: Percentage increase in average fracture aperture with stimulation time for P inj=65MPa, =53.3MPa,

=78.9MPa (Gholizadeh Doonechaly et al. 2012).

From Fig. 5-2 it can be seen that there exists three distinct phases in aperture change
history: 0-38 weeks, 38-50 weeks and 50 weeks and above. As shown in Fig. 5-2, a gradual
increase in average fracture aperture can be observed in the first phase of the stimulation
period (first 38 weeks). Also the corresponding locations of the dilation events are
presented in Fig. 5-3 (a)-(d).

100
 (a) (b)

(c) (d)
Fig. 5-3 : Cumulative location of the shear dilation events after (a) 20 weeks (b) 38 weeks (c) 50 weeks and (d) 60
weeks of stimulation for Pinj=65MPa , =53.3MPa, =78.9MPa.

As the injection of the fluid continues more fractures dilate. At the end of this stage the
average fracture aperture reaches a value of about 14% as shown in Fig. 5-2. The shear
dilation events are primarily distributed near the injection well within 200m away from the
injection well.
101
 Following the first phase pressure builds up reaches a threshold value. At this time the
average fracture aperture increases sharply until about 50 weeks, thus reaching 45%
increase in average fracture aperture. During the second phase of stimulation the shear
dilation events (see Fig. 5-3 (c)) extend further away from the injection well, however,
sparsely populated.

After 50 weeks of stimulation the rate of increase in average fracture aperture decreases
gradually and it reaches a plateau at about 60 weeks. After this period no significant
dilation events can be observed. This infers that for every set of reservoir and stress
parameters as well as injection schedule, an optimum level of shear dilation can be
achieved. At the end of phase 3 the average increase in fracture aperture reaches a value of
about 54% (see Fig. 5-3 (d)). The stimulation results (see Fig. 5-3 (d)) show that the shear
dilation events are distributed uniformly in all directions from the injection well. Also for
this particular case, for the given distribution of fracture, there exist a reasonable
consistency between the fracture density (see Fig. 5-1 (a)) and the occurrence of the shear
dilation events.

Results of this study are compared with that from a previous study (Koh and Rahman et
al, 2011), in which shear dilation events are estimated based on a simplistic model (Willis-
Richards et al, 1996). This comparative study is shown in Fig. 5-4 and Table 5-3
respectively. From Fig. 5-4 a number of distinct features can be observed. Firstly the time
required to overcome the threshold stress is 40 weeks which is about 12 weeks longer than
the previous study. Secondly, the time required to create maximum effective reservoir
volume (rock volume affected by the fluid circulation estimated by the methodology
proposed by (Robinson and Kruger 1992)) is almost 8 weeks longer than the previous study
(52 weeks). Results of this study clearly demonstrate that the material properties, such as
the surface roughness and Modulus of Elasticity used to estimate residual aperture provide
a more reasonable prediction of shear displacement events and the resulting residual
aperture. These results confirm that the reservoir volume (interconnected fracture networks
for the effective heat transfer area) estimated by this study is much smaller (retained
fracture aperture) than that estimated by previous study.

102
 (Koh et al. 2010)

Fig. 5-4: Comparison of the percentage of average aperture increase between the current study and the previous
approach

Table 5-3: Comparative study of the percentage increase in average fracture aperture between the current
study with that of Koh et al 2011.
%increase in average %increase in average
Time (weeks) fracture aperture (Current fracture aperture (Koh et al
Study) 2011)
40 17.64 18.99
45 33.98 39.34
52 48.2 53.1

In a second study the location of the microseismic events produced during field test at
Soultz reservoir are compared with that from this study as shown in Fig. 5-5 (Baria et al.
2005, Evans et al. 2005, Kohl and Megel 2007, Tischner et al. 2007). The unfilled circles
represent events generated due to injection of fluid after 60 weeks during the field trial at a
depth of 3650m. Results of this study show a close match with that from the field test.

The reservoir pressure and stress distribution after 40 and 60 weeks of the stimulation are
shown in Fig. 5-6 and Fig. 5-7 respectively. As shown in the Fig. 5-6 (a), after 40 weeks of
stimulation the injected fluid has reached almost all major fractures. After 60 weeks of
injection (Fig. 5-6 (b)) the pressure is well established in all parts of the reservoir domain. Also
the x- (top) and y-components (bottom) of the effective stress for 40 weeks and 60 weeks are
shown in Fig. 5-7 (a) and (b) respectively. Results of effective stress profile show a significant

103
increase in tensile stress (blue) over the entire reservoir domain towards the end of the
stimulation period.

Fig. 5-5: A comparison between the dilation events in this study (filled circles) with that from the field test trial
(unfilled).

(a) (b)
Fig. 5-6: Pore pressure distribution of the fractured reservoir at different stimulation stages: after (a) 40 weeks
and (b) 60 weeks with σH = 78.9MPa and σh = 53.3MPa, Pinj = 65MPa.

104
 (a) (b)
Fig. 5-7: x (top) and y (bottom) components of effective stress after: (a) 40 weeks and (b) 60 weeks of
stimulation for σH = 78. 9MPa and σh = 53.3MPa, Pinj = 65MPa.

Following the 60 weeks of stimulation, a flow test was carried out at an injection pressure of
51.7MPa and a production pressure of 31.7MPa (at a reservoir impedance of 20MPa between
the injection and production wells) for period of 14 years. During this production period, pore
pressure profile, Log10 root means square (RMS) velocity profile (which is directly
proportional to the permeability) and the matrix temperature drawdowns are monitored. In Fig.
5-8 and Fig. 5-9 the pore pressure and the Log10 RMS fluid velocity profile after 3 years and
14 years of production are presented respectively.

105
 (a) (b)
Fig. 5-8: Pore pressure distribution at different production stages: after (a) 3 years and (b) 14 years of production
for σH = 51.7MPa and σh = 44.8MPa, Pinj = 51.7MPa and Pprod=31.7MPa.

(a) (b)
Fig. 5-9: Log10 RMS velocity profile at different production stages: after (a) 3 years and (b) 14 years of production
for σH = 51.7MPa and σh = 44.8MPa, Pinj = 51.7MPa and Pprod=31.7MPa.

From the results it can be observed that during the early stage of production period (3 years)
high pore pressure is primarily built up around the injection well and the flow of fluid is
primarily through major inter-connected flow paths (Fig. 5-8 (a) and Fig. 5-9 (a)). With the
progress of time the injection pressure advances towards the production well and at about 14
years of production the injection fluid sweeps through almost all parts of the reservoir Fig. 5-8
(b) and Fig. 5-9 (b)).

Production history data for the same period of production (14 years) is also presented in Fig.
5-10.
106
Fig. 5-10: Production rate for pressure draw down of 20MPa over a production period of 14 years after
60 weeks stimulation for Pinj = 51.7MPa and Pprod=31.7MPa (Gholizadeh Doonechaly et al. 2012).

From production history data (as presented in Fig. 5-10) it is evident that the production
rate increased from 2L/s after one month of production to 34 L/s linearly after 3 years.
After this time production rate increases monotonically and reaches a value of 45L/s after
14 years. Such a gradual increase in the production rate can be attributed to the creation of
the secondary flow paths caused by stimulation due to poro-thermo-elasticity. Average
matrix temperature drawdown and produced fluid temperature for the same period of
production (14 years) are presented in Fig. 5-11. As shown in the figure the heat transfer
from matrix to fluid and the resulting thermal drawdown is low at the early stage of
production because of the low fluid and rock matrix contact area. With the pass of time the
fluid sweeps over a large part of the reservoir which increases thermal drawdown. At the
end of the 14 years of production the average matrix temperature drops from 200°C to
140°C which is quite low (drop of 60°C) compared to a previous study by Koh and
Rahman et al. (2011) (see Fig. 5-11) under the same reservoir conditions.
107
 Fig. 5-11: Average matrix temperature and produced fluid temperature vs. time for 14 years of production

Also the results of this study are compared with the results generated by Heidinger et al.
(2006) (see Fig. 5-11). Heidinger et al. (2006) developed a “cost benefit analysis program”
to model the economic aspects of a geothermal reservoir based on their hydrogeological
model. The authors used an analytical model to simulate pressure loss. Fluid flow mainly
occurs inside the interconnected fractures systems. They assumed constant injection flow
rate of 100L/s and the flow test was carried out for 20 years (Heidinger et al. 2006). They
have also simulated hot water production from the Soultz reservoir for 20 years. Reservoir
temperature is set at 200°C at the depth of 4000m. Authors have shown much smaller
produced fluid temperature drop (40°C). The produced fluid temperature profile of this
study matches well with Heidinger et al. (2006). Pressure profile observed by Heidinger et
al. (2006) is characterised by a gradual drop from 200°C to 160°C. This gradual drop is due
to an increased injection rate (100L/s) which is greater than that in this study (average of
40L/s).

Matrix temperature drawdown for 1 year, 3 years, 10 years and 14 years of production are
also presented in Fig. 5-12 (a)-(d) respectively.

108
 200

190

180

170

160

150
(a) (b)

200

190

180

170

160

150
(c) (d)

Fig. 5-12: Reservoir temperature profile after (a) 1 year (b) 3 years, (c) 10 years and (d) 14 years of
production for σH = 51.7MPa and σh = 44.8MPa, Pinj = 51.7MPa and Pprod=31.7MPa..

As shown in the figures, fractures provide the main flow conduits from which it can be inferred
that the heat transfer between the fluid and the matrix is mainly dominated by the convection
because of the high fluid velocity inside the fractures (see Fig. 5-9 (a)). Once the thermal
breakthrough takes place at about 10 years (see Fig. 5-11) the matrix temperature begins to
drop rapidly which reduces the produced fluid temperature. This is also evident in Fig. 5-12
(d) in which cold water swept through most part of the reservoir. Also the x- and y
components of effective stress distribution of the Soultz geothermal reservoir during the

109
same stages of production with and without considering the thermal stresses are shown in
Fig. 5-13 and Fig. 5-14 respectively. As shown in Fig. 5-13 (d) by the end of 14 years of
production the effective stresses throughout the reservoir are significantly reduced, thus
allowing most fractures to open and conduct fluid. The reduction in the effective stresses is
caused by the cold fluid circulation as well as thermal drawdown.

