api.research-repository.uwa.edu.au · ii Abstract Multiphase flow metering provides vital flow...
Transcript of api.research-repository.uwa.edu.au · ii Abstract Multiphase flow metering provides vital flow...
Development of an Earth’s Field Nuclear
Magnetic Resonance Multiphase Flow Meter
Keelan Thomas O’Neill, B. Eng. (Hons)
This thesis is presented for the degree of Doctor of Philosophy of
The University of Western Australia
School of Mechanical and Chemical Engineering
Fluid Science and Resources Division
2019
i
Thesis Declaration
I, Keelan Thomas O’Neill, certify that:
This thesis has been substantially accomplished during enrolment in the degree.
This thesis does not contain material which has been submitted for the award of any other
degree or diploma in my name, in any university or other tertiary institution.
No part of this work will, in the future, be used in a submission in my name, for any other
degree or diploma in any university or other tertiary institution without the prior approval
of The University of Western Australia and where applicable, any partner institution
responsible for the joint-award of this degree.
This thesis does not contain any material previously published or written by another
person, except where due reference has been made in the text and, where relevant, in the
declaration that follows.
The work(s) are not in any way a violation or infringement of any copyright, trademark,
patent, or other rights whatsoever of any person.
This thesis contains published work and/or work prepared for publication, some of which
has been co-authored.
Signature:
Date: 21st April 2019
ii
Abstract
Multiphase flow metering provides vital flow information essential for the
development and operation of oil and gas fields. Current multiphase flow meters
(MPFMs) often do not achieve the desired measurement performance in order to provide
reliable information for reservoir management, production monitoring and flow assurance
purposes. Therefore a market exists to produce a reliable, accurate and robust MPFM.
Nuclear magnetic resonance (NMR) offers attractive technology for multiphase flow
metering. NMR is a non-invasive measurement technique which can provide quantitative
characterisation of fluid behaviour.
This thesis presents the development of the apparatus and measurement
techniques relevant to an Earth’s field nuclear magnetic resonance (EFNMR) MPFM.
The flow meter consists a mobile pre-polarising permanent magnet positioned upstream
of an EFNMR radio frequency coil. The system excites a 1H NMR signal (at a Larmor
frequency of 2.29 kHz) from fluid moving through the flow meter. The free induction
decay (FID) signal is then acquired and appropriate signal interpretation techniques allow
determination of the composition and flowrate of the measured fluid stream.
The initial flow meter development focusses on establishing an appropriate
velocity measurement technique. The FID acquisition time is identified as a suitable
experimental parameter to ensure velocity measurement sensitivity and accuracy.
Velocity probability distributions are determined by fitting the measured FID signal of a
flowing water stream with a NMR fluid model using Tikhonov regularisation. The
accuracy of the measured velocity distributions are validated against in-line rotameter
measurements of water flowrate. The measured velocity distributions have a comparable
shape to theoretical turbulent distributions. The measurement technique demonstrates
robust behaviour with the ability to accurately capture artificially constructed multi-
modal (two-pipe) velocity distributions.
The next stage of system development involved measuring flow information in
two-phase gas/liquid flows. The flow meter is used to accurately measure the liquid
velocity distribution in air/water flows in the stratified and slug flow regimes. The signal
iii
interpretation technique is then applied in a ‘single-shot’ analysis to provide real-time
tracking of the liquid holdup and velocity (at 0.55 – 1.25 Hz). The NMR estimate of
liquid holdup is shown to be comparable against a simultaneous video track of holdup.
Temporal variation in the liquid flowrate can be used as a metric to distinguish stratified
and slug flow.
The final stage of flow meter development involves applying the flow meter to
measure two-phase canola oil/water flows. The metering methodology is extended via
the use of a dual polarisation scheme which utilises the in-built electromagnetic pre-
polarising coil as well as the mobile permanent magnet. The combined polarisation
technique provides sensitivity to a range (50 ms – 3 s) of fluid spin-lattice relaxation times
(T1). The fluid relaxation behaviour is used as a distinguishing mechanism to quantify
the fractions of canola oil and water in two-phase flow. A multi-FID sequence (with
variable polarisation conditions) is developed to provide simultaneous measurement of
fluid velocity and spin-lattice relaxation behaviour. Measured multi-FID signals are fit
with a joint 2D velocity-T1 distribution using the developed NMR flow model via a 2D
Tikhonov regularisation algorithm. Analysis of the measured 2D distributions enables
measurement phase fractions and flowrate. Measurements are performed on two-phase
canola oil/water flows across a range of flow regimes. Flowrate measurements in the
stratified flow with mixing, dispersion of oil-in-water and water, and oil-in-water
emulsion flow regimes demonstrate excellent accuracy relative to independent
measurements of individual oil and water phase flowrates. Furthermore; the measured
2D velocity-T1 distributions captured multiphase flow characteristics such as velocity
slip. However the developed metering system could not be accurately validated for flow
regimes where water-in-oil emulsions are present. Such emulsions cannot be sufficiently
separated during the total measurement time which results in flowrate quantification
errors.
The EFNMR MPFM demonstrates very good flow measurement performance for
single-phase flows, two-phase gas/liquid flows and two-phase oil/water flows. The work
presented in this thesis provides an excellent foundation for further developing the
measurement system towards a fully capable NMR MPFM.
iv
Contents
Thesis Declaration ............................................................................................................ i
Abstract ............................................................................................................................ ii
Contents .......................................................................................................................... iv
List of Figures ................................................................................................................ vii
List of Tables .................................................................................................................. ix
Abbreviations .................................................................................................................. x
Nomenclature ................................................................................................................. xi
Acknowledgements........................................................................................................ xv
Authorship Declaration for Co-authored Publications ........................................... xvii
Chapter 1. Introduction and motivation .................................................................... 1
Chapter 2. Nuclear magnetic resonance .................................................................... 6
2.1 Fundamentals of NMR ....................................................................................... 6 2.1.1. Introduction to NMR ................................................................................... 6
2.1.2. Fundamental NMR Theory ......................................................................... 6 2.1.3. The vector model ........................................................................................ 9 2.1.4. Measuring the free induction decay .......................................................... 10
2.1.5. Spin-lattice relaxation ............................................................................... 12 2.1.6. Spin-spin relaxation .................................................................................. 14
2.2 Application of low-field NMR ......................................................................... 17 2.2.1. Low-field NMR......................................................................................... 17
2.2.2. Earth’s Field NMR .................................................................................... 18
2.3 Inversion techniques in NMR........................................................................... 22
2.3.1. 1D Tikhonov regularisation ...................................................................... 22 2.3.2. 2D Tikhonov regularisation ...................................................................... 24
Chapter 3. Multiphase flow ....................................................................................... 26
3.1 Multiphase Flow Concepts ............................................................................... 26 3.1.1. Multiphase flow information .................................................................... 27 3.1.2. Gas-liquid flows ........................................................................................ 28 3.1.3. Liquid-liquid flows ................................................................................... 31
3.2 Multiphase Flow Metering ............................................................................... 35 3.2.1. Introduction to multiphase flow metering ................................................. 35 3.2.2. Classification of multiphase flow meters .................................................. 36
3.2.3. Velocity measurement principles .............................................................. 37 3.2.4. Phase fraction measurement technologies ................................................ 38 3.2.5. NMR in flow measurement ....................................................................... 43 3.2.6. Validation of flow measurement accuracy ................................................ 46
Chapter 4. Quantitative velocity measurements of single-phase flows ................. 47
4.1 Introduction ...................................................................................................... 47
4.2 Equipment and methodology............................................................................ 47
v
4.2.1. Experimental apparatus ............................................................................. 47 4.2.2. NMR signal model .................................................................................... 51
4.2.3. Velocity probability distributions via regularisation................................. 53 4.2.4. Theoretical turbulent velocity distributions .............................................. 54
4.3 Results and discussion ...................................................................................... 55
4.3.1. Optimum Velocity Determination Methodology ...................................... 55 4.3.2. Single Pipe Experiments ........................................................................... 57 4.3.3. Two pipe experiments ............................................................................... 61
4.4 Conclusions ...................................................................................................... 64
Chapter 5. Characterising two-phase gas/liquid flows ........................................... 65
5.1 Introduction ...................................................................................................... 65
5.2 Equipment and methodology ............................................................................ 66 5.2.1. Equipment for gas/liquid flows ................................................................. 66
5.2.2. Gas/liquid NMR signal interpretation ....................................................... 67
5.3 Results and discussion ...................................................................................... 68 5.3.1. Calibrating the Reference Signal Magnitude ............................................ 68 5.3.2. Gas/liquid time-averaged results ............................................................... 69 5.3.3. Gas/liquid single-scan measurements ....................................................... 72
5.3.4. Statistical analysis of single-scan results .................................................. 79
5.4 Conclusions ...................................................................................................... 81
Chapter 6. Developing a technique for simultaneous oil and water flowrate
measurement .................................................................................................................. 82
6.1 Introduction ...................................................................................................... 82
6.2 Equipment and methodology ............................................................................ 82
6.2.1. NMR fluid quantification of oil/water systems ......................................... 82
6.2.2. Equipment for oil/water measurements .................................................... 83
6.2.3. T1 measurements of stationary fluids ........................................................ 87 6.2.4. Measuring T1 of flowing fluids ................................................................. 91
6.2.5. NMR flow model development for signal polarisation ............................ 95 6.2.6. Velocity–T1 distribution determination via 2D inversion ......................... 97 6.2.7. Component flowrate calculations .............................................................. 99
6.3 Experimental procedure ................................................................................. 100
6.3.1. Choice of experimental parameters ......................................................... 100 6.3.2. Experimental measurement workflow .................................................... 101
6.4 Results and discussion .................................................................................... 104 6.4.1. Single-phase measurements – Water ...................................................... 104 6.4.2. Single-phase measurements – Oil ........................................................... 110
6.5 Conclusions .................................................................................................... 117
Chapter 7. Quantitative characterisation of two-phase oil/water flows .............. 119
7.1 Introduction .................................................................................................... 119
7.2 Experimental methodology ............................................................................ 119 7.2.1. Flow regimes ........................................................................................... 119
7.2.2. Fluid separation ....................................................................................... 122
7.3 Results and discussion .................................................................................... 124 7.3.1. Stratified flow with mixing ..................................................................... 124 7.3.2. Dispersion of oil-in-water & free water layer ......................................... 126 7.3.3. Oil-in-water emulsion ............................................................................. 127
vi
7.3.4. Dispersion of oil-in-water & Dispersion of water-in-oil ........................ 128 7.3.5. Water-in-oil emulsion ............................................................................. 131
7.3.6. Summary and discussion of results ......................................................... 132
7.4 Conclusions .................................................................................................... 136
Chapter 8. Conclusions and future work ............................................................... 137
8.1 Conclusions .................................................................................................... 137
8.2 Future work .................................................................................................... 138 8.2.1. Improving the flow loop equipment ........................................................ 139 8.2.2. NMR magnetic hardware improvements ................................................ 139 8.2.3. Advancing NMR flow measurement techniques .................................... 141
References .................................................................................................................... 143
Appendices ................................................................................................................... 167
Appendix A. Calibrations for NMR flow measurements ..................................... 167
A.1 Magnetic field inhomogeneity........................................................................ 167
A.2 Relaxation in variable magnetic field strengths ............................................. 170
A.3 Signal diminishment during repolarisation .................................................... 173
Appendix B. Fluid relaxometry .............................................................................. 184
B.1 T2 measurements of canola oil and water ....................................................... 184
B.2 Comparative relaxometry measurements ....................................................... 186
Appendix C. Flow system design considerations .................................................. 188
C.1 Separation of oil and water emulsions ........................................................... 188
C.2 Influence of the Halbach array stray field ...................................................... 190
vii
List of Figures
Figure 1.1 Schematic of a typical offshore production scenario. ................................................ 2 Figure 1.2 The Schlumberger Vx multiphase flowmeter. ........................................................... 4 Figure 2.1 Zeeman diagram illustrating the energy states for a 1H atom with the energy
difference between the states indicated. .................................................................... 8 Figure 2.2 Illustration of Larmor precession. ........................................................................... 10 Figure 2.3 Illustration of a pulse and collect sequence used to measure a free induction decay.
................................................................................................................................. 11 Figure 2.4 Classical mechanics representation of the stages of a pulse and collect experiment.
................................................................................................................................. 12 Figure 2.5 Inversion recovery method. ..................................................................................... 14 Figure 2.6 Demonstration of the CPMG experiment. ............................................................... 17 Figure 2.7 Two alternative solutions for pre-polarisation......................................................... 21 Figure 3.1 Example flow regime map for air-water flow in a horizontal pipe. ........................ 30 Figure 3.2 Flow regime map for oil-water flow in a horizontal pipe. ....................................... 34 Figure 3.3 Illustration of the cross-correlation principle. ......................................................... 38 Figure 3.4 Schematic of the gamma-ray attenuation measurement principle. .......................... 39 Figure 3.5 Strip-type electrode configuration for impedance measurements of multiphase flows.
................................................................................................................................. 41 Figure 3.6 A cut-open schematic of the M-Phase 5000 magnetic resonance multiphase flow
meter. ....................................................................................................................... 45 Figure 4.1 Illustrations of the single-phase experimental flow loop apparatus. ....................... 49 Figure 4.2 Illustration of the two-pipe experimental flow loop. ............................................... 50 Figure 4.3 Velocity distribution measurements obtained using separation distance as the
independent variable. ............................................................................................... 56 Figure 4.4 Velocity distribution measurements obtained using FID acquisition time as the
independent variable. ............................................................................................... 58 Figure 4.5 Comparison of the mean velocity predicted using regularisation of NMR data to the
measured mean velocity from in-line rotameter data. ............................................. 59 Figure 4.6 Comparison of velocity probability distribution standard deviations. ..................... 60 Figure 4.7 Velocity measurements for the two-pipe flow system. ........................................... 62 Figure 4.8 Comparison of the predicted mean velocity in pipes A and B to the measured
superficial velocity from the corresponding rotameter. ........................................... 63 Figure 5.1 Reference overall signal magnitude (S0,ref) relative to the measured velocity (single-
phase flows). ............................................................................................................ 69 Figure 5.2 Flow regime map (from Taitel and Dukler (1976)) indicating the relevant flow
regime position of each experimental trial. ............................................................. 70 Figure 5.3 Velocity measurements of air-water flows. ............................................................. 71 Figure 5.4 Comparison of the NMR measurement of time-averaged liquid volumetric flowrate
(ql) to the in-line rotameter measurement of volumetric flowrate. .......................... 72 Figure 5.5 Single-scan velocity measurement of air-water flow. ............................................. 73 Figure 5.6 Tracking the liquid velocity over time for two-phase air/water flows. ................... 75 Figure 5.7 Tracking the liquid holdup over time for two-phase air/water flows. ..................... 76 Figure 5.8 Normalised cross-correlation (rNV) at time lag (τL) between NMR holdup and video
holdup for experimental trials. ................................................................................ 77 Figure 5.9 Liquid flowrate standard deviation for the stratified and slug flow experiments. ... 79 Figure 5.10 Comparison of the NMR predicted average liquid flowrate to the independent
rotameter measurement of liquid flowrate. ............................................................. 80 Figure 6.1 Illustration of the experimental oil/water flow loop system. ................................... 86 Figure 6.2 T1 measurements of oil/water mixtures. .................................................................. 88 Figure 6.3 Phase fraction comparison for oil and water. .......................................................... 91 Figure 6.4 SNR for FID measurements of canola oil flowing through the NMR flow meter with
signal pre-polarisation achieved using the Halbach array. ...................................... 92 Figure 6.5 SNR for FID measurements of canola oil flowing through the NMR flow meter with
signal pre-polarisation achieved using the EPPC. ................................................... 93
viii
Figure 6.6 Model SNR of fluid phases in velocity-T1 space with application of two separate
polarising techniques in the EFNMR flow metering system. ................................. 94 Figure 6.7 Summarisation of the polarising conditions for the 40 FID measurements. ......... 101 Figure 6.8 Flow chart outlining the NMR flow measurement procedure............................... 103 Figure 6.9 An example 2D velocity-T1 measurement for single-phase water. ....................... 106 Figure 6.10 Measured signal data and corresponding velocity and T1 distributions for single-
phase water measurements. ................................................................................... 108 Figure 6.11 Comparison of the NMR and rotameter measurements of water flowrate. .......... 110 Figure 6.12 An example 2D velocity-T1 measurement for single-phase canola oil. ................ 112 Figure 6.13 Measured signal data and corresponding velocity and T1 distributions for single-
phase oil measurement. ......................................................................................... 115 Figure 6.14 Comparison of the NMR and rotameter measurements of oil flowrate. ............... 117 Figure 7.1 The experimental matrix of oil and water superficial velocities for two-phase flowing
measurements conducted in this work. ................................................................. 121 Figure 7.2 An example 2D velocity-T1 distribution for a stratified with mixing flow regime.
.............................................................................................................................. 125 Figure 7.3 An example 2D velocity-T1 distribution for the dispersion of oil-in-water and water
flow regime. .......................................................................................................... 126 Figure 7.4 An example 2D velocity-T1 distribution for the oil-in-water flow regime. ........... 128 Figure 7.5 An example 2D velocity-T1 distribution for the dispersion of water-in-oil and
dispersion of oil-in-water flow regime. ................................................................. 129 Figure 7.6 An example 2D velocity-T1 distribution for the emulsion of water-in-oil flow regime.
.............................................................................................................................. 131 Figure 7.7 Direct comparison of the NMR measured flowrates and rotameter measured
flowrates (for both oil and water) for two-phase oil/water flow measurements. .. 133 Figure 7.8 Statistical summary of NMR flow measurement performance in the five different
flow regimes. ......................................................................................................... 134 Figure 8.1 Model SNR of fluid components in velocity-T1 space using the Halbach array alone
at a separation distance of 5 cm. ........................................................................... 140
Figure A.1 NMR water measurements used for field inhomogeneity function determination.
.............................................................................................................................. 169 Figure A.2 Experimentally acquired FID signals at varying separation distances of 45 – 150 cm.
.............................................................................................................................. 170 Figure A.3 Illustration of the three separate magnetic field strengths present in the flow metering
system. .................................................................................................................. 171 Figure A.4 Measuring T1 in the EPPC polarising field. .......................................................... 172 Figure A.5 Demonstration of the signal diminishment with introduction of the EPPC. ......... 174 Figure A.6 Experimental measurement of the signal diminishment as a function of velocity and
the polarisation-excitation delay (tPE). .................................................................. 175 Figure A.7 Illustration of effective signal development during variable polarisation scenarios
using a pulse and collect experiment. ................................................................... 177 Figure A.8 Illustration of the impact of the non-adiabatic transition on different regions of fluid
within the pipeline. ................................................................................................ 179 Figure A.9 Model fit of equation A.8 to experimentally measured signal diminishment ratios
(SRatio) as a function of fluid velocity and time since excitation pulse (tPE). ......... 183 Figure B.1 Measuring T2 of oil and water in the EFNMR system. ......................................... 185 Figure B.2 Measuring canola oil T1 in the Magritek 2 MHz Rock Core Analyser. ................ 186 Figure B.3 Measuring canola oil T1 in the Magritek 2 MHz Rock Core Analyser. ................ 187 Figure C.1 Photographs of emulsion stability bottle tests for a 50% canola oil/ 50% water
mixture. ................................................................................................................. 189 Figure C.2 Change in the local magnetic field strength as a function of the separation distance
between the polarisation and detection coil. ......................................................... 191
ix
List of Tables
Table 3.1 Summary of gas-liquid flow regimes ...................................................................... 29 Table 3.2 Summary of oil-water flow regimes adapted from Trallero et al. (1997). .............. 33 Table 3.3 Comparison of key flowmeter specifications for the Krohne M-Phase 5000 and the
EFNMR flow meter. ................................................................................................ 46 Table 7.1 Summary of the five relevant oil-water flow regimes. .......................................... 120
Table A.1 Movement of the separate signal variation regions at different times since the
polarisation switch on. ........................................................................................... 181 Table C.1 Summary of emulsion stability bottle test results. ................................................. 190
x
Abbreviations
CPMG – Carr-Purcell-Meiboom-Gill
Do/w & w – dispersion of oil in water and water flow regime
Dw/o & Do/w – dispersion of water in oil and dispersion of oil in water flow regime
DNP – dynamic nuclear polarization
EMFM – electromagnetic flow meter
EFNMR – Earth’s field nuclear magnetic resonance
Eo/w – emulsion of oil in water
EPPC – electromagnetic pre-polarising coil
Ew/o – emulsion of water in oil
FID – free induction decay
GCV – generalized cross validation
GRA – gamma ray attenuation
IR – infrared
ME – mean error
MPFM – multiphase flow meter
MRI – magnetic resonance imaging
NMR – nuclear magnetic resonance
RCA – rock core analyser
r.f. – radio frequency
RMSE – root mean square error
SNR – signal to noise ratio
St – stratified flow regime
St w/ mix – stratified with mixing flow regime
1D – one dimensional
2D – two dimensional
3D – three dimensional
xi
Nomenclature
English alphabet
A – pipe cross-sectional area
B0 – static magnetic field scalar
B0 – static magnetic field vector
B1 – oscillating magnetic field vector
cNV – cross-covariance of two variables
D – pipe diameter
E – spin energy state
fs – slugging frequency
h – Planck’s constant
HIoil – hydrogen index of oil phase
hl – liquid holdup
hl,I – liquid holdup for scan i
hN – liquid holdup from NMR measurement
hV – liquid holdup from video measurment
I – spin quantum number
I(t) – inhomogeneity as a function of time
Im(e) – measured gamma ray intensity at energy level e
Iv(e) – gamma ray intensity for empty pipe at energy level e
ji – volumetric flux of phase i
k – Boltzmann’s constant
K – model kernel matrix
K – Model kernel function
kP – fraction of polarised fluid experiencing non-adiabatic signal decay
kI – fraction of intermediate fluid experiencing non-adiabatic signal decay
L – length
LD – detector length
LI – intermediate length between polarisation zone and detection zone
LP – length of polarisation zone
LPC – polarisation length of pre-polarising coil
LPD – polarisation-detection separation distance
LPH – polarisation length of Halbach array
Ls – sensor separation length
m – nuclear spin state
mi – mass flowrate of phase i
mT1 – number of steps in T1 distribution
mv – number of steps in velocity distribution
M – total mass flowrate
M – net magnetisation vector
MXY – transverse component of magnetisation vector
MZ – longitudinal component of magnetisation vector
M0 – equilibrium magnetisation
N – number of data points
nF – number of points recorded in FID
nL – number of separation distances
np – power law exponent
npp – number of pre-polarising conditions
nT – number of pre-polarising times
xii
Nscans – number of scans or averages applied in an NMR experiment
nw – number of windows in compressed FID
N1/2 – number of spins in spin up state
N-1/2 – number of spins in spin down state
P(x) – probability distribution of a variable x
P(T1) – T1 probability distribution
P(v) – velocity probability distribution
P(vo) – velocity probability distribution of the oil phase
P(vw) – velocity probability distribution of the water phase
P(v,T1) – 2D joint velocity-T1 probability distribution
p – probability distribution vector
qi – volumetric flowrate of phase i
qg – volumetric flowrate of gas phase
ql – volumetric flowrate of liquid phase
ql,I – volumetric flowrate of liquid for scan i
qo – volumetric flowrate of oil
qo,nmr – rotameter measured volumetric flowrate of oil
qw,rot – NMR measured volumetric flowrate of oil
qw – volumetric flowrate of water
qw,nmr – rotameter measured volumetric flowrate of water
qw,rot – NMR measured volumetric flowrate of water
Q – second derivative smoothing parameter matrix
Q – total volumetric flowrate
Qprobe – NMR probe quality factor
r – radial position within pipe cross-section
Re – pipe flow Reynolds number
RI – Gaussian relaxation constant
RT – T1/T2 ratio
RF – ratio of T1 in the polarising field to T1 in the Earth’s field
s – measured signal vector
S – measured NMR signal
SD – NMR signal attenuation during detection
Sdim – Signal diminishment ratio due to non-adiabatic attenuation
SE – NMR signal prior to excitation
Sfit – model signal fit
SI,EPPC – initial signal of FID decay with pre-polarising coil
SI,Hal – initial signal of FID decay with Halbach array
Sloss – initial signal loss between excitation and acquisition
Smodel – overall NMR signal model
SNA – signal variation due to non-adiabatic decay
So – NMR signal of oil phase
SP – polarised NMR signal
SPH – effective signal polarisation due to pre-polarising coil
SPD – NMR signal attenuation between polarisation and detection
SPH – effective signal polarisation due to Halbach just prior to excitation
Sw – NMR signal of water phase
S0 – NMR signal after an infinite magnetisation time
S0C – NMR signal after an infinite magnetisation time in the pre-polarising
coil
S0H – NMR signal after an infinite magnetisation time in the Halbach field
S0,TP – overall signal magnetisation for a two-phase measurement
xiii
S0,ref – reference value for the overall signal magnetisation for single-phase
flow
t – time
T – Temperature
ta – acquisition time
tdelay – excitation to acquisition delay time
te – time since excitation pulse
TE – inter-echo spacing for CPMG experiment
tI – delay time between inversion and excitation pulses
tPE – polarisation to excitation delay time
tPS – time since polarisation switch on
tpolz – time of polarisation pulse
T1 – spin-lattice relaxation constant
T1,C – spin-lattice constant cut-off for oil and water phases
T1,EF – spin-lattice relaxation constant in the Earth’s magnetic field
T1,LM – log-mean spin-lattice relaxation constant
T1,min – minimum spin-lattice relaxation value for distribution
T1,max – maximum spin-lattice relaxation value for distribution
T1,PF – spin-lattice relaxation constant in the polarising magnetic field
T2 – spin-spin relaxation constant
T2* – effective spin-spin relaxation constant
T2,I – spin-spin relaxation from field inhomogeneity
v – velocity
vi – linear velocity of phase i
vl – liquid linear velocity
vM – mean velocity
vmax – maximum velocity bound in velocity distribution
vM,o – mean oil velocity
vM,w – mean water velocity
vsg – superficial gas velocity
vsi – superficial velocity of phase i
vsl – superficial liquid velocity
vsi – superficial velocity of phase i
vso – superficial oil velocity
vsw – superficial water velocity
xD – volumetric fraction in detection zone
xi – volumetric fraction of phase i
xI – volumetric fraction in intermediate zone
xo – volumetric fraction of oil phase
xP – volumetric fraction in polarisation zone
xw – volumetric fraction of water phase
zi,obs – observed measurement of the variable of interest for sample i
zi,ref – reference measurement of the property of interest for sample i
xiv
Greek alphabet
α – regularisation smoothing parameter
γ – Gyromagnetic ratio
ΔB0 – Inhomogeneity of the static field
ΔE – spin energy difference
μ – fluid viscosity
μemulsion – emulsion viscosity
μi – viscosity of phase i
μi(e) – gamma ray attenuation coefficient at energy level e
μo – oil viscosity
μw – water viscosity
ρ – fluid density
ρi – density of phase i
ρo – oil density
ρw – water density
𝜎ℎ𝑁 – standard deviation of the NMR predicted liquid holdup track
𝜎ℎ𝑉 – standard deviation of the video estimated liquid holdup track
𝜎𝑞𝑙 – standard deviation of the video estimated liquid flowrate track
σv – standard deviation of a velocity distribution
τ – time between 90o and 180o pulse in a spin-echo experiment
τC – cross correlation traversal time
τL – cross-correlation time lag
τPC – fluid residence time in polarisation coil
τs – slug flow time period
τPD – residence time between polarisation and detection
ΦGCV – GCV score
ω – frequency
ω0 – Larmor frequency
xv
Acknowledgements
First and foremost, I would like to sincerely thank my primary supervisor
Professor Michael Johns for the tremendous support throughout my time as a PhD
student. You have given me wonderful opportunities, provided fantastic guidance and
helped me develop a great range of skills for the future. I greatly appreciate all of the
time and effort you have put towards myself and my project.
A big thankyou to my co-supervisor, Dr Einar Fridjonsson. Your stimulating
discussions and valuable insights have greatly helped me to understand the complex
science of nuclear magnetic resonance. Many thanks for your excellent guidance and
support throughout the course of my PhD.
I would like to thank the staff in the Fluid Science and Resources Division who
have assisted me along my journey. Thanks to John Zhen for his collaborative work with
the Q-switch as well as providing much needed assistance in magnetic resonance
electronics for the flow loop. Thanks to Dr Paul Stanwix for providing valuable
assistance in building the apparatus and the initial stages of flow metering system
development. Thanks to Professor Eric May and Dr Brendan Graham for providing
feedback and guidance throughout my research as well as assistance across a variety of
tasks.
