Download as pdf or txt
Download as pdf or txt
You are on page 1of 93

Dynamic two-phase flow

Delft University of Technology

modelling of coupled
well-reservoir transients
I.R. Doǧan
Dynamic two-phase flow
modelling of coupled
well-reservoir transients
by

I.R. Doǧan
to obtain the degree of Master of Science
at the Delft University of Technology.

Student number: 4017773


Project duration: October 19, 2015 – October 14, 2016
Thesis committee: dr. ir. M.I. Gerritsma, TU Delft, supervisor
dr. ir. B. Sanderse, Royal Dutch Shell
prof. dr. ir. R.A.W.M. Henkes Royal Dutch Shell/TU Delft
dr. ir. A.H. Van Zuijlen TU Delft
Abstract
The oil and gas industry uses dedicated computer models to design their production systems, which typically
includes the reservoir, wells, pipelines, and processing facilities (such as separators, compressors, pumps). In
these models, the well and the reservoir are often considered as separated systems that communicate with
each other through the near wellbore region. The size and geometry of both systems are often completely
different yielding distinct spatial and temporal characteristic length scales. Typically, well phenomena are in
the order of seconds to hours, whereas reservoir phenomena range between days and years. This motivates
the approach to treat both systems separately, as the dynamic response of one system is hardly influenced by
the other system. However, there are certain dynamic flow phenomena where the transients have compara-
ble time and length scales, such that a coupled simulation approach is necessary.

The goal of this project was to extend the Shell single phase well-reservoir simulator to two-phase flow for
simulation of well reservoir transients. To achieve this goal, four different steps were identified: a literature
study was carried out, a two-phase reservoir model has been developed, an existing well code has been mod-
ified to include a two-phase drift-flux model, and the coupling between the well model and the reservoir
model has been designed. This research is conducted in the team of Shell Projects & Technology, as part of
a master graduation project at the Delft University of Technology. The project has been finalised within a
period of 11 months.

The two-phase immiscible reservoir equations have been discretised in space using a second-order accurate
finite volume method. Second-order time integration was obtained using implicit BDF2 time integration,
which is necessary due to the stiff character of the reservoir equations.
The well model uses a drift-flux formulation. An existing well code, which is a second-order accurate finite
volume method, has been modified in order to be coupled to the reservoir. Again, BDF2 time integration is
employed for the temporal discretisation.
The coupling has been realised in a monolithic fashion. The coupling with respect to single phase flow has
been modified to accommodate the drift-flux model. The coupled simulator showed no difficulties in run-
ning simulations with either very small time steps (seconds) or large time steps (years). This stable behaviour
is greatly desired, since it enables solving different types of applications with one single simulator.

Two different academic cases have been investigated: a production ramp-up (focus on well transients) and a
water flooding case (focus on reservoir transients). Based on these cases, the following can be concluded.
When considering the dynamics in the well, the transients in the reservoir need to be taken into account. The
comparison of the coupling with a dynamic reservoir and an Inflow Performance Curve (IPC) showed that
the transient behaviour in the well was calculated incorrectly when an IPC was employed, stressing the need
for a coupled simulator.
When considering dynamics in the reservoir, the transients in the well have a negligible effect on the dy-
namics in the reservoir. The characteristic time scales of the reservoir are generally larger (several orders in
magnitude) than the time scales of the well. In this case, the need for a coupled simulator is less evident.

The long term goal of Shell is to extend Compas, which is their in-house dynamic multiphase pipeline and
well simulator, with a dynamic reservoir model. The two-phase model for the well that is available in Compas
is different from the drift flux model used in this study. Shell is particularly interested in simulating liquid
loading and in designing intelligent systems, such as smart wells. The coupling through multiple coupling
points is essential. Therefore, it is recommended that the reservoir model should be extended to multiple
layers.
Second, the focus of this study was on development and verification of a dynamic two-phase well-reservoir
code. The performance of the code has been tested using academic test cases. However, the real validation of
the model has not been considered. It is thus recommended to validate the model using experimental and/or
field data.

iii
Preface
As part of my master graduation project at the Delft University of Technology, I decided to apply for a research
internship at Shell. My affection with fluid dynamics and numerical modelling brought me in connection
with the multiphase flow team of Shell Projects & Technology in Amsterdam. Shell challenged me to conduct
a research on coupling methods of the well-reservoir system under multiphase flow conditions. A task which
at that time, seemed to be beyond my knowledge. However, the professional environment and the profound
expertise of the team lured me into the challenge.

My deepest gratitude goes out to my supervisor at Shell: dr. ir. Benjamin Sanderse. Thanks to him, I was
motivated to bring the project to a successful end. I enjoyed both his professional expertise and his energetic
personality.

I have been grateful that the Aerodynamics department of the Delft University of Technology was open to col-
laborate with Shell on this topic. Special gratitude goes out to dr.ir. Marc Gerritsma, who has been assisting
me and made the collaboration possible.

Special gratitude as well to my mentor at Shell: prof.dr.ir. Ruud Henkes, who mentored me, guided me and,
when necessary, challenged me to raise my research to a higher level.

I would like to thank ir. Maurice Hendrix, whom I (accidentally) met thanks to our shared interest in nu-
merical modelling. I particularly enjoyed the intensive collaboration and I deeply appreciate his selfless and
helpful attitude.

Lastly, I would like to thank my friends and family for supporting me during the years I spent in university.
Thanks for the mental support when necessary.

I.R. Doǧan
Delft, October 5, 2016

v
List of Figures

1.1 Schematic of the well-reservoir system, obtained from Economides et al. [19]. . . . . . . . . . . . 2
1.2 Spatial and temporal length scales for well (W) and reservoir (R) processes, obtained from Nen-
nie et al. [40]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Liquid loading of a gas well, obtained from Lea et al. [31]. . . . . . . . . . . . . . . . . . . . . . . . 4
1.4 Schematic of a gas lifted well, obtained from Sinégre et al. [53]. . . . . . . . . . . . . . . . . . . . . 4
1.5 Dynamic IPC for a sinusoidal variations in BHP for various frequencies, obtained from Belfroid
et al. [6]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.6 IPC and TPC for a typical tight gas well. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2.1 Phase diagram for a multi-component hydrocarbon mixture, obtained from Dake [16]. . . . . . 10
2.2 Hysteresis in contact for an imbibition and drainage process and matching the capillary pres-
sure curve, obtained from Dake [16]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.3 The effective (left) and relative (right) permeabilities, as a function of the water saturation S w
for two-phase flow, obtained from Dake [16]. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.4 Relative permeability and capillary pressure curves used in this study. . . . . . . . . . . . . . . . . 15
2.5 Fractional flow function g and its derivative g 0 . The tangent to the curve g originating from
g /S w gives the point S w = α. For values larger than α the solution consists of a shock followed
by a rarefaction wave. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.6 Analytical solutions of the Buckley–Leverett problem with quadratic Corey relative permeabili-
ties (λ=2). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.7 Initial pressure distribution for case I with S Lw = 0.5 and S Rw = 0 at the discontinuity at x = 100 m. 22
2.8 The centres of volume element Ωi are defined on the blue primal grid, which contain he con-
servative variables Ūi . The faces of element Ωi are defined on the red dual grid and contain the
flux variables F i . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.9 Schematic of the volumes and grid nodes near the well interface. . . . . . . . . . . . . . . . . . . . 32

3.1 The radial flow of a typical oil well under steady state flow conditions. . . . . . . . . . . . . . . . . 39
3.2 Analytical verification showing the global ² and local τ truncation error for the single-phase case
using a two-point and four-point stencil for the interior gradients. . . . . . . . . . . . . . . . . . . 41
3.3 MMS verification showing the global ² and local τ truncation error for the single-phase case
using a two-point and four-point stencil for the interior gradients. . . . . . . . . . . . . . . . . . . 44
3.4 MMS verification showing the global ² and local τ truncation error for the two-phase hyperbolic
case using a two-point stencil for the interior gradients. . . . . . . . . . . . . . . . . . . . . . . . . 45
3.5 MMS verification showing the global ² and local τ truncation error for the two-phase parabolic
case using a two-point stencil for the interior gradients. . . . . . . . . . . . . . . . . . . . . . . . . 47
3.6 Computed solutions of the Buckley–Leverett problem with quadratic Corey relative permeabil-
ities (N x = N t , λ=2) at t = 3456 s. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.7 Error plots of the solutions to Buckley Leverett’s equations. . . . . . . . . . . . . . . . . . . . . . . 51

4.1 Schematic drawing of the well, where phase k = g , l . . . . . . . . . . . . . . . . . . . . . . . . . . . 54


4.2 Critical Kutateladze number as a function of the Bond number, obtained from Shi et al. [52]. . . 57
4.3 Depth profiles of pressure, gas hold-up, gas phase velocity and drift velocity of the two-phase
wellbore flow problem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.4 Sketch of the coupling between the wellbore and reservoir through mass and pressure continu-
ity at the well-reservoir interface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
4.5 TPC and IPC for the ramp-up case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.6 Depth profiles of the production ramp-up case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.7 LGR and normalised PI for the production ramp-up case. . . . . . . . . . . . . . . . . . . . . . . . 68
4.8 Water flooding simulation results for t = 0 days to 4.2 days with ∆t = 2.84 × 103 sec. . . . . . . . . 70

vii
viii List of Figures

4.9 Water flooding simulation results for t = 4.2 days to 4.8 days with ∆t = 100 sec. . . . . . . . . . . 71
4.10 WOR and normalised PI for the water flooding case. . . . . . . . . . . . . . . . . . . . . . . . . . . 71
List of Tables

2.1 Typical values of permeability K0 , obtained from Bear [5]. . . . . . . . . . . . . . . . . . . . . . . . 11

3.1 Reservoir model options . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40


3.2 Parameters for single phase test case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.3 Parameters for two-phase (hyperbolic) test case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
3.4 Paramters for two-phase (parabolic) case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
3.5 Parameters for the Buckley-Leverett problem. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
3.6 Experimental order of convergence for the Buckley–Leverett problem. . . . . . . . . . . . . . . . 51
3.7 Order of convergence of the local and global truncation errors. . . . . . . . . . . . . . . . . . . . . 52

4.1 Empirical fitting parameters used in the drift-flux model [41, 52]. . . . . . . . . . . . . . . . . . . 58
4.2 Parameters of the two-phase wellbore flow problem. . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.3 Parameters of the reservoir model. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
4.4 Parameters for the water flooding case. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

ix
Contents

Abstract iii
Preface v
List of Figures vii
List of Tables ix
1 Introduction 1
1.1 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Well-reservoir interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2.1 Liquid loading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2.2 Slug flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.3 Gas lift heading . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.4 Water/gas coning . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.5 Water hammer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.6 Compressibility effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.3 Nodal Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.4 Shell developments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
1.5 Research goal and outline of the report . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2 Reservoir modelling 9
2.1 Darcy’s Law . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2 Fluid properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.2.1 EOS Water . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.2.2 EOS under-saturated oil . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2.3 EOS ideal gas. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Rock properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3.1 Capillary pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3.2 Relative permeabilities. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.3.3 Rock compressibility, consolidation and compaction . . . . . . . . . . . . . . . . . . . 14
2.4 Single phase flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.5 Two-phase immiscible flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.5.1 Phase Pressure Saturation formulation. . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.5.2 Global Pressure IMPES formulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.5.3 The Buckley-Leverett equation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.6 PDE classification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.6.1 General case: Compressible phases and non-zero capillary pressure. . . . . . . . . . . . 24
2.6.2 Incompressible phases and no capillary pressure . . . . . . . . . . . . . . . . . . . . . 26
2.6.3 Compressible phases and no capillary pressure . . . . . . . . . . . . . . . . . . . . . . 27
2.6.4 Incompressible phases and non-zero capillary pressure . . . . . . . . . . . . . . . . . . 27
2.7 Discretisation of the reservoir model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.7.1 Spatial discretisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
2.7.2 Boundary conditions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.7.3 Temporal discretisation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

3 Verification 35
3.1 Error analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.1.1 Taylor expansions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

xi
xii Contents

3.2 Method of manufactured solutions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38


3.3 Steady-state code verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
3.3.2 Test cases steady state . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
3.3.3 Single phase flow (elliptic) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.3.4 Two-phase flow (hyperbolic) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
3.3.5 Two-phase flow (parabolic) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.4 Unsteady code verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.4.2 Buckley Leverett problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.5 Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
4 Well-reservoir coupling 53
4.1 Well modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.1.1 Homogeneous well model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
4.1.2 Drift-flux model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
4.1.3 Reference solution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.1.4 Verification with the two-phase wellbore flow problem . . . . . . . . . . . . . . . . . . 59
4.2 Coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.2.1 Spatial coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4.2.2 Temporal coupling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.3 Test cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
4.3.1 Production ramp-up . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
4.3.2 Water flooding . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
4.3.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
5 Conclusions and recommendations 73
5.1 Conclusions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
5.2 Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
Bibliography 77
1
Introduction

1.1. Background
Reservoirs are subsurface layers containing hydrocarbons in porous and permeable rock formations. Poros-
ity is related to the void spaces in the rock formation, and it is a function of the number and volume of void
spaces. Permeability is the ability of the fluids to flow through the rock formation. A higher porosity and
permeability is thus proficient in terms of oil/gas exploration. Reservoirs experience increased pressure and
temperature values compared to the surface conditions. When a reservoir is connected through a wellbore,
the drawdown, i.e. the pressure difference between reservoir pressure and the Bottom Hole Pressure (BHP),
establishes a natural flow towards the wellbore. The wellbore perceives the reservoir as an inflow, of which
the inflow rate depends on the drawdown. The flow rate in the wellbore on its hand depends on the pressure
difference between the well ends, that is the BHP and the Tubing Head Pressure (THP). A schematic of a well-
reservoir system is presented in fig. 1.1.

The well and the reservoir are often considered as separated systems that communicate with each other
through the near wellbore region. The size and geometry of both systems are often completely different
yielding distinct spatial and temporal characteristic length scales. Nennie et al. [40] provided an estimate
of the spatial and temporal length scales of production-related dynamic phenomena in wells and reservoirs
(see fig. 1.2). Typically, well phenomena are in the order of seconds to hours, whereas reservoir phenomena
range between days and years. This motivates the approach to treat both systems separately, as the dynamic
response of one system is hardly influenced by the other system. However, there are certain dynamic flow
phenomena where the transients have comparable time and length scales, such that a coupled simulation ap-
proach is necessary (overlapping regions in fig. 1.2). Typical transient flow phenomena are wellbore storage,
slugging, liquid loading, coning, water hammer, gas lift heading and compressibility effects [31, 49, 54, 58].
These typical flow transients will be explained in section 1.2.

The need for simulations stems from the requirement to predict well-reservoir performance under different
operating conditions. Oil and gas recovery projects may involve capital investments of several hundreds of
millions of dollars and operations must be performed under save environmental and working conditions [20].
Traditionally, well and reservoir simulations are performed by "standalone" simulators, i.e. dynamic simula-
tors which use a steady-state model for the complementary system. Dynamic well models are combined
with Inflow Performance Curves (IPC), whereas dynamic reservoir models are combined with Tubing Perfor-
mance Curves (TPC) [15]. Often, these are analytic expressions or look-up tables to model the reservoir and
well, respectively. Such simulations are sufficient to predict and monitor steady-state operations, and help
with overall production planning and field life well allocation [49]. However, steady-state models are not able
to capture the transient phenomena. The well and reservoir performance curves will be briefly introduced in
section 1.3.

The depletion of "easy accessible" natural resources turned the oil/gas industry to more complex and expen-
sive recovery projects, like shale gas and Arctic resource recovery. However, large capital investments, high
operational costs and dangerous working environments of such projects, redirected interest to optimisation

1
2 1. Introduction

Figure 1.1: Schematic of the well-reservoir system, obtained from Economides et al. [19].

of existing fields. Recovery optimisation of maturing oil/gas fields brings new challenges in which the under-
standing of well-reservoir transients is essential. A special interest goes out to the understanding and mitiga-
tion of liquid loading of gas wells. Liquid loading is the drainage of a well due to liquid accumulation at the
bottom location of the well, which eventually could lead to killing of production (see also section 1.2.1). Shell,
and in particular the Nederlandse Aardolie Maatschappij (NAM), are interested in mitigation strategies of liq-
uid loading. Shell, through the NAM, operates a significant number of maturing gas fields in the Slochteren
area (Groningen), which are already, or expected to, suffer from liquid loading. Shell and industrial partners
have taken initiative to put research into liquid loading and innovative de-liquification techniques. A brief
overview will be presented in section 1.4.

Finally, the rise of "intelligent" systems provided opportunities for the industry to monitor, automate and
optimise recovery processes. Intelligent systems may help to mitigate risks, optimise planning and mainte-
nance, and increase recovery. The "smart well" is one of these applications. These wells are equipped with
down-hole sensors and Inflow Control Valves (ICV) to monitor and control the reservoir inflow [32]. The in-
teraction between multiple smart wells is yet to be investigated. Dynamic simulation of the well-reservoir
system is essential in the design and the development of these intelligent systems.

1.2. Well-reservoir interaction


In well and reservoir systems transients may occur due to production related phenomena. As a result of tran-
sients developing in one system, transients in the other system can be induced or affected. The interaction
can have a strong effect on the production and may endanger the safety of the production facility due to
high fluctuating loads. This section will summarise the most common encountered flow transients in well-
reservoir systems.

1.2.1. Liquid loading


When the pressure in a reservoir is sufficiently large, the gas velocity in the tubing is high enough to drag the
additionally produced liquids to the surface. Often, liquids can be produced directly from the reservoir, or
condensate liquids can occur due to pressure and temperature changes along the well trajectory. Towards
the end-of-field life, the reservoir pressure reduces such that the liquids can no longer be dragged upwards
by the gas. This leads to accumulation of liquids at the bottom location of the well (see fig. 1.3). This process
is called "liquid loading". Liquid loading can lead to erratic, slugging flow and decreased production [31].
When the well continues to produce at such a reduced rate, the flow is called "meta-stable". The well may
1.2. Well-reservoir interaction 3

Figure 1.2: Spatial and temporal length scales for well (W) and reservoir (R) processes, obtained from Nennie et al. [40].

eventually die if the liquids are not continuously removed.

1.2.2. Slug flow


Slug flow is a flow instability characterised by accumulation of liquid in wells or pipelines. Liquid slugs that
bridge the entire cross-section of the pipe are followed by a gas bubble resulting in an oscillatory flow be-
haviour. The oscillatory behaviour can lead to fluctuations in the THP, yielding unpredicted loads on surface
facilities. Surface facilities are therefore equipped with so-called "slug catchers" to prevent them from flood-
ing. The BHP can also be influenced by slugging, which influences the inflow from the reservoir and can
aggravate the slugging itself [49].

1.2.3. Gas lift heading


In oil producing wells, artificial gas lift can be introduced to increase oil production. A schematic of such an
artificial gas lift mechanism is presented in fig. 1.4. Gas is injected at the tubing head (A) and flows trough the
casing (B) to a new injection point (C). The buoyancy of the gas helps the oil to reach the surface. However, for
low gas lift rates an unstable oil production may occur with periods of reduced or even absent oil production,
followed by large peaks of liquid and gas. Gas lift heading can also occur when the reservoir pressure is below
the oil bubble point pressure, yielding additional free gas in the near wellbore region [54].

1.2.4. Water/gas coning


Coning is a typical natural flow instability that mainly occurs in oil producing wells. When the oil is produced
from a thin oil rim, where the gas cap is close to the perforations of the well, the increase in drawdown may
result in the formation of a gas cone. As a result, gas will enter the tubing in addition to the oil. For some
conditions, this can lead to excessive gas production and a decreased unstable oil production. For other
conditions, only a small amount of gas enters the tubing, creating a natural gas lift, which thus stimulates
the oil production [54]. Equivalent to gas coning, water coning may occur when the well perforations are
close to the water aquifer, rather than the gas cap. Cone formation is a dynamic phenomenon, involving both
reservoir and well dynamics and can have a strong influence on the oil production.

1.2.5. Water hammer


Water hammer is known as the pressure pulse or surge that may occur due to the instant shut-in of a single
phase flow line [12]. The pressure pulse can lead to pipeline rupture, severe damage to surface facilities
and/or failure of formation integrity, resulting in sand production [12]. In mature reservoirs, water can be
4 1. Introduction

Figure 1.3: Liquid loading of a gas well, obtained from Lea et al. [31].

Figure 1.4: Schematic of a gas lifted well, obtained from Sinégre et al. [53].

injected into the reservoir to force the oil to flow to the producing well, which is referred to as "secondary
recovery". In these water injection wells, the reflection of the pressure wave from the reservoir can cause
back flow from the reservoir into the well. This can lead to transport of solid debris and sand, which may
result in a drastic drop of injectivity [50].

1.2.6. Compressibility effects


The reservoir can be modelled by the IPC, which describes the BHP as as a function of the mass flow rate. At
steady-state conditions, a linear relation exists between the mass flow and the BHP (for liquids), as indicated
by the red line in fig. 1.5. However, in dynamic state a phase difference exists between the change in pressure
and mass flow rate, due to the compressibility of the fluid. In fig. 1.5, sinusoidal variations in the BHP and the
corresponding mass flow rates are shown for various frequencies. In contrast to the linear IPC for steady-state
conditions, an elliptical distribution can be observed. The ellipsoids take different shapes, depending on the
frequency of variation in BHP. A higher frequency of variation in the BHP gives rise to a larger phase lag of the
flow rate. Similarly, a higher compressibility yields a larger phase difference. The productivity index (PI) is
defined as:
1.3. Nodal Analysis 5

Figure 1.5: Dynamic IPC for a sinusoidal variations in BHP for various frequencies, obtained from Belfroid et al. [6].

Q
PI = , (1.1)
∆p

where Q is the production rate and ∆p is the drawdown (difference in reservoir pressure and flowing BHP).
The PI is thus the inverse of the slope of the IPC and is proportional to dQ/dp. From fig. 1.5 one can observe
that for higher frequencies (or shorter periods), the slope dp/dQ is decreasing, hence dQ/dp is increasing.
This means that the productivity can become significantly larger (up to a factor 3 for this particular case).
The resulting higher flow rate Q can lead to dangerous situations due to overpressurisation or flooding of the
separator system.

When considering short time dynamics, the oil behaves as a compressible fluid. A similar effect attributed
to the compressibility of the fluids is called "wellbore storage". Wellbore storage is the effect that when a
producing well is shut-in at the surface by closing the tubing head valve, the fluid flow continues to flow from
the reservoir into the wellbore. This can be attributed to the compressibility of the fluid as it takes a finite time
for the bottom hole to feel the pressure change. Additionally, when the fluid in the wellbore is multiphase,
phase redistribution takes place leading to pressurisation of the well [59].

1.3. Nodal Analysis


For the reservoir fluids to reach the surface (e.g. well head or production platform) various obstacles or flow
restrictions have to be overcome, where each obstacle is associated with a pressure loss ∆p. Changing the
flow restrictions by for example altering the THP will have a direct effect on the production rate Q. To de-
termine the overall performance, all components of the entire system have to be considered. This is referred
to as the "Nodal Analysis", where a nodal point is a subsystem of the total system. Although the nodal point
can be considered at any location, for the well-reservoir system it is common to consider the nodal point at
the mid-perforation depth or bottom hole. The curve from the reservoir to this nodal point is then the inflow
curve, whereas the curve from the nodal point to the tubing head is called the outflow curve. The intersection
of the inflow and outflow curve is the operating production point, which gives a certain pressure and mass
flow rate, as presented in fig. 1.6 [31].

The inflow curve is known as the Inflow Performance Curve (IPC) of the reservoir, and describes the mass
inflow rate for a certain BHP. When the BHP equals the (average) reservoir pressure no fluids will be flowing
into the reservoir. This point is depicted by the starting point of the curve. On the other hand, when the BHP
6 1. Introduction

Figure 1.6: IPC and TPC for a typical tight gas well.

is equal to zero, a maximum pressure differential exists, i.e. ∆p = p r es , maximising the inflow rate. This point
is depicted at the end of the curve.

The outflow curve represent the characteristics of the tubing, i.e. the Tubing Performance Curve (TPC). In
steady state the pressure loss over the tubing consists of a hydrostatic pressure loss, frictional pressure loss,
and an additional pressure loss due to the accelerations and redistribution of the fluids. At low gas flow rates
the hydrostatic effects are dominant, whereas at high gas flow rates frictional effects are dominant. One can
observe that the TPC shows a minimum pressure point. For the hydrostatic dominated part of the curve an
increasing mass flow rate would reduce the hydrostatic head, up to the point that friction related pressure
drop increases more rapidly than that the hydrostatic head is reduced. The reason stems from the higher liq-
uid content (or liquid hold-up) for low gas flow rates, which has a larger hydrostatic head than the gas phase.
An increase in mass flow rate reduces the liquid hold-up therefore decreases the overall pressure drop. When
the mass flow rate increases even further, the liquid hold-up may decrease even more, but the pressure drop
due to friction is larger than the "gain" in hydrostatic head. This explains the minimum pressure loss point of
the system. This point is most attractive in terms of required energy to lift the fluids to the surface. Produc-
tion to the right of the minimum is associated with stable production (annular flow), while production to the
left of the minimum results in unstable production (slug flow). The latter situation may result in backflow of
the liquid phase yielding liquid accumulation at the bottom hole which eventually could kill the well (liquid
loading). A more elaborate explanation on stable production rates can be found in Lea et al. [31].

1.4. Shell developments


For more than a decade Shell has an interest in developing an integrated coupled well-reservoir model to
analyse liquid loading and de-liquification techniques, with a particular focus on intermittent production for
tight gas wells (see also literature review of [18, 38]). Here, the key projects are listed which contributed to the
development of such a fully coupled well-reservoir model.
In 2005 Shell, Delft University of Technology and TNO initialised a research framework, named the “Inte-
grated System Approach Petroleum Production” (ISAPP) knowledge centre. Their research aims to increase
hydrocarbon recovery through innovative technologies. Within this research frame work, a number of re-
searchers(e.g. [33, 40, 49]) started investigating the behaviour of wells-reservoirs transients. To simulate dy-
namic phenomena in the well and in the reservoir, they used two existing software packages, respectively
OLGA (Schlumberger) for the well and Mores (Shell) for the reservoir. Due to the explicit coupling approach,
the numerical simulator suffered from stability issues.
1.5. Research goal and outline of the report 7

A collaboration between Shell, the SPT Group and TNO [27] performed a research project using the exist-
ing software package OLGA-ROCX to couple the well and the near wellbore region (a relatively small region
around the well). They showed how coupled simulations could be used for long term production optimisa-
tion of wells, which are almost or already liquid loaded.
In 2014, in the Multiphase flow group, part of Shell’s Pipelines, Flow Assurance and Subsea (PFAS) depart-
ment, a master research project [37] was initialised to develop a fully coupled simulator using the in-house
dynamic multiphase flow simulator, Compas (Shell), and to extended it with a simplified dynamic near well-
bore reservoir model. Prior to coupling the reservoir model to Compas, a prototype code was developed to
investigate numerical aspects of various coupling techniques. A single phase reservoir and single phase well
model have been developed in a Matlab environment to allow more freedom of coupling techniques and to
allow for easy model modifications. Rajkotwala [45] continued with this project, focussing on stability and
numerical efficiency of various coupling techniques. The coupled simulator has been successful in reproduc-
ing water hammer results, obtained from field data of a water injection well in the Gulf of Mexico. However,
before coupling the near wellbore code reservoir code to Compas, it has to be extended to multiphase flow,
since this creates additional modelling challenges that are desired to investigate in the Matlab prototype en-
vironment.