Results also confirms that the thermal stress due to cooling of the reservoir has a significant
bearing in reducing the effective stresses which in turns increases the fracture aperture and
therefore the thermal drawdown of the reservoir (see Fig. 5-13). This can be clearly observed
when the results of effective stress presented in Fig. 5-13 are compared with that presented in
Fig. 5-14, in which thermal stress is ignored. At the end of 14 years of fluid circulation the
effective stress in major parts of the reservoir domain is much lower with thermal effect (see
Fig. 5-13 (d)) than that without thermal effect (see Fig. 5-14 (d)).

Sensitivity Analysis

After assessing the poro-thermo-elastic behaviour of the Soultz geothermal reservoir due
to the cold fluid circulation, a sensitivity analysis is carried out to in order to understand the
degree by which different reservoir parameters affect hot water production. In view of this
matrix drawdown temperature was estimated for minimum and maximum values of the
reservoir parameters which were obtained based on the field data. The parameters that have
been selected for the sensitivity analysis with their corresponding minimum and maximum
values are presented in Table 5-4 (Baria et al. 1992, Sausse et al. 2006).

110
 X-component Y-component

(a)

(b)

(c)

(d)

Fig. 5-13: x (left) and y (right) component of effective stress with thermal stress after (a) 1 year (b) 3
years (c) 10 years and (d) 14 years of production for σH = 51.7MPa and σh = 44.8MPa, Pinj = 51.7MPa and
Pprod=31.7MPa.

111
 X-component Y-component

(a)

(b)

(c)

(d)

Fig. 5-14: x (left) and y (right) component of effective stress without thermal stress after (a) 1 year (b) 3
years (c) 10 years and (d) 14 years of production for σH = 51.7MPa and σh = 44.8MPa, Pinj = 51.7MPa and
Pprod=31.7MPa.

112
 Table 5-4: Reservoir parameters used for the sensitivity analysis (Baria et al. 1992, Kolditz 1995, Sausse et al.
2006, Valley 2007, Watanabe et al. 2010)

Reservoir Parameter Current Study Min. Max.

Fracture Density 0.12 0.1 0.22

Fractal Dimension 1.2 1 2

Matrix Permeability 5.2 x 10-17 8.6 x 10-18 9.6 x 10-16

Fracture Aperture 0.06 0.01 0.25

(i) Fracture Density

Results of the effect of fracture density on the average matrix temperature drawdown are
shown in Fig. 5-15. It is evident from the figure that for about 2 years of production matrix
temperature remained unchanged. Following 2 years of production reservoir with high fracture
density shows a rapid decline in matrix temperature with time. Reservoir with low fracture
density, however, shows a gradual decrease in matrix temperature with time. Considering
Soultz reservoir as a base case scenario one can observe that a matrix temperature difference of
the two reservoirs (low fracture density and high fracture density) is about ±10°C after 6 years
of production. At about seven years of production matrix temperature draw down reaches its
highest value (10°C) when compared with that of the average value (0.12m -1, this study). From
then onward the matrix temperature- drawdown gradually levels out and reaches a plateau at
about 14 years of hot water production. Decrease in fracture density on the other hand retains
the matrix temperature quite high over much longer period (5 years). After this period matrix
temperature gradually drops and reaches a value of 168°C, which is about 18°C higher than this
study.

113
Fig. 5-15: Sensitivity analysis of the effect of uncertainty in fracture density on the average matrix temperature for
a production period of 14 years for σH = 51.7MPa and σh = 44.8MPa, Pinj = 51.7MPa and Pprod=31.7MPa.

(ii) Fractal Dimension

Results of two values of fractal dimension, one higher than this study, i.e. 2 and the other lower
than this study, i.e. 1 are shown in Fig. 5-16. As shown in the figure, the fractal dimension has
a significant bearing on the temperature draw down of the geothermal reservoirs. As observed
with fracture density, fractal dimension has very little effect on production of hot water in early
phase (2years). Matrix temperature begins to decline after 2 years of production. Reservoirs
with high fractal dimension show a rapid matrix temperature decline with time, while the
temperature drawdown of the reservoir with low fractal dimension remains near steady with
time. At about 14 years of production the matrix temperature of the reservoir with high fractal
dimension reaches a value of 125°C which is 25°C lower than this study (150°C).

114
Fig. 5-16: Sensitivity analysis of the effect of uncertainty in fractal dimension on the average matrix temperature
for a production period of 14 years for σH = 51.7MPa and σh = 44.8MPa, Pinj = 51.7MPa and Pprod=31.7MPa.

This means that the fractal dimension has much greater impact on matrix temperature
drawdown than the fracture density. As discussed in section 2.3, fractal dimension is a measure
of the irregularity in self similar objects, such as fractures. Therefore a higher value of the
fractal dimension increases the irregularity in the system of fracture network. Such an
irregularity increases the probability of fragmentation (Turcotte 1986). When compared with
the base case it is evident form Fig. 5-16 that the matrix temperature between the two
reservoirs, one with high fractal dimension and the other with low fractal dimension, varies
+30°C and -20°C.

(iii) Fracture Aperture

Results for the effect of fracture aperture on matrix temperature drawdown are shown in Fig.
5-17. As shown in the figure that the reservoirs with the maximum fracture aperture shows a
very rapid decline in matrix temperature from the early stage of hot water production (1 year)
until 8 years. As shown in Fig. 5-17 after 8 years of production the matrix temperature
decreases monotonically and reaches a value of 127°C after 14 years of hot water production.
Matrix temperature drawdown of the reservoir with low fracture aperture (lower than the base
case) remains near steady and reaches a value of 195°C after 14 years of production. This
means that the temperature drawdown of the two reservoirs can vary from the base case up to
about ±34°C at the end of 14 years of production. This phenomenon is further investigated by

115
analysing the amount of change in the average fracture aperture due to induced fluid pressure.
In this study the case of large average fracture aperture was subjected to fluid pressure and the
change in aperture was monitored with time and compared with the base case.

Fig. 5-17: Sensitivity analysis of the effect of uncertainty in fracture aperture on the average matrix temperature
for a production period of 14 years for σH = 51.7MPa and σh = 44.8MPa, Pinj = 51.7MPa and Pprod=31.7MPa.

The results of this study (Fig. 5-18) show that the change in fracture aperture with time is
similar to that observed in the base case study except that 40% increase in aperture can be
obtained which is 14% lower than the base case. As discussed earlier (Fig. 5-2) for every set of
reservoir (in particular fracture aperture) and stress parameters an optimum level of shear
dilation can be achieved.

(iv) Matrix Permeability

Results of the effect of matrix permeability of reservoirs (one with higher matrix
permeability than the base case and the other lower permeability than the base case) are
presented in Fig. 5-19. From the figure it can be seen that the reservoir with high matrix
permeability exhibits rapid cooling with time. This is attributed to the fact that an increase in
conductivity allows more fluid to sweep through the reservoir thus cooling the matrix more
rapidly. This behaviour is very similar to that observed in Fig. 5-17 (increase in fracture
aperture increases the cooling of the matrix). As expected reservoirs with low matrix
permeability remain almost unchanged until about 8 years of production. After this time matrix
temperature cools down gradually.
116
Fig. 5-18: Sensitivity analysis of the effect of uncertainty in initial fracture aperture on the percentage of increase
in average fracture aperture for the stimulation period of 60 weeks for σH = 78.9MPa and σh = 53.3MPa, Pinj =
65MPa.

Fig. 5-19: Sensitivity analysis of the effect of uncertainty in rock permeability on the average matrix temperature
for a production period of 14 years for σH = 51.7MPa and σh = 44.8MPa, Pinj = 51.7MPa and Pprod=31.7MPa.

117
5.3 Closure

In this chapter, a roughness induced shear displacement model in a poro-thermo-elastic

environment combined with an advanced technique is used to study the poro-thermo-elastic
effects due to induced fluid pressure. Because the shear dilation model is based on
analytical solution it was possible to integrate the model within the fluid flow and heat
transfer models. The poro-thermo-elastic model was applied to the Soultz geothermal
reservoir, France and the effect of water circulation on matrix temperature drawdown and
stress changes over longer term was investigated. It has also been shown that surface
roughness induced shear displacement provides a more realistic prediction of residual
fracture aperture. These results agree well with the experience of existing EGS trials around
the world. An average increase in aperture due to fluid induced shear dilation has been
found to be lower and time required to obtain a maximum stimulated volume is greater.
Results of this study are consistent with that of previous studies: for every geothermal
system there exists an optimum injection schedule (injection pressure and duration). Any
further increases in stimulation effort, i.e. stimulation time for a given stimulation pressure,
does not provide additional permeability enhancement. Also a sensitivity analysis was
carried out to study the effect of different parameters on the reservoir temperature
drawdown. Based on the results of this study it was found that the fracture aperture has a
significant impact on the reservoir temperature drawdown. Fractal dimension is the next
important parameter.

118
Chapter 6

6 CONCLUSIONS AND
RECOMMENDATIONS
6.1 Conclusions

In this study, a roughness induced shear displacement model in a poro-thermo-elastic

environment is used to study the effects of induced fluid pressure and thermal stresses
(cooling effect) on reservoir permeability and consequent increase in hot water production
from naturally fractured geothermal reservoirs.

In the first part of this study a discrete fracture network simulation procedure is
presented by using a hybrid of numerical simulation of tectonic events and stochastic
simulation of field data. Scale error due to lack of sufficient field data was overcome by
integrating a 3D spatial probability of occurrence of fractures in all parts of the reservoir
which was generated by numerical simulation of tectonic events.

Next a combination of single continuum and discrete fracture approach was used to
simulate the fluid flow and heat transfer inside the fractured domain in a poro-thermo-
elastic framework. The coupled process was numerically simulated by using FEM. Also the
Local Thermal Non-Equilibrium concept was used for the heat transfer formulation to
consider the thermal gradient at the matrix/fluid interface.