Also thanks to my student peers in the Fluid Science and Resources Division.
Particular thanks to Nick Bristow; your technical support in the lab as well as your
engineering knowledge has been extremely helpful throughout my project. Furthermore,
you have been a wonderful friend and provided a listening ear to discuss problems as well
as the development of new ideas throughout my PhD.
Thanks to Andy Sederman and the group at the Magnetic Resonance Research
Centre in Cambridge for providing me with a wonderful opportunity and further
developing my research skills.
Thanks to all the housemates and Louise your support and putting up with me
when I came home from tough days at uni.
Finally, a huge thanks to my parents. Your continual encouragement as well as
immeasurable love and support through my entire education has made it possible for me
to be where I am today.
xvi
This research is supported by an Australian Government Research Training Program
Scholarship.
xvii
Authorship Declaration for Co-authored Publications
This thesis contains published work and work prepared for publication, some of
which has been co-authored. The bibliographical details of this work and where it appears
in the thesis are outlined below.
1. Keelan T. O’Neill, Einar O. Fridjonsson, Paul L. Stanwix, Michael L. Johns,
“Quantitative velocity distributions via nuclear magnetic resonance flow
metering”. Journal of Magnetic Resonance 2016, vol. 269, p. 179-185.
I was responsible for conducting experiments, associated data analysis, and
manuscript writing. Einar O. Fridjonsson and Paul L. Stanwix were responsible for the
construction of the experimental apparatus. Einar O. Fridjonsson and Michael L. Johns
provided project guidance and review of the final manuscript prior to submission. This
paper forms the basis of Chapter 4.
2. Keelan T. O’Neill, Adeline Klotz, Paul L. Stanwix, Einar O. Fridjonsson,
Michael L. Johns, “Quantitative multiphase flow characterisation using an Earth’s
Field NMR flow meter”. Flow measurement and Instrumentation 2017, vol. 58,
p. 104-111.
I was responsible for conducting the primary experiments used for publication,
associated data analysis, and manuscript writing. Adeline Klotz was involved with
conducting initial experiments and development of the experimental apparatus. Einar O.
Fridjonsson and Paul L. Stanwix were responsible for the construction of the
experimental apparatus. Einar O. Fridjonsson and Michael L. Johns provided project
guidance and review of the final manuscript prior to submission. This paper forms the
basis for Chapter 5.
3. John Z. Zhen, Keelan T. O’Neill, Einar O. Fridjonsson, Paul L. Stanwix, Michael
L. Johns, “A Resistive Q-switch for low-field NMR systems”. Journal of
Magnetic Resonance 2018, vol. 287, p. 33-40.
This paper involved joint first authorship between John Zhen and I. John Zhen
was responsible for the design and development of the resistive Q-Switch equipment
whilst I was responsible for demonstrating the application of the equipment by conducting
relevant experiments and data analysis. John Zhen and I had equal contribution in
xviii
drafting the manuscript. Einar O. Fridjonsson and Michael L. Johns provided project
guidance and review of the final manuscript prior to submission. This paper is not
incorporated in any single chapter, however the material assists in experimental signal
measurements throughout the thesis.
4. Keelan T. O’Neill, Lorenzo Brancato, Einar O. Fridjonsson, Paul L. Stanwix,
Michael L. Johns, “Two-phase oil/water flow measurement using an Earth’s field
nuclear nuclear magnetic resonance flow meter”. Chemical Engineering Science
2019, vol. 202, p. 222-237.
I was responsible for conducting experiments, associated data analysis, and
manuscript writing. Lorenzo Brancato assisted with conducting experiments as well as
associated data analysis. Einar O. Fridjonsson and Paul L. Stanwix were responsible for
assisting me with construction of the experimental apparatus. Einar O. Fridjonsson and
Michael L. Johns provided project guidance and review of the final manuscript prior to
submission. This paper forms the basis of Chapters 6 and 7.
1
Chapter 1. Introduction and motivation
The global energy demand continues to increase due to world population growth,
economic development and technological advances in society (British Petroleum, 2017,
Holditch and Chianelli, 2008). Oil and gas are predicted to continue to be key sources of
primary energy for the foreseeable future (British Petroleum, 2018, International Energy
Agency, 2017b, US Energy Information Administration, 2017). The world’s conventional
oil and gas reserves are being depleted and therefore petroleum companies are seeking
alternative production techniques to develop hydrocarbons (IBM Institute for Business
Value, 2010, International Energy Agency, 2017a). Stimulation techniques such as enhanced
oil recovery or gas lift are used to improve output from current fields (Alvarado and
Manrique, 2010, Sheng, 2010). Reservoir lifetimes are being extended which is increasing
the quantity of water produced from oil and gas fields (Igunnu and Chen, 2014).
Furthermore; petroleum companies are now developing fields of increasing technical
difficulty, such as marginally economical fields with tighter profit margins or deeper offshore
fields (Millheim, 2008, Speight, 2011). Such difficulties in modern oil and gas production
require improved monitoring of field production and therefore are driving demand for
improving flow metering technology (Fee and O'Dea, 2014).
The measurement of multiphase fluid flowrates and compositions in industrial
pipelines provides vital information for production monitoring, optimisation and flow
assurance purposes (Falcimaigne and Decarre, 2008, Falcone et al., 2009). Metering of
component flowrates from produced multiphase fluid streams has historically been achieved
using conventional test separators (McAleese, 2000). These vessels separate the multiphase
fluid phases (typically oil, gas and water) into individual phase streams using gravimetric
separation. The flowrate of individual phases is then measured by single-phase flow meters
(e.g. orifice-plate meters or Coriolis meters) (Crawford, 1981, Jahn, 1998).
2
Recent advances in flow metering technology have allowed commercial multiphase
flow meters (MPFMs) to provide a competitive alternative to conventional test separators
(Mehdizadeh, 2001, Pinguet et al., 2010, Ross and Stobie, 2010). MPFMs are instruments
which simultaneously measure the individual component flowrates in a multiphase fluid
stream without the need for phase separation (Falcimaigne and Decarre, 2008, Pinguet,
2010). MPFMs were initially developed to meet several challenges in North Sea oil fields
(Falcone et al., 2009); the discovery and development of satellite marginal fields which were
to be connected to existing infrastructure required an alternative cost-effective flow metering
technology where the cost of an additional test separator could not be justified (Falcone et
al., 2002, Miller, 2008, Syre et al., 2013). Furthermore, continual production from maturing
fields has led to changing process conditions (e.g. increasing water and gas fractions) which
require a robust and flexible flow measurement technology (Appel et al., 2011, Bruvik and
Hjertaker, 2016, Pinguet et al., 2006). Today it is widely recognised that MPFMs provide a
considerable range of benefits relative to test separators in terms of well-testing, reservoir
management, production allocation and flow assurance (Pinguet, 2010, Thorn et al., 2013).
The metering positions of MPFMs and test separators are schematically demonstrated in
Figure 1.1 for a typical offshore production scenario.
Figure 1.1 Schematic of a typical offshore production scenario. Oil, water and natural gas produced
from an underground reservoir is transported through a subsea pipeline and riser to an offshore
production platform. The oil, gas and water flowrates of the multiphase stream can be measured; (i)
using a test separator (at the offshore platform) where the fluid streams are separated into individual
components and then metered using single-phase flow meters, and/or (ii) direct in-line measurement
at the manifold using a multiphase flow meter.
The test separator provides multiphase flow measurement at the offshore production
platform whilst the MPFM may be installed in the subsea architecture near or at the
3
production manifold. The MPFM provides a direct in-line measurement of the multiphase
fluid stream close to the wellbore. Conventional test separators have a considerable time
delay between fluids leaving the wellhead and flow measurement occurring on the production
platform (Berge, 2011, Rajan et al., 1993). Furthermore, the equipment can require several
hours to achieve stable separation (Atkinson et al., 2004). In comparison, MPFMs installed
to monitor pipelines at subsea production manifolds can provide a continuous near real-time
measurement of fluid composition and flowrate (Falcone et al., 2001). This can provide
critical information for well-testing, production monitoring and optimisation (Corneliussen
et al., 2005). The correct implementation of MPFMs can result in improved reservoir
management with a 6-9% gain in the value of oil produced (Douglas-Westwood Limited,
2004). MPFMs are considerably cheaper relative to test separators; MPFMs cost
US$100,000 – 500,000 compared to US$1 – 5 million for a conventional test separator layout
(the cost of both instruments can vary with installation location (i.e. onshore, offshore or
subsea) and the size of the apparatus) (Falcone et al., 2009). Several authors have indicated
that the ability to incorporate non-invasive process tomography within a MPFM would be of
considerable benefit in terms of understanding the multiphase flow behaviour for flow
assurance purposes (Chaouki et al., 1997, Thorn et al., 1999).
The multitude of advantages provided by MPFMs has resulted in increasing demand
for these instruments, with the MPFM market anticipated to grow at an annual rate greater
than 5.4 % until 2022 (Research and Markets, 2016, Yoder et al., 2016). Multiple multiphase
flow metering technologies exist which provide unique solutions for measuring the relevant
component phase flowrates. MPFMs typically measure phase flowrates through separate
measurements of the total flowrate (e.g. using Venturi meters) and phase fractions. Gamma-
ray attenuation and electrical impedance are the two most widely used multiphase flow
metering techniques for phase fraction determination and have achieved reasonable
commercial success (Falcone et al., 2009). For example, the Vx multiphase flowmeter
produced by Schlumberger (shown in Figure 1.2) uses a Venturi meter to provide the total
flowrate of the multiphase stream whilst a gamma ray attenuation meter provides phase
fraction measurement (Atkinson et al., 2004).
4
Figure 1.2 The Schlumberger Vx multiphase flowmeter. The meter consists of a venturi meter
with absolute and differential pressure sensors for overall flowrate measurement. A dual energy
gamma-ray detector and source are used to simultaneously measure the phase fractions of oil, gas and
water of the multiphase flow stream (Atkinson et al., 2004).
Current MPFMs often do not meet the desired accuracy required for successful
implementation and thus there is a market for developing a reliable, accurate and robust
MPFM (Thorn et al., 2013). Nuclear magnetic resonance (NMR) is a technology which has
previously been recognised to form the basis for a universal MPFM (Ong et al., 2004). Of
interest are the recent developments in Earth’s field nuclear magnetic resonance (EFNMR)
which offers cost-effective and robust solutions for industrial process control (Mitchell et al.,
2014).
This study outlines the development of the apparatus and associated measurement
techniques required for an Earth’s field nuclear magnetic resonance multiphase flow meter.
Chapter 2 outlines the fundamentals of NMR which are relevant to this thesis. Chapter 3
explores key concepts in multiphase flow and multiphase flow metering.
Chapter 4 describes the development of the EFNMR flow metering system towards
measuring fluid velocity. Quantitative velocity distribution measurements of single-phase
flows are determined using appropriate signal analysis techniques. The accuracy and
robustness of the measurement technique is assessed for flows through a standard single-pipe
as well as a more complex two-pipe system.
5
Chapter 5 details the advancement of the MPFM system towards measuring
velocities and phase fractions in two-phase gas/liquid flows. The measurement technique
demonstrated in chapter 4 is extended in order to quantitatively characterise air-water flows
in both stratified and slug flow regimes; including tracking the liquid velocity and holdup
with time.
Chapter 6 discusses the development of both the flow metering system and signal
interpretation techniques towards measuring two-phase liquid/liquid (oil/water) flows. This
involves incorporating a dual polarisation mechanism for the system to allow the 1H NMR
signal to be measured across a range of fluid relaxation times. The proposed method for
measuring fluid phase fractions is demonstrated for stationary fluids in preparation for two-
phase flowing measurements.
Chapter 7 demonstrates the application of a new signal interpretation methodology
towards the quantitative measurement of two-phase oil/water flows. This involves
simultaneously distinguishing the fluid signal contributions (to quantify phase fractions) and
providing the relevant fluid velocity distributions. The accuracy of the technique in terms of
measuring individual fluid volumetric flowrates is verified.
Chapter 8 summarises the key contributions of this thesis and provides
recommendations for future work.
6
Chapter 2. Nuclear magnetic resonance
2.1 Fundamentals of NMR
2.1.1. Introduction to NMR
The measurement of nuclear magnetic resonance (NMR) within molecular beams was
first reported by Rabi et al. (1938). This was followed by separate demonstrations of NMR
in bulk materials; both in liquid water (Bloch, 1946) as well as in paraffin wax (Purcell et al.,
1946). Since its discovery NMR has developed into a well-established scientific technique
used for a range of academic and industrial purposes. NMR is renowned for its
implementation in the medical industry in the form of magnetic resonance imaging (MRI)
where it can provide in-situ examination of anatomies for clinical diagnosis (Brown et al.,
2014, Luiten et al., 2013, Westbrook and Talbot, 2018). NMR spectroscopy is applied in
chemistry to study the local chemical environment of atomic nuclei; providing structural
information about complex molecules such as proteins (Bovey et al., 1988, Levitt, 2013,
Szilágyi, 1995). The application of NMR to engineering and industrial processes has only
been realised more recently (Gladden and Alexander, 1996, Gladden and Sederman, 2013).
The use of low-field NMR (<1 T) in industrial applications has received considerable
attention, particularly due to recent developments of relevant NMR hardware (Blümich et
al., 2009a, Mitchell et al., 2014). The following sections provide information on NMR
theory relevant to this thesis.
2.1.2. Fundamental NMR Theory
NMR involves the interaction of sub-atomic particles with an external magnetic field.
Nucleons (neutrons and protons) possess a fundamental quantum property known as nuclear
spin which is an intrinsic angular momentum (Callaghan, 1991, Farrar, 1989, Levitt, 2013).
The angular momentum is described by the spin quantum number (I) which is quantised into
increments of ½. Nuclei are able to take one of 2I+1 possible energy states (m) between -I
7
and +I. The spin quantum number relates to the structure of atomic nuclei; nuclei which
contain an even number of both neutrons and protons have zero spin, nuclei which contain
an odd number of both neutrons and protons possess integer spin, whilst nuclei with an odd
number of nucleons (combined total of neutrons and protons) have half-integer spin. In this
work, all measurements performed involve the NMR of hydrogen atoms (1H) which have a
spin quantum number I = ½. Thus a hydrogen atom is able to exist in one of two possible
energy states; m = ½ (spin-up state) or m = -½ (spin-down state).
In the presence of no external magnetic field, the two energy states will possess equal
energy and therefore equal likelihood of a nucleus being in each level. The presence of an
external magnetic field (B0) will orientate the such that they are either generally aligned or
opposed to the external field. This results in each spin energy state having different energy
levels. The interaction between the quantised spin angular momentum and the external
magnetic field is known as the Zeeman interaction (Hore, 2015). The spin states have an
energy (E) which quantifies the Zeeman interaction;
𝐸 = −𝑚ℎ𝛾𝐵0 (2.1)
where γ is the gyromagnetic ratio of the nuclei (for 1H γ is 4.258 × 107 Hz/T), h is Planck’s
constant (6.63 × 10-34 J/s) and B0 is the strength of the external static magnetic field (in T).
Nuclei move between energy states in discrete transitions which can be induced by
electromagnetic radiation of the appropriate energy. For a 1H nuclei with m = ±½; the
transition between energy levels (ΔE) is quantified by (Slichter, 1996);
𝛥𝐸 = ℎ𝑣 = ℎ𝛾𝐵0 (2.2)
where v is the frequency of electromagnetic radiation inducing the energy level transition.
Figure 2.1 provides an illustration of the Zeeman interaction including the relative energy of
the spin-up (m = ½) and spin-down states (m = -½) as well as the energy level difference
(ΔE).
8
Figure 2.1 Zeeman diagram illustrating the energy states for a 1H atom with the energy difference
between the states indicated.
In order to observe magnetic resonant absorption from the nuclear spins, the sample
must be supplied with electromagnetic radiation (typically with a pulse of radio frequency
(r.f.) radiation) exactly matching the energy difference between the spin states. The exact
frequency required is called the Larmor frequency (ω0, in rad/s) and is proportional to the
external static magnetic field (Callaghan, 1991);
𝜔0 = 2𝜋𝛾𝐵0 (2.3)
For example, to detect 1H NMR (γ of 4.258 × 107 Hz/T) in the local Earth’s magnetic
field (~54 µT in Perth, Western Australia), an excitation pulse of ~2.29 kHz is required. The
excitation pertubes the nuclei from their initial equilibrium energy state and causes the
individual nuclear precessions to synchronise to a net coherent precession at the Larmor
frequency. The precession of nuclear moments is known as the free induction decay (FID)
and can be observed using an NMR detection coil.
The relative population of spins in each energy state is described by the Boltzmann
distribution. For 1H with I = ½; the ratio of the number of spins in the spin-down state (N-
1/2) to the number of spins in the spin-up state (N1/2) is given by (Abragam, 1961);
𝑁−1/2
𝑁1/2 = exp (−
𝛥𝐸
𝑘𝑇) (2.4)
Inte
raction
en
erg
y (
E)
Magnetic field strength (B0)
m = -½
m = ½
ΔE = hγB0
9
where ΔE is the energy level transition, k is the Boltzmann’s constant (1.38 × 10-23 J/K) and
T is the absolute temperature. For 1H nuclei in the local Earth’s magnetic field at room
temperature (20 °C); this ratio is extremely small (3.8 × 10-10). The Boltzmann distribution
of spins results in a very slight excess (1 1H proton in every 2.7 × 109) of nuclei in the lower
energy (spin-up) state. These spins contribute towards the net magnetic moment which is
observable during NMR. The very small disproportion of spins between the two energy
states results in a small net magnetic moment and constrains the sensitivity of NMR
experiments. This is the reason why NMR is generally considered as a comparatively low
signal-to-noise-ratio (SNR) technique. However the large number of nuclei available (100
mL of water contains ~6.7 × 1024 1H atoms) results in a net magnetic moment. For Earth’s
field NMR, pre-polarisation of the sample (at a higher magnetic field) is usually required to
achieve a detectable signal; this is discussed in section 2.2.2.
2.1.3. The vector model
A classical mechanics approach is utilised to describe the remainder of the NMR
theory. The net magnetic moment generated by a sample in an external magnetic field can
be considered as a net magnetisation vector (M). The precession of the net magnetisation
vector around the external magnetic field vector (B0) is described using a classical angular
momentum analogy (Bloch, 1946);
𝑑𝐌
𝑑𝑡= 𝐌 × 𝛾𝐁𝟎 (2.5)
where t is time. This description of nuclei spin precession is known as the Bloch vector
model. The net magnetisation vector will align with the direction of the applied static
magnetic field (B0). However if the vector is disturbed from thermal equilibrium (i.e.
perturbed from away the z-axis); the vector is observed to precess about the z-axis at the
Larmor frequency (Keeler, 2005). This behaviour is called the Larmor precession and is
illustrated in Figure 2.2.
10
Figure 2.2 Illustration of Larmor precession. (a) A net magnetisation vector (M) is generated in the
direction of the external static magnetic field (B0), and (b) tilting M away from the direction of B0
(i.e. the z-axis) causes the precession of M about the z-axis at the Larmor frequency (ω0).
The perturbation of the net magnetisation vector into the transverse (X-Y) plane is
necessary in order to detect a signal in a receiver coil and is induced by the application of a
secondary oscillating magnetic field (B1). This is the excitation pulse (of r.f. radiation) which
is applied perpendicular to the direction of B0. The excitation pulse must be applied in
resonance with the spins oscillating about B0 in order to cause M to precess simultaneously
about B0 and B1. The length and amplitude of B1 pulses is used to control the effective
position of M relative to B0. The two commonly applied pulses are; (i) the 90° pulse which
is used to rotate M by 90° into the transverse plane, and (ii) the 180° pulse which inverts M
180°.
2.1.4. Measuring the free induction decay
The detection of the Larmor precession is the observed signal during standard pulse
NMR experiments. The signal is detected by aligning a receiver coil (r.f. coil) near the
sample. The coil must be placed in the transverse plane relative to the static magnetic field;
often a small coil of wire is mounted around the sample perpendicular to the direction of the
static field (Fukushima and Roeder, 1993). The same r.f. coil is often used for excitation and
detection; the coil is switched between transmit and receive during the NMR pulse sequence.
The rotation of the component of M through the transverse plane induces an electromotive
force in the receiver coil; the detected signal is the free induction decay (FID) (Levitt, 2013).
The application of a simple “pulse and collect” experiment, as executed when detecting at
the earth’s magnetic field in practice, is now discussed.
B0
X
Z
Y
(a)
Direction of M B
0
X
Z
Y
ω0
(b)
11
The key components of the pulse and collect experiment are illustrated by the pulse
sequence diagram in Figure 2.3.
Figure 2.3 Illustration of a pulse and collect sequence used to measure a free induction decay. Key
timing parameters are indicated.
The definitions of timing parameters are presented in the pulse sequence, including;
the length of the polarisation pulse (tpolz), the delay between polarisation and excitation (tPE),
the acquisition delay (tdelay) and the time of acquisition (ta). The time since excitation (te) is
the combination of the acquisition delay and acquisition time (i.e. te = tdelay + ta).
The sequence consists of three key stages; polarisation, excitation and detection. A
net magnetisation vector is developed in the sample of interest by imposing a static magnetic
field for a time tpolz. Note that polarisation is not usually applied as a pulse but rather a
permanent static magnetic field; however the Earth’s field nuclear magnetic resonance
(EFNMR) Terranova system (Magritek, New Zealand) utilises an electromagnetic pre-
polarising coil to provide magnetisation (this is discussed in section 2.2.2). Once sufficient
polarisation has been achieved; the B1 excitation pulse at the Larmor frequency is transmitted
using the r.f. coil to rotate M by 90° into the transverse plane. A short acquisition delay is
then required in order to allow ring-down time for the r.f. probe to dissipate unwanted voltage
oscillation due to residual stored energy from the excitation pulse (Zhen et al., 2018). The
r.f. coil is switched to receive and the FID of the excited sample induces an electromotive
force in the coil which is amplified and recorded. Finally the FID will return to thermal
equilibrium according to magnetic relaxation processes (which are discussed in sections 2.1.5
and 2.1.6). The magnetic processes involved in a pulse and collect experiment are captured
using the classical mechanics vector model in Figure 2.4.
te
Polarisation pulse
tpolz
tPE
tdelay
ta
90° excitation pulse Free induction
decay
12
Figure 2.4 Classical mechanics representation of the stages of a pulse and collect experiment. (a)
Polarisation of the sample induces a net magnetisation vector (M) in the direction of the external
magnetic field (B0), (b) M is excited using a 90° r.f. pulse, (c) excitation causes the magnetisation
vector to rotate into the transverse plane which allows observation of the Larmor precession (at a
frequency ω0), (d) The longitudinal component (z) of M relaxes to thermal equilibrium according to
the spin-lattice relaxation time T1, (e) T2* relaxation causes dephasing of M, (f) Net effect of relaxation
processes on M.
The shape of the measured FID signal is characterised by relaxation processes (T1, T2
and T2* relaxation) which will be discussed in the following sections.
2.1.5. Spin-lattice relaxation
Nuclear spins interact with the surrounding environment (“lattice”) resulting in a
process known as spin-lattice relaxation. This process involves energy transfer from the
excited spins to the lattice as a result of molecular motion of spins rotating at the Larmor
frequency. Spin-lattice relaxation is also referred to as longitudinal relaxation and describes
the return to thermal equilibrium for the longitudinal component of the net magnetisation
vector (Mz) with respect to time (t) according to;
𝑑𝑀𝑍
𝑑𝑡=
−(𝑀𝑍−𝑀0)
𝑇1 (2.6)
B0
X
Z
Y
(a)
M r.f. pulse
X
Z
Y
(b)
M
X
Z
Y
(d)
M X
Z
Y
(e)
M X
Z
Y
(f)
M
13
where M0 is the equilibrium magnetisation and T1 is the spin-lattice relaxation constant. The
value of T1 is dependent on the matter of investigation; for example the T1 of water (at 25°C
and at ω0 > 50 kHz) is ~ 2.9 s (Graf et al., 1980). The spin-lattice relaxation behaviour is
dependent on the applied magnetic field as well as the temperature (Korb and Bryant, 2002,
Krynicki, 1966). Integrating equation 2.6 demonstrates that T1 effectively describes the
exponential growth of the longitudinal magnetisation;
𝑀𝑍(𝑡) = 𝑀𝑧(0) exp (−𝑡
𝑇1) + 𝑀0 (1 − exp (−
𝑡
𝑇1)) (2.7)
where Mz(0) is the initial longitudinal magnetisation. Directly after a 90° pulse, Mz(0) = 0
such that the return of Mz to equilibrium is described by;
𝑀𝑍(𝑡) = 𝑀0 (1 − exp (−𝑡
𝑇1)) (2.8)
The T1 relaxation rate can be measured using a range of experimental methods (Fukushima
and Roeder, 1993). One of the most common approaches to measure T1 is the inversion-
recovery method which utilises an initial 180° pulse to invert the magnetization vector M
from the +Z axis into the –Z axis (Vold et al., 1968). This is followed by a pulse and collect
experiment with a 90° pulse used to excite the magnetization into the transverse plane where
finally the FID signal is acquired. The sequence is repeated multiple times with the inversion
delay time (tI; time between the 180° and the 90° excitation pulse) varied in order to capture
signal contrast according to T1. For the inversion recovery pulse sequence; the initial
longitudinal magnetization Mz(0) = -M0 after the application of the inversion pulse, thus the
FID signal amplitude at varying inversion delay times (M(tI)) is described by;
𝑀(𝑡𝐼) = 𝑀0 (1 − 2 exp (−𝑡𝐼
𝑇1)) (2.9)
The measured FID signal amplitudes can be fit using equation 2.9 in order to determine the
T1 in the system of interest.
The inversion recovery method is illustrated in Figure 2.5; with (a) showing the
inversion recovery pulse sequence and (b) the longitudinal magnetisation development for
water (T1 of 2.9 s).
14
Figure 2.5 Inversion recovery method. (a) the inversion recovery pulse sequence, and (b) the
longitudinal magnetisation development during an inversion recovery measurement of a water sample
(T1 ~ 2.9 s).
2.1.6. Spin-spin relaxation
The second relaxation mechanism in NMR is spin-spin relaxation. Each spin in a
Larmor precession interacts with the local magnetic fields of neighbouring spins causing
dephasing of the transverse component of the magnetisation vector (MXY). The decay of MXY
with respect to time is described by (Bloch, 1946);
𝑑𝑀𝑋𝑌
𝑑𝑡=
−𝑀𝑋𝑌
𝑇2 (2.10)
where T2 is the spin-spin relaxation constant. The integral of equation 2.10 demonstrates that
T2 describes the exponential rate of decay for magnetisation in the transverse plane;
𝑀𝑋𝑌(𝑡) = 𝑀𝑋𝑌(0) exp (−𝑡
𝑇2) (2.11)
where MXY(0) is the initial magnetisation in the transverse plane.
In practice two factors influence the decay of MXY; the first being the spin-spin
decoherence according to T2 as described above, whilst the secondary decay mechanism
results from magnetic field inhomogeneity. It is very difficult to produce homogenous
magnetic fields; thus inhomogeneity in the magnetic field results in spins experiencing
different Larmor frequencies in different positions. The distribution of Larmor frequencies
results in a net dispersion of phase in the spin ensemble. The overall or effective spin-spin
relaxation rate (T2*) observed during a FID measurement is described by;
-1
0
1
0.01 0.1 1 10 100
MZ
tI [s]
(b)
te
180° inversion
pulse
tI t
delay t
a
90° excitation
pulse
Free induction decay
(a)
15
1
𝑇2∗ =
1
𝑇2+ 𝛾Δ𝐵0 (2.12)
where ΔB0 quantifies the inhomogeneity of the static field. Dephasing as a result of natural
atomic interactions between neighbouring spins is known as spin-spin relaxation (i.e. T2) and
is an irreversible energy transfer whilst dephasing due to magnetic field inhomogeneity is
reversible.