1.5. Research goal and outline of the report


The research goal of this study is formulated as follows:

Extend the Shell singe phase well-reservoir simulator to two-phase flow for simulation of well-reservoir tran-
sients.

This research is conducted in the team of Shell Projects & Technology, as part of a master graduation project
at the Delft University of Technology. The project has been finalised in a period of 11 months, including a
literature study of two months [18]. The project has been structured in the following way:

• First, a literature study was carried out to (1) review the fundamentals of multiphase flow in well reser-
voir engineering and (2) to review well-reservoir coupling methods found in open literature (see Doǧan
[18]). The literature study identified the need for a coupled well-reservoir model in multiphase flow
conditions. Shell’s desire to optimise existing recovery techniques has two components in which the
coupled simulator simulator is essential: simulation of liquid loading and the design of smart wells.

• Second, the two-phase reservoir model has been derived and discretised in chapter 2. The reservoir
equations are formulated for a radial coordinate system to confine the reservoir model to the near
wellbore region. Since the reservoir model has been extended from single phase to two-phase flow,
its mathematical character has been thoroughly analysed. The spatial and temporal discretisation of
the model has been verified in chapter 3.

• Third, an existing well model has been modified to a drift flux model, as presented in chapter 4. The
drift-flux model is based on the formulations of Hendrix [25], Pan et al. [41, 42], Schutte [51], Shi et al.
[52].

• Last, the coupling between the well and the reservoir has been designed and implemented in sec-
tion 4.2. The benefit of the dynamic coupled simulation over steady state models has been demon-
strated with the aid of two different (academic) test cases in section 4.3.
2
Reservoir modelling
This chapter presents the governing equations for the reservoir model, which will be used in subsequent
chapters to simulate reservoir-well flows. The reservoir model is based on classical fluid dynamics, as given
by the Navier-Stokes equations, which covers the conservation of mass, momentum and energy. However,
in contrast to the Navier-Stokes equations, the momentum balance in the reservoir is given by an empirical
relations, named "Darcy’s Law". Constitutive relations, such as the Equations Of State (EOS), are required to
close the system.
This chapter will be structured as follows. First, an introduction will be given to Darcy’s law in section 2.1,
followed by a brief overview of typical fluid and rock properties in sections 2.2 and 2.3, respectively. The
partial differential equations (PDE) for a single phase reservoir will be presented in section 2.4, and extended
to a two-phase formulation in section 2.5. The classification of these PDEs and their mathematical character
will be discussed in section 2.6. Finally, section 2.7 explains the spatial and temporal discretisation of the
reservoir model.

2.1. Darcy’s Law


The founding father of reservoir engineering is Henry Darcy, who published his most important work Les
Fontaines Publiques de la Ville de Dijon in 1856 [17]. His work described an empirical relation for fluid flow
through porous media. The generalised form of this relation, Darcy’s law, is given by:

K0 ¡
∇p − ρg ,
¢
u =− (2.1)
µ

where u(x, t ) is the velocity vector, K0 (x) is the permeability tensor, which is a property solely dependent on
the nature of the permeable material, µ is the fluid viscosity (assumed to be constant for isothermal condi-
tions), ρ(x, t ) is the fluid density, p(x, t ) is the fluid pressure and g is the gravitation vector.

2.2. Fluid properties


This section presents a brief overview of fluid properties. First, a brief introduction will provided of the
physics of fluid and rock properties, followed by a description of the models used in this study.

The two-phase reservoir model is intended to solve the system for both the gaseous and liquid phases. The
constitutive relations that describe the thermodynamic state of the fluids are called Pressure Volume Tem-
perature (PVT) relations. For ideal gases, the PVT-relation is given by the well-known ideal gas law. For a
real gas, a wide variety of analytical expressions exists, such as the Van der Waals, the Beattie-Bridgeman and
Benedict-Webb-Rubin equations. Unfortunately, no simple EOS exists which describes the PVT-properties of
oil, and therefore the relations have to be determined from experiments. Depending on the reservoir pres-
sure, the hydrocarbons can exist solely in the liquid phase, or in both the gas and liquid phase. This is deter-
mined by the bubble point of the oil mixture. A typical pressure-temperature diagram for a multi-component
hydrocarbon mixture is presented in fig. 2.1. If the mixture conditions are above the bubble point curve, the

9
10 2. Reservoir modelling

Figure 2.1: Phase diagram for a multi-component hydrocarbon mixture, obtained from Dake [16].

fluid exists solely in the liquid phase, whereas below the bubble point curve, gas evaporates from thel liquid
phase yielding hydrocarbons in both the gas and liquid phase. Below the dew point curve the hydrocarbons
are solely gas. When the temperature and pressure are very high, i.e. above the critical point, a fluid mixture
exists in which the fluid is both gas and liquid [16].
During transport from the reservoir along the well to the surface, the total gas production consists of free
gas in the reservoir and additional evaporated gas from the oil phase. This leads to an increase of the total
volume of gas and a reduction of the total volume of liquids of the hydrocarbons. On the other hand, under
high pressure and temperature conditions, water tends to be dissolved in the vapour phase. When the water
vapour is transported upwards, it will condense due to pressure and temperature reduction. As a result,
during the transport of fluids from the reservoir to the surface the following occurs:

• gas evaporates out of the liquid oil phase;

• the volumetric flow rate of liquid oil will reduce;

• In addition to free water in the reservoir (if any), condensate water will be produced due to condensa-
tion of the water vapour;

• the volumetric rate of the gas phase will increase due to the strong decrease of the pressure along the
well towards surface.

In reservoir models the above mentioned effects are described by a hierarchy of increasingly complex fluid
models (see Chen et al. [11]). The reservoir model used in this study has two immiscible phases, which means
that solubility and vaporisation are not allowed. In addition, the model is isothermal, which means that
temperature effects are neglected. The two-phase reservoir model will be further explained in section 2.5.
The subsequent subsections present the EOS for water, under-saturated oil and ideal gas.

2.2.1. EOS Water


Fluids can be considered as slightly compressible. When isothermal conditions are assumed, the fluid com-
pressibility c f and viscosity µ can assumed to be constant. The density changes according to:

ρ w sc
ρw = , (2.2)
1 − c f w (p w − p w sc )

where ρ w sc is the water density at standard pressure, p w sc , and c f w the water isothermal compressibility.
2.3. Rock properties 11

Table 2.1: Typical values of permeability K0 , obtained from Bear [5].

permeability pervious semi- impervious


pervious
aquifer good poor none
well well fine sand, loess, loam
soils
sorted sorted
gravel sand
peat layered unweathered
clay clay
rock highly fractured oil reservoir sandstone limestone granite
rocks rocks
K0 (cm2 ) 10−3 10−4 10−5 10−6 10−7 10−8 10−9 10−10 10−11 10−12 10−13 10−14 10−15
K0 (mD) 108 107 106 105 104 103 102 10 1 10−1 10−2 10−3 10−4

2.2.2. EOS under-saturated oil


Under-saturated oil assumes that the oil exists solely in the liquid phase and that the pressure is far above
the bubble point pressure, i.e. a reduction in pressure does not lead to a co-production of gas. Similar to the
water model the viscosity and fluid compressibility can assumed to be constant under isothermal conditions.
The oil density changes according to:

ρ o sc
ρo = . (2.3)
1 − c f o (p o − p o sc )

2.2.3. EOS ideal gas


The density change for an ideal gases is given by:

pg
ρg = , (2.4)
nRT

where n is the molecular weight of the gas, R the universal gas constant and T the absolute temperature (at
standard conditions for the isothermal models used in this study). The viscosity is assumed to be constant
for isothermal conditions.

2.3. Rock properties


The properties of the rock formation have a large effect on the ability of the fluids to flow through the porous
medium. Depending on the wettability of the fluids the permeability can differ several orders in magnitude.
In table 2.1 an overview is presented of the absolute permeability K0 for various types of soils and rocks.
In this table the soils are considered as unconsolidated materials, whereas the rock types are considered as
consolidated materials. The permeability is given in both SI-units, and in Darcy units D 1 , of which the latter
is commonly used in petroleum engineering. This study is focussed on simulating unloading events for tight
gas reservoirs (sandstone) and flooding for typical oil reservoirs. The permeability is therefore in the range
K0 = 1 − 1 × 104 mD.

2.3.1. Capillary pressure


In immiscible fluids, a discontinuity in the fluid pressure exists across the interface of the respective flu-
ids. This is a consequence of the interfacial tension that exists at the interface [11]. Near the interface, the
molecules are unevenly attracted by their neighbours, since the molecules will experience a preferential at-
traction towards themselves. This results in a free surface energy or interfacial tension [16]. If the interface
is curved, the pressure of the concave side will be larger than the pressure at the convex side. This pressure
difference is defined as the capillary pressure p c .

1 Conversion of Darcy units to SI-units is done by using 1 D = 9.869 233 × 10−13 m2


12 2. Reservoir modelling

(a) Imbibition. (b) Drainage.

(c) Capillary curve.

Figure 2.2: Hysteresis in contact for an imbibition and drainage process and matching the capillary pressure curve, obtained from Dake
[16].

When two immiscible fluids, such as oil and water, are in contact with a rock surface, one fluid will adhere
more to the rock than the other. The fluid that adheres most to the rock is the wetting phase, while the other
is the non-wetting phase. The wetted fluid will displace the non-wetted fluid. However, not all of the non-
wetted fluid will be displaced, since the fluids will also experience an internal attraction that counteracts the
displacing force. As a result, a stable situation will occur, in which the wetted phase will occupy the majority
of the solid surface. At this equilibrium, the contact angle Θ will be the measure of the balance between the
cohesive forces [22].
If water displaces the oil, the wetting phase increases. This is known as imbibition and shown in fig. 2.2a.
When the water saturation decreases, the oil displaces the water, which is called drainage, and is shown in
fig. 2.2b. The contact angle for imbibition and drainage is different and in addition depends on the sequence
in which the processes take place. This is called "hysteresis".
Imbibition and drainage can be visualised using a capillary pressure curve (see fig. 2.2c). Starting at point A
the voids are completely saturated with water. When moving along the dashed curve to point B, the water
is displaced by the oil, which indicates the drainage process. The minimum saturation point B is known as
the connate water saturation point. Irrespective of a capillary pressure increase, the water saturation will not
reduce. When starting from point B and moving along the black curve to point C, the process is described
by an imbibition process. At point C the residual oil saturation is at its minimum. Irrespective of a decrease
in capillary pressure, the oil saturation will not reduce. The drainage and imbibition curves differ due to the
hysteresis of the contact angle of the respective fluids [16]. The connate water saturation and residual oil
saturation are the maxima of the domain in which the capillary curve in practise exists. This will be of impor-
tance when the relative permeability is discussed in section 2.3.2. The capillary pressure curve usually has to
be determined from experiments. However, extensive research has shown that the capillary pressure can be
assumed to be a function of the wetting phase saturation. In the simulations of the reservoir phenomena two
capillary models have been used.
2.3. Rock properties 13

Brooks-Corey’s capillary pressure


This model was introduced by Brooks and Corey [9]:

−1
p c (S w ) = p d S eλw , (2.5a)
S k − S kr
S ek = P , k = n, w, (2.5b)
1 − S αr
α

where p d is the entry pressure, λ a distribution parameter related to the pore sizes, S e k the effective saturation
for phase k and S αr the residual saturations of the respective phases.

Chen’s capillary pressure


This model was introduced by Chen et al. [11]:

S w − S wr + ²
µ ¶
p c (S w ) = B + ln , (2.6a)
1 − S wr
pd
B= , (2.6b)
ln (²/(1 − S wr ))
where ² is a small positive number, p d is the entry pressure and S wr is the residual wetting phase saturation.

2.3.2. Relative permeabilities


The ability of the fluids to flow through the permeable rock is known as the permeability K0 . When mul-
tiple fluids flow simultaneously, the respective fluids limit the mobility of each other, yielding an effective
permeability Kk . The relation between the absolute and the effective permeability is given by:

Kk = k r k K0 , k = n, w, (2.7a)
where k r k is the relative permeability of phase k. The relative permeabilities are directly related to the capil-
lary forces, which result in similar curves as shown in section 2.3.1. A typical relative permeability curve for
oil and water is presented in fig. 2.3. There are four points to be considered. The first point considers the
rock to be completely saturated with water (S w = 1), such that k r w = 1. One the other hand, the rock can
also be completely saturated with oil (S w = 0 or S o = 1), hence k r o = 1. However, in practice, a minimum
saturation exists for which the water and oil will not flow. If the water saturation equals the connate water
saturation (S w = S wc ), the water will not flow and k r w = 0. The minimum oil saturation for the oil to flow is
given by the residual oil saturation (S w = 1 − S or ) and k r o = 0. The saturations below S wc and above 1 − S or
are drawn as dashed lines, since they will never be encountered in practice. This yields the practical range
S wc ≤ S w ≤ 1−S or . The maximum and minimum relative permeabilities are also known as end-point relative
permeabilities:

0
k r o = k r o (S w = S wc ), (2.7b)
0
k r w = k r w (S w = 1 − S or ). (2.7c)
Due to the close relation between the capillary pressure and relative permeability, the relative permeability
is assumed to be a function of the wetting phase saturation. The models used in the simulations performed
later in this report are listed subsequently and shown in fig. 2.4. Note that for these simulations S wc = S or = 0,
unless stated otherwise.

Brooks-Corey’s relative permeability


This model was introduced by Brooks and Corey [9]. The relative permeability functions are given by:

−2+3λ
k r w (S w ) = S e wλ , (2.8a)
2+λ
³ ´
k r n (S w ) = S e n 1 − (1 − S e n ) λ , (2.8b)
14 2. Reservoir modelling

Figure 2.3: The effective (left) and relative (right) permeabilities, as a function of the water saturation S w for two-phase flow, obtained
from Dake [16].

where again λ is a distribution parameter related to the pore sizes and S e k the effective saturation for phase
k.

Corey’s relative permeability


This model is introduced by Corey [13]. The model for a two-phase reservoir is given by:

λ
k r w (S w ) = k r w max S e ww , (2.8c)
λ
k r n (S w ) = k r nmax S e nn , (2.8d)

where k r kmax are the end-point relative permeabilities of the respective phases.

2.3.3. Rock compressibility, consolidation and compaction


The compressibility of rock can be explained as the reduction in bulk volume of material due to the displace-
ment of the gaseous phase. A similar process is called consolidation, which is the densification of the soil due
to the expulsion of liquids without the replacement of the pore volumes by the gas phase. As a result, the soil
particles are packed together more tightly, which increases the mechanical strength of the material due to the
particle interaction. The combination of compressibility and consolidation is called "compaction", which
can be regarded as the decrease in bulk volume due to the removal of the gaseous and liquid phases. When
the compaction is large, the soil may subside, yielding dangerous situations in which earthquakes may occur.
In practice, the compaction is usually neglected and the porosity is assumed to be constant. The reservoir
model of this study assumes constant porosity.

2.4. Single phase flow


Reservoir models describe flow through porous media. The single phase reservoir model is based on con-
servation of mass and conservation of momentum (according to Darcy’s law), and additional constitutive
relations. The equation for conservation of energy is not considered as isothermal conditions have been as-
sumed. Only a part of the porous material is occupied by the fluids, as given by the porosity φ. The governing
equations for conservation of mass and momentum are then given by:


Z Z
ρφ d Ω + ρu · n d S = Q, (2.9a)
∂t Ω S
K0 ¡
∇p − ρg .
¢
u =− (2.9b)
µ
2.4. Single phase flow 15

·104 1
Chen ² = 0.01 k r n Corey λ = 2
8 Brooks-Corey λ = 2 k r w Corey λ = 2
0.8 krn Brooks-Corey λ = 2
kr w Brooks-Corey λ = 2
6
0.6

kr
pc

4
0.4

2
0.2

0
0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
Sw Sw

(a) Capillary pressure curves (b) Relative permeability curves.

Figure 2.4: Relative permeability and capillary pressure curves used in this study.

The flux entering or leaving the control volume can be re-written using Gauss’s Theorem. Taking the time
derivative inside the integral, eqs. (2.9a) and (2.9b) become:

∂(ρφ) ρK0 ¡
Z · µ ¶¸
∇p − ρg d Ω = Q.
¢
−∇· (2.9c)
Ω ∂t µ
Taking the limit of an infinitely small volume, the following differential equations should be respected at all
times:

∂(ρφ) ρK0 ¡
µ ¶
∇p − ρg = q,
¢
−∇· (2.9d)
∂t µ

where q is a source or sink per unit volume, i.e. Q = q d Ω. This resulting non-linear differential equation
R

describes diffusive transport through porous media. The non-linearity of this equations stems from pressure
dependency on the porosity, density, and fluid viscosity. Under the assumptions of isothermal flow, the first
term of eq. (2.9d) can be expanded, such that:

∂(ρφ) ∂φ ∂p ∂ρ ∂p
=ρ +φ . (2.10a)
∂t ∂p ∂t ∂p ∂t
The isothermal compressibility of the rock and the fluid are defined as:

1 ∂φ ¯¯ 1 ∂ρ ¯¯
¯ ¯
cR ≡ cf ≡ , (2.10b)
φ ∂p ¯T ρ ∂p ¯T
such that eq. (2.10a) can be expressed as:

∂(ρφ) ∂p
= ρφ(c R + c f ) , (2.10c)
∂t ∂t
where the total compressibility is defined as c t = c R + c f . The diffusion equation eq. (2.9d) can then be ex-
pressed as a non-linear parabolic differential equation in the pressure p:

∂p ρK0 ¡
µ ¶
ρφc t ∇p − ρg + q,
¢
= ∇· (2.11a)
∂t µ
16 2. Reservoir modelling

with initial and boundary conditions give by:

p(x, 0) = p 0 (x) x ∈ Ω, (2.11b)


p(x, t ) = p d (x, t ) on Γd , ρu · n = Φ(x, t ) on Γn , (2.11c)

where Γd and Γn refer to the Dirichlet (subscript d ) and Neumann boundary (subscript n) on a boundary
segment of the porous medium domain Ω; Γ = ∂Ω, Γd ∪Γn = Γ. Equation (2.11a) is a classic diffusion equation
to the parabolic type, which means that boundary conditions have to be prescribed on all sides of the domain.
Equation (2.11a) is valid for single phase compressible fluids. The mathematical character changes to elliptic
for incompressible fluids, which makes the initial condition superfluous.

2.5. Two-phase immiscible flow


In general multiple phases will flow simultaneously through the porous medium. In this section two immis-
cible phases are considered, i.e. there is no mass transfer between the phases. When the two phases flow
through the porous medium one phase wets the medium more than the other, where wetting is defined as
the ability of the fluid to adhere to a solid surface. The wetting phase will be denoted with the subscript w
(e.g. water) and the non-wetting phase will be denoted with the subscript n (e.g. oil or gas). The reservoir
model assumes that the pore volumes are completely saturated with fluids, and therefore the following rela-
tion should hold for the phase saturations:

S n + S w = 1, (2.12a)

where S n and S w are the phase saturations (volume fractions of the fluid) of the non-wetting and wetting
phase respectively. Equation (2.12b) defines the capillary pressure:

p n − p w = p c (S w ). (2.12b)

Empirical observations have shown that the capillary pressure is a function of the wetting saturation S w [11]
and commonly used models were presented in section 2.3. For two-phase flow, conservation of mass should
hold for each phase. Furthermore, Darcy’s law is assumed to hold on a per-phase basis. Since the phases are
assumed to be immiscible, one obtains:

∂(ρ k φS k )
+ ∇ · (ρ k u k ) = q k k = n, w, (2.12c)
∂t
kr
u k = k v k , v k = −K0 ∇p k − ρ k g ,
¡ ¢
(2.12d)
µk

where k denotes the phases n and w. The effective permeability is defined as the product of the relative
permeability k r k and the absolute permeability tensor K0 (eq. (2.7a)). The relative permeability indicates the
tendency of phase k to wet the porous medium. The relative permeability and capillary pressure have been
explained in section 2.3.

The entire system of eq. (2.12a)-eq. (2.12d) has to be solved for the phase pressures and phase saturations p n ,
p w , S n and S w . Alternative formulations of eq. (2.12a)-eq. (2.12d) exist, which can be beneficial depending on
the discretisation method employed. In addition, the non-linearity and strong coupling between the phase
pressures and saturations can be reduced using these different formulations (with additional assumptions).
Two important formulations will be presented:

• Phase Pressure Saturation formulation (PPS);

• Global Pressure formulation (GPIMPES).

A wider discussion on the applicability for the different formulations can be found in Bastian [4], Chen et al.
[11], but the most important reason for the two chosen formulation will be discussed subsequently. Due to
the shapes of the capillary pressure functions (see section 2.3.1), problems can be expected near the limits
of the saturations, S w = 0 and S w = 1, since p c is there unbounded (see for example fig. 2.4a). According
2.5. Two-phase immiscible flow 17

to Bastian [4], which uses the same capillary pressure models as presented in section 2.3.1, the primary un-
knowns p w and S n should be chosen when S n is bounded away form 1, while the primary unknowns p n and
S w should be chosen when S w is bounded away form 1. The primary unknowns of the PPS formulation are
therefore dependent on the range of expected saturations of the typical problem.
The GPIMPES avoids (some) of these difficulties. This is more widely discussed by Chavent and Jaffré [10].
In addition, the GPIMPES formulation results in a pressure and saturation equation, in which the coupling
between the pressure and saturation is weaker (only trough λ), which makes them typically easier to solve.
This will become clear in section 2.5.2.

The system eq. (2.12a)-eq. (2.12d) has to be closed with closure relations for the fluid properties ρ k , µk and
rock properties φ, k r k . The closure relations have been discussed in sections 2.2 and 2.3.

2.5.1. Phase Pressure Saturation formulation


In the PPS-scheme the system is expressed in one unknown phase pressure and one unknown phase satu-
ration. The remaining unknowns can be solved using the constraints of eqs. (2.12a) and (2.12b). Here, the
unknowns p n and S w are chosen. The system of eq. (2.12a) -eq. (2.12d) becomes:

∂ ρ n φ(1 − S w )
¡ ¢
+ ∇ · ρ n un = qn ,
¡ ¢
(2.13a)
∂t
kr
u n = n v n , v n = −K0 ∇p n − ρ n g ,
¡ ¢
(2.13b)
µn
∂ ρ w φS w
¡ ¢
+ ∇ · ρw uw = qw ,
¡ ¢
(2.13c)
∂t
kr d pc
µ ¶
u w = w v w , v w = −K0 ∇p n − ∇S w − ρ w g . (2.13d)
µw d Sw
Initial and boundary conditions are given by:

p n (x, 0) = p n0 (x) x ∈ Ω, (2.14a)


S w (x, 0) = S w0 (x) x ∈ Ω, (2.14b)
p n (x, t ) = p nd (x, t ) on Γnd ρ n u n · n = Φn (x, t ) on Γnn , (2.14c)
S w (x, t ) = S wd (x, t ) on Γwd ρ w u w · n = Φw (x, t ) on Γwn , (2.14d)

where Γnd and Γnn are the Dirichlet and Neumann boundary for phase n and Γwd and Γwn are the Dirichlet
and Neumann boundary for phase w. The dependency of fluid and rock properties on the primary unknowns
is as follows:

φ = φ(p n ), µn = µn (p n ), ρ n = ρ n (p n ), (2.15a)
µw = µw (p w ), ρ w = ρ w (p w ), (2.15b)
p c = p c (S w ), k r n = k r n (S w ), k r w = k r w (S w ). (2.15c)

2.5.2. Global Pressure IMPES formulation


The governing equations of eq. (2.12a)-eq. (2.12d) are strongly coupled. To reduce the coupling a Global
Pressure can be defined [1]. First we define the phase mobilities:

krk
λk = , k = n, w, (2.16)
µk
and the total mobility λt and fractional flow function f k as:

λt = λn + λw , (2.17a)
λk
fk = k = n, w. (2.17b)
λt
18 2. Reservoir modelling

Note that λk = λk (S w , p k ) and f k = f k (S w , p k ). A global pressure can be obtained by summing the phase
velocities of eqs. (2.13b) and (2.13d) as follows:

u t = un + u w , (2.18a)
= −K0 λn ∇p n + λw ∇p w − (ρ n λn + ρ w λw )g ,
¡ ¢
(2.18b)

and requiring that the sum of the two pressure gradients can be expressed in a single pressure gradient:

λt ∇p = λn ∇p n + λw ∇p w . (2.19)
Rewriting this equation, and substituting eq. (2.12b) gives:

∇p = f n ∇p n + f w ∇p w
= f n ∇p n + f w (∇p n − ∇p c ) (2.20)
= ∇p n − f w ∇p c .

With a global pressure defined as:

Z p c (S w )
f w p c−1 (ξ) d ξ,
¡ ¢
p = pn − (2.21)

the total velocity can be expressed in terms of a single pressure:

u t = −K0 λt ∇p − (ρ n f n + ρ w f w )g .
¡ ¢
(2.22)

Expanding the time derivative in eq. (2.12c) and division by the phase density ρ k leads to:
∂S k 1 ∂ρ k ∂φ 1 qk
φ + φS k + Sk + ∇ · (ρ k u k ) = . (2.23a)
∂t ρ k ∂t ∂t ρ k ρk
Summation over all phases, and employing the constraint eq. (2.12a), leads to:

∂p k ∂φ X 1 X qk
φ
X
S k c f ,k + + ∇ · (ρ k u k ) = . (2.23b)
k ∂t ∂t k ρk k ρk
When assuming incompressibility (c f ,k = 0) and constant porosity φ, and neglecting sources or sinks, conser-
vation of the total velocity u t is obtained:

∇ · ut = 0 . (2.24)
Note that the time derivatives with respect to the global pressure and saturation are eliminated due to the
use of the constraint of eq. (2.12a). Using eq. (2.23a) for the wetting phase, the result of eq. (2.24) and the
additional relations between total velocity and phase velocities of Chen et al. [11] given by:

u w = f w u t + K0 λn f w ∇p c + K0 λn f w (ρ w − ρ n )g , (2.25a)
u n = f n u t − K0 λw f n ∇p c + K0 λw f n (ρ n − ρ w )g , (2.25b)

the saturation equation for the wetting phase is obtained:

∂S w d pc ∂φ q w
µ µ ¶ ¶
φ + ∇ · K0 λn f w ∇S w − (ρ n − ρ w )g + f w u t = −S w + . (2.26)
∂t d Sw ∂t ρ w
Note that λn f w = λw f n and f n + f w = 1. Initial and boundary conditions are given by:

S w (x, 0) = S w0 (x) x ∈ Ω, (2.27a)


p(x, t ) = p d (x, t ) on Γnd ρ n u n · n = U (x, t ) on Γnn , (2.27b)
S w (x, t ) = S wd (x, t ) on Γwd ρ w u w · n = Φw (x, t ) on Γwn . (2.27c)
2.5. Two-phase immiscible flow 19

The fluid and rock properties are defined as:

µn , µw , ρ n , ρ w are constant, (2.28a)


φ = φ(p), (2.28b)
p c = p c (S w ), k r n = k r n (S w ), k r w = k r w (S w ). (2.28c)

The total velocity eq. (2.24) is still depending on the saturation S w via the fractional flow functions f n and f w .
When the fractional flow functions are evaluated explicitly (using S w from the previous time step), eq. (2.24)
can be solved implicitly for the global pressure. Once the global pressure is evaluated eq. (2.26) can be used to
find the saturation of the next time step. This sequential order of solving is referred to as the Implicit Pressure
Explicit Saturation (IMPES)-scheme. The governing equations are now weakly coupled, which is proficient in
terms of solving the system [11]. The explicit evaluation of the saturation on its hand can introduce numeri-
cal stability issues. For the full derivation of the global pressure formulation and IMPES-scheme the reader is
referred to Antontsev [1] and Chen et al. [11].

