Fulltext01

Simulation of Unstable Two-phase Flows
in Long Risers
Andreas Holm Akselsen
Master of Energy and Environmental Engineering

Submission date: June 2012
Supervisor: Ole Jørgen Nydal, EPT
Co-supervisor: Tor Brox Kjeldby, EPT
Norwegian University of Science and Technology

Department of Energy and Process Engineering
Simulation of Unstable Two-phase
Flows in Long Risers
Master Thesis
Author: Supervisor:
Andreas H. Akselsen Prof. Ole Jørgen Nydal
Co-supervisor:
Tor K. Kjeldby
June 13, 2012

Institute of Energy and Process Engineering
Abstract
The principles of the object oriented slug tracking schemes at EPT (Department of Energy and Process
Engineering, NTNU) have been developed and discussed in some detail. Simple bench-mark testing re-
vealed that the LASSI code suffers form a pipe inclination-dependant lack of mass conservation, the cause
of which is presently unidentified. Comparing simplified and non-simplified SLUGGIT simulations with
experimental data published by Taitel et al. [41] mostly indicate a reasonable correspondence, though the
precision is somewhat imprecise. In particular, obtaining stable riser flow (free of significant pressure oscil-
lations) at low liquid flow rates was not managed without excessive gas rates. This is possibly a consequence
of the method’s intrinsic slug flow approximation to dispersed regimes, but further investigation showed that
the methods stability response altered with recent code versions in which alterations to management proce-
dures was identified as the main differences, indicating that the SLUGGIT method’s riser stability properties
are quite sensitive to intuition-based section management routines. Further developing the models to better
accommodate vertical flow regimes is advised.
Supplementary testing was afforded through the development of a steady-state unit-cell type model for
phase fractions in the riser. Excellent accordance with simulation data was found, confirming that the SLUG-
GIT model is capable of reproducing stable, expanding bubble flow. Coarse resolution served to disturb this
process as bubbles becomes longer and are affected by riser entrance and exit effects. It was also found that
pressure oscillations caused by such entrance and exit effects display the typical characteristics of terrain
slugging and may be mistaken as such.
With basis in the P50 Girassol pipeline, a systematic investigation into operational instability phenom-
ena has been carried out using the available boundary conditions. Instabilities rooted in gas accumulation
in jumpers, possibly also influenced by the U-bend, were found when studying the fixed pressure open inlet
condition. The character of these instabilities were of a frequency and intermittency uncongenial to the field
data. Most instability phenomena captured in these simulations were sensitive to changes in geometry, inlet
condition and management parameters.
Severe slugging was initially found to dominate the flow picture with a fixed flow closed inlet condition.
Also this type of operational instability had too high a frequency to be a match with the field data. Nor does
the well-know ‘shark fin’ pressure profile of the terrain slugging liquid build-up and blow-out precesses
match the sinusoidal character of the field data.
It was recently found that the supplied field data needed adjustment for phase transition at the inlet state,
amounting to a considerable reduction in gas flow. This produced predictions of more stable production
compatible with those generated by external participant, though significant pressure fluctuations were still
observed. These fluctuations were found to originate from slugging in the U-bend and entrance effects as
large Taylor bubbles formed through coalescence and penetrated into the riser. This latter cause is believed
to be a feature of the limited, non-dispersed flow objects available in the EPT models.
A production index type boundary condition was implemented to better accommodate the well produc-
tion response. Even so, the conditions under which the Girassol field instability data was recorded could not
be recreated satisfactorily without gas lift and PVT support implemented. Simulations including the pro-
ductivity index inlet indicated, for the most part, that without the presence of a gas lift system, the pipeline
is likely to come to a complete stand-still; unless the liquid in riser and well are strongly aerated at all times,
the well head will not be sufficient to overcome the total system liquid column weight.
2
Sammendrag
Prinsippene bak de objektorienterte slugfølgemodellene ved EPT (institurtt for Energi og Prosessteknikk,
NTNU) blir utledet og diskutert på detaljnivå. Enkel testing viser at LASSI-koden lider av en helningsavheng-
ing og foreløpig uidentifisert feil som fører til at masse ikke blir bevart. Sammenlikning av SLUGGIT-data
med eksperimentell data fra Taitel et al. [41] indikerer stort sett en akseptabel overenstemmelse, dog noe
upresis. Særlig ved lav væskestrømningsrate viste det seg vanskelig å oppnå stabil riserstrømning uten over-
drevent gassinnskudd. Modellenes slugtilnærming til dispersjoner kan ha skyld i dette. Videre undersøkelse
viser at forskjellige, forholdvis nye versjoner av SLUGGIT-koden predikerer forskjellige stabilitetsdomener
for risersystemet. Forskjellen blant versjonene ligger hovedsakelig i prosedyrene tilknyttet objekthåndtering,
hvilket indikerer at modellenes riserstabilitet er følsom for denne intuisjonbaserte delen av simuleringsruti-
nen. Videre modellutvikling med hensyn på vertikalstrømning anbefales.
Videre testing ble gjennomført ved utleding av en stasjonær ‘unit-call’ riser-fasefraksjonsmodell. Ut-
merket overensstemmelse med data fra simuleringer ble observert, hvilket underbygger SLUGGITmodel-
lens evne til å gjenskape stabil, ekspanderende riserstrømning. Grov oppløsning forstyrret derimot denne
prosessen ettersom bobleseksjoner ble lengre, hvilket førte til trykkendringer forbundet med seksjonenes
med inn- og uttreden av riseren. Disse forstyrrelsene i trykk var av en karakter lik det sett ved terrengslug-
ging.
Med utgangspunkt i P50-rørledninged ved Girassol-feltet ble mulige former for produksjonsustabilitet
undersøkt ved bruk av alle tilgjengelige innløpsbetingelser. Ustabilitet med opphav i gassakumulering i
jumpere, samt en mulig U-rørinnvirkning, ble oppdaget under studier med konstant trykk på innløpet. Disse
formene for ustabilitet viste seg derimot å ha for høy frekvens og for skarpt utfall til å samsvare med feltdata.
De fleste av disse simuleringene viste seg å være følsom til endringer i rørkonfigurasjon og simuleringspa-
rametere.
Terrengslugging type I dominerte strømningbildet hvor konstante massekilder ble benyttet som innløps-
betingelse. Også denne typen ustabilitet viste seg å gi oscillasjoner med overdreven frekvens. Heller ikke
den typiske ‘haifinne’-trykkprofilen assosiert med væskeopphopning og utblåsning står i samsvar med den
sinusformede trykkprofilen registrert i feltdataen.
Nylig ble det oppdaget at oppgitt innløpsdata trenger korrigering for faseovergang til innløpstilstanden.
Dette endrer gassraten betydelig. Resulterende simuleringer var mer stabile og dermed mer lik de gjennom-
ført av eksterne deltakere. Allikevel var trykkprofilen preget av en anselig forstyrrelse. Denne forstyrrelsen
viste seg å ha opphav i terrengslugging i U-røret, så vel trykendringer forbundet med at lange Taylor bobler,
formet ved koalesens i horisontalrøret, penetrer inn i riseren. Sistnevnte kan være et resultat av mangel på
dispergerte strømningsobjekter i slugfølgemodellene.
For bedre å tilpasse produksjonsresponsen til brønnen ved innløpet ble produktivitetsindeks imple-
mentert som grensebetingelse. Omstendighetene tilknyttet Girassolfeltets ustabilitet kunne likevel ikke
bli tilfredsstillende gjenskapt, primært grunnet mangelen på gassløft og PVT-tilstandsberegning i kodene.
Simuleringer med produktivitetsindeks på innløpet indikerte hovedsakelig at systemet ville stanse fulls-
tendig opp uten gassløft tilstede; hvis ikke riser og brønn til enhver tid et godt utluftet vil ikke brønntrykket
være tilstrekkelig til å overvinne tyngden av den totale veskesøylen.
3
Contents
1 Introduction 11
1.1 Multiphase flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.1.1 Flow regimes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.1.2 Industrial importance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.1.3 Limitation of exact simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
1.1.4 Common methodology in simulation of multiphase pipe flow . . . . . . . . . . . . . . . 13
1.1.5 Commercial codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.2 Present work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.2.1 The Girassol oilfield . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.2.2 Suspected causes of case instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
1.2.3 Aims . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.2.4 Phenomenological slug tracking models at EPT . . . . . . . . . . . . . . . . . . . . . . . 16
2 Definitions and terminology 17
3 Operational instabilities in vertical risers 18

3.1 Extension driven instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Density wave instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.3 Terrain slugging instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4 Details of Methods 25
4.1 Method structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
4.2 SLUGGIT scheme versions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.3 SLUGGIT – Fundamental Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
4.3.1 Mass Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.3.2 The pressure equation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.3.3 Momentum Balance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
4.4 Implicitness and time regulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.5 Computational expense . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
4.6 Non-locally defined properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.7 Staggerd and non-staggered grids . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
4.8 Slug border velocities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
4.9 Turning point criteria . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
4.10 SLUGGIT v.2/v2s computational sequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
4.11 The LASSI scheme . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.11.1 Abbreviated method presentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.11.2 Domain of well-posedness and slug capturing . . . . . . . . . . . . . . . . . . . . . . . . 42
5 Benchmark tests 44
5.1 Liquid conservation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
5.2 Gas expansion in riser . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
5.3 Influence of gas lift on riser void fraction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
5.4 SLUGGIT and the riser flow map . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
5.5 Comment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
6 Compromises 58
4
7 Field simulation results 59
7.1 Fixed inlet pressure – jumper influence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
7.1.1 Jumper-caused pressure oscillations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
7.1.2 Expansion driven unstable production . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
7.2 Closed inlet, fixed flow rates – Terrain slugging . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
7.3 Full Girassol geometry with well . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72
7.3.1 SLUGGIT v.2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
7.3.2 SLUGGIT v.2s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
8 Discussion and recommendations 80

8.1 Girassol instability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 80
8.2 Method considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
8.3 Recommendations and suggestions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
8.4 Final comments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
9 Conclusion 86
Appendices 90
A A model of void in risers 90

A.1 Discrete calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
A.2 Analytical approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
A.3 Comment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
B Snapshots 95
C Productivity Index inlet 99
D Titbits of MATLAB R code 99
Annexes 104
E The Kelvin-Helmholtz stability criteria 104
F Simplified models of void in riser 105

F.1 Numerical calculation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
F.2 Analytical approximation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5
Nomenclature
Latin symbols
e Unit vector −
n Normal vector of control volume −
Ṁ s Mass rate source term kg/s
Q Volumetric flow rate m3 /s
A Pipe area, perpendicular to flow direction m2
c Speed of sound m/s
D Pipe diameter m
Dh Hydraulic diameter m
√
Fr Froude number = u/ gD −
g Gravitational acceleration m/s2
S ,S T S ,S T
GOR Gas-Oil Ratio = Ug,i /Ul,i −
H Riser height m
h Phase height m
IF M Interphase Friction Multiplier −
K Productivity index Sm3 /(bar · day)
k Wavenumber 1/m
L Pipeline length m
M Control volume mass kg
m Pipe specific mass (extensive) kg/m3
N Total number of samples −
p Pressure Pa
Pwell Well pressure bar
Q ‘Extensive volumetric flux’ – see (4.3) m3 /s
R Gas constant J/(kg · K)
6
r Pipe radius m
Re Reynolds number −
S Wetted perimeter m
sχ Sample standard deviation unit of χ
T Temperature K
u Phase velocity m/s
V Volume of section m3
v Border velocity m/s
VR dz Total phase volume in system, computed using (5.1a) m3
Vio Total phase volume in system, computed using (5.1b) m3
wc water cut = UHS ,S2 O,i

T S ,S T
/Ul,i −
y Vertical axis in pipe cross-section. Origin at pipe bottem m
z Position downstream the pipe m
MAX_BUB_COEF SLUGGIT simulation parameter; maximum allowed bubble length measured in diameter mul-
tiples. −
Greek Symbols
α Section volume fraction (hold-up) −
χ Representation of a property various
∆ Difference (macroscopic) −
δ Difference (differential approximation) −
` Control volume streamwise length m
κ Balance of Bernoulli suction and hydrostatic head gradient (see Eq. (4.28)) m2 /s2
λ Darcy friction coefficient −
λf Film thickness m
µ Dynamic viscosity kg/(m · s)
Ω Control volume domain m3
7
ω Pulsation 1/s
∂Ω Boundary of control volume m2
Φ Total average riser phase fraction −
φ Pipe inclination angle (deg)
ψ Pressure error m3 /s
ρ Density kg/m3
σ Surface tension N/m
τ Shear stress N/m2
Θ Total phase fraction in horizontal pipeline upstream riser −
Subscripts
κ Fluid type: κ ∈ {l, g}
Σ Control volume cell: Σ ∈ {W, E}
σ Control volume border: σ ∈ {w, e}
b ‘bubble’
crit ‘critical’ (criteria limit)
E Eastern (right) control volume cell
e Eastern (right) control volume border
f Liquid film (around bubble)
i ‘inlet’
int ‘interfacial’ (surface)
j Position index along the entire pipe
n ‘nose’
o ‘outlet’
P Present control volume cell
pipe pipe (interior surface)
s ‘slug’
8
t ‘tail’
W Western (left) control volume cell
w Western (left) control volume border
Superscripts
gl ‘gas lift’
n Present time step
n+1 Next time step
S Superficial property (e.g. uSκ = ακ uκ )
s Source
ST (At) standard conditions
Encasements
˙
(·) (dot) Rate of property
d(·)
dχ Total derivative
∂(·)
D(·)
Dz = ∂t + u ∂(·)
∂z Material/particle/substantial etc. derivative
∂(·)
∂χ = ∂χ (·) Partial derivative
h(·)i Temporal average
˜
(·) (tilde) Local property in space (not a cross-section average)
(·) (underline) Vector
(·) (double underline) Tensor
|(·)| Absolute value
Abbreviations
CFL Courant Friedrich Levy (criterion)
DNS Direct Numerical Simulation
EOS Equation Of State
9
EPT Department of Energy and Process Engineering, NTNU
FVM Finite Volume Method
GOR Gas-to-Oil Ratio
IKH Invicid Kelvin-Helmholtz stability criteria
LHS Left Hand Side
NTNU Norwegian University of Technology and Science
RHS Right Hand Side
SSB Slug Stability criterion using the Bendiksen bubble-nose velocity
VKH Viscous Kelvin-Helmholtz stability criteria
Glossary
Gas lift Production technique for reducing the static head and stabilizing the flow by means of pumping gas
into the pipeline.
Hold-up local (liquid) volume fraction. ‘Gas hold-up’ is also sometimes found in the literature.
Horizontal (noun) Part of the P50 pipeline geometry – a 3.5 km segment between well and riser
Manifold Pipeline intersection unit. In the Girassol P50 production line, one is found the the inlet of the long
horizontal.
Production Flow (of a specified phase) through the pipeline outlet.
Riser Long, vertical pipe segment, transporting fluid from seabed to surface.
Stability See Section 2.
10
1 Introduction
1.1 Multiphase flow portray schematically the corresponding regime flow

map. In vertical pipes, slugging and annular flow
1.1.1 Flow regimes will replace the stratified. ‘Chunk flow’ is a term
also sometimes used to characterise some vertical
Multiphase pipe flow is characterised by a wide va-
flow patterns. A combined term for the annular and
riety of flow patterns, or regimes, each characterised
stratified patterns is separated flow.
by different pressure, shear stress, velocity and phase
fraction properties as a result of the flow patterns
being topologically different. In near horizontal Slugs and rolling waves form a particular chal-
pipelines the flow patterns will be separated at low lenge as these have sharp fronts and constitutes near
phase velocities, and all gas and liquid will be sep- discontinuities in cross-section phase fraction. They
arated by a single, relatively flat interface. This is also form jumps in pressure and velocity. Mod-
called a stratified flow regime. Increasing the gas elling these slugs on Eulerian grids tend to demand
flow rate causes waves to appear on the liquid in- fine gridding to avoid numerical diffusion of the slug
terfacial surface due to interfacial friction. At high fronts. Slugs may also be treated in a statistical man-
enough relative velocities these small interface dis- ner using, for instance, unit-cell approaches. Alter-
turbances will grow due to the effects of reduced lo- natively, slug or wave tracking methodology may be
cal pressure at the wave location caused by height- adopted in order retain exact information of where
ened gas velocities due to a narrowing of gas side the slug front discontinuity is positioned, and condi-
cross-section area. Interfacial friction will also act tion the scheme thereafter. The methods considered
unevenly on the interface, increasing wave forma- in this work apply a Lagrangian slug tracking strat-
tion. If the weight in hydrostatic height potential is egy where dynamic grid manipulation if achieved by
not great enough to dampen the wave formation, then means of object oriented programming.
the stratified flow will be hydrodynamically unstable. In the case of waves, an interesting strategy is
This results in the liquid phase at locations spanning to tackling the Riemann problem in a way similar to
the entire cross section area (possibly with gas bub- the Rusanow method [30]. Dynamic grids in strat-
bles entrapped within). A new flow pattern, slugging ified regions may be utilised to promote a control
flow, is now established. This flow pattern is charac- volume which follows the wave front in space and
terised by intermitted flow of sequential liquid slugs time. This strategy is applied in LASSI [37] and will
followed elongated (Taylor) bubbles surrounded by a be discussed in later. Bernoulli suction becomes an
thin liquid film. important mechanism as the waves grow – difference
Should the gas rate be increased even further the in gas and wave velocities account for a relative flow
gas will begin to occupy most of the pipe centre, past waves and therefore across regions of chang-
pushing the liquid out towards the pipe walls. This ing cross-sectional area. The local alteration in gas
flow pattern is known as annular flow. The con- velocity this causes produce a pressure gradient nor-
verse situation, in which the amount of liquid far mal to the stratified interface, producing a suction
exceeds the amount of gas, a fully dispersed bubble force which may overpower the gravitational weight
flow regime appears. Here, small bubbles propa- and cause slug formation. Capturing of these phe-
gate through a continuous liquid phase. From here, nomenon directly is fully possible [13, 7], but, for
void waves of high bubble concentration may form the foreseeable future, too computationally costly for
causing bubble coalescence, returning the flow to large engineering purposes in which the diameter-to-
a slugging pattern. Similarly, rolling waves may pipe length ratio is large. Applying some subgrid
accumulate liquid and span the cross-section, man- model to determine a hydrodynamic slug initiation
ifesting regime transition from the stratified regime criteria is a natural alternative. Several models for
[10]. Figure 1.1 presents a schematic of the differ- slug initiation representing this phenomenon have
ent flow regime in horizontal pipes and Figure 1.2 been proposed during recent years [3, 25], some of
11
Figure 1.1: Flow patterns in horizontal flow. Source: [16] Figure 1.2: Flow map for horizontal pipes. β
is here the pipe inclination. Source: [25]
which will be presented and tested within this text. techniques [20, 38, 40, 14].
The mechanisms which cause the transition to

1.1.2 Industrial importance slugging flow are often divided into two main cate-
gories: hydrodynamic slugging, as described earlier,
Instabilities in multiphase pipe systems are natural and terrain slugging.
phenomena which during the recent decades have
posed challenges to the industry. These instabilities
produce highly transient flow behaviour with large,
undesired, fluctuations in pressure, velocity and flow 1.1.3 Limitation of exact simulation
pattern. For offshore oil and gas production, these
unsteady flow patterns are associated with loss of In single phase flows analytical solutions only exist
production and additional strain and material fatigue in a few, very simple, laminar flow situation. As
on equipment and production facility. Safety also soon at the Reynolds number increases beyond the
quickly becomes an issue, as it always does when- laminar regime the governing equations take on an
ever equipment is place under strains exceeding their infinite number of possible solutions, rendering true
design purpose. predictions impossible [35, 42]. Direct and mod-
In the Nuclear industry the need for accurate pre- elled simulation of the governing equations can re-
diction of fast pressure transients in thermal multi- produce statistical characteristics of both single and
phase cooling systems have been apparent ever since multiphase flow patterns, but often at great computa-
the Three Mile Island accident in 1979. The pres- tional expense. Direct numerical simulation (DNS)
suriser in pressurised cooling reactors is an impor- techniques such as immersed boundary methods in
tant component for controlling the the pressure in the the framework of a Navier Stokes solver [48] or
primary cooling system, and it provides a surge vol- multiphase lattice-Boltzmann codes are examples of
ume for coolant expansion and contraction. [9] methods [39] are examples of approaches aiming to
Multiphase flow stability is also an issue for re- reproduce the dynamic behaviour of the flow by cap-
newable energy systems, for example in thermal so- turing the physics in as an exact a manner as possible.
lar systems and heat and refrigeration pumps.. Though results from quite small, moderate Reynolds
Means with which unstable flow patterns can be number case studies are quite invaluable as an exper-
avoided is therefore immensely valuable to the all in- imental arena, their computational cost makes DNS
dustry operating with multiphase pipe flow systems. inadequate for any appropriate engineering purpose.
Common stabilisation methods in oil fields today in- There seem to be little hope of computers obtaining
clued e.g., back pressure control systems, dynamic the required capacity anywhere in the foreseeable
choking, and intermitted or continuous gas injection future, with the steady development in performance
12
seen to date.1 Indeed, although Moore’s law of long reversing the circumstances which caused transition
term trends in computer development may still be from one state to the other may not necessarily return
said to be valid, the limitation of single processors the flow back to its original state. There is also the
have long since been reached. For the near future problem of the pattern grouping itself; the process
then, DNS to have purely experimental, or laborato- of transition is also be associated with changes in
rial, use as it’s main application. flow characteristics. The introduction of transient
dynamic models around the 1980’s provided an im-
Therefore, in attempting purely mechanistic sim- provement to this condition. Criteria for determining
ulations to large problems, fine scale physics require flow pattern could now be based on local, dynamic
modelling and colsure relations in order to reduce the events rather than the flow state. Still, flow map
computational cost. The mechanistic ideal is still the strategies are still in use.
same; in capturing enough of the physical mecha-
nisms governing the dynamic behaviour of the flow, An alternative to mechanistic strategy is a phe-
a general-purpose method will be obtained capable nomenological modelling. Here, the nature mim-
of responding adequately to each scenario. Again, icking ideal is abandoned in favour of a response-
large problems force coarser grids, demanding more oriented approach. Rather than aiming at capturing
modelling, which in turn generally corrupt the gen- the underlying physics well enough to obtain the ap-
eral purpose ideal. propriate dynamic system behaviour, the phenomena
and dynamic response involved therein directly be-
come the subject of modelling. The slug tracking
1.1.4 Common methodology in simulation of scheme studied in this work are indeed based upon
multiphase pipe flow such a strategy, the object oriented framework being
It has always been an ideal to provide a single ‘grand ideal for the purpose of tailoring dynamic response.
unified’ model with which all flow situations can Admittedly, such procedures can be said to be inca-
be simulated. Particularly when dealing with multi- pable of handling scenarios involving phenomena not
phase flows there seem to be little hope of achieving explicitly included in the model framework. How-
such an ideal, except on the tiny DNS-realm. The ever, the same has already been said about many of
reason for this was seen in Section 1.1.1; creating the mechanistic models as they are usually heavily
a model which intrinsically incorporates all vastly dependant on their empirically determined closure
different flow pattern characteristics without captur- relations [35]. The main goal is usually not to re-
ing the fine-scale mechanics which are source to the trieve exact details, but a wider picture of the overall
differences is simply not done. Instead, the more dynamical system behaviour.
widely adopted strategy is to group flow patterns and
adopt different bespoke models for each flow regime. 1.1.5 Commercial codes
However, this begs the question which regime to sim-
ulate? A priori knowledge of the state of the flow Most commercially successful codes available to-
becomes required, though this information is often a day are in some degree based on phenomenological
part of the motivation for doing the simulation in the modelling, though some more than others. When
first place. For many years this has been a motivation considering multiphase industrial pipe flow, length-
for creating flow maps such as that in Figure 1.2. Un- to-diameter ratios will often be so large as to render
fortunately, these two-dimensional maps have proven any form of three or two dimensional simulation un-
to be of limited applicability, particularly in terms of feasible in terms of computational expense. Rather,
changing geometries. Also, many of the phenomena attention is given to one-dimensional models which
causing phase transitions are transient in nature, and treat properties which are statistically averaged over
1
By assuming that the smallest scales of turbulence are independent of the larger scales and only dependant on dissipation as vis-
cosity, while also assuming that the mean energy transfer through the scales is steady, it is easily shown that η/D ∼ Re−3/4 , η being the
smallest scale to resolve. In three dimensions, including the cost of time step refinement, one finds that the computational cost increases
proportional to Re−3
13
the pipe cross-section. In the mid 1970’s, headway dant empirism plays a large role in how the con-
was made in physical understanding of two-phase tent of each cell is treated in order to derive phase-
flows, leading to development of numerous mecha- specific properties (slip, bubble nose velocity, inter-
nistic models being developed. Most of these were facial mass transfer, friction factors, etc.) from the
initially steady-state models, seeking to provide sta- mixture properties and close the model. These mix-
tistical information on flow rates, phase distribution, ture values for the slower transient, large scale dy-
pressure and temperature. In contrast to empirical namics are found by solving mechanistic consecra-
models, which on their own tend to have a validity tion equations in the system of linked cells. Based on
limited to the pipelines and states at which they are the two-fluid principle, OLGA solves three continu-
calibrated, mechanistic models show far greater po- ity equations (gas, liquid bulk and droplets) together
tential for being applicable in a wide range of flow with two momentum equations (liquid film and gas-
situations. droplet mixture). [5, 13]
Among the most popular early steady-state codes

available are PIPESIM from Baker Jardine, PROS- The TACTILE model, which is based on a drift-
PER GAP from Petroleum Express, PIPEPHASE flux formulation, differentiates between two types of
from SimSci, PEPTITE and WELLSIM Total, and flow patterns: separated and dispersed. Intermitted
TUPPF developed within the Tunisia university fluid flow is treated as a treated as a combination of the
flow projects [11, 35]. In recent years these steady- two using a fraction of separation variable to com-
state models are being replaced by more advanced bine them. This variable is then also included in the
mechanistic and phenomenological transient mod- flow regime transition criteria.
els less dependant upon traditional empirical corre-
lations. These models usually first identify the flow A survey preformed in [11] judged the transient
regime and then solves regime-dependant mass and models OLGA and TACTILE to be superior to the
momentum equations supplied with empirical clo- steady-state models as these were found to be appli-
sure laws. The energy equation may also be included. cable to a wide range of flow situations. This survey
Steady-state models are not able to predict the also found TACTILA to perform better than OLGA.
transition form one flow pattern to another, nor nec-
essarily all phenomena occurring in a pipe [13]. The
nuclear industry was amongst the first to create dy-
namic multiphase simulation codes, such as TRAC, TRIOMPF is another code which also ought be
RELAP_5 and CATHARE [5]. These codes were mentioned. This code is a directly finite volume
mainly motivated by safety protocols in reactor cool- discretisation of the two-fluid model, which consists
ing systems, and focused on capturing the quick pres- of a pair of one-dimensional momentum and energy
sure transients, i.e., smaller systems during shorter equations, one for each phase, applicable to strati-
time intervals. The petroleum industry quickly fol- fied flow. TRIOMPF solves the differential system
lowed, focusing on the slow transients associated in a ‘brute force’ manner, applying finite volume
with mass transfer. upwind discretisation, very fine griddling and itera-
OLGA and TACTILE are perhaps the most com- tion at each time step to solve non-linearities with
mercially successful of these codes. The former has precision. It was used by Issa and Kempf [13] to
been developed in a joint research programme be- demonstrate that is it possible to automatically cap-
tween the Institute for Energy and Technology (IFE) ture hydrodynamic slug formation since the mech-
and SINTEF which applies a unit-cell model. In such anism which causes this phenomenon is present in
models, empirical criteria are used to determine the the two-fluid model (see Section 4.11.) This strat-
flow regime. The dynamics inside a ‘cell’, at the egy (applying enough computational effort to capture
scale of the pipeline diameter, is treated statistically the mechanisms causing flow phenomena, rather than
in a steady-state manner as the control volume frame modelling them) bears resemblance, but should not
of reference follows the flow. Flow pattern depen- be confused with, DNS simulation techniques.
14
1.2 Present work
1.2.1 The Girassol oilfield
Girassol is a deep-water oilfield located 150 km

Northwest of the coast of Luanda in Algeria. The
seabed depth of the Girassol field is situated approx-
imately 1 300 meters, while the reservoir is relatively Figure 1.3: P50 Girassol pipeline stability field tests.
shallow (1 200 meters) [47]. Though the seabed Source: [43]
pipeline is located at high pressures, the relatively
shallow well depth means that the well pressure it-
Another challenge typical of deep water oil fields
self is not sufficient to drive a steady production.
is low temperatures. Although the fields are situ-
Artificial gas injection, known as gas lift, has there-
ated at deep waters, the wells themselves are shallow,
fore been applied to the riser base. Gas lifts are
making the fluid cool. The temperature measured at
commonplace in such fields an their purpose is to re-
the manifold of the P50 pipeline is about 64 ◦ C on av-
duce the hydrostatic riser head by increasing the total
erage, but lower well temperatures are also found in
riser void fraction. Gas injection proves stabilising
this field. Even though insulated against the 4 ◦ C sur-
to production and is often preferable to increasing
roundings, there is a distinct temperature reduction
upstream pressure by means of mixture pumping
associated with transporting the fluid up the riser, as
stations. Some forms of operational instabilities,
internal energy in the fluid is abstracted into potential
commonly called ‘headings’, are associated with the
energy. A lift of 1 300 meters entails a temperature
riser itself (see Section 3.)
reduction of about 5 ◦ C [35]. As wax deposition be-
comes problematic around 40 ◦ C, retaining the liquid
enthalpy quickly becomes an issue. Expensive and
From May 12th to 15th 2004 single-well riser energy consuming electric pipeline hearing may then
stability tests were carried out on the P50 pipeline, become a necessary measure.
while multi-well tests were performed on the P10
pipeline from May 29th to June 3rd, 2004. Gas 1.2.2 Suspected causes of case instability
lift injection rates were stepwise reduced from
190 kSm3/d to 70 kSm3/d through the stages Qgl g ∈ The P50 pipeline offers many possibilities for the
{190, 150, 100, 70} kSm3/d, producing for several hours cause of operational instability seen in Figure 1.3.
at each stage. As the gas lift was reduced to 100 kSm3/d, At the seabed, the pipeline stretches horizontally for
the oscillations form the system disturbance sur- about 4.5 km. It is possible that hydrodynamically
passed 5 bars in amplitude and the oscillation damp- initiated slugging in this region will affect the over-
ing was slow. When reduced to 70 kSm3/d gas lift the all system dynamic. Also, this horizontal has a light
riser became unstable, showing increasing fluctuat- downward inclination (φ ≈ 0.8 deg on average,) mak-
ing amplitudes reaching ∆p = 30 bar before the gas ing it susceptible to terrain slugging (see Section 3.3)
lift was again increased to prevent production system ‘Jumpers’ (small upwards-pointing bends to al-
damage. Pressure, temperature and flow rate data low for slight thermal expansion/contraction of the
were collected form manifolds and riser valves. Fig- pipeline) are positioned along the horizontal part of
ure 1.3 show the pressure and gas lift rate history at the pipeline. Eldar Khabibullin [18] showed that gas
the well manifold. The data was supplied by TOTAL accumulation is these jumpers may cause periodic
[43]. gas releases in the form of void waves, promoting
15
system oscillation. at Kongsberg Oil & Gas Technologies and TOTAL,
The total length of the pipeline, form well bore we seek to understand the field instability better and
to separator, is approximately 7.7 km long. Because highlight differences and similarities in the predic-
the production from the well will be dependant upon tions form the LedaFlow and SLUGGIT codes.
local pressure gradient, and this in turn will be de-
pendant upon the liquid fraction in the riser at the 1.2.4 Phenomenological slug tracking models at
other end of the pipeline, there exist the potential for EPT
a void, density or surge wave type instability [38, 26].
Such types of instabilities are seen as promising can- As a counterpart to the TRIOMPF code, which
didates for the operational instability seen at the P50 sought to reproduce dynamic phenomena by cap-
Girassol pipeline as the long spatial stretches covered turing the mechanisms through sufficiently detailed
by such well-to-riser wave communication would in- mechanistic modelling, the initial SLUGGIT code
volve long periods of oscillation. Oscillation period [34] sought instead to model the phenomena in them-
of the data in Figure 1.3 is three hours. selves. Such phenomenological strategies constitutes
Instability driven by gas expansion in the riser a philosophy different form capturing schemes, fo-
is also likely due to the large change in hydrostatic cusing on response rather than recreation.
pressure across the riser. A gross steady state calcula- The first SLUGGIT code, and the code descend-
tion assuming ideal gas, no phase transfer, and a riser ing from this, utilises object oriented programming
completely filled with liquid reveals that a bubble re- to construct a framework within which dynamic be-
leased at the inlet of a filled riser will expand to about haviour can be hand tailored in compliance with
3.5 times its original volume. In fact, gas expansion experimental observations. Such a strategy pro-
in the riser is fount to play an important role in most vides direct control of model behaviour and arena
of the operational instability mechanisms discussed for extending behavioural capability limited only by
in this work. the detail of information at which the model oper-
This topic will be further discussed in Section 3 ates. Simplicity is a central ideal. In abandoning the
prospect of phenomena capturing, there is a unique
opportunity to push the limits of simplification and
1.2.3 Aims investigate which factors are dominating the macro
The foremost aim of this thesis work is to acquire scale system dynamics. Pushing the limitations of
insight an knowledge about the slug tracking meth- simplicity also implies pushing the boundaries of
ods available at EPT, as well as becoming familiar application as cases of larger scales become man-
with the dynamics of multiphase pipe flow and insta- ageable at less computational expense. The great
bility mechanisms active therein. In achieving these challenge lies of course in determining where the
goals a case study will be undertaken to set a con- limit at which simplification seems to be justifiable
crete physical setting for the investigation. The Gi- and what boundaries may be pushed without damag-
rassol field described in Section 1.2.1 is chosen for ing the main system dynamic.
this purpose Consequently, the methods abilities to
simulate riser flow will take centre stage in this eval- Four codes of varying complexity are available at
uation. The P50 pipeline is chosen. This line become EPT. These are
production unstable as the gas lift injection rates are
• The code implemented by Pascal Klebert [23,
reduced below 70 kSm3/d. Attempts at reproducing
24], in this work referred to as ‘SLUGGIT v.1’.
this instability in simulation using OLGA and SIN-
TEF’s LedaFlow have not proven successful; both • The code implemented by Jørn Kjølaas [21,
these codes predict stable flow [47, 18]. Achieving 22], here referred to as ‘SLUGGIT v.2’.
instability, and possibly gaining an understanding of
which type(s) of instability mechanisms are active in • A simplified version of Jørn’s scheme imple-
this pipeline, is therefore an additional exciting chal- mented in the same code by Tor Kjelby, re-
lenge. Through collaboration with Eldar Khabibullin ferred to as ‘SLUGGIT v.2s’.
16
• A Lagrangian Approximate Scheme for Slug These codes are described in some detail in Sec-
Initiation (‘LASSI’) code [36, 37] by Fabien tion 4.
Renault.
2 Definitions and terminology