The magnetic field inhomogeneity is often described using T2,i which quantifies the
exponential relaxation rate caused by field inhomogeneity. However this assumes that the
magnetic field inhomogeneity induces an exponential signal decay which is not always valid;
therefore assuming an exponential decay can result in significant error during signal analysis
(Grombacher et al., 2013, Grunewald and Knight, 2012). Experimental FID measurements
using the EFNMR system have been observed to produce a half-Gaussian line shape (Zhen
et al., 2018). The field inhomogeneity signal decay (I(t)) can be described by the following
half-Gaussian model (Sukstanskii and Yablonskiy, 2001)
𝐼(𝑡) = exp(−(𝑅𝐼𝑡)2) (2.13)
where RI is a Gaussian relaxation rate constant (in s-1) accounting for the rate of signal decay
introduced by magnetic field inhomogeneity. Therefore a FID signal (S(t)) measured in the
Earth’s magnetic field can effectively be described by;
𝑆𝐹𝐼𝐷(𝑡) = exp (− (𝑡
𝑇2+ (𝑅𝐼𝑡)2)) (2.14)
Most magnetic resonance apparatus are designed to reduce the magnetic field inhomogeneity
using shim systems. The Earth’s magnetic field is homogenous; however it is also very weak
such that nearby ferrous or magnetic materials are able to disrupt the local magnetic field.
The Terranova EFNMR system (Magritek, New Zealand), which is applied in this thesis,
utilises shim coils which introduce small magnetic field gradients in order to negate the
impact of inhomogeneity in the local Earth’s magnetic field. The shim coils are applied in
three orthogonal directions (X, Y and Z) which thus correspond to first-order shims. The
current through each coil controls the shim field strength and is manipulated in order to
optimise the impact of each coil relative to the field inhomogeneity. In more complex NMR
16
systems, higher order shims are utilised to create more complex shim fields allowing for
improved correction of the static magnetic field.
The spin-spin relaxation rate (T2) can be directly measured using a Carr-Purcell-
Meiboom-Gill (CPMG) pulse sequence (Carr and Purcell, 1954, Meiboom and Gill, 1958).
This sequence utilises Spin or Hahn echoes which are produced with the application of a
180° r.f. pulse a time τ after the initial 90° excitation pulse. The 180° pulse inverts the phase
accumulated as a result of the Larmor precession. Spin dephasing which has occurred as a
result of magnetic field inhomogeneity is “flipped” and begins to refocus. At a time 2τ after
the initial pulse, phase dispersion will have been reversed (i.e. field inhomogeneity dephasing
has been fully reversed) resulting in a signal maximum which is the spin echo (Hahn, 1950).
The irreversible dephasing caused by spin-spin relaxation is not reversed and thus spin echoes
are able to remove the effect of field inhomogeneity decay and capture T2 decay in isolation.
The CPMG sequence utilises a train of 180° pulses in order to produce multiple spin echoes
which decay according to equation 2.11. The 180° pulses occur every 2τ implying the time
between the measured echoes or the echo spacing is TE = 2τ. The measurement of T2 using
a CPMG experiment is demonstrated in Figure 2.6; with (a) showing the CPMG pulse
sequence including the 90° excitation and 180° refocussing r.f. pulses, whilst (b) shows the
measured spin echoes and resultant echo amplitude envelope which are fit with equation 2.11
in order to determine the T2 for the sample of interest.
17
Figure 2.6 Demonstration of the CPMG experiment. (a) the CPMG pulse sequence with the 90°
excitation pulse followed by a train of 180° pulses at a spacing of 2τ. (b) The measured signal forms
spin echoes at a spacing of TE = 2τ. The echoes refocus and dephase according to T2* whilst the echo
amplitudes decay according to T2. The echo peak amplitudes can be fit with equation 2.11 in order
to obtain a T2 measurement.
2.2 Application of low-field NMR
2.2.1. Low-field NMR
The science of NMR has been utilised across a vast range of practical applications.
Many implementations of NMR (such as MRI for medical diagnosis or spectroscopy for
chemical analysis) involve the use of high magnetic fields (B0 > 3 T (Mitchell et al., 2014))
to provide sufficient sensitivity and/or resolution. However high-field NMR equipment is
impractical for implementation in industrial and process engineering environments due to the
considerable cost of such systems, lack of portability and safety hazards associated with high
magnetic fields (Blümich et al., 2009a). Low-field NMR (B0 < 1 T (Mitchell et al., 2014))
has recently seen a range of technological developments, including; the design and
construction of magnetic hardware (e.g. Blümich et al., 2009b, Danieli et al., 2010,
McDowell and Fukushima, 2008, Raich and Blümler, 2004), advances in spectrometer
electronics (e.g. Kentgens et al., 2008, Lee et al., 2008, Olson et al., 1995) and improvements
Pulse sequence
Signal
T2 T
2*
90° excitation pulse
180° pulse train
𝑴𝑿𝒀(𝒕) = 𝑴𝒙𝒚(𝟎)𝒆𝒙𝒑 ൬−𝒕
𝑻𝟐൰
(a)
(b)
τ 2τ 2τ 2τ 2τ 2τ 2τ 2τ
TE TE T
E T
E T
E T
E T
E
18
in the signal processing and data interpretation of low-field data (e.g. Holland and Gladden,
2015, Keeler, 2005, Mitchell et al., 2012, Mitchell et al., 2014).
The technological advancement of low-field NMR has led to the utilisation of such
systems across a range of process and industrial engineering applications (Mitchell et al.,
2014). One of the most significant applications of low-field NMR is well logging in the oil
and gas industry which is used to provide petrophysical assessment of reservoir formations
(e.g. Freedman and Heaton, 2004, Hürlimann and Heaton, 2015, Kleinberg, 2001a, Kleinberg
and Jackson, 2001). Low-field NMR laboratory studies of rock cores are also used to provide
insights and guidance for interpretation of well logs (e.g. Borgia et al., 1996, Song, 2010).
NMR has seen uptake in process and quality control in the food industry (Constantinesco et
al., 1997). NMR can provide robust and reliable characterisation of relevant food properties
such as composition or droplet size and has been applied to a variety of products such as;
bread (e.g. Lucas et al., 2005, Lucas et al., 2012), potatoes (e.g. Mortensen et al., 2005,
Thybo et al., 2003) and cheese (e.g. Andersen et al., 2010, Hurlimann et al., 2006).
Quantitative measurements using low-field NMR spectroscopy can be used in process
control and reaction monitoring in industrial processes (e.g. Dalitz et al., 2012, Danieli et al.,
2014, Silva Elipe and Milburn, 2016). Finally low-field NMR has been successfully applied
to characterise oilfield chemistry with droplet sizing in emulsions (e.g. Fridjonsson et al.,
2014a, Ling et al., 2016, Ling et al., 2014, Ling et al., 2017) and quantitative analysis of oil
contamination in produced water (e.g. Wagner et al., 2018, Wagner et al., 2016). These
applications demonstrate the commercial capability of low-field NMR in industrial
environments. The application of NMR towards flow measurement is discussed in section
3.2.5.
2.2.2. Earth’s Field NMR
Earth’s Field NMR (EFNMR) involves the application of NMR experiments where
the nuclear spins are excited and detected in the Earth’s magnetic field (Béné, 1980).
EFNMR is considered a special case of ultra-low-field NMR (B0 < 1 mT (Mitchell et al.,
2014). In this thesis, experiments are conducted in Perth, Western Australia, where the local
Earth’s magnetic field in the laboratory is measured as 54 µT (corresponding to a Larmor
19
frequency of 2.29 kHz). Packard and Varian (1954) first observed free nuclear induction in
the Earth’s magnetic field. Nuclear precession in the Earth’s magnetic field was measured
using a robust and portable experimental apparatus (Waters, 1955). The low cost and
transportable nature has not only contributed towards EFNMR becoming an excellent tool
for teaching the principles of NMR (Callaghan and Le Gros, 1982, Callaghan et al., 2007)
but has also provided an opportunity for developing systems for quantitative analysis (Halse
et al., 2008).
EFNMR has been utilised for a variety of interesting applications and developments
in recent history. The initial nuclear magnetism logging tools operated using the Earth’s
magnetic field during petrophysical reservoir characterisation prior to the development of
logging tools at higher magnetic fields (Dunn et al., 2002). Modern application of EFNMR
in the georesources sector uses a surface NMR technique to conduct hydrogeological surveys
for monitoring and characterising groundwater aquifers (e.g. Grunewald and Knight, 2011,
Knight et al., 2012, Shushakov, 1996). An interesting application of EFNMR utilised an
Earth’s field detection coil to measure the diffusivity of brine in Antarctic sea ice (e.g.
Callaghan et al., 1998, Callaghan et al., 1997).
More recent developments have seen EFNMR provide spectroscopic and imaging
measurements. Earth’s field spectroscopy measurements have been performed which reveal
behavioural information of nuclear spins in 129Xe gas (Appelt et al., 2005, Appelt et al.,
2006). Further development of spectroscopic techniques allowed 2D correlation
spectroscopy measurements to be performed in the Earth’s magnetic field (Robinson et al.,
2006). Hardware developments in pre-polarisation coils, imaging gradients and shim
gradients enabled imaging in an EFNMR system; including self-diffusion imaging, velocity
imaging and slice selection (Mohoric et al., 1999, Planinsic et al., 1994, Stepišnik et al.,
1990). The Magritek EFNMR Terranova system has even been developed to be able to
perform 3D MRI (Halse et al., 2006).
The discussed developments and applications certainly illustrate the potential of
EFNMR. However the application of EFNMR towards quantitative measurement systems is
20
limited due to the inherent lack of measurement sensitivity. The poor signal strength
associated with EFNMR is due to the ultra-low magnetic field strength (SNR α B01.75
(Fukushima and Roeder, 1993)). A number of strategies have been developed in order to
overcome the poor sensitivity and provide signal enhancement (Blümich et al., 2009a). The
first strategy applies dynamic nuclear polarisation (DNP) in the form of the Overhauser effect
(Abragam, 1955, Kernevez and Glenat, 1991). A target nuclear spin (e.g. 1H) receives a net
polarisation transfer from an excited unpaired electron causing considerable polarisation
enhancement (Halse and Callaghan, 2008). The technique requires free electrons to be
introduced using free radicals (e.g. nitroxide radicals) (Armstrong et al., 2009). Whilst this
technique provides excellent signal strength (enhancement factor ~ 3000) it is not feasible to
implement in a subsea multiphase pipeline due to the free radical requirement.
A more practical approach (relative to DNP) for signal enhancement is to use a pre-
polarising magnetic field. There are two approaches for generating a pre-polarising magnet
field; using an electromagnet or a permanent magnet array. The Magritek EFNMR
Terranova system incorporates an in-built electromagnet pre-polarising coil (EPPC) in the
form of a coaxial solenoid. This coil is created from multiple copper windings surrounding
the r.f. detection coil. The EPPC provides a maximum pre-polarising field of 18.7 mT (using
a current of 6 A). This is a relatively easy and cost-effective approach for producing a pre-
polarising field; with the advantage of being able to quickly switch the coil on and off. This
allows a fast transition (delay time is 60 ms) from the relatively inhomogeneous polarising
field to the highly homogenous Earth’s field (used for excitation and detection). The primary
disadvantage of the EPPC is the limit on magnetic field strength. The significant current
required to achieve the pre-polarising field causes resistive heating in the coil; this restricts
the maximum current (and hence the maximum field to 18.7 mT) and the time which the pre-
polarising field is applied (tpolz); which is problematic for long T1 samples.
The alternative option to generate a pre-polarising magnetic field is the use of a
permanent magnet. This has been demonstrated with the use of a Halbach array to
hyperpolarise fluids for ultrahigh resolution NMR spectroscopy in the Earth’s magnetic field
(Appelt et al., 2005, Appelt et al., 2006, Armstrong et al., 2008, Moulé et al., 2003). The
21
cylindrical Halbach array can be designed to generate a strong uniform magnetic field within
the inside of the cylinder whilst a very small stray field is generated externally from the
cylinder (Halbach, 1980). The Halbach array magnetic field geometry is advantageous for
many practical applications (Zhu and Howe, 2001) including mobile low-field NMR (Raich
and Blümler, 2004). Cylindrical Halbach arrays can be produced with an effective magnetic
field strength of up to 3-4 T (Leupold et al., 1993); allowing very considerable pre-
polarisation for EFNMR measurements. The permanent magnet array has a limited stray
field, thus can typically be located within 1 – 1.5 m of the detection coil (Halse and Callaghan,
2008). The primary issue regarding the use of Halbach arrays for EFNMR is the sample
transfer from the permanent Halbach magnet to the EFNMR detection probe. The transfer
must be repeatable with unchanging parameters (e.g. polarisation time and transfer time) in
order to ensure consistent results. Flowing systems (e.g. multiphase pipeline flows) lend
themselves towards such a transfer process; thus the Halbach array is ideally suited for an
EFNMR flow meter. The two pre-polarisation systems are shown in Figure 2.7 with (a)
illustrating the EFNMR system which includes an electromagnetic pre-polarising coil and
(b) showing a cylindrical Halbach array.
Figure 2.7 Two alternative solutions for pre-polarisation. (a) the EFNMR system which includes
an electromagnetic co-axial solenoid (can be seen as copper windings) which provides a magnetic
field of 18.7 mT (with a 6 A current), and (b) a cylindrical Halbach array which provides a permanent
magnetic field of 300 mT.
NMR remote detection systems have been designed to capture information (such as
velocity or composition) from a flowing medium (Granwehr et al., 2005, Harel et al., 2006).
These systems apply a time-of-flight measurement where the fluid of interest is tagged in
order to encode magnetisation information. The fluid then travels to a separate detection coil
in order to recover the appropriate information. The excitation tagging may be optimised in
(a) (b)
22
terms of sequence design (manipulating r.f. pulses and gradients) in order to encode the
relevant information of interest (e.g. spatial encoding for MRI recovering fluid geometry).
2.3 Inversion techniques in NMR
NMR is used to characterise a wide range of substances; providing macroscopic and
microscopic property information. This is particularly useful in chemical engineering, where
NMR can be used to non-invasively probe systems to monitor properties such as diffusion,
velocity and chemical exchange during transport processes (Gladden, 1994, Gladden and
Alexander, 1996, Gladden and Sederman, 2013, Stapf and Han, 2006). However, such
processes can be highly complex; involving distributions of the value of material properties
rather than a single component value. For example; during the petrophysical analysis of
reservoir formations (either from well logging or laboratory based measurements) T2
relaxation distributions are obtained which are reflective of the complex pore geometry
within geological formations (Freedman and Heaton, 2004). In this work, Tikhonov
regularisation is the inversion technique used to capture distributions for relevant parameters
of interest. The following sections (2.3.1 and 2.3.2) will discuss 1D and 2D Tikhonov
regularisation.
2.3.1. 1D Tikhonov regularisation
Tikhonov regularisation is a robust mathematical inversion technique useful in
extracting relevant distributions of parameters. Regularisation is effective at handling noisy
signal data (Mitchell et al., 2014) and does not require any assumptions regarding the shape
of the resultant probability distribution (Hollingsworth and Johns, 2003). The applicability
of regularisation to analysing NMR signal data has previously been demonstrated for
complex systems, primarily in evaluating porous geology in well logging (Coates et al., 1999,
Dunn et al., 2002) and laboratory core analysis (Grunewald and Knight, 2009, Morriss et al.,
1997). Inversion techniques have enabled determination of fluid relaxation distributions, as
well as multi-dimensional relaxation and diffusion distributions within porous media
(Hürlimann et al., 2003, Hürlimann and Venkataramanan, 2002, Mitchell et al., 2012). Such
distributions provide petrophysical characterisation of porous geology such as measurement
of porosity, pore size distribution as well as fluid typing and quantification (Hürlimann et al.,
23
2002, Munn and Smith, 1987). Further applications of Tikhonov regularisation towards
complex systems include droplet sizing of emulsions (e.g. Fridjonsson et al., 2012,
Hollingsworth and Johns, 2003, Hollingsworth et al., 2004, Ling et al., 2014, Ling et al.,
2017) as well the distribution and mobility of water within meat (e.g. (Pearce et al., 2011)).
A probability distribution of a variable x (i.e. P(x)) may be expressed as a function of
an experimentally acquired NMR signal (S(t)) using a first-order Fredholm integral equation
(Delves and Walsh, 1974);
𝑆(𝑡) = ∫ 𝐾(𝑡, 𝑥)∞
0𝑃(𝑥) 𝑑𝑥 (2.15)
where the NMR signal is (S) is a function of an experimental parameter (t) and K(t,x) is a
model kernel function describing the mathematical relationship between the acquired NMR
signal and the relevant probability distribution. Equation 2.15 is discretised and rewritten in
matrix algebra;
s = Kp (2.16)
where s is the measured signal vector, K is the model kernel matrix and p is the discretised
probability distribution vector. Equation 2.16 requires inversion to determine p, however
direct inversion is not possible as the transfer matrix (K) is typically rank deficient (i.e. the
matrix rows are linearly dependent). Equation 2.16 can be solved by minimising the sum of
squared residuals (i.e. least squares fit):
min{‖𝐊𝐩 − 𝐬‖2} (2.17)
however experimental noise in the signal vector produces an unrealistic probability
distribution with large oscillations. Regularisation provides a stable inversion method which
determines a realistic probability distribution in the presence of experimental noise by
applying a penalty function. In Tikhonov regularisation, the following expression is
minimised in order to acquire p (Tikhonov et al., 1995);
min{‖𝐊𝐩 − 𝐬‖2 + 𝛼‖𝐐𝐩‖2} (2.18)
where α is the smoothing parameter and Q is a smoothing operation matrix. In this work, the
Q is designed to calculate the finite second derivative of the resultant probability distribution
to ensure smoothness (Mitchell et al., 2014). The first term in equation 2.18 (‖𝐊𝐩 − 𝐬‖2) is
24
the residual norm whilst the second term (‖𝐐𝐩‖2) is the penalty function. The smoothing
parameter is used to optimise the compromise between finding the true solution (minimising
the residual norm) and limiting the impact of noise on the solution (minimising the penalty
function).
The value of α is selected using the generalised cross-validation (GCV) method
(Golub et al., 1979, Wahba, 1977, Wahba, 1982). This method sequentially removes a data
point in the solution (s) and determines the value of the smoothing parameter which predicts
the removed point with the best accuracy (Hollingsworth and Johns, 2003). This is repeated
for each data point in s, and a GCV score is determined as a function of α. The GCV score
(ΦGCV) may be calculated using;
𝛷𝐺𝐶𝑉 =1
𝑁‖𝐊𝐩−𝐬‖2
൬1−1
𝑁𝑡𝑟(𝐊(𝐊𝑇𝐊 + 𝛼𝐐)−𝟏𝐊𝑻)൰
2 (2.19)
where N is the total number of data measurement points in s, tr(..) denotes the trace of a
matrix and T denotes the matrix transpose. The value of α which minimises this GCV score
is the optimal smoothing parameter; this is a methodology that has been widely validated for
NMR data interpretation (e.g. Hollingsworth and Johns, 2003, Hollingsworth et al., 2004).
2.3.2. 2D Tikhonov regularisation
The extension of NMR measurements into two dimensions requires careful
consideration of appropriate inversion techniques (Mitchell et al., 2012). Two-dimensional
(2D) NMR measurements are useful in probing local surroundings as well as providing
chemical information for complex systems. 2D NMR provides simultaneous measurement
of two parameters of interest (e.g. T1, T2 or self-diffusion coefficients (D)). Such
measurements have been previously been applied to study systems such as; reservoir
structures (e.g Hürlimann et al., 2003, Hürlimann and Venkataramanan, 2002, Mitchell et
al., 2010b, Song et al., 2002), food products (e.g. Godefroy et al., 2003, Hills et al., 2004,
Song, 2009) and biological systems (e.g. Hornemann et al., 2009, Vogt et al., 2013). The
type of 2D measurements commonly applied include; T1-T2 distributions (e.g. Hills et al.,
2004, Song et al., 2002, Vashaee et al., 2018), D-T2 distributions (e.g. Callaghan et al., 2003,
Godefroy and Callaghan, 2003, Hürlimann et al., 2003, Hürlimann and Venkataramanan,
25
2002) and T2-T2 distributions (e.g. Hornemann et al., 2009, Mitchell et al., 2007, Washburn
and Callaghan, 2006).
The measurement of 2D NMR data usually requires two separate independent
variables or experimental parameters; which can be considered as the “direct” and “indirect”
measurements (Mitchell et al., 2012). For example; in the measurement of a T1-T2
distribution, the direct measurement is the single-shot acquisition of the CPMG echo train to
measure T2 whilst the indirect measurement is the determined over multiple experiments with
the variation of pre-polarising conditions (e.g. inversion recovery experiment) to measure T1.
2D NMR signal measurements may be described by extending the first-order Fredholm
integral equation (i.e. equation 2.15) into two dimensions (Van Landeghem et al., 2010);
𝑆(𝑡1, 𝑡2) = ∫ ∫ 𝐾(𝑡1, 𝑡2, 𝑥1, 𝑥2)∞
0𝑃(𝑥1, 𝑥2)
∞
0 𝑑𝑥1𝑑𝑥2 (2.20)
where t1 and t2 are experimental parameters varied in order to see signal contrast in the
measured parameters x1 and x2, with subscripts 1 and 2 denoting the first (direct) and second
(indirect) dimensions respectively. K(t1, t2, x1, x2) is the model kernel function describing the
mathematical relationship between the acquired 2D NMR signal (S(t1,t2)) and the relevant
2D relaxation distribution P(x1,x2). For most 2D NMR measurements, the kernel function is
separated into two components (direct and indirect dimensions) and a singular value
decomposition of each kernel is performed during data processing in order to improve the
computational efficiency of implementing a 2D inversion (Venkataramanan et al., 2002).
However this requires the two data dimensions to be independent and separable (i.e. the
dimensions do not share a common variable) (Mitchell and Fordham, 2011). For 2D
measurements where the dimensions are non-separable, it is often preferable to construct a
full kernel (i.e. non-separable matrix) to avoid undesirable consequences during data analysis
(Mitchell et al., 2010a, Mitchell and Fordham, 2011).
26
Chapter 3. Multiphase flow
3.1 Multiphase Flow Concepts
Multiphase flow is considered the simultaneous flow of two or more phases (or
constituent components of the mixture) within a closed conduit (Corneliussen et al., 2005).
Modern society contains a vast array of multiphase flow systems in both natural and
industrial environments (Kolev, 2011). Multiphase flows are encountered within many
industrial processes, including; oil and gas flows in petroleum production, granular flows in
agricultural processing, ore flows in mineral processing and process flows in chemical plants
(Yadigaroglu and Hewitt, 2017). These processes are described by fundamental physical
principles such as; fluid mechanics, mass transfer, chemical reactions and particle mechanics
(Prud’homme, 2010). The considerable complexity of multiphase flows has caused difficulty
in both characterising and measuring them (Gidaspow, 1994). This has contributed towards
significant research interest in furthering the current understanding of multiphase flow as
well as developing multiphase flow metering technologies.
Within the oil and gas industry, multiphase flow is encountered throughout
production and processing systems (Falcimaigne and Decarre, 2008). Multiphase flow
information of the production stream is crucial for flow assurance, reservoir management,
production allocation and monitoring purposes (Economides et al., 1994, Falcone et al.,
2009). This fluid stream typically consists of oil and natural gas but can also contain water,
condensate (hydrocarbon components which have condensed from the natural gas phase) and
a range of solid phases (e.g. sand, hydrates, waxes and asphaltenes). Most multiphase flow
metering systems focus on measuring the three key phases, being oil, gas and water (Thorn
et al., 2013). Multiphase flows can exist in a variety of flow regimes which describe the
geometric arrangement of the phases within the pipeline. For example gas-liquid flows (e.g.
natural gas and oil) can exist as slug flow or bubble flow, whilst liquid-liquid flows (e.g. oil
27
and water) can exist as stratified liquid layers or emulsified mixtures. The qualitative visual
description of three-phase flow patterns becomes even more complex. The classification of
flow regimes is discussed with reference to gas-liquid flow in section 3.1.2 and liquid-liquid
flow in section 3.1.3.
3.1.1. Multiphase flow information
Multiphase flow is traditionally quantified via the mass or volumetric flow rate of the
individual phases (Thorn et al., 2013). The total volumetric flowrate (Q) of a multiphase
stream is given by (Kolev, 2011);
𝑄 = ∑ 𝑞𝑖𝑁𝑖 = 𝐴 ∑ 𝑥𝑖𝑣𝑖
𝑁𝑖 (3.1)
where i indicates the phase (e.g. oil, water, gas), N is the number of phases present, qi is the
volumetric flowrate of phase i, A is the cross-sectional area of the pipeline, xi is the volumetric
fraction of phase i and vi is the instantaneous velocity of phase i. The total mass flowrate (M)
incorporates the phase densities (ρi) and can be determined by (Brennen, 2005);
𝑀 = ∑ 𝑚𝑖𝑁𝑖 = ∑ 𝑞𝑖𝜌𝑖
𝑁𝑖 = 𝐴 ∑ 𝑥𝑖𝑣𝑖𝜌𝑖
𝑁𝑖 (3.2)
where mi is the mass flowrate of phase i. It is useful to consider the component volumetric
flowrates in terms of cross-sectional averaged volumetric fluxes (ji) which are equivalent to
the superficial phase velocities (vsi) and can be considered as the velocity of the phase if that
was the only phase occupying the pipe cross-section. The superficial velocity for phase i is
given by (Michaelides et al., 2016);
𝑗𝑖 = 𝑣𝑆𝑖 =𝑞𝑖
𝐴= 𝑥𝑖𝑣𝑖 (3.3)
The relative velocity between two phases within a multiphase fluid stream is quantified using
a velocity slip term. The velocity slip ratio (Sab) of phase a relative to phase b is defined as;
𝑆𝑎𝑏 =𝑣𝑎
𝑣𝑏 (3.4)
where va is the velocity of phase a and vb is the velocity of phase b. These multiphase flow
properties provide quantitative information for characterising multiphase flow streams.
28
3.1.2. Gas-liquid flows
Two-phase gas-liquid flows (e.g. natural gas and oil) exhibit interesting
hydrodynamic flow characteristics due to the compressible nature of the gas phase and
naturally occurring fluid turbulence (Yadigaroglu and Hewitt, 2017). The behaviour and
shape of the gas-liquid interface is governed by competing interfacial forces and mechanisms
(Hanratty, 2013). Significant research has previously been undertaken to improve the
understanding of gas-liquid flow regimes (e.g. Al‐Sheikh et al., 1970, Michaelides et al.,
2016, Taitel et al., 1980). There are a range of factors which influence the flow patterns
observed in gas-liquid flow, such as; the phase fractions, velocities, fluid properties (e.g.
densities and viscosities), surface tension, pressure, temperature and flow dynamics (e.g.
steady-state or transient) (Hanratty, 2013, Hetsroni, 1982, Hewitt et al., 2013). The flow
regime is also impacted by the pipe inclination which controls the influence of gravitational
forces on pipe flow (Mukherjee and Brill, 1985a, Mukherjee and Brill, 1985b). In this thesis,
the focus is on horizontal pipe flow due to the orientation of the flow metering apparatus;
therefore only gas-liquid flow regimes for a horizontal pipe are considered.
Flow regimes are typically classified using a flow regime map (Baker, 1953,
Mandhane et al., 1974, Taitel and Dukler, 1976). The qualitative description of each flow
regime is based on visual investigation of the flow patterns and is therefore subject to
interpretation. In this work, the gas-liquid flow regimes are compared to the well-known
flow regime map and classifications presented by Taitel and Dukler (1976). The observed
flow patterns are classified into five separate flow regimes; stratified smooth, stratified wavy,
intermittent (slug and plug), annular and dispersed flows. Table 3.1 provides a summary of
each gas-liquid flow regime in terms of a brief description and visual illustration.
29
Table 3.1 Summary of gas-liquid flow regimes
Name Description Illustration
Stratified
smooth
flow
The gas and liquid exist as segregated phases
with gas flowing above the bottom layer of
liquid.
Stratified
wavy flow
The two phases are separated into two layers
however increased gas velocity introduces
small waves at the interface.
Intermittent
flow
The gas phase exists as long bubbles separated
by liquid “slugs” with small gas bubbles
entrained in the liquid. Intermittent flow can
be separated into “slug flow” and “plug flow”
depending on the size of the gas bubbles and
liquid slugs.
Annular
flow
The liquid flows as an annular film along the
pipe walls whilst the gas flows as a central
core. Liquid droplets are entrained in the gas
phase.
Dispersed
flow
Gas bubbles are dispersed throughout the
continuous liquid phase. Also known as
bubble flow.