2.5.3. The Buckley-Leverett equation


The solution to the two-phase immiscible flow problem has to be found using a numerical solver, since the
analytical solution is generally unknown. However, under particular assumptions, the saturation equation of
eq. (2.26) can be recast into the "Buckley-Leverett" equation, which has an analytical solution. The Buckley-
Leverett equation is used to describe fluid displacement processes for immiscible fluids. In classic reser-
voir engineering problems, the Buckley-Leverett equation is used to describe oil-water displacement. Water
flooding, or secondary recovery, can be used to exploit additional oil recovery by injecting water from an
additional injection well. This will be demonstrated in section 3.4.2 and section 4.3.2.
To derive the Buckley-Leverett equation, we closely follow existing literature [4, 11, 34, 55, 57]. With the as-
sumption of incompressible flow (and rock) and when also neglecting the capillary pressure (p c = 0) and
gravity, eqs. (2.22), (2.24) and (2.26) are considerably simplified:

qn q w
∇ · ut = + , (2.29a)
ρn ρ w
∂S w ¢ qw
φ
¡
+ ∇ · f w ut = . (2.29b)
∂t ρw

Note that the global pressure p, the pressure of the non-wetting phase p n and the wetting phase p w are
identical due to the negation of the capillary pressure and will be further denoted with p. When no sources
or sinks are present, the divergence of the total velocity is zero:

∇ · u t = −∇ · λt K0 ∇p = 0.
¡ ¢
(2.30)

Equation (2.30) is an elliptic equation in the pressure (Laplace equation). There is no time dependence in this
equation. Given a saturation field, λt = λt (S w ) can be evaluated, and p follows from eq. (2.30). The saturation
equation of eq. (2.29b) can be further simplified using the total velocity of eq. (2.30). Expansion leads to:

∂S w qw
φ + ut · ∇ f w = . (2.31a)
∂t ρw
Assuming that no sources and sinks are present and that the fractional flow function f w is a function of the
saturation leads to:

∂S w 1 ∂ f w
+ u t · ∇S w = 0. (2.31b)
∂t φ ∂S w

Equation (2.31b) is a hyperbolic equation in the wetting saturation known as the Buckley-Leverett equation.
In one dimension, and in Cartesian coordinates, the Buckley-Leverett equation reads:
20 2. Reservoir modelling

∂S w u t ∂ f w ∂S w
+ = 0, (2.32)
∂t φ ∂S w ∂x
and u t is constant due the incompressibility constraint. This equation is in the quasi-linear form:

S w t + g 0 (S w )S w x = 0, (2.33)
∂f
where the characteristic velocity is given by g 0 (S w ) = uφt ∂Sww , and the flux function by g (S w ) = u t f w /φ, which
is generally non-convex (see fig. 2.5). A typical form (based on Corey permeability curves, see section 2.3.2)
is given by:

S 2w
g (S w ) = , (2.34)
S 2w + M (1 − S w )2
where M is the ratio of the fluid viscosities M = µw /µn , also known as the frontal displacement ratio. The
non-convexity is important, because in contrast to convex flux functions, the Riemann solution to a non-
linear scalar problem with a non-convex flux function can involve both shock and rarefaction waves [34]. For
convex functions, the derivative with respect S w would be either monotone increasing or decreasing, which
limits the Riemann solution to either shocks or expansion waves. The Riemann solution will be explained
subsequently.
In case a shock is present, the shock velocity is given by:

‚g ƒ
v shock = . (2.35)
‚S w ƒ
Furthermore, the solution is constant along characteristics with

dx ut ∂ f w
= g 0 (S w ) = , (2.36)
dt φ ∂S w
which are straight lines, as long as the solution remains smooth. Note that the solution trajectories (particle
paths) are not straight lines. Given an initial solution S w,0 (x), the solution at a later time follows as:

S w (x, t ) = S w,0 (ξ), (2.37)

where

x = ξ + g 0 (S w,0 (ξ))t . (2.38)

2
g
g0
g /S w
1.5

g 0 (α)
g (S w )

1
g (α)

0.5

0
0 0.2 0.4 0.6 0.8 1
Sw

Figure 2.5: Fractional flow function g and its derivative g 0 . The tangent to the curve g originating from g /S w gives the point S w = α. For
values larger than α the solution consists of a shock followed by a rarefaction wave.
2.6. PDE classification 21

1 1
t = 2025 [sec] t = 2025 [sec]
t = 1377 [sec] t = 1377 [sec]
0.8 t = 729 [sec] 0.8 t = 729 [sec]
t = 81 [sec] t = 81 [sec]

0.6 0.6
S w [-]

S w [-]
0.4 0.4

0.2 0.2

0 0
0 20 40 60 80 100 120 140 0 20 40 60 80 100 120 140
x [m] x [m]
(a) Case I: single shock solution for S Lw = 0.5 and S R
w = 0. this (b) Case II: rarefaction wave followed by a shock solution for
is white text to fill the space S Lw = 1 and S R
w = 0.

Figure 2.6: Analytical solutions of the Buckley–Leverett problem with quadratic Corey relative permeabilities (λ=2).

When a shock appears, a triple-valued solution results. The correct solution is found by the equal area
rule[34], leading to:

g (α)
x shock = t, (2.39)
α
p
where α = M /(M + 1), which can be found graphically at the point where the tangent to the curve g equals
the tangent of the line originating from the origin (g 0 (α) = g (α)/α). This has been demonstrated in fig. 2.5.
Depending on the saturation of the injected fluid, the solution to the Buckley-Leverett equation can be clas-
sified into two categories:

• case I: 0 < S w,0 ≤ α, single shock

• case II: α < S w,0 ≤ 1, rarefaction wave followed by shock

Consider the Riemann problem with a single discontinuity at x = 0 as initial condition:

(
S Lw , x ≤ 0
S w,0 = , S Lw > S Rw . (2.40)
S Rw , x > 0

S Lw is the left state and S Rw is the right state. Given a total velocity u t > 0 the solution is either given by a shock
discontinuity (case I) or a rarefaction wave followed by a shock discontinuity (case II). Both type of solutions
are presented in fig. 2.6, anticipated on the parameter set of section 3.4.

Solution for the pressure


The pressure can be obtained solving eq. (2.30), given an initial saturation distribution. When an initially
discontinuous saturation profile is considered, such as the Riemann problem, the solution to this equation in
Cartesian coordinates consists simply of two linear pressure profiles with different slopes, connected at the
point of discontinuity. The initial pressure distribution of case I for S Lw = 0.5 and S Rw = 0 with the discontinuity
at x = 100 m is presented in fig. 2.7.

2.6. PDE classification


Consider the system of equations describing immiscible two-phase flow in a reservoir in Cartesian coordi-
nates. In the subsequent sections will be demonstrated how the mathematical character of the governing
22 2. Reservoir modelling

·105
2.5 1
p0 S w0

2.4 0.8

2.3 0.6

p [Pa]

S w [-]
2.2 0.4

2.1 0.2

2 0
0 50 100 150 200 250 300
x [m]

Figure 2.7: Initial pressure distribution for case I with S Lw = 0.5 and S R
w = 0 at the discontinuity at x = 100 m.

equations can be evaluated using eigenvalue analysis. In addition, it will be demonstrated that the math-
ematical character changes under particular assumptions. The mathematical character is of importance,
since it describes the nature of a particular process. With respect to fluid flow, the mathematical character
indicates the type of information propagation (e.g. convection, diffusion, acoustic waves). The (dominant)
type of information propagation is important with respect to the discretisation method and solutions strategy
employed. Since the mathematical nature of two-phase reservoir flow changes under particular assumptions,
different discretisation schemes have been employed.
For simplicity we will use the system of equations as presented in section 2.5.1, but neglecting gravity and
source terms. If source terms do not contain derivatives, they do not influence the characteristic analysis.
Gravity, however, leads to derivatives of the mobility, and has an effect on the characteristic analysis (see for
example Cunha et al. [14]). Before we continue with the eigenvalue analysis of the system of equations, a
brief introduction will be given on eigenvalue analysis theory. The theory described below closely follows
Loret [35].

Consider a first-order linear scalar PDE for the unknown function u(t , x) of the form:

∂u ∂u
a +b + c = 0. (2.41)
∂t ∂x
This equation is said to be

• linear if a, b and c are functions of t and x,

• quasi-linear if a, b and c are functions of t , x and u,

• non-linear if these coefficients depend further on the derivatives of u.

We look for characteristic curves along which the PDE becomes an ODE. The characteristic curves are parametrised
as t (s), x(s). Regarding u as a function of s, i.e. u(t (s), x(s)), the change of u with s is given by:

du dt ∂u dx ∂u
= + . (2.42a)
ds ds ∂t ds ∂x
So, if the characteristic curves satisfy

dt dx
=a = b, (2.42b)
ds ds
then the solution u satisfies
2.6. PDE classification 23

du
+ c = 0, (2.42c)
ds
along these characteristic curves. If an initial solution u is known along a curve I 0 (not a characteristic), then
the characteristics can be computed by:

dx b
= , (2.43)
dt a

and the change in solution from eq. (2.42c). If c = 0, then u remains constant along a characteristic; this is
called a Riemann invariant [35].

Now consider a systems of first-order equations in quasi-linear form:

∂u ∂u
A +B + C = 0, (2.44)
∂t ∂x

where A, B and C are matrices which depend on the independent variables and on u(t , x). The mathematical
character of the system can be evaluated by substituting a wave-form, or, alternatively, looking for a combi-
nation of equations such that the a lower-dimensional characteristic system is obtained. As before, we look
for combinations of the n equations such that the system of PDEs reduces to ODEs:

∂u ∂u
µ ¶
du
l· A +B +C = p · + r. (2.45)
∂t ∂x ds
This is similar to diagonalisation of linear equations. Using either wave-form substitution or looking for a
combination to reduce the dimension of the characteristic equation, the following determinant equation is
obtained:

det (An t + B n x ) = 0. (2.46)

After writing σ = −n x /n t , eq. (2.46) can be identified as the following generalised eigenvalue problem:

det (A − σB) = 0. (2.47)

The mathematical character is determined by the eigenvalues σ of the system, i.e. the system is

• elliptic, if the eigenvalues are solely imaginary,

• parabolic, if the eigenvalues are real and the system is defective,

• hyperbolic, if the eigenvalues are real and the system is non-defective.

A system of n equations is said to be defective if it does not generate Rn distinct eigenvectors, i.e. the dimen-
sion of the eigenvector space is fewer than the number of eigenvalues [35].

Many typical PDEs are of a higher-order that first-order in space or time. A reduction to first-order can be
achieved by introducing a new unknown v , e.g.

v = ∂u
(
∂2 u ∂x
f = → (2.48)
∂x 2 f = ∂v
∂x

This technique will be used to re-write the system of equations of two-phase immiscible flow in a reservoir
into its first-order quasi-linear form, and subsequently to determine the characteristics of the equations.
24 2. Reservoir modelling

2.6.1. General case: Compressible phases and non-zero capillary pressure


For convenience, we repeat equation eqs. (2.13a) and (2.13c), formulated in one-dimensional Cartesian co-
ordinates:

∂ ¢ ∂ ρ n krn ∂p n
µ ¶
ρ n φ(1 − S w ) − K0
¡
= 0, (2.49a)
∂t ∂x µn ∂x
∂ ¢ ∂ ρ w kr w ∂p n d p c ∂S w
µ µ ¶¶
ρ w φS w − K0
¡
− = 0, (2.49b)
∂t ∂x µw ∂x d S w ∂x
The vector of unknowns in this formulation is given by:
¶ µ
pn
u= . (2.50)
Sw
We will re-write these equations into a first-order quasi-linear form in terms of extended vector in order to
apply techniques to determine the character of the equations. The time derivatives are expanded as follows:

∂(ρ n φ(1 − S w )) ∂S w ∂φ ∂ρ n ∂p n
µ ¶
= −ρ n φ + ρ n (1 − S w ) + φ(1 − S w ) , (2.51a)
∂t ∂t ∂p n ∂p n ∂t
∂(ρ w φS w ) ∂φ ∂p c ∂ρ w ∂p c ∂S w ∂φ ∂ρ w ∂p n
µ ¶ µ ¶
= ρw φ − ρw Sw − φS w + ρw Sw + φS w . (2.51b)
∂t ∂p w ∂S w ∂p w ∂S w ∂t ∂p w ∂p w ∂t
The thermodynamic definitions of rock and fluid compressibility of the respective phases are given by:

1 ∂ρ n ¯¯ 1 ∂φ ¯¯
¯ ¯
c fn ≡ , c Rn ≡ , (2.52a)
ρ n ∂p n ¯T φ ∂p n ¯T
1 ∂ρ w ¯¯ 1 ∂φ ¯¯
¯ ¯
c fw ≡ , cRw ≡ . (2.52b)
ρ w ∂p w ¯T φ ∂p w ¯T
The total compressibility of the respective fluids is defined as:

c tn = c f n + c Rn , (2.52c)
c tw = c f w + cRw , (2.52d)

Using the thermodynamic definitions eqs. (2.52c) and (2.52d), eqs. (2.51a) and (2.51b) can be written as:

∂(ρ n φ(1 − S w )) ∂S w ³ ´ ∂p ¶
µ
n
= ρn φ − + (1 − S w )c tn , (2.53a)
∂t ∂t ∂t
∂(ρ w φS w ) ∂S w d p c ∂S w ∂p n
µ ¶
= ρw φ − S w ctw + S w ctw . (2.53b)
∂t ∂t d S w ∂t ∂t
Similarly the spatial derivative terms are expanded as follows, (and introducing the mobilities λk = k r k /µk )

∂ ∂p n ∂p n 2 ∂p n ∂S w ∂2 p n
µ ¶ · µ ¶ ¸
− ρ n λn K0 = −ρ n λn K0 c f n + c λn + , (2.54a)
∂x ∂x ∂x ∂x ∂x ∂x 2
∂ ∂p n d p c ∂S w ∂p n 2 d p c ∂p n ∂S w
µ µ ¶¶ · µ ¶ µ ¶
− ρ w λ w K0 − = −ρ w λw K0 c f w + c λw − c f w
∂x ∂x d S w ∂x ∂x d S w ∂x ∂x
d p c ∂S w 2 ∂2 p n d p c ∂2 S w
µ ¶ ¸
− c λw + − , (2.54b)
d S w ∂x ∂x 2 d S w ∂x 2
where

1 d λn
c λn = , (2.55a)
λn d S w
1 d λw
c λw = . (2.55b)
λw d S w
2.6. PDE classification 25

The system of equations is of second-order, and will be rewritten in first-order form by introducing the addi-
tional variables

∂p n ∂S w
, u4 = u3 =. (2.56)
∂x ∂x
Assembling and rearranging terms into matrix format gives:

ρ n φ(1 − S w )c tn ³ −ρ n φ
 
´ 0 0 pn


dp
 ρ w φS w c t w ρ w φ 1 − S w c t w d S wc 0 0 ∂ 
 Sw 
  
0 ∂t p n x
 
0 0 0
  
0 0 0 0 Swx
| {z } | {z }
A(u) u

∂p ∂p n
−ρ n λn K0 c f n ∂xn ³³ −ρ n λn K0 c λ´n ∂x −ρ n λn K0
 
0 pn
 
∂p n ∂p
−ρ w λw K0 c λw − c f w d S wc ∂xn − c λw d S wc ∂S
´
ρ w λw K0 d S wc  ∂ 
dp dp dp 
−ρ λ K c w
−ρ w λw K0  Sw 

+  w w 0 f w ∂x

∂x
 ∂x p n x 

1 0 0 0
 
0 1 0 0 Swx
| {z }
u
| {z }
B(u)

0 0 0 0 pn
  

  S w  = 0.
0 0 0 0  
+ (2.57)
0 0 −1 0   p nx 
0 0 0 −1 Swx
| {z } | {z }
C(u) u

∂p
Note that eq. (2.57) is not strictly the quasi-linear form, as matrix B is dependent on the derivative ∂xn . By
"freezing" this term the quasi-linear form is obtained. The character of eq. (2.57) can be evaluated by deter-
mining the eigenvalues of the system. The eigenvalues of this first-order system can be evaluated using

det (A − Bσ) = 0, (2.58)


dt
where σ is the eigenvalue of the system and refers to planar characteristics with σ = dx . The determinant of
this 4 by 4 matrix can be written as:

a +b ·σ c +d ·σ e ·σ 0
 
f + g ·σ h +i ·σ j ·σ k · σ
det   = e · k · σ4 , (2.59)
 −σ 0 0 0 
0 −σ 0 0
| {z }
D

where
∂p
• a = ρ n φ(1 − S w )c tn , • g = ρ w λw K0 c f w ∂xn ,
∂p n
³ ´
dp
• b = ρ n λn K0 c f n ∂x • h = ρ w φ 1 − S w c t w d S wc ,

• c = −ρ n φ, ³³
dp
´
∂p dp ∂S w
´
• i = ρ w λw K0 c λw − c f w d S wc ∂xn − c λw d S wc ∂x ,
∂p n
• d = ρ n λn K0 c λn ∂x , • j = ρ w λw K0 ,
• e = ρ n λn K0 , dp
• k = −ρ w λw K0 d S wc .
• f = ρ w φS w c t w ,

The determinant then becomes:

d pc 4
−ρ n ρ w λn λw K20 σ = 0. (2.60)
d Sw
26 2. Reservoir modelling

From this characteristic equation one can conclude that this system is purely parabolic, as all eigenvalues are
equal to zero and the matrix D is defect.

2.6.2. Incompressible phases and no capillary pressure


For incompressible flow and no rock compressibility, and when the capillary pressure is ignored, the system
becomes hybrid, i.e. elliptic in pressure and hyperbolic in saturation [11]. In this special case, the satura-
tion equation reduces to the Buckley-Leverett-equation. We will try to show this from the above equations,
although we arrive at a slightly different conclusion. Recognizing that a = b = f = g = k = 0, the system of
equations can be rewritten as:

¶ " ∂p
#
−ρ n φ 0 ∂ Sw −ρ n λn K0 c λn ∂xn −ρ n λn K0 ∂ Sw
· ¸ µ µ ¶
+ ∂p = 0. (2.61)
ρw φ 0 ∂t p n x −ρ w λw K0 c λw ∂xn −ρ w λw K0 ∂x p n x

Note that the vector of unknowns does not contain the pressure any more, but only its derivative. In in-
compressible flow only pressure gradients are important, and the pressure is determined up to a constant
(possibly determined by the boundary conditions). The determinant equation can be written as:

c +d ·σ e ·σ
· ¸
det = 0, (2.62)
h +i ·σ j ·σ

or

σ (d j − i e)σ + c j − he = 0,
¡ ¢
(2.63)

which has one zero root, σ1 = 0, and one real root:

φ(λn + λw )
σ2 = ∂p ¡ ¢. (2.64)
K0 λn λw ∂xn c λn − c λw
To find if the system is parabolic or hyperbolic, we look at the left eigenvectors of the system:

· ¸
c 0
σ1 : l (1) · = 0 → l (1) = κ · ρ w ρn ,
£ ¤
(2.65a)
h 0
c + d σ2 eσ2
· ¸
σ2 : l (2) · = 0. (2.65b)
h + i σ2 j σ2

∂p
" #
φ (2) −ρ n λn λw (c λn − c λw ) + ρ n λn c λn λt ρ n λn λt / ∂xn
l = ∂p = 0, (2.65c)
λn λw (c λn − c λw ) ρ w λn λw (c λn − c λw ) + ρ w λw c λw λt ρ w λw λt / ∂xn
∂p
" #
φ (2) ρ n λn (λw c λw + λn c λn ) ρ n λn λt / ∂xn
→ l (2) = κ · ρ w λw −ρ n λn ,
£ ¤
l = ∂p n = 0
λn λw (c λn − c λw ) ρ w λw (λw c λw + λn c λn ) ρ w λw λt / ∂x
(2.65d)

where κ is a constant scalar value. There are two independent eigenvectors, so the system is hyperbolic. In
most literature, e.g. [4, 7, 34], only the saturation equation (Buckley-Leveret equation) is studied, assuming a
given (constant) total velocity u t , from which it is concluded that the system is hyperbolic. The current form
is more general and can be more easily extended to three-phase flow, to compressible flow, or to a description
that includes the capillary pressure.

To arrive at the same form of the Buckley-Leverett equation, the positive real root σ2 of eq. (2.64) has to be
re-written as:

φλt
σ2 = ´. (2.66)
∂p ∂λn ∂λw
³
K0 ∂xn λw ∂S w
− λn ∂S w
2.6. PDE classification 27

dt
We will compare this root, σ2 = −n x /n t = dx , to the wave velocity given by the Buckley-Leverett equation (see
eq. (2.31b)), here repeated in one dimension:

∂S w u t ∂ f w ∂S w
+ = 0. (2.67)
∂t φ ∂S w ∂x
∂p n ∂S w
where u t = −λt K0 ∂x . The effective wave speed is given by the term in front of ∂x :

dx u t ∂ f w λt K0 ∂p n ∂ λw
µ ¶
= =−
dt φ ∂S w φ ∂x ∂S w λt
à !
λt K0 ∂p n ∂λw 1 ∂λt λw
=− −
φ ∂x ∂S w λt ∂S w λ2t
K0 ∂p n ∂λw ∂λt λw
µ ¶
=− − (2.68)
φ ∂x ∂S w ∂S w λt
K0 ∂p n ∂λw ∂λt
µ ¶
=− λt − λw
λt φ ∂x ∂S w ∂S w
K0 ∂p n ∂λn ∂λw
µ ¶
= λw − λn
λt φ ∂x ∂S w ∂S w

We see that the velocity arising from the characteristic analysis is exactly the same as the one arising from
the Buckley-Leverett equation. Furthermore, we can compute the following equivalent system, obtained by
premultiplying with the left eigenvectors:

∂u ∂ ∂p
µ ¶
l (1) · λ t K0
X
Ak =0 → =0 (2.69)
k ∂x k ∂x ∂x
∂u ∂S w λn λw c λn − c λw ∂p n ∂S w
¡ ¢
l (2) · K0
X
Ak =0 → + =0 (2.70)
k ∂x k ∂t φλt ∂x ∂x
∂S w λn λw c λw − c λn u t ∂S w
¡ ¢
→ + =0 (2.71)
∂t φλ2t ∂x

In the first equation we recognize the incompressibility constraint, and in the second equation the advection
equation for the water saturation with wavespeed σ2 .

2.6.3. Compressible phases and no capillary pressure


In the case that there is no capillary pressure (k = 0), the system of equations can be simplified to a 3 × 3
system, because the second-order derivative of the saturation vanishes. The determinant reads:

a +b ·σ c +d ·σ e ·σ
 
f + g ·σ h +i ·σ j · σ = −σ2 (d j − i e)σ + c j − he = 0,
¡ ¢
(2.72)
−σ 0 0
which has two roots σ1,2 , which are the same as for the incompressible case without capillary pressure from
section 2.6.2, and an additional zero root σ3 = 0. The double zero root prevents the system to have three
distinct eigenvectors, which means that the system is defective. Following the classification as described by
Loret [35], the system of equations is parabolic.

2.6.4. Incompressible phases and non-zero capillary pressure


Again, for incompressible flow we recognise thata = b = f = g = 0, but the inclusion of the capillary pressure
leads to k 6= 0. The determinant is then given by:

c +d ·σ e ·σ
 
0
det  h + i · σ j ·σ k · σ = e · k · σ3 = 0, (2.73)
σ 0 0
28 2. Reservoir modelling

and all roots are zero. This is the "parabolic case" treated in Bastian [4] (see also [55, 57]). The parabolic
nature corresponds to the diffusive effect of the capillary pressure. Rewriting eq. (2.26) (or using eq. (2.57))
for one-dimensional, incompressible flow without sources gives:

∂S w ∂ λw ∂ λn λw d p c ∂S w
µ ¶ µ ¶
φ + ut + K0 = 0, (2.74)
∂t ∂x λt ∂x λt d S w ∂x
which shows that the capillary pressure in the last term leads to second derivatives of the saturation function
and is therefore a diffusive term.

The previous eigenvalue analysis showed that presence or absence of the compressibility and capillary pres-
sure can result in a change of the mathematical character of the governing equations. The mathematical
character should be taken into account to select the most appropriate discretisation scheme. With respect to
the spatial discretisation, this means that for elliptic and parabolic problems central schemes are a natural
choice, while for hyperbolic problems upwinding may be necessary to avoid spurious oscillations. This will
be further discussed in section 2.7.

2.7. Discretisation of the reservoir model


This section describes the spatial and temporal discretisation of the two-phase immiscible reservoir model.
Anticipating the coupling with the well, the reservoir model is confined to the near wellbore region only,
which is defined as a smaller portion of the complete reservoir. The near wellbore region is defined as a
cylinder around the wellbore with a height equal to the reservoir thickness. The equations are therefore solved
on a radial coordinate system, using the PPS-formulation with primary unknowns p n and S w .

2.7.1. Spatial discretisation


To remain close to the conservation laws a Finite Volume (FV) method is applied. FV methods satisfy the
conservation laws exactly2 , independent of the size of the mesh. This property is known as discrete conser-
vation [28]. FV methods are preferred over Finite Difference (FD) methods when discontinuities are present.
Finite Element (FE) methods can have the same desired property of discrete conservation, but their flexibil-
ity requires advanced mathematics and can be time consuming [28]. The FV method starts with the integral
form of the governing equations, which can be expressed as:


Z Z
U dΩ + F · n d S = Q, (2.75)
∂t Ω S
(2.76)

where U is the conservative state vector in volume Ω, F is the flux vector that contains the fluxes flowing
through the boundary S of the volume Ω, n is the outward normal vector and Q an external source or sink.
For the governing equations of eqs. (2.12c) and (2.12d), the conservative state vector U and flux vector F are
given by:

ρ n φS n −ρ n λn K0 ∇p n
· ¸ · ¸
U= , F= . (2.77)
ρ w φS w −ρ w λw K0 ∇p w

The reservoir will be approximated by a one-dimensional cylindrical model using the PPS-formulation, and
gravity effects are neglected. A schematic of a radial cell Ωi is shown in fig. 2.8, where i is the cell counter. The
cell centres are defined on the blue primal grid and contain the conservative volume averages Ūi and Q̄ i :

1
Z
Ūi = U d Ωi , (2.78a)
Ω i Ωi
1
Z
Q̄ i = Qd Ωi , (2.78b)
Ω i Ωi

The fluxes are defined on the red dashed dual grid and contain the surface averages F̄ i :
2 up to machine precision
2.7. Discretisation of the reservoir model 29

Figure 2.8: The centres of volume element Ωi are defined on the blue primal grid, which contain he conservative variables Ūi . The faces
of element Ωi are defined on the red dual grid and contain the flux variables F i .