Unfortunately, in the literature researched during this thesis work a large portion of the terminology encountered
is found to be rather ambiguous depending on the author and topic. Therefore, it is fount helpful and clearly
define a few of the most commonly confused terms and concepts.
Flow patterns
Some ambiguity in the characteristic features of individual flow patterns are also present in the literature, in
particular with the ‘slug’, ‘plug’ and ‘chunk’ patterns. ‘Plugs’ here refer to solid blockage, as by a pig or
hydrate formation. the terminology applied in Figure 1.1 will be adopted here. Due to the limitations of the
slug tracking methods, smooth stratified flow and slug flow (mostly without gas entrainment) will be the most
widely discussed patterns.
‘Stability’
Model stability To be understood in terms of the stability of a model composed of discrete or differential
equations (such as Lyapunov stability.) In quasi-technical terms, a small disturbance to a steady system
should have finite influence on the solution as time progresses towards infinity. Model stability is usually
analysed by finding the criteria for which there exist a diverging Fourier mode.
(Physical) flow stability The conditions for which a specified flow pattern can physically be sustained. Ex-
amples of instability mechanisms in horizontal and vertical pipes may be Kelvin-Helmholtz instability
(see Annex E) and Rayleigh-Taylor instability [8], respectively. These are characteristics of the fluid
dynamics – even though models such as the two-fluid model are used to analyse such phenomena, they
should not be confused with model stability. If, however, teh model captures the physics perfectly, they
terms are equivalent.
Numerical stability The ability of a system of discrete equations to converge towards (close to) a finite solu-
tion not dependant on the discretisations itself (asymptotic convergence). A method spoken of as robust
is a method which remains numerically stable for a large verity of parameter combinations.
Operational stability Sable flow in this context is flow which retains steady properties in time and space.
Operational instabilities are typically termed ‘headings’ in the petroleum industry. This kind of stability
will be the main focus of the present work. As mentioned in Section 1.1.2, operational stability is a hot
topic in industry. Unfortunately, the author has not yet come across any clear definition of a criterion
for when a flow situation may be deemed unstable. Most industrial pipe flows processes operate in a
turbulent regime, which is intrinsically unstable and chaotic on a microscopic level. Industrial processes
usually include dynamic and transient features where small call oscillations or noise is present. Both the
frequency and amplitude of disturbance may be included in a possible definition and could, e.g., represent
fluctuations of any flow or fluid property in the same order of magnitude as those bulk.
Though throwing the terms ‘stable’ and ‘unstable’ about without definition is not unusual, Zakarian and
Larrey applies the usefully, yet arbitrary, ‘5-bar fluctuation’ stability criterion on the pressure upstream
17
the riser in [47] for the Girassol oilfield. In lack good of alternatives, a similar convention is applied here.
However, because risers of different dimensions will be analysed, a slightly more general formulation is
decided upon, namely that the fluctuation peak amplitudes be less than 5% of the unaerated static riser
head. This is about equivalent in the case of the Girassol field.
In the literature, terminological ambiguity concerning this type stability is a nuisance, and no commonly
agreed upon term has been found by the author. The term operational stability has been coined herein.
‘Riser stability’ Whether a small, disturbing gas penetration into the riser will cause a blow-out event – see
Section 3.3
well-posedness The property of a model to have a unique solution continuously dependent upon the initial
conditions.
Steady-state
By steady-state it is in this text meant a state in which all flow variables are constant in time.
Types of operational instability
Many types of operational instability are possible curs at some point, e.g., Cycle C and Cycle D
within the flow regimes of even a simple pipeline- [33, 35, 20, 40, 41]. Yet another take on the term
riser system. This will become apparent in Sec- is a cycle in which the vigorous blow-out event takes
tion 3, where a few of the most important types of place, including cycles like Cycle E (but not Cycle D)
operational instability is discussed. As far as the this [17]. According to Jansen et al., the term severe
author can understand, some ambiguity in the termi- is by definition reserved for terrain slugging events
nology used to describe these instability phenomena in which full blockage occur and the slug length is
is present in the literature, validating the need to for longer or equal the riser height [14].
a definitions. The descriptions given in Sections 3.2
Presently, to avoid contusion, the term ‘severe
and 3.1 should suffice in defining the terms ‘void
slugging’ will be reserved for this latter extreme
wave-’ and ‘expansion driven instability’, respec-
event. ‘Terrain slugging type I’ will be used for
tively. When it comes to terrain slugging choices
all terrain slugging cycles in which the low-point
will have to be made.
becomes completely blocked (Cycle C and Cycle D,
enclosed by the Bøe criterion envelope in Figure 3.3),
In this work the term terrain slugging it used for
while ‘Terrain slugging type II’ is used for cycles
all terrain induced slugging, i.e. changes in pipeline
where gas continuously enters the riser (Cycle E,
elevation causing slug initiation. Slugging due to a
marked ‘unstable oscillations’ in Figure 3.3), as in
pipeline ‘obstacle’, such as a jumper, also falls into
[20, 33]. Further specification (with fall-back, blow-
this category.
out event) is given when necessary.
The term ‘severe slugging’ is by many authors
used synonymous with the term ‘terrain slugging’. Terms like ‘quasi-steady’ and similar, arbitrarily
Other authors use it to characterise all terrain slug- scattered around in many pieces of literature, are a
ging events in which full blockage of the bend oc- nuisance and will be avoided within this text.
3 Operational instabilities in vertical risers
18
Numerous instability mechanisms for operational instability have been doc-
umented from industrial, experimental and theoretical investigations. In the
petroleum industry, many such instabilities are found in connection with gas-
lift systems. Quickly summarise, instability types usually associated with gas-
lift systems are:
• Casing heading
• Formation heading
• Tuning heading
• Pipeline heading
Casing head may occur in gas-lift systems where the gas flow through
into the tubing is subsonic, allowing pressure information from the tubing to
propagate into the annulus between tub and casing. The flow in the tubing may
then be subject to compression of the gas in the annulus. This phenomenon is
similar to oscillations caused by the compressibility of trapped pockets of gas,
which will be visited later.
Formation heading is similar to casing heady, except that the accumulation
takes place in the formation.
Tubing heading is a density wave phenomenon which may occur also when
the gas injection is supersonic. Figure 3.1: Gas-lift well.
Pipeline heading is associated with the upstream pipeline and terrain slug- Source: [38]
ging.
Rather than using this terminology, the more general instability mechanism terms will be adopted, namely
• Expansion driven instability
• Density wave (void wave) instability
• Buoyancy driven instability
• Terrain slugging instability
3.1 Extension driven instability

Expansion driven instability is a term not commonly found in literature. In this work, the term is used to
describe the cyclic entrapment and expulsion of gas in ‘pocket volumes’ upstream the riser in a production
pipeline.
Cycle A. Expansion driven operational instability
A.1 After a gas pocket – in this case a pipeline jumper – has been filled with gas, an increased gas amount
will pass by the pocket and propagating into the riser in the form of a void wave.
A.2 The gas surplus travels up the riser and expands, reducing the riser static head and accelerating the liquid
ahead of it.
A.3 Expansion of the gas trapped in the upstream pocket and/or the acceleration of the system fluid contribute
to further release more of the trapped gas into the main flow and riser base. Consequently, the gas pocket
is drained of gas.
19
A.4 The expended gas surplus exits the riser, the system slows down. The total void fraction in the riser is
reduced as gas form the well is now entrapped in the now flooded gas pocket. As the pocket fills, the
cycle repeats itself.
3.2 Density wave instability

The mechanism of density wave instability in riser systems is, like expansion driven instability, is related to
changes in upstream static head afforded by a variation of the total liquid present in the riser. This phenomenon
is categorised as a dynamic form of instability in that it has a negative-feedback (self regulating) nature where
the driving force of cyclic oscillatory behaviour is a delay in the liquid production response to alterations in
upstream pressure. This delay is associated with the limited propagation velocity of a density changes through
the pipeline. Propagation, or drift, of changes in mixture density is known as a density, consentration or void
wave. The steps of a wave cycle may be described as follows:
Cycle B. Density wave operational instability
B.1 An alteration phase fraction enters the riser base and propagates up the riser. As it propagates upwards
and an altered amount of gas expands, the total void in the riser changes.
B.2 As the total amount of liquid in the riser alters, so does the static pressure upstream the riser. A density
waves of increased liquid hold-up (concentration waves) would cause increased pressure while the oppo-
site is true for waves of decreased hold-up (void waves). This is the riser-to-well communication channel
– pressure changes, and this information propagates at the mixture speed of sound.2
B.3 The well responds to a drop in static head by increasing the well flow rate, and oppositely of pressure
increases (self-regulation or negative feedback). This well responds propagate down the system at the
concentration wave velocity.
B.4 The density wave reaches the riser and the total liquid amount of liquid in the riser changes. Because of
the delay in time from the well responds to pressure changes till the density wave reaches the riser, the
well will over-compensate. This over-compensateion in turn reverses the pressure situation and, through
delay, causes the well to over-compensate in the opposite manner. The cycle continues.
In pipeline a gas lift system is applied a smaller fraction of the total riser void will originate from the well.
Consequently, the expansion of a gas surplus originating form the well will have a smaller impact on the total
riser void fraction Φg , as seen in Section 5.3. Gas lifts therefore has a stabilising effect on the flow, as well as
helping production by compensating for low well pressure.
It is further interesting to note that decreasing the injection rate in the gas lift will produce a ‘jolt’ in riser
liquid fraction which would upset the production-pressure balance and initiate this cycle. Also, fraction of
riser gas originating from the gas lift will directly influence whether this cycle will converge towards stable
production or diverge towards operation instability.
Finally, the delayed response feature of this cycle makes places it amongst the few operation instability
phenomena discussed here which would exhibit a sinusoidal pressure disturbance similar to that seen in the
field data Figure 1.3.
Sinègre and Petit [38] produced a simple model of the density wave. Laplace transformations were
for the pressure change due to changes in produc- applied and an analysis of the underlying character-
tion. The well was treated as a boundary condition istic equation performed using control engineering
with linear delay accounting for the transport time theory. Is was fount that the critical parameter was
2
Though significantly less than the single-phase speeds of sound [16], it is reasonably quick.
20
the gas injection rate and a simple control strategy of more liquid and grow in length. A requirement for
linear feedback to the gas lift from the well pressure, this is that the jump in pressure across the slug is not
a P-regulator, was shown by OLGA simulation to sufficient to move the slug from the low-point up the
stabilize the the system. pipeline. Should the pressure difference across the
slug be great enough to move it as soon as the slug
Apazidis [2] attacks the same problem. Rather is formed, then this terrain slugging will initiate the
than using a delay model for the reaction to the well transition to slugging flow, but not necessarily oper-
response, a more complete model based on mass and ational instability, merely a terrain induced slug flow
momentum balances of each phase was developed. pattern.
A particular air lift pump is considered, and an em-
pirical relation is used for bubble rise velocity. A Should however the opposite be true then gas
linear stability analysis is performed on this model will be prevented form flowing passed the low-point.
using Fourier transformation. Instability for interme- Gas may then accumulate upstream the low-point
diate injection rates were observed which were put blockage, gradually increasing the pressure differ-
down to the density wave mechanism. The size of ence across the slug. Simultaneously, new liquid will
the bubbles form the injection valve was shown to be be fed to the slug, lengthening it and increasing the
significant for the stability domain. required pressure difference needed in order to move
it. One of two things may happen: Either the pressure
The characteristics of void wave propagation resistance provided by the slug and the rest of the
have been studied further in detail by e.g., Lahey pipeline becomes higher than what can be supplied
[26], who performed both linear and nonlinear anal- from the inlet, and the system comes to a stand-still.
ysis of the void-wave phenomena on the two-fluid Alternatively, the accumulated pressure will at some
model (4.26). Different types of void wave profiles point be sufficient to move the entire slug passed the
are found and the conditions necessary for sustain- low-point. This may either be because the pressure
ing them. Smooth, time-invariant void wave profiles accumulation happens faster than slug pressure resis-
(solitons) were shown to be possible only with rela- tance grows due to new liquid supply, or because the
tively large void fractions. It was also found that void upstream end of the slug has reached piping with re-
wave propagation data can be a powerful tool for as- duced inclination or the outlet (‘proper’ severe slug-
sessing interracial momentum transfer laws applied ging). The result is a rapid acceleration of the slug
in the two-fluid model. down the pipeline followed by a high-pressure gas
front. In expelling the slug the upstream gas vol-
ume is abruptly increased causing a sharp, sudden
3.3 Terrain slugging instability
pressure decrease. It is this abrupt, large-scale vari-
Terrain slugging is caused by liquid accumulating at ation in velocity and pressure which characterizes of
local low-points in the pipeline. If the accumulation terrain slugging as unstable in a operational stability
become sufficient for liquid to block the entire pipe sense. After expulsion, a new liquid accumulation
cross section, a slug if formed. As long as this slug may take place at the low-point and the cycle is re-
occupies the low-point position it will accumulate peated. In short:
Cycle C. Terrain slugging operational instability
C.1 Cross section blockage at low-point dip.
C.2 Compression of gas upstream the blockage, increasing the upstream pressure.
C.3 Eventual slug expulsion (blow-out), releasing the gas and plummeting the pressure.
C.4 Possible liquid fall back. New liquid accumulation at low-point and eventual blockage.
A visual example of this cycle can be seen in Figure B.3 in Appendix B.
21
The ‘severity’ of the terrain slugging (i.e., the Bøe criterion has the form [14]
length of the liquid slug at the time of blow-out) is
ρg,o RT S
strongly dependent upon the upstream pipeline vol- Type I slugging ⇒ Ul,iS ≥ U (3.1)
ρl gαl H g,o
ume. During slug build-up there is a race between
the increasing slug column height (liquid supply) and where the notation is consistent with that used
the pipeline pressure increase (gas supply). A larger throughout this work – see the nomenclature.
upstream gas volume means slower pressure build-
up and increased severity/reduced frequency of the The solid, horizontal line mark the divide for
slugging. The ‘severity’ of the terrain slugging can riser stability in the case of a riser completely filled
therefore be said to be proportional to the ratio of with liquid. Again a new take on the stability term;
upstream compressible volume to gas injection rate riser stability is the limit at which, when a small gas
(ALΘg / Ṁg ). Notice that this ratio takes the units amount protrude into the riser, the static head reduc-
ALΘg
"
m3
# "
s
#
T tion due to increased riser void is balanced by the re-
= kg/s
= kg/m3
∼ duction in upstream pipeline pressure due to the ex-
Ṁg %
pansion in gas volume. During steady operation the
i.e., slugging frequency versus upstream gas density riser will contain a constant, total riser liquid fraction
(or multiple of specific volume). Φl . A steady state riser stability model is presented
Figure 3.2 show a typical terrain slugging ge- in the broken line marked ‘steady state stability cri-
ometry. The low-point bend (downward pipeline terion’. This is given by [40]
inclination) is a requirement. Figure 3.3 present a
h i
flow map of stability regimes, including the riser (riser) stable ⇔ Po > Φl Θg /α0g L − H ρl g (3.2)
stability criteria developed by Taitel et al. [40, 41].
Some elaboration of the cyclic modes of operation Θg its the spatially average void fraction in the
instability represented within this flow map is worth pipeline affront the riser. α0g is the void fraction of
including – both because it is informative in light the bubble nose entering the riser, being of the order
of the dynamics active in riser systems without gas 0.9 [41].
lifts, but also because similar cyclic instability pat- In [40] Taitel assumed that flow in the stability region
terns will be encountered later on. of this criterion (above the line) will produce steady
production. A correction to this is presented in [41]
The Bør criterion (marked ‘Boe criterion’) is here and is show in Figure 3.3 as the region marked ‘cyclic
included as a heavy, solid line. This well-known cri- no fall-back’. Here, ‘fall-back’ indicates whether
terion gives a necessary condition for gas blockages the top, or surface interface, of the liquid column
to occur, and it amounts to a simple balance mark- is re-established at the riser outlet or falls back to
ing the limit for when the read ends of the slug can- a level below the outlet after the penetrated gas has
not penetrate back into the pipeline due to the high left the riser. Although blow-out does not take place
relative gas velocity (monotonically increasing part in this region due to riser stability, oscillatory be-
of curve). At sufficiently high liquid flow rates the haviour happens which may result in cyclic slugging
pipeline void fraction approaches zero and the crite- similar to the terrain slugging described in Cycle C.
rion bends off to a horizontal line. The original Bøe The mechanism causing these oscillations are, and
assumption – that inside this criterion envelope slug- the flow pattern itself, is very similar to that of the
ging flow will prevail – is today acknowledged as in- expansion drive instability, Cycle A, and consists of
accurate (see [32] for an alternative criterion.). The the following steps:
Cycle D. Cyclic terrain slugging with blockage but no blow-out
D.1 Cross section blockage at low-point bend.
D.2 Increasing gas pressure forces gas to penetrate into the riser.
22
H
Figure 3.2: Geometry subject to terrain slugging. Moment Figure 3.3: Terrain slugging flow map example.
of blow-out event or transition to steady of oscillatory cycle. Source: [41]
Source: [40]
D.3 The riser is stable in the sense that the liquid flow rate is too great for blow-out to occur. Bubbly flow
initiates at the riser base.
D.4 As new gas fills the riser and expands, the riser void in increased and upstream static head reduced.
D.5 Consequently, the pipeline gas expands, increasing the rate at which gas penetrates into the riser.
D.6 At some point the riser may become sufficiently aerated so that the reduction in static head is not sufficient
to make up for the reduction in upstream gas density. This causes the flow rate into the riser to decrease.
D.7 If the gas flow rate at any point becomes negative doe to this deceleration, the rear liquid front will again
penetrate into the pipeline and cause gas blockage. Gas remaining in the riser exits through the outlet,
reducing the void fraction and increasing static head. At the same time pressure builds up as gas density
increases. At some point (possibly before the riser is gas-free) new gas penetration occurs and the cycle
is repeated.
If, however, the gas flow rate in Step D.6 does not become negative, then Step D.7 will not take place.
Rather, the (expansion driven) oscillations will be sufficiently dampened to bring the flow to a steady state.
A final region of the flow map in Figure 3.3 worth some attention is that marked ‘unstable oscillations’.
According to the Bøe criterion, this a region of steady flow. However, it is also in the region of ‘steady state
instability’, and the liquid velocity is much too low to prevent liquid fall-back after blow-out. The net result
is a cyclic slugging process taking place even though the flow is inherently steady state (a liquid front will not
penetrate back into the riser.) The cycle takes events are the following:
Cycle E. Cyclic terrain slugging with blow-out but no blockage
E.1 Temporarely stable bubbly flow passes through the liquid column. Because the height of the liquid
column is less than the riser outlet, no liquid is produced.
E.2 As new liquid enter the riser base, and as gas expands when the bubbles propagate towards the liquid
column top, the liquid column height is increasesd.
23
E.3 At some point, the liquid column reaches the riser outlet. Because of the inherit lack of stability at
this gas flow rate, a blow-out of the fully aerated liquid will occur rather than obtaining a steady liquid
production.
E.4 Fallback takes place and the system returns to its initial state. The cycle repeats itself.
This region of the flow map is a transition region between steady flow and terrain slugging.
Figure 3.4: Stability map example, func-

tion of GOR and riser height. Source: [35]
Taken from [35], Figure 3.4 shows a stability work. It is evident form the figure that increasing
map similar to that in Figure 3.3, here as a function the GOR has a stabilising effect on the flow. Increas-
of GOR and liquid flow rate, plotting the stability ing the riser height can however be both stabilising
limit for different riser heights. To the left of the sta- and destabilising. The two competing effects in-
bility lies, cyclic terrain slugging dominates. In the fluenced by increasing riser height are the increased
region enclosed to the left of the stability lines, but to gravitational riser pressure drop, promoting blockage
the right of the broken line, oscillation without blow- and destabilisation, and increased pipeline operating
out (Cycle D) dominates. In [35] and Figure 3.4, pressure, stabilising the system.
the terminology ‘severe slugging’ and ‘non-severe
slugging’ was applied, which is not adopted in this
e
tim
tim
e
(a) With blow-out and fall-back (b) Without blow-out and fall-back
Figure 3.5: Phase portrait examples, terrain slugging. Source: [35]
24
Another informative visualisation technique is which the system is at a state to the right of the bro-
given by the phase portraits, as shown if Figures 3.5. ken line in Figure 3.4 (above the horizontal line of
Here, the normalised mass fluxes leaving the riser Figure 3.3.) Here, no fall-back takes place and there
are plotted against the normalised inlet pressure. The is a continuous, though oscillating, liquid production
state is marked at each time step of the slugging (Cycle D), giving the portrait a circular shape.
simulation. From the initial, unstable condition, the
operation state spirals outwards counter-clockwise Jensen et al. [14] takes the analysis of Taitel
until a fixed slugging cycle is established. In Fig- [40, 41] further by also including gas lift into the
ure 3.5a slugging takes place at a state to the left of analysis, including the chock from [40]. It is shown
the broken line in Figure 3.4 and the system diverges that chocking the riser outlet can bring an unstable,
to a cyclic state in which the liquid production is in- slugging system to stable, steady production. The
termitted, giving the phase portrait the ‘flat bottom’ gas lift was also shown to have a stabilising effect by
characteristic of classical severe slugging (Cycle C.) increasing velocities and reducing liquid hold-up.
Figure 3.5b show the phase portrait for a situation in
4 Details of Methods
In this section the common framework of EPT models is established form finite volume analysis, together with
the the empirical closure relations used in all method versions. The main development of the fundamental
equations will focus primarily on the newest SLUGGIT code, though clarifying important and unique features
of each individual method will also be attempted.
The disinterested reader may skip Section 4.3-4.10 without severe loss of comprehension.
During the derivation of this section the following assumptions will be made use of:
Assumptions A. Slug tracking model assumptions
A.1 All key features of the multiphase dynamic can be captured in the stratified or slugging flow patterns.
Bubbles can be treated as closed regions of stratified flow.
A.2 No entrainment of one phase into the spatial domain of the other takes place – each phase κ in a con-
trol volume domain ∂Ω is fully contained within a respective, closed sub-domain ∂Ωκ . In other words,
dispersed flow topologies are not considered.
A.3 The momentum of all pipe-normal movement is neglectable compared with that in the streamwise direc-
tion.
A.4 Liquid phases are assumed incompressible.
A.5 Pipe wall normal density gradients within a phase sub-domain are neglectable.
A.6 Momentum exchange through the fluid interface can be handled analogous to wall friction.
Additional assumptions applied in this work – but which are not crucial to the methods themselves – are:
Assumptions B. SLUGGIT, additional assumptions
B.1 Two-phase flow is assumed.3
3
Although SLUGGIT v.2/v.2s supports three-phase flow.
25
4
B.2 There are no exchange of mass between phases (no evaporation or condensation.)
4
B.3 The ideal gas law is valid.
B.4 Surface phenomena such as rolling waves, surface tension, etc. are not included here.5
B.5 The flow is assumed isothermal – the energy equation will not be touched in the present work. See e.g.
[22] for handling of thermal issues in SLUGGIT.
Similar for all models discussed is that only two types of flow objects are considered: separated and (possi-
ble aerated) slugs. Elongated Taylor bubbles are treated as enclosed regions of stratified flow, and in SLUGGIT
v.2 and LASSI forms of the two-fluid model are used to describe these.
4.1 Method structure rather than operating from the level of the entire flow
system, responses and interactions are given each in-
Central to the dynamic behaviour of the SLUGGIT dividual fluid object, allowing for tailor-made local
and LASSI programmes is the dynamic, object ori- dynamic responses to local bubble-scale events. The
ented programming techniques used in their struc- dynamic behaviour seen from a flow system level is
ture. The fundamental benefit of a completely object thereby the result of the combined interactions of all
oriented programming structure is the way in which objects which individually live and respond accord-
dynamic behaviour is implemented on a local level; ing to individual characteristics.
Bubble section
Bubble-slug border
Bubble unit Slug unit
Pipe
Slug-slug border
Slug-bubble border Bubble-bubble border
Slug section
Figure 4.1: Main object classes
Details on the object oriented code structure will tion. Figure 4.1 illustrates the fundamentals of object
not be given here – see e.g. [22] for this purpose. structure; objects of the ancestor class Section hold
Some description is however deemed necessary as masses and have spatial extent. Those objects of the
differences in programme versions are strongly con- ancestor class Border divide the Section objects at
nected with the object oriented structure. These ob- a spatial position. The movements of these borders
jects are continuously and dynamically moved, de- govern the movement of the fluid objects and change
stroyed, created and transformed during the simula- in section lengths. Pipe properties, such as diam-
4
This assumption may be influential in deep-water case simulations.
5
These topics are however approached and take a central place in the LASSI scheme [37] (Section 4.11) and indirectly in the topic
of hydrodynamic instability and the Kelvin-Helmholtz criterion (Annex E). Rolling wave functionality has been implemented in the
SLUGGIT v.2 framework, though this will not be approached or activated during the present work (see e.g. De Leebeeck [10])
26
eter, roughness and inclination, are extracted from section level in bubbles while the same equations are
the Pipe. Section objects retrieve pipe informa- treated at unit level in slugs, making it possible to
tion from the Pipe object spanning the spatial length connect single equations to whole slugs.
which the section in question occupies. Unit classes
(Unit_slug and Unit_bubble) are ancestors of 4.2 SLUGGIT scheme versions
Section classes, collecting all sequentially similar
Section objects into one. It could be said that sec- The method derived in Section 4.3 is based on the im-
tions are ‘subgrids’ of ‘unit grids’, but in order to plementation [21] by Jørn Kjølaas [22], referred to
underline the strongly object oriented nature of these as ‘SLUGGIT v.2’. This implementation was further
programmes the term ‘grid’ will be avoided. A ‘grid’ developed by Tor Kjelby and Angela De Leebeeck,
suggests something static rather than the a dynamic, and adjusted further by the present author during this
self-sustaining environment wherein the number and thesis work. Alternative implementations of simi-
nature of objects change continuously. Perhaps the lar, yet simpler implementations are also available
most important dynamic feature in this respect is that at EPT; previous to Kjølaas’ version, Pascal Kle-
the pipeline is automatically divided into compress- bert implemented a code [23], documented in [24],
ible (bubble) and incompressible (slug, plug) units, closer to the original model of Nydal and Banerjee
generating regions wherein the system solved may [34]. This code is in the present work referred to as
be compressible or incompressible. Thus, the prob- SLUGGIT v.1. In addition to these versions, Kjelby
lem of having a universal scheme for both compress- has implemented an alternative, simplified proce-
ible and incompressible flow is elegantly avoided. It dure in the framework of SLUGGIT v.2, with the
becomes possible to control the level at which each purpose of merging the different procedures into a
of the governing equations are solved. In particu- single framework.6 Here, this simplified version is
lar, mass and momentum equations may be solved at termed ‘SLUGGIT v.2s’
Here will follow short version descriptions to underline the main differences between versions.
SLUGGIT v.1: In form of implementation, this version has benefited for far less time in development and
debugging than v.2. It makes use of template function structures rather than class inheritance which is
widely adopted in v.2. In terms of the model, uniform pressure (and gas density) throughout every bubble
unit (Taylor bubble and stratified region) is assumed. As a consequence, a slug or pig object is needed
to generate a change in pressure. The benefit of this assumption is that only a single pressure equation
is solved for each bubble unit. Because checker board solutions become impossible with this scheme,
non-staggered grids have been employed – see Figure 4.7a and 4.6a. The mass flux and convection terms
are neglected in the pressure and momentum equations. Gas entrainment in slugs is also included in
the v.1 implementation, promoting a mixture formulation of the slug mass and momentum equations.
No sources, endothermal behaviour or boundary conditions other than fixed inlet flows are presently
supported.
SLUGGIT v.2: Motivated by the need to capture quick transients reactions for sudden pig and plug discharges
in pipelines, SLUGGIT v.2. was implemented with individual pressures for each section of a bubble unit.
In order to avoid unphysical checker board solutions, a staggered grids must then be used (Figure 4.7b.)
The momentum balance, including all terms, is solved for slug units and both phases in all bubble sec-
tions, except those neighbouring a slug unit where the empirical border velocity relations of Section 4.8
dominate – see Figure 4.6b. Gas entrainment in slugs (‘void in slugs’) is not supported in this version,
though thermal behaviour (solving the energy equation), support for rolling wave phenomena [10] and a
wider range of inlet conditions are available. Implicitness is imposed as presented in Section 4.3. Also,
6
A great misfortune of the codes at EPS is that they have been developed separately in individual frameworks, see section 6
27
automatic time step length based on the CFL condition is implemented, optimising the computation time
while ensuring numerical stability.7
SLUGGIT v.2s: This simplified version can be considered a cross between SLUGGIT v.1 and v.2. It is imple-
mented in the v.2 framework, allowing it to benefit form all the work invested into the v.2 implementation
in terms of making the code more robust. Additionally, all extra management and parameter options are
available, such as the automatic time step regulation, possibility of simulation restart and hydrodynamic
slug initiation mechanisms. The assumption of a uniform bubble pressure is employed also here, and
convection terms in the pressure and momentum equations are neglected. The slug momentum and bub-
ble unit pressure equations are coupled and solved implicitly (see Figure 4.6a). For the liquid phase in
bubbles, an explicit momentum balance is solved at the sections (non-staggered,) while the border ve-
locities sat to the mean of that in neighbouring sections. The gas velocities are not determined by the
momentum equation at all, but form an explicit mass balance at the borders. Here, the velocities are
governed by the fact that gas velocity equal border velocity at bubble unit borders. As the gas density,
due to the pressure, is uniform throughout the bubble unit, this determines the gas velocity as inversely
dependant on the void fraction of the cross section area αg . Neighbour means are used for the gas section
velocities. Mass equations are treated as in v.2. Void in slugs is presently not supported. Suggested
documentation: [24, 34, 33].
4.3 SLUGGIT – Fundamental Equations Notation

The following subscript conventions will be used
(see Figure 4.2)
σ ∈ {w, e} Σ ∈ {W, E} κ ∈ {l, g}

In this section the the basis for the SLUGGIT model
is derived and presented in some detail. The aim where l and g denote ‘liquid’ and ‘gas’, respectively.
here is to present the model in details fine enough W , P and E denote ‘western’, ‘present’ and ‘eastern’
to establish an overview of the fundamental ideas control volume cell, while w and e denote the western
and differences between the models. This is done by and eastern control volume face values, respectively.
first deriving the general balances used in the meth- This is in accordance with the conversion used in
ods and then examining the way in which the grids many CDF applications (e.g. [45]). ‘Left’ and ‘right’
are defined and quantities approximated within the are also standard notations, though letters denoting
different frameworks. In order to obtain new in- ‘left’ and ‘liquid’ may easily be confused. As the
sight, for both reader and writer into the methods, model to be developed is 1D, dimensional indexa-
the present section seeks to establish the undamen- tion is deemed superfluous and non-bold vector and
tal equation applying control volumes directly to the tensor counterparts are streamwise components by
∆ ∆
discrete scheme and derive the numerics from there. conversion (e.g., u = uz , τ = τzz ). All variables not
In contrast, [16, 22, 37, 24] proceed by deriving or marked with a tilde ( ˜· ) represents cross section aver-
presenting the continuous two fluid model (4.26) and ages unless otherwise stated. Some attempt is made
simplify and discretise from there. The notation used of distinguishing the main variables stemming from
presently is chosen to correspond reasonably to the the conservation equations from those of a paramet-
model sources [22, 37, 34, 24, 20] and CFD conver- ric or algebraic nature by using capital symbols for
sion [45, 30]. This derivation was developed during the latter.
the present work.
7
As with the other EPS slug tracking codes, SLUGGIT v.2 has issues with robustness, but this is mainly due to the flow object
management operations. Only during the present thesis work SLUGGIT v.2 has improved in stability greatly through continuous
debugging and management measures.
28
Figure 4.2: Control volume Figure 4.3: Pipe cross section
4.3.1 Mass Balance

By applying unity as the transported property to the Reynolds transport theorem over the control volume in
Figure 4.2 the continuous momentum equation takes the form
* *
d
ρ̃ dV + ρ̃ ũ − ṽ · ñ dA = ˙ s dV

m̃ (4.1)
dt
Ωκ ∂Ωκ Ωκ
The tilde indicated that these are local variables in all dimensions of space, that is, they are continuous and
their integral values will be approximated by control volume averages. Based on Assumption A.5, a key
approximation used in the FVMs is:
" " "
1
ρ̃(ũ · ñ) dA ≈ ρ̃ dA · ũ − ṽ · ñ dA = nρκ ακ A(uκ − v) σ

(4.2)
(ακ A)σ
∂Ωσ,κ ∂Ωσ,κ ∂Ωσ,κ
ακ in (4.2) is the section volume fraction of metric flux Q is defined by

fluid κ, i.e. ακ = Vκ /V , Vκ being the volume occu-
pied by fluid species κ. ακ A thereby represent the Qσ,κ = (uκ − v)An σ

(4.3)
mean stream-normal area occupied by fluid κ. nσ is
the z-component of the ∂Ωσ surface normal; ne = 1, The term ‘extensive’ is here, perhaps improperly, ap-
nw = −1. plied because Q is defined using the total pipe area A
while retaining the intrinsic velocity averages, and so
only appropriate for the flux of pipe-specific superfi-
In the following, the convection term is slightly cial quantities (m, mu). Beware that the area of Qn+1
reformulated in terms of superficial mass and flux is still taken from the previous time step.
quantities8 m, Q; rather than expressing the flux of m is nearly equivalent to the density, except that it
mass ρκ Vκ across a border sub-area (ακ A)σ , the total is the specific mass based on total pipe area A. Also,
area A and superficial specific masses mκ = Mκ /V = ρ should not be confused with m in that the notation ρ
ρκ ακ are used, Mκ and V being the total specie mass will be reserved for densities derived form the equa-
and control volume. Likewise, the extensive volu- tion of state (EOS), used in the pressure equation,
8
The term ‘superficial’ is equivalent to extensive (as opposed to intrinsic) spatial averages, e.g., using the notation of [16]: ρκ = hρ̃κ ii
and mκ = hρ̃κ i = ακ hρ̃κ ii
29
while Mκ (and mκ ) originate from the conservative mass conservation equations. In practice, the volume
mass equation. V n+1 is taken after the border positions are corrected
The mass flux over a border in (4.2) can now be with zn+1 = zn + vn δt, but before the slug border po-
written on superficial form (mQ)σ,κ : sitions have been accurately adjusted with respect to
  the new liquid mass [21], making it only a prelimi-
 dM p X nary future volume. Implicitness between the control
+ (nmQ)σ = ṀP 
s