Gas-liquid flow regime maps are presented as a function of the superficial liquid
velocity (vsl) and superficial gas velocity (vsg). Figure 3.1 shows the flow regime map for
air-water flow in a horizontal pipe (25 mm in diameter) at 25°C as presented by Taitel and
Dukler (1976).
30
Figure 3.1 Example flow regime map for air-water flow in a horizontal pipe. The pipe has a 25 mm
internal diameter and the fluid temperature is 25°C (Taitel and Dukler, 1976).
A brief description of the physical mechanisms which cause flow regime transitions
will now be discussed starting with the base case of stratified smooth flow at very low liquid
and gas velocities. In stratified smooth flow, the low fluid velocities result in a stable
equilibrium between the two phases which remain as separate layers. If either the liquid or
gas velocity is increased beyond velocity thresholds, then instability is introduced at the gas-
liquid interface. The instability initiates wave propagation resulting in capillary waves at the
interface (i.e. stratified wavy flow). When the liquid velocity is substantially increased the
liquid level rapidly rises forming a liquid bridge and a blockage. The blockage results in
pressure oscillations in the gas phase and subsequently intermittent slug or plug flow will be
observed. At very high liquid flowrates and low gas flowrates, the turbulent mixing forces
between the gas and liquid layer overcome the buoyancy forces of the gas resulting in
dispersion of the gas phase as small bubbles within a continuous liquid layer.
Increasing the gas velocity from the stratified smooth flow regime also introduces
instability at the gas-liquid interface which initiates wave propagation resulting in a transition
to the stratified-wavy flow regime. At very high gas velocities, there is insufficient liquid to
maintain a liquid bridge; thus the liquid in the waves is swept up and around the pipe to form
an annular film along the pipe walls resulting in the annular flow regime. Some of the liquid
is entrained as droplets within the central core gas phase.
0.01
0.1
1
10
0.1 1 10 100
Su
pe
rifi
cia
l li
qu
id
velo
cit
y,
vs
l[m
/s]
Superficial gas velocity, vsg [m/s]
Dispersed flow
Intermittent flow Annular flow
Stratified wavy flow
Stratified smooth flow
31
3.1.3. Liquid-liquid flows
Liquid-liquid flow exhibits different hydrodynamic characteristics relative to gas-
liquid flow as the two phases are generally incompressible. One of the most common forms
of liquid-liquid flow is oil-water flow. The oil and water typically have similar densities
relative to the large density difference present in gas-liquid flow. The turbulent mixing forces
which control the dispersion between oil and water is more influential in determining the
flow characteristics in liquid-liquid flow compared to buoyancy forces. This results in a
variety of flow regimes involving oil/water emulsions.
During crude oil production, water is coproduced with oil from the reservoir
formation (Fink, 2016). Emulsions are encountered through oil production and processing
including in the reservoir, pipelines and equipment (e.g. at wellhead choke points) (Kokal,
2005). These locations are subject to high shear conditions providing sufficient turbulent
force for the formation of stable oilfield emulsions. Water-in-oil emulsions typically form
(depending on the liquid water-cut) with a rigid interfacial film surrounding water droplets
preventing coalescence (Tadros, 2013). Emulsion stability is dependent on a range of process
conditions, such as; crude oil composition, temperature, presence of solids (e.g. asphaltenes,
resins and waxes), pH, salinity and droplet size (Becher, 1996).
Understanding emulsion formation as well as developing strategies for breaking
emulsions is an area of considerable research interest. A variety of demulsification
techniques can be applied to destabilise oilfield emulsions for oil and water phase separation.
These techniques include; chemical demulsifiers, thermal treatment, electro-coalescence,
biological treatment or membrane separation (Sjoblom, 2001, Zolfaghari et al., 2016).
Treating oilfield emulsions requires large processing equipment resulting in significant
capital and operational expenditure (Stewart and Arnold, 2008). The successful operation of
such equipment requires knowledge of the emulsion characteristics, such as the emulsion
droplet size distribution (Johns, 2009, Schramm, 2006). The hydrodynamics of emulsions
can also influence the degree of pipeline corrosion during crude oil transportation (Becerra
et al., 2000).
32
In the context of MPFM development, the presence of emulsions is important for the
design and operation of flow measurement techniques. For example; electrical impedance
flow measurements utilise either electrical resistance or capacitance measurements
depending on whether the flow is water or oil continuous (Dykesteen et al., 1985). Emulsion
formation influences the distribution of phases within the pipeline and therefore the operation
of electrical impedance MPFMs. A robust MPFM should be independent of the oil/water
flow regime and provide accurate measurement across a wide range of oil and water flowrates
(Thorn et al., 2013). Furthermore; newer MPFM technologies are developing measurement
techniques for identifying flow regimes (Salgado et al., 2010, Tjugum et al., 2002, Wu et al.,
2001). Therefore it is necessary to understand the formation and flow characteristics of the
various oil/water flow regimes (including emulsions) in order to develop a robust MPFM.
Two-phase oil/water flows regimes have been extensively studied due to their
prevalence in the petroleum industry. However there is difficulty in comparing experimental
results due to the complex nature of such flows and variety of possible measurement
conditions (Xu, 2007). Similar to the gas/liquid flow regimes, the description and
classification of oil/water flow regimes is based on visual description and therefore subject
to interpretation. Flow regimes in two-phase oil/water pipe flows are influenced not only by
the velocities of oil and water but a range of flow conditions including; the density difference
between the two phases, the fluid viscosities and wetting properties of the wall (Angeli and
Hewitt, 2000, Charles et al., 1961, Lovick and Angeli, 2004).
The oil/water flow regime map proposed by Trallero et al. (1997) is presented as an
example of liquid-liquid flow regime classification. Trallero et al. (1997) used a mineral oil
and water flow system with a 50.13 mm (internal diameter) pipeline. The two-phase oil and
water fluid properties (measured at 25.6 °C) include; a viscosity ratio (μo/μw) of 29.6, a
density ratio (ρo/ρw) of 0.85 and an interfacial surface tension (γo/w) of 36 mN/m. Six flow
regimes were identified; stratified flow (St), stratified flow with mixing at the interface (St
w/ mix), a dispersion of oil-in-water over a free water layer (Do/w & w), a dual dispersion
of water droplets in oil over oil droplets in water (Dw/o & Do/w), a full oil-in-water emulsion
33
(Eo/w) and a full water-in-oil emulsion (Ew/o). A variety of other oil/water flow regimes
are also possible (e.g. annular flow); however different flow conditions (e.g. a much higher
viscosity ratio) are required in order to achieve such flow regimes (Charles and Lilleleht,
1966, Oliemans et al., 1987, Yao et al., 2009). Table 3.2 provides a summary in terms of a
brief description and visual illustration of each oil-water flow regime.
Table 3.2 Summary of oil-water flow regimes adapted from Trallero et al. (1997).
Name Description Illustration
Stratified flow Two separate fluid phases with oil at the
top of the pipe cross-section above water.
The oil-water interface is smooth.
Stratified flow
with mixing
Fluids exist as segregated layers with the
presence of small capillary waves at the
interface. Small quantities of droplets of
each fluid are dispersed into the alternate
phase.
Dispersion of
oil-in-water and
water
A dispersion layer exists at the top of the
pipe with oil droplets entrained within a
continuous water phase. The bottom layer
consists of free water alone.
Dispersion of
water-in-oil and
dispersion of
oil-in-water
A dual dispersion with; water droplets
dispersed into a continuous oil layer (top)
above oil droplets dispersed into a
continuous water layer (bottom).
Oil-in-water
emulsion
A full emulsion of oil droplets dispersed
in a continuous water layer.
Water-in-oil
emulsion
A full emulsion of water droplets
dispersed in a continuous oil layer.
Oil-water flow regime maps are presented as a function of the superficial oil and
water velocities (vso & vsw respectively). Figure 3.2 shows the flow regime map for oil-water
flow in a horizontal pipe (50.13 mm internal diameter) at 25.6°C presented by Trallero et al.
(1997).
34
Figure 3.2 Flow regime map for oil-water flow in a horizontal pipe. The pipe is 50.13 mm inside
diameter and the fluid temperature is 25.6 °C presented by Trallero et al. (1997).
A brief description of the physical mechanisms relating to the flow regime transitions
will now be discussed beginning with the base case of stratified flow at very low oil and
water velocities. A small increase in either the oil or water velocity leads to stratified flow
with mixing at the interface; the increase in flow rate introduces shear at the interface
inducing instability in the form of small capillary waves. Small quantities of droplets begin
to disperse between the phases at the interface.
A further increase in superficial oil or water velocities pushes the system into the
semi-dispersed flow regimes. For an increase in water flowrate – the system transitions from
the stratified with mixing flow regime to the dispersion of oil-in-water and water flow regime.
The increase in shear provided by the water phase forces the small quantity of oil to discretise
into small droplets and disperse within the water phase. For large superficial water velocities
(vsw>1.5 m/s) and low oil fraction the mixture becomes a full emulsion with oil dispersed into
the water phase. For mixtures which are just within the oil-in-water emulsion flow regime;
the oil droplets will have a size distribution which is dependent on radial position and height;
with larger oil droplets observed towards the top of the pipe due to buoyancy forces as well
as towards the pipe wall due to frictional forces. At very high superficial water velocities the
droplet distribution is more homogenous throughout the pipeline as the turbulent shear forces
dominate.
0.01
0.1
1
0.01 0.1 1
Su
pe
rfic
ial w
ate
r ve
loc
ity,
vs
w
[m/s
]
Superficial oil velocity, vso [m/s]
Stratified
Stratified with mixing
Oil-in-water emulsion
Disp. o/w & w
Water-in-oil emulsion
Disp. w/o & Disp. o/w
35
An increase in superficial oil velocity from the stratified with mixing flow regime
will cause the mixture to transition into the dual dispersion of water-in-oil and oil-in-water
flow regime. The turbulent shear forces in both layers are sufficient to overcome buoyancy
forces and induce dispersion of droplets of each component within the alternate phase. For
large superficial oil velocities, the turbulent mixing is sufficiently high such that a full water-
in-oil dispersion exists with water droplets suspended in the continuous oil phase. The
droplet size distribution gradients which are observed in oil-in-water emulsions are similarly
observed for the water-in-oil emulsions; with larger water droplets sinking to the bottom of
the pipe due to gravitational forces. At very high superficial oil velocities for a water-in-oil
emulsion, the water droplets will be dispersed homogenously across the pipe cross-section.
3.2 Multiphase Flow Metering
3.2.1. Introduction to multiphase flow metering
The purpose of a flow meter is to measure the flowrate of a fluid stream of interest
(Baker, 2016). MPFMs measure the flowrate for each individual constituent phase within a
multiphase fluid stream; providing information on the velocity, fraction and flowrate for each
phase (Corneliussen et al., 2005, Falcone et al., 2009). This information has historically been
provided by test separators on production platforms; where the multiphase stream is
separated into individual components and the component flowrates are determined by single-
phase flowmeters after separation. Whilst test separators provide accurate flowrate
information, they are expensive and measurements are time-consuming (Thorn et al., 1999).
A typical test separator is often designed to be larger than initially required in order to
accommodate for changes in flow conditions, using valuable space on offshore production
platforms (Atkinson et al., 2004). The range of flow regimes and multiphase conditions often
cause problems for test separators. For example, slug flow, tight emulsions (emulsions with
small, closely distributed droplets which are difficult to separate), foams and viscous oils all
create difficulty in obtaining reasonable separation therefore can often result in inaccurate
flowrate measurements (Falcimaigne and Decarre, 2008, Pinguet, 2010, Rajan et al., 1993).
Finally, the oil and gas industry are seeing increased subsea completions with multiphase
pipelines developed over long distances. Such installations require monitoring of multiphase
flowrates subsea for flow assurance and production monitoring (Falcone et al., 2001, Lunde
36
and Hoseth, 1994). Installation of conventional test separators subsea is too costly and more
compact measurement technology is required for subsea multiphase flow metering.
3.2.2. Classification of multiphase flow meters
MPFMs can be classified into several categories; in-line flow meters, separation type
meters and wet gas flow meters (Corneliussen et al., 2005). In-line flow meters provide a
direct measurement of the multiphase flow stream without phase separation; these are the
MPFMs of interest in this work. Separation type meters involve the full or partial separation
of components into individual phases prior to obtaining flow measurements (Falcone et al.,
2009, Thorn et al., 1997). Separating the components is expensive and pushes the cost of
separation type meters towards that of test separators. In this work a separation type meter
is not considered a true MPFM as it does not directly measure the multiphase fluid stream
(Falcone et al., 2009). Wet gas meters are MPFMs with the measurement technology
specifically targeted towards accurately metering multiphase fluid streams with high gas
content (Mehdizadeh et al., 2002). Wet gas meters can be either in-line type or separation
type MPFMs (Corneliussen et al., 2005).
A secondary classification for in-line type MPFMs considers the intrusive nature of
flow metering technologies. There is difficulty for MPFMs to accurately meter flow under
some of the complex flow regimes (Rajan et al., 1993). Therefore many MPFMs homogenise
the multiphase stream of interest such that accurate volumetric flow measurements may be
performed on a well-mixed stream (Falcone et al., 2002). MPFMs utilising homogenisation
or disrupting the natural pipeline flow (e.g. Venturi or orifice style flow meters) are
considered invasive flow meters. There is motivation to move towards non-invasive flow
meters which provide a true and direct measurement of the multiphase fluid stream (Thorn
et al., 2013). A non-invasive MPFM could provide information of multiphase flow
behaviour such as flow regime identification or velocity slip characteristics which is useful
for flow assurance diagnostics as well as production optimisation (Falcone et al., 2009).
Many MPFMs utilise a combination of velocity measurement and phase fraction
measurement techniques in order to determine the phase volumetric flowrates. The velocity
37
and phase fraction measurement principles for in-line type MPFMs will now be discussed
separately in sections 3.2.3 and 3.2.4.
3.2.3. Velocity measurement principles
Velocity measurement in MPFMs is categorised into two classes; differential pressure
measurement and the cross-correlation method.
Differential pressure measurement
Many MPFMs measure differential pressure across pipeline constrictions to
determine fluid velocity. The most common differential pressure meter for multiphase flow
is the Venturi flow meter (Falcone et al., 2002), although alternative flow constrictions such
as orifice plates are also used (Lin, 1982, Murdock, 1962). Venturi flow meters apply the
Venturi effect in a pipeline choke; fluid will accelerate at a constriction whilst its dynamic
pressure will decrease according to Bernoulli’s principle (Baker, 2016). The application of
differential pressure measurements towards multiphase flows requires appropriate
correlations to account for the fluid properties of each phase (Steven, 2002). Such
correlations are often proprietary information, although some well-used correlations have
been published publicly (De Leeuw, 1997). Multiphase Venturi correlations are often
designed and calibrated towards specific flow conditions such as wet gas applications with
low liquid content (Xu et al., 2011). Therefore there is difficulty applying these correlations
across a range of multiphase flow conditions resulting in a lack of robustness in Venturi
MPFMs (Steven, 2008).
The cross-correlation method
The fluid stream velocity can be determined using a signal processing technique
known as the cross-correlation method (Beck and Pląskowski, 1987). This technique
measures a fluid property of a dispersed phase (which depends on the sensor type; e.g.
acoustic signal measured in an ultrasonic flowmeter) at two different locations (Lysak et al.,
2008). The cross-correlation method determines the time taken (τC) to traverse the known
sensor separation length (Ls) from the signal evolution between the sensors. The fluid
velocity may then be determined from the traversal time. The principle of the cross-
correlation technique is illustrated in Figure 3.3.
38
Figure 3.3 Illustration of the cross-correlation principle. The signals captured via separate upstream
and downstream sensors are input into a cross-correlation function to determine the traversal time and
subsequently fluid velocity.
The measurement of the carrier phase velocity requires a tracer for detection or an
empirical model for the slip between the carrier and dispersed phases (Rodriguez et al.,
2011). An empirical model will describe the turbulent velocity profile of the system in order
to relate the linear velocity of the dispersed phase to the overall average phase velocity (Lysak
et al., 2008, Sanderson and Yeung, 2002). The use of a tracer can be problematic for subsea
flow metering; particularly when the carrier phase requires a unique tracer (Thorn et al.,
2013). There can be considerable difficulty in applying the cross-correlation method to
measure velocity under slug flow (Falcone et al., 2002). When the cross-correlation method
is used with gamma or capacitance meters, a distinct flow feature such as a liquid slug is
required. The relationship between slug velocity and the average fluid velocity is highly
complex (Nydal et al., 1992) and thus in this case robust measurements based on slug flow
velocity are difficult to implement.
3.2.4. Phase fraction measurement technologies
A variety of measurement principles have been applied to non-invasively measure
phase fractions (composition). A discussion is provided for three commercially established
measurement techniques; gamma-ray attenuation, electrical impedance and microwave
attenuation meters. This is followed by a brief description of some of the less prominent
multiphase volumetric fraction measurement techniques.
Ls
Flow
Upstream sensor
Downstream sensor
Receivers
Transmitters
Cross-correlation τC v = Ls/τc
39
Gamma-ray attenuation meters
Gamma-ray attenuation (GRA) or densitometry is one of the most common phase
fraction measurement technique in multiphase flow metering (Falcimaigne and Decarre,
2008). Numerous examples demonstrate the multiphase flow measurement capability of
GRA in both laboratory (Blaney and Yeung, 2008, Tjugum et al., 2002) and field conditions
(Pinguet et al., 2010, Pinguet et al., 2006). The attenuation of gamma-rays is dependent on
the density of the medium of propagation (Pedrotti et al., 2017). GRA multiphase flow
systems consist of an ionising radiation source (e.g. 137Cs) and collimator which transmit
gamma-rays across the pipe cross section (Åbro and Johansen, 1999). The transmitted beams
are attenuated by the multiphase fluid stream within the pipeline according to the
photoelectric effect, positron-electron pair production and the Compton effect (Falcone et
al., 2009). The attenuated signal is measured at a detector diametrically opposite the
radiation source (Scheers and Noordhuis, 1999). The GRA measurement principle for
multiphase flow measurement in a pipe is illustrated in Figure 3.4.
Figure 3.4 Schematic of the gamma-ray attenuation measurement principle.
The signal decay is quantified by the Beer-Lambert law for the attenuation of
radiation through a material (Ingle and Crouch, 1988). For gamma radiation transmitted from
a source of energy e through a pipe of diameter D containing a multiphase fluid, the radiation
decay is described by (Abouelwafa and Kendall, 1980);
𝐼𝑚(𝑒) = 𝐼𝑣(𝑒) exp(− ∑ 𝑥𝑖𝜇𝑖(𝑒)𝑁𝑖 𝐷) (3.5)
where Im(e) is the transmitted intensity measured at the detector, Iv(e) is the calibrated
intensity for an empty pipe, N is the number of components in the multiphase stream, xi is
the volumetric fraction of phase i and μi(e) is the linear attenuation coefficient for phase i for
the given source radiation energy. A single radiation source is used to solve the phase
fractions in two-phase flow. For more complex multiphase flows; two or more radiation
Multiphase flow
Gamma-ray source
Detector
Collimator
Pipe wall
40
sources are used at different energy levels (e.g. 137Cs at 662 keV and 170Tm at 84 keV)
(Pinguet et al., 2006, Scheers and Slijkerman, 1996). The detection of two or more
attenuation measurements can thus be used to determine volumetric fractions for three or
more phases (Frøystein et al., 2005). Developments towards multi-beam and fan collimators
have also extended the range of flow regimes over which GRA meters can operate (Åbro and
Johansen, 1999, Arvoh et al., 2012, Sætre et al., 2010, Tjugum et al., 2002), nonetheless
there is still evidence of a lack of versatility of gamma ray meters under varying flow
conditions (Corneliussen et al., 2005, Folgerø et al., 2013). There is also motivation to move
away from ionising radiation sources (Thorn et al., 2013). For example; uncertainty in the
actual values of gamma ray attenuation coefficients will lead to systematic errors in measured
fractions and flowrates (Bruvik and Hjertaker, 2016).
Electrical impedance meters
Electrical impedance flow meters have also achieved reasonable commercial success
in the MPFM industry (Thorn et al., 1997, Thorn et al., 1999). Such meters operate on the
principle that the measured impedance is dependent on the fluid dielectric properties (i.e.
conductivity and permittivity) (Ceccio and George, 1996). The measured electrical
impedance across a pipe containing a multiphase flow stream is therefore dependent on the
conductivity and permittivity of the individual components and their respective volumetric
fractions (Chun and Sung, 1986). Impedance is a function of conductance and capacitance;
however most systems operate such that only one electrical property dominates the
measurement (e.g. operating at high frequency such that capacitance dominates) (Falcone et
al., 2009). The functional relationship between the measured impedance and the multiphase
component fractions is dependent on the electrical sensor configuration (Dykesteen et al.,
1985) thus considerable work has focussed on optimising impedance sensor design (e.g.
García-Golding et al., 1995, Jin et al., 2008, Klug and Mayinger, 1994). Several authors
have indicated the potential for electrical based meters to provide tomographic images of
multiphase pipe flows through the use of electrical capacitance tomography and/or electrical
resistance tomography (e.g. Da Silva, 2008, Huang et al., 2003, Ismail et al., 2005).
41
The simplest sensor design consists of two strip-type electrodes mounted on the pipe
diametrically opposite each other (Green and Cunliffe, 1983). The electrical impedance of
the multiphase mixture is measured across the two electrodes. Figure 3.5 presents a
schematic of the strip-type electrode configuration for impedance measurements in
multiphase flow metering.
Figure 3.5 Strip-type electrode configuration for impedance measurements of multiphase flows.
The most significant problem for impedance type multiphase flow meters is the flow
regime dependence. In gas/liquid flows the impedance measurements are observed to vary
considerably as a function of the geometry of the fluid within the pipeline (e.g. between
stratified and annular flow) (Chun and Sung, 1986). Specific sensors have been designed to
reduce the parameter sensitivity to the flow regime (Merilo, 1977). In oil/water mixtures the
impedance sensors are often designed to switch between capacitance measurement for oil-
continuous mixtures and conductivity measurements for water-continuous mixtures
(Dykesteen et al., 1985). However such systems require correct identification of the phase
inversion point in oil/water emulsions; which is a complex process (Piela et al., 2008) and
therefore the incorrect meter response can result in considerable measurement errors (May et
al., 2008). Electrical impedance MPFMs are also susceptible to systematic errors due to
uncertainty in the values of electric permittivity (Bruvik and Hjertaker, 2016).
Microwave attenuation meters
Microwave attenuation meters are similar to electrical impedance meters in terms of
measuring dielectric fluid properties. The attenuation of microwave radiation through a
multiphase fluid is dependent on permittivity and phase fraction of each multiphase
component (Wu et al., 2009). There are two types of microwave sensors used for MPFMs;
transmission and resonator sensors (Nyfors and Vainikainen, 1989). Transmission sensors
Multiphase mixture
Z
Electrode
Pipe wall
42
use two antennas to transmit and receive microwaves through a medium (Corneliussen et al.,
2005). The transmitted waves may be at a single or variable frequency. Resonator sensors
measure the resonant frequency of microwaves through a medium. Microwave resonance is
a function of fluid permittivity thus the measured resonant frequency can be related to the
multiphase fluid (Khan, 2014).
The commercial application of microwave attenuation MPFMs is much more limited
compared to gamma-ray attenuation or electrical impedance. Several examples illustrate
successful application of microwave-based MPFM technology in laboratory and field trials
in combination with other technologies (Ashton et al., 1994, Mehdizadeh et al., 2009, Xie,
2007). Microwave attenuation meters need careful calibration and are suggested to be
problematic in the presence of gas (Falcone et al., 2009). Consequently, microwave
attenuation meters lack the robustness required for accurate flowrate measurement across
variable multiphase flow conditions in the modern oil and gas industry.
Alternative multiphase flow metering technologies
A variety of alternate technologies have been applied to analyse multiphase flows,
often providing solutions to unique multiphase flow metering problems. The use of acoustic
emission in the form of ultrasonic waves has been used to measure two-phase gas/liquid
flows (Al-Lababidi et al., 2012). Acoustic emission is particularly useful in bubble flow
where the attenuation of ultrasonic waves is influenced by the gas void fraction and the size
of bubbles (Bensler et al., 1987, Jones et al., 1986). However application to three-phase flow
measurement is complicated by the array of propagation mechanisms for sound waves (e.g.
diffraction or reflection) through complex media (Hoyle, 1996) such that ultrasonic MPFMs
are only sensitive to flows with small phase fractions (Goncalves et al., 2011).
Electromagnetic flow meters (EMFM) have had success as single-phase flow meters
(Baker, 2016). EMFM operate on the principle that an electrically conductive fluid passing
through a magnetic field induces an electrical potential perpendicular to the magnetic field
and velocity (Shercliff, 1962). The measurement principle has been extended to analyse
gas/liquid flows (Bernier and Brennen, 1983) and to determine fluid velocity profiles
43
(Leeungculsatien and Lucas, 2013, Wang et al., 2007). However there is difficulty in
measuring non-conductive fluids such as hydrocarbons and therefore further research is
required in order to develop the measurement principle towards full multiphase flow
metering (Falcone et al., 2009).
Infrared (IR) spectroscopy has received some attention in terms of monitoring
multiphase flows; particularly as a water-cut meter. The adsorption of IR radiation by a
material is dependent on the wavelength of the emitted light; thus infrared adsorption
spectroscopy (across a range of IR wavelengths) can be used to distinguish oil and water
(Sarkodie et al., 2018). Analysis of IR spectra has been shown to give accurate phase fraction
measurements for flows with high water content (>85 %) (Falcone et al., 2009). A MPFM
which incorporated infrared technology showed positive results for water-cut measurement
during field trials (Nassereddin et al., 2010). However IR MPFMs require further research
in order to measure flow at lower water content, analyse gas/liquid flows (Sarkodie et al.,
2018) and to quantify the influence of emulsified droplets (Borges et al., 2015).
3.2.5. NMR in flow measurement
NMR has seen extensive application as a tool for examining flow characteristics
(Fukushima, 1999). The ability to provide non-invasive interrogation of complex flow
behaviour is particularly advantageous (Caprihan and Fukushima, 1990). This is evident in
the application of high-field MRI to provide tomographic characterisation of process flows.
For example, MRI has been used to provide quantitative hydrodynamic characterisation of
gas-liquid bubbly flow (Holland et al., 2011, Leblond et al., 1998, Tayler et al., 2012), oil-
water homogenisation (Lakshmanan et al., 2016) and gas-solid flow in packed beds (Müller
et al., 2006, Muller et al., 2007). This is very useful in improving our understanding of such
complex flows and for validation of computational fluid dynamic simulations (Lemonnier,
2010). NMR has also been used to examine hydrodynamic and diffusive flow behaviour in
porous media (Seymour and Callaghan, 1997, Song, 2013) as well as examining the fluid
dynamics and rheology of complex fluids (Callaghan, 2006, Callaghan, 1999).
44
The use of NMR as an industrially capable flow meter is much more limited relative
to its application to studying flow behaviour. This can be attributed to the high cost and low
portability of high-field NMR equipment (Danieli et al., 2010) which makes it impractical as
a measurement technology for industrial multiphase flow metering. The first NMR flow
meter was used to measure the flowrate of blood (Battocletti et al., 1981, Singer, 1959). The
initial NMR flow meter designed towards industrial applications was developed by Vander
et al. (1968) for laminar flow whilst the first NMR flow meter for multiphase flow probed
the water velocity and fraction in two-phase nitrogen-water flows (Krüger et al., 1996,
Krüger et al., 1984). Recent developments in low-field NMR (Blümich et al., 2009a,
Mitchell et al., 2014) provides attractive technology for developing a low cost and
transportable instrument for industrial multiphase flow measurement.
The most prominent low-field NMR MPFM is the commercially available M-Phase
5000 developed from a collaboration between Shell, Krohne and Spinlock. The system
operates at a 1H resonance of 8 MHz and utilises a remote detection system with separate
regions for polarisation and detection (Hogendoorn et al., 2015). The polarising field is
produced by a series of Halbach magnets which can be rotated to modify the strength and
orientation of the magnetic field (Appel et al., 2011). The detection field is generated by
another series of Halbach magnets which are fixed. The system utilises a CPMG
measurement of an “effective signal decay” which incorporates a linear flush-out effect (Ong
et al., 2004). The velocity is then measured from the effective rate of signal decay of the
measured CPMG (Osán et al., 2011). Gradient coils are also incorporated into the detection
region in order to produce 1D MRI measurements to capture flow profiles (Hogendoorn et
al., 2015). The flow meter uses differences in fluid pre-polarisation (according to fluid T1)
to quantify phase fractions (Hogendoorn et al., 2013). The system has undergone prototype
testing in commercial multiphase flow loops as well as a field trial (Hogendoorn et al., 2016,
Hogendoorn et al., 2014).