1
Z
F̄ i ±1/2 = F d S i ±1/2 . (2.78c)
S i ±1/2 S i ±1/2

When the continuous domain is discretised into a finite number of volumes, eq. (2.75) can be expressed by
the semi-discrete form:

dUi
Ωi = − (F (U )i +1/2 − F (U )i −1/2 ) + Ωi Q i , (2.79)
dt
where the flux vector F is a function of U and represents the spatial discretisation. Note that the volume and
surface properties are approximated using uniform properties for the volumes and faces. This approximation
is second-order accurate. Using the PPS-formulation the discrete system of equations becomes

∂p n ∂p n
µµ ¶ µ ¶ ¶
d (Un )i
Ωi = − S K0 ρ n λn − S K0 ρ n λn + Ωi Q n i , (2.80a)
dt ∂r i +1/2 ∂r i −1/2
∂ p n − p c (S w ) ∂ p n − p c (S w )
µµ ¡ ¢ ¶ µ ¡ ¢¶ ¶
d (U w )i
Ωi = − S K0 ρ w λ w − S K0 ρ w λ w + Ωi Q w i . (2.80b)
dt ∂r i +1/2 ∂r i −1/2

where the subscript n refers to the non-wetting phase ¡ and w to the ¢ wetting phase. The surface areas and
volumes are given by S i ±1/2 = 2πr i ±1/2 h and Ωi = πh r i2+1/2 − r i2−1/2 , respectively, where h is the height of the
volume (reservoir thickness).
Strictly speaking, the PPS-formulation is solved for the unknowns p n and S w . However, the formulation of
eqs. (2.80a) and (2.80b) contains the unknown p c = p c (S w ) rather than S w itself. Since the capillary pressure
p c is a known function (see section 2.3.1) it follows directly from the unknown wetting phase saturation S w .
When the saturation is explicitly required the following form can be used:

∂p n ∂p n
µµ ¶ µ ¶ ¶
d (Un )i
Ωi = − S K0 ρ n λn − S K0 ρ n λn + Ωi Q n i , (2.81a)
dt ∂r i +1/2 ∂r i −1/2
∂p n dp c ∂S w ∂p n dp c ∂S w
µµ µ ¶¶ µ µ ¶¶ ¶
d (U w )i
Ωi = − S K0 ρ w λ w − − S K0 ρ w λ w − + Ωi Q w i .
dt ∂r dS w ∂r i +1/2 ∂r dS w ∂r i −1/2
(2.81b)

The additional term dS wc ∂S


dpw
∂r allows one to formulate eqs. (2.80a) and (2.80b) explicitly in the unknowns p n
and S w . The formulation of eqs. (2.80a) and (2.80b) and eqs. (2.81a) and (2.81b) are identical in the contin-
uous domain, however the numerical approach yields a difference in accuracy. The primary unknowns p n
dp
and S w and fluid and rock properties ρ k , λk , dS wc , and K0 are all mid-point evaluated quantities. The flux
30 2. Reservoir modelling

function requires the rock and fluid properties to be evaluated on the faces. The formulation of eqs. (2.81a)
dp
and (2.81b) requires an additional interpolation due to the term dS wc and is therefore slightly less accurate.
Since a non-linear implicit solver is used (see section 2.7.3), the formulation of eqs. (2.80a) and (2.80b) is pre-
ferred and used in this study.

∂p ∂ p −p (S )
The gradients ∂rn and ( n ∂rc w ) (or alternatively ∂S w
∂r ) can be approximated using a two or four-point sten-
cil using Langrange polynomials to increase the order of accuracy (see Marino [37]). Despite the (possible)
higher order stencil of the interior gradients, the discretisation is limited to second-order accuracy, due to
the assumption of constant cell properties. This will be demonstrated in chapter 3. Using the second-order
central differencing scheme eqs. (2.80a) and (2.80b) become:

d ρ n φ(1 − S w ) i
¡ ¢
p ni +1 − p ni ³ p ni − p ni −1
µ³ ´ ´ ¶
Ωi = − SK0 ρ n λn − SK0 ρ n λn + Ωi Q n i , (2.82a)
dt i +1/2 r i +1 − r i i −1/2 r i − r i −1

d ρ w φS w i
¡ ¢ ¡ ¢ ¡ ¢
p ni +1 − p c (S w i +1 ) − p ni − p c (S w i )
µ³ ´
Ωi = − SK0 ρ w λw
dt i +1/2 r i +1 − r i
¡ ¢ ¡ ¢¶
³ ´ p ni − p c (S w i ) − p ni −1 − p c (S w i −1 )
− SK0 ρ w λw + Ωi Q w i . (2.82b)
i −1/2 r i − r i −1

Interpolation between the cell centres and the faces can be done using several methods, e.g. arithmetic av-
eraging, harmonic averaging or higher-order methods, such as cubic splines. In reservoir engineering a dis-
tinction has to be made between rock properties, e.g. the permeability K0 tensor, and fluid properties, e.g.
the phase densities ρ k and phase mobilities λk (for k = n, w). The permeable rock is often considered as a
piecewise constant material K = K (x). Between two cells one searches for a permeability K ∗ that replaces the
discontinuous K (x) [23, 43]. It can be shown that the harmonic mean of K (x) is required to find such a value.
Using n permeabilities to approximate the face values, the harmonic mean is given by:

n
Q
n· Kj
j =1
K i +1/2 = Ã !. (2.83)
n n
P 1 Q
Ki Kj
i =1 j =1

The harmonic mean of the permeability for two adjacent cells is then given by:

2K i K i +1
K i +1/2 = 1
. (2.84)
Ki + K1
i +1

The phase densities and mobilities should be treated differently. Depending on the mathematical character
of the system of equations, different interpolation schemes should be applied. According to section 2.6, the
system of equations is parabolic in character. Due to its dominantly diffusive character, central (or arithmetic)
averaging can be applied to interpolate both the phase densities and mobilities to the face to obtain second-
order accuracy (this will be further explained in chapter 3). However, for incompressible fluids and without
capillary pressure, the pressure equation becomes elliptic and the saturation equation hyperbolic. The phase
densities are a function of the pressure, so the diffusive character of elliptic equations allows one to maintain
a central averaging scheme for the phase densities. The saturation on its hand, has a hyperbolic character,
hence information travels in particular directions. The mobilities, which are a function of the saturation, i.e.
λk = λk (S w ), should be computed based on the direction from which the flow is coming from to reach the
particular cell face. For one-dimensional flow this gives

(
λki , if u > 0
λki +1/2 = (2.85a)
λki +1 , if u < 0

which is called upstream weighting, or upwinding. For a more elaborate explanation is referred to Aziz and
Settari [3], Hajibeygi and Matei [23], Pettersen [43]. Upwinding for the fluid properties is consistent with the
2.7. Discretisation of the reservoir model 31

approach of Aziz and Settari [3], who demonstrated that upwinding for the transmissibilities ¢ is necessary to
avoid spurious oscillations, where the transmissibility Ti ±1/2 is given by Ti ±1/2 = SK0 ρ k λk i ±1/2 . Aziz and
¡

Settari [3] treads the absolute permeability, density and mobility in the same way. However harmonic av-
eraging for (a piecewise constant) permeability, central averaging for the densities and upwinding for the
saturations is also applied by Chen et al. [11], Hajibeygi and Matei [23], Pettersen [43].

Equations (2.80a) and (2.80b) contain different forms of information, such as topological, metrical and phys-
ical information. These various types of information are collected in matrices. The semi-discrete form can
then be written in matrix format:

d(ρ n ◦ φ ◦ (1 − S w ))
= Ω−1 DSRn K0 Γn ZGp n + Ω−1Q n , (2.86a)
dt
d(ρ w ◦ φ ◦ S w )
= Ω−1 DSRw K0 Γw ZG p n − p c (S w ) + Ω−1Q w ,
¡ ¢
(2.86b)
dt
where

• D and G contain topological information and represent the discrete divergence and gradient operator
respectively.

• Ω,S and Z contain only metric information. They represent the volume sizes, lateral surfaces of the
volumes and denominator of the gradient operator respectively.

• Rn and Rw contain fluid densities depending on the phase pressures.

• K0 contains the absolute permeabilities.

• Γn and Γw are the phase mobilities and contain physical fluid depending on the saturation.

• ρ n , ρ w are φ are the vectors containing the unknown phase densities and porosity.

• p n and S w are the vectors containing unknown pressure and saturation.

• Q n and Q w are the vectors containing the phase source values.

The system of eqs. (2.86a) and (2.86b) is solved for the entire domain, including the boundary nodes. How-
ever, the boundary nodes are "known" values, so by including them to the "unknown" solution vector, addi-
tional equations have to be added to the system. These additional equations are of an algebraic form, rather
then the differential equations which have to be solved for the interior domain. This will be further elaborated
upon in section 2.7.2. In compact form the semi-discretised system reads:

dUn
= Ω−1 An p n +Q n ,
¡ ¢
(2.87a)
dt
dU w
= Ω−1 Aw p n − p c (S w ) +Q w ,
¡ ¡ ¢ ¢
(2.87b)
dt
where

• An = DSRn K0 Γn ZG, which is still a function of p n and S w .

• Aw = DSRw K0 Γw ZG, which is still a function of p n and S w .

Equation (2.79) requires the flux function F (U ) to be a function of the conserved variable U = [ρ n φS n ρ w φS w ]T .
However, the spatial discretisation has been based on the primitive variables W = [p n S w ]T . The map-
ping from the primitive variables to the conservative variables is straight forward, but vice versa it is not,
i.e. W → U , but U 9 W . Expressing W in terms of U is not possible, hence the system has been solved for
the primitive variables. This will have a consequence for the non-linear solver, as will be explained in sec-
tion 2.7.3, which solves the system:

dU
= F (W , t ). (2.88)
dt
32 2. Reservoir modelling

Figure 2.9: Schematic of the volumes and grid nodes near the well interface.

2.7.2. Boundary conditions


The boundary nodes can be described by Dirichlet, Neumann or Robin(mixed) boundary conditions. These
various types of boundary conditions are included by adding additional algebraic equations for the boundary
nodes and writing them down in the form of eq. (2.88), such that the non-linear solver in addition "solves"
for the boundary nodes. This approach is particularly convenient when the coupling with the well model is
considered. This will be further explained in chapter 4.

Figure 2.9 provides a schematic of the spatial discretisation near the boundary. The boundary node coincides
with the left interface of the first volume, and is referred to as node i = 1/2. The discretisation scheme near
the boundary is slightly different from the interior, due to half volume distance ∆r /2 between node i = 1/2
and node i = 1. This affects the local accuracy, which will be discussed in chapter 3. Note that for N volumes,
N + 2 unknowns exist, including the boundary nodes.

Dirichlet boundary conditions are given by the pressure and saturation values at respectively the wellbore
r = r w f and reservoir end r = r e . Neumann conditions effectively describe the gradients of the pressure and
saturation. For a radial coordinate system the Neumann condition for the pressure is given by the mass flow
rate q, which follows directly from q = ρu A:

q k = −2πr hρ k K0 λk ∇p k · n k = n, w, (2.89)

where n is the normal vector. The Neumann conditions for the saturation is given by

Θw = ∇S w · n, (2.90)

but does not have a clear physical meaning. Alternatively, the split between the mass flow rate of the wetting
phase and the non-wetting phase can be prescribed, i.e. Liquid-to-Gas Ratio (LGR), which is more commonly
applied in reservoir engineering. The LGR is given by:

qw
LGR = . (2.91)
qn

The non-linear solver solves the system of eq. (2.88). The vector function F (W ) contains the spatial discreti-
sation according to eqs. (2.80a) and (2.80b). However, the boundary conditions are known values and do not
follow from the (discretised) differential equation. By adding algebraic equations for the boundary nodes, i.e.
node i = 1/2 and i = N + 2, the non-linear solver can be used to "solve" additionally for the boundary values.
These algebraic equations for Dirichlet and Neumann boundary conditions are summarised by:
2.7. Discretisation of the reservoir model 33

Dirichlet
At r = r w f : At r = r e :

0 = p n (1) − p n w f , 0 = p n (N + 2) − p ne ,
0 = S w (1) − S w w f . 0 = S w (N + 2) − S w e .

Neumann
At r = r w f : At r = r e :

0 = Θw (1) − Θw w f , 0 = Θw (N + 1) − Θw e ,
q w (1) q w (N + 1)
0= − LGR w f . 0= − LGR e .
q n (1) q n (N + 1)

2.7.3. Temporal discretisation


This section describes the temporal discretisation. The classification of time integration methods is two-fold:

• implicit methods use information from the current and next time step. This means that a system of
equations needs to be solved which can require expensive operations, such as matrix inversion. The
main advantage of implicit methods is their unconditional stability. For highly non-linear systems a
time-step restriction can be required, but this depends on the particular non-linear solver.

• explicit methods use only information from the previous time step, which makes evaluation of the next
time step straight-forward and relatively cheap to compute. However, explicit methods are condition-
ally stable, which means that the time step is restricted. As a result, a large number of time steps may
be required for a particular simulation.

Finally, there are mixed methods available, such as the IMPES-scheme as described in section 2.5.2. These
methods try to optimise the balance between computational cost and number of iterations.

The reservoir model uses a second-order implicit Backward Difference method (BDF2). This method approx-
imates the time derivative using information of the two previous time levels:

U n+1 − 43 U n + 13 U n−1 2
= F (W n+1 , t n+1 ). (2.92)
∆t 3

Convection-diffusion equations, such as the two-phase immiscible reservoir equations, can have a stiff char-
acter. The BDF2 scheme is an appropriate choice for stiff problems, thanks to its large stability character [29].
The BDF2 scheme has been chosen as a first start and can be particularly convenient when the coupling with
the well is considered, since the stability of the coupled system is still not assessed. Other time-integration
methods have not been considered and can be a subject of further investigation. Note that the BDF2-scheme
can not be applied for the first time step. The first time step is therefore computed with the first-order Back-
ward Euler method.

Equation (2.92) is a non-linear equation, which requires an iterative solver. Well-known iterative solvers are
the Picard method and the Newton-Raphson method. The Newton-Raphson method has the property of
"quadratic convergence", which means that the rate of convergence is quadratic, while the Picard method
converges linearly. As a result, less iterations are required to solve the system which makes the Newton-
Raphson methods a popular choice. Note that the Newton-Raphson method has requirements with respect
to the initial guess and smoothness of the particular function, but the discussion on those requirements is
omitted.
Newtons-Raphson’s method uses a linear approximation of the non-linear system at each sub-iteration. The
sub-iterations are continued until a desired convergence criterion is met. Let ν be the sub-iteration counter,
then eq. (2.92) becomes:
34 2. Reservoir modelling

U ν+1 − 34 U n + 13 U n−1 2
= F (W ν+1 , t n+1 ) (2.93)
∆t 3
The linearisations of W , U and F (W , t ) are defined as:

W ν+1 = W ν + ∆W , (2.94a)
∂U ν
µ ¶
ν+1 ν
U =U + ∆W , (2.94b)
∂W
∂F ν
µ ¶
F (W ν+1 , t n+1 ) = F (W ν , t n+1 ) + ∆W , (2.94c)
∂W
³ ´ ³ ´
∂U ∂F
where ∂W and ∂W are the Jacobians of the vectors U and F with respect to W respectively. Substitution of
eq. (2.94a)-eq. (2.94c) in eq. (2.93) then leads to

à ν 3 n 1 n−1 !
1 ∂U ν 2 ∂F ν U − 4U + 3U
· µ ¶ µ ¶ ¸
2
− ∆W = − + F (W ν , t n+1 ), (2.95)
∆t ∂W 3 ∂W ∆t 3

or in compact form:

Aν ∆W = f ν . (2.96)

The right-hand side f ν is the residual of the system at sub-iterations ν. Upon convergence we have || f ν ||∞ <
², such that U ν+1 ≈ U n+1 . The algorithm is given by algorithm 1.

Algorithm 1 Newton Raphson algorithm


initialise W ν = W n
while || f ν ||∞ > ² do ³ ´ν ³ ´ν
∂U ∂F
compute matrix Aν = 1
∆t ∂W − 32 ∂W
compute F (W ν )
map W ν → U ν
U ν − 3 U n + 1 U n−1
compute residual f ν = − 4
∆t
3
+ 32 F (W ν )
ν ν ν
solve A ∆W = f
update properties W ν+1 = W ν + ∆W ν
n +1 ← ν+1
end while

Note that solving³in terms


´ of W requires two Jacobian evaluations in stead of one evaluation when solving in
∂F
terms of U (only ∂U required). This additional Jacobian evaluation for each Newton-Raphson sub-iteration
can have an important effect on the computational time if the system of equations become large.
3
Verification
This chapter presents the verification of the reservoir simulator. The reservoir simulator numerically dis-
cretises the set of PDEs, as presented in chapter 2, and when subjected to appropriate initial and boundary
conditions, it solves the system of equations. Since the numerical discretisation of the code introduces a dis-
cretisation error with respect to the continuous solution, it is of interest to know the accuracy of the code. This
is the purpose of code verification, i.e. to "convincingly demonstrate that the equations are solved correctly,
usually with the same order of accuracy and always consistently"[48].
First, an introduction will be presented on error analysis and verification techniques in sections 3.1 and 3.2
respectively. The code verification has been divided into verification of the steady-state solution and verifi-
cation of the unsteady solution. Since in steady state the temporal component has no influence, steady state
verification is only concerned with the spatial discretisation. This will be presented in section 3.3. Subse-
quently, the unsteady verification will be presented in section 3.4, focussing on the temporal discretisation.

3.1. Error analysis


In the FV approach the discretistation of the spatial part is "discrete conservative", which means that the con-
servation laws are conserved exactly1 [28]. The approximation comes in when, for example, gradients within
the fluxes are expressed as differences between quantities using adjacent cells and when mid-point evalu-
ated quantities are interpolated to the faces. This is demonstrated by the discretisation scheme of eqs. (2.82a)
and (2.82b). To evaluate the consistency of the discretisation scheme, Taylor expansions are used to find the
local truncation error. The order of the local truncation error is a measure for the convergence rate, i.e. when
the grid spacing is refined, ∆r → 0, the error should decrease according to the order of the local truncation
error. The global error is defined as the difference between the exact solution Ŵ and computed solution W∆r :

²∆r = Ŵ − W∆r . (3.1)

The error vector of eq. (3.1) is defined in the grid points r i . To quantitatively compare the error for various
grid spacings the L p -norm of the error is computed using:

à !1
1 X N p
||²∆r ||p = |Ŵi − W∆r i |p . (3.2)
N i =1

A distinction has been made between the local truncation error τ∆r and the global truncation error ²∆r . The
semi-discrete form of the reservoir model is given by eqs. (2.87a) and (2.87b) or in compact form by eq. (2.88).
However, in steady state the time derivative will vanish, such that the system becomes:

F (W ) = 0, (3.3)
1 up to machine precision

35
36 3. Verification

and the semi-discrete form is given by the non-linear system:

Ω−1 A(W )W = −Ω−1Q. (3.4)

Note that the division by the volume matrix (Ω) is important, when the spatial discretisation of the PDEs of
eq. (2.13a) - eq. (2.13d) is compared to discretisation of the FV method. The local truncation error can be
defined as the residual to eq. (3.4):

τ∆r = −Ω−1 AŴ −Q .


¡ ¢
(3.5)

Using eq. (3.1), the global truncation error ²∆r can be related to the local truncation τ∆r error through the
relation:

τ∆r = −Ω−1 A ²∆r (3.6)

In the remainder of this section the local truncation error and global truncation error are simply referred to
as τ and ², which are the respective errors taken in the L ∞ -norm. Exclusively in section 3.4, the L 1 -norm and
L 2 -norm have been used as well.

In the unsteady case the truncation error has a different meaning. The local truncation error is now defined
as the error made during one time iteration. The global truncation error is defined as the accumulation of the
local truncation errors over all time iterations. The unsteady code verification will be discussed in section 3.4.

3.1.1. Taylor expansions


The local and global truncation errors will be used to verify the proposed discretistaion scheme, as was ex-
plained in section 2.7. Section 2.6 demonstrated that the mathematical character can change due to the
presence or absence of compressibility and capillary pressure. As a result, the governing equations can have
a elliptic, parabolic or hyperbolic character. The discretisation schemes have been modified according to the
mathematical character by changing the interpolation scheme (see section 2.7.1 ). This will be demonstrated
in three different test cases:

• elliptic problem

• hyperbolic problem

• parabolic problem

These modifications of the discretisation schemes can have an effect on the local and global truncation error.
The difference between the cases can be explained by using Taylor expansion analysis. Next, an example will
be presented explaining the approach of the Taylor expansion analysis, which will be applied to the three
different test cases.

Consider the schematic of the reservoir of fig. 2.9. The grid spacing is uniform on a radial grid, with exception
of the boundaries. For on a one-dimensional radial grid the gradient and divergence operator are given by


∇(•) ≡ (•) · er , (3.7a)
∂r
1 ∂
∇ · (•) ≡ (r (•)) , (3.7b)
r ∂r
where er is the unit vector in radial direction. We will consider the governing equations of eq. (2.13a) to
eq. (2.13d). For clarity, we will simplify eq. (2.13b) by considering the pressure gradient only, and assuming
steady state, such that eq. (2.13b) becomes:

1 ∂ ∂p n
µ ¶
r = 0. (3.8)
r ∂r ∂r
3.1. Error analysis 37

Expanding eq. (3.8) leads to:

1 ∂p ∂2 p
+ 2 = 0. (3.9)
r ∂r ∂r
Now consider an interior volume i on fig. 2.9. The discrete equivalent of eq. (3.8) using the central discretisa-
tion scheme is given by:

p i +1 − p i p i − p i −1
µµ ¶ µ ¶¶
1
2πhr i +1/2 − 2πhr i −1/2 = 0. (3.10)
πh r i2+1/2 − r i2−1/2
¡ ¢
r i +1 − r i r i − r i −1

The local truncation error can be found by expanding around the point i and substituting the Taylor expan-
sions into eq. (3.10). First we introduce ∆r = r i +1 − r i , such that:

r i +1 = r i + ∆r, (3.11a)
r i −1 = r i − ∆r, (3.11b)

The Taylor expansions of the exact solution of p i ±1 around the point p i are given by:

∆r 2 ∆r 3 ∆r 4
p i +1 = p i + p i0 ∆r + p i00 + p i000 + p i0000 + O (∆r 5 ), (3.12a)
2 6 24
∆r 2 ∆r 3 ∆r 4
p i −1 = p i − p i0 ∆r + p i00 − p i000 + p i0000 + O (∆r 5 ), (3.12b)
2 6 24
∂p
where p i ±1 = p(r i ± ∆r ), p i = p(r i ) and p i0 = ∂r (r i ). By inserting the Taylor expansions into eq. (3.10), we
obtain the approximation of eq. (3.9):

p i0 000
à !
1 pi 1 0000
+ p i00 + + p i ∆r 2 + O (∆r 3 ). (3.13)
r 6 r 12

The first to two terms of eq. (3.13) can be recognized as the PDE eq. (3.9), while the remaining terms form the
truncation error. The result is a second-order accurate approximation of eq. (3.9), and consistent with the
applied central discretisation scheme.

The pressure gradient at the boundary is evaluated slightly differently. Again, consider the schematic of the
grid near the well interface of fig. 2.9. The right gradient of the first cell is computed in the same way as in the
interior stencil, but the left gradient is computed using node 1 and node 1/2 and distance ∆r /2. This requires
a Taylor expansion of p 1/2 around the point p 1 :

1 2 1 3 1 4
1 ( ∆r ) ( ∆r ) ( ∆r )
p 1/2 = p 1 − p 10 ∆r + p 100 2 − p 1000 2 + p 10000 2 + O (∆r 5 ) (3.14)
2 2 6 24
The stencil at the left boundary is given by:

à à !!
1 ³ p2 − p1 ´ p 1 − p 1/2
2πhr 3/2 − 2πhr 1/2 1 = 0. (3.15)
πh r 3/2 ∆r
¡ 2 2
2 ∆r
¢
− r 1/2

Substitution of the Taylor expansions into eq. (3.15) gives:

p 10 5 p 100
µ ¶
3 3
+ p 100 + + p 1000 ∆r + O (∆r 2 ). (3.16)
r 4 48 r 64

Since the first two terms do not correspond with the PDE of eq. (3.9), also for ∆r → 0, the discretisation is in-
consistent. The above-mentioned techniques will be used to explain the different local and global truncation
errors of the three different test cases in section 3.3.2.
38 3. Verification

3.2. Method of manufactured solutions


In the code verification one tries to identify whether the solution of the solver converges to the exact solution
as the temporal and spatial mesh are refined [47, 48]. In addition, the order of convergence should be consis-
tent with the chosen numerical discretisation scheme. The best way to test this is to compare the computed
solution with an exact analytical solution. However, the exact analytical solution of a set of PDEs is often not
known (e.g. Navier-Stokes equations, multiphase convection-diffusion equation). In order to impose an an-
alytical solution, the Method of Manufactured Solutions (MMS) can be used [47]. The basic idea of the MMS
is to choose an exact analytical solution and substitute this into the original PDE. Since the chosen solution
is not an actual solution to the PDE, additional source terms will be generated. By adding these source terms
to the original PDE, the solution is forced to the chosen analytical solution. In order to evaluate all "phys-
ical" aspects of the code, the predetermined solution should be sufficiently complex such that all terms in
the governing equations are tested [48]. For the solution to be continuous, differentiable on the entire do-
main, and to allow for boundary conditions of all kinds (Dirichlet, Neumann, Robin), periodic functions are
advised [48].

3.3. Steady-state code verification


This section will present the code verification of the spatial discretisation in steady state. The accuracy and
consistency of the discretisation scheme will be investigated using the MMS. Three different test cases will be
considered and their relative differences will be explained using Taylor expansion analysis.

3.3.1. Introduction
The system of equations for the two-phase immiscible reservoir are solved for the unknown phase pressure
p n and saturations S w , as given by eq. (2.13a)-eq. (2.13d). This system of equations does not have an ex-
act analytical solution, such that the MMS is employed to assess the convergence rate of the discretisation
scheme.
Alternatively, we can choose a solution for the two-phase case by using an analytical solution available for
single phase flow. Single phase flow can be forced by assuming either S w = 1 or S w = 0. The considered
test cases assume a constant producing well. In a radial coordinate system the exact single phase analytical
solution for steady state is then of the form:

qµ re
µ ¶
p̂ = p w f + ln , (3.17a)
2πkh rw f

where q is the production rate, µ the fluid viscosity, k the effective permeability and h the perforation depth.
The flowing bottom hole pressure (p w f ) is defined at the well interface (r = r w f ). When instead of the pro-
duction rate (Neumann), the boundary pressure at the far end (p e at r = r e ) and flowing bottom hole pressure
(p w f ,Dirichlet) are prescribed, the analytical solution is given by:

pe − p w f r
µ ¶
p̂ = ³ ´ ln . (3.17b)
ln rr e rw f
wf

The solution has been shown in fig. 3.1. The constant pressure assumption near the outer boundary appears
somewhat artificially, but is realistic when the near wellbore region is considered, or in case of secondary re-
covery in which one aims to keep the pressure constant by water injection [16].