In the third part of this study, an analytical technique is presented to simulate the
hydraulic stimulation of the fractured reservoirs caused by the induced fluid pressure.
Because the model is based on an analytical solution it was possible to integrate the shear
displacement model with the fluid flow and heat transfer model (with less computation
effort) to predict stress changes due to induced fluid pressure and thermal drawdown of
matrix.

119
 The integrated methodology was applied to the Soultz geothermal reservoir, France, and the
effect of induced fluid pressure as well as water circulation on permeability changes of the
reservoir were investigated. From the results of this study, the following conclusions were
drawn:

- Simulation of tectonic events has helped to generate probability of occurrence of

fracture throughout the structure. These data were valuable for generating a detailed
field-wide fracture map. Neural network provided a continuum map of the fracture
density and fractal dimension in both wellbore scale and reservoir scale. Further, Back
Propagation Neural Network (BPNN) helped to establish a complex relationship
between the fracture density and fractal dimension and all available field data.
- The integrated roughness induced shear displacement model provided an efficient
framework to investigate the effect of different phenomena, such as water flow rate,
thermal stresses, Local Thermal Non-Equilibrium (LTNE) heat transfer and
convection dominated heat transfer on thermal drawdown of Soultz geothermal
reservoir. These effects are briefly discussed below.
(a) Three distinct phases of reservoir stimulation were observed: (i) A gradual
increase of 14% in average fracture aperture during the 40 weeks of injection,
(ii) a sharp increase in average fracture aperture reaching to about 45% in next
ten weeks (40th – 50th week) and (iii) a gradual increase reaching a plateau of
54% after about 60 weeks of stimulation. These data agree well with the shear
dilation events observed during stimulation of Soultz reservoir.
(b) Thermal induced stresses caused an average increase in tensile stresses by 10%
within the reservoir after 14 years of hot water production. This resulted in an
average increase in permeability by 30%, which in turn increased the production
rate from 35L/s to 45L/s for constant pressure drawdown between injection and
production wells of 20MPa for 14 years of circulation. During this period of
time, the produced fluid temperature decreased from 200°C to 140°C and the
average matrix temperature dropped from 200°C to 155°C.
(c) When results of Local Thermal Non-Equilibrium (LTNE) compared with that of
Local Thermal Equilibrium (LTE), it was observed that the average matrix

120
 temperature draws 10°C more down for LTNE than that for LTE after 14 years
of water circulation.
(d) It was also observed that the heat transfer in the reservoir, in particular fractures,
was mainly dominated by convection which was up to 3 orders of magnitude
more than that of conduction. This increase in heat transfer is attributed to the
high fluid velocity inside the fractures.

6.2 Scope for Future Research

The following recommendations are suggested for further studies:

1- The heat transfer model used in this study is based on the heat transfer coefficient
value obtained from the literature. However, further investigations, both numerical
and experimental, is required to obtain more accurate estimates of the heat transfer
coefficient. Also the spatial variation of the heat transfer coefficient in a reservoir
scale problem need to be considered.
2- The mechanism of brittle-ductile to ductile fracturing caused by the rock creep
(Regenauer-Lieb et al. 2009, Weinberg and Regenauer-Lieb 2010) need to be
integrated with the presented subsurface fracture map generator in order to consider
the effect of growth of microscale voids in creating interconnected fracture systems.
3- Chemical dissolution/precipitation of the minerals inside the fractures in geothermal
reservoirs can also be considered to have a more accurate estimate of the change in
permeability of the fractures during the cold fluid circulation in geothermal
reservoirs.
4- The methodology used in this study to simulate fluid flow and heat transfer in
naturally fractured reservoirs is in a 2-D framework which ignores the spatial
orientation of the fracture system. Therefore 3-D fluid flow and heat transfer model
can help to have a better understanding of the effect of spatial orientation of
fractures.

121
References
Aguilera, R. (1980). Naturally fractured reservoirs, Petroleum Pub. Co.
Ameen, M. S. and G. S. o. London (2003). Fracture and In-situ Stress Characterization of
Hydrocarbon Reservoirs, Geological Society.
Arbogast, T., J. Douglas, J. and U. Hornung (1990). "Derivation of the Double Porosity Model of
Single Phase Flow via Homogenization Theory." SIAM Journal on Mathematical Analysis
21(4): 823-836.
Asquith, G. B. (1985). Handbook of log evaluation techniques for carbonate reservoirs, American
Association of Petroleum Geologists.
Assaad, F. A., P. E. LaMoreaux and T. Hughes (2004). Field Methods for Geologists and
Hydrogeologists, Springer.
Babadagli, T. (2001). "Scaling of Cocurrent and Countercurrent Capillary Imbibition for Surfactant
and Polymer Injection in Naturally Fractured Reservoirs." SPE Journal(4): 465-478.
Bahar, A. H. A. K. a. A., Inc; Maged H Al-Deeb; Salem E Salem; Hussein Badaam, ADCO; Mohan
Kelkar, The University of Tulsa (2003). "Practical Approach in Modeling Naturally Fractured
Reservoir: A Field Case Study."
Bai, M. (1999). "On equivalence of dual-porosity poroelastic parameters." Journal of Geophysical
Research: Solid Earth 104(B5): 10461-10466.
Baria, R., J. Baumgärtner, A. Gerard and P. Moore (1992). {GPK1 - Preliminary results obtained
during the drilling operations at Soultz}.
Baria, R., S. Michelet, J. Baumgärtner, B. Dyer, J. Nicholls, T. Hettkamp, D. Teza, N. Soma, H.
Asanuma and J. Garnish (2005). "Creation and Mapping of 5000 m deep HDR/HFR Reservoir to
Produce Electricity." Proceedings, Paper 1627. pdf.
Barton, N., S. Bandis and K. Bakhtar (1985). "Strength, deformation and conductivity coupling of
rock joints." International Journal of Rock Mechanics and Mining Sciences & Geomechanics
Abstracts 22(3): 121-140.
Barton, N. and V. Choubey (1977). "The shear strength of rock joints in theory and practice." Rock
mechanics 10(1-2): 1-54.
Bathe, K. J. (1996). Finite element procedures, Prentice Hall.
Beckham, W. E. (1996). "Seismic anisotropy and natural fractures from VSP and borehole sonic
tools---A field study." Geophysics 61(2): 456-466.
Beckner, B. L., K. Ishimoto, S. Yamaguchi, A. Firoozabadi and K. Aziz (1987). Imbibition-
Dominated Matrix-Fracture Fluid Transfer in Dual Porosity Simulators. SPE Annual Technical
Conference and Exhibition. Dallas, Texas, 1987 Copyright 1987, Society of Petroleum
Engineers.
Beer, G. and J. O. Watson (1992). Introduction to finite and boundary element methods for
engineers, Wiley.
Berry, M. D. (1991). "Palm Valley No.7, Core Analysis, Summary and Interpretation of Results."
Magellan Petroleum Australia Limited (unpublished).
Berry, M. D., D. W. Stearns and M. Friedman (1996). "The Development of a Fractured Reservoir
Model for the Palm Valley Gas Field." Journal of the Australian Petroleum Production and
Exploration Association Ltd: 82-103.
Berryman, J. (2002). "Extension of Poroelastic Analysis to Double-Porosity Materials: New
Technique in Microgeomechanics." Journal of Engineering Mechanics 128(8): 840-847.
Biot, M. A. (1961). "Theory of Folding of Stratified Viscoelastic Media and Its Implications in
Tectonics and Orogenesis." Geological Society of America Bulletin 72(11): 1595-1620.
Birch, F., J. F. Schairer and H. C. Spicer (1942). Handbook of physical constants, The Society.