45
Figure 3.6 A cut-open schematic of the M-Phase 5000 magnetic resonance multiphase flow meter.
The key components of the magnetic resonance measurement section are highlighted (Hogendoorn
et al., 2013).
Despite generally positive results for the M-Phase 5000 there are still several
drawbacks. The key problems which are limiting commercial application include; the size
(3.5 m long, 1150 kg), power consumption (180 W) and cost of the system (Krohne, 2017).
There is difficulty capturing accurate flow information under dynamic flow regimes; for
example in the presence of slug flow the system requires considerable signal averaging to
obtain time-averaged fluid properties and has difficulty in measuring the instantaneous flow
features (Hogendoorn et al., 2015). The system requires a-priori knowledge of the magnetic
relaxation properties (e.g. T1, T2) which are sensitive to process conditions such as pressure,
temperature and oil compositions. This can lead to flow measurement errors when the
calibrated properties are incorrect (relative to the real properties), however measurement of
the changing relaxation parameters has been demonstrated to be useful for process
diagnostics (Hogendoorn et al., 2017).
An alternative NMR flow meter in development utilises the Earth’s magnetic field
(Fridjonsson et al., 2014b). This system utilises a remote detection system with a mobile
pre-polarising Halbach array upstream of an Earth’s field radio frequency detection coil. A
parameter-free model is presented which is shown to accurately describe the NMR signal of
a fluid flowing through the system at velocities from 0 – 1.0 m/s. The Earth’s field NMR
flow meter is the basis of the work presented in this thesis.
46
Table 3.3 compares some of the key flow meter specifications for the commercially
available Krohne M-Phase 5000 and the alternative EFNMR flow meter.
Table 3.3 Comparison of key flowmeter specifications for the Krohne M-Phase 5000 and the
EFNMR flow meter.
Flowmeter Dimensions
(L x W x H, in m)
Larmor Frequency
(kHz)
Velocity range
(m/s)
Krohne M-Phase
5000 3.5 x 0.8 x 0.9 m 8000 0.1 - 6
EFNMR flow
meter 2.0 x 0.4 x 0.4 m 2.29 0.1 - 3
3.2.6. Validation of flow measurement accuracy
In this work, NMR flow measurement techniques are developed across a range of
flowing conditions and are validated against reference flow measurements to quantify
accuracy. Two sample statistics are used in order to compare the NMR measurements
relative to reference measurements. The mean error (ME) is a sample statistic which
quantifies measurement accuracy in terms of how well an observed measurement (e.g. NMR
measurement of water flowrate) matches a reference measurement (e.g. rotameter
measurement of water flowrate). The mean error for a sample (of size N) is defined as;
𝑀𝐸 = ∑ 𝑧𝑖,𝑜𝑏𝑠− 𝑧𝑖,𝑟𝑒𝑓
𝑁𝑛=1
𝑁 (3.6)
where zi,obs is the observed measurement of the variable of interest for sample i and zi,ref is the
reference measurement of the property of interest for sample i.
The root-mean-square error (RMSE) is a sample statistic which quantifies accuracy
in terms of the quadratic mean of the residual errors of observed measurements relative to a
reference measurement. The RMSE for a set of flow measurements (of sample size N) is
calculated using;
𝑅𝑀𝑆𝐸 = √∑ (𝑧𝑖,𝑜𝑏𝑠− 𝑧𝑖,𝑟𝑒𝑓)2𝑁
𝑛=1
𝑁 (3.7)
where zi,obs and zi,ref have the same meanings as in equation 3.6.
47
Chapter 4. Quantitative velocity measurements of single-
phase flows
4.1 Introduction
The first stage of developing the Earth’s field nuclear magnetic resonance (EFNMR)
multiphase flow metering system will focus on establishing a velocity measurement
technique. This chapter demonstrates the capability of the EFNMR system in terms of
measuring velocity probability distributions of single-phase fluid flows. The EFNMR flow
metering system is detailed in terms of the apparatus and associated NMR signal model. The
optimal method of velocity encoding is examined with respect to appropriate experimental
parameters. The velocity measurement methodology is then demonstrated; with the
instrument being applied to determine the velocity probability distribution of a flowing
stream. Determination of the velocity distribution is a necessary precursor to application of
the instrument to multi-phase flow; it is required to distinguish flow regime (e.g. stratified
flow or slug flow) and to determine the velocities of multiple phases within the measured
flowing stream.
4.2 Equipment and methodology
4.2.1. Experimental apparatus
Figure 4.1(a) shows a photo of the flow metering measurement section including the
pre-polarising magnet and detection coil whilst Figure 4.1(b) shows a schematic of the NMR
flow meter and the associated flow loop for experimental validation. The pre-polarisation
magnet (0.3 T Halbach array provided by Magritek, New Zealand) of length LPH is located a
variable distance, LPD, upstream of the EFNMR radio frequency (r.f.) detection coil (of length
LD). The pre-polarisation magnet is required to boost the NMR signal in the r.f. coil to
detectable levels (Appelt et al., 2006, Halse and Callaghan, 2008). The Halbach array is
contained within a mu-metal enclosure (Magnetic Shield Corp., USA) in order to minimise
48
stray magnetic field interference at the detection coil. The r.f. detection coil (Magritek, New
Zealand) is used to excite and detect the NMR signal at a frequency of ~2.29 kHz (the 1H
NMR Larmor frequency associated with the local Earth’s magnetic field of ~54 µT). The
detection coil is orientated with its axis perpendicular to the Earth’s magnetic field and is
shielded from external electrical fields by a Faraday cage. A photo of this flow metering
instrument (magnet and coil) is shown in Figure 4.1(a), in which the Faraday cage, pre-
polarisation magnet, detection coil and mu-metal shielding are indicated. The flow loop
includes a 44 mm internal diameter acrylic pipe in the measurement section (E-Plas,
Australia) and a centrifugal pump (CHI4-40, Grundfos, Australia). A 4 to 60 L/min glass
rotameter (Cole Parmer, Australia) is used to independently measure the volumetric flow
rate.
49
Figure 4.1 Illustrations of the single-phase experimental flow loop apparatus. (a) Photo of the
measurement section of the flow loop with flow direction indicated. The detection coil is protected
by a simple Faraday cage. (b) Schematic of experimental flow loop, which includes a mobile pre-
polarisation magnet (Halbach array) and an EFNMR excitation and detection coil. The distances
relevant to the model, i.e.; polarisation coil length (LPH), distance between polarisation and detection
(LPD) and the detection coil length (LD) are indicated.
In Figure 4.1(b), the video analysis region just prior to the EFNMR detection coil and
the air inlet line are relevant for gas/liquid flow experiments are discussed in section 5.2.1.
In order to access multi-modal velocity distributions, the flow loop was also modified to
accommodate two separate pipes (29 mm internal diameter, both with independent flow
control and rotameter velocity measurement) through the NMR flow meter. A photo of this
two-pipe system is shown in Figure 4.2(a) with a schematic of the instrument configuration
shown in Figure 4.2(b).
(a)
Pre-polarising Halbach array
EFNMR detection coil
Flow direction
Mu-metal shielding
Faraday cage
(b) LPD LPH LD
50
Figure 4.2 Illustration of the two-pipe experimental flow loop. (a) Photo of the two-pipe system
with flow direction indicated. (b) Schematic of two-pipe flow loop. Ball valves are located after the
pipe split in order to control the flowrate in each pipe.
The NMR pulse sequence used was a basic ‘pulse and collect’, as shown in Figure
2.3. Note that no pre-polarisation pulse is applied; polarisation is provided by the upstream
Halbach array alone. The EFNMR r.f. detection coil was shimmed prior to running a
sequence of experiments (at a given separation distance, LPD) in order to optimise the signal
obtained relative to the surrounding magnetic environment. The following key sequence
parameters were used for the ‘pulse and collect’ experiments: an acquisition delay between
excitation and detection (tdelay) of 25 ms (see further discussion regarding Q-switch below);
a total acquisition time per scan (tacq) of 0.5 s, a repetition time of 0.8 s and 64 scans (Nscans)
per experiment.
Pre-polarising Halbach array
EFNMR detection
coil
Liquid rotameters
Flow direction
(a)
(b)
51
Nuclear magnetic resonance systems observe unwanted voltage oscillation (referred
to as ringdown) within the r.f. probe. Ringdown is a consequence of residual stored energy
following the 90o r.f. pulse (Hoult, 1984). The effective efficiency of NMR probes is
quantified by the probe quality factor (Qprobe) which is generally designed to be high in order
to improve the SNR (SNR √𝑄𝑝𝑟𝑜𝑏𝑒) (Andrew and Jurga, 1987). A consequence of using
a high Qprobe is increased ringdown time (the ringdown time constant τ Qprobe). This
problem can be overcome by the use of a Q-switch damping circuit which switches on a
resistive load immediately after the r.f. excitation pulse to accelerate energy dissipation
during ringdown (Hopper et al., 2011, Peshkovsky et al., 2005). A Q-switch circuit was been
incorporated in the EFNMR system used for the MPFM developed within this thesis which
allows the excitation-acquisition time delay (tdelay) to be reduced from 25 ms to 9 ms (Zhen
et al., 2018). The Q-switch was primarily implemented to allow earlier FID acquisition for
the measurements with oil and was completed prior to the work presented in Chapters 6 and
7 (results presented in Chapters 4 and 5 do not incorporate a Q-switch; except for the single-
pipe results in section 4.3.2).
4.2.2. NMR signal model
In the following section, a model of the NMR signal detected with the apparatus
shown in Figures 4.1 and 4.2 is detailed. This expands from the initial model development
and validation presented by Fridjonsson et al. (2014b). The measured NMR signal is a
composite of three contributions: signal accumulation due to polarisation (SP), signal
attenuation from intermediate decay between the polarisation magnet and the EFNMR
detector (SPD), and signal attenuation during detection (SD). Each of these components are
dependent on the residence time in the relevant section of the system and are detailed below.
When a fluid containing NMR active nuclei (nuclei with a net magnetic moment (i.e.
spin quantum number I ≠ 0)) is placed in a static magnetic field, the detectable NMR signal
(S) develops according to (Krüger et al., 1996, Pendlebury et al., 1979):
𝑆 = 𝑆0 (1 − exp (−𝑡
𝑇1)), (4.1)
52
where S0 is the NMR signal after an infinite time in the magnetic field (which is related to
the static magnetic field strength (B0), usually by S0 B01.75 (Fukushima, 1999)), t is the time
spent in the magnetic field and T1 is the fluid spin-lattice relaxation time. Consider a fluid
element travelling through the flow meter with a velocity v. The development of the potential
signal due to polarisation (SP) may be determined by applying;
𝑆𝑃 = 𝑆0 (1 − exp (−𝐿𝑃𝐻
𝑣𝑇1)), (4.2)
where LPH is the length of the pre-polarising Halbach array and S0 is the potential NMR signal
after an infinite residence time in the polarisation field. Following polarisation, the signal
attenuation between polarisation and detection (SPD) due to spin-lattice relaxation is modelled
by;
𝑆𝑃𝐷 = 𝑆𝑃 exp (−𝐿𝑃𝐷
𝑣𝑇1), (4.3)
where LPD is the separation distance between the polarisation magnet and detection coil.
Once the fluid element reaches the detection coil, its NMR signal is excited by
applying a r.f. pulse via the EFNMR coil at the relevant Larmor frequency. The free
induction decay (FID) signal is detected in the EFNMR coil, however an acquisition delay
time (tdelay) is required between excitation and detection to allow transient currents to
dissipate during coil ring-down. The resultant FID signal measured under flow is affected
by; (i) stationary FID behaviour, and (ii) a ‘flush-out’ effect as excited fluid leaves the
detection volume. The FID behaviour for stationary fluid involves spin-spin relaxation (T2)
as well as relaxation due to magnetic field inhomogeneity and is given by equation 2.14. The
consequential signal behaviour during detection (SD) is modelled as:
𝑆𝐷(𝑡𝑒) = (1 −𝑡𝑒𝑣
𝐿𝐷) exp (− (
𝑡𝑒
𝑇2+ (𝑅𝐼𝑡𝑒)2)), for 𝑡𝑒 ≤
𝐿𝐷
𝑣, (4.4)
where te is the time since excitation, LD is the length of the detection coil and RI is the
Gaussian relaxation rate constant accounting field inhomogeneity signal decay (and is a
function of polarisation-detection separation distance).
53
By combining equations, 4.2, 4.3 and 4.4, the overall model signal (Smodel) at the
detection coil is obtained:
𝑆𝑚𝑜𝑑𝑒𝑙(𝐿𝑃𝐷 , 𝑣, 𝑡𝑒) = 𝑆𝑜 (1 − exp (−𝐿𝑃𝐻
𝑣𝑇1)) exp (−
𝐿𝑃𝐷
𝑣𝑇1) (1 −
𝑡𝑒𝑣
𝐿𝐷) exp (− (
𝑡𝑒
𝑇2+ (𝑅𝐼𝑡𝑒)2))
for 𝑡𝑒 ≤ 𝐿𝐷
𝑣, (4.5)
thus by measuring S as a function of either LPD or te, the velocity (v) may be determined. The
following section considers how to determine the probability distribution of v for a flowing
sample, P(v).
4.2.3. Velocity probability distributions via regularisation
One-dimensional Tikhonov regularisation will be applied in order to extract velocity
probability distributions (P(v)) from the measured FID signals. Further details regarding 1D
Tikhonov regularisation are discussed in section 2.3.1. In order to measure a probability
distribution using Tikhonov regularisation, an experimental parameter must be varied in
order to observe signal contrast with respect to the parameter of interest (i.e. velocity for the
flow meter). For the EFNMR flow metering system outlined in section 4.2.1 either the
polarisation-detection separation distance (LPD) or the time since excitation (te) may be varied
in order to observe signal contrast with respect to velocity. If the separation distance is used
as the independent variable, then the appropriate first-order Fredholm equation is given by;
𝑆(𝐿𝑃𝐷) = ∫ 𝐾(𝐿𝑃𝐷 , 𝑣)∞
0𝑃(𝑣) 𝑑𝑣 (4.6)
where S(LPD) is the NMR signal measured as a function of separation distance and K(LPD,v)
is the kernel transfer function which describes the NMR signal attenuation as a function of
separation distance and velocity. However if the velocity is measured with respect to the
time since excitation (i.e. a FID measurement), then the appropriate first order Fredholm
equation is;
𝑆(𝑡𝑒) = ∫ 𝐾(𝑡𝑒 , 𝑣)∞
0𝑃(𝑣) 𝑑𝑣 (4.7)
where S(te) is the NMR signal measured as a function of time since excitation and K(te,v) is
the kernel transfer function which describes the NMR signal attenuation as a function of time
since excitation and velocity.
54
For both scenarios; the velocity probability distribution is determined using the
Tikhonov regularisation optimisation expression given in equation 2.18. The appropriate
model kernel matrix must be used according to the relevant independent parameter. The
smoothing parameter (α) is selected using the generalised cross-validation method (as
detailed in section 2.3.1).
The following parameters (in Equation 4.5) are experimentally measured prior to
flowing measurements and are fixed for when performing regularisation on signals measured
in flowing experiments; LPH, LD, T1, T2 and RI. Measurement of relaxation constants for
water (T1 and T2) are discussed in section 6.2.3 as well as Appendix A.2 whilst measurement
of the Gaussian relaxation rate constant (RI) is discussed in Appendix A.1. The physically
measured length of the Halbach array (20 cm) is used as the length of the Halbach (LPH). A
hall effect sensor was used to measure the strength of the Halbach magnetic field just outside
the Halbach cylinder and has been used to verify that the Halbach magnetic field length
approximately corresponds to the physical length of the Halbach. A 1D MRI (with the pipe
full of stationary water) along the longitudinal axis of the detector coil in order to capture a
weighted profile of the detector length (LD). The final variable in Equation 4.5 is S0; the
NMR signal after an infinite time in the Halbach magnetic field. S0 can effectively be
considered a free, dependent variable in regularisation analysis and can be determined via
appropriate interpretation of the model fit (see section 5.2.2). It is possible to fix and
constrain S0 prior to inversion, however this was found to introduce artefacts into the
probability distribution. Furthermore, allowing S0 to be a dependent variable was determined
to be useful in two-phase gas/liquid measurements (discussed in Chapter 5).
4.2.4. Theoretical turbulent velocity distributions
Theoretical turbulent velocity distributions are determined for comparative purposes.
Turbulent velocity flows are often approximated by a power law velocity profile, where the
velocity (v) at a radial position (r) in a pipe of diameter D is given by (Iguchi and Ilegbusi,
2013);
𝑣(𝑟) = 𝑣𝑀𝑛𝑝+1
𝑛𝑝(1 −
2𝑟
𝐷)
1
𝑛𝑝 (4.8)
55
for a mean velocity vM and a power law exponent np. An np value of seven is often used for
fully developed turbulent flow, leading to the well-known 1/7th power law distribution
(Iguchi and Ilegbusi, 2013). In this work, the theoretical turbulent velocity distributions are
determined using the correlation for the power law exponent (np) presented in Zagarola et al.
(1997);
1
𝑛𝑝=
1.085
ln (𝑅𝑒)+
6.535
(ln(𝑅𝑒))2 (4.9)
where Re is the pipe flow Reynolds number of the fluid.
4.3 Results and discussion
4.3.1. Optimum Velocity Determination Methodology
Fridjonsson et al. (2014b) demonstrated that equation 4.5 (with te = tdelay and
assuming exponential decay behaviour for T2*) could reliably describe acquired NMR signals
as a function of mean velocity (vM) and separation distance (LPD). Initial experiments
progress from these results, with NMR data acquired over a range of mean velocities (0.11
to 0.55 m/s) and separation distances (68 to 128 cm). A minimum separation distance of 62
cm between the pre-polarising magnet and detection coil was required to ensure that
distortion of the Earth’s magnetic field in the detection region was sufficiently low as to allow
consistent NMR signal acquisition. Figure 4.3(a) shows the resultant data, along with the fit
of equation 4.6 – agreement is excellent. The resultant distributions of P(v), as produced
using regularisation, are shown in Figure 4.3(b).
56
Figure 4.3 Velocity distribution measurements obtained using separation distance as the
independent variable. (a) Experimental data and the model fit (determined using equation 4.6) for
varying NMR signal as a function of separation distance. The measured superficial velocities range
from 0.11 to 0.55 m/s. (b) Resultant velocity probability distributions (P(v)) with clear regularisation
artefacts at larger velocities (~2.5 m/s).
The distributions in Figure 4.3(b) clearly contain erroneous artefacts, evident in the
appearance of peaks at unfeasibly large velocities (~2.5 m/s). These artefacts occur due to
the inability of the methodology to access NMR data at low values of LPD, for reasons detailed
above, and to a much lesser extent large values of LPD, due to physical constraints on
instrument size. Such limitations are a lot less stringent when acquiring the NMR signal (as
per equation 4.7) as a function of te (i.e. the FID); this was shown to be the case in the data
that follows and thus is exclusively used henceforth.
0
2
4
6
8
10
12
0.6 0.8 1 1.2 1.4
NM
R s
ign
al [µ
V]
Separation distance [m]
Expt. data
Model fit
vM
[m/s]0.110.220.330.440.55
0
1
2
3
4
5
6
0 0.5 1 1.5 2 2.5 3
P(v
)
Velocity [m/s]
0.11m/s
0.22m/s
0.33m/s
0.44m/s
0.55m/s
(b)vM
57
4.3.2. Single Pipe Experiments
For the single pipe system, five separation distances (LPD) from 62 to 132 cm and
velocities (vM) of 0.18 m/s to 1.83 m/s were explored (corresponding to pipe flow Reynolds
numbers (Re = ρvMD/μ, for density ρ, mean velocity vM, pipe diameter D and dynamic
viscosity μ) of 9030 – 90300; indicating turbulent flow for all velocities). Figure 4.4(a) shows
sample NMR signal data as a function of te (i.e. the FID) for a separation distance of 62 cm
and mean velocities ranging from 0.37 to 1.83 m/s. The regularised fit of equation 4.7 for
each mean velocity is also shown. Figure 4.4(b) shows a comparison of velocity probability
distributions, P(v); with measured distributions determined via regularisation of the FID
signals, whilst theoretical distributions are determined using from the power law velocity
profile (equation 4.8) with the power law exponent determined from the Reynold’s number
correlation (Zagarola et al., 1997).
58
Figure 4.4 Velocity distribution measurements obtained using FID acquisition time as the
independent variable. (a) Equation 4.7 fit to FID experimental data at a separation distance of 62 cm
and mean velocities ranging from 0.37 to 1.83 m/s. (b) Velocity probability distributions (P(v))
returned by this regularisation analysis, as well as comparative theoretical velocity distributions.
The model fit to the experimental FID signals is excellent for all measurements in
Figure 4.4(a). Furthermore; the fit is determined across the majority of the signal decay (tdelay
= 9 ms); which allows a velocity probability distribution to be determined without the
presence of erroneous signal artefacts. The measured distributions in Figure 4.4(b) compare
favourably to the theoretical distributions in terms of the mean value and width of the
distributions. Perfect agreement is not expected as the theoretical distributions correspond
to instantaneous velocity whereas the NMR experimental velocity distributions are a
complex average over the polarisation and acquisition time-frame. In addition, the second
0
10
20
30
40
50
0 0.1 0.2 0.3 0.4 0.5
NM
R S
ign
al [µ
V]
Time since excitation pulse [s]
Expt. data
Model fit
(a)
vM
[m/s]0.370.731.101.461.83
0
2
4
6
8
0 0.5 1 1.5 2 2.5 3
P(v
)
Velocity [m/s]
vM
[m/s]0.370.731.101.461.83
(b)Measureddistribution
Theoretical distribution
59
derivative penalty function imposed during regularisation will smooth out the sharp
interfaces evident in the theoretical probability distributions.
The accuracy of the NMR velocity measurement technique is now considered with
respect to the measured mean velocity. Figure 4.5 compares the mean velocity determined
as the expected value of each experimental probability distributions shown in Figure 4.4(b)
with the mean velocity measured from the in-line rotameter. There are five sets of data
corresponding to the five sets of experiments run at different separation distances (ranging
from 62 to 132 cm).
Figure 4.5 Comparison of the mean velocity predicted using regularisation of NMR data to the
measured mean velocity from in-line rotameter data. Five sets of data were obtained at separation
distances of 62 to 132 cm.
In Figure 4.5, the NMR predicted mean velocity compares reasonably well to the
superficial velocity measured using the in-line rotameter. There is no obvious systematic
deviation with separation distance in the accuracy of the mean velocity predicted by the NMR
measurement. For all fifty velocity measurements (across five separation distances and ten
velocities); the mean error (calculated according to equation 3.6) is 0.006 m/s and the root
mean square error (RMSE, calculated according to equation 3.7) is 0.089 m/s indicating good
agreement between the NMR and rotameter measurements. A small error is observed at the
highest velocity (1.83 m/s); the impact of signal loss during the initial acquisition delay (tdelay
= 9 ms) is prevalent. For lower velocities the signal loss during the acquisition delay is
0
0.5
1
1.5
2
0 0.5 1 1.5 2
NM
R m
ea
su
red
ve
loc
ity [
m/s
]
Rotameter measured velocity [m/s]
627893109132
LPD [cm]
60
minimal (e.g. 3.1 % signal loss for 0.37 m/s) however at higher velocities the signal loss from
fluid flush-out during the acquisition delay is much more significant (e.g. 15.3 % at 1.83
m/s). This leads to error in fitting the appropriate velocity distribution to the measured signal
data such that there is discrepancy between the NMR and rotameter velocity measurements
(at higher velocities).
The accuracy of the NMR velocity measurement methodology with respect to
capturing the width of the velocity distribution is now considered. The standard deviation of
each velocity probability distribution is calculated for the velocity measurements at a
separation distance of 62 cm. Figure 4.6 compares the NMR measured velocity distributions
to the theoretical prediction for turbulent velocity distributions in terms of their standard
deviations’ (σv).
Figure 4.6 Comparison of velocity probability distribution standard deviations. Experimentally
determined standard deviations (σv) for the NMR measured velocity distributions (at LPD = 62 cm)
and the corresponding standard deviations of theoretical turbulent velocity distributions.
Figure 4.6 demonstrates that the measured distribution width is underpredicted
relative to the theoretical distribution width (in terms of standard deviation). However the
measured distributions are still competent in capturing the appropriate trend and relative
magnitude of the velocity distribution standard deviation. As highlighted previously (with
respect to Figure 4.4(b)), perfect agreement between the measured and theoretical
distributions is not anticipated. The measured distributions capture a complex time-average
of the flow profile and the smoothing imposed by the regularisation penalty function in the
0
0.05
0.1
0.15
0.2
0.25
0.3
0 0.5 1 1.5 2
σv
[m/s
]
Rotameter measured velocity [m/s]
Measured distribution
Theoretical distribution
61
measured distributions will restrict the ability to capture sharp interfaces observed for the
theoretical distributions.
4.3.3. Two pipe experiments
The second set of experiments conducted involved the two-pipe flow loop shown in
Figure 4.2, in order to test the velocity measurement methodology with an artificially created
bimodal velocity distribution. The ball valves used to control the flow in each pipe were
adjusted such that the velocity in pipe A was 33% of the velocity in pipe B. For the two-pipe
experiments; separation distances of 68, 78, 88 and 98 cm were used and the total water
flowrate through the system varied from 0.9 to 3.0 m3/h (in 0.3 m3/h intervals). Figure 4.7(a)
shows an example of the model fit (obtained through regularisation using equation 4.7) to
the experimental FID data for a separation distance of 68 cm and overall flowrates of 1.8, 2.4
and 3.0 m3/h respectively. Figure 4.7(b) shows the resultant velocity probability distributions
returned by the regularisation analysis. Note that the FID measurements captured in the two-
pipe experiments do not use a Q-switch; thus the initial acquisition delay is 25 ms.
62
Figure 4.7 Velocity measurements for the two-pipe flow system. (a) The model fit (equation 4.7)
to FID experimental data for the two-pipe experiments. The polarisation-detection coil separation
distance is 68 cm and overall flowrates are 1.8 m3/h (with measured velocities of 0.18 and 0.56 m/s
in pipes A and B respectively), 2.4 m3/h (with measured velocities of 0.24 and 0.75 m/s) and 3.0 m3/h
(with measured velocities of 0.32 and 0.96 m/s). (b) The resultant velocity probability distributions
returned by regularisation analysis of the two-pipe NMR data shown in (a).
Figure 4.7(a) demonstrates very good agreement between the experimentally
measured FID signals and the regularised model fit; as would be expected. The two-pipe
velocity distributions in Figure 4.7(b) show two distinct peaks; clearly the methodology is
able to deal with a multi-modal velocity distribution. Each peak is representative of the
contribution from the respective pipe (i.e. the lower velocity peak is from pipe A and the
higher velocity peak is from pipe B). The overall probability distributions were split into the
0
5
10
15
20
25
0 0.1 0.2 0.3 0.4 0.5
NM
R s
ign
al [µ
V]
te [s]
Model fit
1.8 m3/h
2.4 m3/h
3.0 m3/h
Expt. data
(a)Overall flowrate
0
1
2
3
4
5
6
0 0.2 0.4 0.6 0.8 1
P(v
)
Velocity [m/s]
30 l/min
40 l/min
50 l/min
1.8 m3/h
2.4 m3/h
3.0 m3/h
(b)
63
two peaks (using the minima between them) and the mean velocity calculated for each peak.
These mean velocity data are shown in Figure 4.8 plotted against the superficial velocity as
measured using the respective rotameter on each pipe.
Figure 4.8 Comparison of the predicted mean velocity in pipes A and B to the measured superficial
velocity from the corresponding rotameter. Predicted mean velocities are obtained using
regularisation of acquired NMR data, followed by velocity peak splitting. Four sets of data are
obtained at separation distances of 68, 78, 88 and 98 cm with the overall flowrate being varied from
1.8 to 3.0 m3/h .