 (4.4)
dt σ volume masses and the velocities is lost. However,
κ
the mass continuity equation is still incorporated im-
The implicit Euler method is used for the time dis-
plicitly into the momentum equation in two ways.
cretization:
Firstly, the mass equation (4.5) is used to simplify
 
 (mV)n+1 − MPn X h i the momentum equation, allowing the transient vol-
+ = ṀP 
P n+1 s

n(mQ) (4.5)
δt
 σ
σ
ume integral to be evaluated for velocity rather than
κ
momentum. Secondly, the pressure, which is solved
The volumetric fluxes for the current time step Qn+1 implicitly with the velocity, originates from a refor-
is available as the phase momentum equations and mulation of the continuity equation, incorporating a
pressure equation are computed prior to the phase compressibility relation for the gas phase.
4.3.2 The pressure equation

The pressure equation is obtained directly by reformulating the mass balance (4.4) by applying the chain rule
on the first transient term, assuming ρg = ρg (p, T ):
∂ρ dp ∂ρ dT
" #
dMκ dρκ dVκ dVκ
= Vκ + ρκ = Vκ + + ρκ
dt dt dt ∂p dt ∂T dt dt
Since the main contributions to the absolute hydrostatic pressure is found from streamwise pressure gradients,
and only pressure gradients have a fine scale impact on the momentum equation, there is no ambiguity is placing
this EOS pressure p at the fluid interface. As the liquid phase is assumed incompressible (Assumption A.4)
this equation only becomes informative for the gas phase. Still, a combined phase control volume formulation
(overall continuity) allows for the simplest form:
n 
X VP,κ  ∂ρ pn+1 n
∂ρ dT 
!  X Ṁ s !
P − pP
X X nσ h i X
+  + n
m Q n+1
+ (nv )σ AP =
n
+ ψns,P (4.6)
ρn 
∂p δt ∂T ρn σ,κ ρn

κ P,κ T dt P κ σ P,κ σ κ P κ
Dividing by the respective density serves to bal- density in upholding the Equation Of State (EOS),
ance the equation with respect to the fluid densi- which in this case is the ideal gas law. Because sim-
ties. Doing otherwise would cause the liquid phase plifications and non-conservative inaccuracies occur
to dominate the equation, possibly leading to conver- in this non-conservative mass formulation, the abso-
gence problems [13]. In this work only isothermal lute pressure will eventually diverge from that pres-
systems will be considered (Assumption B.5). Be- sure which is consistent with the EOS and densities
cause of this the temperature term in (4.6) is disre- from the conservative mass equation (4.5). The fi-
garded and no energy equation will be solved. Equa- nal term ψ is a countermeasure for this; a correction
tion (4.6) is Laplacean in nature, meaning that it gov- source term accounting for the discrepancy between
erns the streamwise change in pressure as a function EOS pressure as found form the conservative masses,
of the change in mass. Such a formulation is useful and the pressure change from the non-conservative
as it is the pressure gradient which appears in the mo- pressure equation 4.6. This ‘trick’ of enforcing mass
mentum equation. However, absolute pressure is also conservation by explicitly supplementing the pres-
important as pressure must be in agreement with gas sure with the error from the previous time step allows
30
for consistency to remain between pressure and gas discrepancy in the momentum equation can originate
mass. The volume error takes the form: only from the present time iteration.
V

X mκ 
  The SLUGGIT v.2 scheme uses a staggered grid
ψs =

 − 1 (4.7) in order to avoid the notorious checker-board pres-
δt ρκ

κ
sure solution. An uniform interface pressures across
By including this error correction source term the er- entire bubble units is assumed in SLUGGIT v.1. Ac-
ror sustained during one time iteration is removed in cordingly, SLUGGIT v.1 solved (4.6) at bubble unit
the iteration which follows. Thus, pressure-density level, while SLUGGIT v.1 solves it at section level.
4.3.3 Momentum Balance
By applying the momentum to the Reynolds transport theorem (e.g., [16, 46, 45]) over the control volume in
Figure 4.2 the continuous momentum equation takes the form
* *
d
ρ̃ũ dV + ρ̃ũ ũ − ṽ b · n dA = τ̃ · n − p̃n dA + ρ̃g + m̃
˙ s ũ s dV (4.8)
dt
Ωκ ∂Ωκ ∂Ωκ Ωκ
Here g · n = −g sin φ.
The pressure integral require some extra attention. It evaluates as a surface integral of local pressure p̃, but
only the pressure p at the gas-liquid interface generated from (4.6) is available. Transverse pressure gradients
must also be included. Because there may be streamwise gradients in the liquid height hl , transverse pressure
consideration will affect the pressure acting in the streamwise momentum balance also. The two-fluid model,
applies a hydrostatic approximation to include the gradients in fluid height. In fact, the two fluid model, on
which this derivation is based, becomes ill-posed if this hydrostatic pressure contribution is neglected (the
‘constant pressure model’) [16].
Figure 4.4: Pipe cross section Figure 4.5: Hydrostatic approximation of liquid
height gradient influence
The pressure integral in (4.8) may be evaluated hydrostatically, formulating the western and eastern inte-
31
grals around the centre hight hP :
X " h i
− p̃n dA = − p(z) + h(z) − y ρl ez · g dA

ez · nσ
σ
∂Ωκ ∂Ωσ,κ
X "
 
X Zhe 
(4.9)
= (nακ p)σ A +  hσ dA − y b̃(y) dy ρl g cos φP
 
σ  σ

∂Ωσ,κ hw
| {z }
Vh
where b̃(y) is the pipe breadth orthogonal to the flow plain (see Figure 4.4) and p is the pressure at the interface
found from the pressure equation (4.6). Vh (the term inside the square brackets) require some extra attention. In
a differential formulation its value would follow from integration using Leibnitz’ theorem. Here one must trod
carefully
h i h i
Vh (hακ A)e − (hακ A)w − 21 b̃(hP ) · h2e − h2w = he (ακ A)P + 12 ∆hb̃P − hw (ακ A)P − 21 ∆hb̃P − hP b̃P ∆h
| {z }
(he +hw )(he −hw )
where ∆h = he − hw . The western and eastern areas have been expressed as deviations around AP,κ – see the
schematic of Figure 4.5. ∆h d has here been assumed, allowing the pipe breadth b̃(y) to be approximated
constant equal to b̃(hP ) within the narrow integral region. Rearranging yields
Vh AP,κ ∆h + b̃P 21 (he + hw ) ∆h − b̃P hP ∆h = AP,κ ∆h

| {z }
hP
accordingly, the hydrostatic non-constant pressure model (4.9) evaluates as

X
− p̃n dA = − (nακ )σ Aκ pσ + hσ ρl g cos φP

ez · (4.10)
σ
∂Ωκ
The result is easily understood from Figure 4.5 as wall normal density gradients within the phase sub-
a correction to the gravity term for the liquid height domains, the fluxes may be assumed averaged and
gradient. There he > hw have arbitrarily been as- the final simplification with regard to the convection
sumed, which does not affect the result. Expres- term is
sion (4.10) was derived with the liquid phase in fo- "
ρ̃ũ ũ − ṽ · ñ dA

cus (σ → l), though it also holds for the gas phase
(σ → g).9 It is however only in quite special cases ∂Ωσ,κ
" "
that the contribution of the gas weight will be impor- 1
ρ̃ũ dA ·

≈ ũ − ṽ · ñ dA
tant. (ακ A)σ
∂Ωσ,κ ∂Ωσ,κ
In the final paper of [37] a simplified transverse
momentum balance is developed as an alternative to = (ρuQ)σ,κ
hydrostatic approximation.
The shear force integral in (4.8) evaluates as:
The convection integral (‘momentum flux’) is now ez ·
h
τ̃ · ñ dA = (τA) pipe,k + (τA)int
i
(4.11)
sat in focus. Again, the base assumptions of the fi- P
∂Ωκ
nite volume method is that integrated properties can
be approximated by a uniform control volume aver- where the subscripts pipe and int respectively indi-
age. Also applying Assumption A.5 of neglectable cate the pipe wall and fluid interface control volume
9
In the gas region, both y ≥ h and p̃(y) ≤ p, changing the signs twice and producing the same result.
32
surfaces. The shear forces τ must be evaluated from Friction Multiplier (IF M ) to the friction factor. The
the local strain across these surfaces. Simple Darcy friction forces used are computed from the gas phase
friction factor relations will be used to approximate
1
these, and so they will be linearised functions of the τint = · IF M · λg ρg ug − ul ug − ul
8
relative difference in kinetic energy across radial sur-
faces. and so the interface perimeter is included in the gas
1
τ = λρ|u|u (4.12) hydraulic diameter
8
Many different friction correlations are have been πD2 πD2
Dh,g = , Dh,l = (4.14)
developed throughout the years. Issa et al. proved Sg + Sl Sl
these to be of some influence in their slug capturing
scheme [13]. Here, the Darcy friction coefficient λ is Since the convection term of the momentum equa-
determined form the explicit Haaland correlation: tion introduces non-linearities (momentum is trans-
ported by means of fluid motion,) the convective flux
!1.11  terms are evaluated using velocities from the previ-

1  6.9 ε
√ = −1.8 log  +

 (4.13) ous times step Qn . The same non-linearity is found in
λκ Reκ 3.7Dh,κ
the friction factor definition (4.12) and also here the
where ε is the pipe wall roughness and the phase squared velocity is approximated as the product of
Reynolds number Reκ is computed using the hy- the velocity form the precious time iteration and the
draulic diameter Reκ = ρκ Dh,k uκ /νκ . The Dh is the present velocity for which the system is being solved
hydraulic diameter – the length scale analogous to – un+1 |un |. An iterative approach is also possible here.
a pipe diameter evaluated from the surface periph- This would require expensive re-evaluation of the
ery and cross section area. Momentum exchange be- pressure-momentum system in each times step and
tween phases across the interface is strongly simpli- has for this reason not been implemented. The possi-
fied and use the same kind of friction factor, which ble benefits in terms of accuracy are small in light of
is equivalent to approximating the phases as solid the overall model simplifications As before, the first
surfaces in relation to each other (Assumption A.6). order Euler method is used for the time derivative.
Possibility for including some linear deviation from Collecting all the above methods of evaluation, (4.8)
this assumption is supplied by adding an Interphase takes the form:
 X

 − nσ mn Qn un+1 + αnκ Apn+1
σ,κ σ



σ





 1
− (Sκ Lλκ ρκ )nP unP,κ un+1




 P,κ


 8
(Mu)n+1 − (Mu)n
" # 
1

=

− · IF M · (Sint Lλg ρg )nP unP,κ − unP,↑κ↓ un+1 n+1 (4.15)

δt P,κ − uP,↑κ↓
8


P,κ 

hnl,e − hnl,w


 " #

− gMP,κ sin φ +
n
cos φ



`nP






 s
+ Ṁu




P,κ
where the subscript ↑κ↓ indicates a property of the other phase. The last source term is usually of little relevance
as mass source inn and outflows are in most cases orthogonal to the main stream.
In multiplying the discrete mass balance (4.5) with the velocity at the computed time step un+1
P and subtract-
33
ing this from (4.15) the momentum balance simplifies to10
X h i
+ nσ (mQ)nσ,κ (uP − uσ )n+1 − αnκ Apn+1

κ

σ


σ





1


− (Sκ Lλκ ρκ )nP unP,κ un+1



P,κ

8



n+1

− un
" #
nu

 1
=

M − · IF M · (Sint Lλg ρg )nP unP,κ − unP,↑κ↓ un+1 n+1 (4.16)

δt P,κ − uP,↑κ↓
P,κ



 8
hnl,e − hnl,w


 " #
− gMP,κ sin φ + cos φ


 n
`n




i P



 h
+ Ṁ s u s − un+1



P,κ
In slugs, Equation 4.16 is solved for the liquid phase only (k = l), meaning that the interfacial friction term
(3rd term on RHS) vanishes. If gas entrainment is included in the scheme (SLUGGIT v.1,) then gas propagation
is determined by a slip relation. [7, 24, 10].
The complete system of Equations (4.5), (4.6) and (4.16) is first order accurate in both time and space.
4.4 Implicitness and time regulation unity

u j,κ δt
A purely explicit scheme may appear attractive is Cmax = max <1
j, κ `j
terms of simplicity of implementation. However, for
stability reasons explicit procedures are impractical For accuracy reasons, choosing Cmin in the range of
for the purpose of simulating slow transients. Tra- 0.1 to 0.01 is appropriate for large cases using SLUG-
ditionally, most nuclear reactor safety codes would GIT. For the TRIOMPF slug capturing code Bonizzi
be implemented explicitly as these were designed to et al. operated with Cmax = 0.5 [7], which for that
simulate small time invervals [5]. code is acceptable as accuracy is ensured through it-
However,the Courant Friedrich Levy (CFL) cri- eration at each time step.
terion for explicit time iteration imposes a time step
restriction of δt ≤ min j,κ ` j / u j,κ ± c j,κ where c j,κ is the 4.5 Computational expense
speed of sound for fluid κ at position z j .11 This re-
flects the non-relative propagation of information in Let ∆T be the (real) time the SLUGGIT programme
the system of hyperbolic Euler equations. For im- needs perform a simulation of the time interval ∆t of
plicit procedures however, only the dynamic mass- a pipeline of length L. The length of a computational
transport criterion applies: time step δt will, due to the CFL criterion, be linearly
dependant on the smallest bubble section length:
`j
δt ≤ min
j, κ u j,κ δt ∝ min ` j , j ∈ [1, N]
j
which basically is the assurance that the advection Because the system is one dimensional, forming tri-
will not be able to pass entirely through, or bypass, a and heptadiagonal band systems, the physical com-
cell within a single time step. In practice, the Courant putation time δT needed to compute a single simula-
number chosen to determine the time step is less than tion time step δt will be proportional to the number
10
As a sidenote, applying the same procedure to the differential formulation of mass and momentum equations produce a convection
term on the form ρ (u − v) ∂z u for each phase. The upwind discretisation of un+1σ,κ in (4.16) will then, in a finite difference formulation,
be equivalent to moving the differential ∂z u upwind.
11
The speed of sound originate from the eigenvalues of the Euler equations and the CFL criterion can either be understood by decou-
pling this equation system through transforming it to eigenvalue variables and performing a von Neumann stability analysis, or simply
by realizing that for the information required at time n + 1 to be available from neighbouring cells at time n, the time step must be small
enough so that cells beyond neighbouring cells cannot be reached along the system characteristics in a single iteration [45, 30].
34
fluid sections: 4.7 Staggerd and non-staggered grids
L
δT ∼ N · δτ ≈ · δτ Where properties are defined depends on how the
`
scheme grid is constructed, or rather, where the con-
where δτ is the physical computation time needed per trol volume is placed. To avoid the notorious checker
section per simulation time step and δz the average board solution, staggered grids must be applied in
section length. The total physical computation time schemes where pressure-velocity variable pairs are to
becomes solved in each node. In SLUGGIT v.1 and v.2s this
δT δτ L/`
∆T = ∆t ≈ ∆t issue is avoided solving for a single bubble unit pres-
δt min j ` j
sure rather than respective pressures for each section
By assuming that the average and minimum section
velocity. Accordingly, non-staggered grids are cho-
lengths are proportional it is found that
sen. Here, all properties are defined in each section.
∆T ∝ `−2 (4.17) SLUGGIT v.2 and LASSI are based on a two-fluid
formulation and must apply staggered grids in which
4.6 Non-locally defined properties the control volumes are shifted for the velocity and
pressure equations, respectively. This places implic-
Some properties included in the control volume itly computed velocities at borders in SLUGGIT v.2
equations are not defined at all locations where they while pressures are places at borders in LASSI. Fig-
art needed. As a consequence,approximations these ure 4.7 illustrates the difference for the momentum
quantities at other locations are needed. balance control volume. In evaluating the mass and
All slug tracking schemes discussed here uses the pressure equations both versions apply the grid over
upwind scheme to evaluate the advection terms and the section object, similar to Figure 4.7a
central averaging for the flux terms. The benefit with
Shifting the pressure rather than the velocity in
using an upwind approach is that it is both simple and
a staggered scheme, as is done in LASSI, may in-
numerically robust. The most prominent drawbacks
troduce some benefits. Such schemes will have a
are that it is only first order accurate and subject to
fully defined flux term in the section. In dips, hav-
severe numerical diffusion. The same goes for the
ing the pressure defined at the borders may also
first order Euler discretisation in time. Numerical
be beneficial as pressure gradients here will be of
diffusion would be especially damaging to the so-
importance for the seepage and fluid accumulation
lution had the model been purely Eulerian, which is
which affects terrain slugging characteristics. Also,
standard in most unit-cell and capturing codes. Often
as will be discussed later, Renault [37] found that the
Eulerian schemes compansate by enforcing a higher
velocity staggered grid in conjunction with the up-
order spatial discretisation or solving with a high
wind scheme tend to shift the pressure drop caused
resolution in space. However, the Lagrangian slug
by Bernoulli suction downstream the hold-up protru-
tracking approach avoids diffusion at slug and wave
sion.
fronts by where it would be most damaging. This is
Table 1 lists the variables not locally defined
the main idea of slug tracking.
when using the different grid structures in a stratified
area.
A border property χ evaluated by an upwind
scheme (indicated by a ‘hat’) is taken from upstream
the flux: Figure 4.6 illustrate the schematic differences in
  momentum control volume systems between the sim-
χW ; Qw ≥ 0 χP ; Qe ≥ 0
 
  plified and non-simplified versions of SLUGGIT v.2.
χw =  χe = 
 
χP ; Qw < 0 χE ; Qe < 0
b b

 
 Key features in the simplified scheme are single unit
bubble pressures and neglect of momentum convec-
while centrally defined properties χ are simply eval- tion between bubbles and slugs. This allows the mo-
uated as mentum equation to be formulated as a system of
1
χσ = (χΣ + χP ) slug control volumes only. Pressures and border ve-
2
locities are formulated into the the momentum sys-
35
tem, making bubble units similar to control volume attempted in Table 2. This information has been
face values. An attempt at surmising the all evalu- (painstakingly) collected form the source code [21]
ation situations in SLUGGIT v.2 and v.2s has been for the purpose of familiarisation.
Mass Eq Pressure Eq Momentum Eq

σ , uσ
mn+1 mnσ , un+1 mnσ , unσ , un+1
σ , α σ , hσ
n+1 n n
non-staggered σ
Velocity-staggered (SLUGGIT v.2) mn+1
σ mnσ mnP , unσ , un+1
σ , vσ
n+1
Pressure-staggered (LASSI) mσ , un+1

n+1
σ σ , mP (in ψ )
vn+1 n n
mσ , uσ , uσ , ασ , hnσ
n n n+1 n
Table 1: Variables not locally defined
SLUGGIT version: v.2 v.2s(VI)

Variables Slug(I) immersed bubble section bubble unit border
(III) (IV)
Mln
[ n+1

Mσ,κ
(mQ)n+1
σ,l ρl Qd
n+1
σ,l n+1 (ul − v)σ
`σ
n+1
(u − v)n+1
`n+1 l
Same as v.2
Σ
[ n+1
Mσ,g
(mQ)n+1
σ,g Not solved `n+1
(ug − v)n+1
σ 0(I) Not solved(V)
σ
(a) Mass equation

Variables immersed bubble section bubble unit border
n Mln n
[ [
Mσ,κ Mσ,κ
mn Qn+1 `σ
n (ul − v)σn+1 `n (ul − v)n+1 n+1
`n (ul − v)σ
σ,l Σ σ
n
[
Mσ,κ (I)
n n+1
m Q `n (ug − v)σ
n+1
0 0(I)
σ,g σ
(b) Pressure equation

Variables Bubble Slug(I) Bubble(III) Slug(I)
M n n
(mQ)nσ,κ (ακ A)nσ κ
uκ (II) − v Aw ρl unP,l − vnσ Ql = 0, gas not solved(VII) 0
V σ σ
un+1
σ,κ
n+1
ud
σ,κ
n+1
ud σ,l 0 0
n
αnσ Apσn+1 defined 1 n
2 (Aw + Ane )pn+1
σ 0 V
` pn+1
σ
κ P
n n
MP,κ MP,κ defined defined defined
(c) Momentum equation
Table 2: Undefined variables

(I) Special case of no void in slugs; no gas flux through unit border as uσ,g = vσ .
(II) for the liquid phase, unσ,κ is computed from an explicit section momentum balance.
(III) Explicit computation
(IV) Due to explicit slug treatment this term takes the old mass
V
(V) Gas mass equation not solved; section gas mass computed by Mg,section = Mg,unit Vg,section
g,unit
(VI) Uniform bubble pressure assumed
(VII) Bubble gas velocities are determined from mas balance assuming constant uniform density
(pressure) and the the gas velocity to equal the border velocity at the slug front.
When evaluating the momentum equation of the simplified scheme v.2s in bubble units, a non-staggered
36
(a) Simplified scheme: SLUGGIT v.2s
(b) Non-simplified scheme: SLUGGIT v.2
Figure 4.6: Momentum equation control volumes included in implicit system momentum computation. Note
that pn+1 = p(un+1 ) is included as the pressures in the momentum equation (4.16) are eliminated using the
pressure equation (4.6)
(a) Non-staggered grid (b) Velocity-staggered grid
Figure 4.7: Momentum equation control volume
37
grid (Figure 4.7a) is used for the liquid phase while staggered grids (Figure 4.7b) are used for the gas phase
[21]. In this way, ug = v is maintained at the bubble-slug/slug-bubble borders. Undefined border and section
velocities are then later approximated by the means u.
4.8 Slug border velocities cal and horizontal pipes, respectively [5, 4]:
U0,ν = 0.35 gD U0,h = 0.54 gD

p p
The mechanics governing the propagation of bubbles
through slugs, or slugs through stratified sections, are
essentially three dimensional and influenced by mix- The relation’s dependency on the slug Froude num-
√
ing effects, wake effects, bubble shape and surface ber Fr s,m = u s,m / gD is an effect of bubble position in
tension. Obviously, a coarse grid, one dimensional the pipe cross-section. Increasing the Froude num-
method cannot obtain the relevant detailed informa- ber tend to move the bubble nose centre towards the
tion needed to capture bubble propagation warranting pipe cross-section h centre. For
i Reynolds numbers in
the need for empirically modelled closure relations the range Re ∈ 4 · 104 , 105 , the centre field veloc-
for the slug and bubble unit border velocities. ity is about 1.21 times the cross-sectional average.
Firstly, a distinctive classification is needed in or- This supplies a limiting bubble-bulk liquid velocity
der to characterise the behavioural properties of the ratio C0 of about 1.2 in turbulent flows – the bubble
border. Two basic types of borders are defined – bub- moves faster than the surrounding liquid. In hori-
ble noses (alternatively ‘slug tails’) and bubble tails zontal pipes, complete bubble centring takes place
(alternatively ‘slug fronts’). The basic characteristics for Fr > 3.5. [4] When the slug velocity is oppo-
behind this distinction is the physical bubble shape; a site that of the bubble (sign U0 , sing u s,m ) the rela-
bubble front will be the ‘spearhead’ of bubble pene- tive flow across the bubble is from nose to tail, giv-
tration through slugs, entrapping liquid from the slug ing a different bulk drag on the bubble. Bendiksen’s
tail. The relations imposed to model bubble prop- experiments indicate that a C0 value close to unity
agation applied in all versions of the slug tracking (C0 = 0.98) is then appropriate [4], which is also in-
schemes follows the assumption of Nicklin et al. that cluded in the implementation.
the bubble front velocity can be expressed as a linear ξ± in (4.19) is a variable added to indicate the
relation from the slug mixture velocity sign logic influencing the horizontal drift. This logic
is a bit cumbersome and depends on whether the pipe
vb,n = C0 u s,m + U0 (4.18) inclination is positive, negative or zero, and whether
the nose is pointing to the left or the right. These fea-
where subscripts b and s denote ‘bubble’ and ‘slug’, tures also influence – and should be seen in light of –
respectively. n denotes ‘nose’ while m the ‘mixture’ whether a unit border is defined as a bubble nose or
velocity, i.e. um = κ ακ uκ . The work of Bendiksen
P
tail (Section 4.9). As an illustration, the logic applied
[4] for long Taylor bubbles in inclined tubes is ap- for SLUGGIT v.2 and v.2s (taken form source code
plied to close the model [34, 22, 20, 24, 37] 12 [21]) is

C0 = 1.05 + 0.15 sin2 φ sign φ φ , 0


3.6
ξ± = 

Fr s,m ≤ cos φ (4.20a)
U0 = U0,ν sin φ + ξ± U0,h cos φ nb

 φ=0
(4.19) 
+1 bubble-slug border

C0 = 1.2

nb = 

(4.20b)
Fr s,m > 3.6
cos φ −1 slug-bubble border

U0 = U0,ν sin φ

where U0,ν and U0,h are the bubble drift velocities in A similar expression may be used for the slip
stagnant liquid (neglecting surface tension) for verti- velocity for gas entrained in slugs. Such slip relation
12
The presented drift model (4.19) is not identical in all publications [4, 34, 22, 37, 20, 24] and some alterations have been made in
recent time to the source codes. For instance, smoothing functions have been implemented in SLUGGIT v.2 similar to those presented
in [15], expanded to handle flow in both directions. [37] contains an error in the presentation of (4.19).
38
often have the form ug,s = S d (ul,s − U0 ), S d being the α s,g will be substituted with the void fraction in the
distributed slip ratio [24, 10]. A similar slip relation wave section. Also analogous to slugs, a wave tail
approach is used in the case of three-phase flow. relation will be used similar to that of the bubble
nose expression (4.19), except that the wave liquid
As slugs shed liquid at the slug tails (to the trail- hold-up, rather than wave section length, will be al-
ing bubble), they may retrieve liquid at the slug fronts tered as a consequence of unequal liquid fluxes at tail
(from the tail of the downstream bubble.) This hap- and front. Rolling waves will not be considered in
pens when the slug mixture velocity is larger than this work and accordingly disabled in all simulations.
the liquid film velocity around the bubble ahead of it,
which will be the case for steady state bubble propa- In SLUGGIT v.2 and v.2s, the border velocities
gation due to continuity. Obtaining a relation for the v are expressed as multiples of the gas and liquid ve-
slug front velocity from a mass balance in this case locities
is straight forward and schematically demonstrated vn+1 = Cg un+1
g + Cl un+1
l +U (4.22)
in Figure 4.8.
allowing for the border fluxes to be expressed with
full implicitness in the mass and pressure equations
(4.5), (4.6). The momentum equation applies the old
border velocity; doing otherwise would make it non-
linear.
4.9 Turning point criteria

Relation (4.19) only holds for bubble nose borders
while (4.21) is only appropriate for slug fronts mov-
ing faster than the liquid film ahead. Therefore, some
Figure 4.8: Principle used when computing slug turning point criterion is needed in order to deter-
front velocities mine whether the border in question is a bubble nose
or bubble tail. Empirical studies indicate that the
This reveal that sign of the pressure gradient constitutes an appropri-
dzb,t u s,l − (αu)b,l ate criterion – i.e. bubble fronts points the opposite
vb,t = = (4.21a) direction of the pressure gradient [4, 22].
dt αb,g
or, more generally, if the slug entrains gas, the rela- Slightly different turning point criteria are used
tion becomes a balance of mixture velocities: in the different method versions. All rely on a turn-
ing a friction-gravity balance based on the general
(αu) s,l − (αu)b,l assumption is that the bubble always moves opposite
vb,t = (4.21b)
αb,g − α s,g of the pressure gradient [34].
subscript t denoting ‘tail’. If the liquid film veloc-

2 λl ρl U crit |Ucrit | S = ρl gA sin φ
1
(4.23)
ity is greater than the slug liquid velocity, the shock
expression (4.21) would be entropy violating, and so which constitutes a criterion based on the (liquid
(4.21) is disregarded in such cases [22, 37]. film) Froude number, as in the bubble noes model
In the case of rolling wave objects, which is im- (4.19). Bendiksen [4] applied this criterion in a
plemented in SLUGGIT v.2/v.2s, expression (4.21b) mostly analytical investigation and found this to be
will also be used for wave fronts [10]. In this case, in good agreement whit experimental data.
SLUGGIT v.1 uses Nydal and Banerjee’s original procedure [34] where, if the pipe is inclined upwards along
the general flow direction, all bubble-slug borders are assumed to be bubble noses and all slug-bubble borders
are assumed to be slug fronts. The Froude number condition (4.23) is solved iteratively.
39
isNose: φ>0 φ<0
bubble-slug true u s,m > Ucrit
slug-bubble false u s,m < Ucrit
If isNose == true then v = vb,n and Bendiksen’s empirical bubble nose velocity (4.19) is used. Otherwise
the border is assumed to be a slug front and (4.21) is applied.
SLUGGIT v.2n and v.2s uses the Blasius friction coefficient in turbulent pipe flow to find a simplified explicit
critical velocity such that no iteration is needed (see source code [21]). Applying Blasinus’ friction coefficient
to (4.23) yields  
 !1/4 1/1.75 
2gD | sin φ| ρl D

 
|Ucrit | = max 
   
|U0 |, (4.24)
 
0.316 µl
  

 


The following logic is applied for determining the boolean isNose [21]:
isNose: φ>0 φ<0 φ=0
bubble-slug u s,m > −|Ucrit | u s,m > |Ucrit | u s,m > −|Ucrit |
slug-bubble u s,m < −|Ucrit | u s,m < |Ucrit | u s,m < |Ucrit |
or, using the logical variables (4.20)
isNose = nb u s,m > −nb ξ± |Ucrit | (4.25)
LASSI uses the same critical velocity (4.23), only LASSI allows for bubbles to travel towards the inlet in
inclined pipes. Rather than basing the formulation on whether the Bendiksen correlation (4.19) is appropriate
or not, LASSI bases the formulation on when the slug front formulation (4.21) is not appropriate, i.e. when the
slug liquid velocity is not great enough to capture liquid from the neighbouring bubble’s liquid film: u s,l < ub,l
isNose: u s,l < ub,l u s,l > ub,l
bubble-slug u s,m > Ucrit true
slug-bubble true u s,m < Ucrit
As with SLUGGIT v.1, an iterative approach is used for obtaining Ucrit .
4.10 SLUGGIT v.2/v2s computational sequence

The sequence undergone at each simulation time step is the following:
1. Border velocity coefficients (4.22) are computed from bubble nose (4.18) and slug front (4.21) relations.
2. The coefficients in the combined pressure (4.6) and momentum (4.16) equation system are found from
previous time step values and stored in an external matrix. Each object is connected to a row of this
matrix and individually takes care of updating this row without management interference (for details, see
[22] and source codes [21, 23, 36]). In the SLUGGIT v.2 scheme, this forms a linear, heptadiagonal band
systems of three upper and three lower co-diagonals, along with the main diagonal. Tridiagonal band
systems are formed in the case of the other two SLUGGIT versions.
3. The pressure-momentum system is solved using Gauss-elimination. u, v, z and p are updated to n + 1.
4. The coefficients in the mass equation for bubble sections are found from available quantities and stored
in external matrices, one for each bubble unit. In the simplified scheme, this is done for the liquid phase
only. If not simplified, this forms two, tri-diagonal systems (one for gas and one for liquid) for each
bubble unit consisting of more than two bubble sections.
40
5. The mass systems are solved through Gauss-elimination without pivoting (Thomas or TDMA algorithm.)
Bubble masses are updated.
6. Liquid mass equations for slugs are solved explicitly and slug lengths are updated.
7. Gas densities ρg are updated form the EOS and a test is preformed to see whether the volume fraction
error ψ from the pressure equation (4.6) is too big. If it is, values are reset and the time step computation
is re-preformed using a smaller time step.
8. Management operations are preformed. This is the only part of the routine which takes place above object
level. Here, too long section objects are split in two, too small sections are merged with neighbouring
objects, and the nature of objects may be converted according to initiation criteria.
9. n := n + 1: the cycle is repeated.
The computational sequence of the LASSI scheme is similar, but slightly different in due to less time in
development and a different set of equations (Section 4.11). Mainly, the difference lies in the shallow water
wave equation solving procedure forming a central component is this scheme. For details, see [37, 36].
4.11 The LASSI scheme and hκ (ακ ) the phase height (seeFigure 4.2). Fur-
ther details on these variables are given later in Sec-
Although also the LASSI scheme is based on the tion 4.3
same fundamental two-fluid equations as SLUGGIT
v.2, it developers them into a rather different sys- The LASSI scheme employs a few simplifying
tem of equations. LASSI can be seen as an inter- assumptions in order to decouple the gas and liquid
mediary between a fully resolved two-fluid model phases in (4.26). These assumptions are
as that of Issa and Kempf [13], and the SLUGGIT
schemes originating from [34]. Including the full Assumptions C. LASSI
model derivation also for this scheme would make
the this text too extensive in terms of methodology. C.1 Assumptions A and B holds. Wave phenom-
Rather, an abbreviated presentation is given and the ena are not neglected.
interested reader may refer to the thesis source [37]
C.2 In regard to the momentum equation (4.26b)
for further details.
for liquid, the gas phase may be considered in-
compressible
4.11.1 Abbreviated method presentation
C.3 Derivatives of the mixture velocity can be
LASSI is based on a differential formulation of the neclecten in (4.26b) for the liquid.
two-fluid model:
C.4 Gas momentum is neglectable compared to
∂ ∂
(ρα)κ + (ραu)κ = 0 (4.26a) liquid momentum (i.e., ρl ul ρg ug )
∂t ∂x
 (τS ) By eliminating the pressure in the liquid mo-
pipe,κ (τS )int
− ∓κ mentum equation (4.26b) through substitution with



A A



its gas equation counterpart, and then applying the

∂ ∂ ∂p



− ακ

(ραu)κ + ραu = 
2

(4.26b)

∂t ∂x κ 
 ∂x " assumptions, a decoupled, modified shallow water

∂hκ equation is produced for the liquid phase:


 #
 − ρκ gακ sin φ + cos φ



∂x

∂ ∂
αl + (αu)l = 0 (4.27a)
which constitutes four equations. The friction τ is ∂t ∂x
defined so that ∓κ is (+) for the liquid phase and (−) ∂ ∂ h 2 1 2 i αl
(αu)l + αu + 2 καl = FV (4.27b)
for the gas phase. S symbolise the wetter perimeter ∂t ∂x l ρl
41
where the modification appears in the form of a sec- transition to model instability takes place when there
ond spatial derivative involving κ: exist a disturbance wavenumber whose correspond-
ρl − ρg dαl
ing pulsation has a negative imaginary part (see An-
1 ρg
!−1
κ= g cos φ − (ug − ul )2 (4.28) nex E). The continuous two-fluid model is used for
ρl dhl αg ρl
such analysis. Renault went on to prove for the dis-
which is the balance between the Bernoulli suction crete LASSI scheme that the effect of finite border
1 ρg ρl −ρg
dα −1
2
αg ρl (ug − ul ) and hydrostatic head ρl g dhl
l
cos φ . spacing ` on model stability is a new term propor-
FV is the volumetric forces acting on the liquid phase: tional to that spacing, along with a ‘numerical sur-
! face tension’ term. In terms of model stability, the
X (τS )int (τS )κ,pipe latter has no significant impact. The former increases
FV = ∓κ − − ρκ g sin φ (4.29)
κ
ακ A ακ A the discrete model stability domain all over as bor-
In the LASSI scheme, (4.26) is first solved im- der spacing ` increases. Enlarging the pipe diameter
plicitly for the gas velocities and pressures along reduce this effect.
the pipe. The border velocities are then sat as de-
scribed in Section 4.8. A modified Lagrangian shal- Further, a character of the continuous two-fluid
low water scheme is then employed to solve (4.27) model is that relative velocities leading to the sys-
for the liquid velocities and hold-up. This shallow tem being ill-posed as a hyperbolic problem coin-
water scheme solves the Reimann problem at each cides with the inviscid Kelvin-Helmholtz (IVH) two-
bubble border, accurately determining wave propa- fluid model stability criterion [13]. 13
gation. The mobility of borders is utilized in mov-
ing bubble-bubble borders to coincide with the wave This means that in the inviscid case, the transition
shock fronts, allowing the inherit discontinuity to be to a non-stratified flow takes place as the eigenvalues
handled without much numerical diffusion (smear- – the slopes of the characteristics – of the two-fluid
ing) while avoiding excessively refined grids. As in model become complex. Unique solutions continu-
the SLUGGIT schemes, an upwind spatial discreti- ously dependant upon the initial conditions is then
sation is applied combined with Euler discretisation no longer guaranteed. Rather, numerical solutions
in time – see [37, p. 87-89] for details on the discrete strongly influenced by the numerical discretisation
scheme. may be found. Ill-posedness is usually manifested in
a discrete model by the solutions produced therefrom
failing to become independent of grid spacing as this
4.11.2 Domain of well-posedness and slug cap-
is continuously reduced. To illustrate this, Figure 4.9
turing
show two plots borrowed from a publication of Issa
A characteristic of the discrete LASSI scheme which and Kampf wherein one is well-posed and and the
is of some importance is that it retains nearly the other is ill-posed. The growth rate increases asymp-
same model stability domain as the original two- totically with decreasing grid spacing in the ill-posed
fluid model, converging towards the Viscous Kelvin- case, whereas this becomes independent of grid spac-
Helmholtz criterion (VKH). In stability analysis, the ing in the well-posed case.
The domain of well-posedness is determined by tem and finding its eigenvalues. For the two-fluid
formulating the two fluid model as an equation sys- model, the well-posedness criterion reads [16, ch. 5]
αl αg
" # !−1
dαl
well posed ⇔ (ug − ul )2 ≤ (ρl − ρg )g + cos φ (4.30)
ρl ρg dhl
13
Though very similar and, perhaps, both arising from the mechanism causing flow regime transition, the issue of well posedness
should not be entirely confused with stability of the stratified flow regime. While the former is a mathematical property of the continu-
ous two fluid model (4.26) the latter is physical character of the fluid dynamics, often analysed using that same model.
42
(a) Well-posed (b) Ill-posed
Figure 4.9: Illustrating of manifestation of ill-posedness. Source: [13]
which bear striking resemblance to the IKH cri- marked by the IKH-criterion, viscous forces prove
terion (E.3), mounting to a balance between the destabilizing to the flow [3], causing flow transi-
Bernoulli suction and the hydrostatic forces. Also, tion to a new flow regime in viscous fluids to take
for ρl ρg , well-posedness becomes equivalent to place at lover relative velocities, marked by the
κ > 0, where κ emerged naturally in the wave formu- VKH-criterion [13]. A state in between these cri-
lation (4.28) of LASSI. Infinite liquid depth need not teria should therefore lead to flow transition without
be assumed in this criterion, contra to the regime sta- becoming ill-posed. This has in recent years boosted
bility condition. Bonizzi et al. [7] also showed that the confidence in the two fluid model as a means by
their extended two-fluid model, which includes gas which automatic slug capturing, rather than applying
entrainment in slugs, share the same criterion. initiation mechanisms, may be utilised. Practising
A problem in using numerically dissipative pro- some moderation in the choice of relative velocities
cedures such as the upwind scheme (used in all mod- will however be required.
els discussed here) is that numerical diffusion tent
to dampen and mask the symptoms of ill-posedness. Experiments have further confirmed that insta-
For many years this has damaged the credibility of bility in the continuous two-fluid model corresponds
methods utilising the two-fluid model [13]. Physi- well with physically observed instability and transi-
cally, ill-posedness in the two-fluid model can be ex- tion to a non-stratified flow regime. Some scepticism
plained by the the model’s neglection of certain as- still remains over what happens when approaching
pects, such as surface tension and more complex vis- the limits of these domains. Finally, as was demon-
cous effects [37]. strated by Bonizzi [6], also numerical issues tend to
Indeed, after Ramshow and Trapp raised the arise as the two-fluid model approach the limit of
problem of ill-posedness in 1978, doubts about the well-posedness. Thus, the domains of physical in-
applicability of the two-fluid model became wide stability, model instability, numerical instability and
spread. In later years, as ill-posedness of the two- ill-posendess of the two-fluid model are similar and
fluid model was seen in light of the model (strati- overlapping. These relationships are illustrated in
fied flow) stability limit, confidence has been some- Figure 4.10
what restored; though the limit of well-posedness is
Figure 4.10: Slug initiation principle in the LASI scheme. Figure source: [37]
43
As LASSI scheme retains the same model sta- In this respect, none of the versions of SLUGGIT
bility domain of the original two-fluid model, the solve the Reimann problems of wave propagation.
approach implemented in LASSI is then to assign Since SLUGGIT v.2 applies a two-fluid model in
a minimum value for κ ≥ κmin 14 to ensure well- stratified regions, it may be slug capturing, but only
posedness while the model instability will ensure with very fine border spacing. The other methods
slug initiation. (v.1, v.2s) are simplified to the extent that their model
These considerations supplies the LASSI method stability domain probably does not match the physi-
with an advantage over methods which are to dif- cal one, and so cannot be used as capturing methods.
ferent from the two-fluid model to share its stability Instead, the hydrodynamic slugging phenomena have
domain; hydrodynamic slug initiation happen auto- to be modelled through a criterion and clugs artifi-
matically driven by the appropriate physical mech- cially initiated. Such models are often questionable
anisms which are retained in the method through in complex geometries [37] and the terms compos-
the Reimann-solving wave procedure. This makes ing such criteria often come at large computational
LASSI a slug capturing method, yet far less compu- expense15 . The same goes for the perhaps most com-
tationally expensive than the finely gridded, iterative mercially popular multiphase pipe flow code on the
Eulerian two-fluid representation model of Issa et al. market: OLGA [5].
5 Benchmark tests
5.1 Liquid conservation

This section concerns the methods’ ability to conserve mass. For reasons of simplicity, only liquid conservation
is considered.
Introducing two ways of computing the liquid volume contained in the pipes:
Zzo N
X
V dz,l (t) = A αl (z, t) dz = A αl, j δz j (5.1a)
R
zi j=1
and
Zt
Vio,l (t) = VR +A αl (zi , t∗ ) ul (zi , t∗ ) − αl (zo , t∗ ) ul (zo , t∗ ) dt∗

dz,l (t0 ) (5.1b)
t0
where subscripts i and o indicate ‘inlet’ and ‘outlet’, respectively. The V-section geometry of Figure 5.1 is
tested. It is composed of two 2 m long pipes with opposite inclination φ ∈ ± {0◦ , 30◦ , 60◦ }. The pipe diameter is
1.2 cm.
Figure 5.1: Configuration used to thest liquid consercation in Figures 5.2 and 5.3.
14
After inspecting the source code [36] it is found that the value of κmin currently implemented is a fraction of the hydrostatic force:
ρ −ρ
−1
κmin = 0.2 l ρl g g cos φ dαl
dhl
15
Experienced from the SLUGGIT v.2 code.
44
Figures 5.2 and 5.3 depict the pipe liquid vol- ing the inlet liquid velocity and TargetLength pa-
umes computed using (5.1b) and (5.1a). The times rameter.16 The gas velocity is the same as in the previ-
tmid and tout presented in the captions are respectively ous case, i.e. 0.44 m/s, and the inlet hold-up is kept at
the times at which the liquid fronts first reach the a constant αl = 0.5. The timestep used is δt = 0.01 s.
mid section (the bend) and the outlet of the pipeline, Geometrically, the pipe configuration is a horizontal
which initially is free of liquid. It is clearly visible pipe section of 10 m followed by a 85◦ inclined riser
that, while SLUGGIT appears to conserve the liquid long enough so that the liquid front never reach the
mass appropriately, LASSI produce a biased liquid outlet during a simulation period.
error highly dependant on the pipe inclination. This
is visible in the plots as rates of error accumulation
change at t = tmid , which is the time at which liquid
reach the intersection change of the first and second Figure 5.4: Riser geometry used to
pipe segment. For horizontal pipes and pipes where test LASSI liquid volume error in
1000 m
the inclination angle is small, this error is not large Figures 5.5.
enough to be directly visible from the hold-up ani-
mations generated in PLOTIT [31]. Increased angles
are associated with increased volumetric errors. Er- 20 m
rors occurring with steep pipe inclinations, |φ| ≥ 60◦ ,

are clearly visible in the animations. Simulations per-
formed wherein only a single inclined pipe section Varying the time step appears to have some mi-
have produced the same error as if it had been pre- nor impact on the results, also in the horizontal pipe
ceded by a horizontal pipe section, indicating that the section, but this influence is small compared to that
bend itself is not the main source of error. of the inlet velocities and the TargetLength pa-
The computed liquid flux at the outlet is slightly rameter. Figures 5.5 show that increasing the liq-
imprecise due to the slugging and momentary nature uid inlet velocity Ul,i reduces VR dz,l relative to Vio,l .
of the output data, which is why the time series are Likewise, increasing TargetLength reduces VR dz,l
cut around when liquid first reaches the outlet. This relative to Vio,l . The reasonable appearance of Fig-
occurs at t = tout ure 5.5c (Ul,i = 0.50 m/s, TargetLength = 0.1 m)
should not be confused with a solution to the volume
It should be pointed out that the V-section geom- error problem; it is merely the intersection of over-
etry and all simulation parameters corresponds to the and under-predictions of VR dz,l – reducing either time
simulations presented in the second paper of [37]. step or TargetLength will not lead to an asymptotic
In these simulation the inclination angles range from reduction in error.
φ ∈ [−15◦ , 30◦ ]. As is evident from Figures 5.2b
and 5.2c, the errors are small in this inclination The consequences of this failure to conserve liq-
regime. However, in the third paper of [37] a S- uid mass in the available LASSI code [36] is fatal, as
riser is simulated whose upstream rise is close to 55◦ seen in Figure 5.5e. Here, the liquid in the pipe VR dz,l
and downstream rise near 75◦ . reaches a constant value long before any liquid ever
reaches the riser outlet. In other words, even as new
In order to investigate the errors apparent in liquid is pushed in through the inlet ad infinitum, no
LASSI further, and illustrate some of the problems liquid will ever reach further than the first metres of
inconsistent masses will impose on simulations of a the riser, all new inserted liquid mysteriously disap-
deep water riser case, the same mass conservation pearing from the system.
test is applied to a riser case of similar dimension
to the Girassol field. Figure 5.4 presents a schematic. The κ-values (Equation (4.28), κ < 0 ⇔ well-
Figure 5.5 depict the volume calculations (5.1) vary- posedness criterion (4.30) for ρl ρg ) are giver in
16
TargetLength is not an universal section length, but is the length of sections at the inlet and half the length at which sections split
in two. The timestep also influence the average section length through the CFL criterion [37, paper 1, p. 60]
45
−4
x 10 SLUGGIT
V
io,l
V
Liquid Volume in Pipeline [m3]

∫ dz,l
0
0 2 4 6 8 10
Time [s]
(a) φ = 0◦ , tmid = 4.65 s, tout = 9.55 s

−4
x 10 SLUGGIT −4
x 10 SLUGGIT
Vio,l Vio,l
V∫ dz,l V∫ dz,l
1 1
0 0
0 2 4 6 8 10 0 2 4 6 8 10
Time [s] Time [s]
(b) φ = ∓30◦ , tmid = 1.85 s, tout = 7.15 s (c) φ = ±30◦ , tmid = 5.40 s, tout = 7.35 s
−4
x 10 SLUGGIT x 10
−4 SLUGGIT
Vio,l Vio,l
V∫ dz,l V∫ dz,l
0 0
0 2 4 6 8 10 0 2 4 6 8 10
Time [s] Time [s]
(d) φ = ∓60◦ , tmid = 1.80 s, tout = 7.15 s (e) φ = ±60◦ , tmid = 5.25 s, tout = 7.65 s
Figure 5.2: Liquid volume conservation, SLUGGIT v2. V-section configuration, as in Figure 5.1. (δt = 0.01,
D = 0.012, αl,i = 0.275)
46
−4
x 10 LASSI
V
io,l
V

∫ dz,l
3
0
0 5 10 15
Time [s]
(a) φ = 0◦ , tmid = 5.80 s, tout = 13.30 s

−4
x 10 LASSI −4
x 10 LASSI
Vio,l Vio,l
V∫ dz,l V∫ dz,l
0 0
0 2 4 6 8 10 0 2 4 6 8 10
Time [s] Time [s]
x 10
−4 LASSI −4 LASSI
x 10
Vio,l Vio,l
V∫ dz,l
Liquid Volume in Pipeline [m ]
V∫ dz,l
3
0 0
0 1 2 3 4 5 0 2 4 6 8 10
Time [s] Time [s]
Figure 5.3: Liquid volume conservation, LASSI. V-section configuration, as in Figure 5.1. (δt = 0.01, D = 0.012,
αl,i = 0.275)
47
Fiugres B.4 and B.5. It is seen that the V-section N is the total number of temporal samples. The ba-
simulations, for which lack of mass conservation sic steady-state model code can be found in List-
was documented, is well posed. Unsurprisingly, the ing 7, Appendix D, while Listing 5 presents the pro-
same can not be said about the riser case due to the cedure for averaging the SLUGGIT simulation data
steep riser inclination. Since the flow is slugging by means of interpolating projection. Gas and liq-
here, the ‘stratified’ Taylor bubbles are in any event uid mass sources representative of the Girassol oil
no longer than a few sections long. field [47, 43] are applied to a closed inlet. Four
values of the parameter MAX_BUB_COEF were tested:
It is evident that the available LASSI code [36] MAX_BUB_COEF ∈ {1 000, 200, 100, 50}. This pa-
cannot be applied to the riser stability study under- rameter dictates how long a bubble section can be,
taken within this work. In consideration of the rea- in terms of pipe diameter multiples, before it is split
sonable results published in [37] for slightly inclined into smaller sections. MAX_BUB_COEF = 100 is not in-
pipelines, and in reviewing the details of the scheme, cluded in Figure 5.6 as it overlaps MAX_BUB_COEF =
it seems probable that the absence of liquid conserva- 50 and 200 nearly perfectly.
tion is due to some error in the code implementation
– possibly within the section management procedure. 0.7
Steady state computetion, film included
Steady state computetion, film neglected
0.6 SLUGGIT; MAX BUB COEF = 1000
SLUGGIT; MAX BUB COEF = 200
5.2 Gas expansion in riser SLUGGIT; MAX BUB COEF = 50
0.5
The slug tracking schemes discussed in this work all Ṁg = 33.69 kg/s
Ṁl = 1.687 kg/s
have two basic flow elements – the Taylor bubble and hαg i
0.4
the slug. It has been assumed that with these two el-
0.3
ements most flow regimes can be approximated, and
the appropriate basic flow system characteristics ob- 0.2
tained. In this section, the methods ability to handle
riser expansion will be tested. A short, closed inlet 0.1
with mass sources will be used to make the physical

0
regime bubbly and challenge the slug-flow approx- −1400 −1200 −1000 −800
z
−600 −400 −200 0
imation. Mass sources are chosen to correspond to

those active in the larger field case study to come. Figure 5.6: Gas volume fraction profile αg (z). Dy-
namic simulations vs. steady state analyses. Closed
In Appendix A, a unit-cell type steady-state inlet. Ṁg = 33.69 kg/s, Ṁl = 1.687 kg/s
method developed for analysis of the total void frac-
tion in the riser. A discrete procedure for finding the The numerical steady-state model including the
profile of the mean void fraction hαg i(z) is presented liquid bubble film from Appendix A.1 produce a
in Section A.1. These models have been developed near perfect match with the finely spaced section
during the present work. simulations, even though friction effects have been
neglected in the steady-state models. These are
Figure 5.6 shows the gas volume profiles from in any case low in the simulations conducted here
the steady state computation of Appendix A.1 (µ ∼ 10−3 Pa/s). The simplified version, SLUGGIT
against temporally averaged dynamic simulations v.2s, give equally good results, except that the ampli-
with SLUGGIT v.2. The mean profile is here deter- tude of the sinusoidal MAX_BUB_COEF = 1 000 curve
mined by computing is less, suggesting that the simplified scheme is less
ZtN affected by slug entrance and exit effects. A possible
N
1 1 X explanation for this is a damping of the entrance and
hαg i(z) = αg (t0 , z) dt0 = αg (ti , z) δti
t N − t0 tN − t0 i=1 exit disturbance obtained since pressure equalisation
t0
in the pipeline bubble unit happens instantaneously,
becoming a sample average in discrete space. Here, not dependent on internal gas flux.
48
x 10
−3 LASSI x 10
−3 LASSI
7 7
Vio,l Vio,l
6 V∫ dz,l 6 V∫ dz,l

5 5
4 4
3 3
2 2
1 1
0 0
0 100 200 300 400 500 0 100 200 300 400 500
Time [s] Time [s]
= =
   
 TargetLength 0.1 m   TargetLength 0.01 m 
(a)  ul,i = 0.125 m/s (b)  ul,i = 0.5 m/s
   
 
tmid = 138 s tmid = 44 s
 
LASSI
0.01
V
io,l
V∫ dz,l
0.008
0.006
0.004
0.002
0
0 50 100 150 200 250 300
Time [s]
=
 
 TargetLength 0.1 m 
(c)  ul,i = 0.5 m/s
 

tmid = 52 s

LASSI LASSI
0.035 0.015
Vio,l Vio,l
0.03 V∫ dz,l V∫ dz,l
0.025
0.01
0.02
0.015
0.005
0.01
0.005
0 0
0 50 100 150 200 250 300 0 100 200 300 400 500
Time [s] Time [s]
= =
   
(d)  ul,i = 2.00 m/s (e)  ul,i = 0.5 m/s
   
 
 
Figure 5.5: Liquid volume conservation, geometry as in Figure 5.4, LASSI. (δt = 0.01, Ug,i = 0.44, D = 0.012,
αl,i = 0.5)
49
Studying the animations generated form the sim-
Although the present result is quite satisfying, it ulation data provides an alternative explanation.
should not be forgotten that both SLUGGIT and this As the maximum bubble section length decreases,
steady-state model rely on the same bubble nose ve- the model is capable of capturing the rise in hold-
locity expression for closure. The ancestor model on up similar to a hydraulic jump at the inlet to the
the other hand, in which the liquid film is neglected, first horizontal pipe segment. This was seen for
grossly over-predicts the void fraction profile, indi- MAX_BUB_COEF = 100 and MAX_BUB_COEF = 50 (see
cating that this model was too simple. Figures 5.7c), and allows for some hydrodynamic
to take place, similar to the finely gridded capturing
methods of, e.g., Issa and Kempf [13]. This makes
Interestingly, section length dependant profile
the average bubble length shorter. Therefore, when
oscillations are uncovered. Large amplitude, low
a bubble enters the riser, a smaller amount of gas
frequency oscillations dominate the large-bubble-
mass is subtracted from the riser inlet pipe section.
section simulation profile (MAX_BUB_COEF = 1 000).
Conversely, with a large value on the MAX_BUB_COEF
A low amplitude, high frequency component is also
parameter, all slug initiation upstream of the riser is
visible here. In the small-bubble-section profile
killed off by numerical damping. A similar influ-
(MAX_BUB_COEF = 200 and 50,) it is the high fre-
ence hydrodynamic upstream slug initiation on the
quency oscillations that dominate, while these are far
riser dynamic code has already been documented by
less severe than in the aforementioned case. From
Kjeldby [19] using the same SLUGGIT v.2 code.
Figure 5.6 one can see that the MAX_BUB_COEF = 200
simulation bears larger oscillations near the riser in-
and outlet, indicating that this is a slug entrance and Because the inlet is closed, a bubble being sucked
exit phenomenon increasing with slug length.17 into the riser inlet will cause a significant reduction
of gas in the horizontal riser inlet pipe, causing a
pressure reduction at the inlet. Upstream pressure
A possible complimentary explanation for the
reductions causes a deceleration of the bubble once
profile oscillations’ dependency on the parameter
in the riser, making it’s occupation time in the lower
MAX_BUB_COEF may lie in how bubble volumes
section of the riser longer (first local maximum if
themselves have a tendency oscillate in the deep
Figure 5.6.) As the gas source once again fill the
water cases, especially when the pressure conditions
first horizontal pipe with gas, the pressure rises and
are altered as bubbles enter and exit the riser. This
the bubble accelerates through the middle part of the
is rooted in the large hydrostatic heads pressing on
riser (middle local minimum in Figure 5.6,) before a
the bubbles and the inertias of the liquid columns.
new bubble enters the riser, again causing a pressure
Allowing for larger bubble lengths increases the av-
drop and a deceleration in the riser (final local max-
erage bubble volume in the riser (see Figures 5.7),
imum.) This effect is further reflected in the phase
increasing the bubble compressibility. These big-
portraits of Figures 5.8a and 5.9b, and described in
ger bubbles causes longer periods of oscillation with
Cycle F.
larger total changes in volume during the bubble
rise. Also, the times steps are significantly larger for
simulations allowing longer bubbles as this places a In Figures 5.8 the Φl − p phase portraits (nor-
less strict CFL criterion on the simulation. It is im- malised scatter-plots of total riser liquid fraction vs.
portant to note that the SLUGGIT simulations were riser inlet pressure) are shown for the two cases
performed in a simple geometry with only four pipe MAX_BUB_COEF = 100 (b) and MAX_BUB_COEF =
segments, allowing for bubbles to take on the maxi- 1 000 (a) – two of the three simulations whose hαg i-
mum length. Including short pipe sections ahead of profile was shown in Figure 5.6.
the riser (such as small jumpers) would have a simi- For the case of MAX_BUB_COEF = 100, and all
lar effect as reducing the parameter MAX_BUB_COEF. smaller values of this coefficient, the phase diagram
appears fairly random, though askewed, indicating a
17
This is more clearly seen in simulations of intermediate MAX_BUB_COEF values, not shown here.
50
(a) MAX_BUB_COEF = 1 000 (b) MAX_BUB_COEF = 200 (c) MAX_BUB_COEF = 50
Figure 5.7: Snapshot of bubbles in simple riser system. Image pipe diameter rescaled for visibility. (This is
also done in all pipeline images throughout this text.)
0.15
Slug front 16
reaches
riser outlet 0.1 15
Bubble noes 14
enters riser 0.05

13
12
Φl − hΦl i
Time [h]
0
hΦl i
11
10
−0.05
Slug created −0.1

8
at inlet
7
Slug tail −0.15
exits riser 6
−0.1 −0.08 −0.06 −0.04 −0.02 0 0.02 0.04 0.06 0.08

pi − hpi i
hpi i
(a) MAX_BUB_COEF = 1 000 (b) MAX_BUB_COEF = 100
Figure 5.8: Φl − p phase portraits of simulations in Figures 5.6 and 5.7.
51
Slug tail
exits riser 6
16
15
5
14
4
13
Ṁl − hṀl i 3 12
Time [h]
Slug created hṀl i
at inlet 11
2
Slug front reaches 10
pipeline outlet
1 9
Slug front
reaches 8
riser outlet Bubble noes
0
Slug tail reaches
enters riser 7
pipeline outlet
−1
6
−0.1 −0.08 −0.06 −0.04 −0.02 0 0.02 0.04 0.06 0.08

pi − hpi i
hpi i
(a) MAX_BUB_COEF = 1 000 – Snapshot Figure 5.7a (b) MAX_BUB_COEF = 100
Figure 5.9: Ṁl − p phase portraits of simulations in Figures 5.6 and 5.7.
correlation between inlet pressure and riser void. cle is repeated. Specific events form the four cor-
Only analysing the riser system in a hydrostatic ners of phase portrait shape as the system state moves
frame, one would rather expect a perfect correla- clockwise along a thin line in Φl − p space. The time
tion (points forming a positively inclined line,) but spans t ∈ [35, 100] hours, after the system has come to
transient momentum effects and a near constant riser a steady-state cycle18 . Smaller data point marker than
void fraction due to small bubbles causes scattering. those in Figure 5.8b have here been used to indicate
It is likely that, if run long enough, the simulation how narrow the line of this pattern is. Even so, nearly
with MAX_BUB_COEF = 100 would also converge to a only the point from the first hours are visible as these
steady cycle pattern. However, time in computation overlap those following. A single cycle takes about
this would require can simply not be afforded. three minutes, making the number of cycles shown
more than one-thousand. The events forming the cy-
The Φl − p phase portrait for MAX_BUB_COEF = cles are manifested in the corners of the shape and
1 000 (Figure 5.8a) is significantly different. Here, are the following:
the data points form a distinct bacon shape as a cy-
Cycle F. Φl − p cycle; Figure 5.8a
F.1 Lower left corner: A slug is formed in the horizontal upstream the riser and this slug front enters the riser
inlet. The liquid fraction in the riser increases and the inlet pressure increases as the inlet is blocked and
upstream volume is reduced. This corner is not ‘sharp’ because the cross section area of the gas phase
αg A at the riser inlet narrows before slug initiation (αg > 0.95 at inlet).
F.2 Upper left corner: The front of the slug highest positioned in the riser reaches the riser outlet and enters
the downstream horizontal. Riser liquid fraction reduces as the topmost slug is blown out.
F.3 Upper right corner: The slug entering the riser in Step F.1 has completely entered the riser. The column
of slugs and bubbles in the riser is pushed higher towards the riser outlet and the inlet volume is increased,
reducing the pressure. The topmost slug is pushed out of the riser
F.4 Lower right corner: The slug exerting the riser outlet in Step F.2 has completely left the riser. Liquid is
no longer drained from the riser and the liquid fraction increases form the riser inlet.
18
Characteristics of entire cycle is time independent.
52
Snapshot 5.7a is taken in the narrow time slot between Step F.2 and F.3, marked and magnified in Fig-
ure 5.8a.
The MAX_BUB_COEF = 1 000 Ṁl − p phase portrait the upper horizontal is less.
shown in Figure 5.9a becomes discontinuous and Code samples used in producing these portraits
only vaguely similar to the phase portrait example are given in Listings 3 and 4.
Figure 3.5a presented in the introductory discussion
(outer ring in this image). Still, the process has a cir- In Figure 5.10, the pressure times series corre-
cular, counter-clockwise moving portrait pattern with sponding to the MAX_BUB_COEF = 100 and
a flat bottom where production becomes zero. The MAX_BUB_COEF = 1 000 cases are shown. It is clear
discontinuity is a characteristic of the slug tracking that the larger bubble units causes the amplitude
method; the discontinuity of hold-up in slug fronts of the pressure fluctuations to be 3 – 4 times that
and tails produce jumps in production as unit bor- of the case with more finely spaced bubble sec-
ders pass through the outlet. Here, the ‘production’ tions. Also apparent is the fact that the simulation
Ṁl is taken form the pipeline outlet. Measuring Ṁl of MAX_BUB_COEF = 100 has not reach an equally
at the riser outlet will give a somewhat different por- steady cycle as at this time interval as for the coarser
trait where the production delay due to transport in section case.
140
MAX_BUB_COEF = 1000
MAX_BUB_COEF = 100
135 MAX_BUB_COEF = 50
130
Pressure
125
120
115
110
105
4 4.05 4.1 4.15 4.2 4.25 4.3 4.35 4.4 4.45 4.5
Time [hours]
Figure 5.10: Pressure profiles of simulations in Figures 5.6 to 5.9.
Sharp pressure fluctuation profile like those from tive and marks the riser as securely stable according
the MAX_BUB_COEF = 1 000 case is characteristic of to Taitel’s model.
terrain slugging, but is slugging flow the physically Likewise, the Bøe criteria (3.1) can be checked.
accurate solution? The severe slugging result can be It may be rewritten:
seen in connection with the theory of Taitel from Sec-
RT
tion 3.3. Criterion (3.2), telling whether the riser is Type I slugging ⇒ Ṁls ≥ Ṁ s (5.2)
gαg L g
table and can easily be checked: Approximating the
properties of (3.2) from the plots available through Feeding inn the numbers the RHS evaluates to
PLOTIT [31], it is found that L = 600 m, H ≈ 1300 m, 0.4970 kg/s – the necessary criterion for terrain slug-
hΘg i ≈ 0.6, Po = 37 bar, hΦl i ≈ 0.69 and α0g was found ging type I is satisfied. Accordingly, the system is
to be of the order 0.9 in [41]. Due to the shortness of in the enclosed rectangular region of the flow map
the horizontal inlet, the RHS of (3.2) becomes nega- in Figure 3.3. Therefore, it is likely that the physi-
53
cally accurat solution should be either cyclic without 5.3 Influence of gas lift on riser void frac-
fall-back, or steady flow.19 tion
Judging from the pressure time series in Fig- Using the models from Appendix A, the void frac-
ure 5.10, it seems likely that the the physical sys- tions representative for the Girassol field case [47]
tem would be stable. All simulations have come have been computed (H = 1278 m, d = 0.2032 m,
to a reasonably steady state, in the sance that the po = 37 bar, T = 59 ◦ C – isothermal approxima-
average signal amplitudes and statistical moments tion.) The spatially discretised model A.1 with
have been established. Only the simulation with the Girassol production state Ṁl = 33.7 kg/s, Ṁg =
MAX_BUB_COEF = 1 000 is operationally unstable Ṁgwell = 1.69 kg/s (152 kSm3/d), finds a average void frac-
per present definition. Further, no consistent pres- tion hΦg i = 0.345, while the void fraction found
sure fluctuation pattern or cycle is found, making the for Ṁg = Ṁgwell + Ṁggl = 2.35 kg/s is hΦg i = 0.422.
signal appear as a steady state pressure with white The void fraction attributed to the gas lift is then
noise. Obviously, the small variations in pressure for 0.3645 − 0.290 = 0.0774, or a head reduction from the
MAX_BUB_COEF 1 000 are caused by the same slug pure liquid case of ∆H well = 98.9 m.
entrance and exit effects as in Cycle F, though the Using the analytical model of Appendix A.2
riser bubbles are far smaller and far more numerous, yields ∆H gl = H(0.4064 − 0.3254) = 103.5 m.
causing a more complex dynamic and supplying the
signal with a random element.
The pressure population standard deviation (sec- 1
Numerical computation, film included
ond moment) s p of the inlet pressure time series are Analytical computation, film included
0.8 Numerical computation, film neglected
Analytical computation, film neglected
sp = 9.1 bar sp = 3.8 bar s p = 2.7 bar 0.6

1 000 100 50 hΦg i
Ṁg =0.658kg/s
0.4
where the numbers refer to the value of
MAX_BUB_COEF. This amounts to a reduction in 0.2
standard deviation proportional in order of magni-
tude to the reduction in bubble section length, which 0
is not unexpected. The pressure fluctuations will be 0 0.5 1S 1.5 2 2.5
ul [m/s]
proportional to the slug lengths as these are formed.
Bubbly flow will be dominant in the physical steady Figure 5.11: Void fraction as a function of superficial
state case which the slug-stratified flow model con- liquid velocity, Ṁg = 0.658 kg/s (70 kSm3/d)
verge towards as the bubble sections approach in-
finitesimal lengths. Figure 5.11 shows the total void fraction’s de-
pendency on the liquid flow rate for the gas. In
The influence of section lengths on the system comparing this with Figures A.4 in the appendix, it
dynamics makes it clear that the section length pa- is clear that the simpler liquid film neglecting analy-
rameters does not only affect accuracy and numerical sis in Annex F supplies an acceptable approximation
stability. Even though coarse resolution (vast sec- only when liquid flow rates are high relative to gas
tions) may give reasonable results (as they do in the flow rates.
Girassol field case), these should be validated against
more finely resolved simulations to ensure that the If there is no liquid flow and the system has come
main dynamics are not significantly altered. to a stand-still, the void fraction produced by a gas
lift will be fare more significant. In a stand still sit-
19
note that Figure 3.3 is only an example – the the domain of steady flow inside the Bøe criterion may be larger than it appears here.
54
uation ( Ṁl = Ṁgwell = 0) with a gas lift flow rate slowly than Taylor bubble. The analysis still provide
Ṁglg = 0.658 kg/s = 70 kSm3/d the numeric liquid-film- an order of magnitude estimate, approving in accu-
including analysis of Appendix A.1 predicts a total racy as coalescence ensures Taylor bubble formation.
void fraction hΦg i(0.658, 0) = 0.4193, and the ana-
lytic approximation (A.10) of Appendix A.2 predicts 5.4 SLUGGIT and the riser flow map
hΦg i(0.658, 0) = 0.4031. This constitutes a gas lift
riser head reduction of ∆H gl = 535.9 m and ∆H gl = The testing so far has raised the question of whether
515.1 m, respectively. Reducing the static head by the slug tracking methods are capable of predicting
40% will in most cases be enough to prevent the appropriate operational stability situation due to the
system from comming to a stand-still, even at rel- ‘lack’ of a dispersed flow regime (Assumption A.1.)
atively low well pressures. In the dayly operation Here then, the ability of SLUGGIT v.2 to reproduce
of the Girassol field, the gas lift flow rate is about the four types of stable and unstable riser flow dis-
150 kSm3/d = 1.41 kg/s – hΦg i(1.41, 0) = 0.59, or more cussed in Section 3.3 will quickly be tested. This
than half the static head at stand-still. study will aim to find the simulation flow map, but
only at a few points to give an overall impression. To
In Section 7.3, SLUGGIT simulation results of evaluate the results, experimental data published by
full-geometry simulations with a productivity index Taitel et al. will be used for comparison. The pipe
boundary condition are presented. The entire sys- diameter is d = 2.54 cm, the inclination is φ = 5◦ , the
tem coming to a stand-still is amongst the results length form pipeline inlet to riser inlet is L = 19.1 m,
encountered. Obviously, were gas lift to be included whereat gas is injected at the pipeline inlet itself and
in these simulations, the steady state outcome would liquid injected at z = 10 m downstream this inlet. Fig-
be severely affected. Probably, a system stand-still ure 5.12 shows the stability map for this geometry,
would not be amongst the simulation outcomes. where the data points are tabularise in the original
publication [41]. Those data points marked with dia-
Strictly speaking, Taylor bubble Assumption D.4 mond markers have been simulated using SLUGGIT
in the model derivation is not entirely valid in this v.2 without active slug initiation at the bend. Pres-
case as gas usually enter the riser base as tiny bub- sure time series from these simulations are shown in
bles which, driven by buoyancy, will propagate more Figure 5.13.
Figure 5.12: Flow map, L = 19.1, φ = 5◦ ,

d = 2.54 cm. Liquid injection at z = 10 m.
Source: [41]
Diamonds mark points simulated with SLUG-
GIT cf. Figure 5.13
S
Ug,o = {0.280, 0.185, 0.157, 0.245, 0.313}
UlS = {0.071, 0.066, 0.249, 0.527, 0.385}
Filled diamonds mark additional simulations
cf. Figure 5.14
The following can be said about these simulations from viewing their inlet pressure time series and their
PLOTIT [31] animations:
‘Steady flow’ point Cyclic flow pattern wherein the slug tail never penetrate far into the upstream pipeline
(equivalent of continuous gas penetration). Cycle appears equivalent dt that described in Cycle E, Sec-
tion 3.3, which is termed ‘Unstable oscillations’ in [40, 41, 14] and Figure 5.12. Snapshots of this cycle
55
"Steady flow" point; US
g,o
= 0.280, US
l
= 0.071
pi [bar]
1.2
1
50 60 70 80 90 100 110 120 130 140 150
"Unstable oscillation" point; US
g,o
= 0.185, US
l
= 0.066
pi [bar]
1.2
1
50 60 70 80 90 100 110 120 130 140 150
"Cyclic + fallback" point; US
g,o
= 0.157, US
l
= 0.249
pi [bar]
1.2
1
50 60 70 80 90 100 110 120 130 140 150
"Cyclic no falback" point; US
g,o
= 0.245, US
l
= 0.527
pi [bar]
1.2
1
50 60 70 80 90 100 110 120 130 140 150
g,o
= 0.313, US
l
= 0.385
pi [bar]
1.2
1
50 60 70 80 90 100 110 120 130 140 150
Time [s]
(a) SLUGGIT v.2