To exploit the MMS for two-phase flow, an imposed analytical solution for the unknown phase pressure p̂ n is
chosen similar to the logarithmic solution of eq. (3.17b). Since the analytical solution is very close to the single
phase case, it is expected that the forcing term is small. In addition, the function is sufficiently smooth and
continuously differentiable on the entire domain (r w f > 0). The imposed analytical solution of the wetting
3.3. Steady-state code verification 39

·107

1.98

1.96

p [Pa] 1.94

1.92

1.9

0 50 100 150 200 250 300


r [m]

Figure 3.1: The radial flow of a typical oil well under steady state flow conditions.

saturation Ŝ w has a similar form:

p ne − p n w f µ r ¶
pˆn = ³ ´ ln , (3.18a)
ln rr e rw f
wf

S we − S w w f µ r ¶
Sˆw = ³ ´ ln , (3.18b)
ln rr e rw f
wf

3.3.2. Test cases steady state


This section presents three different test cases to assess the accuracy and consistency of the spatial discreti-
sation. The general system of equation is given by eq. (2.13a)-eq. (2.13d). In addition, gravity is neglected and
source terms are equal to the MMS source terms.

As discussed in chapter 2 the mathematical character of the two-phase flow equations can change from
parabolic to elliptic in the pressure and hyperbolic in the saturation when the fluids are incompressible and
the capillary pressure is neglected. This change of mathematical character has an impact on the number of
boundary conditions that have to be prescribed. For the elliptic and parabolic cases, boundary conditions
on all sides of the domain have to be prescribed, while in the hyperbolic case it depends on the direction
of information propagation. In addition, the elliptic case does not require an initial pressure condition, but
the pressure follows from the initial saturation distribution according to eq. (2.30). In the current test case
of a producing well, the pressure difference between the flowing bottom hole pressure and the reservoir-end
pressure pushes the fluids towards the wellbore. This means that in the hyperbolic situation (incompressible
fluids, p c = 0), the saturation boundary condition at the well follows from the numerical domain. To fulfil this
requirement a Neumann boundary conditions has been prescribed for the saturation.

A considerable "freedom" is present in simulating the reservoir. Especially the capillary pressure function
and the relative permeability functions can take various shapes (see sections 2.3.1 and 2.3.2). Table 3.1 lists
the options that can be implemented in the reservoir model. From the listed options, an elliptic, hyperbolic
and parabolic problem has been chosen, as explained in sections 3.3.3 to 3.3.5.

The results of the test cases are presented in the subsequent subsections. The figures show the global (²) and
local truncation error (τ) in the L ∞ -norm of respectively the total domain (left) and interior domain (right),
using a two-point or four-point stencil to approximate the interior gradients. The distinction between total
and interior domain has been made since the gradients on the boundary are differently computed than the
gradients in the interior domain. In the interior domain, the gradients are computed using information of
40 3. Verification

Table 3.1: Reservoir model options

variable option
coordinate system Cartesian, radial
grid spacing uniform, logarithmic
boundary gradient representation 2-pt, 3-pt, 4-pt
fluid incompressible, compressible
permeability function Constant, Corey, Brooks-Corey
capillary pressure function Zero, Chen, Brooks-Corey
Boundary conditions Dirichlet, Neumann

both adjacent grid cells, while the gradients at the boundaries use only one-sided information. This can have
a significant effect on the local truncation error. It is important to realise that the interior domain is defined
as the cells whose discretisation stencil is unaffected by the boundaries. This means that for the two-point
boundary stencil, the first and last volume are excluded, for the three-point stencil, the first and last two
volumes are excluded etc.

3.3.3. Single phase flow (elliptic)


The first case resembles single phase flow in which the saturation is trivially satisfied (S w = 1). In addition,
incompressible flow, constant viscosity (µ) and constant absolute permeability (K0 ) are assumed, and the
capillary pressure is neglected (p c = 0). The single phase case has an analytical solution such that the MMS
does not have to be employed. The governing equations of eq. (2.13a) to eq. (2.13d) reduce to

K0
µ ¶
(∂ρφ)
− ∇ · ρ ∇p = 0, (3.19)
∂t µ

Since ρ, φ, µ and K 0 are constant, they can be divided out of eq. (3.19) and the derivative with respect to time
is equal to zero. This results in an elliptic Laplace equation for the pressure. In a radial coordinate system, the
exact analytical solution is then given by eqs. (3.17a) and (3.17b).

The incompressible single phase problem is solved on a radial coordinate system using the two-phase solver.
The domain is give by Ω = (10, 300) m with are reservoir height of h = 75 m, and the following parameter set:

Table 3.2: Parameters for single phase test case.

fluids: capillary pressure:


incompressible pc = 0
ρ n = ρ w = 1000 kg/m3
µn = µw = 0.001 Pa s

rock: boundary conditions:


φ = 0.2 p n = 1.9 × 105 Pa for r = 10 m
K 0 = 1 × 10−7 m2 p n = 2.0 × 105 Pa for r = 300 m

relative permeability: initial conditions:


constant, k r k = 1 p n (r , 0) = p̂ n for x ∈ Ω

Analytical solution
The first test case resembles the single phase flow case and serves as a benchmark with the single phase
model from Marino [37]. The results of the first case are presented in fig. 3.2. The upper two figures show
the truncation errors of the two-point interior stencil, while the bottom two figures represent the four-point
interior stencil.
The global truncation error of the two-point interior stencil is second-order accurate on both the total (left)
and interior (right) domain. For the local truncation error one can observe that the two- point boundary
3.3. Steady-state code verification 41

101 101
τn -2pt τn -2pt
0
100 τn -3pt 100 τn -3pt
τn -4pt τn -4pt
10−1 −1 ²p n -2pt
10−1 ²p n -2pt
−1
10−2 ²p n -3pt 10−2 ²p n -3pt
²p n -4pt ²p n -4pt
10−3 10−3 −2
², τ

², τ
10−4 10−4
10−5 10−5
10−6 10−6
10−7 10−7
−2 −2
10−8 10−8 −2
10−9 1 10−9 1
10 102 103 104 10 102 103 104
N N
(a) Error total domain using a two-point interior stencil. (b) Error interior domain using a two-point interior stencil.
101 101
0 τn -2pt τn -2pt
100 τn -3pt 100 τn -3pt
10−1 −1 τn -4pt 10−1 τn -4pt
10−2 −2
²p n -2pt 10−2 ²p n -2pt
10−3 ²p n -3pt 10−3 ²p n -3pt
²p n -4pt −2 ²p n -4pt
10−4 10−4
10−5 10−5
², τ

², τ

−4
10−6 10−6
10−7 −2 10−7 −2
10−8 10−8
10−9 −3 10−9 −3
10−10 −4 10−10 −4
10−11 10−11
10−12 1 10−12 1
10 102 103 104 10 102 103 104
N N
(c) Error total domain using a four-point interior stencil. (d) Error interior domain using a four-point interior stencil.

Figure 3.2: Analytical verification showing the global ² and local τ truncation error for the single-phase case using a two-point and
four-point stencil for the interior gradients.

stencil results in zero-order accuracy, and the three-points and four-point boundary stencil in first-order ac-
curacy. The local truncation error in the interior domain is always second-order accurate, as can be observed
in the upper right figure. The reductions in order are thus related to the boundary nodes. The proof will be
given at the end of this subsection.
The global truncation error is always second-order accurate and, apparently, does not "suffer" from the local
order reduction at the boundaries. This can be attributed to the elliptic nature of the Laplace operator, which
essentially diffuses the error over the adjacent cells. These results are consistent with the results from for
example Mattheij et al. [39], who showed that for diffusion equations the local truncation error may be up to
two orders less than the global truncation error. Note that this is solely true for diffusive equations. When the
mathematical character changes to for example a hyperbolic character, errors can propagate throughout the
domain. This will be further explained in section 3.3.4.

The bottom two figures of fig. 3.2 show the results of the four-point interior stencil. Second, third and fourth-
order accuracy of the global truncation error is obtained for both the total and interior domain. The local
truncation error is zero, first and second-order accurate for respectively the two, three and four-point bound-
ary stencils. The local truncation error of the two-point interior and four point boundary stencil is one order
less accurate than the truncation error of the four-point interior and four point boundary stencil (see star
42 3. Verification

dashed lines of figs. 3.2a and 3.2c), even though the discretisation of the boundary node seem identical (four
point boundary stencil). The differences will be explained using Taylor expansions.

We will now demonstrate the order of the local truncation error by using a Taylor expansion. The truncation
error of the interior domain for both the two-point and four-point stencil are consistent with the proposed
discretisation schemes, so the derivations of those have been omitted. Consider the first volume of see fig. 2.9.
The permeability (K0 ) and density (ρ) are constant, so the do not influence the truncation error and are left-
out of the equation. The stencil using a two-point interior and two-point boundary is then given by:

à à !!
1 ³ p2 − p1 ´ p 1 − p 1/2
¢ 2πhr 3/2 − 2πhr 1/2 1 = 0. (3.20)
πh r 3/2 ∆r
¡ 2 2
2 ∆r
− r 1/2

The expansion of r i ±1/2 , p 1/2 and p 2 are given by eq. (3.11a), eq. (3.11b), eq. (3.14) and eq. (3.12a) respectively.
Substitution of the Taylor expansions into eq. (3.20) gives:

p 10 5 p 100
µ ¶
3 3
+ p 100 + + p 1000 ∆r + O (∆r 2 ). (3.21)
r 4 48 r 64

The discretisation is inconsistent, as was observed in the verification plots of fig. 3.2. A similar derivation can
be done for the three-point boundary stencil. The discretisation scheme for the three-point boundary stencil
is given by:

− 31 p 2 + 3p 1 − 83 p 1/2
à à !!
1 ³ p2 − p1 ´
2πhr 3/2 − 2πhr 1/2 = 0, (3.22)
πh r 3/2 ∆r ∆r
¡ 2 2
¢
− r 1/2

where the coefficients in front of the pressure points have been obtained using Lagrange polynomials (Marino
[37]). Substituting the Taylor expansions of p 1/2 and p 2 gives:

p 10 1
+ p 100 + p 1000 ∆r + O (∆r 2 ) (3.23)
r 6

This is a first-order approximation of eq. (3.19). The discretisation scheme with a four-point boundary stencil
and two-point interior is given by:

3 5 15 46
à à !!
1 p2 − p1 ´ 20 p 3 − 6 p 2 + 4 p 1 − 15 p 1/2
³
¢ 2πhr 3/2 − 2πhr 1/2 = 0, (3.24)
πh r 3/2 ∆r ∆r
¡ 2 2
− r 1/2

where the Taylor expansion of p 3 around point p 1 is given by:

(2∆r )2 (2∆r )3 (2∆r )4


p 3 = p 1 + p 10 2∆r + p 100 + p 1000 + p 10000 + O (∆r 5 ) (3.25)
2 6 24
Substituting the Taylor expansions of p 1/2 , p 2 and p 3 gives:

p 10 1 000
+ p 100 + p ∆r + O (∆r 2 ) (3.26)
r 24 1

The discretisation is still first-order accurate, but the remainder term is a factor 4 smaller compared to the
three-point stencil. These results are consistent with the local truncation errors observed in fig. 3.2a.

Finally, the four-point boundary stencil with four-point interior is second-order accurate (compare fig. 3.2a
with fig. 3.2c). The stencil is given by:
3.3. Steady-state code verification 43

1
p 3 + 76 p 2 − 54 p 1 + 15
2 3 5 15 46
ÃÃ ! Ã !!
1 − 20 p 1/2 20 p 3 − 6 p 2 + 4 p 1 − 15 p 1/2
2πhr 3/2 + − 2πhr 1/2 = 0,
πh r 3/2 ∆r ∆r
¡ 2 2
¢
− r 1/2
(3.27)

Substituting the Taylor expansions of p 1/2 , p 2 and p 3 gives:

p 10 1 p 1000
µ ¶
1
+ p 100 + − p 10000 + ∆r 2 + O (∆r 3 ), (3.28)
r 24 8 r

which is indeed second-order accurate. The higher approximation of the right face thus increases the order
of the discretisation. These results are consistent with the results of Marino [37].

MMS solution
The single phase case has also been verified using the MMS, as presented in fig. 3.3 to identify possible dis-
crepancies in handling source terms. Note that a different forced analytic function has been chosen than
eq. (3.18a), since this forced solution is identical to the actual analytical solution of the PDE, hence the MMS
source terms would be equal to zero. To remain close to the logarithmic profile of the analytical solution, a
quarter period sine-function have been used for this particular case. The results are are shown in fig. 3.3.
In this case the two-point and four-point interior stencil show almost identical results, with exception of the
global truncation error, which is now limited to second-order accuracy.
The limitation to second-order accuracy can be attributed to the volume approximation of the cells. The
source terms of the MMS are obtained when the forced exact solution Ŵ is evaluated by the PDE, i.e.

Q mms = L (Ŵ ), (3.29)

where L (W ) is the differential operator of the PDE of eq. (2.13a) to eq. (2.13d). These source terms are a
direct result of the PDE and are therefore valid in any point of the domain. The FV method on its hand solves
the system of eqs. (2.80a) and (2.80b) for the volumes Ωi . Since the source terms are evaluated in a point,
the solution of the FV method has be to divided by the volumes. The current method assumes constant
properties for the volumes. This limits the order of accuracy to second-order, as earlier mentioned in sec-
tion 2.7.1. This could be resolved in two ways. The (MMS) source terms could be integrated exactly, but this
can be cumbersome. Alternatively, multiple nodes could be included to increase the accuracy of the volume
approximations.
The second-order limit was not observed in the analytical test case, as it showed fourth-order accuracy for
the four-point stencil. This particular test case neglected source terms, which implied that the volume matrix
can be divided out of the equation. The limiting effect of the volume matrix was therefore not present. Since
the remaining test cases will include source terms, the effect of the volume approximation will be present.
The remaining cases are therefore only solved using a two-point interior stencil.

3.3.4. Two-phase flow (hyperbolic)


The incompressible two-phase problem without capillary pressure and Corey’s relative permeability function
is solved on a radial coordinate system with domain Ω = (10, 300) m and a reservoir height of h = 75 m. The
remaining parameters are given in table 3.3.

As presented in section 2.6, the assumption of zero capillary pressure and incompressible fluids have changed
the mathematical character of the governing equations from parabolic to elliptic in pressure and hyperbolic
in saturation. To avoid spurious oscillations in the saturation, the mobilities λk = λk (S w ) have been up-
winded using a first-order upwind scheme (see section 2.7.1). This will have a direct effect on the global
trunction error, as will be shown subsequently.

The results of this test case are presented in fig. 3.4. On fig. 3.4a one can observe that the local truncation
errors τn and τw are respectively zero, first and first-order accurate for the two, three and four-point boundary
stencil, similar to the verification plots of the single phase case.
44 3. Verification

10−2 10−2
τn -2pt τn -2pt
τn -3pt τn -3pt
10−3 0 τn -4pt 10−3 τn -4pt
²p n -2pt ²p n -2pt
10−4 ²p n -3pt 10−4 ²p n -3pt
²p n -4pt ²p n -4pt
10−5 −1 10−5
², τ

², τ
−1
10−6 10−6
−2
10−7 10−7

10−8 −2 10−8 −2

10−9 1 10−9 1
10 102 103 104 10 102 103 104
N N
(a) Error total domain using a two-point interior stencil. (b) Error interior domain using a two-point interior stencil.
10−2 10−2
τn -2pt τn -2pt
−3
τn -3pt −3
τn -3pt
10 0 τn -4pt 10 τn -4pt
²p n -2pt ²p n -2pt
10−4 ²p n -3pt 10−4 ²p n -3pt
²p n -4pt ²p n -4pt
10−5 10−5
−1
², τ

², τ

10−6 10−6
−2 −2
10−7 10−7
−2 −2
10−8 10−8
−2 −2
10−9 1 10−9 1
10 102 103 104 10 102 103 104
N N
(c) Error total domain using a four-point interior stencil. (d) Error interior domain using a four-point interior stencil.

Figure 3.3: MMS verification showing the global ² and local τ truncation error for the single-phase case using a two-point and four-point
stencil for the interior gradients.

The global truncation error for the pressure p n and saturation S w are respectively second and first-order
accurate for all boundary stencils, which is different in comparison with the single phase case. The order dif-
ference stems from the different interpolation schemes, i.e. central averaging (second-order) for the pressure
dependent variables and upwinding (first-order) for the saturation dependent variables to avoid spurious
oscillations.
Note that the local truncation error of in the interior domain is first-order accurate (dashed lines of fig. 3.4b).
Upwinding of the mobilities have reduced the order of accuracy. The pressure follows from the elliptic equa-
tion of eq. (2.30). The diffusive nature reduces the error introduced at the boundaries. In addition, the total
mobility in eq. (2.30) is constant, so no approximation comes in using the upwinding scheme. This explains
why the pressure is second-order accurate. The saturation is found solving eq. (2.31b), which requires inter-
polation of the wetting phase mobilities resulting in first-order accuracy.

The boundary condition of the saturation at the wellbore interface (r = r w f ) is of a Neumann type. Since the
saturation equation is now a first-order PDE (see section 2.5.3), boundary conditions have to be prescribed
on one side of the domain. The pressure gradient forces the flow to flow in the direction of the wellbore, which
means that the boundary condition at the wellbore follows from the numerical domain. A Dirichlet boundary
condition would yield an overdetermined problem and leads to spurious oscillations. These oscillations grow
3.3. Steady-state code verification 45

Table 3.3: Parameters for two-phase (hyperbolic) test case.

fluids: capillary pressure:


incompressible pc = 0
ρ n = ρ w = 1000 kg/m3
µn = µw = 0.001 Pa s

rock: boundary conditions:


φ = 0.2 p n = 1.9 × 105 Pa, Θw =0 for r = 10 m
K 0 = 1 × 10−7 m2 p n = 2.0 × 105 Pa, S w = 0.7 for r = 300 m

residual saturation: initial conditions:


S wc = 0.2 p n (r , 0) = p̂ n for r ∈ Ω
S nr = 0.1 S w (r , 0) = Ŝ w for r ∈ Ω

relative permeability:
Corey, λ = 2.0

over time and eventually lead to blow-up of the simulation.

100 0 100
τn -2pt τn -2pt
τn -3pt τn -3pt
10−1 10−1
−1 τn -4pt τn -4pt
0 τw -2pt −1 τw -2pt
10−2 −1 10−2 −1
τw -3pt τw -3pt
10−3 −1 τw -4pt 10−3 τw -4pt
²p n -2pt ²p n -2pt
−1
², τ

², τ

10−4 ²p n -3pt 10−4 ²p n -3pt


²p n -4pt ²p n -4pt
10−5 ²S w -2pt 10−5 ²S w -2pt
²S w -3pt ²S w -3pt
−2 −2
10−6 ²S w -4pt 10−6 ²S w -4pt

10−7 10−7

10−8 1 10−8 1
10 102 103 104 10 102 103 104
N N
(a) Error total domain using a two-point interior stencil. (b) Error interior domain using a two-point interior stencil.

Figure 3.4: MMS verification showing the global ² and local τ truncation error for the two-phase hyperbolic case using a two-point stencil
for the interior gradients.

The observed truncation errors of fig. 3.4 will now be explained using Taylor expansions. In the hyperbolic
case, the addition of the mobility (λk = λk (S w )), effects the local truncation error near the boundary. The
local truncation error τn and τw are zero-order accurate for the two-point boundary stencil and first-order
accurate for both the three and four-point boundary stencil. The proof will be presented subsequently.

The interior scheme is first-order accurate and seems inconsistent with the proposed second-order scheme.
In the current derivation, the fluid density (ρ) and absolute permeability (K0 ) are constant and therefore
neglected from the discretisation scheme for clarity. The interior scheme is given by

1 ³³ p i +1 − p i ´ ³ p i − p i −1 ´´
2πhr i +1/2 λk − 2πhr i −1/2 λk = 0, for k = n, w. (3.30)
πh r i +1/2 − r i2−1/2 ∆r ∆r
¡ 2 ¢ i +1 i

The mobilities are upwinded from right to left, as is appropriate for a production scenario. The Taylor expan-
sion of λki +1 around point λki is given by
46 3. Verification

∆r 2 ∆r 3 ∆r 4
λki +1 = λki + λ0ki ∆r + λ00ki + λ000
ki + λ0000
ki + O (∆r 5 ). (3.31)
2 6 24
Substitution then gives:

p i0
à !
λki p i00 + + λ0ki p i0 + O (∆r ), (3.32)
r

which is indeed a first-order approximation of the PDE. Upwinding of the relative permeabilities reduces the
interior scheme to first-order accuracy. The boundaries are again slightly different evaluated. The two-point
boundary stencil is given by

à à !!
1 ³ p2 − p1 ´ p 1 − p 1/2
¢ 2πhr 3/2 λk2 − 2πhr 1/2 λk1 1 = 0, for k = n, w. (3.33)
πh r 3/2 ∆r
¡ 2 2
2 ∆r
− r 1/2

Substitution then gives the inconsistent result:

3 00 p 10
µ ¶
λk1 p + + λ0k1 p 10 + O (∆r ), (3.34)
4 1 r

3
due to the term 4 in front of p 100 . The three-point and four-point boundary stencil are given by

− 31 p 2 + 3p 1 − 83 p 1/2
à à !!
1 ³ p2 − p1 ´
¢ 2πhr 3/2 λk2 − 2πhr 1/2 λk1 = 0, for k = n, w, (3.35)
πh r 3/2 ∆r ∆r
¡ 2 2
− r 1/2

and

3 5 15 46
à à !!
1 p2 − p1 ´ 20 p 3 − 6 p 2 + 4 p 1 − 15 p 1/2
³
¢ 2πhr 3/2 λk2 − 2πhr 1/2 λk1 = 0, for k = n, w, (3.36)
πh r 3/2 ∆r ∆r
¡ 2 2
− r 1/2

which afters substitution of the Taylor expansions both simplify to the first-order result:

p0
µ ¶
λk1 p 100 + 1 + λ0k1 p 10 + O (∆r ). (3.37)
r

3.3.5. Two-phase flow (parabolic)


The compressible two-phase problem with Corey’s relative permeability and Brooks-Corey’s capillary pres-
sure function is solved on a radial coordinate system with domain Ω = (10, 300) m and a reservoir height of
h = 75 m. The remaining parameters are presented in table 3.4.
The results of this test case are presented in fig. 3.5. The interior domain is second-order accurate for both
the local and global truncation errors, consistent with the expected discretisation scheme. The total domain
shows again zero, first and first-order accuracy for respectively the two, three and four-point boundary stencil,
consistent with the elliptic and hyperbolic cases. The global errors of the pressure and saturation are not
effected by the local order reduction, similar to the single phase case, since the mathematical character is
purely parabolic and dominantly diffusive.
The truncation errors observed in fig. 3.5 will now be explained using Taylor expansions. The fluids are as-
sumed to be compressible, hence the densities have to be included in the discretisation scheme. In addition,
the fluid densities and mobilities are interpolated using both a central averaging scheme, since the parabolic
nature does note require upwinding for the interpolation. Lastly, due to the capillary pressure, two different
fluid pressures exist, denoted with p ki . The interior scheme is second-order accurate and consistent with the
3.3. Steady-state code verification 47

Table 3.4: Paramters for two-phase (parabolic) case.

fluids: capillary pressure:


compressible Brooks-Corey, λ = 2.0
oil,water
µn = 0.01 Pa s, µw = 0.001 Pa s

rock: boundary conditions:


φ = 0.2 p n = 1.9 × 105 Pa, S w = 0.3 for r = 10 m
K 0 = 1 × 10−7 m2 p n = 2.0 × 105 Pa, S w = 0.7 for r = 300 m

residual saturation: initial conditions:


S wc = 0.2 p n (r , 0) = p̂ n for r ∈ Ω
S nr = 0.1 S w (r , 0) = Ŝ w for r ∈ Ω

relative permeability:
Corey, λ = 2.0

100 10−1
τn -2pt τn -2pt
−1 0 τn -3pt τn -3pt
10 τn -4pt 10−2 τn -4pt
0
τw -2pt τw -2pt
10−2 −1 τw -3pt τw -3pt
τw -4pt 10−3 τw -4pt
−1
10−3 ²p n -2pt ²p n -2pt
², τ

², τ

²p n -3pt 10−4 ²p n -3pt


−2
10 −4 −2 ²p n -4pt ²p n -4pt
²S w -2pt ²S w -2pt
10−5 −2
²S w -3pt ²S w -3pt
10−5 −2
²S w -4pt ²S w -4pt
−2
10−6
−2 10−6

10−7 1 10−7 1
10 102 103 104 10 102 103 104
N N
(a) Error total domain using a two-point interior stencil. (b) Error interior domain using a two-point interior stencil.