122
Blumenthal, M., M. Kuhn, H. Pape, V. Rath and C. Clauser (2007). "Hydraulic model of the deep
reservoir quantifying the multi-well tracer test. ." EHDRA Scientific Conference, Soultz-sous-
Forets.
Bogatkov, D. S. B., Tayfun , SPE, University of Alberta (2008). "Integrated Modeling and
Statistical Analysis of 3-D Fracture Network of the Midale Field."
Borg, I. and J. Handin (1966). "Experimental deformation of crystalline rocks." Tectonophysics
3(4): 249-367.
Borja, R. I. (2006). "On the mechanical energy and effective stress in saturated and unsaturated
porous continua." International Journal of Solids and Structures 43(6): 1764-1786.
Bourne, S., L. Rijkels, B. Stephenson and E. Willemse (2000). "Predictive modelling of naturally
fractured reservoirs using geomechanics and flow simulation." GeoArabia (Manama) 6(1): 27-
42.
Bourne, S. J., F. Brauckmann, L. Rijkels, B. J. Stephenson, A. Weber and E. J. M. Willemse
(2000). Predictive modelling of naturally fractured reservoirs using geomechanics and flow
simulation. Abu Dhabi Interntional Petroelum Exhibition and Conference. Abu Dhabi, United
Arab Emirates.
Bourne, S. J. and E. J. M. Willemse (2001). "Elastic stress control on the pattern of tensile
fracturing around a small fault network at Nash Point, UK." Journal of Structural Geology
23(11): 1753-1770.
Brown, S. R. (1987). "Fluid Flow Through Rock Joints: The Effect of Surface Roughness." J.
Geophys. Res. 92(B2): 1337-1347.
Bunge, R. J. (2000). Midale Reservoir Fracture Characterization Using Integrated Well and Seismic
Data, Weyburn Field, Saskatchewan, Colorado School of Mines.
Buzzi, O., J. Hans, M. Boulon, F. Deleruyelle and F. Besnus (2007). "Hydromechanical study of
rock–mortar interfaces." Physics and Chemistry of the Earth, Parts A/B/C 32(8–14): 820-831.
Carlson, R. (2003). Practical Reservoir Simulation: Using, Assessing, and Developing Results,
PennWell.
Chen, Z., S. P. Narayan, Z. Yang and S. S. Rahman (2000). "An experimental investigation of
hydraulic behaviour of fractures and joints in granitic rock." International Journal of Rock
Mechanics and Mining Sciences 37(7): 1061-1071.
Chiaramonte, L. and S. University (2009). Geomechanical Characterization and Reservoir
Simulation of a Carbon Dioxide Sequestration Project in a Mature Oil Field, Teapot Dome, WY,
Stanford University.
Chilès, J.-P. (2005). Stochastic Modeling of Natural Fractured Media: A Review
Geostatistics Banff 2004. O. Leuangthong and C. V. Deutsch, Springer Netherlands. 14: 285-294.
Choi, E. S., T. Cheema and M. R. Islam (1997). "A new dual-porosity/dual-permeability model
with non-Darcian flow through fractures." Journal of Petroleum Science and Engineering 17(3–
4): 331-344.
Clark, R. M., S. J. D. Cox and G. M. Laslett (1999). "Generalizations of power-law distributions
applicable to sampled fault-trace lengths: model choice, parameter estimation and caveats."
Geophysical Journal International 136(2): 357-372.
Cosgrove, J. W. (1999). "Forced folds and fractures: An introduction." Geological Society, London,
Special Publications 169(1): 1-6.
Courant, R. (1943). "Variational methods for the solution of problems of equilibrium and
vibrations." Bull. Amer. Math. Soc. 49: 1--23.
Cuvelier, C., A. Segal and A. A. van Steenhoven (1986). Finite Element Methods and Navier-
Stokes Equations, Springer.
Da Prat, G. (1990). Well Test Analysis for Fractured Reservoir Evaluation, Elsevier Science.
Darling, T. (2005). Well Logging and Formation Evaluation, Elsevier Science.
123
De Borst, R., M. Crisfield, J. Remmers and C. Verhoosel (2012). Nonlinear Finite Element
Analysis of Solids and Structures, Wiley.
Detournay, E. and A. Cheng (1993). Fundamentals of poroelasticity, Comprehensive Rock
Engineering, Vol 2: Analysis and Design Methods, Eds. Brown, ET, Fairhurst, CH and Hoek, E,
Pergamon.
Do Rozario, R. F. (1991). "The Palm Valley Gas Field, Amadeus Basin, central Australia." Bureau
of Mineral Resources Bulletin 236: 477-492.
Do Rozario, R. F. (1991). "The Palm Valley Gas Field, Amadeus Basin, Central Australia. In:
Korsch, R.J. and Kennard, J.M. (eds.), Geological and Geophysical Studies in the Amadeus
Basin, Central Australia." BMR Bulletin 236: 477-492.
Doonechaly, N. G., S. S. Rahman and A. Kotousov (2013). "A New Approach to Hydraulic
Stimulation of Geothermal Reservoirs by Roughness Induced Fracture Opening."
Ellis, D. V. and J. M. Singer (2007). Well Logging for Earth Scientists, Springer.
Evans, K. F., H. Moriya, H. Niitsuma, R. H. Jones, W. S. Phillips, A. Genter, J. Sausse, R. Jung and
R. Baria (2005). "Microseismicity and permeability enhancement of hydrogeologic structures
during massive fluid injections into granite at 3 km depth at the Soultz HDR site." Geophysical
Journal International 160(1): 389-412.
Fahad, M. (2013). SIMULATION OF FLUID FLOW AND ESTIMATION OF PRODUCTION
FROM NATURALLY FRACTURED RESERVOIRS, The University of New South Wales.
Fahad, M., S. S. Rahman and Y. Cinar (2011). "A Numerical and Experimental Procedure to
Estimate Grid Based Effective Permeability Tensor for Geothermal Reservoirs." Geothermal
Resources Council Transactions.
Fanchi, J. (2010). Integrated Reservoir Asset Management: Principles and Best Practices, Elsevier
Science.
Fisher, N. I. (1995). Statistical analysis of circular data, Cambridge University Press.
Fisher, N. I., T. Lewis and B. J. J. Embleton (1993). Statistical Analysis of Spherical Data,
Cambridge University Press.
FitzGerald, E. M., C. J. Bean and R. Reilly (1999). "Fracture-frequency prediction from borehole
wireline logs using artificial neural networks." Geophysical Prospecting 47(6): 1031-1044.
Fjar, E., R. M. Holt, A. M. Raaen, R. Risnes and P. Horsrud (2008). Petroleum Related Rock
Mechanics: 2nd Edition, Elsevier Science.
Fletcher, R. C. (1991). "Three-dimensional folding of an embedded viscous layer in pure shear."
Journal of Structural Geology 13(1): 87-96.
Fortin, M. (1981). "Old and new finite elements for incompressible flows." International Journal for
Numerical Methods in Fluids 1(4): 347-364.
Fu, P., S. M. Johnson and C. R. Carrigan (2011). "Investigation of Stimulation-Response
Relationships for Complex Fracture Systems in Enhanced Geothermal Reservoirs." GRC
Transactions 35.
Gangi, A. F. (1978). "Variation of whole and fractured porous rock permeability with confining
pressure." International Journal of Rock Mechanics and Mining Sciences & Geomechanics
Abstracts 15(5): 249-257.
Gartner, J. and J. Suau (1980). "Fracture Detection From Well Logs." The Log Analyst XXI(2).
Gelet, R., B. Loret and N. Khalili (2012). "A thermo-hydro-mechanical coupled model in local
thermal non-equilibrium for fractured HDR reservoir with double porosity." Journal of
Geophysical Research: Solid Earth 117(B7): B07205.
Genter, A. and H. Traineau (1996). "Analysis of macroscopic fractures in granite in the HDR
geothermal well EPS-1, Soultz-sous-Foreˆts, France." Journal of Volcanology and Geothermal
Research 72(1–2): 121-141.

124
Gentier, S., X. Rachez, T. D. T. Ngoc, M. Peter-Borie and C. Souque (2010). "3D Flow Modelling
of the Medium-Term Circulation Test Performed in the Deep Geothermal Site of Soultz-Sous-
forêts (France)." Proceedings World Geothermal Congress, Bali, Indonesia, 25-29 April.
Ghassemi, A. and S. Tarasovs (2006). FRACTURE SLIP AND OPENING IN RESPONSE TO
FLUID INJECTION INTO A GEOTHERMAL RESERVOIR.
Ghassemi, A., S. Tarasovs and A.-D. Cheng (2007). "A 3-D study of the effects of
thermomechanical loads on fracture slip in enhanced geothermal reservoirs." International
Journal of Rock Mechanics and Mining Sciences 44(8): 1132-1148.
Ghassemi, A., S. Tarasovs and A. H. D. Cheng (2007). "A 3-D study of the effects of
thermomechanical loads on fracture slip in enhanced geothermal reservoirs." International
Journal of Rock Mechanics and Mining Sciences 44(8): 1132-1148.
Ghassemi, A. and X. Zhou (2011). "A three-dimensional thermo-poroelastic model for fracture
response to injection/extraction in enhanced geothermal systems." Geothermics 40(1): 39-49.
Gholizadeh Doonechaly, N. and S. S. Rahman (2012). "3D hybrid tectono-stochastic modeling of
naturally fractured reservoir: Application of finite element method and stochastic simulation
technique." Tectonophysics 541–543(0): 43-56.
Gholizadeh Doonechaly, N., S. S. Rahman and A. Kotousov (2012). "An Innovative Stimulation
Technology for Permeability Enhancement in Enhanced Geothermal System--Fully Coupled
Thermo-Poroelastic Numerical Approach." 36th Geothermal Resources Council Transactions.
Gholizadeh Doonechaly, N., S. S. Rahman and A. Kotousov (2012). "A Realistic Assessment of
Recoverable Thermal Energy from Australian Geothermal Reservoirs: A Simulation Study."
Proceedings of the 2012 Australian Geothermal Energy Conference
Gong, B., M. Karimi-Fard and L. J. Durlofsky (2008). "Upscaling Discrete Fracture
Characterizations to Dual-Porosity, Dual-Permeability Models for Efficient Simulation of Flow
With Strong Gravitational Effects." SPE Journal 13(1): pp. 58-67.
Gray, D. and K. Head (2000). "Fracture detection in the Manderson Field: A 3D AVAZ case
history." SEG Technical Program Expanded Abstracts 19(1): 1413-1416.
Griggs, D. (1939). "Creep of Rocks." The Journal of Geology 47(3): 225-251.
Gringarten, E. (1998). "FRACNET: Stochastic simulation of fractures in layered systems."
Computers & Geosciences 24(8): 729-736.
Groshong, R. H. (1999). 3-D Structural Geology: A Practical Guide to Surface and Subsurface Map
Interpretation, Springer.
Harstad, H., L. W. Teufel and J. C. Lorenz (1995). Characterization and Simulation of Naturally
Fractured Tight Gas Sandstone Reservoirs. SPE Annual Technical Conference and Exhibition.
Dallas, Texas, 1995 Copyright 1995, Society of Petroleum Engineers, Inc.
Hast, N. (1979). "Limits of stress measurements in the Earth's crust." Rock mechanics 11(3): 143-
150.
Haupt, P. (2000). Continuum mechanics and theory of materials, Springer.
He, N., S. H. Lee and C. L. Jensen (2001). Combination of Analytical, Numerical and Geostatistical
Methods to Model Naturally Fractured Reservoirs. SPE Western Regional Meeting. Bakersfield,
California, Copyright 2001, Society of Petroleum Engineers Inc.
Heidinger, P., J. Dornstädter and A. Fabritius (2006). "HDR economic modelling: HDRec
software." Geothermics 35(5–6): 683-710.
Helmig, R., A. Mielke and B. I. Wohlmuth (2006). Multifield Problems in Solid and Fluid
Mechanics, Springer-Verlag.
Hossain, M. M. (2001). Reservoir Stimulation by Hydraulic Fracturing: Complexities and
Remedies with Reference to Initiation and Propagation of Induced and Natural Fractures,
University of New South Wales.