S S
"Steady flow" point; Ug,o = 0.280, Ul = 0.071
pi [bar]
1.2
1
50 60 70 80 90 100 110 120 130 140 150
S S
"Unstable oscillation" point; Ug,o = 0.185, Ul = 0.066
pi [bar]
1.2
1
50 60 70 80 90 100 110 120 130 140 150
"Cyclic + fallback" point; US
g,o
= 0.157, US
l
= 0.249
pi [bar]
1.2
1
50 60 70 80 90 100 110 120 130 140 150
g,o
= 0.245, US
l
= 0.527
pi [bar]
1.2
1
50 60 70 80 90 100 110 120 130 140 150
"Cyclic no falback" point; US = 0.313, US = 0.385
g,o l
pi [bar]
1.2
1
50 60 70 80 90 100 110 120 130 140 150
Time [s]
(b) SLUGGIT v.2s (simplified)
Figure 5.13: Pressure time series simulated with SLUGGIT v.2, MAX_BUB_COEF = 5, no active slug initiation
in bend. Flow conditions cf. Figure 5.12 – diamond markers.
56
are presented in Figure B.2, Appendix B.
‘Unstable oscillations’ point Unstable oscillations as described in Cycle E. Compared to the ‘Steady flow’
point, the gas penetration rate is less and the gas-liquid interface re-establishes itself at a lower height
after blow-out.
‘Cyclic + fall-back’ point Severe slugging (slug longer than riser at blow-out), as described in Cycle C. In-
terface re-established about midway up the riser after blow-out.
‘Cyclic no fall-back’ points Severe slugging with some fall-back. The fall-back is less than in the ‘Cyclic +
S
fall-back’ point, and smallest in the point Ug,o = 0.313 m/s, UlS = 0.385 m/s, which is visible in the pressure
time series by shorter oscillation periods and disappearance of the flat-topped profile.
Evidently, the lack of a non-intermittent dis- present work. On the contrary, it has proven to be im-
persed flow regime tend to make bubbly flow pat- mensely valuable (though time consuming) in reveal-
terns in the riser less stable. Consequently, the region ing unexpected shortcomings in the newest SLUG-
called ‘Unstable oscillations’ if Figure 5.12 extends GIT code versions. As the SLUGGIT v2/v2s code
further to contain a wide range of gas flow rates. [21] is being continuously developed by more than
Also a consequence is the region in which liquid once participant, its behaviour is altered with time,
fall-back takes place after the blow-out, even though though precise track of what changes are is easily
the experiments indicate no fall-back. This latter lost. The reasonable results presented so far stem
may be explained by too much liquid being cleared from simulations generated by a SLUGGIT code ver-
from the riser since gas can only travel in the form sion dated 19th of Match. When first carried out,
of Taylor bubbles which pushes more liquid ahead of later versions of the code were used which, surpris-
them than would smaller bubble formations. Adopt- ingly, gave significantly different results. A ver-
ing the simplified scheme SLUGGIT v.2s, as seen sion dated to the beginning of June managed to cor-
in Figure b does not affect the time series signifi- rectly give steady flow in the point Ug,oS
= 0.280 m/s,
cantly. Somewhat larger amplitudes, longer periods Ul = 0.071 /s, something the older version failed
S m
and fine-scale disturbances are here observed. to do. Unfortunately, this version also predicted
all other selected points to be stable. Even newer
The most important observation to take from versions again exaggerated the instability domains.
these brief tests is that is a word of warning. In These versions also compared less favourably the the
the later sections of this work the SLUGGIT pro- steady state model from Section 5.2.
gramme will be utilised in an attempt to understand Figure 5.15 show the Φl − p and Ṁl − p phase por-
operational instability in a real-life oilfield, yet it traits for the same MAX_BUB_COEF = 1 000 case, as in
has just been shown that this method may fail in Section 5.2, using the version from early June which
detecting some stable flow states at low liquid flow gave steady flow at every flow map point. The anima-
rates. Figure 5.14 give an impression of the further tion of this run looks quite the same, except for two
flow pattern development as the gas flow rate is in- elongated bubbles occupying the riser rather than just
creased form the ‘Steady flow’ point. The second the one. Also, it appears as if the slug is initiated at a
pressure moment is not reduced below 5% of the un- slightly different location near the riser inlet. Similar,
earated riser pressure (0.015 bar) until the gas velocity jet more diffused phase portraits are seen. Since two
is about twice that of the original data point. bubbles are active in the riser simultaneously, a cycle
consists of two ‘loops’ in the Φl − p portrait. Impor-
5.5 Comment tantly, the scales in these phase portraits are about
an order of magnitude less than those in Figure 5.8
The type of benchmark testing preformed in this sec- and 5.9, indicating the much lower fluctuations and
tion of the SLUGGIT v2/v2s codes may seem ex- higher resilience towards model instability.
tensive and somewhat diverging from the aims of the
57
Pressure standardS deviation vs. gas flow rate.
Ul 0.071 m/s
US = 0.280
1.2 0.035
g,o
1
50 55 60 65 70 75 80 85 90 95 100
0.03
1.2 0.025
S
Ug,o = 0.485
sp [bar]
1
0.02
50 55 60 65 70 75 80 85 90 95 100
0.015
1.2
US
g,o
= 0.809
0.01
1
50 55 60 65 70 75 80 85 90 95 100
0.005
1.2
US = 1.294
g,o 0
0.2 0.4 0.6 0.8 1 1.2 1.4
1
50 55 60 65 70 75
Time [s]
80 85 90 95 100
US [m/s]
g,o
(a) Pressure time series (b) Pressure standard deviation (second moment)
Figure 5.14: Attempt at obtaining steady flow by increasing gas flow rate (filled diamonds in Figure 5.12)
What causes these differences is version be- issue is strongly dependant on somewhat arbitrarily
haviour is, unfortunately, unknown. The alterations chosen management procedures.
are numerous and included changes in evaluation
of friction terms and management operations such To ensure compatibility, the code dated 19th of
as splitting and merging procedures. Despite time March, without any severe modifications, will be
consuming investigation, a precise understanding of used in the reminder of the present work as this ver-
what causes these differences has yet not been es- sion gave reasonable stability behaviour for riser sys-
tablished. Future investigation seems necessary. All tems.
investigation so far suggest that the model stability
0.1 1.5
45 45
1
40 40
0.05
35 35
0.5
0 30 30
Φl − hΦl i Ṁl − hṀl i

Time [h]
Time [h]
hΦl i 25 hṀl i 25
0
−0.05 20 20
−0.5
15 15
−0.1 10 10
−1
5 5
−0.15 −1.5
−0.03 −0.02 −0.01 0 0.01 0.02 0.03 −0.03 −0.02 −0.01 0 0.01 0.02 0.03
pi − hpi i pi − hpi i
hpi i hpi i
(a) Φl − p phase portrait (b) Ṁl − p phase portrait (c) Snapshot
Figure 5.15: Analogue to Figure 5.8a and 5.9a using newer version giving more steady solutions
6 Compromises
58
A great misfortune with respect to the slug track- computed in sections rather than borders and gas en-
ing codes available at EPT is that there codes as im- trainment in slugs. In future code development it is
plementations of similar schemes, yet implemented highly recommended by the author that the develop-
in totally separate and independent frameworks. This ment takes place in a unified framework, preferably
means that little time has been afforded to make each the v.2/v.2s code [21]. For these reasons, the v.2s
implementation robust, and the further development implementation will be studied in favour of the v.1
of one implementation will not benefit the others. code in the remainder of this work.
Another resulting disadvantage is that in exploiting
all methods, familiarisation with three vast codes In Section 5.1 it was shown that the LASSI code
rather than just one is required. [36] is presently not at a state which allows for analy-
sis of riser systems. The amount of work required to
At the present state of each code, the time re- remedy this problem (find the bug) is unknown. The
quired to preform the adjustments and ‘tinkering’ scheme was originally developed for smaller lab-
needed to perform both simulation and analysis for scale systems using fairly high grid resolution. Also
all versions is simply not available. Some compro- the LASSI code lacks many of the inlet, manage-
mise is therefore required. Presently, the SLUGGIT ment and automated timestep adjustment procedures
v.1 code [23] does not compile and is significantly available in SLUGGIT v.2/v.2s, making comparisons
less developed than the SLUGGIT v.2/v.2s code [21]. more challenging. For these reasons the study of the
Many of the special feature of the v.1 scheme is in- LASSI scheme will be confined to the theoretical de-
corporated into the simplified v.2s alternative pro- scription hitherto presented, keeping the scheme con-
cedure, as discussed in Section 4.2. The main re- cepts and ideas in mind during the continued analy-
maining differences are liquid velocities implicitly sis.
7 Field simulation results
The goal of this section is to gain insight into

what type(s) of instability(ies) causes to oscillations
observed in the P50 Girassol pipeline as the gas in-
jection rate is decreased below 100 kSm3/d (see Fig-
ure 1.3.) Simulation capability of the SLUGGIT
v.2/v.2s code [21] in deep-water cases is also tested.
Figure 7.1 shows the gas and liquid Ṁ − p phase por-
traits generated with the field data [43]. 20
20
This point is fraught with some controversy; the value of h Ṁl,i i = 33.7 kg/s corresponds to that reported by Zakarian and Larrey
in [47]. On the other hand, this is inconsistent with the supplied field data [43] from which all the pressure information originates.
Taking the average over the time period during which the gas lift rare was 70 kSm3/d, liquid mean volumetric flow rate is here found to
be 1 288 Sm3/d, which is about 14.9 kg/s, or less than half half what has been taken as the ‘proper’ liquid flow rate.
59
14.05.2004 07:50 −− 14.05.2004 17:50
0.5 10 7.1 Fixed inlet pressure – jumper influence
9
0.4
Khabibullin showed in [18] that the SINTEF multi-
8
0.3 phase dynamic pipe flow simulation code LedaFlow
7
show oscillatory pressure profiles for vertical riser
0.2
6
QST ST
g − hQg i systems that are dependent upon whether jumpers
Time [h]
hQST
g i
0.1 5
are included upstream with riser. The riser is here
0
4
similar to that of the Girassol P50 pipeline (see [47]
−0.1
3 and Section 1.2.1). A fixed pressure inlet boundary
2 condition (a von Neumann type boundary condition)
−0.2
1 is applied with inlet pressure is pinlet = 145 bar (a bit
−0.3
−0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15
0 higher than found in the field data, though there is
pi − hpi i
hpi i no gas lift in this simulation) and an outlet pressure
poutlet = 37 bar. The same dynamics as observed by
(a) QSg T − p phase portrait
14.05.2004 07:50 −− 14.05.2004 17:50 Khabibullin have here been recreated using SLUG-
0.5 10
GIT v.2.
9
0.4
8
0.3
7
0.2
6
Ṁl,i − hṀl,i i
Time [h]
hṀl,i i
0.1 5
4
0
3
−0.1
2
−0.2
1
−0.3 0
−0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15
pi − hpi i
hpi i
(b) Ṁl,i − p phase portrait
Figure 7.1: Girassol data. Wells on manifold P5011
Note that, contrasty to the flow rate phase por- (a) Jumper excluded (b) Jumper included
traits of Sections 5.2 and 7.2, which are defined at
the outlet, field data is only available near the inlet. Figure 7.2: Short open inlet with fixed out- and in-
let pressures poutlet = 37 bar, pinlet = 145 bar. Slight
As in the the pure terrain slugging phase portraits positive inclination of horizontal pipe upstream riser.
exemplified in Figure 3.5, the field data portraits take
on a circular form, spiralling anti-clockwise with Figures 7.2 present two snapshots – one with,
time. They do however also take on a diagonal and one without, a jumper afront the riser. A steady
trend, flow rates decreasing with increasing pressure. stream of bubbles and slugs are observed in the no-
Whereas the former suggests a terrain slugging dy- jumper case, and the pressure profile measured at the
namic, as seen in Section 3.3, the latter suggests an riser inlet is approximately constant. When a small
influence of the well characteristic; increasing pres- jumper was included ahead of the riser, short bursts
sure at the inlet will reduce well production. This lat- of gas occasionally released form the jumper, fol-
ter possibly indicates a void, concentration or surge lowed by streams of smaller bubbles, were observed.
wave phenomenon. These bursts appear to take place at regular intervals.
60
7.1.1 Jumper-caused pressure oscillations whole horizontal of the Girassol geometry was in-
cluded in simulations using SLUGGIT v.2. To avoid
Comment The analysis of this section, carried out
the system coming to a stand still, it was found nec-
at an early stage of this work, is sensitive to changes
essary to increase the inlet pressure to pinlet = 167 bar
in parameters and geometry, and also by the rather
so that both increased friction and the static head (no
large maximum section length applied. For this rea-
gas lift) could be overcome. The geometries used are
son the results themselves should be considered with
shown in figure 7.3 and the pressure time series at the
some reservations. Still, they do indicate that the U-
riser inlet in Figures 7.4. To demonstrate the influ-
bend in the P50 pipeline geometry may play a part in
ence of the jumpers and their placement, three cases
an upstream dynamic – a notion which is not to be
are presented: One with the jumpers placed approxi-
discredited.
mately as in the field geometry [43], one in which the
jumper placement in altered slightly, and one where
Simulation To avoid the fixed inlet pressure in- the jumpers have been removed altogether.
fluencing the upstream pressure measurements, the
Figure 7.3: Simplified Girassol geometries. Pressure profiles at riser inlet shown in Figure 7.4. Pipe diameter
rescaled by a factor 75
In the simulation where the jumpers are placed geometry in Figure 7.3.) Here the regular pressure-
in accordance with field geometry (Figure 7.4a,) the plunges continue throughout the time series. To
pressure occasionally plunges sharply with about demonstrate that the jumpers are sources to pres-
16 bar during the initial stages of simulation as gas sure disturbances, the pressure time traces when all
expulsions form the jumpers enter the inlet. These jumpers are removed are included in Figure 7.4c.
events occur on average at intervals of about half an As expected, this simulation show no sign of pro-
hour. After two hours these events abruptly cease nounced pressure fluctuation.
and a transient/annular steady state is reached. The
dynamic changes as the positions of the jumpers are
slightly altered, the jumper closest to the riser be- The manifold pressure time series form the field
ing moved further upstream (Figure 7.4b and middle data [43] as also included for comparison, contin-
uing the pursuit of Khabibullin in [18]. The simi-
61
170 190 160
180
160
150 SLUGGIT, no jumpers
170
150
SLUGGIT, jumpers at Girassol locations 140
160
140
150
pressure [bara]
pressure [bara]
pressure [bara]
130
130 140 SLUGGIT, jumpers slightly moved
120
130
120
120 110
110
110
100
100
100
Girassol field data transient Girassol field data transient
Girassol field data transient
90 90 90
0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9
time [hours] time [hours] time [hours]
(a) Jumpers placed according to geom- (b) Jumpers moved slightly (c) Jumpers removed
etry field data
Figure 7.4: Pressure time series at riser inlet. SLUGGIT simulations with simplified Girassol geometries
(Figure 7.3, top ↔ a, middle ↔ b, top ↔ c) fixed out- and inlet pressures poutlet = 37 bar, pinlet = 167 bar. Also
shown are field data [43] as gas lift is quickly reduced from 100 to 70 kSm3/d at t = 0.
larities are not convincing. The pressure series in More precisely, the jumpers has to be moved further
Figure 7.4b are amongst the few results managing to upstream, indicating that the field instability involve
obtain oscillations where both period and amplitude also some length of the horizontal pipeline.
are of the same order of magnitude, though pipeline
geometry manipulation was needed to achieve this.
Oscillation dynamics The simulation data from this open inlet jumper case have been studied using visual-
ization provided by the EPT PLOTIT visualization tool [31]. The jumpers, which are always filled with gas,
are observed to periodically go through the following stages:
Description A.
A.1 Separated flow through the jumper and pipeline.
A.2 The gas volume in the jumper begin to oscillate. Volume fluctuation amplitude and frequency increase
progressively in the jumper.
A.3 A small, additional amount of gas is released into the horizontal pipe from the jumper. This is immedi-
ately followed by a short slug and travels downstream the horizontal as a density wave.
A.4 As the additional gas reaches the riser and begins to expand, the oscillations in the jumper stabilize and
the cycle repeat itself.
It is also found that the cycle in a single-jumper system is probably quite regular. The simulation which
gave the pressure trace in Figure 7.4b has three jumpers. These influence each others’ cycles and cause the
observed disregularity.
In Section 1.2.2, a suspected causality between mations that the activity in the jumpers become less
jumper gas release and riser expansions was men- vigilant as the void wave progresses up the riser (see
tioned. The idea was that a gas expulsion from the accompanying animations.) The animations further
jumper would travel up the riser and reduce the hy- unveils that a long slug forms ahead of the void wave
drostatic head, causing upstream expansion and fur- in the riser. When this slug reaches U-bend near the
ther gas expulsion. Instead, it appears form the ani- outlet, U-bend terrain slugging will cease. As U-
62
bend slugging stops, also the high-frequency density esting jumper-riser-U-bend negative feedback cou-
oscillation seen in the jumper form which the void pling – a possible cycle of causality is hypothesised:
wave originated cease. This may suggest and inter-
Description B.
B.1 From steady state, terrain slugging initiate and progress in intensity and frequency in the upper U-bend.
B.2 This produces oscillations in the jumper void through pressure fluctuations propagating upstream. These
oscillations allow for irregular gas entrainment in the jumper.
B.3 Pressure oscillations, and therefore jumper gas volume oscillations, progress in relation with the U-bend
dynamics until a gas surplus is released into the pipeline form the jumper.
B.4 When this void wave reaches the riser, the disturbance is sufficient to initiate a slug formation quickly
travelling up the riser.
B.5 Once the slug front (propagating much faster than the bubble noes) reaches the outlet, the U-bend is
flooded and U-bend oscillations cease.
B.6 In turn, the pressure fluctuations will come to and end and the jumper volume will steady with little
entrapped gas and at relatively higher static riser head.
B.7 As the slug is expelled, a separated regime is again resurrected and slugging flow may begin to developed
in the U-bend. The cycle is repeated.
This presents an interesting prospect where oscillatory phenomena are coupled across approximately 1 500
meters of piping by means of pressure propagation upstream and void wave advection downstream. Figure 7.5
presents two illustrative snapshots.
(a) Jumper oscillations possibly caused by U- (b) U-bend slugging ceased by slug formation in
bend produced pressure fluctuations front of void wave disturbance
Figure 7.5: Open inlet, jumper-riser-U-bend oscillations
63
Figure 7.6 presents three hold-up time series – flow regime should occur at the middle jumper.
one at the jumper closest the the riser inlet, one at
the riser inlet itself, and one at the U-bend, at a point
where the inclinations is positive and slugging oc- 7.1.2 Expansion driven unstable production
curs. Two hours of the eight hour simulation time
is depicted, during which no void wave travels form From here, the simple representation of the U-bend
any of the two jumpers further upstream. It is clear from the previous Section is replaced by a more de-
that as soon as the gas enters the riser base (negative tailed U-bend representation of the Girassol P50 pipe
spikes in the middle graph) periods of calmer jumper using field geometry data [43]. The well piping is
hold-up lasting about 10 minutes can be observed still not included, but will be studied in Section 7.3.
(top graph). As this happens, a quickly expelled slug Figure 7.8 display the geometry. The horizontal sec-
floods the riser and U-bend (visible in the bottom tion is retained somewhat simplified (pipe segments
graph). The expulsion of this riser slug manifest it- of φ = 0◦ and φ = 0.4◦ inclination where field geome-
self as a large drop in hold-up in the U-bend as the try has slightly negative inclination.)
expanded gas emitted form the jumper reaches the
outlet. These downward spikes in U-bend hold-up Quite accidentally, expansion driven instability,
(bottom) infallibly marks the re-initiating of oscilla- as described in Cycle A, is here stumbled upon.
tions in the upstream jumper (top). This is consistent The best way to understand the predicted dynamic
with the hypothesis, but does not prove any causality. is by viewing an animation of the simulation, and so
the reader is encourage to view the accompanying
PLOTIT output files. Also helpful are the snapshots
Naturally, simulations without any bend atop the in Figure B.1, Appendix B
riser have been performed to strengthen or kill the
aforementioned hypothesis. Figure 7.7 shows the Pressure time traces are shown in Figure 7.9.
found steady state solution. These pressure series are similar to those fount for
the simplified U-bend (Figure 7.7b), but the mecha-
This simulation quickly reaches a steady state nism responsible for the pressure fluctuations appear
solution of stratified flow from the inlet to the second to be different when viewing the animations.
jumper, and bubble-slug flow from there to the outlet. In the case of the geometry where the jumper po-
Pressure fluctuations at the riser base (Figure 7.7b) sitions are slightly altered (the last jumper is moved
are on order of two bars – the flow is considered upstream), the cycle is more regular as jumper place-
stable. The horizontal in Figure 7.7a has zero incli- ment coincide better with bubble positions at the ex-
nation all over, so it is interesting that the change of pansion event.
Description C. Expansion driven unstable production

C.1 No production occurs. Gas is entrapped in and around the jumpers. There are a couple of bubbles at the
bottom of the riser. The liquid column surface in the riser is slightly too low to reach the outlet U-bend.
C.2 As the bubbles travel up the riser, their size increase, riser void increases, bottomhole hydrostatic pressure
is reduced and the liquid hight in the riser increases. When the bubbles are about half-way up the riser,
the riser liquid height is heigh enough for the liquid to reach the outlet.
C.3 The liquid production increases the liquid flow, transporting gas from the rightmost jumper into the riser,
replacing the riser inlet bubble in Step C.1. This bubble enters the riser about the same time as the
expanding, liquid producing bubble exits the riser and the riser again floods with a liquid column surface
below the riser outlet.
C.4 Production is sustained as gas escaping a jumper is replaced by gas travailing from the upstream jumper
neighbour.
64
Jumper
1
Hold−up
0.5
0
4 4.2 4.4 4.6 4.8 5 5.2 5.4 5.6 5.8 6
Riser base
Hold−up
0.9
0.8
0.7
4 4.2 4.4 4.6 4.8 5 5.2 5.4 5.6 5.8 6

U−bend
1
Hold−up
0.5
0
4 4.2 4.4 4.6 4.8 5 5.2 5.4 5.6 5.8 6
Time [h]
Figure 7.6: Hold-up at last jumper (top), riser base (middle) and slugging portion of U-bend (bottom)
160
150
140
130
Pressure [bar]
120
110
100
90
80
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.8 2
Time [hours]
(a) Geometry (b) Short pressure time series.
Figure 7.7: Open inlet. Geometry equivalent to that in Figure 7.4b without U-bend atop the riser.
65
(a) Jumpers slightly displaced
(b) Jumpers placed according to field data
Figure 7.8: Simplified Girassol geometry with complete U-bend
Figure 7.9: Pressure time series at riser base. Top graph ⇔ geometry Figure 7.8a, bottom graph ⇔ geometry
Figure 7.8b
66
See Figure B.1b of animation snapshots to get up the riser a distance far enough to expand suffi-
a visual impression of the dynamic. The period of ciently. This expansion volume equals the volume
steady riser pressure seen in the top graph of Fig- in between the riser outlet and an unaerated liquid
ure 7.9 happens as there is no liquid production (the column surface. As is evident from top plot graph
liquid does not reach the top of the riser) and the in Figure 7.9, this time interval is fairly constant and
amount of liquid in the riser remains unchanged. periodic. Variations in duration of steady pressure
This lasts for about 30 minutes, which is not the time are caused by uneven amounts of gas being left in
needed for the base bubbles to travel all the way up the riser base after a cycle.
the riser, but the time needed for the bubbles to travel
In the case where the jumper is placed close to the riser (Figure 7.8b) the dynamic can be described as
follows:
Description D.
D.1 Liquid blocks the riser inlet. The jumper and piping upstream the riser contains gas that trickles forward.
D.2 The gas trickles into the riser. At some point, the trickle is large enough to suck the entire gas amount
entrapped in the last jumper into the riser. The system may stay idle for hours before this happens.
D.3 As the gas expands, the liquid production increases and gas is pulled from the inlet. Gas present in the
horizontal is pulled further downstream.
D.4 Depending on the amount of released gas and the momentum of the liquid, gas pulled from upstream
regions reach the riser inlet before the gas in the riser may reach the outlet. If it does, the liquid production
continues by means of the buoyancy and expansion provided by the newly arrived gas.
D.5 Eventually, new gas will not reach the riser inlet. Instead, the gas transported from upstream the riser
will reach the final jumper and accumulate a gas storage here. The cycle is repeated.
– see Figure B.1a.
The duration of system idleness is here quite un- possible. In the present case, the inlet pressure was
steady and unpredictable. From the bottom plot of pi = 167 bar. Should however the inlet pressure be
Figure 7.9 it is seen that this period can last as long higher than this then the pressure inlet-outlet differ-
as three hours, all depending on the tiny movement ence is high enough to allow for production even with
necessary to transport a sufficient amount of gas the no void in the riser. The dynamic then will be a fast
short distance from jumper to riser inlet. Basically, flow of slugs and bubbles, as previously seen in the
the main jumper-caused difference between the be- beginning of this section. Though this may itself be
haviour seen in Descriptions C and D is that the subject to density wave instability and other forms of
jumper placement does not correspond equally well instability, the dynamic presented in Description D
with bubble movement for replacement of the riser and C cannot be replicated.
inlet bubble to be guaranteed – Step C.3 and D.4. In reducing the inlet pressure too far below the
static head of a pure-liquid riser column, the void
Some important features of these simulations produced by gas expanding while travelling up the
should be underlined: riser will not increase the riser liquid surface height
sufficiently to reach the outlet. In this case, no liquid
Inlet pressure In a case study such as this, with production will take place at all, the gas will leave
fixed inlet pressure, there is only a narrow interval of the riser, and the system will settle to a stand-still.
inlet pressures within which this type of instability is The upper snapshots of Figures B.1a and b shows
67
how little additional head is needed on the inlet to into the dynamic. Compared with the open inlet,
drive the system into continuous production. fixed pressure boundary condition, the closed inlet
promotes a quite different system response.
Inclination of horizontal pipeline Attempts have
been made to reproduce the type on instability de- Again, the Girassol field P50 pipeline will be in
scribed above using the exact Girassol P50 pipeline focus as a test case. The P50 pipeline is perhaps
geometry. This has proved impossible. The reason a more natural test case choice for the closed inlet
is that the longest segments of piping on the P50 boundary as the combination of compressibility and
pipeline have an inclination of φ ≈ −0.8◦ . Negative downward-sloping terrain produce a setting prone to
inclination makes the pipeline prone to terrain slug- terrain slugging. Terrain slugging has already been
ging, but buoyancy constantly works to prevent new discussed introductorily in Section 3.3 and has been
gas from entering the riser. From simulations it was past of benchmark testing in Section 5.4. As be-
found that some promising expansion slugging could fore, the well is excluded from geometry. This time,
be seen at the initial stages of the simulations, as gas the exact horizontal and riser geometry is used [43],
was pulled into the riser by the momentum of the liq- which is shown in Figure 7.10 – notice the downward
uid. At some point however, the liquid will come to inclination of the horizontal, which is φ = −0.74 up-
rest and the gas may begin to evacuate up towards stream the two jumpers, and φ = −0.83 downstream.
the inlet. Again then, if the inlet head is less than
the riser height the system will come to a stand still. Mass sources are added close to the inlet; at
Alternatively, were will be continuous production in z = 1.0 m a gas source pumping Ṁgs = 1.69 kg/s is
which buoyancy forces are does not drive the pro- placed, while a liquid source Ṁls = 33.7 kg/s is placed
duction (buoyancy may on the other hand still affect at z = 250 m. These sources ensure mass fluxes corre-
operational stability.) In other words, since periods sponding the average well production measurements
of no liquid production is a part of the expansion published by Zakarian and Larrey in [47]. The pres-
driven unstable production seen in this section, the sure response time series at the riser base is presented
system must be able to endure periods of near stand- in Figure 7.11
still while allowing gas to enter the riser.
7.2 Closed inlet, fixed flow rates – Terrain 200

slugging 180
160
Inlet pressure [bara]
140
120
100
80
Figure 7.10: Closed inlet geometry
60 Simulation
Field measurement
The fixed flow rates inlet boundary condition is avail- 40
0 2 4 6 8
able in all EPT slug tracking codes. Here, all mass Time [h]
fluxes through the inlet are fixed, either by fixing
S
the inlet superficial velocities Uk,i and pressure (den- Figure 7.11: Pressure time series upstream riser
sity), or by denying any flow through the inlet and from P50 pipeline geometry. Closed inlet, constant
fixing mass sources somewhere in the pipeline. This mass sources Ṁgs = 1.69 kg/s, Ṁls = 33.7 kg/s. SLUGGIT
is a Dirichlet type boundary condition that forces v2 simulation vs. Girassol manifold measurements
changes in gas density, accommodating elasticity [43, 18, 47].
68
Again, in order to draw a parallel to Ṁg = 0.295 kg/s, which correspond to a pressure
Khabibullin’s project thesis work [18], the pressure pi = 122 bar and temperature T i = 63.5 ◦ C. This
time trace is plotted against the Girassol P50 mani- amounts to a gas mas flow rate less than a fifth of
fold pressure measurements just after the gas lift has the original inlet condition.
been reduced form 100 to 70 kSm3/d. A description of
the dynamic is here not necessary; a textbook case of Unsurprisingly, the character of the flow now
severe slugging plays out, and is already described in changes significantly. The horizontal is now mostly
Cycle C in Section 3.3. Snapshots of animation can filled with liquid, the gas forming Taylor bubbles in
be found in Figure B.3 in Appendix B. a liquid filled horizontal rather than a stratified re-
gion stretching from inlet to the bend-blocking slug.
The severe slugging result can be seen in connec- The temporally and spatially averaged liquid hold-
tion with the theory of Taitel from Section 3.3. Crite- up in the horizontal is found to be hΘl i = 0.84. In
rion (3.2), telling whether riser is table, can easily be comparison, the same average in the previous case,
checked. Approximating the properties of (3.2) from with Ṁgs = 1.69 kg/s and severe slugging, was found
the plots available through PLOTIT [31], it is found to be hΘl i = 0.60. This means that, with the new
that L ≈ 4600 m, H ≈ 1300 m, hΘi ≈ 0.5, Po = 37 bar, gas flow rate, the horizontal compressible volume is
hΦg i = 0 and α0g was found to be of the order 0.9 about halved. Upstream compressibility is a key in-
in [41]. This makes the RHS of (3.2) equal 125 bar, gredient in the terrain slugging cycle C and reducing
which is significantly more than Po , confirming riser this will reduce the severity of the slugging.
instability. With the new reduced gas injection, liquid pro-
Likewise, the Bøe criteria (3.1) may be checked, duction is uninterrupted. The riser is mostly filled
using the re-written formulation (5.2). This produce with un-aerated liquid, occasionally disturbed by a
a RHS of 7.93 kg/s – the necessary criterion for terrain Taylor bubble penetrating from the horizontal. Total
slugging type I is satisfied. Using the alternative to average liquid riser hold-up is hΦl i = 0.95
the Bøe criterion from [32] give than same result.
200
Figure 7.12 show the riser phase portraits of the 180

severe slugging simulation. The portraits show dis-
160
tinct resemblance to the portraits in Figures 5.8a and