Figure 3.5: MMS verification showing the global ² and local τ truncation error for the two-phase parabolic case using a two-point stencil
for the interior gradients.

proposed scheme. The proof for the interior will therefore be omitted. The absolute permeability is assumed
to be constant and has no effect on the discretisation, hence it is neglected from discretisation scheme for
clarity. The discretisation scheme for the parabolic case using a two-point boundary stencil is given by:

õ ¶ à !!
λk1 + λk2 ρ k1 + ρ k2 p k2 − p k1
¡ ¢¡ ¢
1 p k1 − p k1/2
2πhr 3/2 − 2πhr 1/2 λk1/2 ρ k1/2 ∆r
= 0, for k = n, w.
πh r 3/2 ∆r
¡ 2 2
¢
− r 1/2 2 2
2
(3.38)

Equation (3.38) requires an additional Taylor expansion of ρ k2 around point ρ k1 :

∆r 2 ∆r 3 0000 ∆r
4
ρ k2 = ρ k1 + ρ 0k1 ∆r + ρ 00k1 + ρ 000
k1 + ρ k1 + O (∆r 5 ), (3.39)
2 6 24
48 3. Verification

and Taylor expansions of ρ k1/2 around point ρ k1 and λk1/2 around point λk1 :

1 ( 1 ∆r )2 ( 21 ∆r )3 ( 21 ∆r )4
ρ k1/2 = ρ k1 − ρ 0k1 ∆r + ρ 00k1 2 − ρ 000
k1 + ρ 0000
k1 + O (∆r 5 ), (3.40a)
2 2 6 24
1 ( 1 ∆r )2 ( 21 ∆r )3 ( 1 ∆r )4
λk1/2 = λk1 − λ0k1 ∆r + λ00k1 2 − λ000
k1 + λ 0000 2
k1 + O (∆r 5 ). (3.40b)
2 2 6 24

Substitution of the Taylor expansion gives the inconsistent result:

1 ρ k1 λk1 3 p k1
à 0 0 00 !
ρ k1 λk1 p k0 1 + + + + O (∆r ). (3.41)
r ρ k1 λk1 4 p k0
1

The three point boundary stencil is given by:

λk1 + λk2 ρ k1 + ρ k2 p k2 − p k1
µµ ¡ ¢¡ ¢ ¶
1
¢ 2πhr 3/2 −
πh r 3/2 ∆r
¡ 2 2
− r 1/2 2 2
− 13 p k2 + 3p k1 − 83 p k1/2
à !!
2πhr 1/2 λk1/2 ρ k1/2 = 0, for k = n, w, (3.42)
∆r

which after substitution of the Taylor expansions simplifies to the first-order accurate result:

1 ρ k1 λk1 p k1
à 0 0 00 !
ρ k1 λk1 p k0 1 + + + + O (∆r ). (3.43)
r ρ k1 λk1 p k0
1

Finally, the four point boundary stencil is given by

λk1 + λk2 ρ k1 + ρ k2 p k2 − p k1
µµ ¡ ¢¡ ¢ ¶
1
2πhr 3/2 −
πh r 3/2 − r 1/2 ∆r
¡ 2 2
¢
2 2
3 5 15 46
à !!
20 p 3 − 6 p 2 + 4 p 1 − 15 p 1/2
2πhr 1/2 λk1/2 ρ k1/2 = 0, for k = n, w, (3.44)
∆r

which gives again a first-order accurate result:

1 ρ k1 λk1 p k1
à 0 0 00 !
ρ k1 λk1 p k0 1 + + + + O (∆r ). (3.45)
r ρ k1 λk1 p k0
1

3.4. Unsteady code verification


This section presents the code verification of the unsteady case in which the temporal error will be investi-
gated. Using the assumptions of section 3.3.4 the saturation equation can be recast in the form of the hy-
perbolic Buckley-Leverett equation, as explained in section 2.5.3. The exact solution and analysis have been
demonstrated for a Cartesian coordinate system. The subsequent analysis will therefore be performed on a
one-dimensional Cartesian coordinate system. The grid spacing is indicated with ∆x rather than ∆r , as was
used for the radial coordinate system used in section 3.3. Time integration has been implemented using the
second-order Backward Difference method (BDF2).
3.4. Unsteady code verification 49

3.4.1. Introduction
In the unsteady case the error consists of both a spatial and temporal component. Taylor analysis allows
one to express the error amongst the spatial grid points at end time t end in terms of the grid spacing ∆x and
arbitrary time step ∆t :

²∆x,∆t = Ŵ − W∆x,∆t , (3.46a)


p q
= C 1 (x i , t ) O (∆x ) +C 2 (x i , t ) O (∆t ), i = 1..N , (3.46b)

where C 1 and C 2 are coefficients that may depend on space and time, and p and q. Taylor expansion anal-
ysis requires the solution W to be continuous and differentiable (up to the order of interest). In the special
case of shock solutions Taylor expansion analysis may fail, since the solution is not differentiable at the shock
location. As a result, the suggested order of accuracy may not be achieved. This will be demonstrated in sec-
tion 3.4.2.

The convergence rate of the total error is determined by the lowest order of the spatial and temporal com-
ponent if the ∆x and ∆t are linked during mesh refinement. Since the spatial error is first-order accurate
(first-order upwinding of mobilities), this error will be dominant, hence first-order accuracy for the total er-
ror is expected. By expressing the grid spacing ∆x in terms of the time step ∆t the total error can be expressed
as a function of the time step only. The relation between the grid spacing and the time step has been given in
terms of the Courant-Friedrich-Lewy (C F L) condition, which is a necessary stability condition for hyperbolic
PDEs when explicit time integration schemes are considered. Although an implicit time integration scheme
is considered (BDF2), such that the C F L condition does not have to be met to guarantee stability, it is still a
helpful condition to link the characteristic time scales to the characteristic length scales. The C F L condition
is given by

u∆t
CFL ≡ < 1, (3.47)
∆x
where u is the characteristic velocity of the system. Given a fixed C F L number the grid spacing can be ex-
pressed in terms of time step:

u
∆x = ∆t , (3.48)
CFL
This analysis does not give information on the temporal error, but on the combined error of both the spatial
and temporal discretisation. By subtracting the spatial error from the exact solution, the temporal error can
be assessed. The temporal error can then be defined as:

²∆t = W∆x,∆t →0 − W∆x,∆t , (3.49)

where W∆x,∆t →0 is the solution of the semi-discretised system using a very small time step. This "exact"
solution is different from the exact analytical solution of eq. (3.46a) as it still contains the spatial error.

3.4.2. Buckley Leverett problem


The incompressible two-phase problem without capillary pressure and Corey’s relative permeability func-
tion is solved on the domain Ω = (0, 300) m with a reservoir height of h = 75 m. The remaining parameters are
given in table 3.5. For the initial condition, the Riemann problem is considered with a single discontinuity at
x = 0 m. The left saturation state is equal to one (S Lw = 1), while the right state is equal to zero (S Rw = 0). The
initial pressure field follows directly from the initial saturation distribution according to eq. (2.30).

The solutions to the Buckley Leverett equation at t end = 3456 s using BDF2 time integration have been plotted
in fig. 3.6. The computed solutions converge to the exact analytical solution (blue) as the number of volumes
and number of time steps (N x = N t ) increase from 16 to 256. The Courant number is given by C F L = 1.2 with
u shock = 0.054 m/s. This case is (qualitatively) compared with the results of Bastian [4].
50 3. Verification

Table 3.5: Parameters for the Buckley-Leverett problem.

fluids: capillary pressure:


incompressible pc = 0
ρ n = ρ w = 1000 kg/m3
µn = µw = 0.001 Pa s

rock: boundary conditions:


φ = 0.2 φn = −1.0 × 10−5 kg/s, S w = 1 for x = 0 m
K 0 = 1 × 10−7 p n = 2.0 × 105 Pa, Θw = 0 for x = 300 m

residual saturation: initial conditions:


(
1, x ≤ 0
S wc = S nr = 0 S w,0 =
0, x > 0

relative permeability:
Corey, λ = 2.0

1
analytic
16 volumes
32 volumes
0.8
64 volumes
128 volumes
256 volumes
0.6
S w [-]

0.4

0.2

0
0 20 40 60 80 100 120 140 160 180 200 220 240 260 280 300
x [m]

Figure 3.6: Computed solutions of the Buckley–Leverett problem with quadratic Corey relative permeabilities (N x = N t , λ=2) at t =
3456 s.

For quantitative analysis the L 1 -norm and L 2 -norm of the total error have been computed and shown in
fig. 3.7a. The slope of the error plots gives the convergence rate. The convergence rate of the L 1 -norm and
L 2 -norm are 0.9 and 0.66, respectively. This is slightly lower than the expected convergence rate of 1 (first-
order upwinding). Bastian [4] obtains convergence rates of 0.8 and 0.4 for the L 1 -norm and L 2 -norm respec-
tively, using first-order upwinding and Backward Differencing for the time integration with Courant number
C F L=0.8. The difference between the L 1 -norm and L 2 -norm is not explained in the work of Bastian [4]. The
observations could be explained as follows.
The first-order upwinding scheme gives a first-order approximation given the Taylor expansion analysis.
However, Taylor expansions hold for continuous differentiable solution, which is not the case at the shock
location. The (slightly) lower convergence rate of the L 1 -norm and L 2 -norm can be related to the inaccurate
position of the numerically computed shock location. The computed solutions seem to intersect slightly be-
yond the actual shock location of the analytical solution. This point does not converge to the actual shock
location with increasing number of grid points and time steps, therefore stalls the overall convergence be-
haviour. This assumption is confirmed by analysing the L ∞ -norm, which gives the maximum absolute dis-
tance of the error. The convergence rate of the L ∞ -norm is indeed equal to zero.
3.5. Conclusions 51

The convergence rate of the temporal truncation error can be assessed when the exact solution is computed
using eq. (3.49). The "exact solution" still contains the spatial error, so subtraction of the computed solution
from this (semi)-exact solution gives the temporal error. The L 1 -norm and L 2 -norm of the temporal error
has been plotted in fig. 3.7b. The convergence rate of 2 for both norms is consistent with the second-order
BDF2-scheme. An overview of the convergence rates is given in table 3.6. The convergence rates have been
determined between two sequential time steps.

L 1 -norm
100
100 L 1 -norm
L 2 -norm L 2 -norm
−1
10

10−1
0.66 10−2
²∆x,∆t

²∆t
0.9 10−3
10−2
2.3
10−4

10−3 0 10−5 0
10 101 102 103 10 101 102 103
∆t [sec] ∆t [sec]

(a) Total error. (b) Temporal error.

Figure 3.7: Error plots of the solutions to Buckley Leverett’s equations.

Table 3.6: Experimental order of convergence for the Buckley–Leverett problem.

time L1, rate L2, rate L 1 , ²∆t rate L 2 , ²∆t rate


steps ²∆x,∆t ²∆x,∆t
16 1.86 × 10−1 2.86 × 10−1
32 9.96 × 10−2 0.88 2.05 × 10−1 0.48 3.04 × 10−2 5.66 × 10−2
64 5.43 × 10−2 0.75 1.48 × 10−1 0.47 8.67 × 10−3 1.8 1.90 × 10−2 1.6
128 2.93 × 10−2 0.89 1.04 × 10−1 0.50 1.95 × 10−3 2.2 4.76 × 10−3 2.1
256 1.74 × 10−2 0.88 9.27 × 10−2 0.17 4.47 × 10−4 2.1 1.15 × 10−3 2.0
512 9.41 × 10−3 0.90 5.87 × 10−2 0.66 8.80 × 10−5 2.3 2.33 × 10−4 2.3

3.5. Conclusions
The code verification has been divided into steady state code verification and unsteady code verification.
In the steady state code verification the code was verified against a forced analytical steady state solution
using the MMS, discarding the temporal component. In steady state, the temporal component is of no influ-
ence, hence this approach assesses the truncation error of the spatial discretisation. The system of PDEs was
solved on a radial coordinate system and slightly altered (by including a source term) to force the solution to
a "manufactured" analytical solution. A discrepancy was observed in the local and global truncation error,
which have been explained according to Taylor expansion analysis.

The convergence rates for the global and local truncation error are presented in table 3.7. The most important
lessons are:

• The global truncation error is not affected by the reduction in local truncation error for the elliptic
and parabolic case, due to the diffusive effect of their mathematical nature. For the hyperbolic case,
52 3. Verification

convective behaviour is dominant, hence errors introduced at the boundary can propagate throughout
the entire computational domain.

• The hyperbolic case (incompressible fluid, no capillary pressure), requires the saturation dependent
mobilities to be interpolated using an upwinding scheme. Since the applied upwinding scheme was
first-order accurate, the global truncation error of the saturation is reduced to first-order.

• Volume averaging (uniform properties over a volume ) limits the spatial discretisation to second-order
accuracy.

One can conclude that the second-order differencing scheme is consistently applied. The local truncation
error is inconsistent due to different discretisation scheme applied at the boundary nodes. Local improve-
ment can be obtained by implementing higher-order boundary stencils, but the global truncation error is not
affected by this. An overview of the convergence rates for the spatial discretisation is given in table 3.7, which
supports the conclusions made in this section. A notification should be made by the convergence rates of the
interior domain in the parabolic case. The convergence rates of the local truncation error are give by respec-
tively 1.6 and 1.4 for non-wetting and wetting phase. However, these slopes advance to a convergence rate of
2 when more volumes are considered.
Second, implementation of higher-order interior schemes has no effect on the order of the global truncation
error, due to the limiting effect of the volume approximation. For further analysis and simulations of reservoir
phenomena the central differencing scheme using a two-point boundary and two-point interior will be used.

In the unsteady code verification the temporal integration (BDF2) has been investigated. The convergence
rates of the total error were consistent with the results of Bastian [4]. The temporal error is second-order
accurate for both the L 1 and L 2 -norm .

Table 3.7: Order of convergence of the local and global truncation errors.

total domain, rate L ∞ interior domain, rate L ∞


test case interior boundary ² p n ²S w τ n τw ²p n ²S w τn τw
gradient gradient
single phase (analytical) 2 pt 2 pt 2.0 - 0.0 - 2.0 - 1.9 -
single phase (analytical) 2 pt 3 pt 1.9 - 0.9 - 1.9 - 1.8 -
single phase (analytical) 2 pt 4 pt 2.0 - 1.1 - 2.0 - 1.7 -
single phase (analytical) 4 pt 2 pt 2.0 - 1.1 - 2.0 - 1.9 -
single phase (analytical) 4 pt 3 pt 2.9 - 1.0 - 2.9 - 3.6 -
single phase (analytical) 4 pt 4 pt 3.9 - 1.9 - 3.9 - 3.5 -
single phase (MMS) 2 pt 2 pt 2.0 - 0.0 - 2.0 - 2.0 -
single phase (MMS) 2 pt 3 pt 2.0 - 1.0 - 2.0 - 1.9 -
single phase (MMS) 2 pt 4 pt 2.0 - 1.1 - 2.0 - 1.9 -
single phase (MMS) 4 pt 2 pt 2.0 - 1.1 - 2.0 - 1.9 -
single phase (MMS) 4 pt 3 pt 2.0 - 1.0 - 2.0 - 1.9 -
single phase (MMS) 4 pt 4 pt 2.0 - 2.0 - 2.0 - 1.9 -
two-phase hyperbolic (MMS) 2 pt 2 pt 2.0 1.0 0.0 0.0 1.9 1.0 0.6 1.0
two-phase hyperbolic (MMS) 2 pt 3 pt 2.0 1.0 0.9 0.9 2.0 1.0 0.6 1.0
two-phase hyperbolic (MMS) 2 pt 4 pt 2.0 1.0 1.0 0.6 2.0 1.0 0.6 1.0
two-phase parabolic (MMS) 2 pt 2 pt 2.0 2.0 0.0 0.0 2.0 2.0 1.6 1.4
two-phase parabolic (MMS) 2 pt 3 pt 2.0 2.0 0.9 1.0 2.0 2.0 1.6 1.4
two-phase parabolic (MMS) 2 pt 4 pt 2.0 2.0 1.0 1.0 2.0 2.0 1.6 1.4
4
Well-reservoir coupling
This chapter describes the coupling between the reservoir model and the well model. The well has been
modelled using a a drift-flux model. First, for a practical understanding, the governing equations of the ho-
mogeneous model will be introduced, followed by the drift-flux model in section 4.1. The drift-flux model will
be verified using an reference solution in section 4.1.3. The spatial and temporal coupling will be explained
in section 4.2. Finally, section 4.3 presents two academic problems using fully coupled dynamic simulations.

4.1. Well modelling


In this study the well has been modelled using both steady state and dynamic models. The steady state model
serves as a benchmark to the dynamic model and is referred to as the Tubing Performance Curve (see also
section 1.3). For the dynamic well model two different formulations have been used: the homogeneous well
model and the drift-flux model. The homogeneous model assumes that the two phases are perfectly mixed,
such that the fluid can be regarded as one pseudo-fluid using mixture properties. There is no slip between
the phases. The homogeneous model will be described in section 4.1.1.

The application of the homogeneous model is limited, since it assumes a single phase flow. To include the
effect of different flow regimes, slip between the phases should be taken into account. For two-phase flow,
there are two different models described in the open literature that include slip between the phases: the two-
fluid model (mechanistic model) and the drift-flux model. The two-fluid model describes the slip by solving
mass and momentum for the respective phases (and energy when non-isothermal flow is considered). This
model is in general most accurate, since it introduces models based on detailed physics of the different flow
patterns [52]. Alternatively, the drift-flux model can be used. The drift flux model can be seen as is an exten-
sion of the homogeneous model, but it allows for slip between the phases using an algebraic slip relation. The
drift-flux model is faster to compute than the two-fluid model and well- suited for the coupled simulations of
near wellbore flow. For a more elaborate description of well modelling the reader is referred to the literature
overview ([18]) and to the introduction of Hibiki and Ishii [26], Shi et al. [52].

The drift-flux model was first proposed by Zuber and Findlay [60] and has been refined by many researchers
(e.g. Hibiki and Ishii [26], Kabir et al. [30]). The formulation used in this study is based on the formulation of
Hendrix [25], Pan et al. [41, 42], Schutte [51], Shi et al. [52]. The governing equations of the model are briefly
described in section 4.1.2.

The drift-flux model used in this study serves as initial approach to demonstrate the dynamic coupling be-
tween the well and the reservoir, including transition between flow patterns. The empirical parameters used
in the drift-flux model are based on values reported by Pan et al. [42], Shi et al. [52]. However, small varia-
tions of these fitting parameters result in large variations of the solution. Therefore, the parameters should
be closely studied and tuned for industrial applications. The focus of this work was to demonstrate and in-
vestigate the coupling. Further investigation of the fitting parameters is outside the scope of this study.

53
54 4. Well-reservoir coupling

s i 2
Ω i +1/
uk
i
i ,α k
ρ k, p
i
2
i −1/
uk

i
Ak

Figure 4.1: Schematic drawing of the well, where phase k = g , l .

4.1.1. Homogeneous well model


A schematic description of an interior volume of the well model is presented in fig. 4.1. The spatial coordinate
s is defined along the well trajectory and the inclination angle θ is defined with respect to the horizon, i.e.
for vertical flow θ = 90 deg. The velocities (u k ) are defined at the faces, while the density (ρ k ), pressure (p)
and the hold-up (αk ) are defined at the volume centres (staggered grid). The governing equations of the
homogeneous well model are based on the Euler equations with an additional term for the wall friction. For
isothermal fluids and one-dimensional flow, the equations read:

∂ ¡ ¢ ∂ ¡
ρg Ag + ρ g u m A g = 0,
¢
(4.1a)
∂t ∂s
∂ ¡ ¢ ∂ ¡
ρl Al + ρ l u m A l = 0,
¢
(4.1b)
∂t ∂s
∂ ¡ ¢ ∂ ¡ 2
ρ m um A + ρ m u m A + p A + ρ m A g sin θ + T w = 0.
¢
(4.1c)
∂t ∂s

Initial and boundary conditions are given by:

p(s, 0) = p 0 (s) s ∈ Ω, (4.2a)


u k (s, 0) = u k0 (s) s ∈ Ω, (4.2b)
p(s, t ) = p d (s, t ) on Γd , ρ m u m · n = Φmn (s, t ) on Γn , (4.2c)

where Γd and Γn are the Dirichlet and Neumann boundary. The gas hold-up A g and liquid hold A l have to fill
up the total area of the tubing A, which yields an additional constraint between the gas and liquid hold-up:

A g + A l = A. (4.3)

The wall friction T w is expressed as:

Γ
Tw = f w ρ m u m |u m |, (4.4)
2

where Γ = πD is the pipeline perimeter and u m is the mixture velocity. The Fanning friction factor f w for
laminar and turbulent flow in a pipe is given by the Churchill correlation:

µµ ¶12 ¶1/12
8
fw = 2 + (a + b)−3/2 , (4.5a)
Re D

where
4.1. Well modelling 55

¶¶16
7 0.9 κ
µ µµ ¶
a = 2.457 ln + 0.27 , (4.5b)
Re D D
µ ¶16
37530
b= . (4.5c)
Re D

Here, κ is the surface roughness of the tubing and D is the diameter. The Reynolds number is based on the
diameter and mixture fluid properties

ρ m |u m |D
Re D = . (4.6)
µm

The mixture fluid properties are defined as

ρ m = ρ g A g + ρ l A l /A,
¡ ¢
(4.7a)
µm = µg A g + µl A l /A.
¡ ¢
(4.7b)

The model is closed by equations of state for both phases, which were presented in section 2.2.

4.1.2. Drift-flux model


The governing equations of the drift-flux model for one-dimensional isothermal flow read:

∂ ¡ ¢ ∂ ¡
ρg Ag + ρ g u g A g = 0,
¢
(4.8a)
∂t ∂s
∂ ¡ ¢ ∂ ¡
ρl Al + ρ l u l A l = 0,
¢
(4.8b)
∂t ∂s
∂ ¡ ¢ ∂ ³ ´
ρ g u g A g + ρ l ul A l + ρ g u g2 A g + ρ l u l2 A l + p A + ρ m A g sin θ + T w = 0. (4.8c)
∂t ∂s
Initial and boundary conditions are given by:

p(s, 0) = p 0 (s) s ∈ Ω, (4.9a)


u k (s, 0) = u k0 (s) s ∈ Ω, (4.9b)
p(s, t ) = p d (s, t ) on Γd , ρ k u k · n = Φkn (s, t ) on Γkn , (4.9c)

where Γd and Γkn are the Dirichlet and Neumann boundary for phase k. The mixture velocity u m can now
be found by dividing the total mass flux per unit area, G = (ρ g u g A g + ρ l u l A l )/A, by the mixture density of
eq. (4.7a):

G
um = . (4.10)
ρm

Again, the wall friction T w is given by the Churchill correlation of eq. (4.5a) and the mixture properties are
given by eqs. (4.7a) and (4.7b). The phase densities are given by the relations of state as expressed in sec-
tion 2.2.

In vertical flows, the gas tends to concentrate in the centre of the tubing, and therefore, the average gas veloc-
ity tends to be higher than the average liquid velocity. The slip velocity between the gas and liquid is further
increased by buoyancy effects, since the gas density is significantly lower than the liquid density [51]. Zuber
and Findlay [60] proposed the drift-flux model to account for these effects:

u g = C 0 j + ud . (4.11)
56 4. Well-reservoir coupling

Here, u g is the flow velocity of the gas phase (or the bubble rise velocity), C 0 is a profile parameter which
describes the effect of the velocity and concentration profiles, u d is the drift velocity of the gas describing
the buoyancy effect and j is the volumetric flux of the mixture, given by the sum of the superficial phase
velocities:

j = u sg + u sl = (A g u g + A l u l )/A, (4.12)

where u sg and u sl are the superficial gas and liquid velocity, respectively. Note that j is different from the
mixture velocity u m (see eq. (4.10)). The drift velocity, as it was implemented by Pan et al. [42] and Schutte
[51], is based on the formulation of Shi et al. [52]. They assumed that the drift velocity was equal to the bubble
rise velocity of gas in a stagnant liquid medium, if the gas hold-up was small (αg < a 1 ). For a large gas hold-up
(αg > a 2 ), the drift velocity is assumed to be such that the gas velocity is just able to drag a small liquid film in
upward motion, therefore preventing it from flowing downwards [51]. The values for a 1 and a 2 are given in
table 4.1. Note that αk = A k /A for k = l , g . The drift-flux velocity of the gas has the following form:

(1 − αg C 0 )K (αg , Kuc ,C 0 )u c m(θ̂)


ud = , (4.13)
ρg
q
αg C 0 ρ + 1 − αg C 0
l

where m(θ) accounts for the inclination of the well:

¢n1 ¡ ¢n 2
m(θ̂) = m 0 cos θ̂ 1 + sin θ̂ .
¡
(4.14)

Note that the inclination angle θ̂ is defined with respect to the vertical, rather than the horizontal, as was
demonstrated in fig. 4.1 and eq. (4.8a)-eq. (4.8c). The difference stems from the alternative formulation of Shi
et al. [52] and Pan et al. [41]. Since the well is assumed to be vertical, the inclination angle (θ̂) is now equal
to zero. The fitting parameters m 0 , n 1 and n 2 are given in table 4.1, obtained from experimental data of Pan
et al. [41].

The Kutateladze number represents a balance between the inertial, buoyancy and surface tension forces.
The critical Kutateladze number Kuc marks the onset of flooding, i.e. the gas velocity is unable to drag the
liquid film in upward motion. The critical Kutateladze number is a function of the Bond number, which is a
dimensionless parameter describing the ratio between body forces and surface tension forces, and is used to
characterise the shape of bubbles in a moving surrounding fluid [51]:

g (ρ l − ρ g )
µ ¶
Nb = D 2 , (4.15)
σg l

where σg l is the gas/liquid interfacial tension. The critical Kutateladze number is given by:

à Ãs !! 1
2
C ku Nb
Kuc = p 1+ 2
−1 , (4.16)
Nb C ku Cw

where C ku is a constant, and C w is the wall friction factor, both given in table
p 4.1. A typical curve of the critical
Kutateladze number as a function of the dimensionless pipe diameter Nb is given by fig. 4.2, as originally
formulated by Richter [46]. He found a value of C ku = 75, which was used by Shi et al. [52] and Schutte [51],
while Pan et al. [42] found that C ku = 142 was a better value to fit to the original experimental data of Richter
[46]. The model presented in this work is based on the fit of Pan et al. [42].

The function K (αg , Kuc ,C 0 ) is used to make a smooth transition of the drift velocity between the bubble rise
stage and the film flooding stage. Different from the linear interpolation suggested by Shi et al. [52], the
following smooth function is used [41]:
4.1. Well modelling 57

4
C ku = 75
C ku = 142

Kuc
2

0
0 20 40 60 80
p
Nb

Figure 4.2: Critical Kutateladze number as a function of the Bond number, obtained from Shi et al. [52].


1.53

 if αg < a 1
αg −a 1
³ ³ ´´
C 0 Kuc −1.53
K = 1.53 + 2 1 − cos π a2 −a1 if a 1 ≤ αg ≤ a 2 (4.17)

if αg > a 2

C K
0 uc

The characteristic velocity u c of eq. (4.13) is given by:

!1/4
σg l g (ρ l − ρ g )
Ã
uc = . (4.18)
ρ 2l

Finally, the distribution parameter C 0 has to be specified to close the system. Zuber and Findlay [60] proposed
values ranging between 1.0 and 1.5, depending on the flow regime. Several drift-flux models, (e.g. Aziz and
Govier [2] and Hasan and Kabir [24]) use a value of 1.2 for the bubbly and slug flow regime, and values close
to 1.0 for the annular mist flow regime. In addition, when the gas hold-up αg approaches 1 or the mixture
velocity becomes high, the distribution parameter C 0 should approach 1. These requirements have been
satisfied in the following relation, as proposed by Shi et al. [52]:

S A (1 − SB )2
C0 = ¢2 . (4.19)
(1 − SB )2 + (S A − 1) β − SB
¡

where β is given by;

αg |u m |
µ ¶
β = max αg SF . (4.20)
u sg f

The parameters S A , SB and SF are given table 4.1, as given by Shi et al. [52]. u sg f is the critical gas velocity that
marks the onset of flooding, which is related to the characteristic velocity u c (see eq. (4.18)) as:

ρl
s
u sg f = Kuc uc , (4.21)
ρg

For a more elaborate description of the drift-flux model the reader is referred to Hendrix [25], Pan et al. [41,
42], Schutte [51], Shi et al. [52].
58 4. Well-reservoir coupling

Table 4.1: Empirical fitting parameters used in the drift-flux model [41, 52].

fitting parameter value


a1 0.06
a2 0.21
a 10 0.4
a 20 0.7
m0 1.85
n1 0.21
n2 0.95
C ku 142
Cw 0.008
C0 1.0
SA 1.2
S 0A 1.2
SB 0.3
SF 1.0

Oil/Water flows
For oil/water flows a slightly different relation is used, as proposed by Hasan and Kabir [24], Shi et al. [52].
The oil velocity (u o ) is now given by:

u o = C 00 u l + u d0 . (4.22)

Shi et al. [52] suggested a continuous form for the profile parameter (C 00 ). The profile parameter equals 1.2
for a low oil hold-up (αo ≤ 0.4) and should approach 1.0 when oil becomes the continuous phase (αo ≥ 0.7).
This has been satisfied in the following function:


S0 if α0 < a 10
 A

 ³ α −a 0 ´
C 00 = S 0A − (S 0A − 1) a 0 −a 01
o
if a 10 ≤ α0 ≤ a 20 (4.23)
 2 1
if αg > a 20

1.0

where S 0A , a 10 and a 20 can be found in table 4.1. The oil drift velocity u d0 is give by the relation:

u d0 = 1.53 u c0 (1 − αo )2 , (4.24)

where u c0 is the characteristic velocity, now for oil and water properties (eq. (4.18)).