The mean velocity determined using NMR regularisation agrees reasonably well with
the rotameter velocity measurements for both pipes respectively. For the two pipe
measurements; the RMSE (calculated according to equation 3.7) for pipe A is 0.055 m/s (or
a relative RMSE of 23.9 %) whilst the RMSE for pipe B is 0.062 m/s (or a relative RMSE of
11.1 %). The larger relative error with respect to pipe A is due to the poorer signal-to-noise
ratio (SNR) values at such low velocities. For example, for the 1.2 m3/h two pipe experiment
(with rotameter measured velocities of 0.12 and 0.38 m/s in pipes A and B respectively) at a
separation distance of 68 cm the effective SNR is 16.3 for pipe A and 40.1 for pipe B. This
is as the residence time (and therefore signal decay) between polarisation and detection
becomes significant at lower velocities (v < 0.2 m/s), such that the signal measured at the
detector will have a considerably poorer SNR compared to the SNR at higher velocities. For
each bi-modal velocity probability distribution in Figure 4.7(b), the two peaks are not equal
in magnitude, even though the pipes are of equivalent diameter. The relative magnitude of
the velocity peak attributed to pipe B diminishes in intensity as flowrate increases. This is
partially a consequence of the inability to acquire data for te < tdelay, the initial 25 ms following
0
0.2
0.4
0.6
0.8
1
1.2
0 0.2 0.4 0.6 0.8 1 1.2
NM
R p
red
icte
d m
ea
n v
elo
cit
y
[m/s
]
Measured velocity [m/s]
Pipe A Pipe B
0.68 m
0.78 m
0.88 m
0.98 m
64
signal excitation. As flowrate increases this region of the FID (te < 25 ms) becomes more
influential in determining the relative contribution of pipe B (the larger velocity) to the
overall velocity distribution.
4.4 Conclusions
In this chapter, the EFNMR flow meter demonstrated the ability to accurately
quantify velocity distributions in single-phase pipe flows. Details of the flow metering
equipment is outlined featuring detection using an EFNMR r.f. coil positioned downstream
of a mobile pre-polarisation magnet. In principle, velocity distributions could be determined
by varying this separation distance, however this was shown to be impractical primarily due
to restraints on the minimum distance between the pre-polarisation magnet and the detection
coil. Analysis of the acquired FID signal, however, proved to be much more successful, with
measured velocity distributions demonstrated to be consistent with theory and whose mean
is consistent with independent rotameter measurements for flow through a single pipe.
Furthermore, this approach was successfully demonstrated for artificially constructed multi-
modal (two-pipe) velocity distributions, a key step towards realising MPFMs capable of
measuring the velocity distribution of multi-phase flow.
65
Chapter 5. Characterising two-phase gas/liquid flows
5.1 Introduction
The ability to identify and quantify different gas/liquid flow regimes (e.g. slug flow)
is a considerable challenge in MPFM development (Falcone et al., 2009, Falcone et al., 2002,
Thorn et al., 2013). Fluids transported in subsea pipelines during oil and gas production
exhibit a range of multiphase flow regimes. The most problematic flow regime is slug flow,
which is characterised by the intermittent flow of liquid slugs separated by large gas bubbles.
Slug flow has been shown to reduce the integrity of pipelines and equipment through
increased corrosion (Hill et al., 1996, Jepson et al., 1996, Sun and Jepson, 1992, Villarreal
et al., 2006) and pipe vibration (Zhong et al., 2007). Additionally, slug flow can cause
significant pressure oscillations, with severe pressure fluctuations potentially leading to
safety trips and platform shutdown, increasing operational and capital expenditure as
processing equipment (e.g. separators and compressors) require larger operating margins
(Havre et al., 2000).
Significant research has previously been undertaken to improve the knowledge base
of slug flow through experimental measurement and modelling of multiphase flow
characteristics. The measurement of key slug flow characteristics (e.g. liquid holdup and
slug frequency) has traditionally been achieved by the use of conductance and impedance
probes (Andreussi et al., 1988, Costigan and Whalley, 1997, Nydal et al., 1992) or pressure
transducers (Lin and Hanratty, 1987, Matsui, 1984). Such measurements have been used to
develop a number of widely used slug flow correlations (Andreussi et al., 1993, Barnea et
al., 1980, Dukler and Hubbard, 1975, Issa and Kempf, 2003). However the considerable
range of parameters that influence the complex nature of slug flow has often resulted in
difficulty applying such correlations, particularly in industrial pipelines (Al-Safran, 2009,
66
Zabaras, 2000). Furthermore, these measurement techniques are intrusive, which is an
undesirable characteristic for a true non-invasive MPFM (Falcone et al., 2009).
In this chapter, the EFNMR measurement system is extended to analyse two-phase
air-water flows with the ability to identify and measure flow characteristics in gas/liquid
flows (stratified and slug flow). Continuous monitoring of multiphase flow (particularly slug
flow) is inherently more difficult due to the complex time-varying flow behaviour. NMR
signal interpretation (via Tikhonov regularisation) has been applied to monitor the liquid
holdup and velocity with time. The liquid holdup measurement is compared to a video
analysis of the liquid height of the flowing stream at a transparent pipe section. Finally, the
accuracy of the full NMR measurement procedure applied to two-phase flow is verified by
comparing the average liquid volumetric flowrate (determined by combining the liquid
velocity and holdup tracking) to an in-line rotameter measurement of the liquid volumetric
flowrate in the flow loop water feed.
5.2 Equipment and methodology
5.2.1. Equipment for gas/liquid flows
The EFNMR flow metering apparatus was previously outlined in section 4.2.1
(including a photo and schematic in Figure 4.1). In this section, details relevant to the
introduction of a gas stream to the flow loop apparatus are discussed. Compressed air (20
psi) is injected at a controlled flowrate prior to the pre-polarisation magnet and a 10 – 100
L/min rotameter (Cole Parmer, Australia) is used to measure the gas flowrate prior to mixing.
The compressed air is injected into the flowing water stream and the gas/liquid fluid stream
flows through a transparent acrylic pipe (31 mm internal diameter) in the NMR measurement
section. The upstream Halbach array is maintained at a separation distance of 62 cm relative
to the EFNMR detection coil for the results presented in this chapter. NMR measurements
of the gas/liquid stream acquire free induction decay (FID) signals using the ‘pulse and
collect’ sequence (illustrated in Figure 2.3). The following key parameter ranges were used
for the ‘pulse and collect’ experiments; an acquisition delay between excitation and detection
(tdelay) of 25 ms; a total acquisition time per scan (ta) of 0.5 – 1.0 s and a repetition time of
0.80 – 1.80 s (corresponding to a scan frequency of 0.55 – 1.25 Hz).
67
Video analysis is used to obtain an independent measurement of the liquid holdup. A
video (at 30 frames per second) of the flowing stream is captured during the same
experimental time as the NMR measurements. The video captures a 10 cm transparent
section of the pipe just prior to the detection coil (indicated in Figure 4.1(b)). The flowing
water stream is dyed red when required (using food colouring) to assist video interpretation.
The video is processed using an analysis code developed in Matlab R2017b. The analysis
identifies the colour of the pixels in order to determine the liquid height in the region. The
known cross-sectional area of the pipe is used to adjust the measured liquid height to an
estimate of the liquid holdup.
5.2.2. Gas/liquid NMR signal interpretation
In this chapter, FID measurements are obtained of two-phase air/water flows. During
measurements of air/water flows, only the liquid is contributing to the measurable NMR
signal. This has two important implications; firstly the velocity probability distribution
determined from the FID signals are representative of the liquid velocity distribution alone.
This observation means that the FID signals acquired from two-phase gas/liquid flows may
be described by the model for the NMR signal detected using the flow meter (equation 4.5)
which was developed in Chapter 4. The velocity probability distribution of the liquid phase
within the air/water flows may be measured using the Tikhonov regularisation procedure
described in section 4.2.3; with distributions determined by fitting to the FID signal (i.e.
signal as a function of te)
The second implication of only the liquid phase contributing to the NMR signal is
that the overall level of signal magnetisation (S0) is in principle directly proportional to the
liquid holdup in the detector. This observation can be used to estimate the liquid holdup (hl)
during two-phase flow via the following equation (Krüger et al., 1996);
ℎ𝑙 =𝑆0,𝑇𝑃
𝑆0,𝑟𝑒𝑓 (5.1)
where S0,TP is the overall signal magnetisation for a two-phase measurement, and S0,ref is a
reference value for the overall signal magnetisation determined from single-phase
68
experiments with the pipe full of equivalent liquid. S0 can be obtained from the model signal
fit following regularisation analysis. The model signal fit (Sfit) is determined by combining
the contribution of all velocities to the measured signal via the velocity probability
distribution vector, P(v).
𝑆𝑓𝑖𝑡(𝑡𝑒) = 𝑆0[𝐾(𝑣, 𝑡𝑒)]P(𝑣) (5.2)
where K(v,te) is the model kernel matrix as a function of velocity and time since excitation
described by equation 4.5. S0 can be determined by back-extrapolation from the first point
of acquisition (i.e. at time te = tdelay):
𝑆0 =𝑆𝐹𝑖𝑡(𝑡𝑑𝑒𝑙𝑎𝑦)
[𝐾(𝑣,𝑡𝑒)]𝑃(𝑣) (5.3)
Following calculation of the liquid holdup, the liquid flowrate may be determined. The mean
linear liquid velocity (vl) is determined as the expected value of the liquid velocity probability
distribution (as was done in 0). Then the time-averaged liquid volumetric flowrate ql is;
𝑞𝑙 = 𝑣𝑙ℎ𝑙𝐴 = 𝑣𝑙𝑆0,T𝑃
𝑆0,𝑟𝑒𝑓𝐴 (5.4)
where A is the pipe cross-sectional area.
5.3 Results and discussion
5.3.1. Calibrating the Reference Signal Magnitude
The reference signal magnitude (or full liquid overall signal magnitude, S0,ref) is
determined at a range of velocities (0.09 to 1.14 m/s corresponding to water flowrates of 0.24
to 3.12 m3/h respectively, as measured using the in-line liquid rotameter) for single-phase
water flows. The value should be independent of the fluid velocity as the velocity-related
contributions have been removed from the measured signal. Figure 5.1 displays the measured
S0 values as a function of velocity.
69
Figure 5.1 Reference overall signal magnitude (S0,ref) relative to the measured velocity (single-
phase flows).
The results shown in Figure 5.1 confirm that the S0 is independent of velocity over an
intermediate mean liquid velocity range. The mean value of 352 µV is used as the reference
S0 value for two-phase analysis. The measured reference signal magnetisation at low mean
liquid velocities (v<0.24 m/s) shows some variation. The FID signals acquired at these low
velocities have significantly poorer signal-to-noise (SNR) ratio compared to the SNR of FID
signals at higher velocities (as discussed in section 4.3.3). This is a consequence of the
significant polarisation to detection residence time (τPD = LPD/v) at low velocities. The
reference signal magnitude is also observed to reduce at higher mean liquid velocities (v >
1.00 m/s). The signal loss during the initial acquisition delay (t < tdelay) becomes increasingly
significant at higher velocities as the ‘flush-out’ effect (i.e. excited fluid leaving the detection
region) is more substantial. This signal loss results in an artificial reduction of the predicted
reference signal magnitude for such higher velocities.
5.3.2. Gas/liquid time-averaged results
Two-phase air-water trials were conducted at liquid superficial velocities (vsl) of 0.09
– 0.26 m/s (corresponding to liquid flowrates of 0.24 – 0.72 m3/h) and gas superficial
velocities (vsg) of 0.22, 0.44 and 0.88 m/s (corresponding to gas flowrates of 0.6, 1.2 and 2.4
m3/h). Figure 5.2 shows the position of the fifteen experimental trials on a flow regime map.
0
100
200
300
400
500
0 0.2 0.4 0.6 0.8 1 1.2
Re
fere
nc
e o
ve
rall
sig
nal
ma
gn
eti
sati
on
[µ
V]
Measured mean liquid velocity [m/s]
+ 5%
- 5%
Mean value
70
The visually confirmed flow regimes (i.e. stratified smooth or slug flow) correspond to the
predicted flow regimes from Taitel and Dukler (1976) for all experimental trials.
Figure 5.2 Flow regime map (from Taitel and Dukler (1976)) indicating the relevant flow regime
position of each experimental trial. The trials visually observed to be stratified smooth flow are
indicated by blue circles whilst the trials visually observed to be slug flow are indicated by red crosses.
The experimentally acquired FID signals from each of the fifteen trials were initially
time-averaged (across Nscans = 128) and subsequently analysed using the NMR signal
interpretation procedure discussed in section 5.2.2. Figure 5.3 displays sample time-
averaged analysis for superficial liquid velocities of 0.09 – 0.26 m/s and a superficial gas
velocity of 0.44 m/s. Figure 5.3(a) shows time-averaged NMR signal data as a function of te
(the FID signal) as well as the model fit according to equation 4.5. Figure 5.3(b) shows the
corresponding liquid velocity probability distributions, P(v).
0.01
0.1
1
0.1 1 10
vs
l[m
/s]
vsg [m/s]
Dispersed flow
Intermittent flow
Annular flow
Stratified wavy flow
× – Slug flow
(from visual observation)
– Stratified
smooth flow (from visual observation)
Stratified smooth flow
71
Figure 5.3 Velocity measurements of air-water flows. (a) Equation 4.5 fit to FID experimental data
for two-phase air/ water flows with vsl = 0.09 – 0.26 m/s and vsg = 0.44 m/s. (b) Resultant liquid
velocity probability distributions (P(v)) returned by regularisation analysis of the FID signals.
The stratified flow experiments (vsl = 0.09 and 0.13 m/s) show a single peak
corresponding to the continuous liquid linear velocity distribution anticipated for time-
invariant flow. The slug flow experiments (vsl = 0.18, 0.22 and 0.26 m/s) clearly exhibit a
bimodal velocity distribution, with the larger lower velocity peak corresponding to the
background stratified liquid layer whilst the smaller higher velocity peak corresponds to the
faster moving slugs.
0
5
10
15
20
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
NM
R s
ign
al [µ
V]
Time since excitation pulse [s]
(a)Expt.data
Model fit
vSL
[m/s]
0.090.130.180.220.26
0
4
8
12
16
20
0 0.2 0.4 0.6 0.8 1
P(v
)
Liquid velocity [m/s]
vSL [m/s] Flow regime
0.09 stratified flow
0.13 stratified flow
0.18 slug flow
0.22 slug flow
0.26 slug flow
Contribution of slug component
(b)
72
For each of the fifteen experimental trials shown in Figure 5.2, the time-
averaged liquid volumetric flowrate is calculated using equation 5.4. Figure 5.4
compares the NMR measured liquid volumetric flowrates to the in-line water
rotameter measurements of volumetric flowrate. Excellent agreement is observed
between the two flowrate measurements with a mean error (calculated using equation
3.6) of -0.004 m3/h and a root mean square error (calculated using equation 3.7) of
0.029 m3/h.
Figure 5.4 Comparison of the NMR measurement of time-averaged liquid volumetric flowrate (ql)
to the in-line rotameter measurement of volumetric flowrate.
5.3.3. Gas/liquid single-scan measurements
The results presented in section 5.3.2 demonstrate that the NMR signal interpretation
procedure is accurate and can resolve the presence of slugs from the measured FID signal
(via the liquid velocity probability distribution). However this procedure does not provide
time resolved information of flow characteristics (e.g. liquid velocity and liquid holdup). In
order to provide such time-variant information, the individual FID scans must be interpreted
in a ‘single-shot’ analysis (i.e. each single FID signal acquired is analysed individually as
opposed to analysing the overall time-averaged FID signal as done in section 5.3.2).
0
0.2
0.4
0.6
0.8
0 0.2 0.4 0.6 0.8
NM
R m
ea
su
red
ql[m
3/h
]
Rotameter measured ql [m3/h]
Strat.flow
Slug flow
qg
[m3/h]
0.6
1.2
2.4
73
Figure 5.5(a) shows an example of two 'single-shot’ NMR signal measurements as a
function of te obtained during two-phase flow with vsl = 0.18 m/s, vsg = 0.44 m/s. The first
scan (blue crosses) is obtained when a stratified water layer (flowing underneath an elongated
gas bubble) is detected whilst the second scan (red circles) is obtained when a liquid slug is
flowing through the detection region. The regularised model fit of equation 4.5 to each scan
is also shown. Figure 5.5(b) shows the corresponding liquid velocity probability
distributions, P(v).
Figure 5.5 Single-scan velocity measurement of air-water flow. (a) Equation 4.5 fit to non-time-
averaged experimental signal scans for two-phase air/water flows with vsl = 0.18 m/s, vsg = 0.44 m/s.
The first signal is acquired when the background stratified liquid layer is visually observed to be in
the detection region, whilst the second signal is obtained when a slug is visually observed to be in the
0
2
4
6
8
10
12
14
16
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
NM
R s
ign
al [µ
V]
Time since excitation pulse [s]
Model fit
Expt. data
Stratified layer
Liquidslug
(a)
Photo
0
1
2
3
4
0 0.5 1 1.5 2
P(v
)
Liquid velocity [m/s]
Stratified layer
Liquid slug
(b)
74
detection region. Photos (obtained from the video analysis of two-phase flow) are also shown to
indicate the relevant flow regimes. (b) The corresponding liquid velocity probability distributions
(P(v)) returned by regularisation analysis.
Two key differences can be observed by comparing the detectable signals from the
stratified layer and a liquid slug in Figure 5.5. Firstly, the overall signal magnitude is
considerably higher for the liquid slug signal relative to the stratified layer signal. This is
anticipated due to the higher liquid holdup, which corresponds to an increase in the number
of observable nuclear spins in the detection region. The second difference is that the
observable signal decays faster when detecting a liquid slug due to the increased ‘flush-out’
effect at higher velocity. This is captured in the regularised liquid velocity distribution that
shows the liquid slug is travelling at a higher velocity relative to the stratified liquid layer.
The NMR signal interpretation procedure for two-phase flows discussed in section
5.2.2 and demonstrated in section 5.3.2 for the time-averaged NMR signal data is now
applied to all ‘real-time’ NMR signal scans (i.e. for each scan in Nscans (= 128)) for each of
the fifteen experimental trials. Regularisation is used to fit equation 4.5 to each individual
non-time-averaged FID signal acquired. The corresponding velocity distributions are
returned via the regularisation analysis. The mean of each velocity probability distribution
is used as a measure of the liquid velocity with time. Figure 5.6 shows this liquid velocity
tracked over 60 s at superficial liquid velocities of 0.09, 0.18 and 0.26 m/s and at a gas
superficial velocity of 0.44 m/s (as measured using the in-line rotameters).
75
Figure 5.6 Tracking the liquid velocity over time for two-phase air/water flows. The superficial
liquid velocities are 0.09, 0.18 and 0.26 m/s, whilst the superficial gas superficial velocity is 0.44 m/s
for all three trials. The visually observed flow regime is indicated for each superficial liquid velocity.
The liquid velocity tracking can be observed to be relatively constant (a standard
deviation of 0.018 m/s for 128 scans) during the stratified flow experiment (vsl = 0.09 m/s).
The slug flow experiments show greater fluctuations in the velocity tracking due to the
periodic presence of fast-moving slugs (relative to the slower background stratified liquid
layer). In the first slug flow experiment (vsl = 0.18 m/s) the periodic fluctuations in liquid
velocity are less frequent compared to the second slugging experiment (vsl = 0.26 m/s). This
observation corresponds to longer but less frequent (0.07 Hz) liquid slugs occurring at lower
velocities and shorter but more frequent (0.16 Hz) slugs occurring at higher velocities (the
method for estimating the slug frequency from the liquid holdup is discussed below with
respect to Figure 5.8).
The overall signal magnetisation values (S0,TP) are extracted from the resultant fit of
each individual scan (via equation 5.3) and used to estimate the liquid holdup (using equation
5.1). Figure 5.7 shows the liquid holdup being tracked over 60 s for superficial liquid
velocities of: (a) 0.09, (b) 0.18 and (c) 0.26 m/s. The superficial gas velocity is 0.44 m/s for
all three trials. The NMR measurement of liquid holdup is compared to the simultaneous
video analysis of liquid holdup (described in section 5.2.1).
0.0
0.2
0.4
0.6
0.8
1.0
0 10 20 30 40 50 60
Liq
uid
ve
loc
ity [
m/s
]
Experimental Time [s]
vsl [m/s] Flow regime
0.09 stratified flow
0.18 slug flow
0.26 slug flow
76
Figure 5.7 Tracking the liquid holdup over time for two-phase air/water flows. The superficial
liquid velocities are; (a) 0.09 m/s (stratified flow), (b) 0.18 m/s (slug flow) and (c) 0.26 m/s (slug
flow), whilst the superficial gas velocity is 0.44 m/s for all trials. The liquid holdup determined from
NMR analysis is compared to the liquid holdup determined from video analysis.
A time series cross-correlation between the NMR holdup (hN) and video holdup (hV)
has been applied in order to verify the periodic correlation in the slugging data. The cross-
covariance (cNV) of the NMR holdup (hN) and video holdup (hV) for a time lag of τL is given
by (Chatfield, 2016);
0.00
0.25
0.50
0.75
1.00
0 10 20 30 40 50 60
Liq
uid
Ho
ldu
p
Experimental Time [s]
NMR holdup
Video holdup
(a)
0.00
0.25
0.50
0.75
1.00
0 10 20 30 40 50 60
Liq
uid
Ho
ldu
p
Experimental Time [s]
NMR holdup
Video holdup
(b)
0.00
0.25
0.50
0.75
1.00
0 10 20 30 40 50 60
Liq
uid
Ho
ldu
p
Experimental Time [s]
NMR holdup
Video holdup
(c)
77
𝑐𝑁𝑉(𝜏𝐿) = {
∑(ℎ𝑁,𝑡−ℎ𝑁 )(ℎ𝑉,𝑡+𝜏−ℎ𝑉 )
𝑁
𝑁𝑡=1
𝑓𝑜𝑟 𝜏𝐿 = 0, 1, … , 𝑁 − 1
∑(ℎ𝑁,𝑡−ℎ𝑁
)(ℎ𝑉,𝑡+𝜏−ℎ𝑉 )
𝑁
𝑁𝑡=1−𝜏𝐿
𝑓𝑜𝑟 𝜏𝐿 = −1, −2, … , −(𝑁 − 1)
(5.5)
where t is the video frame number, N is the total number of video frames analysed, ℎ𝑁 is the
mean NMR liquid holdup and ℎ𝑉 is the mean video liquid holdup. The normalised cross-
correlation (rNV) is given by (Chatfield, 2016);
𝑟𝑁𝑉(𝜏) = 𝑐𝑁𝑉(𝜏)
𝜎ℎ𝑁𝜎ℎ𝑉
(5.6)
where 𝜎ℎ𝑁 is the standard deviation of the NMR predicted liquid holdup track and 𝜎ℎ𝑉
is the
standard deviation of the video estimated liquid holdup track. Figure 5.8 shows example
normalised cross correlation relationships for the following experimental trials; (a) vsl = 0.18
m/s, vsg = 0.44 m/s and (b) vsl = 0.26 m/s, vsg = 0.44 m/s.
Figure 5.8 Normalised cross-correlation (rNV) at time lag (τL) between NMR holdup and video
holdup for experimental trials. Measurements are conducted at (a) vsl = 0.18 m/s, vsg = 0.44 m/s and
(b) vsl = 0.26 m/s, vsg = 0.44 m/s.
The maximum peak (at zero time lag) in Figure 5.8 is representative of the similarity
between the two signals (with a value of 1 representing an exact match). The time lag
difference between the peaks in the cross-correlation plot is indicative of the average time
period of the slug units (τs) and can be used to estimate the slugging frequency (fs) for a given
experimental trial (e.g. for trial (a) τs = 15.5 s and fs = 0.07 Hz, whilst for trial (b) τs = 6.1 s
and fs = 0.16 Hz).
-1
-0.5
0
0.5
1
-20 -10 0 10 20
No
rmali
sed
cro
ss-c
orr
ela
tio
n
Time lag [s]
0.18 m/s
0.26 m/s
vsl
(a) (b)
78
From Figures 5.7 and 5.8 it can be observed that the NMR holdup estimate correlates
relatively well to the video holdup. The NMR measurement procedure is successfully able
to identify the presence of slug flow (relative to stratified flow) and give a reasonable estimate
of the liquid holdup. The NMR measurements underestimate the liquid holdup relative to
the video holdup; minor deviations are expected in the video holdup estimate as the analysis
does not account for gas bubbles entrained in the liquid phase and measures the upper edge
of a concave meniscus for the liquid height. Both of these physical reasons cause a slight
over-estimation of the liquid holdup in the video analysis. Furthermore, the presence of gas
bubbles can increase the observable rate of signal decay (as the effective NMR spin-spin
relaxation is faster) in gas-liquid flow, leading to an overestimation of liquid velocity and an
underestimation of liquid holdup using the NMR signal interpretation procedure.
The difference between the holdup estimates is particularly significant immediately
following the measurement of a slug in the system (e.g. at 6 s in Figure 5.7(b)). This
difference can be attributed to the model assumption that the linear velocity of fluid elements
is constant whilst they travel through the flow metering system. This is obviously not true
for time-variant slug flow. The fluid detected immediately following the slug (the
‘subsequent fluid’) will have been subject to shear drag forces by the slug layer. The
subsequent fluid is predominantly composed of fluid shed from the back of the slug layer and
has therefore undergone a complex change in velocity whilst it travels through the flow
metering apparatus. The regularisation procedure will fit the fluid velocity distribution that
was observed at the detection coil. However this ‘subsequent fluid’ will have been travelling
at a significantly higher velocity through the polarisation magnet when subjected to the shear
drag of the slug. The assumed degree of signal polarisation (SP) will therefore be
considerably over-estimated, leading to a significant under-estimation of the liquid holdup
using the back-extrapolation estimation method (equations 5.1 and 5.3).
The NMR holdup measurement procedure is currently unable to detect all slugs at
higher velocities. The current NMR repetition time (0.8 s) limits the measurable velocity
range of the system. As the superficial liquid and gas velocities increase, the residence time
79
of the slug in the EFNMR detection region is reduced. For example, for the experiment trial
shown in Figure 5.7(c) where vsl = 0.26 m/s and vsg = 0.44 m/s, the slug residence time in the
detection coil ranges from 0.2 – 0.4 s (determined from video analysis) meaning such slugs
may be missed or only partially excited (e.g. at 18 s in Figure 5.7(c)).
5.3.4. Statistical analysis of single-scan results
Important flow characteristics can be determined from the analysis of the liquid
volumetric flowrate track. For an individual FID signal scan i, the liquid volumetric flowrate
(ql,i) can be calculated from the liquid holdup for scan i (hl,i) and the liquid velocity for scan
i (vL,i) using equation 5.4.
The gas/liquid flow regimes can be distinguished by analysing the variation (i.e.
standard deviation) in the liquid flowrate track for a given experiment across all scans. Figure
5.9 plots the standard deviation of liquid flowrate (𝜎𝑞𝑙) across each experiment (Nscans = 128)
for all fifteen trials.
Figure 5.9 Liquid flowrate standard deviation for the stratified and slug flow experiments.
Experimental trials in the stratified flow regime are observed to have very little
variation in the liquid flowrate (𝜎𝑞𝑙<0.06 m3/h) as anticipated for time-invariant flow,
whereas trials in the slug flow regime have more significant spread in the liquid flowrate
0.01
0.1
1
10
0 0.2 0.4 0.6 0.8
NM
R m
ea
su
red
qls
tan
da
rd
devia
tio
n [
m3/h
]
Rotameter measured ql [m3/h]
Stratified flow
Slug flow
qg [m
3/h]
0.6 1.2
2.4
80
(𝜎𝑞𝑙>0.27 m3/h). The significant difference in the spread of flowrates that is highlighted here
may be used as a practical measurement technique to differentiate stratified and slug flow.
The accuracy of the NMR velocity measurement procedure applied to individual
acquisitions can be verified using the liquid flowrate measurement. The average liquid
volumetric flowrate (ql) for an entire NMR experiment (i.e. for Nscans = 128 scans) is
determined by:
𝑞𝑙 = ∑ 𝑞𝑙,𝑖
𝑁𝑠𝑐𝑎𝑛𝑠𝑖=1
𝑁𝑠𝑐𝑎𝑛𝑠 (5.7)
where ql,i is the NMR predicted liquid volumetric flowrate for scan i. Figure 5.10 compares
the NMR predicted average liquid flowrate (determined from analysis of individual scans) to
the measured liquid flowrate from the in-line water rotameter.
Figure 5.10 Comparison of the NMR predicted average liquid flowrate to the independent rotameter
measurement of liquid flowrate. Experimental trials visually observed to be in the stratified (strat.)
flow regime are indicated by open markers whilst slug flow trials are indicated by closed markers.