125
Hossain, M. M., M. Rahman and S. Rahman (2002). "A shear dilation stimulation model for
production enhancement from naturally fractured reservoirs." Spe Journal 7(2): 183-195.
Hossain, M. M., M. K. Rahman and S. S. Rahman (2002). "A Shear Dilation Stimulation Model for
Production Enhancement From Naturally Fractured Reservoirs." SPE Journal 7(2): 183-195.
Hughes, T. J. R. (2000). The Finite Element Method: Linear Static and Dynamic Finite Element
Analysis, Dover Publications.
Iverson, W. P. (1992). Fracture Identification From Well Logs. SPE Rocky Mountain Regional
Meeting. Casper, Wyoming, 1992.
Jaeger, J. C., N. G. W. Cook and R. Zimmerman (2009). Fundamentals of Rock Mechanics, Wiley.
Jäger, P., S. M. Schmalholz, D. W. Schmid and E. Kuhl (2008). "Brittle fracture during folding of
rocks: A finite element study." Philosophical Magazine 88(28-29): 3245-3263.
Jensen, J., L. W. Lake, P. W. M. D. Corbett and Goggin (2010). "STATISTICS FOR
PETROLEUM ENGINEERS AND GEOSCIENTISTS." 2ND EDITION.
Jing, L. (2003). "A review of techniques, advances and outstanding issues in numerical modelling
for rock mechanics and rock engineering." International Journal of Rock Mechanics and Mining
Sciences 40(3): 283-353.
Johnson, A. M. (1970). Physical processes in geology: a method for interpretation of natural
phenomena; intrusions in igneous rocks, fractures, and folds, flow of debris and ice, Freeman,
Cooper.
Johnson, D. E. and K. E. Pile (2002). Well Logging in Nontechnical Language, PennWell Pub.
Josnin, J.-Y., H. Jourde, P. Fénart and P. Bidaux (2002). "A three-dimensional model to simulate
joint networks in layered rocks." Canadian Journal of Earth Sciences 39(10): 1443-1455.
Jupe, A. J., D. Bruel, T. Hicks, R. Hopkirk, O. Kappelmeyer, T. Kohl, O. Kolditz, N. Rodrigues, K.
Smolka, J. Willis-Richards, T. Wallroth and S. Xu (1995). "Modelling of a european prototype
hdr reservoir." Geothermics 24(3): 403-419.
Kadkhodaie-Ilkhchi, A., H. Rahimpour-Bonab and M. Rezaee (2009). "A committee machine with
intelligent systems for estimation of total organic carbon content from petrophysical data: An
example from Kangan and Dalan reservoirs in South Pars Gas Field, Iran." Computers &
Geosciences 35(3): 459-474.
Kaus, B. J. P. and Y. Y. Podladchikov (2001). "Forward and reverse modeling of the
three&#8208;dimensional viscous Rayleigh&#8208;Taylor instability." Geophys. Res. Lett.
28(6): 1095-1098.
Kaviany, M. (2008). Heat transfer physics, Cambridge University Press Cambridge, UK.
Kazemi, H., J. R. Gilman and A. M. Elsharkawy (1992). "Analytical and Numerical Solution of Oil
Recovery From Fractured Reservoirs With Empirical Transfer Functions (includes associated
papers 25528 and 25818)." SPE Reservoir Engineering(05).
Kim, J., E. L. Sonnenthal and J. Rutqvist (2012). "Formulation and sequential numerical algorithms
of coupled fluid/heat flow and geomechanics for multiple porosity materials." International
Journal for Numerical Methods in Engineering 92(5): 425-456.
Kocher, T. and N. S. Mancktelow (2005). "Dynamic reverse modelling of flanking structures: a
source of quantitative kinematic information." Journal of Structural Geology 27(8): 1346-1354.
Koh, J., N. Gholizadeh Doonechaly, A. R. Shaik, S. S. Rahman and L. Mortimer (2010). "Reservoir
characterisation and numerical modelling to reduce project risk and maximise the chance of
success. An example of how to design a stimulation program and assess fluid production from
the Soultz geothermal field, France." Australian Geothermal Conference.
Koh, J., H. Roshan and S. S. Rahman (2011). "A numerical study on the long term thermo-
poroelastic effects of cold water injection into naturally fractured geothermal reservoirs."
Computers and Geotechnics 38(5): 669-682.

126
Kohl, T. and T. Megel (2007). "Predictive modeling of reservoir response to hydraulic stimulations
at the European EGS site Soultz-sous-Forets." International journal of rock mechanics and
mining sciences 44(8): 1118-1131.
Kolditz, O. (1995). "Modelling flow and heat transfer in fractured rocks: Conceptual model of a 3-
D deterministic fracture network." Geothermics 24(3): 451-470.
Kotousov, A., L. Bortolan Neto and S. Rahman (2011). "Theoretical model for roughness induced
opening of cracks subjected to compression and shear loading." International Journal of Fracture
172(1): 9-18.
Kouider El Ouahed, A., D. Tiab, A. Mazouzi and S. .Jokhio (2003). Application of Artificial
Intelligence to Characterize Naturally Fractured Reservoirs. SPE International Improved Oil
Recovery Conference in Asia Pacific. Kuala Lumpur, Malaysia, Society of Petroleum Engineers.
Kuwahara, Y., H. Ito and T. Kiguchi (1991). "Comparison between natural fractures and fracture
parameters derived from VSP." Geophysical Journal International 107(3): 475-483.
Lan, L. and P. J. Hudleston (1991). "Finite-element models of buckle folds in non-linear materials."
Tectonophysics 199(1): 1-12.
Lee, W. B. (1993). Advances in Engineering Plasticity and its Applications, Elsevier Science.
Lei, W. and B. Chen (1995). "Fractal characterization of some fracture phenomena." Engineering
fracture mechanics 50(2): 149-155.
Li, X.-Y. (1997). "Fractured reservoir delineation using multicomponent seismic data." Geophysical
Prospecting 45(1): 39-64.
Lockner, D. A., J. B. Walsh and J. D. Byerlee (1977). "Changes in seismic velocity and attenuation
during deformation of granite." Journal of Geophysical Research 82(33): 5374-5378.
Lonergan, L. (2007). Fractured Reservoirs, Geological Society.
Long, J. C. S., J. S. Remer, C. R. Wilson and P. A. Witherspoon (1982). "Porous media equivalents
for networks of discontinuous fractures." Water Resources Research 18(3): 645-658.
Lough, M. F., S. H. Lee and J. Kamath (1997). "A New Method To Calculate Effective
Permeability of Gridblocks Used in the Simulation of Naturally Fractured Reservoirs." SPE
Reservoir Engineering 12(3): 219-224.
Lu, W. and Y. Y. Xiang (2012). "Analysis of the Instantaneous Local Thermal Equilibrium
Assumption for Heat Exchange between Rock Matrix and Fracture Water." Advanced Materials
Research 594-597: 2430-2437.
Luthi, S. (2001). Geological Well Logs: Their Use in Reservoir Modeling, Springer.
Lynn, H. B., R. Bates, M. Layman and M. Jones (1995). Natural Fracture Characterization Using P-
Wave Reflection Seismic Data, VSP, Borehole Imaging Logs, and the In-Situ Stress Field
Determination. Low Permeability Reservoirs Symposium. Denver, Colorado, 1995 Copyright
1995, Society of Petroleum Engineers, Inc.
Maerten, L., P. Gillespie and D. D. Pollard (2002). "Effects of local stress perturbation on
secondary fault development." Journal of Structural Geology 24(1): 145-153.
Mandelbrot, B. B. (1982). The Fractal Geometry of Nature, Henry Holt and Company.
Mandelbrot, B. B. (1983). "The fractal geometry of nature/Revised and enlarged edition." New
York, WH Freeman and Co., 1983, 495 p. 1.
Markner-Jäger, B. (2008). Plate Tectonics, Tectonics and Faultings. Technical English for
Geosciences. B. Markner-Jäger, Springer Berlin Heidelberg: 28-33.
Martinez-Torres, L. P. (2002). Characterization of Naturally Fractured Reservoirs from
Conventional Well Logs, University of Oklahoma.
Martinez-Torres, L. P. (2002). Characterization of naturally fractured reservoirs from conventional
well logs.
McClure, M. W. and R. N. Horne (2013). Discrete Fracture Network Modeling of Hydraulic
Stimulation: Coupling Flow and Geomechanics, Springer Science & Business.