4.1.3. Reference solution


The drift-flux model of eq. (4.8a) - eq. (4.8c) has been verified by using a reference solution for steady-state,
compressible flow, which is given by Pan et al. [42]. Hence, we follow the derivation of Pan et al. [42]. In addi-
tion, profile effects over the cross-section are assumed to be negligible (C 0 = 1).

When the gas velocity is of the form of eq. (4.11), the steady-state momentum equation of eq. (4.8c) can be
written as:

αg ρ g ρ l 2 Γ f G2
µ ¶
d 2
ρ m um + ud = − − ρ m g cos θ̂, (4.25)
ds 1 − αg ρ m 2ρ m A

where G = ρ m u m is the total mass flow rate per unit cross-sectional (see the appendix of [42] for useful re-
lationships to express eq. (4.8c) into the form of eq. (4.25)). The solubility of one phase in the other phase
is assumed to be very low, which implies that the mass fraction of the gas, X = ρ g αg u g /G, is assumed to be
4.1. Well modelling 59

independent of position, and the densities are assumed to be a function of pressure and temperature [42].
Therefore, by definition, the gas hold-up αg can be related to the gas mass fraction X as:

jg G X /ρ g X
αg = = = , (4.26)
j + ud G X /ρ g +G(1 − X )/ρ l + u d X + a1 ρ g

where a 1 = (1 − X )/ρ l + u d /G. Since the phase densities are assumed to be a function of the temperature and
pressure only, a solution to the gas hold-up (eq. (4.26)) and drift velocity (eq. (4.13)) can be found for a given
pressure and temperature. By introducing a new unknown h = G 2 +m g ρ l u d2 /a 1 , and expanding the left-hand
side of eq. (4.25), the following form can be obtained:

∂ h Γ f G2 ∂ h
µ µ ¶ ¶ µ ¶
dP
+1 =− − ρ m g cos θ̂ − gT . (4.27)
∂P ρ m ds 2ρ m A ∂T ρ m

The temperature gradient g T is equal to zero, since isothermal flow is considered. Equation (4.27) can be
integrated, such that the following inverse pressure distribution is obtained [42]:

³ ´
∂ h
Z P 1 + ∂P ρm
s(P ) = s 0 − dP, (4.28)
P0 Γ f G
2
2ρ m A + ρ m g cos θ̂

where P 0 is the pressure at reference location s 0 . The integral relation of eq. (4.28) is solved using trapezoidal
numerical integration with an integration step size, such that the error in the "analytical" reference solution
is small enough compared to the numerical solution. A general procedure to obtain the analytical reference
solution is as follows:

• Step 1: calculate the densities ρ g and ρ l using the EOS for a series of pressures (starting at P 0 ).

• Step 2: Solve for αg and u d (see eq. (4.13) and eq. (4.26)) for the computed densities ρ g and ρ l .

• Step 3: Calculate ρ m , h(P ) and the integrand of eq. (4.28) for each pressure point P .

• Step 4: Calculate s(P ) by integrating eq. (4.28).

• Step 5: Calculate u m u g and u d from αg and u d .

4.1.4. Verification with the two-phase wellbore flow problem


The numerical implementation of the drift-flux formulation (see [25]) has been verified using the reference
solution of section 4.1.3. A well is considered with two immiscible phases, being air and water, for isothermal
conditions. The wellbore is vertical (θ̂ = 0 deg) with a length of L = 1000 m. The phase densities are com-
puted according to the EOS of section 2.2. An upward flow with a total flux per area of 50 kg/s/m2 is assumed,
identical to the case presented by Pan et al. [42]. The integration step size to obtain the reference solution is
defined as dP = (P max − P mi n )/N wel l . The remaining parameters are specified in table 4.2.

The results of the numerical and reference solution are presented in fig. 4.3. The depth profiles of pressure (a),
gas hold-up (b), gas phase velocity (c) and drift-velocity (d) show good agreement between the reference and
numerical solution. In addition, for increasing number of grid points (N wel l ) the error reduces consistently
with the trapezoidal integration approach. Note that the distribution parameter (C 0 ) is assumed to be equal
to 1, as was used by Pan et al. [42].

As mentioned at the end of the introduction to this section, the empirical fitting parameters are based on
values reported by Shi et al. [52] and Pan et al. [42]. However, small variations of the (fitting) parameters in-
volved in the drift-flux model (e.g. C 0 , m(θ̂), C ku , a 1 and a 2 ) result in large variations in the solution, possibly
non-physical. These parameters should therefore be closely studied and tuned for industrial applications.
The investigation of the fitting parameters was outside the scope of this study.
60 4. Well-reservoir coupling

0 0
Analytical Analytical
Numerical Numerical
−200 −200
Depth [m]

−400 −400

text
−600 −600

−800 −800

−1,000 −1,000
1 2 3 4 5 6 7 8 0.8 0.85 0.9 0.95 1
p [Pa] ·10 5 αg [-]

(a) pressure profile (b) gas hold-up profile


0 0
Analytical Analytical
Numerical Numerical
−200 −200
Depth [m]

−400 −400
text

−600 −600

−800 −800

−1,000 −1,000
0 5 10 15 20 25 0.2 0.3 0.4 0.5 0.6
u g [m/s] u d [m/s]

(c) gas phase velocity profile (d) drift velocity profile

Figure 4.3: Depth profiles of pressure, gas hold-up, gas phase velocity and drift velocity of the two-phase wellbore flow problem.
4.2. Coupling 61

Table 4.2: Parameters of the two-phase wellbore flow problem.

parameter value unit


L 1000 m
D 0.1 m
G 50 kg/s/m2
X 0.5 −

T (isothermal) 20 C
p wh 1 bar
p sc 1 bar
ρ g sc (air) 1.2 kg/m3
ρ l sc (water) 1000 kg/m3
σg l 72.86 × 10−3 N/m
κ (wall roughness) 2.4 × 10−5 m
friction model Churchill −
P max 20 bar
P mi n 1 bar
N wel l 200 −

4.2. Coupling
This section describes the coupling between the well and reservoir model. The coupling has a spatial and
temporal component, which will be discussed in sections 4.2.1 and 4.2.2. The coupling was earlier investi-
gated by Marino [37] and Rajkotwala [45] for single phase flow. The current coupling strategy is based on this
approach, but requires small modifications to ensure convergence of the solver. Therefore, the coupling for
single phase flow will be briefly repeated.

4.2.1. Spatial coupling


The well and the reservoir (superscript W and R respectively) interact with each other through the near well-
bore region, modelled as a radial symmetric portion of the complete reservoir. The interaction can be ex-
plained as follows. The well perceives the reservoir as an inflow condition. The mass flow rates are calculated
by the reservoir model, according to Darcy’s law, and communicated to the well model. The reservoir on the
other hand perceives the well as a pressure boundary condition. The BHP from the well is communicated to
the reservoir model. Therefore, the following conditions must hold at the well-reservoir interface:

• continuity of pressure (possibly with an additional pressure loss to account for skin effects);

• continuity of mass flow for each phase.

For single phase flow, the spatial coupling of the well and reservoir model is described by Marino [37] and
Rajkotwala [45], as visualised in fig. 4.4a. The coupling of the reservoir to the well model occurs via a source
injection term. Since the reservoir model is one-dimensional, the flow entering from the reservoir only has a
radial component, and is therefore orthogonal to the flowing direction in the well. As a result, the mass flow
entering the well will not add momentum to the flow present inside the well, since the velocity component
in well direction is equal to zero (u · n = 0). The coupling occurs thus solely by conservation of mass and
continuity of pressure[37]. This is summarised as follows:

• The well model is considered as a Dirichlet pressure boundary condition for the reservoir. The reservoir
R
pressure at the interface p 1/2 equals the pressure in the first cell of the well p 1W .

• The reservoir model is considered as a mass source term in the well, flowing into the side of the first
W
volume (see fig. 4.4a). The mass source term Q 1/2 is identical to the flux at the left face of the first cell in
R
the reservoir F 1/2 .
W
• The flux at the bottom of the well F 1/2 is equal to zero. This has been realised by putting the velocity at
W
the boundary equal to zero, i.e. u 1/2 = 0. Note that the pressure is defined at the mid-points of the vol-
umes, and the velocities at the faces of the volumes (staggered grid), with exception of the boundaries,
where both the pressure and velocity are defined.
62 4. Well-reservoir coupling

×
W
F 5/2

U2W

ΩW
2
×
W
F 3/2 ΩR1 ΩR2

R R
U1W U1/2 , F 1/2 U1R U2R
× × ×
interface R R
F 3/2 F 5/2

W
ΩW
1
U1/2
×
W
F 1/2

(a) Spatial coupling single phase flow.

×
W
F 5/2

U2W

ΩW
2
×
W
F 3/2

U1W

ΩW
1 interface
×
W W R R
U1/2 , F 1/2 ,U1/2 , F 1/2

U1R

×
ΩR1
R
F 3/2

(b) Spatial coupling two-phase flow.

Figure 4.4: Sketch of the coupling between the wellbore and reservoir through mass and pressure continuity at the well-reservoir inter-
face.
4.2. Coupling 63

Mathematically, the above-mentioned conditions can be prescribed by formulating conservation of mass of


the first volume in the well and the reservoir as follows:

¢W d ρW
1
ΩW
+ ρu A 3/2 = Q 1W ,
¡
1 (4.29a)
dt
¡ ¢R õ
d ρφ 1
¶R ¶ !
K0 ∂p K0 ∂p R
µ
R
Ω1 − 2πr hρ − 2πr hρ = 0. (4.29b)
dt µ ∂r 3/2 µ ∂r 1/2
W
¢W
= ρu A 1/2 is omitted from equation eq. (4.29a), since there is no momentum coming from the
¡
The flux F 1/2
bottom of the well. The mass flow condition reads:

K0 ∂p R
µ ¶
Q 1W = F 1/2
R
= 2πr hρ . (4.30)
µ ∂r 1/2
Continuity of the pressure is defined as:

R
p 1/2 = p 1W , (4.31)

Using a two-point formulation for the discretisation of the pressure gradient, the mass flow Q 1W is given by:

K0 R p 1R − p 1W
µ ¶
Q 1W = 2πr hρ . (4.32)
µ 1/2 r 1 − r 1/2
In two-phase flow the spatial coupling is different from the single phase coupling. A schematic drawing of
the spatial coupling in two-phase flow is shown in fig. 4.4b. The coupling is now relocated to the bottom
interface. Rather than putting the first interface flux to zero, the well interface flux (F kW ) coincides with
1/2
the first reservoir interface flux (F kR ). This modification has been made, since the drift-flux model showed
1/2
problems when very low velocities (u k << 1) are employed. This problem remains unresolved and is a subject
for further investigation, since it is envisioned in the future to include multiple reservoir layers, which requires
a sequence of coupling points in vertical direction to the side of the well. Mathematically, this leads to an
identical relation as was given by eq. (4.29a) and eq. (4.29b), but now for the two respective phases k:

¢W
d ρk Ak 1
¡
¢W ¡ ¢W
ΩW + ρ k u k A k 3/2 − ρ k u k A k 1/2 = 0, for k = g , l ,
¡
1 (4.33a)
dt
¢R
d ρ k φS k 1
õ ¶ !
∂p k R ∂p k R
¡ ¶ µ
ΩR1 − 2πr hρ k λk − 2πr hρ k λk = 0 for k = n, w. (4.33b)
dt ∂r 3/2 ∂r 1/2
The mass flow conditions for eq. (4.33a) read:

∂p k R
µ ¶
¢W
ρ k u k A k 1/2 = F kR1/2 = 2πr hρ k λk
¡
for k = n, w. (4.34)
∂r 1/2
Note that the phases k = l , g of the well correspond to the phases k = n, w in the reservoir, respectively.
Continuity of pressure is given by:

p kR1/2 = p 1W for k = n, w, (4.35)

where p nR1/2 = p wR
1/2
due to negation of the capillary pressure. When capillary effects are taken into account,
the different fluid pressures of the reservoir need to be matched to the single fluid pressure of the well.
This can be done by for example simply averaging the fluid pressures and coupling to the well pressure, i.e.
(p nR + p w
R
)/2 = p W . However, most studies with respect to well-reservoir coupling neglect capillary effects in
the near wellbore region. This subject needs to be further investigated when capillary effects are important.

Lastly, the geometry of fig. 4.4b suggests that momentum that is coming from the reservoir is in the same
direction as the flow inside the well. However, from a physical perspective it is plausible to assume that the
flow coming from the reservoir is orthogonal to the flow inside the well, hence momentum is neglected.
64 4. Well-reservoir coupling

4.2.2. Temporal coupling


The temporal coupling of the well and the reservoir can be done by using fully implicit methods, partially
implicit-explicit (IMEX) methods [56] and fully explicit methods (see section 2.7.3). A primary classification
is taken from fluid-structure interaction (see Felippa et al. [21], Piperno and Farhat [44]):

• monolithic

• partitioned

– partitioned explicit
– partitioned implicit

In this study the coupling has been done in a monolithic fashion as a first approach. The monolithic ap-
proach solves the well-reservoir system simultaneously. The underlying PDEs are collected into one system,
forming a single set of equations which are solved simultaneously. A property of the monolithic approach is
that the system can be expected to be unconditionally stable if the non-linearities are mild. Unconditionally
stable schemes have no restriction on the admissible time step. This is a desired property since the large
characteristic time scales of the reservoir would require many time steps when the admissible time step is
restricted. The disadvantage of the monolithic approach is that generally larger systems have to be solved,
which increases the computational costs. Note that the efficiency of the temporal coupling has not been in-
vestigated in this study, but can be a subject for future research.

Recall that the compact form of the semi-discretised equations for the well model and the reservoir model
are given by:

dU W
+ F W (W W , t ) = 0, (4.36)
dt
dU R
+ F R (W R , t ) = 0. (4.37)
dt

In the monolithic approach the semi-discretised equations of eqs. (4.36) and (4.37) are combined into a single
set of equations, and the system is integrated using the BDF2-scheme, as explained in section 2.7.3. The
system of equations is then given by:

dU t ot
+ F t ot (W t ot , t ) = 0, (4.38)
dt

where

· W¸ · W¸
FW
· ¸
W U
W t ot = , U t ot = , F t ot = R . (4.39)
WR UR F

Note that F W = F W (Wt ot ) and F W = F W (Wt ot ), due to the coupling of the well and the reservoir.

4.3. Test cases


So far, the well model and the reservoir model have been discussed individually. The models have been ver-
ified using MMS and/or analytical solutions, and the verification showed that the discretisation of the equa-
tions and the implementation in the code were done correctly. Then, the coupling between the well-reservoir
model has been discussed, both spatially and temporally, such that the coupled well-reservoir model is able
to predict fluid flow for steady-state and transient conditions. This section will present two different scenar-
ios in which the focus lies on transients in the well (section 4.3.1) and in the reservoir (section 4.3.2). These
test cases will demonstrate the added value of fully coupled dynamic simulations with respect to steady state
performance curves, as explained in section 1.3.
4.3. Test cases 65

·106
0.7 4
αl
T PC p wh =10 bar
0.6 T PC p wh =9 bar
t = 40 sec I PC
3.5
t = 0 sec dynamic model
0.5 PI-model
3
steady state (dynamic)
0.4 steady state (PI)
αl [−]

p [Pa]
time instance
2.5
0.3
2
0.2
t = 350 sec
0.1 1.5
t = 1 day

0 1
50 100 150 200 250 300 350 400 450 500 550 600
G [kg/s/m2 ]

Figure 4.5: TPC and IPC for the ramp-up case.

4.3.1. Production ramp-up


The steady state operating point of a well can be predicted with a nodal analysis, as explained in section 1.3.
In this section, the operating point, which can be found by the intersection of the steady-state IPC and TPC,
is being used to verify the coupled simulator under steady-state conditions. Second, the coupling of the
dynamic well model with the dynamic reservoir model is compared to the coupling with a linear PI-model
for the reservoir.
The IPC has been obtained by simulating the reservoir for various BHPs (p w f ), while prescribing the LGR at
the reservoir end (LGR e ). The LGR is given by eq. (2.91). Similarly, the TPC has been obtained by calculating
the pressure drop for a range of mass flow rates, while maintaining the THP (p wh ) constant and prescribing
the same LGR as used for the reservoir. Figure 4.5 shows the steady state curves of the well and the reservoir.
The initial steady state operating point can be found by the intersection of the two curves, given by the blue
dot at t = 0 sec. On can observe that a second operating point is possible at the right bottom corner (black
diamond). The intersection point to the left of the minimum of the TPCs is associated with unstable flow,
while the intersection point to the right of the minimum is associated with stable flow. This will be discussed
subsequently.

An unloading event is simulated by a ramp-up of the production. The well is originally producing at a low
production rate (G = 60 kg/s/m2 ), with a relatively high liquid content in the well (αl = 0.4). The production
is ramped-up by reducing the THP p wh = 10 bar to p wh = 9 bar. A new steady-state production point will be
obtained after about t = 1 day , which can be found by the intersection of the new TPC (green curve of fig. 4.5)
and IPC. The black diamond is the numerical solution of the coupled simulator when the new steady state
with reduced THP is obtained. The numerical solution is in good agreement with the predicted intersection
of the THP and IPC. This example verifies that the the coupling is well implemented, at least for steady state.
However, more interesting is the transient from the original steady state to the new steady state, which cannot
be predicted with a nodal analysis, but requires a coupled dynamic simulation. We will come back to this at
the end of this section.

The reservoir has been modelled as a radial near wellbore region using Nr es = 64 volumes, with an outer ra-
dius of r e = 100 m and a perforated section of h = 10 m. The permeability of the reservoir is K0 = 250 mD and
the porosity is φ = 0.2. The LGR is prescribed at the reservoir end, LGR e = 0.5, and the reservoir end pressure
is set to p e = 40 bar. The parameters of the reservoir are specified in table 4.3. The same well properties have
been used as presented in section 4.1.4. The parameters of the well are given in table 4.2, with exception of
66 4. Well-reservoir coupling

Table 4.3: Parameters of the reservoir model.

parameter value unit


r w f = D/2 0.05 m
re 100 m
h 10 m
φ 0.2 −
K0 250 mD
krk Corey (λ = 2) −
pc 0 bar
S nr = S wr 0 −
pe 40 bar
Nr es 64 bar

the number of well volumes (N wel l = 64) and the THP. The initial steady state THP is given by p wh = 10 bar
(see red curve of fig. 4.5). Subsequently, the THP is reduced to p wh = 9 bar (see green curve of fig. 4.5) using a
hyperbolic tangent decay function with ∆t = 10 sec.

The results of the production ramp-up case are presented in fig. 4.6. Figures 4.6a to 4.6d show the depth
profiles of the liquid hold-up αl , well pressure p W , gas velocity u g and drift velocity u d for various moments
in time.
At t = 5 sec the THP is reduced. The change in pressure results in an immediate change of the phase velocities,
and an expansion wave propagates to the bottom of the well. At t = 15 sec (see fig. 4.7), the bottom of the well
is reached. A prediction of the acoustic wave velocity for multiphase flow (in steady state) can be calculated
with the following relation, obtained from Brennen [8]:

!!−1/2
αg
à Ã
αl
c m = (ρ g αg + ρ l αl ) + . (4.40)
ρ g c g2 ρ l c l2

For the given conditions at t = 5 sec, the average multiphase acoustic velocity (changes along the well path)
is given by c m ≈ 100 m/s, such that the expansion wave reaches the bottom of the well in ∆t = L/c m ≈ 9.9 sec.
This is consistent with the observations of t = 15 sec, since the THP is reduced after t = 5 sec.
When the expansion wave reaches the bottom of the well, the BHP is reduced. The reduction in BHP increases
the drawdown of the reservoir, with an increase in production as a result (see relative productivity of fig. 4.7).
Since the mass flow rate increases, and the well diameter remains constant, the phase velocities will increase
as well, as can be seen in fig. 4.6c. The larger phase velocities give rise to a lower liquid hold-up in the well.
To obtain this new state of reduced liquid hold-up, a large amount of liquid has to be produced out of the
well. After t = 15 sec the liquid hold-up starts to reduce rapidly at the bottom of the well. At t = 100 sec, a
large amount of liquid has already travelled roughly 200 m (depth is −800 m) in upward direction. This liquid
"slug" (not filling the entire cross-section) propagates to the top of the well in roughly t = 350 sec. This can be
seen by the various "snap-shots" of time instances of the liquid hold-up in fig. 4.6b.
The reduction in liquid hold-up in the well reduces the hydrodynamic head. This can be seen in fig. 4.6a.
Initially, the well has a pressure drop of roughly ∆p = 25 bar, dominated by the hydrodynamic head of the
heavier liquid phase, mainly located in the lower sections of the well. This explains the convex pressure dis-
tribution at t = 0 sec. The liquid hold-up continues to increase when we move into deeper sections of the well.
This gives rise to an additional hydrodynamic head in the lower sections, therefore increasing the pressure
drop even further (d2 p/dx 2 > 0). When the liquid slug travels upwards, the convex pressure profile becomes
concave, depicted by the cyan distribution at t = 200 sec. The hydrodynamic head is mainly built-up in the
upper sections of the well, and then increases marginally, due to the little contribution of the friction related
pressure drop (d2 p/dx 2 < 0). Eventually, after t = 350 sec, most of the liquid has been produced. The total
pressure drop over the well reduced to ∆p = 10 bar and is now dominated by friction, rather than the hydro-
dynamic head. From this point, the well slowly reaches its new steady state after t = 1 day.

The characteristic time scales of the reservoir are much slower than characteristic time scales of the well. Fig-
ure 4.6e and fig. 4.6f show the pressure and saturation distribution for the same time instances as the well.
4.3. Test cases 67

0 0
t = 0 sec t = 0 sec
t = 15 sec t = 15 sec
−200 t = 100 sec −200 t = 100 sec
t = 150 sec t = 150 sec
t = 200 sec t = 200 sec
Depth [m]

−400 t = 250 sec −400 t = 250 sec

text
t = 350 sec t = 350 sec
−600 −600

−800 −800

−1,000 −1,000
1 2 3 4 0 0.1 0.2 0.3 0.4 0.5
p [Pa] ·106 αl [−]

(a) Depth profiles of the well pressure. (b) Depth profile of the gas hold-up fraction.
0 0
t = 0 sec t = 0 sec
t = 15 sec t = 15 sec
−200 t = 100 sec −200 t = 100 sec
t = 150 sec t = 150 sec
t = 200 sec t = 200 sec
Depth [m]

−400 t = 250 sec −400 t = 250 sec


text

t = 350 sec t = 350 sec


−600 −600

−800 −800

−1,000 −1,000
0 10 20 30 40 50 60 0 0.2 0.4 0.6 0.8 1 1.2
u g [m/s] u d [m/s]

(c) Depth profiles of the well velocities. (d) Depth profiles of the drift velocity.
6
·10
4

0.52
3.5
S w [−]
p [Pa]

3 t = 0 sec
t = 0 sec 0.5 t = 15 sec
t = 15 sec
t = 100 sec
t = 100 sec
2.5 t = 150 sec
t = 150 sec
t = 200 sec
t = 200 sec
0.48 t = 250 sec
t = 250 sec
2 t = 350 sec
t = 350 sec
0 2 4 6 8 10
0 2 4 6 8 10
r [m]
r [m]
(f) Reservoir saturation distribution.
(e) Reservoir pressure distribution.

Figure 4.6: Depth profiles of the production ramp-up case.


68 4. Well-reservoir coupling

1.5 0.65

1.4
0.6
1.3

P I /P I 0 [−]

LGR [−]
1.2
0.55

1.1

P I /P I 0
1 0.5
LGR
0 100 200 300 400 500
t [sec]

Figure 4.7: LGR and normalised PI for the production ramp-up case.

The reduction of the BHP (slightly) changes the overall pressure profile, but in this relatively brief period of
time, the saturation profile is hardly influenced. Saturation fronts, especially of liquid, may travel only several
meters a day (see also section 3.4.2), which explains the almost unaffected saturation distribution.

We already mentioned that the operating point to the left of the minimum of the TPC in fig. 4.5 is associated
with unstable flow, while the new operating point after the reduction in THP, is associated with stable flow.
When considering the unstable point, the small reduction in THP "triggered" the well in the following way.
The expansion wave originating from the tubing head reduces the BHP and leads to an increase in drawdown
and thus to an increase in production. Since the increase in mass flow rate is associated with a reduction in the
liquid hold-up (dαl /dQ < 0, see also fig. 4.5), the hydrodynamic head reduces even further (dp h yd /dQ < 0),
aggravating the drawdown and production from the reservoir. As long as the reduction in liquid hold-up re-
sults in an increase in drawdown, the production continues to grow. However, the higher production rate
leads to additional friction loss (dp f r i c /dQ > 0), which in turn increases the BHP, therefore counters the re-
duction in BHP due to the reduction in hydrodynamic head. At the minimum of the TPC, which can be
regarded as the resultant of the hydrodynamic and friction related pressure drop, the two mechanisms are
in balance. Further increase of the mass flow rate leads to a higher friction drop, negatively influencing the
drawdown and thus the production. Fluctuations in the pressure or the mass flow rate are then encountered
by the system itself. This is associated with stable production.

We argued that the transient behaviour, which is found at the movement from the unstable production point
to the stable production point, could only be calculated by simulating both the well and the reservoir dynam-
ically. To support this statement, both the coupling with a dynamic (black) and steady state PI model (gray)
have been simulated. The results have been presented in fig. 4.5.
The new steady state operating point of the dynamic model and the PI model are different.The constant PI
model calculates the corresponding inflow for a given BHP with a linear relation (eq. (1.1)). The IPC of the
reservoir is linear, as is used in the PI model. Therefore, the operating point of the (green) TPC and linear PI
curve is incorrect.
The two models give a different prediction of the transient behaviour from the original steady state to the new
steady state. Figure 4.7 shows the PI of the dynamic model for the time interval t = 0 sec to 500 sec. The PI
has been normalised with the PI at time instant t 0 = 0 sec, i.e. P I 0 . One can observe that The PI changes over
time, and is certainly not constant. The variation in PI is related to the change in LGR, shown by the red line
in fig. 4.7. The constant PI model assumes a fixed LGR. The dynamic simulation clearly showed that the LGR
changes considerably. The linear PI-model is thus not capable of predicting the transient behaviour correctly
and at least for gas/water flow is not capable of predicting the correct new steady state. A better prediction
of the new steady state can be obtained when a look-up table is constructed before hand, containing the PIs
for a range of mass flow rates and BHPs of the reservoir. Simulation of the transient behaviour requires a
dynamic coupled simulation.
4.3. Test cases 69

Table 4.4: Parameters for the water flooding case.

reservoir parameter value unit well parameter value unit


r w f = D/2 0.05 m L 1000 m
re 100 m D 0.1 m

h 5 m T (isothermal) 20 C
φ 0.2 − p wh 1 bar
K0 1100 mD p sc 1 bar
krk Corey (λ = 2) − ρ o sc (comp. oil) 800 kg/m3
pc 0 bar ρ w sc (inc. water) 1000 kg/m3
S nr = S wr 0 − µo 0.5 × 10−3 Pa s
pe 120 bar µw 1 × 10−3 Pa s
Nr es 128 M 2 −
σow 40 × 10−3 N/m
κ (wall roughness) 2.4 × 10−5 m
friction model Churchill −
N wel l 32 −

4.3.2. Water flooding


Primary hydrocarbon recovery occurs due to the initial energy that is present in the reservoir [36]. The high
pressure and temperature conditions result in a natural flow when a well is connected to the surface. How-
ever, when the reservoir matures, the average reservoir pressure declines, resulting in a reduction in produc-
tion. Secondary recovery involves the introduction of artificial energy into the reservoir by means of fluid
injection. One conventional method of secondary recovery is water flooding [36] (see also the Buckley Lev-
erett equation of section 2.5.3). Water flooding is the injection of water by means of an injection well, pushing
the oil to the production well. This test case aims at simulating a typical water flooding scenario for an oil pro-
duction well using a fully coupled well-reservoir.