The average liquid volumetric flowrate predicted by analysis of the individual NMR
scans agrees reasonably well with the rotameter measurement of liquid flowrate. The root
mean square error (RMSE) for the fifteen velocity measurements is 0.058 m3/h. The RMSE
obtained here (from analysis of non-time-averaged scans) is higher than the RMSE of 0.029
m3/h determined in Figure 5.4 (from analysis of the time-averaged scans). The reduction in
0
0.2
0.4
0.6
0.8
0 0.2 0.4 0.6 0.8
NM
R p
red
icte
d q
l[m
3/h
]
Rotameter measured ql [m3/h]
Strat.flow
Slug flow
qg
[m3/h]
0.6
1.2
2.4
81
accuracy is anticipated due to the substantial reduction in SNR (e.g. for ql = 0.48 m3/h, qg =
1.2 m3/h the time-averaged SNR is 73 whilst the average SNR for non-time-averaged scans
is 9). Whilst the regularisation procedure is relatively robust, this significant reduction in
SNR will cause a loss of accuracy. However the slight loss of accuracy is an acceptable
compromise given the ability to track the liquid velocity and holdup with respect to time
when applying the interpretation procedure to analyse non-time-averaged scans.
5.4 Conclusions
In this chapter, the application of the EFNMR multiphase flow metering system to
monitor gas/liquid stratified and slug flow regimes is demonstrated. Appropriate signal
interpretation via the application of Tikhonov regularisation allows the determination of the
velocity probability distribution of the liquid phase in air-water flows. Signal analysis
applied to individual FID signal measurements allows tracking of the liquid holdup and
velocity with time (at 0.55 – 1.25 Hz). The NMR estimate of liquid holdup is quite
comparable with a video analysis of the holdup. Furthermore, the accuracy of the liquid
velocity distributions are validated by comparison with an independent rotameter
measurement of the liquid volumetric flowrate. Finally, the variation in the liquid flowrate
may be used to distinguish stratified and slug flow regimes.
82
Chapter 6. Developing a technique for simultaneous oil and
water flowrate measurement
6.1 Introduction
This chapter details the development of an appropriate measurement technique for in-
situ quantification of the oil and water component flowrates. The EFNMR flow loop is
extended to be capable of performing oil/water flows experiments. The appropriate fluid
distinguishing mechanism and the relevant signal sensitivity for NMR measurement of
oil/water mixtures is discussed. The measurement system and data analysis are appropriately
adapted and extended for simultaneous velocity measurement and component distinction (via
T1 relaxometry). The 1H NMR signal of the multiphase fluid stream is acquired at varying
pre-polarisation conditions using a dual polarisation mechanism. The acquired FIDs are
interpreted via a two-dimensional signal inversion in order to provide the joint velocity-T1
probability distribution. Subsequent analysis allows the determination of individual phase
flowrates. Finally the application of the developed two-phase measurement procedure is
validated against in-line fluid rotameters both for single-phase oil and water flows.
6.2 Equipment and methodology
6.2.1. NMR fluid quantification of oil/water systems
NMR provides multiple measurement options in terms of quantifying the relevant
phase fractions of a two-component system (such as oil and water). NMR spectroscopy can
be applied to quantify phase fractions due to the chemical shift of oil and water components
(e.g Wagner et al., 2016). However it is not possible to achieve the spectral resolution for
distinguishing oil and water at the Earth’s magnetic field, or at any practical field strength
for multiphase flow metering. Therefore a different approach must be applied in order to
quantify the relevant phase fractions.
83
NMR fluid relaxometry (i.e. relaxation rate measurement) is used across a range of
applications to study material properties and to distinguish relevant phase populations. For
example, in NMR well-logging fluid volumes (clay bound, capillary bound and free water)
are characterised using petrophysical analysis of T2 distributions (Freedman and Heaton,
2004, Kleinberg, 2001b). The T2 distributions of flowing oil/water emulsions have been used
to quantify the emulsion phase fractions (Allsopp et al., 2001, Kantzas, 2004, Kantzas et al.,
2009). This has been achieved by sampling flow and analysing a stationary sample such that
there is sufficient acquisition time to accurately perform a relaxometry measurement. There
is difficulty applying T2 measurements in-situ (i.e. directly on the flowing sample), since the
acquisition time is limited by velocity and various flow effects can impact the acquired NMR
signal (Mueller et al., 1986).
In contrast to T2 relaxometry, analysis of T1 relaxation is a more robust and flexible
approach to differentiating oil and water. T2 measurements are achieved using multi-pulse
acquisition sequences (e.g. CPMG measurements), however T1 signal contrast can be
achieved in our instrumentation through variation of the pre-polarisation conditions and thus
is not limited in terms of acquisition time. Furthermore, T2 measurements are more sensitive
to internal magnetic field gradients arising from magnetic susceptibility differences between
phases in emulsified flows (Fridjonsson et al., 2014a). T1 differentiation will be used as the
fluid phase distinguishing mechanism in this work. Further development of signal contrast
mechanisms is discussed in sections 6.2.3 – 6.2.5.
6.2.2. Equipment for oil/water measurements
The EFNMR flow metering system presented in Chapter 4 is extended in order to be
capable of performing two-phase oil/water flow measurements. The advanced flow loop is
shown in a photo (Figure 6.1(a)) as well as a process and instrumentation diagram (Figure
6.1(b)). The EFNMR measurement system (i.e. the Halbach array and r.f. detection coil) are
the same as previously detailed in Chapter 4. A shorter separation distance (45 cm) relative
to the original separation distance (62 cm) for results presented in Chapters 4 and 5 was
achieved by using additional mu-metal shielding of the Halbach array as well as improved
shimming of the detection coil relative to the stray field. Shorter distances (less than 45 cm)
84
are not achievable due to the stray field from the Halbach array significantly interfering with
the detection coil. An upper limit of 150 cm is imposed on the Halbach separation distance
as larger distances are impractical for realistic flow metering system construction as well as
current practical lab constraints. The EFNMR detection system includes an electromagnetic
pre-polarising coil (a coaxial solenoid of diameter 10 cm, operating at 2.5 V and 6 A) which
can be used to provide a polarisation field to generate magnetisation at the detection coil.
The combined polarisation scheme (simultaneously using the Halbach and the
electromagnetic coil) is discussed in section 6.2.4. The EFNMR detection coil was also
modified to include a resistive Q-switch (Zhen et al., 2018). The application of the 90°
excitation pulse during NMR measurements results in residual stored energy causing
unwanted voltage oscillation (coil ring-down) in the r.f. probe. The Q-switch enhances the
rate of energy dissipation after the excitation pulse which allows reduction of the probe ring-
down time. This enables the acquisition delay time (time between excitation and initial signal
acquisition) to be reduced from 25 ms to 9 ms; allowing earlier acquisition of free induction
decay (FID) signals. This is essential for capturing the short relaxation behaviour of oil as
well as high velocity flows where the effect of fluid flush-out during the acquisition delay is
prominent.
Canola oil (purchased from Oil2U, Australia) and tap water are stored separately in
two intermediate bulk containers (1.0 m3 capacity). Canola oil was selected as an appropriate
oil for flow metering validation measurements primarily due its magnetic resonance
relaxation properties (T1 ~ 115 ms and T2 ~ 82 ms; see section 6.2.3) which allow the fluid
to be readily distinguished from water via NMR relaxometry. Furthermore canola oil has
physical properties (density of 920 kg/m3 and viscosity of 55 cP) which allow a range of
two-phase oil/water flow regimes (e.g. stratified flow, emulsions) to be achieved within the
flow metering system. Canola oil was transported via a close-coupled centrifugal pump
(Calpeda NM32-20A) and the individual phase flowrates monitored using an in-line
rotameter (Stubbe DFM350, 2.5 – 25 m3/h) prior to mixing. The tap water is concentrated
with sodium chloride (12 wt %) in order to reduce the separation time of oil-in-water
emulsions during two-phase flowing experiments (Klaus et al., 2012, Zolfaghari et al., 2016).
The addition of salt causes an increase in the water density (1.09 kg/m3) and viscosity (1.25
cP). Saline water is pumped using a centrifugal pump (Calpeda MXHLM803). The water
85
flowrate is monitored using a second in-line rotameter (Stubbe DFM350, 1.0 - 10 m3/h). Both
of the rotameters were re-calibrated using gravimetric measurement of fluid outflow in order
to account for the used fluid viscosities and densities. The individual phase flowrates are
adjusted using ball valves (V-02 for water and V-05 for oil in Figure 6.1(b)). During single-
phase experiments (i.e. only using water or only using oil alone), bypass lines are used such
that fluid returns directly to the appropriate tank after the NMR measurement section. During
two-phase flow (i.e. water and oil flowed simultaneously through the system) a gravimetric
oil/water separator (separation volume of 128 L) is used in order to split the two phases before
they return to their individual storage tanks.
86
Figure 6.1 Illustration of the experimental oil/water flow loop system. (a) A photo of the flow loop
apparatus with key system components indicated. The direction of flow through the NMR
measurement section is also indicated. (b) A process and instrumentation diagram of the flow loop.
((a)
EFNMR detection coil
Oil storage tank
Water storage tank
Water pump
Flow direction
Oil/water Separator
Pre-polarising Halbach magnet
Oil pump
(b)
87
6.2.3. T1 measurements of stationary fluids
The flow metering system will use T1 differentiation in order to quantify oil and water
(as discussed in section 6.2.1). Initial stationary T1 measurements of oil/water mixtures are
conducted to gain an understanding of the fluid magnetic relaxation properties prior to
flowing measurements. Experiments are conducted for eleven samples with the oil
volumetric fraction stepped from 0 – 100% (in 10 % increments). Each 500 mL sample is
placed in a bottle in the EFNMR detection coil. During stationary measurements of mixtures,
the liquids exist as stratified layers; the oil phase sits on top of a water layer in the bottle.
The mixture T1 distributions are measured using an excitation-delay method. FID signals are
acquired using multiple pulse and collect sequences (illustrated in Figure 2.3) with varying
polarisation-excitation delay time (i.e. tPE). Note that a minimum polarisation-excitation
delay of 60 ms is required due to instrument ring-down following the electromagnetic
polarisation pulse. For all mixtures except the 100 % oil sample; FID measurements are
acquired at 64 polarisation-excitation delay times logarithmically spaced from 65 ms to 6.5
s (with Nscans = 4). Improved accuracy was desired for the 100 % canola oil sample; therefore
the 64 polarisation-excitation delay times were logarithmically spaced between 65 and 600
ms (with Nscans = 16). Longer delay times are not necessary for canola oil due to the shorter
T1.
FID signals are back-extrapolated to determine the polarised signal prior to excitation
(SE). The measured polarised signal is modelled using the following equation;
𝑆𝐸(𝑡𝑃𝐸 , 𝑇1) = 𝑆𝑃 exp (−𝑡𝑃𝐸
𝑇1) (6.1)
where SP is the effective NMR signal after pre-polarisation. The measurement effectively
observes excited protons returning to their thermal equilibrium state in the Earth’s magnetic
field, thus is measuring T1 in the Earth’s field (further details regarding relaxation in variable
magnetic field strength is provided in appendix A.2). For each mixture; a T1 distribution is
fit to the measured signal decay with equation 6.1 using 1D Tikhonov regularisation
(described in section 2.3.1). Figure 6.2(a) shows the model fits to experimental signal data
88
and Figure 6.2(b) shows the resultant probability distributions for T1.
Figure 6.2 T1 measurements of oil/water mixtures. (a) the experimental signal decay as a function
of polarisation-excitation delay fit with the appropriate decay model (equation 6.1) using Tikhonov
regularisation, and (b) the T1 probability distributions for the oil water mixtures returned from
regularisation.
Figure 6.2(b) demonstrates that water exhibits a mono-exponential decay for T1 of
3.1 s. In comparison the canola oil exhibits bi-exponential T1 behaviour with the first peak
representing 64% of the sample with a log mean T1 of 81 ms and the second peak representing
0
20
40
60
80
0 1 2 3 4 5 6 7 8
NM
R S
ign
al [µ
V]
Polarisation - excitation delay [s]
Oil Fraction [%]0
102030405060708090100
Expt data
Model fit
(a)
0
1
2
3
4
5
0.01 0.1 1 10
P(T
1)
T1 [s]
0%10%20%30%40%50%60%70%80%90%100%
Oil fraction
(b)
89
36% of the sample with a log mean T1 of 217 ms. The bi-exponential relaxation behaviour
has been previously observed for canola oil (Luthria, 2004). Note that the two peaks of the
oil distribution are only effectively resolved for the 100 % sample where sufficient SNR is
achieved and more appropriate timing parameters are utilised (e.g. for the 90 % oil mixture
with Nscans = 4, the effective SNR of the oil is 35.8, whilst for the 100 % oil mixture with
Nscans = 16, the SNR is 74.5). Flowing measurements will observe a low SNR (similar to that
of the mixtures) such that the two canola oil peaks will in all likelihood merge into a single
peak. The overall distribution log-mean T1 (for the 100 % canola oil sample in Figure 6.2(b))
of 115 ms is thus a more useful comparison to results with lower SNR. The canola oil T1
relaxation behaviour measured in the EFNMR system is validated in appendix B.2 with
comparison to relaxometry measurements using a Magritek 2 MHz rock core analyser.
The relaxation distributions of oil/water mixtures presented in Figure 6.2(b)
demonstrate that the individual components can be readily differentiated via relaxometry.
All mixtures display two distinct peaks; the first peak at a short T1 of ~ 100 ms (associated
with canola oil) and a second peak at a higher T1 of ~ 3 s (associated with water). In Figure
6.2(a) the considerable difference in the initial signal amplitude between canola oil and water
(for single-phase samples) can be attributed to the initial signal delay between polarisation
and excitation (60 ms). The shorter T1 of canola oil results in a significant signal loss of 55
% during the initial delay (relative to the minor 1.9 % signal loss for water).
The accuracy of the T1 measurements of stationary mixtures in terms of quantifying
oil composition is now considered. The oil and water phases are distinguished using a T1
cut-off value (T1,C; similar to T2 cut-offs used to differentiate bound fluid and free fluid in
NMR analysis of rock cores (e.g. Hürlimann and Heaton, 2015)). For this work, a cut-off of
T1,C = 600 ms is used (calculated as the log-mean of the T1 relaxation rate for oil (115 ms)
and the T1 relaxation rate for water (3.1 s)). The signal contribution of each phase is
calculated by integrating over the relevant region of the T1 probability distribution;
𝑆𝑜 = ∫ 𝑃(𝑇1) 𝑑𝑇1𝑇1,𝐶
𝑇1,𝑚𝑖𝑛 (6.2)
𝑆𝑤 = ∫ 𝑃(𝑇1) 𝑑𝑇1𝑇1,𝑚𝑎𝑥
𝑇1,𝐶 (6.3)
90
where So and Sw are the oil and water signal contribution to the model fit respectively whilst
T1,min and T1,max define the bounds of the discretised T1 range. The individual phase fractions
are calculated from the relevant signal contribution of each phase in a mixture (Kleinberg
and Vinegar, 1996);
𝑥𝑤 =𝐻𝐼𝑜𝑖𝑙
𝐻𝐼𝑜𝑖𝑙+ 𝑆𝑜𝑆𝑤
(6.4)
𝑥𝑜 = 1 − 𝑥𝑤 (6.5)
where xw is the water phase fraction (water-cut), xo is the oil phase fraction and HIoil is the
hydrogen index for canola oil. The value for HIoil (0.96) has been experimentally determined
by comparing the measured signal intensity of canola oil and water samples obtained from
CPMG measurements in a Magritek 2 MHz NMR rock core analyser. The hydrogen index
is dependent on the atomic density of hydrogen in canola oil relative to water and is not
expected to be frequency dependent. Therefore, the value determined at 2 MHz is assumed
to be valid in the 2.29 kHz EFNMR system.
The NMR measured oil and water phase fractions are determined for each of the 9
stationary oil/water mixtures from 10 – 90 % (i.e. excluding the pure oil and pure water
samples). Figure 6.3 compares the NMR measured phase fractions to the volumetric cylinder
measurements of phase fractions (i.e. the fluid volume measurements obtained using a
volumetric cylinder during sample preparation) for both oil and water.
91
Figure 6.3 Phase fraction comparison for oil and water. The NMR measured component phase
fractions are validated against volumetric cylinder measurements obtained during sample preparation.
Equality lines are included for reference.
Figure 6.3 demonstrates good agreement between NMR and volumetric cylinder
measurements of phase fractions. The root mean square error for the nine mixtures is 2.3%.
This demonstrates that NMR T1 measurements can be used for accurate in-situ phase
quantification for oil/water mixtures. T1 measurements will now be incorporated into the
flow metering measurement procedure in order to quantify phase fractions of flowing
mixtures.
6.2.4. Measuring T1 of flowing fluids
The multi-exponential relaxation behaviour observed for the canola oil is useful for
demonstrating the applicability of the system to measuring complex multi-modal relaxation
behaviour anticipated for crude oils (Fridjonsson et al., 2014a). However the short relaxation
behaviour for oil introduces an additional difficulty in terms of obtaining measurements with
reasonable SNR. The signal attenuation during the intermediate decay (SPD) will be
significant for oil due to its low T1 (relative to water), particularly at low velocities (<1.0
m/s). In order to demonstrate the considerable signal decay, measurements of the SNR
(calculated as initial FID signal relative to average measured noise) obtained with canola oil
(alone) flowing through the system at velocities of 0.17 – 1.83 m/s are presented in Figure
6.4. FID measurements are obtained using 32 scan averages (Nscans = 32) at the shortest
separation distance used of 45 cm.
0%
20%
40%
60%
80%
100%
0% 20% 40% 60% 80% 100%
NM
R m
ea
su
red
fra
cti
on
Cylinder measured oil fraction
Oil
Water
92
Figure 6.4 SNR for FID measurements of canola oil flowing through the NMR flow meter with
signal pre-polarisation achieved using the Halbach array. The oil is travelling at velocities of at
velocities of 0.17 – 1.83 m/s and the polarisation-detection separation distance is 45 cm.
Figure 6.4 shows that at low canola oil velocities (<1.0 m/s) the SNR is poor (<5).
The low signal measured from the oil is a result of the significant intermediate signal decay
between polarisation and detection (SPD, as in Equation 4.3). Obtaining measurements with
such low SNR values was found to produce unreasonable solutions during signal processing
due to data inversion instability. Therefore, an alternative methodology must be considered
in order to obtain reasonable signal from the canola oil phase.
The electromagnetic pre-polarising coil (EPPC) which is associated with the EFNMR
detection coil (and used for stationary fluid measurements) was previously considered
unsuitable for flowing measurements (Fridjonsson et al., 2014b). The fluid flush-out effect
restricts the residence time in the coil such that the EPPC does not achieve sufficient
polarisation of flowing water. However the application of the EPPC can be considered useful
for fluids at low velocity and low T1. FID measurements of canola oil are obtained using
only the EPPC (i.e. without the upstream pre-polarising Halbach) with a pre-polarising time
of 600 ms. Figure 6.5 shows the measured SNR determined from FID measurements of
canola oil flowing at velocities of 0.17 – 1.83 m/s.
0
5
10
15
20
0 0.5 1 1.5 2
Me
as
ure
d S
NR
Velocity [m/s]
93
Figure 6.5 SNR for FID measurements of canola oil flowing through the NMR flow meter with
signal pre-polarisation achieved using the EPPC. The oil is travelling at velocities of 0.17 – 1.83 m/s
and the EPPC pre-polarising time is 600 ms.
Figure 6.5 demonstrates that the SNR is sufficient (>5) for low velocities (<2.0 m/s)
where the fluid outflow effect is reduced. At higher velocities, the outflow effect begins to
constrain the residence time of fluid in the pre-polarising region, therefore the measurable
signal is poorer (relative to low velocities).
By comparing Figures 6.4 and 6.5, it can be observed that the two polarising
mechanisms (i.e. the upstream Halbach array and the EPPC) are effective at different oil
velocity ranges. To further demonstrate this, the model for NMR signal polarisation of a
flowing fluid (which is fully defined below in section 6.2.5) has been used to calculate the
initial NMR FID signal (i.e. the first point of acquisition after the excitation-acquisition
delay) for the two separate polarisation conditions. The model is applied across a velocity-
T1 space of 0 – 3 m/s (for velocity) and 10 ms to 10 s (for T1). Figure 6.6 displays the model
output in terms of a 2D SNR map across the velocity-T1 space. Figure 6.6(a) shows the result
when using the Halbach array for pre-polarisation (at a minimum separation distance of 45
cm), whilst Figure 6.6(b) shows the result when applying the EPPC for pre-polarisation (with
a polarisation time of 600 ms). Figure 6.6(c) is the combined polarisation (i.e. the effective
sum of the two contributions). The model incorporates a noise value based on FID
measurements with 32 scan averages (Nscans = 32). Contour lines (in pink) for an SNR of 5
are included to indicate the appropriate regions where reasonable signal interpretation is
anticipated.
0
5
10
15
20
0 0.5 1 1.5 2
Me
as
ure
d S
NR
Velocity [m/s]
94
Figure 6.6 Model SNR of fluid phases in velocity-T1 space with application of two separate
polarising techniques in the EFNMR flow metering system. (a) The model SNR when applying the
Halbach magnet as the pre-polarising mechanism (with a separation distance of 45 cm), and (b) The
model SNR when applying the electromagnetic pre-polarising coil (with a polarisation time of 600
ms). (c) Combined polarisation incorporating the effective sum of the two contributions; however,
this is clearly dominated by the Halbach contribution. The pink lines are contours for SNR = 5;
highlighting the regions where reasonable interpretation of the measured signals can be anticipated.
From Figure 6.6 it can be observed that in general the Halbach permanent magnet
provides a much stronger signal across a broad range of velocities compared to the EPPC.
However the EPPC is successfully able to fill the void for low velocity (< 1.0 m/s) and low
T1 (50 – 600 ms) fluids (e.g. oil) where the Halbach polarisation has a poor SNR due to
intermediate signal decay between polarisation and detection. Therefore by combining the
polarising methods in a dual polarisation mechanism, the flow metering system can more
effectively measure signals across a range of velocities (0.1 – 3 m/s) and fluid T1 values (50
ms – 10 s). Considering that it is impractical to remove the Halbach array from the pipeline
during flow measurements, the combined measurements always incorporate Halbach array
pre-polarisation, with the option of additional re-polarisation using the EPPC once the fluid
reaches the EFNMR system.
Measurement of the fluid T1 under flow requires an independent variable to observe
signal contrast according to T1. For measurements obtained using the Halbach array alone;
the obvious independent variable is the separation distance which can easily be adjusted via
movement of the mobile Halbach magnet. Signal contrast with the EPPC can be observed
95
by varying the pre-polarisation time (tpolz). Thus variable separation distances (45 – 150 cm)
will be used to measure signal contrast for fluids with higher T1 (and/or at high velocities)
whilst the variable pre-polarisation time (10 – 600 ms) will be used to measure signal contrast
for low T1 fluids with low velocity. The selection of appropriate sequence parameters will
be discussed in more detail in section 6.3.1.
6.2.5. NMR flow model development for signal polarisation
The model of the NMR signal of a flowing fluid is now extended; focussing on the
incorporation of the secondary polarisation mechanism. The dual polarisation mechanism
involves the application of the EPPC in some of the FID measurements to provide additional
re-polarisation, whilst the permanent Halbach array will be applied for all experiments as it
is fixed around the pipeline. For experiments where the EPPC is not applied (i.e. the Halbach
magnet alone is used for pre-polarisation), the effective signal polarisation due to the Halbach
just prior to excitation (SPH), including the intermediate signal decay term, is:
𝑆𝑃𝐻(𝐿𝑃𝐷 , 𝑣, 𝑇1) = 𝑆0𝐻 (1 − exp (−𝐿𝑃𝐻
𝑣 𝑇1)) exp (−
𝐿𝑃𝐷
𝑣𝑇1) (6.6)
where S0H is the overall signal magnetisation after an infinite time in the Halbach field and
LPH is the length of the Halbach magnet. Therefore the fraction of fluid which is not polarised
(xNP) upon reaching the EFNMR detection coil is:
𝑥𝑁𝑃 = 1 −𝑆𝑃𝐻
𝑆0𝐻 (6.7)
This represents fluid which was either not polarised at the Halbach array or has decayed to
its original energy state during the residence time between polarisation and detection. The
effective signal polarisation of a stationary fluid (SPC) due to the EPPC can be described by
the T1 signal development (Pendlebury et al., 1979):
𝑆𝑃𝐶(𝑡𝑝𝑜𝑙𝑧, 𝑇1) = 𝑆0𝐶 (1 − exp (−𝑡𝑝𝑜𝑙𝑧
𝑇1)) exp (−
𝑡𝑃𝐸
𝑇1) (6.8)
where S0C is overall signal magnetisation after an infinite time in the EPPC field, tpolz is the
polarisation time and tPE is the polarisation-excitation delay (60 ms). For measurements
conducted on flowing fluids; the effective polarisation time is limited by the fluid residence
time in the polarising coil (Krüger et al., 1996, Krüger et al., 1984). A fluid element
96
travelling with velocity v will leave the EPPC region (with an effective length LPC of 27 cm)
after a residence time τPC = LPC/v, thus equation 6.8 is modified for flowing fluids:
𝑆𝑃𝐶(𝑡𝑝𝑜𝑙𝑧, 𝑇1) = 𝑆0𝐶 (1 − exp (−𝜏𝑃𝐶
𝑇1)) exp (−
𝑡𝑃𝐸
𝑇1)
where 𝜏𝑃𝐶 = min (𝑡𝑝𝑜𝑙𝑧,𝐿𝑃𝐶
𝑣− 𝑡𝑃𝐸) (6.9)
Note that the polarisation-excitation delay term is included in the minimisation expression to
account for fluid which would leave the coil during the delay time. The incorporation of the
minimisation term effectively restricts the range of useful polarisation times according to the
fluid velocity; larger pre-polarisation times (tpolz > 600 ms) are ineffective as the fluid will
flush through the coil before sufficient polarisation can be achieved. When combining the
dual effects of Halbach polarisation and the EPPC, only fluid which is not polarised upon
reaching the EFNMR (quantified by equation 6.7) can be re-polarised by the EPPC.
Therefore, the combined signal polarisation (SP) is modelled by:
𝑆𝑃 = 𝑆𝑃𝐻 + ൬1 −𝑆𝑃𝐻
𝑆0,𝐻൰ 𝑆𝑃𝐶 (6.10)
Note that if the pre-polarising coil is not applied (i.e. τPC = tpolz = 0) then SP = SPH, meaning
that the overall signal polarisation is just the signal polarisation due to the Halbach. The
overall model for the NMR signal of a flowing fluid (previously outlined in equation 4.5) can
now be considered as:
𝑆(𝐿𝑃𝐷 , 𝑡𝑝𝑜𝑙𝑧, 𝑡𝑒 , 𝑣, 𝑇1) = 𝑆𝑃 (1 −𝑡𝑒𝑣
𝐿𝐷) exp (− (
𝑡𝑒
𝑇2+ (𝑅𝐼𝑡𝑒)2)) for 𝑡𝑒 ≤
𝐿𝐷
𝑣, (6.11)
where SP is defined in equation 6.10.
The final FID component to be considered for flow measurements is the fluid T2
signal decay under flow. The effective T2 relaxation during flowing experiments will be a
function of the fluid composition considering the differing T2 relaxation rates for canola oil
and water. The measurement of fluid T2 distribution simultaneously with velocity and T1
under flow would be very difficult; therefore a T1/T2 ratio is introduced in order to link the
modelled T2 decay to the measured T1 decay. The T1/T2 ratio is defined as; RT = T1/T2 and is
specified about the oil/water T1 cut-off (T1,C = 600 ms, as discussed in section); with oil (T1
< 600 ms) having RT = 1.35 and water (T1 ≥ 600 ms) using RT = 1.63 (determined from
97
stationary fluid relaxation measurements). The T1/T2 ratio (RT) is introduced into the model
for NMR signal of a flowing fluid:
𝑆(𝐿𝑃𝐷 , 𝑡𝑝𝑜𝑙𝑧, 𝑡𝑒, 𝑣, 𝑇1) = 𝑆𝑃 (1 −𝑡𝑒𝑣
𝐿𝐷) exp (− (
𝑡𝑒∗𝑅𝑇
𝑇1+ (𝑅𝐼𝑡𝑒)2)) for 𝑡𝑒 ≤
𝐿𝐷
𝑣 (6.12)
This effectively removes T2 from the model equation, leaving velocity (v) and spin-lattice
relaxation (T1) as the only dependent variables.