5.9a in Section 5.2 where oscillations are caused by 140
coarse resolution.
120
100
Adjusted gas injection rate
80
During the final weeks of work the present master
60 Simulation
thesis it was found, through communication with
Field measurement
Eldar Khabibullin, that there has been a misun- 40
0 2 4 6 8
derstanding concerning the input data used in the Time [h]
above considerations. It would appear that the flow
rates supplied ( Ṁoil = 14.36 kg/s, ṀH2 O = 19.33 kg/s, Figure 7.13: Pressure upstream the riser from P50
Ṁg = 1.69 kg/s) are in fact the phase flow rates the fluid pipeline geometry (Figure 7.10). Closed inlet, con-
mixture would obtain under standard condition pres- stant mass sources Ṁgs = 0.295 kg/s, Ṁls = 35.55 kg/s.
sure and temperature, not under the inlet conditions. SLUGGIT v.2 simulation vs. Girassol manifold mea-
In other words, these flow rates are not adjusted surements.
for phase transition associated with the state change
from standard to inlet conditions. The appropriate A pressure time series of a simulation performed
inlet mass flow rates as recently supplied by Eldar using these new inlet boundary conditions is pre-
Khabibullin are Ṁoil = 15.96 kg/s, ṀH2 O = 19.59 kg/s, sented in Figure 7.13. The pressure fluctuations have
69
30
0.2
7 7
25
0.1
6 6
20
0
Ṁl,o − hṀl,o i 15 5 Φl − hΦl i −0.1 5
Time [h]
Time [h]
hṀl,o i hΦl i
−0.2
10
4 4
−0.3
5
3 3
−0.4
0
2 −0.5 2
−5
−0.4 −0.3 −0.2 −0.1 0 0.1 0.2 0.3 −0.3 −0.25 −0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15 0.2
hpi i hpi i
(a) Ṁl,o − p phase portrait (b) Φl − p phase portrait
Figure 7.12: Closed inlet with fixed mass sources Ṁgs = 1.69 kg/s, Ṁls = 33.7 kg/s. Geometry as in Figure 7.10
12 0.05
8 8
10
0
7 7
8
6 −0.05 6
Ṁl,o − hṀl,o i 6 Φl − hΦl i
Time [h]
Time [h]
hṀl,o i hΦl i
5 5
4 −0.1
4 4
2
−0.15
3 3
2 2
−0.2
−2 −0.1 −0.05 0 0.05
−0.15 −0.1 −0.05 0 0.05 0.1
hpi i hpi i
Figure 7.14: Closed inlet with fixed mass sources Ṁgs = 0.295 kg/s, Ṁls = 35.55 kg/s. Geometry as in Figure 7.10
70
a period of about 10 min and peak-to-peak amplitude contrary to what was seen with the U-bend in Fig-
around 20 bar. These fluctuations are again ladened ure 7.14b, and confirms that the divergence from a
with more noise of higher frequency and smaller straight line riser void/upstream pressure relationship
smaller amplitude. in the former simulation was caused by slugging in
Figure 7.14 present the phase portraits of the adjusted the U-bend. This, of course, makes sense since pres-
SLUGGIT simulation. Though some trends may be sure differences across an downstream slug is not ac-
discearnable, it is clear that these portraits does not counted for in the riser liquid column.
show patterns in the way seen in the severe slugging
simulation, Figure 7.12. It is quite interesting to note 200
that even under these flow conditions, fluctuations 180
with amplitude in the order of those measured at the
Girassol field are seen. The oscillations in pressure 160

are formed by gas occasionally penetrating into the 140
riser and expanding therein. PLOTIT animation [31]
120
further reveals that the Tailor bubbles in the horizon-
tal tends to ‘shake’ back and forth as they slowly 100
propagate downstream. This happens as bubble tails 80
are temporarily turned into bubble noses so that the
bubbles move a short distance upstream. Interest- 60 Simulation
Field measurement
ingly, the temporary changes in bubble propagation 40
0 2 4 6 8 10
direction appears to be caused by the disturbances Time [h]
generated as gas slugs in the U-bend atop the riser. It
is difficult to tell whether this dynamic has influence Figure 7.15: Pressure upstream riser from P50
on the pressure fluctuations, as well as whether it is pipeline geometry – as in Figure 7.10, only U-
physically appropriate or caused by the bubble nose bend of pipeline geometry replaced with a horizontal
criteria not being entirely appropriate when bubbles pipe segment. Closed inlet, constant mass sources
are nearly stationary in horizontal pipes. From the Ṁgs = 0.295 kg/s, Ṁls = 35.55 kg/s. SLUGGIT v.2 simu-
basic description of the SLIGGUT v.2/v.2s bubble lation vs. Girassol manifold measurements.
nose logic, Section 4.9, it should be evident that cap-
turing correct bubble propagation behaviour in near Simulations performed by Eldar Khabibullin us-
horizontal pipes can be somewhat tricky. ing SINTEF’s LedaFlow model and the commer-
cially popular OLGA code predicts the same high-
To investigate to what degree the pressure dis- frequency fluctuations as seen Figure 7.13, only
turbances are caused by the bubble-riser penetration these fluctuations were slightly more regular [18].
effect or terrain slugging in the U-bend, a new sim- These simulations included a gas lift injection rate
ulation is carried out. This time, the U-bend atop of 70 kSm3/d. Predicted fluctuation period of the
the riser replaced with a 400 m long perfectly hori- LedaFlow simulation was 5 min with a peak-to-peak
zontal pipe segment. All other parameters remains amplitude of about 11 bar. OLGA predicted sim-
as in the previous simulation. The result is shown ilar fluctuations of 10 bar peak-to-peak amplitude
in Figure 7.15. A part of the fluctuation has disap- and 24 min periods. The oscillations from OLGA
peared and only the less frequent bubble-riser en- were composed of steady double-pulse signal with
trance events remain. The amplitude of fluctuation a smaller pulsation followed by a larger one, sim-
is here found to be slightly increased, possibly ex- ilar to a heart beat. Compared with the 3 hour pe-
plained by the Taylor bubbles being larger since there riod and 35 bar amplitude of the field data [43], Fig-
is no flow disturbance from the U-bend. The corre- ure 1.3, these simulations does not appear to cap-
sponding Φl − p phase portrait shown in Figure 7.16b ture the appropriate instability phenomena. 35 bar is
form nearly a straight line, which is consistent with the fluctuation amplitude reached before the gas lift
a purely hydrostatic system consideration. This is injection rate was restored – it is unknown to what
71
8
8 8
7 0
6 7 7
−0.05
5
6 −0.1 6
4
Ṁl,o − hṀl,o i Φl − hΦl i
Time [h]
Time [h]
hṀl,o i hΦl i −0.15
3 5 5
2 −0.2
4 4
1
−0.25
3 3
0
−0.3
−1
2 2
−0.35
−2 −0.25 −0.2 −0.15 −0.1 −0.05 0 0.05
−0.25 −0.2 −0.15 −0.1 −0.05 0 0.05 0.1
hpi i hpi i
Figure 7.16: Closed inlet with fixed mass sources Ṁgs = 0.295 kg/s, Ṁls = 35.55 kg/s. Geometry as in Figure 7.10,
only U-bend is replaced with a horizontal pipe segment
amounts these fluctuations would grow if given time, 7.3 Full Girassol geometry with well
nor what would happen if gas lift was eliminated al-
together. In fact, the slugging high-frequency fluctu-
ations found whit LedaFlow were found to be a con- Figure 7.17 depicts the full Girassol geometry in-
sequence of predicted terrain slugging in the U-bend cluding the sub-seabed well pipe sections.
at the pipeline outlet, and not a riser slugging phe-
nomenon [18]. With SLUGGIT v.2, the same U-bend
terrain slugging is found, but also Tailor bubbles pen-
etrating into the riser appears to generate pressure
disturbances. Most likely it is these latter pressure
disturbances who case the SLUGGIT fluctuations to
be less regular that that of LedaFlow and OLGA.
LedaFlow further predicted bubbly flow in the

horizontal and slugging flow in the riser.21 Bubbly
flow, though intermittent on the small scale within Figure 7.17: Full Girassol pipe geometry with well
the flow itself, physically adopts a continuous char-
acter in cross-sectionally averaged properties. A
steady and continuous transport of gas into the riser
is therefore possible with simulation tools including During the present work a linear pressure-flow
dispersed flow regimes. No dispersed flow patterns rate inlet coupling has been implemented similar to
are (presently) available in SLUGGIT, these instead that used in the commercial multiphase code OLGA
manifesting themselves as bubble-slug flow. Coales- [12]. Details of this performance index implementa-
cence thereby tend to form all gas in the horizontal tion can be found in Appendix C. The productivity
into elongated bubbles, distributing the gas discon- index was implemented in order to increase the sys-
tinuously and irregularly along the horizontal. These tem similarity to the real Girassol field and compa-
will disturb the pressure when entering the riser. rability to the simulation results of collaborators who
work with other models (see Section 1.2.3 and 1.2.1.)
21
LedaFlow animations supplied by Khabibullin.
72
7.3.1 SLUGGIT v.2 criteria will therefore be tested: The Slug Stability
(Bendiksen) (SSB) criterion and the Invicid Kelvin-
Three simulations will here be described. All have
Helmholtz (IKH) criterion. The former is a criterion
the an inlet productivity index K = 210 Sm3/d·bar,
seen from the slugging regime, allowing slug ini-
a well pressure Pwell = 252 bar, Gas-Oil Ratio
tiation when the conditions are such that slugging
GOR∗ = 48.316 Sm3/m3 and Water Cut wc∗ = 0 m3/m3
flow can be sustained. This is simply achieved by
[47, 18]. The GOR and water cut are here adjusted
initiating slugs when the gas velocity exceeds what
from the field values of GOR = 103 and wc = 0.531
would be the bubble nose velocity vb,n from Equa-
to approximate both oil and water as a single liquid
tion (4.19). The latter of these criteria is seen form
phase.22 Because it was found in Section 5.2 that al-
the stratified region, and concerns when stratified
lowing bubble sections larger than 200 times the pipe
flow can no longer fluid-dynamically be sustained.
diameter may cause significant ‘numerical numerical
It is presented in Annex E. The inviscid version is
terrain slugging’, 200 will be the above limit for the
chosen in order to reduce the computational expense
MAX_BUB_COEF parameter in this section.
of checking the criteria at each section, every time
step. For further details on hydrodynamic slugging
The relative velocities under the present flow
criteria, see e.g. [25, p. 66-74].
conditions are thought to be sufficiently high to
make hydrodynamic slug creation a possibly sig-
nificant influence. Two hydrodynamic slug initiation Two simulation cases are here given attention:
Case I: MAX_BUB/SLUG_COEF = 200. No modelled hydrodynamic slug initiation.
Description:
t < 0.3 h: Initial transient phase where the pipeline is filled by a stratified liquid hold-up front propagates
through the pipeline.
0.3 < t < 0.5: Slugging initiated from the first jumper. Liquid is transported through slugs up the riser.
0.5 < t < 0.8: Slugging from jumpers cease (for some reason.) Separated flow throughout system with-
out liquid production. Liquid builds up in riser as riser hold-up increases.
0.8 < t: Instantaneous formation of a slug in riser and well, blocking both with a full, un-aerated liquid
column. The static head rises abruptly at the inlet and becomes larger than the well pressure ( pi >
Pwell .) As a result, the inflow through the inlet shuts down and the system comes to a stand still (no
backflow through the inlet is allowed, as in OLGA [12].)
Case II: MAX_BUB/SLUG_COEF = 200. SSB modelled hydrodynamic slug initiation is active.
Description:
t < 0.25 h: Initial phase. Hydrodynamic slug initiation at beginning of the horizontal pipeline. The
system is filled with liquid by a propagating slug-bubble front.
0.25 < t: Steady state production with slugging flow in the horizontal, riser and U-bend, and annular
flow in the well. The slug-to-bubble length ratio in the horizontal is on average in the order of 0.1.
Consequently, the point at which an increase in well production is balanced by the pressure drop
due to friction is relatively high (high production).
22
Even thought three phase simulations are supported in SLUGGIT, this is outside the scope of the present work.
73
z = 200m, D = 15.7cm z = 200m, D = 15.7cm
270 10 220 10
260
pressure
Superficial elocity uS [m/s]
Superficial elocity u [m/s]

250 uS
Inlet pressure p [bara]

l pressure
210
S
240 uS uS
g l
230 5 uS 5
g
220
200
210
200
190 0 190 0
0 0.5 1 1.5 0 0.1 0.2 0.3 0.4 0.5
Time [h] Time [h]
(a) Case I (b) Case II
Figure 7.18: SLUGGIT simulated pressure and phase velocities using the performance index inlet boundary
conditions. Time trace taken close to well inlet.
The velocity and pressure time traces close to the blocks the riser. This is because the well pressure is
well inlet are shown for cases I and II in Figure 7.18. less than the maximum static head in the Girassol ge-
A simulation similar to case II has also been carried ometry, and the SLUGGIT code lacs gas lift support.
out using the IKH-criterion (E.3) to actively initiate The event causing such a blockage usually takes the
hydrodynamic slugs. The same dynamic behaviour form of a ‘collapse’ in the riser; if the liquid frac-
as seen in case II was then predicted, only slugs were tion in an annular or lightly slugging vertical flow
more frequently initiated for then to be destroyed regime is allowed to built up, the upstream pressure
shortly after due to decreasing slug lengths. Conse- will increase and there may come a point where liq-
quently, the net difference in choosing the IKH crite- uid hold-up locally reaches unity. As this happens,
rion was larger computational expense. the blockage causes a sudden flow retardation and
For comparison with the mass source boundary fall-back throughout the riser, turning all riser liquid
conditions supplied by Eldar Khabibullin and ap- into a long, un-aerated slug nearly instantaneously.
plied in Section 7.2, the steady-state average mass The large hydrostatic pressure increase cased by this
flow rates are computed from the case II simula- event occurring in the riser immediately causes the
tion. The average liquid production was found to same event in the well, and to the hydrostatic pres-
be h Ṁl,o i = 56.9 kg/s, compared with Ṁl,i = 33.7 kg/s as sure suddenly increases beyond the well pressure.
supplied by Khabibullin.20 For the gas phase mass
flow rate is found to be h Ṁg,o i = 1.97 kg/s, contrary Alternatively, the system always manages to
to the supplied boundary condition h Ṁg,i i = 1.69 kg/s. clear enough liquid to avoid liquid build-up and
Since phase transition is not presently included in the blockage. In this situation, a positive feedback takes
SLUGGIT code, this steady-state flow rate is con- place during the initial stages; a reduction in riser
stant throughout the pipeline (more on this later.) and well liquid reduces the hydrostatic head, which
Naturally, the stand-still in case I results in no pro- in turn increases the well production. In the case
duction. studied here, increasing well production reduces the
It would appear that, depending on minor alter- overall system liquid fraction, reducing the inlet head
ations on the simulation parameters, the predicted even further. At some point, the inlet pressure will
steady state solutions take on one of two opposite be dominated by frictional rather than gravitational
extremes. Either a complete stand-still solution is forces, forming a high-velocity equilibrium point.
found if, at any point, a large enough slug blocks Compensating for gas lift by e.g., reducing riser and
74
well length has the effect of eliminating the former is assumed to be zero. This may in part explain the
flow solution, all predictions becoming the high- difference in prediction.
velocity ones.
It is also interesting to investigate whether finer
These solutions are contrary to solutions found resolution affects the results. A simulation with
when computing the same case with LedaFlow or MAX_BUB/SLUG_COEF = 100 is carried out without
OLGA [18]. The solutions found using these codes hydrodynamic slug initiation, along with the exag-
are steady, but with less void in the system than gerated case of MAX_BUB/SLUG_COEF = 1 000. It is
the high velocity solutions found with SLUGGIT.23 hoped that the high relative phase velocities possible
These simulations predict bubbly flow regimes in the in this case study may be sufficient to produce some
long horizontal and the annular regime in the riser. natural hydrodynamic slug capturing, though obvi-
Bubbly and droplet flow regimes are not yet imple- ously not sufficient to be undampened.
mented in SLUGGIT v.2, and the void in all slugs
Case III: MAX_BUB/SLUG_COEF = 50. No modelled hydrodynamic slug initiation.
Description:
t < 0.27 h: Again, initial transient phase where the pipeline is filled by a stratified liquid hold-up front
propagates through the pipeline. Slug initiates form first jumper.
0.27 < t < 0.4: Formation of small slugs in the riser. These are blown out. All slug generation ceases.
0.4 < t < 0.5: All-over separated flow. A surge wave of high hold-up propagates up the riser.
0.5 < t < 0.6: Annular flow in the riser collapses and the riser is flooded with un-aerated liquid. The
wave front has by then reached about 500 m up the riser.
0.6 < t Annular flow regime in the well collapse. pi > Pwell and the system comes to a stand-still.
Case IV: MAX_BUB/SLUG_COEF = 1 000. No modelled hydrodynamic slug initiation.
Description:
t < 0.3 h: Initial phase. Separated flow throughout entire system. Vaguely increasing liquid hold-up in
riser and well.
0.3 < t: Equilibrium obtained. Steady state production without any slug formation.
Pressure and superficial phase velocity time case VI.

traces are shown in Figures 7.19. Obviously, The gas and liquid production rates in case VI are
MAX_BUB/SLUG_COEF = 50 is not a sufficiently fine found to be h Ṁg,o i = 0.994 kg/s and h Ṁl,o i = 28.7 kg/s,
resolution to obtain any form of hydrodynamic slug respectively. Compared with case II, the flow rates
capturing. This should perhaps not be surprising as are a little less.
Issa, Kempf and Bonizzi [7, 13] found that a ratio
`/D ≥ 0.4 would be necessary for capturing to be rea-
sonably unaffected by further refinement. What is In cases I and III the flow remained stratified
interesting, however, is that a separated flow steady- some time before a collapse occurred in the riser,
state solution was found for the coarsely resolved while in case VI the flow obtained a steady-state so-
lution with stratified flow throughout the horizontal.
23
Private communication with Eldar Khabibullin.
75
z = 200m, D = 15.7cm z = 200m, D = 15.7cm
260 10 240 10
250
pressure 230

uS

240 l
S
uS pressure
g 220
230
5 uS 5
l
220
210 uS
g
210
200
200
190 0 190 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Time [h] Time [h]
(a) Case III (b) Case VI
Figure 7.19: SLUGGIT v.2 simulated pressure and phase velocities using performance index inlet boundary
conditions. Time series taken close to well inlet.
Due to the high relative velocities in these cases, one (see Section 4.11.2). However, due to numerical dif-
would expect a non-stratified regime in the physical fusion, they seldom manifest themselves in discre-
flow. Figures 7.20 and 7.21 display the left and right tised sachems unless grid resolution is exceptionally
hand side of the inviscid Kelvin-Helmholtz criterion high24 [13]. Consequently, the ill-posedenss (of the
(E.3) from Annex E at t = 0.25 h and t = 1.0 h from underlying continuous two-fluid model) remain un-
cases III and VI, respectively. The tricky term dα l
dhl detected and its influence, if any, unknown. Case
is handled by a geometric consideration of the sketch VI, where the ‘grid’ is very coarse indeed, may be
in Figure 4.3, where the interface angle γ is found by a prime example of this, wherein no symptoms are
solving 2π1 (γ − sin γ) − αl = 0 in MATLAB R
using the visible, yet the continuous two-fluid model would be
sin γ2 is
dα
built-in iterative function fzero. dhl = πD
l 4
ill-posed and the physical solution should, according
then obtained. Alternatively, a linearisation provided to the IKH-criterion, be non-stratified.
in [25, p. 49] may be applied within an accuracy of
±2 %. Also shown in Figures 7.20 and 7.21 are the pipe
Evidently, the IKH-criterion is violated through- inclination angles, of which the RHS of the IKH-
out the pipeline. Because the domain of VKH sta- criterion is dependant. In vertical pipe segments, the
bility is smaller than that of the IKH-criterion, a RHS of (E.3) reduce to zero and the criterion is un-
stratified flow regime should not be able to persist conditionally violated. Physically, this means that
physically as a steady state flow pattern. Some form a stratified flow cannot be sustained vertically. In-
of hydrodynamic slug initiation mechanism, as in stead, if separated, the regime will be annular. In
case II, is therefore justified. Further, as was proven the SLUGGIT model however, only the wetted and
by Bonizzi et al. [6, 13, 7], the IKH-criterion also interfacial perimeters Sκ , Sint are adjusted. These are
marks a necessary criterion for the stratified two- weighted as a function of pipe inclination to give a
fluid model to have real characteristics (be well- smooth transition to an annular flow pattern topol-
posed) and produce physically reliable solutions. ogy. At the same time, the stratified two-fluid model
Similar figures of the well-posendess criterion (4.30) is solved as if the flow is stratified, including the
are not shown here as they are nearly identical to ∂z h term in the momentum equation (4.16). Though
Figures 7.20 and 7.21. Ill-posedness can manifest it- this term vanishes as the inclination approaches ±90◦ ,
self in un-physical, resolution-dependant disturbance the mid-region of high inclinations is poorly de-
24
It is the continuous formulation of the two-fluid model which becomes ill-posed.
76
fined, assuming partially annular surface perimeters values as the scheme basically incorporates a coarse
while retaining a phase hight hydrostatic pressure two-fluid model [16], all be it with adjusted friction
term based on stratified flow. This aside, a sepa- terms. Though appearing unproblematic from expe-
rated flow regime, as modelled in the SLUGGIT v.2 rience running the code, the effect of ill-posedness in
scheme in vertical risers, will have complex eigen- this scenario is not altogether understood.
30
IKH−criterion
Time = 0.25 h
25
|ug − ul |
√
20 RHS
[m/s]
15
10
0
0 1000 2000 3000 4000 5000 6000 7000 8000
100
z [m]
φ [(deg)]
50
−50
−100
0 1000 2000 3000 4000 5000 6000 7000 8000
z [m]
Figure 7.20: Case III simulation. (Square roots of) left and right hand side of inviscid Kelvin-Helmholtz
inequality criterion (E.3)
30
IKH−criterion
Time = 1 h
25
|ug − ul |
√
20
RHS
[m/s]
15
10
0
0 1000 2000 3000 4000 5000 6000 7000 8000
100
z [m]
φ [(deg)]
50
−50
−100
0 1000 2000 3000 4000 5000 6000 7000 8000
z [m]
Figure 7.21: Case VI simulation. (Square roots of) left and right hand side of inviscid Kelvin-Helmholtz
inequality criterion (E.3)
77
7.3.2 SLUGGIT v.2s
For comparison, case I is re-simulated using the simplified scheme SLUGGIT v.2s:
Case I-s: MAX_BUB/SLUG_COEF = 200. No modelled hydrodynamic slug initiation. Simplified scheme.
Description:
t < 0.25 h: Slug formation takes place only some 50 meters away from the inlet. Slugging flow (slugs
and bubbles of about equal length) propagates through the pipeline and up the riser.
0.25 < t < 0.4: Slugs in the horizontal die away as the total liquid fraction in riser increases. The flow
slows down, increasing the liquid fraction in the well also.
0.4 < t: In a gradual manner, the inlet pressure approaches the well pressure and the system comes to a
stand-still. No liquid production has taken place.
Figure 7.22 shows the pressure and superficial So why the difference in prediction? The expla-
velocities close to the inlet. The simulation shows nation lies of course in the main method simplifica-
much higher fluctuations than in the non-simplified tions as they are described in Section 4.2 and Ta-
case of the previous section. The reason for this is bles 2.
onset of slugging close to the inlet. Notice also that Most importantly, assuming uniform pressure in
data is taken form the location z = 20 m, where the stratified units denies gas acceleration thought pres-
diameter is D = 14 cm, rather than z = 200 m, where sure transients, which are likely to be important in
D = 15.7 cm, as was done previously. The reason for the initial formation of the separated regime seen in
choosing z = 200 m was the large section lengths of the unsimplified model. Transient pressure forces
Case VI, while the location z = 20 m in Case I-s was counteract liquid build-up. This is demonstrated in
chosen because further away from the inlet the flow Figure 7.23, which shows the spatial pressure profile
was slugging, which gave very intermitted and untidy during the initial stages (t = 20 s) of the Case I and
curves. Case I-s simulations. Here, both simulations show
liquid propagation by means of a negative pressure
gradients. The non-simplified scheme manages to
create this gradient in the stratified/annular regime as
gas accelerates and column weight decreases. Due
to uniform stratified unit pressure, this is not a pos-
sible solution in the simplified scheme. Rather, the
formation of slugs are required in order to manifest
a pressure gradient. Also seen in Figure 7.23 is how
the pressure at the most upstream slug front in the
simplified scheme always will equal the outlet pres-
sure. Consequently, this warrants huge mass flow
rates initially
Uniform stratified pressure also dismisses the

Bernoulli suction phenomenon and the possibility of
slug capturing – the area of model stability now dif-
Figure 7.22: Case I-s simplified simulation. pressure fers from that of the two-fluid model. An interesting
and phase velocities using performance index inlet note to add here is Renault’s discovery that first order
boundary conditions. Time series taken close to well upwind schemes with velocity-staggered grids (Fig-
inlet. ure 4.7b), will tend to ‘misplace’ this Bernoulli suc-
78
tion one section downstream [37]. Velocity staggered the supplied GOR value is to find the gas-oil-water
grids are quite common in transient codes, and is separation at standard conditions and compute the
found in e.g., SLUGGIT v.2, TRIOMPF and OLGA. separation of this mixture back to the inlet condi-
A consequence of this is that suction physically caus- tions using PVT software. The main difference here
ing increasing hold-up locally will numerically in- is that the gas mass at the inlet in considerably less
duce a pressure gradient in the counter-flow direc- as phase transition causes gas condensation at the
tion, which produce a suction ahead of single-section increasing pressure.
slugs, possibly promoting liquid propagation and a
separated regime in the non-simplified scheme. This Khabibullin provided a new approximation
gives an incentive for considering pressure-staggered of the GOR at the inlet under the conditions
schemes as a substitution. p = 122 bar, T = 63.5 ◦ C, namely GOR = Vg /Voil =
Further, in excluding the convection terms in the 259 m3 /d/1839 m3 /d = 0.14 m3/m3 . Since all PVT sup-
liquid momentum equation, injected liquid is given port is yet to be implemented in the SLUGGIT
no initial momentum with which it is transported code25 , the same procedure as earlier must be em-
away from the inlet. ployed, only the standard conditions are redefined
to closer match the inlet. The ‘adjusted standard
Contrary to the intention and intuition, the sim- ∗ ∗
conditions’ are pS T = 122 bar, T S T = 65.3 K. Wa-
plified scheme ends up requiring a significantly ter cut is assumed unchanged at 53% (though it in
longer computation time than the non-simplified reality slightly increased). As before, the GOR is
scheme because of the increased complexity accom- further adjusted to an all-oil single phase liquids
panied with the slugging flow (particularly through GOR∗ = (1 − 0.53) · 0.14 = 0.0658, wc∗ = 0
the jumpers.)
The simplified scheme afforded a significantly
Figure 7.24 show pressure and superficial veloc-
longer computation time than the non-simplified
ity plots for the new GOR and ‘adjusted standard con-
scheme due to increased the complexity of the slug-
ditions.’ The profiles are seen to be similar to Case I
ging flow
and Case III in Figures 7.18a and 7.19a, respectively.
The main difference lies in the reduced volumetric
Adjusted GOR gas flow rate due to the correction for phase transi-
Near the end of Section 7.2, an adjustment was made tion. Overall liquid fraction increases therefore, also
to the inlet flow rate conditions to correct for a mis- when slugs are hydrodynamically initiated (bubbles
understanding that had arisen where mass rates had between slugs become shorter). Consequently, the
been defined at standard conditions and so not cor- liquid fraction in the riser and well become too high,
rected for phase transfer. This mishap also affects the and the flow collapse once the front of the slug-
present section as the GOR allegedly also is defined bubble regression has reached some distance up the
in terms of gas-liquid separation at standard condi- riser.
tions, and not, as was initially though, as the separa-
tion at the inlet. Due to the present intrinsic lack of It should be emphatically pointed out that adjust-
any PVT support in the models codes, the interpre- ing the GOR-value as done here is no a solution to
tation of the GOR as a boundary condition computes the present lack of PVT-management in the SLUG-
the mass flow rates under standard conditions and as- GIT codes. Earlier it was seen that the difference
sumes these phase mass rates to be the mass rates at in gas mass flow rates due to phase transition at the
the inlet. Without PVT support, the conversion form well and outlet conditions are large. Changing the
standard cubic meters was done by finding the ratio gas mass flow at the inlet does not account for the
of density at inlet and standard conditions – see Ap- phase transition which occurs in the riser as local
pendix C. pressure decreases. Consequently, adjusting for a
Conversely, the ‘appropriate’ way of applying correct gas mass flow rate at the well will generate an
25
This is being done presently by Tor Kjelby.
79
Non−simplified scheme. Time = 20 s
200
Pressure [bar]
150
100
50
0
0 500 1000 1500
Simplified scheme. Time = 20 s

150
Pressure [bar]
100
50
0
0 500 1000 1500
z [m]
(a) Case I (b) Case I-s (c) Pressure
Figure 7.23: Initial development, simplified vs. non-simplified SLUGGIT scheme t = 20 s
under-prediction of gas mass flow rate near the out- in deep-water simulation tools.
let. PVT-support is therefore a necessary component
z = 200m, D = 15.7cm z = 200m, D = 15.7cm

240 10 240 10
230 pressure 230 pressure


uS
l uS
l
S
220 uS 220 uS
g g
5 5
210 210
200 200
190 0 190 0
0 0.1 0.2 0.3 0.4 0.5 0.6 0 0.1 0.2 0.3 0.4 0.5 0.6
Time [h] Time [h]
(a) No hydrodynamic initiation (b) SSB hydrodynamic initiation
Figure 7.24: Adjusted GOR computation. MAX_BUB/SLUG_COEF = 200
8 Discussion and recommendations
8.1 Girassol instability