In contrast to the test case of section 4.3.1, this case focusses on the transient in the reservoir. The well and
reservoir are initially under steady-state conditions. The injection of water is modelled by an instantaneous
increase of the saturation (from S w = 0.3 to S w = 1 ) at the reservoir end (r = r e ). A (water) saturation front will
move into the direction of the well. As explained in section 2.5.3, the saturation front may contain a shock
and/or expansion wave solution. This is dependent on the flux function (eq. (2.34)), which contains the
characteristic velocity of the saturation front (eq. (2.36)). The shape of the saturation front is determined by
the frontal mobility ratio M = µw /µn . This test case assumes incompressible water and slightly compressible
oil, with M = 2. This makes the test case comparable to a Buckley-Leverett problem (which requires strictly
incompressible flow). The remaining reservoir properties are specified in table 4.4.
The drift flux model for oil/water flows is different from gas/water flows. The drift velocity of the oil (u o0 ) is
presented at the end of section 4.1.2. The remaining parameters of the well are specified in table 4.4.
Lastly, the water flooding case has been simulated with different time steps to investigate the transients in
both the well and the reservoir. To that end, first a coupled simulation of t end = 8 days is carried out with
∆t = 8/128 days. This simulation is used to identify the moment at which the saturation front reaches the
bottom hole of the well. In a second simulation, the time step has been reduced to ∆t = 100 sec to investigate
the transients in the well. This "second" simulation runs from t = 4.2 days to 4.8 days.

The results for the reservoir are presented in fig. 4.8. An expansion wave followed by a shock solution moves
in the direction of the well, indicated by the different time instances of fig. 4.8b. The pressure distribu-
tion of fig. 4.8a changes as the saturation front propagates through the domain, but the BHP remains (al-
most) unaffected. After t = 4.2 days, the front reaches the bottom hole. Until this moment, a time step of
∆t = 4.2/128 days is used. The dynamics in the well are expected to take place at a different time scale, hence
the simulation is continued for half a day, with a time step of ∆t = 100 sec. The water cut in the bottom of the
well will now rapidly increase.

The results of the second simulation for t = 4.2 days to 4.8 days are presented in fig. 4.9. The water hold-
up (αw ) is presented in fig. 4.9b. The water hold-up gradually increases as the Water-to-Oil Ratio (WOR)
70 4. Well-reservoir coupling

·107 1.2
1.2 t = 0 day
t = 1.9 days
1.18 1 t = 2.9 days
t = 3.6 days
t = 4.2 days
1.16 0.8

S w [−]
p [Pa]

1.14 0.6
t = 0 day
t = 1.9 days
1.12 t = 2.9 days 0.4
t = 3.6 days
t = 4.2 days
1.1 0.2
0 10 20 30 40 0 20 40 60 80 100
r [m] r [m]

(a) Reservoir pressure distribution. (b) Reservoir saturation distribution.

Figure 4.8: Water flooding simulation results for t = 0 days to 4.2 days with ∆t = 2.84 × 103 sec.

grows over time. In contrast to the production ramp-up case of section 4.3.1, no slug flow behaviour can be
observed. This can be explained in three ways.
First, the flow conditions are such that they are not in the slug flow regime.
Second, the drift-flux model for oil-water flows is totally different from gas-water flows. Both models have
been obtained from Shi et al. [52] and are not thoroughly investigated. The dynamics of the oil/water drift
flux model may not include "slug-flow-like" behaviour.
Third, the time step of ∆t = 100 sec might be too large to capture transients in the well. This has been inves-
tigated by reducing the time step to ∆t = 1 sec and continuing the simulation from the point t = 4.2 days. It
appeared that for such a time step, the change in WOR is too slow. The reservoir can be regarded as a constant
inflow with fixed WOR. From this we can conclude that the water hold up gradually increases over time.
The pressure distribution in the well is presented fig. 4.9a. One can observe that the pressure remains con-
stant during this time interval. Even though the WOR changes significantly, the productivity is hardly influ-
enced (see fig. 4.10). Since the total mass flow rate, or PI, is coupled to the BHP, it remains unaffected as well.
Apparently, the redistribution in the well does not affect the pressure distribution. This can be explained by
the fact that the density of oil and water are relatively close to each other and that the pressure drop over the
well is mainly due to the gravity component.
Figures 4.9c and 4.9d present the pressure and saturation distribution in the reservoir during the time interval
t = 4.2 days to 4.8 days. Again, the pressure distribution is almost uninfluenced. The saturation near the well
increases from S w = 0.4 to S w = 0.7, which thus gives a significant change of the WOR. This observation is
supported by the steep slope of the WOR in the interval t = 4.2 days to 4.8 days, observed in fig. 4.10.

This test case was used to investigate the effect of a dynamic well on typical transients in the reservoir. We
can conclude that the dynamics in the reservoir are hardly affected by the well. Investigating the transients
using a time step of ∆t = 1 sec showed that the reservoir properties remain constant. From the perspective of
the reservoir, this implies that the well can be modelled as a steady relation (TPC).

4.3.3. Conclusions
We have considered two test cases for the transients in the well (section 4.3.1) and in the reservoir (sec-
tion 4.3.2). The characteristic time scales of the two test cases are completely different. For the production
ramp-up case, the transients take place in the order of seconds, whereas the transients in the water flooding
case take place in the order of days. In addition, both cases have been investigated for steady state conditions,
for which the dynamic simulator has been used. The coupled simulator showed no difficulties for simulations
with either small or large time steps. This stable behaviour is greatly desired, since it enables solving different
types of applications with one single simulator.
Considering the transient behaviour in the well, the production ramp-up case of section 4.3.1 showed that
4.3. Test cases 71

0 0
t = 4.30 days t = 4.30 days
t = 4.33 days t = 4.33 days
−200 t = 4.34 days −200 t = 4.34 days
t = 4.47 days t = 4.47 days
t = 4.80 days t = 4.80 days
Depth [m]

−400 −400

text
−600 −600

−800 −800

−1,000 −1,000
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 1.2
p [Pa] ·107 αw [−]

(a) Depth profiles of the well pressure. (b) Depth profile of the water hold-up.
·107 1
1.2

0.8
1.15
S w [−]
p [Pa]

0.6
t = 4.30 days t = 4.30 days
1.1 t = 4.33 days t = 4.33 days
t = 4.34 days t = 4.34 days
0.4 t = 4.47 days
t = 4.47 days
t = 4.80 days t = 4.80 days

0 10 20 30 40 0 20 40 60 80 100
r [m] r [m]

(c) Reservoir pressure distribution. (d) Reservoir saturation distribution.

Figure 4.9: Water flooding simulation results for t = 4.2 days to 4.8 days with ∆t = 100 sec.

1.5 10
P I /P I 0
W OR
8

1
P I /P I 0 [−]

6
W OR [−]

4
0.5

0 0
0 2 4 6 8
t [days]

Figure 4.10: WOR and normalised PI for the water flooding case.
72 4. Well-reservoir coupling

the dynamic coupled approach is necessary to predict the well flow. The comparison of a dynamic reser-
voir model with a linear PI-model showed that both the transients and the new steady state were predicted
inaccurately by the PI-model. The steady state can be predicted more accurately, when a look-up table is
constructed up-front, including a range of PIs for various combinations of pressures and mass flow rates.
The water flooding case showed that the transients of the reservoir are not affected by the dynamics of the
well. The well quickly adapts to a change in bottom hole conditions, at a much faster rate than the transients
in the reservoir. Dynamic reservoir simulations can thus be carried out with steady state TPC models.
It is recommended to test the new coupled simulator for other well-reservoir transients. Transients such
as water/gas coning may have comparable characteristic time and length scales for both the well and the
reservoir. Further analysis of other (production related) transient phenomena can point out when a dynamic
fully coupled simulation is necessary.
5
Conclusions and recommendations

5.1. Conclusions
The goal of this thesis project was to extend the Shell single phase well-reservoir simulator to two-phase flow
for simulation of well reservoir transients. Four steps were identified to achieve this goal:

• a literature study on well-reservoir engineering to (1) review the fundamentals of multiphase flow in
wells and reservoirs and (2) to review well-reservoir coupling methods found in open literature;

• the development and verification of the two-phase reservoir model;

• the modification and verification of the two-phase drift-flux model of the well;

• the coupling between the well and the reservoir and the demonstration of two academic test cases
showing the benefit of dynamic models over the coupling with steady state model.

The conclusions will be discussed according to these four different steps.

literature study
The literature study identified the need for a coupled well-reservoir model in multiphase flow conditions.
Shell’s desire to optimise existing recovery techniques has two components in which the coupled simulator
simulator is essential. First, for simulation of liquid loading to ensure future production of the maturing gas
fields. Second, the design of smart wells. The potential of smart wells is to increase hydrocarbon recovery
by optimising the Inflow Control Valve (ICV) settings of multiple down hole valves. Simulation of the near
wellbore flow is essential to design and optimise such a configuration.

reservoir model
The two-phase immiscible reservoir equations have been discretised in space using a second-order accurate
finite volume method. Second-order time integration was obtained using implicit BDF2 time integration,
which is necessary due to the stiff character of the reservoir equations.
However, the accuracy of the discretisation method depends on the mathematical character of the two phase
reservoir equations, which changes under particular assumptions. The character changes with the presence
or absence of the capillary pressure and compressibility of the fluids. For compressible flows, the governing
equations are parabolic. For incompressible flows, and without the capillary pressure, the two-phase reser-
voir equations become elliptic in pressure and hyperbolic in saturation. This change in character affects the
spatial discretisation, boundary conditions and initial conditions.
For the parabolic case, a central discretisation scheme has been applied to calculate the pressure gradients
and a central averaging scheme has been used to interpolate fluid and rock properties to the faces. This re-
sulted in a second-order accurate spatial discretisation. Higher-order spatial discretisations have been con-
sidered, but due to the assumption of uniform volume properties the spatial discretisation was limited to
second-order accuracy. In addition, boundary conditions have to be prescribed on both sides of the domain
and both the pressure and saturation require an initial condition.

73
74 5. Conclusions and recommendations

For the hyperbolic case, the saturation dependent variables have been upwinded to avoid spurious oscilla-
tions. This reduces the overall spatial discretisation to first-order accuracy. The pressure dependent variables
can maintain a central averaging approach.
The elliptic pressure equation makes the initial pressure condition superfluous, and has to follow from the
initial saturation distribution. In addition, the order of the saturation equation reduced from second-order
to first-order. As a result, only one boundary condition has to be prescribed. Depending on the direction
of information propagation, the "second" boundary condition follows from the numerical domain. This has
been realised by prescribing a Neumann condition.

In contrast to existing literature, an analysis of the mathematical character has been presented using eigen-
value analysis of the governing equations in quasi-linear form. This is a robust approach that is applicable
irrespective of the assumptions made (e.g. compressibility, capillary pressure, etc.). Most literature is limited
to analysing the decoupled pressure and saturation equations, rather than starting from the fully coupled
equations. The current approach also allows for easy extension to three-phase flow.

well model
The well model uses a drift-flux formulation. This formulation uses an algebraic equation to determine the
slip between the phases. This approach is preferred over the two-fluid model, due to its reduced computa-
tional time, especially when the coupling with the reservoir is considered. An existing well code, which is
a second-order accurate finite volume method, has been modified in order to be coupled to the reservoir.
Again, BDF2 time integration is employed for the temporal discretisation.
The drift-flux model implementation has been successfully verified by comparing to a reference solution.
However, investigation of the empirical (fitting) parameters in the drift flux model showed that small varia-
tions of these parameters resulted in large variations of the solution, which seem to be non-physical. There-
fore, the parameters should be carefully chosen when industrial applications are considered and further re-
search might be necessary.

coupling
The coupling has been realised in a monolithic fashion. The coupling with respect to single phase flow has
been slightly modified to accommodate the drift-flux model. The coupled simulator showed no difficulties
in running simulations with either very small time steps (seconds) or large time steps (years). This stable
behaviour is greatly desired, since it enables solving different types of applications with one single simulator.
Two different academic cases have been investigated: a production ramp-up (focus on well transients) and a
water flooding case (focus on reservoir transients). Based on these cases, the following can be concluded.
When considering the dynamics in the well, the transients in the reservoir need to be taken into account. The
comparison of the coupling with a dynamic reservoir and an Inflow Performance Curve (IPC) showed that
the transient behaviour in the well was calculated incorrectly when an IPC was employed, stressing the need
for a coupled simulator.
When considering dynamics in the reservoir, the transients in the well have a negligible effect on the dy-
namics in the reservoir. The characteristic time scales of the reservoir are generally larger (several orders in
magnitude) than the time scales of the well. In this case, the need for a coupled simulator is less evident.

5.2. Recommendations
The long term goal of Shell is to extend Compas, the in-house dynamic multiphase pipeline and well sim-
ulator, with a dynamic reservoir model. Shell is particularly interested in simulating liquid loading and in
designing intelligent systems, such as smart wells. The coupling through multiple coupling points is there-
fore essential, hence the reservoir model should be extended to multiple layers. The coupling then has to
be modified with respect to the demonstrated coupling approach. It is envisioned to couple the reservoir
to the well model via source terms at various locations (perforation heights) in the well. The problem arising
currently is that the drift-flux model has problems in solving the system for very low phase velocities, encoun-
tered at the bottom of the well. This problem should be solved before coupling via multiple source terms can
be realised.
5.2. Recommendations 75

Apart from including more physics in the model via multiple reservoir layers, the efficiency of the current
coupling approach can be improved. The monolithic approach has been taken as first starting point, but is
subject for further improvement.
This study showed that the characteristic time scales of the well and the reservoir may differ several orders
in magnitude. To reduce computational time, a partitioned approach is recommended, using different time
steps for the well and the reservoir. Alternatively, the monolithic approach can be used with dynamic time
stepping. A similar approach has already been used in the water flooding case. The simulation was split-up
into two parts using different time steps. Note that the BDF2 scheme uses two previous time steps to approx-
imate the time derivative. The use of dynamic time stepping requires modification of the time discretisation
scheme.
In the non-linear solver two processes can be identified, which are most time consuming: Jacobian evaluation
and solution of the linear system of equations. Currently, two Jacobians have to be evaluated for each Newton
sub-iteration. In the partitioned approach, computational time of the Jacobian evaluation may be reduced,
due to its reduced dimensions. In addition, the Jacobians are calculated using a numerical approach. By
deriving algebraic derivatives for the Jacobians, computational time can be reduced.
Second, when a partitioned approach is considered, the solution of the linear system requires less computa-
tional time, again due to its reduced dimensions (although the number of iterations might increase). There-
fore, the partitioned approach has the potential to reduce computational time considerably as both the di-
mensions of the Jacobians and the solve matrix are reduced. Note that partitioned approach may require
more iterations. An efficiency study of the two methods should point out which method is preferred.

The focus of this study was on development and verification of a dynamic two-phase well-reservoir code. The
performance of the code has been tested using academic test cases. However, validation of the model has not
been considered. It is recommended to validate the model using experimental and/or field data.
Bibliography
[1] S.N. Antontsev. On the solvability of boundary value problems for degenerate two-phase porous flow
equations. Dinamika Splosnoi Sredy Vyp, 10:28–53, 1972.

[2] K. Aziz and G.W. Govier. Pressure drop in wells producing oil and gas. Journal of Canadian Petroleum
Technology, 11(03):38–48, 1972.

[3] K. Aziz and A. Settari. Petroleum Reservoir Simulation. Applied Sience Publishers LTD, London, 1979.

[4] P. Bastian. Numcerical computation of multiphase flow in porous media. PhD thesis, Technischen
Fakultät der Christian–Albrechts–Universität Kiel, 1999.

[5] J. Bear. Dynamics of Fluids in Porous Media. Dover Publications, 1972.

[6] S.P.C. Belfroid, W.L. Sturm, G.J.N. Alberts, M.C.A.M. Peters, and W. Schiferli. Prediction of Well Perfor-
mance Instability in Thin Layered Reservoirs. In 2005 SPE Annual Technical Conference and Exhibition,
Dallas, Texas, U.S.A., 9 - 12 October 2005. SPE 95835.

[7] John B Bell, John A Trangenstein, and Gregory R Shubin. Conservation laws of mixed type describing
three-phase flow in porous media. SIAM Journal on Applied Mathematics, 46(6):1000–1017, 1986.

[8] C. E. Brennen. Fundamentals of multiphase flow. Cambridge university press, 2005.

[9] H. Brooks, R and A.T. Corey. Hydraulic properties of porous media and their relation to drainage design.
Transactions of the ASAE, 7(1):26–0028, 1964.

[10] G. Chavent and J. Jaffré. Mathematical models and finite elements for reservoir simulation: single phase,
multiphase and multicomponent flows through porous media. North-Holland, 1986.

[11] Z. Chen, G. Huan, and Y. Ma. Computational methods for multiphase flows in porous media, volume 2.
Siam, 2006.

[12] S.K. Choi and W. Huang. Impact of Water Hammer in Deep Sea Water Injections Wells. In SPE Annual
Technical Conference and Exhibition, Denver, Colorado, U.S.A., 30 October - 2 November 2011. SPE
146300.

[13] A.T. Corey. The interrelation between gas and oil relative permeabilities. Producers monthly, 19(1):38–41,
1954.

[14] M.C.C. Cunha, M.M. Santos, and J.E. Bonet. Buckley–Leverett mathematical and numerical models
describing vertical equilibrium process in porous media . International Journal of Engineering Science,
42(11–12):1289–1303, 2004.

[15] D.V.A. da Silva and J.D. Jansen. A Review of Coupled Dynamic Well-Reservoir Simulation. IFAC-
PapersOnLine, 48(6):236–241, 2015.

[16] L.P. Dake. Fundamentals of Reservoir Engineering. Elsevier Science, 1978.

[17] H.P.G. Darcy. Les fontaines publiques de la ville de Dijon. Exposition et application á suivre et des formules
á employer dans les questions de duistribution d’eau. Victor Dalmont, 1856.

[18] I.R. Doǧan. Literature Study. Technical report, Delf University of Technology, 2016.

[19] M.J. Economides, A.D. Hill, C. Ehlig-Economides, and D. Zhu. Petroleum production systems. Pearson
Education, 2012.

[20] T Ertekin, J.H. Abou-Kassem, and G.R. King. Basic applied reservoir simulation. SPE textbook series,
2001.

77
78 Bibliography

[21] C. A Felippa, K.C. Park, and C. Farhat. Partitioned analysis of coupled mechanical systems. Computer
methods in applied mechanics and engineering, 190(24):3247–3270, 2001.

[22] P.W.J. Glover. Petrophysics, Course Notes. University of Laval, 2000.

[23] H. Hajibeygi and T. Matei. Lecture Notes Reservoir Simulation, 2016.

[24] A.R. Hasan and C.S. Kabir. A Study of Multiphase Flow Behavior in Vertical Wells. SPE Production Engi-
neering, 3:263–272, May 1988.

[25] M.H.W. Hendrix. Overview of the principles of by-pass pigging with speed control. Technical report,
Shell Global Solutions International B.V., 2014.

[26] T. Hibiki and M. Ishii. One-dimensional drift-flux model and constitutive equations for relative motion
between phases in various two-phase flow regimes. International Journal of Heat and Mass Transfer, 46:
4935–4948, 2003.

[27] B. Hu, K. Veeken, R. Yusuf, and H. Holmas. Use of Wellbore-Reservoir Coupled Dynamic Simulation to
Evaluate the Cycling Capability of Liquid-Loaded Gas Wells. In SPE Annual Technical Conference and
Exhibition, Dijon, France, 19 - 22 September 2010. SPE 134948.

[28] S.J. Hulshoff. CFD II Part 1: Discretisations for Compressible Flows, 2012.

[29] W. Hundsdorfer. Partially implicit BDF2 blends for convection dominated flows. SIAM Journal on Nu-
merical Analysis, 38(6):1763–1783, 2001.

[30] C.S. Kabir, A.R. Hasan, D.L. Jordan, and Z. Wang. A Wellbore/Reservoir Simulator for Testing Gas Wells
in High-Temperature Reservoirs. SPE Formation Evaluation, pages 128–134, June 1996.

[31] J.F. Lea, H.V. Nickens, and M.R. Wells. Gas Well Deliquification. Elsevier Inc, 2008.

[32] A. P. Leemhuis, E. D. Nennie, S. Belfroid, G. Alberts, L. Peters, and G. J.P. Joosten. Gas Coning Control for
Smart Wells Using a Dynamic Coupled Well-Reservoir Simulator. In SPE Intelligent Energy Conference
and Exhibition, Amsterdam, The Netherlands, 25 - 27 February 2008. Society of Petroleum Engineers.
SPE 112234.

[33] A.P. Leemhuis, S.P.C. Belfroid, and G.J.N. Alberts. Gas Coning Control for Smart Wells. In SPE Annual
Technical Conference and Exhibition, Anaheim, California, U.S.A., 11 - 14 November 2007. SPE 110317.

[34] Randall J. LeVeque. Finite Volume Methods for Hyperbolic Problems. Cambridge Texts in Applied Math-
ematics. Cambridge University Press, 2002. ISBN 9780521810876.

[35] B. Loret. Lecture Notes on Partial Differential Equations , 2009.

[36] C.W. Lyons. Standard Handbook of Petroleum & Natural Gas Engineering - Volume 2. Gulf Professional
Publishing, 1996.

[37] M. Marino. Dynamic Flow Modelling for Well-Reservoir Interaction. Master’s thesis, Delft University of
Technology, December 2014.

[38] M. Marino and B. Sanderse. Dynamic Multiphase Flow Modelling for Well-Reservoir Interaction - A
Literature Study . Technical report, Shell Global Solutions International B.V., 2014.

[39] R.M.M. Mattheij, S.W. Rienstra, and J.H.M Ten Thije Boonkkamp. Partial differential equations: model-
ing, analysis, computation. Siam, 2005.

[40] E.D. Nennie, G.J.N. Alberts, S.P.C. Belfroid, E. Peters, and G.J.P. Joosten. An Investigation Into the Need
of a Dynamic Coupled Well-Reservoir Simulator. In SPE Annual Technical Conference and Exhibition,
Anaheim, California, U.S.A., 11 - 14 November 2007. SPE 110316.

[41] L. Pan, C.M. Oldenburg, S. Wu, and K. Pruess. T2Well/ECO2N Version 1.0: Multiphase and Non-
Isothermal Model for Coupled Wellbore-Reservoir Flow of Carbon Dioxide and Variable Salinity Water.
Earth Sciences Division, Lawrence Berkeley National Laboratory University of California, 2011.
Bibliography 79

[42] L. Pan, S.W. Webb, and C.M. Oldenburg. Analytical solution for tho-phase flow in a wellbore using the
drift-flux model. Advances in Water Resources, 34(12):1656–1665, 2011.

[43] Ø. Pettersen. Basics of Reservoir Simulation With the Eclipse Reservoir Simulator. Technical report,
Department of Mathematics, University of Bergen, 2006.

[44] S. Piperno and C. Farhat. Partitioned procedures for the transient solution of coupled aeroelastic prob-
lems. Part II: energy transfer analysis and three-dimensional applications. Computer Methods in Applied
Mechanics and Engineering, 190(24):3147–3170, 2001.

[45] A.H. Rajkotwala. Coupling Methods for Well Reservoir Interaction. Master’s thesis, University of Twente,
August 2015.

[46] Horst J. Richter. Flooding in tubes and annuli. International Journal of Multiphase Flow, 7(6):647 – 658,
1981.

[47] P.J. Roache. Verification and validation in computational science and engineering. Hermosa, 1998.

[48] P.J. Roache. Code Verification by the Method of Manufactured Solutions. Journal of Fluids Engineering,
124:4–10, 2002.

[49] J. Sagen, T. Sira, S. Selberg, M. Chaib, and H. Eidsmoen. A coupled dynamic reservoir and pipeline model
- development and initial experience. BHR Group Multiphase Production Technology, 13:359–372, 2007.

[50] F.J. Santarelli, E. Skomedal, P. Markestad, H.I. Berge, and H. Nasvig. Sand Production on Water Injec-
tors: How Bad Can It Get? In SPE/ISRM Rock Mechanics in Petroleum Engineering. Society of Petroleum
Engineers, 2000.

[51] K.C.J. Schutte. A Hydrodynamic Perspective on the Formation of Asphaltene Deposits. PhD thesis, Tech-
nical University of Delft, Delft, The Netherlands, 2016.

[52] H. Shi, J.A. Holmes, L.J. Durlofsky, K. Aziz, L.R. Diaz, B. Alkaya, and G. Oddie. Drift-Flux Modeling of
Two-phase Flow in Wellbores. SPE Journal, March 2005. SPE 84228.

[53] L. Sinégre, N. Petit, P. Lemétayer, P. Gervaud, and P. Ménégatti. Casing heading phenomenon in gas lifted
well as a limit cycle of a 2D model with switches. IFAC Proceedins Volumes, 38(1):531–536, 2005.

[54] W.L. Sturm, S.P.C. Belfroid, O. van Wolfswinkel, M.C.A.M. Peters, and F.J.P.C.M.G. Verhelst. Dynamic
Reservoir Well Interaction. In SPE Annual Technical Conference and Exhibition, Houston, Texas, U.S.A.,
26-29 September 2004. SPE 90108.

[55] CJ Van Duijn, Lambertus A Peletier, and Iuliu Sorin Pop. A new class of entropy solutions of the Buckley-
Leverett equation. SIAM Journal on Mathematical Analysis, 39(2):507–536, 2007.

[56] A. H. van Zuijlen and H. Bijl. Implicit and explicit higher order time integration schemes for struc-
tural dynamics and fluid-structure interaction computations. Computers and Structures, 83(2–3):93–
105, 2005. ISSN 0045-7949. Advances in Analysis of Fluid Structure Interaction.

[57] Y Wang and C. Kao. Central schemes for the modified Buckley–Leverett equation. Journal of Computa-
tional Science, 4(1):12–23, 2013.

[58] T. Wen and G. Zabaras. Water hammer analysis perdido water injection well. Shell Report SR.12.14070,
2012.

[59] P.H. Winterfeld. Simulation of Pressure Buildup in a Multiphase Wellbore/Reservoir System. SPE For-
mation Evaluation, 4(2):247–252, 1989.

[60] N. Zuber and J.A. Findlay. Average Volumetric Concentration in Two-Phase Flow Systems. Journal of
Heat Transfer, ASME, 87:456–468, 1965.

You might also like