127
Mildren, S., J. Burgess and J. Meyer (2005). "Multiple Application of Image Log Data to EOR
Operations in the Cooper Basin, Australia." SPE International Improved Oil Recovery
Conference in Asia Pacific.
Milne, N. A. and D. C. Barr (1990). "Subsurface fracture analysis, Palm Valley gas field." APPEA
Journal 30(1): 321-341.
MIT (2006). "he Future of Geothermal Energy." An Interdisciplinary MIT Study.
Mogi, K. (2007). Experimental Rock Mechanics, Taylor & Francis.
Mohebbi, A., M. Haghighi and M. Sahimi (2007). Using Conventional Logs for Fracture Detection
and Characterization in One of Iranian Field. International Petroleum Technology Conference.
Dubai, U.A.E., International Petroleum Technology Conference.
Moore, J. C. and D. E. Karig (1976). "Sedimentology, structural geology, and tectonics of the
Shikoku subduction zone, southwestern Japan." Geological Society of America Bulletin 87(9):
1259-1268.
Murphy, H. D. and M. Fehler (1986). Hydraulic fracturing of jointed formations. International
Meeting on Petroleum Engineering.
Nakagome, O., T. Uchida and T. Horikoshi (1998). "Seismic reflection and VSP in the kakkonda
geothermal field, japan: fractured reservoir characterization." Geothermics 27(5–6): 535-552.
Narayan, S. P., Z. Yang, S. S. Rahman and Z. Jing (1998). "Propant Free-Shear Dilation: An
Emerging Techology for Exploiting Tight to Ultra-Tight Gas Resources." SPE Annual Technical
Conference and Exhibition, 27-30 September 1998, New Orleans, Louisiana.
Nelson, P. P. and S. E. Laubach (1994). Rock mechanics models and measurements: challenges
from industry ; proceedings of the 1st North American Rock Mechanics Symposium, the
University of Texas at Austin, 1-3 June 1994, A.A. Balkema.
Nelson, R. A. (2001). Geologic analysis of naturally fractured reservoirs, Gulf Professional Pub.
Neves, F. A., A. Al-Marzoug, J. J. Kim and E. L. Nebrija (2003). "Fracture characterization of deep
tight gas sands using azimuthal velocity and AVO seismic data in Saudi Arabia." The Leading
Edge 22(5): 469-475.
Nicholson, R. (1994). "Structural Geology by R. J. Twiss and E. M. Moores. W. H. Freeman & Co.,
San Francisco, 1992. No. of pages: 532. Price: $47.95 (hardback). ISBN 0 7167 2252 6."
Geological Journal 29(4): 382-383.
O'Driscoll, E. S. (1962). "Experimental patterns in superimposed similar folding." J. Alberta Soc.
Pet. Geol. 10: 145-167.
Olson, J. E., P. H. Hennings and S. E. Laubach (1998). Integrating Wellbore Data and
Geomechanical Modeling for Effective Characterization of Naturally Fractured Reservoirs.
SPE/ISRM Rock Mechanics in Petroleum Engineering. Trondheim, Norway, 1998 Copyright
1998, Society of Petroleum Engineers Inc.
Olsson, W. A. and S. R. Brown (1993). "Hydromechanical response of a fracture undergoing
compression and shear." International Journal of Rock Mechanics and Mining Sciences &
Geomechanics Abstracts 30(7): 845-851.
Ouenes, A., G. Robinson, A. Zellou, D. Balogh and U. Araktingi (2005). Seismically Driven
Fractured Reservoir Characterization, World Petroleum Congress.
Parney, R. C. T. L. P., P; Dershowitz, W; Curran, B (2000). "Fracture and Production Data
Integration Using Discrete Fracture Network Models for Carbonate Reservoir Management,
South Oregon Basin Field." SPE60306, Society of Petroleum Engineers Rocky Mountain
Regional/Low Permeability Reservoirs Symposium and Exhibition, Colorado, USA,
Mar.12-15.
Pastor, M., M. Quecedo and O. C. Zienkiewicz (1997). "A mixed displacement-pressure
formulation for numerical analysis of plastic failure." Computers & Structures 62(1): 13-23.

128
Pengcheng, F. and C. R. Carrigan (2012). "Modeling Responses of Naturally Fractured Geothermal
Reservoir to Low-Pressure Stimulation." GRC Transactions 36.
Piggott, A. R. and D. Elsworth (1991). A Hydromechanical Representation of Rock Fractures, A.A.
Balkema. Permission to Distribute - American Rock Mechanics Association.
Pogacnik, J., P. Leary and P. Malin (2012). "Computational Framework for EGS Fracture
Stimulation." GRC Transactions 36.
Price, N. J. (1966). Fault and joint development in brittle and semi-brittle rock / by Neville J. Price.
Oxford ; New York :, Pergamon Press.
Pride, S. R. and J. G. Berryman (2003). "Linear dynamics of double-porosity dual-permeability
materials. I. Governing equations and acoustic attenuation." Physical Review E 68(3): 036603.
Rafiee, A. V., M; (2008). "Application of Geostatistical Characteristics of Rock Mass Fracture
Systems in 3D Model Generation." International Journal Of Rock Mechanics &Mining Sciences
45: 664-652.
Rahman, M. K., M. M. Hossain and S. S. Rahman (2000). "An analytical method for mixed-mode
propagation of pressurized fractures in remotely compressed rocks." International Journal of
Fracture 103(3): 243-258.
Rahman, M. K., M. M. Hossain and S. S. Rahman (2002). "A shear-dilation-based model for
evaluation of hydraulically stimulated naturally fractured reservoirs." International Journal for
Numerical and Analytical Methods in Geomechanics 26(5): 469-497.
Ramberg, H. (1963). "Fluid dynamics of viscous buckling applicable to folding of layered rocks."
AAPG Bulletin 47(3): 484-505.
Ramsay, J. G. (1983). The techniques of modern structural geology / John G. Ramsay, Martin I.
Huber. London ; New York :, Academic Press.
Ramsay, J. G. and M. I. Huber (1983). The Techniques of Modern Structural Geology: Strain
analysis, Academic Press.
Rasmussen, M. L. and F. Civan (2003). Full, Short-, and Long-Time Analytical Solutions for
Hindered Matrix-Fracture Transfer Models of Naturally Fractured Petroleum Reservoirs. SPE
Production and Operations Symposium. Oklahoma City, Oklahoma, Society of Petroleum
Engineers.
Rasmussen, T. C., J. Yeh and D. Evans (1987). "Effect of variable fracture permeability/matrix
permeability ratios on three-dimensional fractured rock hydraulic conductivity." Proceedings of
the Conference on Geostatistical, Sensitivity, and Uncertainty Methods for Ground-Water Flow
and Radionuclide Transport Modeling, San Francisco, California, September 1987, B. E.
Buxton, Batelle Press, Columbus, OH, 1987, 337.
Rawnsley, K. and P. Swaby (1996). "A new approach to fracture modelling in reservoirs using
deterministic, genetic and statistical models of fracture growth." Journal Name: AAPG Bulletin;
Journal Volume: 80; Journal Issue: 8; Conference: American Association of Petroleum
Geologists (AAPG) international conferences and exhibition, Caracas (Venezuela), 8-11 Sep
1996; Other Information: PBD: Aug 1996: Medium: X; Size: pp. 1327b-1328.
Rayner, B. L. (1995). "Palm Valley 10 and 10A FMS Interpretation Report." Schlumberger, for
Magellan Petroleum Australia Limited (unpublished): 762-774.
Reddy, J. N., R. J. Stein and Wickham. (1983). "Finite-element modelling of folding and faulting : J
S Int J Num Anal Meth Geomech, V6, N4, Oct-Dec 1982, P425-440." International Journal of
Rock Mechanics and Mining Sciences & Geomechanics Abstracts 20(2): A39-A39.
Regenauer-Lieb, K. (1998). "Dilatant plasticity applied to Alpine collision: ductile void growth in
the intraplate area beneath the Eifel volcanic field." Journal of Geodynamics 27(1): 1-21.
Regenauer-Lieb, K., T. Poulet, D. Siret, F. Fusseis, J. Liu, K. Gessner, O. Gaede, G. Morra, B.
Hobbs, A. Ord, H. Muhlhaus, D. Yuen, R. Weinberg and G. Rosenbaum (2009). First Steps

129
 Towards Modeling a Multi-Scale Earth System. Advances in Geocomputing, Springer Berlin
Heidelberg. 119: 1-25.
Regenauer-Lieb, K., R. F. Weinberg and G. Rosenbaum (2006). "The effect of energy feedbacks on
continental strength." Nature 442(7098): 67-70.
Richards, H. G., R. H. Parker, A. S. P. Green, R. H. Jones, J. D. M. Nicholls, D. A. C. Nicol, M. M.
Randall, S. Richards, R. C. Stewart and J. Willis-Richards (1994). "The performance and
characteristics of the experimental hot dry rock geothermal reservoir at Rosemanowes, Cornwall
(1985–1988)." Geothermics 23(2): 73-109.
Riley, M. (2004). "An Algorithm for Generating Rock Fracture Patterns: Mathematical Analysis."
Mathematical Geology 36(6): 683-702.
Robinson, B. A. and P. Kruger (1992). Pre-test estimates of temperature decline for the LANL
Fenton Hill Long-Term Flow Test, Los Alamos National Lab., NM (United States).
Roe, L. E. (1995). "Palm Valley Seismic Survey, Interpretation Report." Magellan Petroleum
Australia Limited (unpublished).
Sahimi, M. (1993). "Flow phenomena in rocks: from continuum models to fractals, percolation,
cellular automata, and simulated annealing." Reviews of modern physics 65(4): 1393.
Sahimi, M. (1995). "New Models for Natural and Hydraulic Fracturing of Heterogeneous Rock."
SPE Western Regional Meeting, 8-10 March 1995, Bakersfield, California.
Sahimi, M. (2000). "Fractal-wavelet neural-network approach to characterization and upscaling of
fractured reservoirs." Computers & Geosciences 26(8): 877-905.
Satter, A., G. M. Iqbal and J. L. Buchwalter (2008). Practical Enhanced Reservoir Engineering:
Assisted with Simulation Software, PennWell Corporation.
Sausse, J. (2002). "Hydromechanical properties and alteration of natural fracture surfaces in the
Soultz granite (Bas-Rhin, France)." Tectonophysics 348(1–3): 169-185.
Sausse, J., M. Fourar and A. Genter (2006). "Permeability and alteration within the Soultz granite
inferred from geophysical and flow log analysis." Geothermics 35(5–6): 544-560.
Schatzinger, R. A. and J. F. Jordan (1999). Reservoir Characterization: Recent Advances, American
Association of Petroleum Geologists.
Schmalholz, S. M. (2008). "3D numerical modeling of forward folding and reverse unfolding of a
viscous single-layer: Implications for the formation of folds and fold patterns." Tectonophysics
446(1-4): 31-41.
Serra, O. (1984). The Acquisition of Logging Data, Elsevier Science.
Serra, O. (2007). Well Logging and Reservoir Evaluation, Editions Technip.
Serra, O. (2008). The Well Logging Handbook, Atlasbooks Dist Serv.
Shaik, A. R., J. Koh, S. S. Rahman, M. A. Aghighi and N. H. Tran (2009). "Design and evaluation
of well placement and hydraulic stimulation for economical heat recovery from enhanced
geothermal systems." GRC Annual Meeting October 4–7, 2009, conference proceedings.
Shaik, A. R., S. S. Rahman, N. H. Tran and T. Tran (2011). "Numerical simulation of Fluid-Rock
coupling heat transfer in naturally fractured geothermal system." Applied Thermal Engineering
31(10): 1600-1606.
Shanks, R. T., B. S. Kwon, M. R. DeVries and P. A. Wichmann (1976). A REVIEW OF
FRACTURE DETECTION WITH WELL LOGS. SPE Annual Fall Technical Conference and
Exhibition. New Orleans, Louisiana, 1976 Copyright 1976.
Shen, F., X. Zhu and M. N. Toksöz (2002). "Effects of fractures on NMO velocities and P-wave
azimuthal AVO response." Geophysics 67(3): 711-726.
Siratovich, P. A., I. Sass, S. Homuth and A. Bjornsson (2011). "Thermal Stimulation of Geothermal
Reservoirs and Laboratory Investigation of Thermally Induced Fractures." GRC Transactions
35.