Finding the origin of the P50 Girassol pipeline instability has proven to be challenging task. Though multiple
forms of operational instability were generated, none had a period of oscillation in the range of the field data.
80
Over-predicting slugging frequency seems to be a common problem in most multiphase codes [35]. For the
SLUGGIT code, the following obstacles have shown to be particularly troublesome:
Gas lift No field data is available in which the gas injection is switched off altogether. Neither does the data
from which it is at its lowest (70 kSm3/d) ever reaches a steady amplitude of fluctuation. In fact, it is not
altogether know whether the P50 pipeline has any production at all without active gas lift. Neither any of
the SLUGGIT codes nor the LASSI code has in them any well-functioning support for generating a gas
lift in the riser (i.e., placing a gas source at the riser base.) The reason for this is that technical challenges
arise when a slug object passes the point at which the gas source is placed. Kjeldby implemented a
method for tackling these challenges in his master thesis work [20], though this method caused some
pressure disturbances and did not function ideally. Modified gas lift procedures are possible and are
likely to be implemented in the near future. An alternative and perhaps preferable way of overcoming
the gas source challenge is to implement gas entrainment into slugs. Void in slugs is also recommended
to improve the methods’ capability of predicting dispersed flows.
PVT treatment It has already been demonstrated, though quite unintentionally, that retrieving the appropriate
thermodynamic PVT fluid mixture properties is important in fields where pressure changes are signifi-
cant. Mainly the issues are those of phase transition, which in this field is found to be of the same order
as the produced gas. The influence of this may very well affect the system dynamic as a whole.
Field data controversy The for publication reported by Zakarian and Larrey [47] does not altoghther match
the field data [43] from which all the pressure information originates. Where the former supplies a liquid
flow rate h Ṁl,i i = 33.7 kg/s, the mean volumetric flow rate is found to be 1 288 Sm3/d in the latter, or 14.9 kg/s.
Though there probably exist a natural explanation for this, it does serve to reduce confidence in the data.
Limited flow regimes All of the EPT slug tracking methods presently rely on the base assumption that all
simulated flow regimes can be represented by slug and stratified flow objects. In Section 5 and 7, it was
found that the limited number of flow regimes – particularly the absence of a dispersed flow – may reduce
the domains of model stability.
In Section 5.4 it was found that the SLUGGIT v.2/v.2s codes had difficulty in obtaining steady flow at
low liquid, high gas flow rates. This is particularly disconcerting since flow stability is here the main
focus.26 Similarly, it was found is Section 7.3 that the SLUGGIT v.2/v.2s codes predicted high pressure
fluctuations in comparison to LedaFlow and OLGA simulations carried out by Khabibullin [18]. Occa-
sional release of coalesced tailor bubbles into the riser was found to cause a part of the fluctuations in the
SLUGGIT simulations, while LedaFlow and OLGA had predicted bubbly flow regimes in the horizontal.
On the other hand, due to lack of gas lift and confusion surrounding inlet boundary conditions, the basis
for comparison is questionable.
Implementation of gas entrainment in slugs is likely to improve this condition. The SLUGGIT v.1 code
by Pascal Klebert already have this feature installed, but could not be tested during this work due to
technical problems with the code. Bonizzi et al. [7] investigate the influence of entrainment into their
slug capturing code TRIOMPF. In the limited and simple cases investigated in this publication it was
found that the effect was not severe, but corrected an otherwise erroneous trend where increasing mixture
velocity would also increases mean hold-up in the V-section case.
Limited focus on vertical flow All investigated slug tracking methods have been developed based on the prin-
ciple of the stratified two-fluid model. During the present work, these methods have unquestioningly been
26
Some reserves should here be made. It is not entirely clear whether this is due to the model fundamentals or short-comings in
the current management procedure. The SLUGGIT v.2/v.2s are being continuously developed and differences in stability predictions
within resent versions may suggest the latter.
81
applied to vertical risers. In the case where the flow regime is predicted to be slugging with short units
(often method-equivalent of bubbly flow,) the SLUGGIT methods reduce to an empirical slug specifica-
tion model in which the bubble propagation correlations and mass balances govern the flow.
Some assurance is provided by the fact that test simulations provided excellent correspondence to the
steady-state model in Appendix A, provided a reasonable maximal section length. With high gas rates
however, the riser flow will on the other hand be annular. Even with some correction for the friction
perimeters, the liquid height gradient term ∂z h cos φ will disappear form the (stratified) two-fluid model.
Consequently, the pressure treatment will be equivalent to applying the constant cross-sectional pressure
assumption in a simplified two-fluid. As was shown in [16, p. 69-70], or in expression (4.30) for that
matter, this simplified two-fluid model is then unconditionally ill-posed. Though this topic is somewhat
controversial, it seems to this author at least that some effort must be made, either in ensuring well-
posedness, or in proving an evidential basis, for it to be fully accepted in a scientific community.
Another point which presently seems weak in the LASSI and SLUGGIT codes is the 90◦ bend at the riser
inlet. Appropriate slug formation here has been found to be vital for obtaining the correct flow pattern
in the riser, which in turn has a strong effect on the system dynamic as a whole. Bend initiation criteria
have not proven to be all together successful in this respect, partially because this criterion requires inflow
towards a low-point from both directions. The presently developed pipe class for bent pipes may improve
these matters, but would still require high section length refinement in bend regions.
Although the type of pressure fluctuations ob- with it’ additional gas from jumpers (Cycle A) seems
served in the field data [43] (Figure 1.3) has not been very unlikely for two reasons. Firstly, the proximity
recreated, much has been learned about the instabil- of the gas congestion would be too close to the riser
ity characteristics and how they apply to the Girassol to produce periods of three hours. Secondly, the en-
field case. In Section 7.1, various forms of oscilla- trapment volume of the risers and pipeline does not
tory behaviour related to the fixed-pressure ‘open in- seem sufficient.
let’ boundary conditions were discovered. Here, geo- Gas release caused by upstream expansion is also
metric features of the pipeline, centrally the jumpers a possibility. Again though, the only upstream en-
and U-bend, were given special attention. Reproduc- trapment volume big enough is likely to be the well
ing oscillations particularly caused by gas expansion itself. If so, the mechanism is actually equivalent to
in the riser was attempted successfully, including the the favoured void wave Cycle B.
dynamic described as ‘expansion driven instability’
in the assignment description of this thesis. Also a All the instabilities observed in the open in-
surprising feature of this investigation was that os- let configuration appeared to be very sensitive to
cillations in the U-bend may play a role in operation changes in geometry, and inlet hold-up and pressure.
instability dynamics involving pipeline upstream of Small changes have been found to cause or reverse
the riser. Despite these findings, it was in general operational instability mechanise, and even uncover-
observed that the oscillating pressure profiles cor- ing unexpected new ones.
responded poorly with the field data measurements.
Periods of gas accumulation in jumpers, and oscilla-
In Section 7.2, the fixed fluid flow rates bound-
tion periods generated from the void waves created
ary condition was tested. Severe slugging obtained
therefrom, were both found to be short (usually less
using Ṁg,i = 1.69 kg/s and Ṁl,i = 33.7 kg/s was found
than an hour.) These would yield pressure fluctua-
to give slugging periods of about 30 min and tip-to-
tions that were occasional and spiky, rather than pos-
tip amplitudes of approximately 60 bar. Compared to
sessing a slow, sinusoidal character of the field data.
the 3 h periods and 35 bar tip-to-tip amplitudes of the
Although riser expansion is likely to be a key com-
field data, the match is poor. The flat-top form sinu-
ponent in the field instability, the suggested mecha-
soidal of the field data also compares poorly with the
nism of expansion driven fluid accelerations ‘pulling
sharp zig-zag pressure pattern of terrain slugging. In
82
terms of amplitude, it must again be underlined that will exceed the well head. If not at a stand still, the
it is unknown how the field pressure would develop system would show no signs of operational instabil-
if allowed to grow. ity.
Recently made corrections of the mass source
conditions gave a different response altogether. Here, Again, these results appear to have been affected
the rate of gas injection was so low as to make gas by the misunderstanding concerning the inlet con-
production only occasional, as Tailor bubbles from ditions (GOR definition in terms of phase transfer).
time to time penetrated into the riser. The restricted Only stand-still solutions were found when apply-
gas amounts reduced the volume of upstream coming an approximate GOR based on ‘adjusted standard
pressibility, while the riser was mostly un-aerated. conditions’. Adjusting the GOR and standard con-
Pressure fluctuations were found to originate both ditions is not a solution to the PVT problems, and
form the U-bend and the events wherein a Taylor the differences in prediction demonstrate that the pro-
bubble enter the riser. Again, fluctuation periods and ductivity index implementation is not complete until
amplitudes were too short. such support is available. Naturally, the mix-up in
inlet conditions also serves as a reminder of the im-
The full geometry simulations with the produc- portance of supplying detailed, accurate and verified
tivity index inlet boundary condition of Section 7.3 case information.
was though to offer the greatest hope of reproduc-
ing the field instability. The main reason for believ- 8.2 Method considerations
ing so is the long oscillation periods in the field data.
Should these oscillation be caused by a transport phe- LASSI: The investigatory part of this thesis be-
nomena, as opposed to some sort of build-up as in gan by testing the codes’ ability to handle the ver-
terrain slugging type I or gas pocked build-up, the tical riser geometry. The LASSI code [36] failed
distance of transport need to be vast. It is likely that these tests brutally, the cause rooted in a inclination-
the entire pipeline – from inlet outlet – is involved. dependant failure to conserve mass. Since the formu-
Wave phenomena similar to that described in Cy- lation of the LASSI model is presented in an entirely
cle B seems the most plausible explanation in this re- mass-conservative manner in [37], it is assumed that
spect, particularly due to the way such a cycle would the lack of mass conservation is due to an imple-
generate a sinusoidal pressure profile, as well as be mentation error, possibly in the section management
strongly dependant upon the operational state of gas procedure. Whether finding and solving this bug
lift injection – both in terms of how gas injection af- would be a small or large task is unknown, but it
fects the flow in a stabilising manner, and in term of is imperative if the scheme is to be applied for any
how a change in gas lift state would cause a distur- future purpose.
bance kick-starting this type of operational instabil-
ity. This was discussed in more detail in Section 3.2. Based on the results published in [37], one may
Since well response is crucial for such density-wave postulate how this scheme may fare in comparison to
instability dynamics to take place, further investiga- the SLUGGIT schemes for this case study. In the
tion into well-replicating boundary conditions on the horizontal of the Girassol case the LASSI scheme
inlet may be beneficial in investigating this field case. would probably hold certain advantages, such as
waves and slug capturing. The method is how-
Disappointingly, unless liquid accumulation in ever not developed for the purposes of vertical flow,
the riser was continuously counteracted by upstream and so would have little advantage in the riser and
hydrodynamic slugs, the system would come to a well. Further, simulating with the resolution required
stand-still. Most simulation indicated the stand-still for grid-independent wave functionality would en-
outcome, which is an easily verifiable solution as the tail significant computational cost in a pipeline of the
un-aerated inlet static head of filled riser and well field dimension. In [37], section lengths at the or-
27
Even though the LASSI code [36] is more simply implemented and so therefore runs significantly quicker than SLUGGIT for the
same number of sections and time steps.
83
der of the pipe diameter is normally applied, which implementing gas entrainment into slugs is the pre-
would be impossible in a field study.27 On the other ferred approach since gas entrainment in itself also a
hand, also SLUGGIT simulations are clearly affected suggested improvement. Further, the slug initiation
by section length when pushed beyond a limit. In routine during terrain slugging needs to be investi-
field tests this was found to begin affecting the sim- gated with focus on operational stability regime.
ulations as MAX_BUB_COEF passed 200. More spe- Because density wave instabilities, as described
cific statistical properties, such as slug frequency in Section 3.2, seems a plausible cause of the field in-
and length, are likely to be affected by resolution in stability, further investigation into density waves may
SLUGGIT for far smaller section lengths. prove fruitful. Further development of the well rela-
tion applied to the inlet boundary may also be neces-
SLUGGIT v.2s: In the full-geometry simulations, sary in this endeavour.
the simplified scheme gave slugging flow starting Incorporating gas entrainment in slugs may re-
at the inlet due to intrinsically uniform pressure in duce the fluctuations form Taylor bubbles entering
stratified units. Though it should be mentioned that the riser in low-gas systems, such as those seen in
it is not really known what flow regime is physically Figure 7.24. It may possibly also improve confor-
appropriate at the inlet of the Girassol field case, it mity to the terrain slugging flow map tested in Sec-
is an important drawback of the simplified scheme is tion 5.4.
that achieving an separated flow solutions in highly All slug tracking method discussed here have
inclined pipe segments becomes nearly impossible. been developed for near-horizontal flow. It may be
Only with massive interfacial friction forces can liq- necessary with further development of the method
uid be driven upwards without the assistance of a intrinsics with particular focus on vertical flow. In
stratified/annular pressure gradient. Otherwise, the essence, this concerns expanding the extent of sup-
simplified scheme only really support slugging flow. ported flow patterns so that they are appropriately de-
Therefore, the simplified scheme is most appropri- fined in vertical pipes. On the other hand, expanding
ate when the physical flow regime consists of rela- the methods excessively may go against the ideal of
tively short bubbles. If bubble units are single sec- simplicity and ‘reducing to essentials’ on which the
tion sized, than, apart from neglecting the convective methods are founded.
terms in the momentum equation, the simplified and During the flow map testing in Section 5.4, it
non-simplified schemes are equivalent, removing the was found that both criteria and the procedure by
benefit of simplification altogether. which slugs were initiated at the riser inlet dip were
determining factors in the method’s ability to repro-
Perhaps it is possible to create a hybrid environ- duce experimentally determined operational regimes.
ment within which the simplified treatment changes Criteria and initiation procedures are however de-
if bubble units grow too long. This would however veloped and changes continuously, affecting stabil-
cause management challenges in the transition from ity behaviour both for better and worse. In the opin-
‘short’ to ‘long’ bubbles which probably would sab- ion of the author, finding a general-purpose model
otage the intended simplicity. which will be capable of generating appropriate ter-
rain slugging for all possible pipeline configurations,
similar to the bend initiation procedure, would be
8.3 Recommendations and suggestions
difficult and require much special case treatment.
In order to perform reliable deep-water simulations Rather, these areas of the geometry, which are subject
the challenges listed in the beginning of Section 8.1 to much activity and influence the system as a whole,
must be addressed. need the appropriate resolution. Presently, the only
In particular gas lift and PVT support seems vital. way of reducing allowable section length is to do so
Alternative border restriction procedures for allow- across the entire pipeline.
ing gas sources to be placed independently of slug It should, however, be a simple task to allow for
presence have been suggested. Though this would location dependant length parameters. This could
be interesting to test, enabling gas lift by means of reduce computational time significantly while retain-
84
ing an appropriate resolution where it is needed. An equately familiarised with each code. Nor, for that
important feature in this respect however, is that the matter, to find and remove enough bugs for the case
dynamic time step regulation sets the time step ac- simulation to run. It would appear unlikely, or at leas
cording to the smallest Courant number found across not recommendable, that the hours required in devel-
the entire pipeline. Broadly stated, this implies that oping each of these codes to a robust level ever be af-
a local section refinement would still be accompa- forded. Rather, the methodologies of each can be im-
nied by a proportional increase in computation time plemented in a unified framework, utilising as much
(which is better than the quadratic increase found as possible of the same management code and regu-
otherwise – see Section 4.5). An idea in this con- lation features. This is also a vital point for the valid-
cern is an intermediate computation; in a region of ity of comparison; are core method differences to be
pipe where resolution is sought to be increased, say appreciated with clarity, all other features of manage-
a distance reaching form one slug-bubble border to ment should remain the same (grid, time step, inlet,
another, the local section lengths could be halved lo- outlet, exception handling, closure relations, turning
cally. Rather than making extra computations across and hydrodynamic slug initiation criteria, etc.. )
the entire pipeline, an intermediate computation n+1/2
is computed in this region only. For the mass equa- 8.4 Final comments
tion, this would be unproblematic since mass fluxes
across bubble-slug borders are already computed ex- It should not be forgotten that the field data to which
plicitly. The implicitness of the pressure-momentum a comparison has been attempted is the field response
equation system would however be disturbed. This to a particular event, namely the momentary with-
latter obstacle holds equally for the trouble of paral- drawal of gas lift support from a stable to an unstable
lelising the code, which would also involve a ‘seam’ state. Possibly the oscillations seen is the beginning
across which implicitness is broken. of a highly transient process which eventually would
bring the system, from initial steady, gas-lift sup-
ported conditions, down to a stand-still. In this case,
When considering the simplified v.2s versus the obtaining similar transients while initiating from a
non-simplified v.2 SLUGGIT schemes, it is the au- stand-still condition without gas lift would indeed be
thors opinion that the simplified scheme will be impossible. It may also be that the oscillations from
suited for some special slugging cases only, and that the field data would continue to grow. In any event,
in these cases the simplification benefit in terms of since the data presents merely an uncompleted tran-
computational savings is limited. Instead, there is sition from one state to another, only comparison of
at the moment thought to exist a significant poten- the same transition process can be said to be appro-
tial in optimising and improving the efficiency of the priate. This is not possible at present.
SLUGGIT v.2/v.2s code [21] as a measure for reduc-
ing computational costs. Also, though it may be that
The goals put forth in this thesis work was to
the simplified scheme has an advantage in terms of
investigate the EPT slug tracking schemes’ ability
robustness, the non-simplified scheme has benefited
to recreate real-life oilfield slugging dynamics us-
from more development, debugging and testing.
ing the P50 Girassol field pressure data as a bench-
mark for comparison. At the same time, finding the
Lastly, the author is absolutely adamant that the hitherto undocumented caus(es) of the oilfield’s un-
continued development should take place within one stable operation was put forth as a supplementary
single unified framework. It is a great misfortune goal. Accordingly, it has here been attempted to ver-
that the three codes, SLUGGIT v.1 [23], SLUGGIT ify model behaviour using field data, and investigate
v.2/v.2s [21] and LASSI [36], have all developed in- operational instability mechanisms using those same
dependently. Even the time afforded within a master models. This constitutes a type of ‘mutual bench-
thesis work did not prove sufficient to become ad- marking’ – always a questionable business indeed.
85
9 Conclusion
The slug tracking models, in combination with the PLOTIT visualisation tool, have shown to be particularly
well suited in helping to understand pipeline flow dynamics. Not only in the clarity of visualisation does this
become apparent, but also in the way in which the code is open and easily modifiable. This offers an unique
ability to investigate the influence of phenomena and flow mechanisms by making them accessible for direct
phenomenological manipulation.
At present, the SLUGGIT and LASSI codes does not seem mature enough handle deep water riser systems
reliably. In particular, these codes still lack
• Support for gas sources capable of handling passing slugs (gas lift).
• PVT-support accounting for high-pressure compressibility and phase transition.
• Models developed for vertical and dispersed flow regimes.
Prediction sensitivity to mostly intuition-based management procedures is also an issue.
Although not directly reproducible, the present study indicates that the field instability is produced, at
least in part, by some wave phenomena involving long stretches of pipeline, and probably also the well. The
sinusoidal character of pressure measurements suggests density or surge waves, similar to Cycle description B.
Acknowledgements
The author would like to thank his supervisor, Ole Jørgen Nydal, found to always be available for
a discussion, and co-supervisor/colleague Tor Kjeldby, who helpfully and cheerfully exposed himself to
regular and frequent bombardments of technical questions.
Further, the author is indebted to Eldar Khabibullin for all help regarding the oil field and boundary
conditions, as well as to TOTAL for approving the use of their field data.
This thesis was written in LATEX R by Erdinc˛ Özkan, the freeware IncScape R was used for all drawn
figures and the software MATLAB R for all plots and data analysis.
86
References
[1] Andreas H. Akselsen and H.I. Andersson. The Effect of Near Wall Resolution on Turbulence Statistics
using the Lattice Boltzmann method, 2011. Project thesis, NTNU.
[2] N. Apazidis. Influence of bubble expansion and relative velocity on the performance and stability of an
airlift pump. Int. J. Multiphase Flow, 11(4):459–479, 1985.
[3] D. Barnea and Y. Taitel. Kelvin-Helmhotz stability criteria for stratified flow: viscous versus non-viscous
(invicid) approaches. Int. J. Multiphase Flow, 19(4):639–649, 1993.
[4] K.H. Bendiksen. An experimental investigation of the motion of long bubbles in inclined tubes. Int. J.
Multiphase Flow, 10(4):467–483, 1984.
[5] K.H. Bendiksen, D. Malens, R. Moe, and S. Nuland. The Dynamic Two-Fluid Model OLGA: Theory and
Application. SPE Production Engineering, pages 171–180, 1991. Inst. for Energy Technology, Norway.
[6] M. Bonizzi. Transient one-dimensional modelling of multi-phase slug flows. 2002.
[7] M. Bonizzi and R.I. Issa. A model for simulating gas bubble entrainment in two-phase horizontal slug
flow. Int. J Multiphase Flow, 29:1685–1717, 2003.
[8] S. Chandrasekhar. Hydrodynamic and Hydromagnetic Stability. Oxford University Press, 1961.
[9] Y.-H. Chenga, J.-R. Wangb, H.-T. Linb, and C. Shih. Benchmark calculations of pressurizer model for
Maanshan nuclear power plant using TRACE code. Nucl. Eng. Des., 2009.
[10] A. De Leebeeck and O.J. Nydal. Simulation of large amplitude waves in a slug tracking scheme compared
to roll wave experiments at high pressure. Int. J. Multiphase Flow, 36:40–50, 2010.
[11] H. Dhulesia and D. Lopez. Critical evaluation of mechanistic two-phase flow pipeline and well simulation
models. In SPE Annual Technical Confrence and Exhibition, pages 393–402, oct. 1996.
[12] Houston: SPT Group. OLGA user manual - transient multiphase flow simulator, 2006.
[13] R.I. Issa and M.H.W. Kempf. Simulation of slug flow in horizontal and nearly horizontal pipes with the
two-fluid model. Int. J. of Multiphase Flow, 29:69–95, 2003.
[14] F.E. Jansen, O. Shoham, and Y. Taitel. The elimination of severe slugging – Experiments and modeling.
Int. J. Multiphase Flow, 22(6):1055–1072, 1996.
[15] Monika Johansen and O.J Nydal. An experimental study of the bubble propagation velocity in 3-phase
slug flow. PhD thesis, Norwegian University of Science and Technology. Department of Energy and
Process Engineering, Trondheim, 2006.
[16] S.T. Johansen, T. Ytrehus, and K.E. Einarsrud. Modeling of Multiphase Flows. Compendium of PhD
course EP8404, Multiphase Flow Modeling, Norwegian University of Science and Technology.
[17] E. Kaun and O.J. Nydal. Expansion driven Unstable Two Phase Flows in Long Risers and Wells. Master’s
thesis, Norwegian University of Science and Technology. Department of Energy and Process Engineering,
Trondheim, 2011.
[18] Eldar Khabibullin. Multiphase pipe transport of oil and gass: Dynamic simulation of Girassol field case,
2011. Project thesis, NTNU.
87
[19] T.K. Kjeldby, R. Henkes, and O.J. Nydal. Slug tracking simulations of severe slugging experiments. In
World Academy of Science, Engineering and Technology 78, pages 928–933, 2011.
[20] Tor K. Kjeldby and O.J. Nydal. Simulation of expansion driven flow instabilities in long risers. Master’s
Trondheim, 2010.
[21] J. Kjølaas, T.K. Kjelby, and A.H. Akselsen. Sluggit source code, 2012. Referred to as “SLUGGIT v.2”
and “v2s” in the present work, documentation: [22].
[22] Jørn Kjølaas. Plug propagation in multiphase pipelines: Modeling and small scale experiments. PhD
Trondheim, May 2007.
[23] P. Klebert and O.J. Nydal. Sluggit source code, 2004. Referred to as “SLUGGIT v.1” in the present work,
documentation: [24].
[24] Pascal Klebert and O.J. Nydal. SLUGGIT, Models and numerical implementation, October 2004. Post-
Doc work.
[25] Olav Kristiansen and O.J Nydal. Experiments on the transition from stratified to slug flow in multiphase
pipe flow. PhD thesis, Norwegian University of Science and Technology. Department of Energy and
Process Engineering, Trondheim, 2004.
[26] R.T. Lahey Jr. Void Wave Propagation Phenomena in Two-Phase Flow (Kern Award Lecture). AIChE,
37(1):123–135, 1991.
[27] E.W. Llewellin, E. Del Bello, J. Taddeucci, P. Scarlato, and S.J. Lane. The thickness of the falling film of
liquid around a taylor bubble. Proceedings of the Royal Society, 468(A):1041–1064, 2012.
[28] T.S. Mayor, A.M.F.R. Pinto, and Campos J.B.L.M. Hydrodynamics of gas-liquid slug flow along vertical
pipes in turbulent regime – A simulation study. IChemE, 2007.
[29] T.S. Mayor, A.M.F.R. Pinto, and Campos J.B.L.M. On the gas expansion and gas hold-up in vertical
slugging columns – A simulation study. Chemical Engineering and Processing, 2007.
[30] B. Müller. Introduction to Computational Fluid Dynamics. Lecture Notes in Computational Heat and
Fluid Flow, August 2011.
[31] O. Novak. PLOTIT (SLUGGIT, LASSI and OLGA data visualization programme), March 2012. Plotit
is using and/or is based in part on the work of the Qt library, Qwt project and QwtPlot3D project. EPT,
NTNU.
[32] O.J. Nydal. Excerzise 2 – severe slugging. Course TEP4250 - Multiphase Transport and the Norwegian
University of Science and Technology.
[33] O.J. Nydal, M. Audibert, and M. Johansen. Experiments and modelling og gas-liquid flow in an s-shaped
riser. 1998.
[34] O.J. Nydal and S. Banerjee. Dynamic slug tracking simulations for gas-liquid flow in pipelines. Chem.
Eng. Comm., 141-142:13–39, 1996.
[35] P.F. Pickering, G.F Hewitt, M.J. Watson, and C.P. Hale. The prediction of flows in production risers –
truth & myth? Published online, http://www.feesa.net, 2001.
88
[36] Fabien Renault. Lassi source code, 2007. documentation: [37].
[37] Fabien Renault and O.J. Nydal. A Lagrangian slug capturing scheme for gas-liquid flows in pipes. PhD
Trondheim, 2007.
[38] L. Sinègre and N. Petit. Distributed delay model for density wave dynamics in gas lifted wells. In 44th
IEEE Conference on Decision and Control, and the European Control Conference, pages 7390–7397, des.
2005.
[39] R. Skartlien, E. Sollum, P. Meakin, K. Furtado, and I.E. Smith. The Lattice Boltzmann Method with
applications to multi-phase flow containing interfacially active components. Des. 2008. Institute for
Energy Technology (IFE), Norway.
[40] Y. Taitel. Stability of severe slugging. Int. J. Multiphase Flow, 12(2):203–217, 1986.
[41] Y. Taitel, S. Vierkandt, O. Shoham, and J.P. Brill. Severe slugging in a riser system: Experiments and
modeling. Int J. Multiphase Flow, 16(1):57–68, 1990.
[42] H. Tennekes and J. L. Lumley. A First Course in Turbulence. The Massachusetts Institute of Technology,
1972.
[43] TOTAL, 2005. Field data from the Girassol oil field. Received with permission from TOTAL.
[44] M. van Dyke. An Album of Fluid Motion. The Parabolic Press, Stanford, California, 1982.
[45] H.K. Versteeg and W. Malalasekera. An Introduction to Computational Fluid Dynamics: The Finite
Volume Method. Pearson / Prentice Hall, Harlow, 2 edition, 2007.
[46] F.M. White. Fluid Mechanics. McGraw-Hill, sixth edition, 2008.
[47] E. Zakarian and D. Larrey. A Systematic Investigation of Girassol Deepwater-Field Operational Data To
Increase Confidence in Multiphase Simulation. Feb. 2009.
[48] M. Zastawny, G. Mallouppas, F. Zhao, and B. van Wachem. Derivation of drag and lift force and torque
coefficients for non-spherical particles in flow. Int. J. of Multiphase Flow, 39:227–239, 2012.
89
Appendices
A A model of void in risers

In this section, a steady state model designed to compute the void fraction in vertical risers is developed from
base principles. Although somewhat of a digression from the main topic of the EPT dynamic multiphase mod-
els, it is ventured upon as it will provide some basis of model comparison and evaluation. Though experimental
measurements are perhaps preferred for the purpose of validation, the basis for comparison with the large Gi-
rassol field case [47] is not ideal. Also, this theoretical analysis may be used to gain some insight into how gas
lift, or the lack such, affects the result when comparing with field data (Section 5.3.) A theoretical model with
which to compare is, therefore and in of itself, valuable. Additionally, it opens for the inclusion of theoretical
analysis into the present thesis work.
Figure A.1: Riser with gas lift Figure A.2: Bubble-slug unit
Figure A.1 presents a sketch of the layout for this analysis. The gas inflow stem from both the well and the
gas lift, so
Ṁg = Ṁggl + Ṁgwell
while liquid is only supplied from the well.

The following assumptions are used:
Assumptions D. Riser steady state void model
D.1 The liquid phases can be approximated as a single phase (two fluids)
D.2 The production is at steady state, i.e. the mass fluxes of both fluids, are constant in time and space.
D.3 The riser is fully vertical.
D.4 The gas propagates as large Taylor bubbles so that the empirical bubble nose velocity from Section 4.8
can be used to approximate the gas velocity – i.e., ub = vb,n
D.5 The local bubble area fraction αb (z) never reach unity.
90
D.6 Bubble film thickness acquire a practically constant value some short distance downstream from the
bubble nose.28
This analysis will be based on the schematic of and Uc1 = Ṁg RT/A and Uc2 = Uν + 1.2 Ṁl /(Aρl ). A to-
Figure A.2. The ratio of bubble and slug lengths tal riser void fraction Φg (neglecting the gas weight)
will be important in this respect, and so the follow- may then be defines:
ing local, mean bubble length fraction Γ is defined, Z H
together with the local average bubble cross-section ∆ 1
Φκ = ακ dz (A.3)
area fraction αb H 0
∆ `b ∆ Ab (r − λ f )2 and, by Fubini’s theorem:

Γ= αb = =
` A r2 Z t Z H
1 1
hΦg i = lim αg dz dt0
Applying the steady state assumption D.2, mass bal- t→∞ t − t0 t H 0
0
ances may be applied using to mass rates to provide 1
Z H
1
Z t
two equations. Also, doing a liquid mass balance = lim αg dt0 dz (A.4)
H 0 t→∞ t − t0 t0
across slug and bubble film, one can retrieve expres- Z H
1
sion (4.21) in Section 4.8. As declared in Assump- = hαg i dz
H 0
tion D.4, the empirical bubble nose velocity expres-
sion of Section 4.8 is used to express average bubble A slight digression: Before it was known that
velocity: the variables αb and Γ would disappear from (A.2),
it was originally thought that a final relation on the
Ṁg = ρg ub AΓαb (A.1a) bubble film thickness λ f would be necessary in or-
Ṁl = ρl u s A(1 − Γ) · 1 + ρl u f AΓ · (1 − αb ) (A.1b) der to achieve closure. Before this was known, some
u s − (1 − αb )u f effort was put into finding an appropriate expression
ub = (A.1c)
αb for the film thickness λ f . Searching through the lit-
ub = 1.2u s − Uν (A.1d) erature provided some such information. In particu-
lar, Llewellin et al. [27] give a nice summary of this
As before, b, s and f denote ‘bubble’, ‘slug’, and topic and showed that the dimensionless film thick-
(liquid) ‘film’, respectively. ness λ f /r ispa function of the dimensionless inverse
Equations (A.1) produce a system of four equa- viscosity ρl gd3 /µ only. For example, should bubble
tions and five variables {u s , ub , u f , αb , Γ}. As it film thickness be a interesting characteristic of the
turns out, the variables Γ and αb collapse into the dynamic simulation models, Brown’s (1965) laminar
variable of interest – the average gas volume frac- analyses, attacking Navier-Stokes equations, could
tion hαg i = αb Γ. Solving system (A.1) is a tedious be applied:
task, the details of which the reader will be spared. √
1 + ND − 1  16 ρ2l g 
 
The procedure is to contract (A.1d) and (A.1c) into λf = , where N = 
9Frb µ2

(A.1a), and then attack (A.1b), solving for Γ and mul- N
tiplying by αb . Gas density is handled by substitution
where Frb = ub / 2grb is the bubble Froude number
p
using the ideal gas law. Using caution, this produces:
with bubble radius rb = r − λ f . Brown assumed a
hαg i =
Uc1
(A.2) constant bubble Froude number Frb = 0.351 making
Uc2 p + 1.2Uc1 the film thickness constant throughout the riser.
where the mean gas volume (void) fraction hαg i =
Expression (A.2) provide and express for the av-
1 − hαl i is defined by
erage phase volume fraction hακ i and total void frac-
∆ 1
Z t tion hΦg i as a function of the local pressure. How-
hαg i(z) = lim α(t0 , z) dt0 = αb Γ(z) ever, the local hydrostatic pressure is itself an integral
t→∞ t − t0 t0
28
Empirically found to be valid [4]
91
function of the mean liquid volume fraction hαl i: Note that only the fluid in the cells above the present
Z z cell j is included. It would be more precise to also
p(z) = po + gρl ρm dz0 include half the liquid weight of the liquid in the
Z z 0h i (A.5) present section. This would however make the ex-
= po + g hαg iρg + 1 − hαg i ρl dz0 pression implicit and warrant the deed for iteration.
0
Two approaches, one discrete and one analytical ap- Instead of iteration, a sequential approach may be
proximation, are presented to tackle this challenge. take, regaining the accuracy by a finer grid. The pro-
cedure is then simply to loop through j = 1 to N (from
riser outlet, down towards the inlet). At each step j,
A.1 Discrete calculation
p j is computed from (A.6), using the values hαg iζ ,
In this approach, the riser is divided into small, finite ρg,ζ = pζ /RT ; ζ < j stored from the previous compu-
sections. Equation (A.2) is solved for each section tations. Then, hαg ij (p j−1 , p j−2 , . . . ) may be computed
using the information in the above sections; the pres- from (A.2) and stored as a profile. The total riser void
sure integral (A.5) for section j is evaluated as fraction can the be found by:
j−1
X N
p j = po + g ρm,ζ δzζ 1 X
hΦg i = hαg ij δz j
ζ=1 H j=1
j−1
(A.6)
X h i
= po + g 1 − hαg iζ ρl + hαg iζ ρg,ζ δzζ
ζ=1
0.6 600
Ṁg =2.35 kg/s
Ṁg =1.69 kg/s
Ṁl = 33.7 kg/s, UlS = 1.04 m/s 500
0.5 Ṁg =0.658 kg/s
400
pressure [bara]
0.4
s
2 .3 5 kg/
hαg i Ṁg = /s
0.3 1 .6 9 kg 300
Ṁ g =
0.2 g/s 200
0 .6 58 k
Ṁ g =
0.1 100
Ṁl = 33.7 kg/s, UlS = 1.04 m/s
0 0
−1200 −1000 −800 −600 −400 −200 0 −1200 −1000 −800 −600 −400 −200 0
z z
(a) Mean local gas volume fraction profile hαg i (b) Pressure profile p(z)
Figure A.3: Numerically computed profiles, Ṁl = 33.69 kg/s, or UlS = 1.04 m/s.
H = 1278 m, d = 0.2032 m, T = 59 ◦ C, po = 37 bar
A.2 Analytical approximation

Even though the numerical procedure of Section A.1 is straight forward, a single analytical expression for the
total riser void fraction may be useful. Here, this is acheived by attacking Equation (A.2) with the pressure
from (A.5). Contrary to the numerical procedure, the gas weight will be neglected. Combining these equations,
rearranging slightly and integrating from z = 0 to z = H (outlet to inlet) yields:
Z H Z z Z H Z H
Uc2 gρl hαg i hαg i dz0 dzUc2 gρl zhαg i dz − (Uc2 po + 1.2Uc1 ) hαg i dz + Uc1 H = 0 (A.7)
0 0 0 0
92
From here, the technique employed is to express (A.7) in terms of integrals of Γ from 0 to H , which is already
defined in (A.3) as the sought-after average total riser void R z fraction
hΦg i. The first term is handles easily as
either integration by parts or substitution of dz by hαg i−1 d 0 hαg i dz0 yields
Z H
Z z Z H !2
1 1 2
hαg i(z) hαg i(z0 ) dz0 dz = hαg i(z) dz = hΦg i H
0 0 2 0 2
RH RHRz
The second term, 0 zhαg i dz = hΦg iH 2 − 0 0 hαg idz0 dz , is far more troublesome – the author has not found
any way of solving for it analytically as the running variable z requires information in the void profile itself.
From the results of the numerical approach of Section A.1 (see Figure 5.6) it is observed that the slope of hαg i
is smooth and does not change greatly across the riser. The approximation
Z H Z H Z H
1 1
zhαg i dz ≈ z dz · hαg i dz = hΦg iH 2 (A.8)
0 H 0 0 2
is applied, which is an approximation analogue to the key approximation of the handling of the flux terms in
the finite volume method – e.g. Equation (4.2). Because the slope of hαg i slightly decreases as the outlet is
approach, approximation (A.8) underpredicts the true term value by a small amount.
All terms may now be expressed in hΦg i and known variables, producing a quadratic equation. Further
rearranging gives the simplest form
hΦg i2 − 2bhΦg i + c = 0 (A.9)
where
1 1
b= po + 1.2Uc1 Uc2 +
gρl H 2
2 Uc1
c=
gρl H Uc2
The quadratic expression (A.9) has the well-know solution

√
hΦg i = b ± b2 − c (A.10)
(± → −) providing the physical solution.
Figures A.4 shows the numerical and analytical void model’s response to changing liquid velocity into the
riser at a constant gas mass flow rate of Ṁg = 1.69 kg/s (152 kSm3/d). This is a gas lift flow rate taken from the
Girassol P50 pipeline field data [43], along with the properties H = 1278 m, d = 0.2032 m, T = 59 ◦ C, po = 37 bar.
The liquid well flow rate associated with this field is Ṁl = 33.69 kg/s or uSl = 1.04 m/s [18, 43, 47]29 .
Figures A.4 also include computations from the further simplified models presented in Annex F. These
models were developed leading up to the present one, and were simpler in that the the liquid film was neglected
– i.e. the slug liquid velocity was computed directly from the liquid flux and mean hold-up. As can be seen
from the figures, these models gave fairly similar predictions when the liquid mass rate is high relative to the
gas mass rate. However, when a larger mean fraction of the riser cross section is occupied by gas, the accuracy
decreases, notably – see Figure 5.6. The wish to improve this spatial void fraction profile near the riser outlet
was the motivation for the present model.
A.3 Comment
29
Temperature downstream riser choke applied.
93
1 0.7
Numerical computation, film included Numerical computation, film included
Analytical computation, film included 0.6 Analytical computation, film included
Numerical computation, film neglected Numerical computation, film neglected
0.8
Analytical computation, film neglected Analytical computation, film neglected
0.5
0.6 0.4
hΦg i hΦg i
Ṁg =1.69kg/s
0.3
0.4
0.2
0.2
0.1 Ṁl = 33.7 kg/s, UlS = 1.04 m/s
0 0
0 0.5 1 1.5 2 2.5 0 0.5 1 1.5 2 2.5 3
uSl [m/s] Ṁg [kg/s]
(a) Void fraction as a function of superficial liquid veloc- (b) Void fraction as a function of gas mass flow, Ṁl =
ity, Ṁg = 1.69 kg/s (152 kSm3/d) 33.69 kg/s or uSl = 1.04 m/s
Figure A.4: Approximated void fraction in riser
After the completion of this section, a dynamic expansion model by

Mayor et al. [29, 28] was found in the literature. Encouragingly, this
model applies much of the same methodology as employed here. The
model of Mayor et al. differs in that they numerically organize the com-
putation around bubbles which are tracked up the riser, quite similar to
the slug tracking models which compose the main theme of the present
work. Mayor et al. applies a Box Muller algorithm to set the bubble
lengths at the riser inlet from a normal distribution, and the slug length
from a dependant normal distribution. Additionally, iteration is needed
in this method, making it altogether more complicated than the method
of this section.
Figure A.5: Bubble expan-

sion schematic of Mayor et al.
Source: [29]
94
B Snapshots
(a) Jumpers moved, see Description C. The snapshots are taken at times (top to bottom)
t ∈ {3 220 s, 4 840 s 5 040 s, 5 460 s, 5 960 s, 6 570 s}
(b) Jumpers unmoved, see Description D. The snapshots are taken at times (top to bottom)
t ∈ {12 290 s, 12 460 s 12 870 s, 13 100 s, 135 700 s, 13 450 s}
Figure B.1: Snapshots of expansion driven instability (see Section 7.1.2 and pressure times series in Figure 7.9).
Circles included to underline void propagation and period of production
95
Figure B.2: Riser stability flow map bench-mark test – see Section 5.4. Ug,o S
= 0.280 m/s, UlS = 0.071 m/s. The
snapshots are taken at times (top to bottom) t ∈ {37.5 s, 46.5 s, 50.0 s, 55.0 s, 61.0 s}. Experimental data from
Taitel et al show this point to be stable.
Figure B.3: Terrain slugging type I (severe slugging) – see Section 7.2. The snapshots are taken at times (top
to bottom) t ∈ {5 180 s, 5 330 s, 5 920 s, 6 050 s, 6 160 s, 7 030 s}. Notice the speed of blow out.
96
LASSI
Time = 6 s
0.2
0.15
κ 0.1
0.05
0
0 0.5 1 1.5 2 2.5
z [m]
(a) φ = 0◦ , tmid = 5.80 s, tout = 13.30 s

LASSI LASSI
Time = 6 s Time = 8 s
0.2 0.2
0.15 0.15
0.1 0.1
κ
0.05 0.05
0 0
0 1 2 3 4 0 1 2 3 4
z [m] z [m]
LASSI LASSI
0.2 0.2
0.15 0.15
0.1 0.1
κ
0.05 0.05
0 0
0 1 2 3 4 0 0.5 1 1.5 2
z [m] z [m]
Figure B.4: κ as in (4.28), LASSI. V-section configuration, as in Figure 5.1. (δt = 0.01, D = 0.012, αl,i = 0.5).
Generated during the mass conservation testing of Section 5.2
97
LASSI LASSI
0.2 0.2
0.15 0.15
0.1 0.1
0.05 0.05
0
κ
κ
−0.05 −0.05
−0.1 −0.1
−0.15 −0.15
−0.2 −0.2
0 10 20 30 40 50 60 0 50 100 150 200
z [m] z [m]
= =
   
(a)  ul,i = 0.125 m/s (b)  ul,i = 0.5 m/s
   
 
 
LASSI
Time = 200 s
0.2
0.15
0.1
0.05
0
κ
−0.05
−0.1
−0.15
−0.2
0 20 40 60 80
z [m]
=
 
 TargetLength 0.1 m 
(c)  ul,i = 0.5 m/s
 

tmid = 52 s

LASSI LASSI
0.2 0.2
0.15 0.15
0.1 0.1
0.05 0.05
0
κ
0
κ
−0.05 −0.05
−0.1 −0.1
−0.15 −0.15
−0.2 −0.2
0 20 40 60 80 100 0 10 20 30 40
z [m] z [m]
= =
   
(d)  ul,i = 2.00 m/s (e)  ul,i = 0.5 m/s
   
 
 
Figure B.5: κ as in (4.28), geometry as in Figure 5.4, LASSI. (δt = 0.01, D = 0.012, Ug,i , αl,i = 0.5). Generated
during the mass conservation testing of Section 5.2
98
C Productivity Index inlet
The productivity indec boundary conditions is simplicity (despite the large system pressures). The
formulated as [12]: phase velocities are found by regarding (C.1) as an
expression for the mass flux and assuming this flux
S ,S T
Ul,i = K(Pwell − pi )/A (C.1) remain unchanged across the variation of pressures
found at the inlet
where K is the productivity index – a empirically de-
termined well parameter. It is often given in the units ∂ Ṁk,i Ug,i ρSg T
= 0 ⇒ Aρg Ug = AρSg T UgS T ⇒ S T =
[Sm3 /(bar · day)]. Water cut wc and Gas-Oil-Ratio ∂ρ Ug,i ρg,i
GOR are input parameters used to compute the inlet
water and oil superficial velocities, as in OLGA [12, additionally assuming isothermal flow
p. 46]: Ug,i PS T
ST
= (C.3)
S ,S T S ,S T Ug,i pi
Uoil,i = Ul,i (1 − wc) (C.2a)
UHS ,S2 O,i
T S ,S T
= Ul,i wc (C.2b) or alternatively found by linearising around the stan-
S ,S T
Ug,i = S ,S T
Uoil,i GOR (C.2c) dard conditions.30
Here, S denote ‘superficial’, i.e. UκS = ακ Uκ .

S T indicate that the fluid is at standard conditions. Correction In section 7.2 and 7.3, is was seen that
The same standard conditions are used as those im- this implementation introduces large errors in high-
plemented in OLGA, which are representative for pressure cases because the GOR is apparently defined
those widely used in the oil industry [12, p. 136]: as the gas-oil separation that would take place at stan-
PS T = 1 atm = 1.01325 Pa, T S T = 15.56 ◦ C = 288.71 K. dard conditions, not the separation actually found at
As the well relation (C.1) apply to the state vari- the inlet. Consequently, phase transition affects the
ables as standard conditions, a transformation to the mass rate and therefore Ṁk,i , Ṁkst . PVT-support
local state is required at the inlet. In the present is thereby required for accurate implementation of a
work, the ideal gas law is used for the PVT states for productivity index boundary condition.
D Titbits of MATLAB R code

Some useful segments of code is shown here. The code is intended to be simple and illustrative. Optimisation
has not been a main focus and loops have been used, contrary to the previous work [1].
Listing 1: General input data load

1 path = [ m a i n _ d i r , f o l d e r , ’ \ ’ , f i l e n a m e ] ;
2 if load_data ;
3 f i l e _ i d = f o p e n ( path , ’ r ’ ) ;
4 a s s e r t ( f i l e _ i d ~= −1 , ’ i n p u t d a t a n o t f o u n d ’ ) ;
5
6 d a t a = t e x t s c a n ( f i l e _ i d , [ ’%f 3 2 %f 3 2 %f 3 2 %f 3 2 %f 3 2 %f 3 2 %f 3 2 ’ , . . .
7 ’%d %f 3 2 %f 3 2 %f 3 2 %f 3 2 %f 3 2 %f 3 2 %f 3 2 %f 3 2 %f 3 2 %f 3 2 %f 3 2 ’ , . . .
8 ’%f 3 2 %f 3 2 %f 3 2 %f 3 2 ’ ] , ’ D e l i m i t e r ’ , ’ , ’ ) ;
∂Ug Ug
30
linearising around the specific volume produce (C.3) straight away as ∂(1/ρg )
= ρg
and all higher derivatives are identically zero,
U P ρSg T −ρg ζ
while linearising about ρ would produce U SgT = ∞
ζ=0 ρS T
g g
99
9 i f i s e m p t y ( d a t a { 1 } ) % Data b e g i n s a t s e c o n d l i n e o f i n p u t f i l e
10 d a t a = t e x t s c a n ( f i l e _ i d , [ ’%f 3 2 %f 3 2 %f 3 2 %f 3 2 %f 3 2 %f 3 2 %f 3 2 ’ , . . .
11 ’%d %f 3 2 %f 3 2 %f 3 2 %f 3 2 %f 3 2 %f 3 2 %f 3 2 %f 3 2 %f 3 2 %f 3 2 %f 3 2 2 ’ , . . .
12 ’%f 3 2 %f 3 %f 3 2 %f 3 2 ’ ] , ’ D e l i m i t e r ’ , ’ , ’ , ’ H e a d e r L i n e s ’ , 1 ) ;
13 end
14 fclose ( file_id );
15 end

Listing 2: Sorting data in single-time cells

1 v a r _ n a m e s . SLUGGIT = {
2 ’ Time ’ %1
3 ’z ’ %2
4 ’ Length ’ %3
5 ’ u_b ’ %4
6 ’ id ’ %5
7 ’ Hold −up ’ %6
8 ’ Pressure ’ %7
9 ’ P i p e number ’ %8
10 ’ u_l ’ %9
11 ’ u_g ’ %10
12 };
13 len_dat = length ( data {1});
14 i _ l a s t = find ( d a t a { 1 } ( 2 : l e n _ d a t ) , d a t a { 1 } ( 1 : len_dat −1) >0);
15 times = [ data {1}( i _ l a s t ) ; data {1}( len_dat ) ] ;
16 len_time = length ( times ) ;
17 prop = c e l l ( len_time , 1 ) ;
18 p r o p {1} = d a t a { p r o p _ i n d e x } ( 1 : i _ l a s t ( 1 ) ) ;
19 f o r i = 2 : l e n _ t i m e −1
20 prop { i } = d a t a { prop_index }( i _ l a s t ( i −1)+1: i _ l a s t ( i ) ) ;
21 end
22 prop { len_time } = d a ta { prop_index }( i _ l a s t ( i )+1: l e n _ d a t ) ;

Listing 3: Creating mass rate - pressure phase portraits

1 % times defined in Listing 2
2 j t _ s t a r t = f i n d ( t i m e s >= t _ s t a r t , 1 , ’ f i r s t ’ ) ;
3 dotM_out = z e r o s ( l e n _ t i m e −1− j t _ s t a r t , 1 ) ;
4 p _ i n = dotM_out ;
5 count = 0;
6 f o r j t = j t _ s t a r t : l e n _ t i m e −1
7 c o u n t = c o u n t +1;
8 % z , u , p and h o l d u p g e n e r a t e d f r o m " p r o p " i n L i s t i n g 2
9 j s t a r t = f i n d ( z { j t }> z _ i n l e t _ p , 1 , ’ f i r s t ’ ) ;
10 j e n d = f i n d ( z { j t }< z _ o u t l e t , 1 , ’ l a s t ’ ) − 1 ;
11 dotM_out ( c o u n t ) = a r e a ∗ r h o l ∗ h o l d u p { j t } ( j e n d ) ∗ u { j t } ( j e n d ) ;
12 p_in ( count ) = p{ j t }( j s t a r t ) ;
13 end
14 dotM_out_mean = mean ( dotM_out ) ;
15 p_in_mean = mean ( p _ i n ) ;
16 Mmark = ( dotM_out − dotM_out_mean ) / dotM_out_mean ;
17 pmark = ( p _ i n − p_in_mean ) / p_in_mean ;

100
Listing 4: Total liquid riser fraction - pressure phase portrait
1 j t _ s t a r t = f i n d ( t i m e s >= t _ s t a r t , 1 , ’ f i r s t ’ ) ;
2 p h i = z e r o s ( l e n _ t i m e −1− j t _ s t a r t , 1 ) ;
3 p_in = phi ;
4 count = 0;
5 for j t = j t _ s t a r t : len_time
6 c o u n t = c o u n t +1;
7 % z , s e c _ l e n , p and h o l d u p g e n e r a t e d f r o m " p r o p " i n L i s t i n g 2
8 j i n l e t = f i n d ( z { j t }> z _ i n l e t _ p , 1 , ’ f i r s t ’ ) ;
9 j s t a r t = f i n d ( z { j t }>= z _ r i s e r _ i n l e t , 1 , ’ f i r s t ’ ) ;
10 j e n d = f i n d ( z { j t }<= z _ r i s e r _ i n l e t +H, 1 , ’ l a s t ’ ) ;
11
12 %I n t e r p o l a t e s e c t i o n s a t r i s e r i n l e t / o u t l e t
13 dz_in = z{ j t }( j s t a r t )− z _ r i s e r _ i n l e t ;
14 d z _ o u t = ( z _ r i s e r _ i n l e t +H) − z { j t } ( j e n d ) ;
15 weight_in = dz_in / sec_len { j t }( j s t a r t −1);
16 weight_out = dz_out / sec_len { j t }( jend ) ;
17
18 p_in ( count ) = p{ j t }( j i n l e t ) ;
19 p h i ( c o u n t ) = ( w e i g h t _ i n ∗ h o l d u p { j t } ( j s t a r t − 1 ) . ∗ s e c _ l e n { j t } ( j s t a r t −1) . . .
20 + weight_out ∗ holdup { j t }( jend ) . ∗ sec_len { j t }( jend ) . . .
21 + sum ( h o l d u p { j t } ( j s t a r t : j e n d − 1 ) . ∗ s e c _ l e n { j t } ( j s t a r t : j e n d − 1 ) ) ) / H ;
22 end
23 phi_mean = mean ( p h i ) ;
24 p_in_mean = mean ( p _ i n ) ;
25 p h i m a r k = ( p h i − phi_mean ) / phi_mean ;
26 pmark = ( p _ i n − p_in_mean ) / p_in_mean ;

Listing 5: Integrating liquid fraction in riser through time

1 % I n t h i s code , t h e l i q u i d f r a c t i o n any e v e r t y t i m e s t e p i s
2 % p r o j e c t e d down t o an a v e r a g e r i s e r l e n g t h H w i t h s e c t i o n
3 % l e n g t h d z (= 1 ) . I n t e r p o l a t i o n i s u s e d a s d a t a b o r d e r s do
4 % n o t c o i n s i d e w i t h " mean r i s e r " b o r d e r s .
5 dz = 1 ;
6 d t = t i m e s ( 2) − t i m e s ( 1 ) ; %d t c o n s t i n o u t p u t f i l e
7 x = 0 : dz : H ;
8 z _ r i s e r = l e n g t h _ h o r i z o n t a l : dz : l e n g t h _ h o r i z o n t a l +H ;
9 holdup_profile_sum = zeros (1 , length ( z _ r i s e r ) ) ;
10 includecount = 0;
11 notincludedcount = 0;
12
13 f o r j t = round ( t _ s t a r t / d t ) : l t
14 % z and h o l d u p g e n e r a t e d f r o m " p r o p " i n L i s t i n g 2
15 z _ r i s e r = z { j t }− l e n g t h _ h o r i z o n t a l ;
16 j = 1;
17 f o r i = f i n d ( z _ r i s e r <= x ( 1 ) , 1 , ’ l a s t ’ ) : l e n g t h ( z _ r i s e r ) −1
18 j n e x t = f i n d ( x >= z _ r i s e r ( i + 1 ) , 1 , ’ f i r s t ’ ) ;
19 i f i s e m p t y ( j n e x t ) , break ; end
20 h o l d u p _ p r o f i l e _ s u m ( j : j n e x t −2) = h o l d u p _ p r o f i l e _ s u m ( j : j n e x t −2)+ h o l d u p ( i ) ;
21 j = jnext ;
22 h o l d u p _ p r o f i l e _ s u m ( j −1) = h o l d u p _ p r o f i l e _ s u m ( j −1) + . . .
23 ( ( z _ r i s e r ( i +1)−x ( j − 1 ) ) ∗ h o l d u p ( i ) + ( x ( j ) − z _ r i s e r ( i + 1 ) ) ∗ h o l d u p ( i + 1 ) ) / dz ;
24 end
101
25 h o l d u p _ p r o f i l e _ s u m ( j : end ) = h o l d u p _ p r o f i l e _ s u m ( j : end ) + h o l d u p ( i + 1 ) ;
26 i n c l u d e c o u n t = i n c l u d e c o u n t +1;
27 end
28 alpha_profile = 1 − holdup_profile_sum / includecount ;

Listing 6: IKH/well posedness criteria at time t across pipeline

1 % z , Ul , Ug , s e c _ l e n , p and h o l d u p g e n e r a t e d f r o m " p r o p " i n L i s t i n g 2
2 % s e e same l i s t i n g f o r d e f . o f ’ t i m e s ’
3 t_index = find ( times > t , 1 , ’ f i r s t ’ ) ;
4 % remove s l u g s e c t i o n
5 H = holdup { t_index };
6 s t r a t i f i e d = H<1;
7 H _ s t r a t = H( s t r a t i f i e d ) ;
8 z = z{ t_index }( s t r a t ) ;
9 L = sec_len { t_index }( s t r a t ) ;
10 r h o g = p { t _ i n d e x } ( s t r a t ) / ( R∗T ) ;
11 u l = Ul { t _ i n d e x } ( s t r a t ) ;
12 ug = Ug{ t _ i n d e x } ( s t r a t ) ;
13
14 [ pipezs , angles ] = g e t p h i ( ) ; % f u n c t i o n holding geometry data in array
15 phi = zeros ( length ( H_strat ) , 1 ) ;
16 for k = 1: length ( H_strat )
17 for j = 1: length ( pipezs )
18 i f z (k) < pipezs ( j )
19 phi ( k ) = angles ( j ) ;
20 break
21 end
22 end
23 end
24 cosphi = cosd ( phi ) ;
25
26 % interphase angle through s e r i e s expansion :
27 gamma = p i ∗ H _ s t r a t + ( ( 3 ∗ p i / 2 ) ^ ( 1 / 3 ) ) ∗ . . .
28 ( 1−2∗ H _ s t r a t + H _ s t r a t . ^ ( 1 / 3 ) − ( 1 − H _ s t r a t ) . ^ ( 1 / 3 ) ) ;
29 % or through i n t e r a t i v e procedure :
30 % for j = 1: length ( H_strat )
31 % f _ t h e t a = @( gamma ) ( 2 ∗ p i ) \ ( gamma − s i n ( gamma ) ) − H _ s t r a t ( j ) ;
32 % theta = fzero ( f_theta , pi / 2 ) ;
33 % end
34
35 d A l d h l = 2∗ r ∗ s i n ( gamma ) ;
36 u g l s q = ( ug− u l ) . ^ 2 ;
37 % IKH
38 RHS = ( r h o l ∗(1 − H _ s t r a t )+ r h o g . ∗ H _ s t r a t ) . ∗ . . .
39 ( ( r h o l −r h o g ) . / ( r h o l . ∗ r h o g ) ) ∗ G. ∗ c o s p h i ∗ a r e a . / d A l d h l ;
40 % well_posed
41 %RHS = ( r h o l −r h o g ) . ∗ G . ∗ ( H / r h o l + (1 −H ) . / r h o g ) . ∗ c o s p h i ∗ a r e a . / d A l d h l ;

Listing 7: Numeric riser void fraction model

1 f u n c t i o n [ mean_alpha , z ] = a l p h a _ f i l m _ n u m e r i c _ m i x t u r e ( Mdotg , Mdotl )
2
3 g l o b a l H a r e a R T G Po Uv r h o _ l
102
4
5 N = 5∗H ;
6 dz = H / N ;
7 z = −dz : − dz : −H ;
8 Gz = G∗ dz ;
9 m e a n _ a l p h a = z e r o s (N , 1 ) ;
10 r h o _ g = z e r o s (N , 1 ) ;
11
12 Uc1 = Mdotg ∗R∗T / a r e a ;
13 Uc2 = Uv+ 1 . 2 ∗ Mdotl / ( a r e a ∗ r h o _ l ) ;
14 f o r j = 1 :N
15 p = Po + Gz ∗ ( sum ( m e a n _ a l p h a ( 1 : j − 1 ) . ∗ ( r h o g ( 1 : j −1) − r h o _ l ) ) + r h o _ l ∗ ( j −1) ) ;
16 r h o _ g ( j ) = p / ( R∗T ) ;
17 m e a n _ a l p h a ( j ) = Uc1 / ( Uc2∗ p + Uc1 ) ;
18 end
19 end

Listing 8: Analytic riser void fraction model

1 f u n c t i o n [ mean_alpha , z ] = a l p h a _ f i l m _ n u m e r i c _ m i x t u r e ( Mdotg , Mdotl )
2
3 g l o b a l H a r e a R T G Po Uv r h o _ l
4
5 N = 5∗H ;
6 dz = H / N ;
7 z = −dz : − dz : −H ;
8 Gz = G∗ dz ;
9 m e a n _ a l p h a = z e r o s (N , 1 ) ;
10 r h o _ g = z e r o s (N , 1 ) ;
11
12 Uc1 = Mdotg ∗R∗T / a r e a ;
13 Uc2 = Uv+ 1 . 2 ∗ Mdotl / ( a r e a ∗ r h o _ l ) ;
14 f o r j = 1 :N
15 p = Po + Gz ∗ ( sum ( m e a n _ a l p h a ( 1 : j − 1 ) . ∗ ( r h o g ( 1 : j −1) − r h o _ l ) ) + r h o _ l ∗ ( j −1) ) ;
16 r h o _ g ( j ) = p / ( R∗T ) ;
17 m e a n _ a l p h a ( j ) = Uc1 / ( Uc2∗ p + Uc1 ) ;
18 end
19 end

103
Annexes
E The Kelvin-Helmholtz stability criteria

Figure E.1 shows a photo of the onset of Kelvin-Helmholtz instability of stratified shear flow. The main phys-
ical mechanism behind the phenomenon is the Bernoulli suction which appears in the gas cross section as the
gas is accelerated past a small perturbation of the smooth liquid surface. If the conditions are such that this
suction is sufficient to amplify the perturbation, the flow conditions are unstable.
Figure E.1: Kelvin-Helmholtz instability of stratified

shear flow. Source: [44].
Barnea and Taitel presented in 1993 a linear stability analysis of the two-fluid model [3, 37]. As all models
discussed in this work rely on the fundamental equations of this model, its stability domain is of interest.
The two-fluid model consists of two pairs of mass and momentum conservation equations, one for each
fluid phase (4.26). A small disturbance in the stratified flow regime is considered. This disturbance is presented
as a small perturbation in the state variables from their equilibrium states. This perturbation can be expressed
as a Fourier sum of modes
∞
X n o
χ(t, z) = χ + χζ exp i ωζ t − kζ z ;
e χ ∈ {ug , ul , αg , αl , hl , p}
ζ=0
where χ is representative of the state variables. kζ and ωζ are respectively the mode wavenumber and its
corresponding pulsation. It is clear from this expression that if there exists a wavenumber k with a corresponding
pulsation ω whose imaginary part is negative, then this imaginary pulsation part will cause an amplification of
this mode over time, resulting in instability. To find whether such a wavenumber exists, all frequencies of
the disturbance may be considered individually. Therefore, the analysis proceeds by inserting a single Fourier
mode and allow its wavenumber k to take on any value form zero to infinity. If the state equilibrium variables
χ are such that there exists a wavenumber k whose corresponding pulsation ω consists of a negative imaginary
part. Accordingly, one proceeds by eliminating the pressure term in the two-fluid model (4.26) by combining
the two momentum equations, inserting the mode
χ(t, z) = χ + e
χ exp {i (ωt − kz)} ; χ ∈ {ug , ul , αg , αl , hl , p}
for the state variables, and deduct the equilibrium part. After linearising around the resulting volumetric force
FV on the liquid phase (see Equation (4.29),) the following dispersion equation for the pulsation ω is obtained
ω2 − 2[ak − ib]ω + ck2 − dk4 − iek = 0 (E.1)
where a, b, c and e are combinations of the equilibrium state variables using the original notation of Barnea and
104
Taitel [3].
ρl ρg
"
ρl ul 2 ρg ug 2
dα −1 #
ρ∗ = αl + αg c= 1
ρ∗ αl + αg − ρl − ρg g cos φ dh l
l
ρg ug
dα −1
ρ l ul σ
a= 1
ρ∗ +
αl αg d= ρ∗
l
dhl
 
 ∂FV
∂FV
∂F
b= 1
e = − ρ1∗
 V
2ρ∗
 S
∂ul
− ∂uSg α ,uS
 ∂αl α ,uS
αl ,uSg g l l g
adopting the liquid volumetric force FV from (4.29) and σ is the surface tension. Since the neutral stability limit
occurs where the imaginary part of the pulsation ω turns from a positive to a negative value, the critical states
will have ={ω} = 0. In the following then, ω can be considered a purely real value.
Solving (E.1) for its imaginary parts yields a criterion critical wave velocity ω/k
ω e
=
k 2b
and solving for the real part of (E.1), using this critical wave velocity, yields a stability criterion:
e 2
stable ⇔ For ∀ k : − a + c − a2 − dk2 < 0
2b
The last LHS term is the only term involving surface tension, and is also the only term which is a multiple of the
wavenumber. Consequently, surface tension is only important for large wavenumbers (short waves). Obviously,
since d is always positive, k → 0 is the determining wavenumber, producing the final viscous Kelvin-Helmholtz
(VKH) criterion:
e 2
stable ⇔ − a + c − a2 < 0 (E.2)
2b
The inviscid Kelvin-Holmholtz criterion (IKH) is easily found inserting the equilibrium state variables into
(E.2), and neglecting the viscous and surface tension terms. The result is a simple imbalance where the hydro-
static forces must outweigh the Bernoulli suction for stability [3, 13]:
ρl − ρg
!−1
dαl
stable ⇔ (ug − ul )2 < (αg ρl + αl ρg ) g cos φ (E.3)
ρl ρg dhl
In the viscous Kelvin-Helmholtz criterion (VKH), the RHS of (E.3) is multiplied with a viscosity-dependant
factor less than unity, which constitutes a smaller model stability domain. Rewriting slightly reveals the simi-
larity to the well-posedness criterion (4.30):
αg αl
" # !−1
dαl
stable ⇔ (ug − ul ) < (ρl − ρg )g + 2
cos φ
ρl ρg dhl
F Simplified models of void in riser
Preceding the riser void model of Section A, a rate (see e.g. Figure 5.11), but gave poor predictions
model was made in which the liquid film around Tay- otherwise. Since it was already documented, and
lor bubbles was neglected altogether. It was found because its character gives some insight to the liquid
that this approximation was good for cases in which film influence, it has not been removed completely
the gas flow rate was low relative to the liquid flow from this text.
105
where
The Assumptions D from Appendix A are re- 1.2 · Ml Ṁg
a(p j ) = 1 + +
tained while including an assumption on the bubble ρl AUν p j AUν
liquid film: Ṁg
b(p j ) =
p j AUν
Assumptions E. Simplified riser steady state void
model and (± ⇒ −) is the only physical solution.31 Pres-
sure p j and total average riser void fraction hΦg i can
E.1 Assumptions D still holds. then be obtained the same way as in Appendix A.1.
Void fraction profiles of this procedure can be found
E.2 The local slug liquid velocity can be approxi- in Figures A.4, 5.11 and 5.6.
mated by the local average liquid velocity (i.e.
the influence of the velocity difference of the As a final note, the special case of Ṁl → 0 re-
bubble liquid film is neglected) quires some special attention. Figure A.4a shows
Assumption D.1 gives: that the numerical solution to the void fraction ap-
proaches unity as Ṁl → 0. The reason is that the the
Ṁκ = A(ρuhαi)κ = const (F.1) liquid velocity is intrinsically a part of this model so
that ul and therefore ug increases proportionally. For
and Assumption D.4 allows for ug = vb,n where cases where Ṁl = 0 and Ṁg > 0.5292, other states
vb,n is the bubble nose velocity from Section 4.8. To being the same, will give a gas volume at the out-
compute it, the Bendiksen model from Section 4.8 let larger than can be accompanied in the pipe sec-
is used, assuming no backflow and a 90◦ inclination, tion at velocity ug = Uν (i.e. b > 1), for which the
along with (F.1) for the liquid phase to approximate model is not defined (violating Assumption D.5.)The
u s,m (Assumption E.2) elementary solution hαg i = 1 ∀ z is then found. Fig-
ure F.1 show the development of the gas volume frac-
Ṁl tion hαg (z)i with no liquid flow as the gas mass flux
ug = 1.2 − Uν (F.2)
Aρl (1 − hαg i) increases towards where it is ill-defined. The void
fractions are respectively 0.085, 0.184, 0.303, 0.461,
where Uν = 0.35 gd was the empirically determined
p
0.750, 1.000 for increasing gas flux, where the last is
bubble nose velocity in vertical pipes in still liquid. not a valid solution.
As before, both a numerical and an analytical ap-
1.4
proximation is found.
Gas volume fraction profile
1.2
Ṁl = 0
Ṁg
F.1 Numerical calculation 0.6
1
As before, the riser is discretised j ∈ {0, N} from out-

0.8
let to inlet as in scetch A.1. Equation (F.1) gives, αg
when inserting the ideal gas law, (F.2) for the gas ve- 0.6
locity and (F.1) for the liquid velocity: 0.5
0.4
p !
j Ṁl
Ṁg = Ahαg ij 1.2 + Uν 0.2
0.4
RT (1 − hαg ij )ρl A 0.3

0.2
0.1
0
After some rearranging, this yields an expression for −1400 −1200 −1000 −800
z [m]
−600 −400 −200 0
hαg i
Figure F.1: Gas volume fraction hαg i profiles with no
q
aj ± a2j − 4b j
hαg ij = (F.3) liquid flow
2
In the case were Ṁl is exactly zero and Ṁg is larger than about 0.5 kg/s, the root becomes (1 − b)2 with b > 1 and so (± ⇒ +) is
p
31
the hydrostatic solution, but gives an ill-defined αg > 1
106
where Uν = 0.35 gd was the empirically determined
p
F.2 Analytical approximation
bubble nose velocity in vertical pipes in still liquid.
Simplified expressions may also be developed using
To simplify matters, z is defined positive up-
only inlet and outlet properties. Using the superficial
stream, along the gravitational force (see Fig-
gas velocities uSb = ub hαg i and (F.1) yields:
ure A.1). Accordingly, the pressure gradient be-
∂ Ṁg 1 ∂uS 1 ∂ρg 1 ∂p comes:
=0 ⇔ S b = = (F.4) ∂p
∂z ub ∂z ρg ∂z RT ρg ∂z = gρl (1 − hαg i) (F.6)
∂z
To compute the bubble nose velocities ub = vb,n ,
the Bendiksen model from Section 4.8 is used, as- Using (F.1) for ρg and (F.6) for ∂z p, (F.4) can be writ-
suming no backflow and a 90◦ inclination, along with ten
1 ∂uSb (Aub )αg (1 − hαg i)Gρl
(F.1) for the liquid phase to approximate u s,m (As- =−
sumption E.2) v ∂z
S
RT Ṁg
Ṁl which can be rearranged and integrated to give the

ub = 1.2 − Uν (F.5) lift gas volume in the riser
Aρl (1 − αg )
S (H)
ZH ZH ∂uSb vZ
RT Ṁg ∂z dz RT Ṁg duSb
Vg = hΦg iAH = hαg iA dz = − = (F.7)
vS ub (1 − hαg i)

gρl gρl vS 2 1 −
1
0 0 uSb (0) hαg i
The final integral in (F.7) is not trivial to solve analytically. To simplify matters, the gas fraction αg is substituted
with a frozen mean approximation value
1h i
hαg (z)i ≈ hαg i = hαg (H)i + hαg (0)i (F.8)
2
which gives
uSb (H)
duSb RT Ṁg αg
 
RT Ṁg hαg i
Z
1 1 
=

hΦg iAH ≈ − (F.9)
gρl 1 − αg
 
gρl 1 − hαg i vS 2
uSb (H) uSb (0)

uSb (0)
This is not a valid assumption when the gas vol- An iterative solving procedure can then be used in
ume fraction approaches unity! Further, (F.1) gives which (F.5) is solved for both phases. The riser
an implicit expressions for αg (0) and αg (H) with re- inlet pressure p(H) is found using (F.11), which is
spect to the outlet pressure po when substituting with in turn used in finding the gas fractions hαg i from
the ideal gas law: (F.10). From here, the mean fraction hαg i can be
Ṁg RT 1 computed from (F.8), and the riser gas volume
hαg (0)i = (F.10a) (from the gas lift) Vg from (F.9). About ten itera-
Apo ub (0)
tions seems to be sufficient to reach an accuracy for
Ṁg RT 1 1
hαg (H)i = (F.10b) |∆hΦg ii /hΦg ii | < 10−6
A p(H) ub (0)
The riser gas volume fraction Vg from (F.9) is

For inlet liquid mass flow rates less than 10 kg/s
also an implicit expression in terms of the riser inlet
( Ṁg = 0.658 kg/s), or gas mas flow rate grater than
pressure p(H), which, for simplicity, is approximated
1.8 kg/s ( Ṁl = 33.69 kg/s), the iteration procedure with,
by neglecting the gas mas
more or less, arbitrary initial values begin to diverge.
Z H
p(H) ≈ po + ρg g(1 − hαg i) dz
This probably has to do with the fact that the gas
0
(F.11) volume fraction modelled by Ṁg RT/(Apub ) becomes
= po + ρl gH 1 − hΦg i larger than one at the outlet as the liquid velocity
107
approaches zero. Also, the mean gas volume fraction which need no simplification as in (F.9). The log-
hαg i simplification may influence the result. arithm arise as keeping v constant produce a void
gi
R dhα
integral on the form hαg i(1−hα g i)
which can easily be
As an ending note, for steady liquid, ub = Uν =

solved by substituting dαg with d 1/hαg i − 1 . . .
0.35 gd and ∂z ub = 0 giving a system on the form
p
Ṁg RT hαg (0)i hαl (H)i

hΦg i = ln (F.12)
AHUν gρl hαg (H)i hαl (0)i
108

Fulltext01

Uploaded by

Copyright:

Available Formats

Fulltext01

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Fulltext01

Uploaded by

Copyright:

Available Formats

Simulation of Unstable Two-phase Flows

Andreas Holm Akselsen

Master of Energy and Environmental Engineering

Norwegian University of Science and Technology

June 13, 2012

2 Definitions and terminology 17

3 Operational instabilities in vertical risers 18

8 Discussion and recommendations 80

A A model of void in risers 90

C Productivity Index inlet 99

D Titbits of MATLAB R code 99

E The Kelvin-Helmholtz stability criteria 104

F Simplified models of void in riser 105

n Normal vector of control volume −

Ṁ s Mass rate source term kg/s

Q Volumetric flow rate m3 /s

A Pipe area, perpendicular to flow direction m2

c Speed of sound m/s

g Gravitational acceleration m/s2

IF M Interphase Friction Multiplier −

K Productivity index Sm3 /(bar · day)

M Control volume mass kg

m Pipe specific mass (extensive) kg/m3

N Total number of samples −

Pwell Well pressure bar

Q ‘Extensive volumetric flux’ – see (4.3) m3 /s

R Gas constant J/(kg · K)

sχ Sample standard deviation unit of χ

u Phase velocity m/s

v Border velocity m/s

VR dz Total phase volume in system, computed using (5.1a) m3

Vio Total phase volume in system, computed using (5.1b) m3

wc water cut = UHS ,S2 O,i

y Vertical axis in pipe cross-section. Origin at pipe bottem m

z Position downstream the pipe m

α Section volume fraction (hold-up) −

χ Representation of a property various

δ Difference (differential approximation) −

` Control volume streamwise length m

λ Darcy friction coefficient −

µ Dynamic viscosity kg/(m · s)

Ω Control volume domain m3

∂Ω Boundary of control volume m2

Φ Total average riser phase fraction −

φ Pipe inclination angle (deg)

σ Surface tension N/m

τ Shear stress N/m2

Θ Total phase fraction in horizontal pipeline upstream riser −

κ Fluid type: κ ∈ {l, g}

Σ Control volume cell: Σ ∈ {W, E}

σ Control volume border: σ ∈ {w, e}

crit ‘critical’ (criteria limit)

E Eastern (right) control volume cell

e Eastern (right) control volume border

f Liquid film (around bubble)

int ‘interfacial’ (surface)

j Position index along the entire pipe

P Present control volume cell

pipe pipe (interior surface)

Vh AP,κ ∆h + b̃P 21 (he + hw ) ∆h − b̃P hP ∆h = AP,κ ∆h