130
Sivakugan, N., S. K. Shukla and B. M. Das (2012). Rock Mechanics: An Introduction, Taylor &
Francis.
Smith, R. L. and J. P. McGarrity (2001). "Cracking the fractures---seismic anisotropy in an offshore
reservoir." The Leading Edge 20(1): 18-26.
Snow, D. T. (1968). "Rock fracture spacings, openings, and porosities." Journal of Soil Mechanics
& Foundations Div.
Solberg, P., D. Lockner and J. D. Byerlee (1980). "Hydraulic fracturing in granite under geothermal
conditions." International Journal of Rock Mechanics and Mining Sciences & Geomechanics
Abstracts 17(1): 25-33.
Somerton, W. H., I. C. o. R. Mechanics, I. C. f. R. Mechanics and B. C. o. E. University of
California (1970). Rock Mechanics: Theory and Practice ; Proceedings, Eleventh Symposium on
Rock Mechanics, Held at the University of California, Berkeley, California, June 16-19, 1969,
Society of Mining Engineers, American Institute of Mining, Metallurgical, and Petroleum
Engineers.
Song, X., P. Shen, S. Yuan and H. Cao (2000). The Integrated Characterization Techniques on
Reservoir Fractures. International Oil and Gas Conference and Exhibition in China. Beijing,
China, Copyright 2000, Society of Petroleum Engineers Inc.
Song, X., Y. Zhu, Q. Liu, J. Chen, D. Ren, Y. Li, B. Wang and M. Liao (1998). Identification and
Distribution of Natural Fractures. SPE International Oil and Gas Conference and Exhibition in
China. Beijing, China, 1998 Copyright 1998, Society of Petroleum Engineers Inc.
Sonnenthal, E., A. Ito, N. Spycher, M. Yui, J. Apps, Y. Sugita, M. Conrad and S. Kawakami
(2005). "Approaches to modeling coupled thermal, hydrological, and chemical processes in the
drift scale heater test at Yucca Mountain." International Journal of Rock Mechanics and Mining
Sciences 42(5–6): 698-719.
Stephansson, O., J. Hudson and L. Jing (2004). Coupled Thermo-Hydro-Mechanical-Chemical
Processes in Geo-systems, Elsevier Science.
Taleghani, A. D. (2009). Analysis of Hydraulic Fracture Propagation in Fractured Reservoirs: An
Improved Model for the Interaction Between Induced and Natural Fractures, The University of
Texas at Austin.
Tamagawa, T. M., T; Anraku, T; Tezuka, K; Namikawa, T; (2002). "Construction of Fracture
Network Model Using Static and Dynamic Data."
Tarahhom, F., K. Sepehrnoori and F. Marcondes (2009). A Novel Approach to Integrate Dual
Porosity Model and Full Permeability Tensor Representation in Fractures. SPE Reservoir
Simulation Symposium. The Woodlands, Texas, Society of Petroleum Engineers.
Teimoori, A., Z. Chen, S. S. Rahman and T. Tran (2005). "Effective Permeability Calculation
Using Boundary Element Method in Naturally Fractured Reservoirs." Petroleum Science and
Technology 23(5-6): 693-709.
Tester, J. W., B. J. Anderson, A. S. Batchelor, D. D. Blackwell, R. DiPippo, E. M. Drake, J.
Garnish, B. Livesay, M. C. Moore, K. Nichols, S. Petty, M. Nafi Toksoz, R. W. Veatch, R.
Baria, C. Augustine, E. Murphy, P. Negraru and M. Richards (2007). "Impact of enhanced
geothermal systems on US energy supply in the twenty-first century." Philosophical
Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences
365(1853): 1057-1094.
Tischner, T., M. Schindler, R. Jung and P. Nami (2007). HDR project Soultz: Hydraulic and
seismic observations during stimulation of the 3 deep wells by massive water injections.
Proceedings, 32nd workshop on Geothermal Engineering, Stanford University, Stanford,
California.
Torsæter, O. and J. Silseth (1985). "The Effects of Sample Shape and Boundary Conditions on
Capillary Imbibition." North Sea Chalk Symposium, Stavanger 1.

131
Tran, N. H. (2004). "CHARACTERISATION AND MODELLING OF NATURALLY
FRACTURED RESERVOIRS." Thesis.
Tran, N. H. (2007). "Fracture orientation characterization: Minimizing statistical modelling errors."
Computational statistics & data analysis 51(6): 3187-3196.
Tran, N. H., Z. Chen and S. S. Rahman (2006). "Integrated conditional global optimisation for
discrete fracture network modelling." Computers &amp; Geosciences 32(1): 17-27.
Tran, N. H. C., Z; Rahman, S S (2006). "Integrated conditional global optimization for discrete
fracture network modeling." Computers & Geosciences.
Tsenn, M. C. and N. L. Carter (1987). "Upper limits of power law creep of rocks." Tectonophysics
136(1–2): 1-26.
Turcotte, D. (1986). "Fractals and fragmentation." Journal of Geophysical Research: Solid Earth
(1978–2012) 91(B2): 1921-1926.
Vafai, K. (2009). Handbook of porous media, Crc Press.
Valley, B. C. (2007). The relation between natural fracturing and stress heterogeneities in deep-
seated crystalline rocks at Soultz-sous-Forêts (France), Diss., Naturwissenschaften,
Eidgenössische Technische Hochschule ETH Zürich, Nr. 17385, 2007.
Voss, R. F. (1985). Random fractal forgeries. Fundamental algorithms for computer graphics,
Springer: 805-835.
Walsh, J. J. and J. Watterson (1993). "Fractal analysis of fracture patterns using the standard box-
counting technique: valid and invalid methodologies." Journal of Structural Geology 15(12):
1509-1512.
Wang, F., X. Cai, Y. Su, J. Hu, Q. Wu, H. Zhang, J. Xiao and Y. Cheng (2012). "Reducing
cytotoxicity while improving anti-cancer drug loading capacity of polypropylenimine
dendrimers by surface acetylation." Acta Biomaterialia 8(12): 4304-4313.
Wang, X. and A. Ghassemi (2012). "A Three-Dimensional Poroelastic Model for Naturally
Fractured Geothermal Reservoir Stimulation." GRC Transactions 36.
Warren, J. E. and P. J. Root (1963). The Behavior of Naturally Fractured Reservoirs.
Watanabe, K. and H. Takahashi (1995). "Fractal geometry characterization of geothermal reservoir
fracture networks." Journal of geophysical research 100(B1): 521-528.
Watanabe, K. and H. Takahashi (1995). "Fractal geometry characterization of geothermal reservoir
fracture networks." J. Geophys. Res. 100(B1): 521-528.
Watanabe, N. (2012). Finite element method for coupled thermo-hydro-mechanical processes in
discretely fractured and non-fractured porous media, PhD Thesis, Technische Universität
Dresden, Chair of Applied Environmental System Analysis, Helmholtz Centre for
Environmental Research UFZ, Department of Environmental Informatics.
Watanabe, N., W. Wang, C. McDermott, T. Taniguchi and O. Kolditz (2010). "Uncertainty analysis
of thermo-hydro-mechanical coupled processes in heterogeneous porous media." Computational
Mechanics 45(4): 263-280.
Wealthal, G. P., B. H. Kueper and D. N. Lerner (2001). "Fractured Rock-mass Characterisation for
Predicting the Fate of DNAPLS " Conference Proceedings of Fractured Rock 2001, Toronto,
Ontario, Canada.
Weinberg, R. F. and K. Regenauer-Lieb (2010). "Ductile fractures and magma migration from
source." Geology 38(4): 363-366.
Wells, A., D. Forman, L. Ranford and P. Cook (1970). "Geology of the Amadeus Basin, Central
Australia." Bureau of Mineral Resources, Geology and Geophysics, Canberra, Australia(Bulletin
100).
Weng, M. C., L. S. Tsai, Y. M. Hsieh and F. S. Jeng (2010). "An associated elastic–viscoplastic
constitutive model for sandstone involving shear-induced volumetric deformation." International
Journal of Rock Mechanics and Mining Sciences 47(8): 1263-1273.

132
Wickham, J. S., G. S. Tapp and J. N. Reddy (1982). "Finite-element modelling of fracture density
in single layer folds." International Journal for Numerical and Analytical Methods in
Geomechanics 6(4): 441-459.
Wickham, J. S. T., G S; Reddy, J N (1983). "Finite element modelling of fracture density in single
layer folds : Int J Num Anal Meth Geomech, V6, N4, Oct-Dec 1982, P441-459." International
Journal of Rock Mechanics and Mining Sciences & Geomechanics Abstracts 20(2): A39-A39.
Williamson, G., A. Lovell, P. K. Harvey and G. S. o. London (1999). Borehole Imaging:
Applications and Case Histories, Geological Society.
Willis-Richards, J., K. Watanabe and H. Takahashi (1996). "Progress toward a stochastic rock
mechanics model of engineered geothermal systems." J. Geophys. Res. 101(B8): 17481-17496.
Wong, P., F. Aminzadeh and M. Nikravesh (2002). Soft Computing for Reservoir Characterization
and Modeling, Physica-Verlag HD.
Xing, H. (2009). Advances in Geocomputing, Springer.
Xu, C. and P. Dowd (2010). "A new computer code for discrete fracture network modelling."
Computers & Geosciences 36(3): 292-301.
Yükselen, M. A. and M. Z. Erim (1996). "Basic equations for incompressible, non-Newtonian
fluids in curvilinear, non-orthogonal and accelerated coordinate systems." Acta Mechanica
118(1): 39-54.
Zhang, X., R. G. Jeffrey and E. Detournay (2005). "Propagation of a hydraulic fracture parallel to a
free surface." International Journal for Numerical and Analytical Methods in Geomechanics
29(13): 1317-1340.
Zienkiewicz, O. C. T., R L; Nithiarasu, P; (2000). "The Finite Element Method for Fluid Dynamics,
Sixth Edition."
Zimmermann, G., I. Moeck and G. Blöcher (2010). "Cyclic waterfrac stimulation to develop an
Enhanced Geothermal System (EGS)—Conceptual design and experimental results."
Geothermics 39(1): 59-69.

133
Appendix A: Probability of Occurrence of Fractures
In this appendix a general formula is derived to have an approximation of the probability
of the presence of fracture based on the thermodynamic equilibrium equation (Reddy et al.
1983, Wickham 1983). Thermodynamic equilibrium for a body needs that:

d
(U  V  T  I )  0 (A. 1)
dt

where t is the time, U is the strain energy (Griffith, 1921), V is the potential energy, T is the
kinetic energy and I is the sum of all irreversible energies. The above equation can be
written in another form if we assume that all the energy dissipation takes place in the
periphery of a crack in the rock.

dA d
(U  V  T  I )  0
dt dA (A. 2)

For semi-steady state problems like the one in this study, kinetic energy term can be
omitted from the above equation. Also the irreversible energy term can be written in the
form:

dI dF dD
 
dA dA dA (A. 3)

dF
where is the energy required to produce a unit surface of fracture and is usually
dA
denoted by  f . D is the sum of all other energies unrelated to fracture propagation like heat

generation etc.

From Eq. (A.2) and Eq. (A.3) and also considering the elimination of kinetic energy
term we have:
134
 d dD
(U  V )   f (A. 4)
dA dA

This study is done under nearly incompressibility conditions and so the potential energy
term (dV ) can also be eliminated from the equation. After that we have:

dA dU dD
f   (A. 5)
dt dt dt

dU dD
It can be seen that if is bigger than then the fracture propagates and if the first
dt dt
part is smaller than the second part, the fracture closes and if it is zero there is no
movement. In the above equation  f is known for a material under consideration. So the

only thing which should be done is to relate the right hand side of the above equation to the
stress tensor of the material.

The strain energy is a function (f1) of the stress invariants. Also we can assume that the
dissipation energy is a function (f2) of the stress invariants with a proportionality constant
K, which is a material property. Then Eq. (A.5) can be written as:

dA
f  f1  Kf 2 (A. 6)
dt

Based on the generalized von Mises criteria we have:

Toct  f3 (1 ,  2 ,  3 ) (A. 7)

where Toct is the octahedral shearing stress and f3 is a function of principal stresses. Since
the strain energy density is proportional to octahedral shearing stress, Eq. (A.6) can be
easily converted into an energy equation as follow (Murrel 1963):
135
 2 2 8
( I1  3I 2 )  K 0 I1  0 (A. 8)
3 

Comparing Eq. (A.6) and (A.7), we have:

2 2
f1  ( I1  3I 2 ) (A. 9)
3

8
f2  I1 (A. 10)


Then it can be easily understood that the Eq. (A.6) can be written in the form:

dA 2 2 8K
f  ( I1  3I 2 )  I (A. 11)
dt 3  1

dA
Based on the above equation, can be calculated for each node and a value
dt
representing the rate of formation of new fractures of the presence of fractures will be
assigned to all nodes. Also this probability value will be updated during each time step and
finally it will be normalized to be prepared as an input to the stochastic simulation part.

136
Appendix B: Artificial Neural Network
The Back propagation Neural Network (BPNN) finds a complex relationship between a
set of data (which are known to have a relationship) through an iterative training process
(Winston, 1993). A very simple BPNN consists of three different layers namely: input
layer, hidden layer and the output layer. Each node is connected to all the adjacent layer’s
nodes via a typical weighted link.

The input layer nodes represent the field data including core descriptions, well log
parameters etc. The number of nodes in the middle layer varies depending on the
complexity of the problem. The output layer nodes represent the desired parameters which
are going to be calculated based on the relationship between them and the input parameters.

For the training purpose, a randomly selected conditional data composed of all available
data types are collected from one sampling trial is assigned to the input layer. Based on the
weights which have been selected arbitrarily at the beginning of the process, the values of
the middle layer and output nodes are calculated respectively and then, an error function is
used to calculate the difference between the obtained and the actual output data. The values
of the middle layer nodes are calculated based on the corresponding weights as follow:

n
y j   wij xi  b j (B. 1)
i 1

where y is the value for the jth middle layer node, and w is the weight factor for the ith input
layer node and b is the bias value for the jth middle layer node. The bias value
(representing all biases for one node) in one node of a layer is connected to all the nodes in
the next layer, equals to 1 and its corresponding weight factor is corrected during the
iteration process as is discussed below (Rumelhart et. al., 1986).

After that, y values calculated in each of the middle layer nodes are transformed via the
Radial Basis Function (RBF) known as the activation function as follow:

137
 h j  f (d j ) (B. 2)

RBFs have strong functionality (Park and Sandberg, 1991) and Gaussian function is the
commonly used RBF for this purpose which is a symmetric exponential function as shown
below:

x
f ( x)  exp( ) (B. 3)
2

Which  is the control constant of the function that represents the width of the spatial
attribute. This constant is the same as the range parameter used in Gaussian covariance
models (Deutsch and Journel, 1997).

After that each output node’s value is calculated based on a weighted sum of all the
middle layer nodes as follow:

m
ok   w jk h j (B. 4)
j 1

where ok is the kth output node, wjk is the weight of the link connecting the jth middle layer
node to kth output node. Also h is calculated with Eq. (B.3).

Obtained values in the output layer nodes are again converted with the activation
function to calculate the final output value. Combining Eqs. (B.1-4) it can be easily shown
that:

ok  f ( w jk  f ( wij  xi  t j )  bk ) (B. 5)
i k

where ok is the kth output value, f is the activation function, xi is the ith input value, t and b
are the biases and w are the corresponding weights.

138
 After calculating the values of the output nodes, quadratic error function is used
compare them with the actual field data used for the training purpose as follow:

1 n
E 
2 k 1
(d k  ok )2 (B. 6)

where dk and ok are the actual and target output values for the kth output node. As can be
seen the error values for all output nodes are cumulatively added together.

Substituting Eq. (B.5) into (B.6) results in:

1
E [dk  f (j w jk  f (i wij  oi  t j )  bk )]2
2 k
(B. 7)

Then all the weights used in the process are adjusted according to the computed error
value with the gradient descent method. In another word, the whole system of weights
could be imagined as a virtual sphere which should be rolled on the error surface in the
direction of steepest descent (negative of the function gradient) to reach the global
minimum finally. So the main requirement is that the activation function (Eq. B.3) does not
have any discontinuity or flat section.

All the weights are adjusted with the following equation:

wnew  wold  w (B. 8)

The values of w for the each weighted links are calculated with Eq. (B.9).

E
wij    (B. 9)
wij

where wij is the weight used from the ith to the jth node of two adjacent layers and  is a
positive learning rate parameter bounded by [0,1] which determines the step size in

139
gradient descent. For the weight connecting nodes from middle layer to the output layer,
E
is obtained with Eq. (B.7) after some modifications and derivation of the function f as
wij

follow:

E
  (d k  ok )  f k  o j (B. 10)
wkj k

Also the value of wij for the weighted links connecting the ith input nodes to the jth

middle layer node are obtained based on the Eq. (B.9) incorporating the chain rule as
follow:

E o j
wij     (B. 11)
o j wij

After taking the corresponding derivatives from Eq. (B.7), Eq. (B.11) can be written as
follow:

wij     (d k  ok )  f k  w jk  f k  oi (B. 12)

k

For each set of sample data, the whole above mentioned process is repeated and the new
adjusted weight values are used for the next iteration step. Every single iteration step, in
which the weights are updated, is called an epoch (Masters, 1993). The training process
stops when any of the following criteria met at the end of corresponding iteration:

 Cumulative error value obtained with Eq. (B.6) is below the user-defined threshold
 Maximum number of iterations (user-defined) reached
 Over-training happens (when the training patterns are memorized instead of being
generalized)

140
Appendix C: Finite Element Shape Function
Shape functions are used to interpolate an unknown parameter over the domain of an
element. For a scalar variable, u, interpolation can be written as (Zienkiewicz 2000):
n
u   Ni ai
i 1
(C. 1)

where n is the total number of functions and a is the unknown parameter. Shape functions
also follow the property named ‘partition of unity’ that:
n

N
i 1
i 1
(C. 2)

It is also convenient to use the normalised coordinates in the basic formulation. For this
purpose local coordinated (  , ) are used to locate the centre of an element at the origin
(Zienkiewicz 2000). A systematic method to generate the shape functions in any order is to
use the product of the appropriate polynomials. Polynomials which have the
abovementioned property are called Lagrange polynomials and can be written as:
(  0 )  (   n )
lkn ( ) 
(k  0 )  (k   n ) (C. 3)

For two dimensions the Lagrange multiplier method can be extended by multiplying the
Lagrange multiplier for each coordinate as:

Nij  lln ( )l mj ( ) (C. 4)

where n and m are the number of subdivisions in each dimension.

It is also efficient to generate the shape functions for the nodal values placed on the
boundary of the element.

The above mentioned procedure is applied to generate the shape functions for the 3- and
6-nodded triangular elements. The shape functions are listed below.

141
 Three-Node Triangular Element
N1  1    (C. 5)
N2   (C. 6)
N3   (C. 7)

Six-Node Triangular Element

6-nodded triangular element is composed of two types of nodes, corner nodes and mid-
size node. Therefore two sets of shape functions are used as follow (nodes 1-3 as corner
nodes and 4-6 as mid-side nodes):
N1  (1    )  (1  2  2 ) (C. 8)
N2    (2  1) (C. 9)
N3    (2  1) (C. 10)
N4  4    (1    ) (C. 11)
N5  4 (C. 12)
N6  4   (1    ) (C. 13)

142