6.2.6. Velocity–T1 distribution determination via 2D inversion
The objective of the flow meter is to measure the phase flowrates of water and oil;
this will be achieved by measurement of the velocity distribution as well as phase fraction
determination via T1 differentiation. Therefore the velocity and T1 distributions must be
measured simultaneously as a joint 2D velocity-T1 distribution. This requires careful
consideration in the application of 2D inversion techniques.
The model for NMR signal of a flowing fluid (equation 6.12) now effectively
describes the relationship between the experimental parameters (i.e. LPD, tpolz and te) and the
measured parameters (i.e. v and T1) and is used as the model kernel function for 2D inversion.
The kernel function can be considered in terms of the “direct” and “indirect” dimensions; the
direct measurement is obtained from the single-shot FID signal which is detected (SD), whilst
the indirect measurement corresponds to variation in the pre-polarising conditions (SP). The
model kernel function is simplified as:
𝐊(𝐿𝑃𝐷 , 𝑡𝑝𝑜𝑙𝑧 , 𝑡𝑒 , 𝑣, 𝑇1) = 𝑆𝑃𝑆𝐷
where 𝑆𝐷 = (1 −𝑡𝑒𝑣
𝐿𝐷) exp (− (
𝑡𝑒∗𝑅𝑇
𝑇1+ (𝑅𝐼𝑡𝑒)2)) for 𝑡𝑒 ≤
𝐿𝐷
𝑣 (6.13)
where SP has been previously described in equation 6.10. The kernel function is non-
separable as the measured parameters (v and T1) are present in both the direct and indirect
dimensions. Variable substitution is used in 2D regularisation inversions in order to allow
kernel separability; however this results in data truncation and back-extrapolation to fill
missing data which can cause the creation of artificial components in the resultant 2D
distribution (Leu et al., 2005, Mitchell and Fordham, 2011). Therefore a full kernel matrix
98
(i.e. non-separable matrix) is constructed to avoid such undesirable consequences during data
analysis (Mitchell et al., 2010a). A pseudo-1D inversion is applied (using appropriate matrix
manipulation) with the full kernel matrix in order to determine the 2D velocity-T1 probability
distribution. The relevant first-order Fredholm integral is now written as:
𝑆(𝐿𝑃𝐷 , 𝑡𝑝𝑜𝑙𝑧, 𝑡𝑒) = ∫ ∫ 𝐾(𝐿𝑃𝐷 , 𝑡𝑝𝑜𝑙𝑧, 𝑡𝑒 , 𝑣, 𝑇1)∞
0𝑃(𝑣, 𝑇1)
∞
0 𝑑𝑣 𝑑𝑇1 (6.14)
where S(LPD, tpolz, te) are the experimentally acquired NMR signals, K(LPD, tpolz, te, v, T1) is
the model kernel matrix described in equation 6.13 and P(v, T1) is the joint 2D probability
distribution to be determined. The NMR signals consist of FID measurements (with nF points
recorded for each FID) acquired at variable pre-polarising conditions. The FID
measurements are compressed using window-averaging (Mitchell et al., 2014) in order to
obtain signal matrices of reasonable size for inversion processing; the FID signals are divided
into nw equally sized windows or bins (FID signals are generally dominated by a linear
outflow effect; therefore using equally spaced windows is appropriate (Istratov and Vyvenko,
1999)). The data points within each bin are averaged to provide a compressed FID signal of
size nw (in this work nw = 32). If FID measurements are acquired at nL different separation
distances (for variable Halbach position measurements) and nT different polarisation times
(for additional pre-polarisation at the electromagnetic coil) then there are npp = nL + nT pre-
polarisation conditions. Thus the effective NMR measured signals will be of size nw×npp
(i.e. a 2D data matrix of npp compressed FIDs each of length nw). The signal matrix (S) is
unwrapped into a 1D signal column vector (s). If the solution matrix (P) for the probability
distribution is to be of size mv×mT1 then the relevant model kernel matrix (K) will be of size
n×m (where n = nwnpp and m = mvmT1). The row elements of the model kernel matrix describe
changes in signal with respect to the experimental parameters (i.e. LPD, tpolz and te) and
correspond to the signal vector components. The column elements of the model kernel matrix
describe changes in the measured signal with respect to the measured parameters (i.e. v and
T1) and correspond to the solution vector components. The measured signals (s) have now
been appropriately manipulated such that the signal vector is of length n = nwnp and the model
kernel (K) has been manipulated to provide a matrix of size n×m in order to solve the solution
vector (p) of length m with 1D Tikhonov regularisation (using equation 2.18). The vector
and matrix are also of appropriate size to ensure reasonable computational efficiency
(Mitchell et al., 2012). Note that the smoothing operation matrix (Q) must be carefully
constructed to calculate the finite second difference across 2D solution space. The resultant
99
solution is returned as a vector (p) is of length m and is reshaped into the final 2D distribution
(P of size mv×mT1). Useful graphical representations of the described matrix manipulation
procedures are presented by Mitchell et al. (2012). The 1D distributions (P(v) and P(T1))
may be determined by projecting the 2D distribution onto the relevant 1D axis (velocity axis
for P(v) and T1 axis for P(T1)).
6.2.7. Component flowrate calculations
The joint 2D velocity-T1 probability distribution can now be used to calculate
individual phase flowrates (of canola oil and water). The oil and water phases are
distinguished with a T1 cut-off value (T1,C = 600 ms, as described in section 6.2.3). The signal
contribution of each phase is calculated by integrating over the relevant region of the joint
2D probability distribution:
𝑆𝑜 = ∫ ∫ 𝑃(𝑣, 𝑇1)𝑣𝑚𝑎𝑥
0𝑑𝑣 𝑑𝑇1
𝑇1,𝐶
𝑇1,𝑚𝑖𝑛 (6.15)
𝑆𝑤 = ∫ ∫ 𝑃(𝑣, 𝑇1)𝑣𝑚𝑎𝑥
0𝑑𝑣 𝑑𝑇1
𝑇1,𝑚𝑎𝑥
𝑇1,𝐶 (6.16)
where So and Sw are the oil and water signal contribution to the model fit respectively, T1,min
and T1,max define the bounds of the discretised T1 range and vmax is the maximum value in the
discretised velocity range. The phase fractions (xw and xo) are calculated from the relevant
signal contribution of each phase according to equations 6.4 and 6.5. The individual velocity
distributions of each phase (P(vo) and P(vw) for oil and water respectively) are determined by
integrating the relevant regions of the 2D distribution:
𝑃(𝑣𝑜) = ∫ 𝑃(𝑣, 𝑇1) 𝑑𝑇1𝑇1,𝐶
𝑇1,𝑚𝑖𝑛 (6.17)
𝑃(𝑣𝑤) = ∫ 𝑃(𝑣, 𝑇1) 𝑑𝑇1𝑇1,𝑚𝑎𝑥
𝑇1,𝐶 (6.18)
The mean velocity for each phase (vM,o and vM,w for oil and water, respectively) is then be
determined by calculating the expected value for each phase velocity distribution:
𝑣𝑀,𝑖 = ∫ 𝑣𝑖𝑃(𝑣𝑖)𝑣𝑚𝑎𝑥
0𝑑𝑣𝑖 (i = oil or water) (6.19)
Finally, individual volumetric flowrates for each phase (qo and qw for oil and water
respectively) are calculated:
𝑞𝑖 = 𝑥𝑖𝑣𝑀,𝑖𝐴 (i = oil or water) (6.20)
100
where A is the internal cross-sectional area of the pipe. The phase flowrates which are
measured from the NMR signal analysis methodology are verified against independent
flowrate measurements from in-line rotameters of single-phase flow which are obtained prior
to mixing in the flow loop.
6.3 Experimental procedure
6.3.1. Choice of experimental parameters
Experimental NMR measurements require the selection of appropriate sequence
parameters for the independent variables relevant to signal polarisation (i.e. pre-polarisation
time and separation distance). The EPPC uses variable polarisation time in order to observe
signal contrast in flowing fluids; twenty FID measurements are conducted with
logarithmically spaced pre-polarising times ranging from (10 to 600) ms (at a fixed separation
distance of 70 cm). Polarisation times longer than 600 ms are not useful with current
hardware as the effective polarisation time is limited by the residence time of fluid within the
coil. A further twenty FID measurements are conducted at variable separation distances
logarithmically spaced between (45 and 150) cm (with zero pre-polarisation). Therefore, 40
FID measurements are conducted for each flowrate (20 with variable pre-polarising time and
20 with variable separation distance). Each FID is signal averaged with 32 scans (Nscans).
The FID acquisition time (ta) is varied from 0.205 – 0.819 s and the repetition time is varied
from 0.6 – 1.8 s depending on the flow velocity and pre-polarising time. The relevant
sequence parameters for the 40 FIDs are summarised in Figure 6.7.
101
Figure 6.7 Summarisation of the polarising conditions for the 40 FID measurements. For the first
20 measurements, both the Halbach and the EPPC are used for polarisation, with the Halbach position
held constant (70 cm separation distance) whilst the polarisation time is logarithmically varied from
10 – 600 ms. For the second 20 FID measurements, only the Halbach is used for polarisation with
separation distances logarithmically ranging from 45 – 150 cm.
The overall measurement procedure, which has not been ‘total acquisition time’
optimised, takes ~21 minutes to capture the required FID data. The above set of FID
conditions will be referred to as the “NMR flow measurement sequence” from here onwards.
6.3.2. Experimental measurement workflow
Figure 6.8 provides a flow chart outlining the key steps in the procedure applied to
measure the fluid flowrates using the NMR flow measurement sequence. The NMR flow
measurement sequence is applied with the EFNMR system using the associated Prospa
software (Magritek, New Zealand), capturing 20 FID measurements at varying separation
distance and 20 FID measurements at varying pre-polarisation time (as discussed in section
6.3.1). Once the data has been recorded, it is imported into Matlab R2017b for signal
processing and analysis. A series of pre-processing steps are applied to the imported FID
data such that they are suitable for inversion. A Gaussian noise filter is implemented on the
measured FID spectrums primarily to remove the influence of 50 Hz mains noise. Each FID
signal is then truncated at the point where the SNR reaches 1 (i.e. data with SNR < 1 is
removed); such that the baseline noise does not influence the signal and introduce artefacts
in the resulting 2D velocity-T1 distribution. The truncated signals are then window averaged
0
20
40
60
80
100
120
140
160
5 10 15 20 25 30 35 40
0
100
200
300
400
500
600
700
800
Se
pa
rati
on
dis
tan
ce [
cm
]
FID Number
Pre
-po
lari
sati
on
tim
e [
ms
]
Variable separationdistance
Constant separation distance (70 cm)
Variable polarisation time
102
such that the data is of appropriate size for computationally efficient inversion (Istratov and
Vyvenko, 1999).
The processed FID data is then fit with the appropriate 2D velocity-T1 distribution
using the NMR flow model (equation 6.13) via Tikhonov regularisation inversion (equation
2.18). The 2D regions of the velocity-T1 distribution are appropriately integrated in order to
determine the relevant signal contribution of each phase (according to equations 6.15 and
6.16). The signal contributions are corrected for the oil hydrogen index in order to quantify
volumetric fluid fractions (according to equations 6.4 and 6.5). The expected value of each
of the individual phase velocity distributions are calculated to determine the mean velocity
(according to equation 6.19). Finally, the individual volumetric flowrates for water and oil
are calculated from the volumetric phase fractions and phase mean velocities (using equation
6.20).
103
Figure 6.8 Flow chart outlining the NMR flow measurement procedure.
Flowing measurements
Measure 20 FIDs with varying tpolz
Measure 20 FIDs with varying LPD
Record data
Import data into Matlab
Signal pre-processing
Experimental data fit with model using 2D Tikhonov inversion
2D velocity-T1
distribution measured
Integrate relevant regions to determine oil and water NMR signal
contribution
Correct for oil hydrogen index
Calculate volumetric flowrate of each phase
Validate against inline rotameters
Calculate mean velocity of each
phase
1. Filtering
2. Truncation
3. Window averaging
0
5
10
15
20
0 0.05 0.1 0.15 0.2
NM
R S
ign
al
[µV
]
te [s]
Direction of increasing LPD
0
5
10
15
20
0 0.05 0.1 0.15 0.2
NM
R S
ign
al
[µV
]
te [s]
Direction of increasing tpolz
0
5
10
15
20
0 200 400 600
NM
R S
ign
al
[µV
]
tpolz [ms]
Expt. data
Model fit
0
5
10
15
20
40 80 120 160
NM
R S
ign
al
[µV
]
LPD [cm]
Expt. data
Model fit
104
6.4 Results and discussion
6.4.1. Single-phase measurements – Water
The flow measurement procedure is first demonstrated for single-phase water flows.
An example result is shown for water flowing at 6.0 m3/h (corresponding to a velocity of
1.11 m/s). The pipe flow Reynolds number (Re = ρvMD/μ) is calculated as 54,600 for this
flowrate, indicating turbulent flow. The flow loop system (illustrated in Figure 6.1(a)) is
configured for single-phase water use (via appropriate valves) and then water is flowed
through the system with the rotameter set at 6.0 m3/h. The NMR flow measurement sequence
is applied to capture the 40 FID measurements (as outlined in section 6.3.1). Figure 6.9(a)
shows the 20 FID measurements with variable separation distance whilst Figure 6.9(b) shows
the 20 FID measurements with variable pre-polarisation time. For each FID measurement,
the effective polarisation is determined by back-extrapolating the measured FID signal to
zero time since excitation (i.e. te = 0). Figure 6.9(c) shows the effective polarisation as a
function of separation distance whilst Figure 6.9(d) shows the effective polarisation as a
function of pre-polarisation time. For each of the measured signal plots (i.e. Figure 6.9(a-d))
the relevant model fit determined by fitting equation 6.13 to the experimental data using
Tikhonov regularisation is also shown. Figure 6.9(e) displays the resultant 2D velocity-T1
probability distribution obtained from the inversion procedure as well as the projected 1D
distributions (P(v) and P(T1)).
105
0
10
20
30
40
0 0.05 0.1 0.15 0.2
NM
R S
ign
al
[µV
]
Time since excitation [s]
LPD
[cm]45485154586266707580859096
103109116124132141150
Expt. data
Model fit
Direction of increasing LPD
(a)
0
10
20
30
40
0 0.05 0.1 0.15 0.2
NM
R S
ign
al
[µV
]
Time since excitation [s]
tpolz
[ms]1012151924293645567086
107133165204253314390484600
Expt. data
Model fit
Direction of increasing tpolz
(b)
0
10
20
30
40
50
60
70
80
40 80 120 160
NM
R S
ign
al
[µV
]
Separation distance [cm]
Experimental data
Model fit
(c)
0
10
20
30
40
50
60
70
80
0 200 400 600
NM
R S
ign
al
[µV
]
Pre-polarising time [ms]
Experimental data
Model fit
(d)
106
Figure 6.9 An example 2D velocity-T1 measurement for single-phase water. The water is flowing
at 6.0 m3/h (corresponding to a water velocity of 1.11 m/s). Measured experimental signals; (a) FID
measurements at variable polarisation-detection separation distance, (b) FID measurements at
variable electromagnetic coil pre-polarisation time, (c) effective polarisation as a function of
polarisation-detection separation distance and (d) effective polarisation as a function of pre-
polarisation time. For each of the measured signal plots (a-d) the model fit is also displayed. (e) The
resultant 2D velocity-T1 probability distribution as well as the 1D distributions (P(v) and P(T1)). The
dashed line indicates the oil-water T1 cut-off (600 ms).
Figure 6.9(a) and (c) illustrate the signal decay as a function of the polarisation-
detection separation distance; the effective polarisation exhibits an exponential decay as a
function of LPD according to exp(-LPD/(vT1)) (as in equation 6.6). Figure 6.9(b) and (d)
illustrate measured signal decay as a function of the EPPC pre-polarisation time; the effective
polarisation increases exponentially according to (1- exp(-τPC/vT1)). The limiting effect of
the fluid residence time in the coil (τPC = LPC/v – tPE = 0.27/1.11 – 0.06 = 183 ms) is
observable during pre-polarising time experiments; at tpolz > 183 ms the fluid residence time
is less than the pre-polarising time, thus the effective polarisation remains constant. Note
that the initial effective polarisation (e.g. tpolz = 10 ms) in the pre-polarising experiments (in
Figure 6.9(d)) is offset to 26.6 μV corresponding to the initial polarisation due to the Halbach
(at a separation distance of LPD = 70 cm). This will not exactly match the effective
polarisation at LPD = 70 cm in Figure 6.9(c) due to non-adiabatic signal attenuation during
application of the polarisation pulse (see Appendix A.3 for further detail). The 2D velocity-
Oil-water cut-off
(e)
107
T1 distribution displays a single 2D peak exhibiting measured properties anticipated for the
experiment; the mean velocity from the 2D distribution is 1.09 m/s whilst the log-mean T1 is
2.68 s. This demonstrates excellent agreement between the reference flowrate measurement
(6.0 m3/h) and the NMR water flowrate measurement (5.95 m3/h).
To further demonstrate the accuracy of the NMR flow measurement sequence, the
measurement procedure is conducted for ten different water flowrates of 1.0 – 10.0 m3/h
(corresponding to mean velocities of 0.18 – 1.83 m/s). The pipe flow Reynolds number range
for these flows is determined to be 9030 – 90300, therefore all ten flowrates are considered
to be turbulent. Figure 6.10(a) shows the effective polarisation as a function of separation
distance for each flowrate (the same as Figure 6.9(c)) whilst Figure 6.10(b) shows the
effective polarisation as a function of pre-polarisation time (the same as Figure 6.9(d)). For
each measurement, the acquired 2D velocity-T1 distribution has been separated into
individual 1D distributions (corresponding to the projected 1D distributions in Figure 6.9(e))
with the velocity distributions (P(v)) shown in Figure 6.10(c) and the T1 distributions (P(T1))
displayed in Figure 6.10(d).
108
Figure 6.10 Measured signal data and corresponding velocity and T1 distributions for single-phase
water measurements. Flow measurements are obtained at flowrates of 1.0 – 10.0 m3/h (corresponding
to mean velocities of 0.18 – 1.83 m/s). (a) Effective polarisation as a function of polarisation-
detection separation distance and (b) Effective polarisation as a function of pre-polarisation time. For
each of the measured signals in (a) and (b) the model fit is also displayed. (c) The 1D velocity
distributions and (d) 1D T1 distributions for each water flowrate obtained using the inversion
procedure.
Figure 6.10(a) demonstrates the effective polarisation as a function of polarisation-
detection separation distance; the exponential rate of decay is decreasing as velocity increases
as anticipated according to the relationship; exp(-LPD/(vT1)) (as in equation 6.6). Figure
6.10(b) shows the effective polarisation as a function of the pre-polarising time. The
effective polarisation becomes constant at earlier pre-polarising times as the velocity
increases (e.g. for 3.0 m3/h effective polarisation is constant at tpolz > 0.44 s whilst for 10.0
m3/h effective polarisation is constant at tpolz > 0.09 s). This is a consequence of the fluid
residence time in the coil reducing as velocity increases. In general, the larger fluid residence
0
10
20
30
40
50
60
70
80
40 80 120 160
NM
R S
ign
al
[µV
]
Separation distance [cm]
(a)
0
10
20
30
40
50
60
70
80
0 200 400 600
NM
R S
ign
al
[µV
]Pre-polarising time [ms]
Flowrate[m3/h]
12345678910
Expt. data
Model fit
(b)
0
2
4
6
0 1 2 3
P(v
)
Velocity [m/s]
12345678910
Flowrate [m3/h]
(c)
0
1
2
3
0.01 0.1 1 10
P(T
1)
T1 [s]
12345678910
Flowrate [m3/h]
(d)
109
time observed for lower velocities results in higher effective polarisation (relative to flows at
higher velocities). This is not always the case (e.g. a flowrates of 1.0 and 2.0 m3/h) due to
complex flush-out effects observed due to the non-adiabatic signal diminishment (see
Appendix A.3 for further details).
The velocity distributions shown in Figure 6.10(c) each display a single peak. The
distributions exhibit a turbulent velocity distribution shape very similar to those measured
using the 1D FID analysis measurement discussed in section 4.3.2. The accuracy of the
means of the velocity distributions is validated via the flowrate comparison in Figure 6.11.
The T1 distributions in Figure 6.10(d) each display a mono-exponential peak at a similar
position. The average log-mean T1 across the ten distributions is 2.55 s. The T1 distributions
tend to broaden as the velocity increases. The rate of exponential decay with respect to the
separation distance decreases (as shown in Figure 6.10(a)) as the velocity increases which
results in the relative fraction of the observable effective polarisation decay being reduced.
The sensitivity of Tikhonov regularisation inversion to accurately fit the appropriate
dependent parameter distributions (i.e. v and T1) is influenced by the range and number of
each independent parameter (i.e. te, LPD and tpolz) (Istratov and Vyvenko, 1999, Mitchell et
al., 2014). For example; at a water flowrate of 1.0 m3/h (mean velocity of 0.18 m/s), the
effective polarisation decays from 61.6 μV at LPD = 45 cm to 7.9 μV at LPD = 150 cm
(corresponding to an 87 % signal decay) thus the inversion is sensitive to longer T1
components and a narrow T1 distribution is observed. However, at a water flowrate of 10.0
m3/h (mean velocity of 0.183 m/s), the effective polarisation decays from 26.9 μV at LPD =
45 cm to 19.6 μV at LPD = 150 cm (corresponding to a 27 % signal decay). This results in
the inversion being relatively less sensitive to the longer T1 components (compared to flows
at lower velocity) and a broad T1 distribution is observed. In order to be more sensitive to
longer T1 components at high velocities, the separation distance would need to be
substantially increased; however this results in an impractically large flowmeter. In any case,
the necessity of accurately measuring T1 is not strictly important if the relevant components
(e.g. oil and water) are distinguishable for two-phase flow measurements.
110
The accuracy of the NMR flow measurement procedure for single-phase water flows
can be validated against the rotameter flowrate measurements. The NMR measurement of
water flowrate is extracted from the 2D velocity-T1 distributions captured in Figure 6.10
using the component flowrate calculation procedure outlined in section 6.2.7. Figure 6.11(a)
shows a comparison of the NMR measured water flowrate (qw,nmr) to the rotameter measured
water flowrate (qw,rot). Figure 6.11(b) shows the deviations of the NMR measured flowrate
(i.e. qw,nmr - qw,rot).
Figure 6.11 Comparison of the NMR and rotameter measurements of water flowrate. (a) Direct
comparison of flowrates, and (b) Deviations of the NMR measured flowrate.
Figure 6.11 shows excellent agreement between the NMR measured flowrate and the
rotameter measured flowrate. The mean error (calculated using equation 3.6) is 0.011 m3/h
and the root mean square error (calculated using equation 3.7) is calculated as 0.068 m3/h for
the ten water flowrate measurements.
6.4.2. Single-phase measurements – Oil
Flowing measurements are now shown for single-phase canola oil flows. An example
result is demonstrated for canola oil flowing at 6.0 m3/h (corresponding to a velocity of 1.11
m/s). The pipe flow Reynolds number is determined as 820; therefore the flow is laminar.
The flow loop system is configured for single-phase oil use and then oil is flowed through
the system with the rotameter set at 6.0 m3/h. The NMR flow measurement sequence is
captured with the 40 FID measurements (as outlined in section 6.3.1). Figure 6.12(a) shows
the 20 FID measurements with variable separation distance whilst Figure 6.12(b) shows the
0
2
4
6
8
10
0 2 4 6 8 10
NM
R m
easu
red
flo
wra
te [
m3/h
]
Rotameter measured flowrate [m3/h]
(a)
-0.2
-0.1
0.0
0.1
0.2
0 2 4 6 8 10
NM
R f
low
rate
measu
rem
en
t d
ev
iati
on
[m
3/h
]
Rotameter measured flowrate [m3/h]
(b)
111
20 FID measurements with variable pre-polarisation time. For each FID measurement, the
effective polarisation is determined by back-extrapolating the measured FID signal to zero
time since excitation (i.e. te = 0). Figure 6.12(c) shows the effective polarisation as a function
of separation distance whilst Figure 6.12(d) shows the effective polarisation as a function of
pre-polarisation time. For each of the measured signal plots (i.e. Figure 6.12(a-d)) the
relevant model fit determined by fitting equation 6.13 to the experimental data using
Tikhonov regularisation is also shown. Figure 6.12(e) displays the resultant 2D velocity-T1
probability distribution obtained from the inversion procedure as well as the 1D distributions
(P(v) and P(T1)).
0
2
4
6
8
10
0 0.02 0.04 0.06 0.08 0.1
NM
R S
ign
al
[µV
]
Time since excitation [s]
LPD
[cm]45485154586266707580859096
Expt. data
Model fit
Direction of increasing LPD
(a)
0
2
4
6
8
10
0 0.02 0.04 0.06 0.08 0.1
NM
R S
ign
al
[µV
]
Time since excitation [s]
tpolz
[ms]1012151924293645567086107133165204253314390484600
Expt. data
Model fit
Direction of increasing tpolz
(b)
112
Figure 6.12 An example 2D velocity-T1 measurement for single-phase canola oil. The oil is flowing
at 6.0 m3/h (corresponding to an oil velocity of 1.11 m/s). Measured experimental signals; (a) FID
measurements at variable polarisation-detection separation distance, (b) FID measurements at
variable electromagnetic coil pre-polarisation time, (c) effective polarisation as a function of
polarisation-detection separation distance and (d) effective polarisation as a function of pre-
polarisation time. For each of the measured signal plots (a-d) the model is also displayed. (e) The
resultant 2D velocity-T1 probability distribution obtained from the inversion procedure as well as the
1D distributions (P(v) and P(T1)). The dashed line indicates the oil-water T1 cut-off (600 ms).
Note that not all variable separation distance trials are incorporated in inversion
analysis; FID signals with an initial signal below the minimum allowable SNR (SNR cut-off
of 1) are removed to essentially avoid fitting to noise. This can be observed in Figure 6.12(a)
0
2
4
6
8
10
12
14
16
40 80 120 160
NM
R S
ign
al [µ
V]
Separation distance [cm]
Experimental dataModel fit
(c)
0
2
4
6
8
10
12
14
16
0 200 400 600
NM
R S
ign
al [µ
V]
Pre-polarising time [ms]
Experimental data
Model fit
(d)
(e)
Oil-water cut-off
113
and (c); only FIDs at a separation distances less than or equal to 96 cm are used in the
inversion. The FID measurements at variable separation distance in Figure 6.12(a) show that
the measurable signals are substantially lower relative to those from the water measurements
(shown in Figure 6.9(a)). For example; the maximum signal for the oil FIDs at varying
separation distance in Figure 6.12(a) is 3.5 μV, which is much less than the maximum signal
of 30.7 μV in Figure 6.9(a). This can be attributed to the considerable intermediate signal
decay when the fluid flows from the polarising Halbach array to the r.f. detection coil (as
discussed in section 6.2.4). This justifies the use of the EPPC for low T1 fluids; Figure 6.12(b)
demonstrates an improvement in the measurable signal (up to 7.0 µV for tpolz = 600 ms).
The rate of decay in the measured FIDs is faster for oil measurements relative to
water. The oil signal fully decays to the SNR cut-off after ~0.09 s at maximum polarisation
(for the FID at tpolz = 600 ms in Figure 6.12(b)). For the same FID measurement for water
with a velocity of 1.11 m/s (FID with tpolz = 600 ms in Figure 6.9(b)), the signal has decayed
to the SNR cut-off at ~0.16s. The difference in FID attenuation rates is a result of the
differing T2 decay constants for canola oil and water. The log-mean T2 for canola oil is 82
ms relative to measured water T2 of 1.96 s (see Appendix A.3). This illustrates how T2 decay
is much more influential on the FID for oil measurements relative to water. This will restrict
the ability of the inversion to accurately resolve the velocity distribution in flowing oil
measurements.
The measured 2D velocity-T1 distribution in Figure 6.12(e) displays a broad 2D peak
exhibiting measured properties anticipated for the experiment; the mean velocity from the
2D distribution is 1.11 m/s whilst the log-mean T1 is 0.88 s. The measured velocity
distribution is much broader than the corresponding distribution for water at the same
velocity (shown in Figure 6.9(e)). This can be attributed to the difference in the single-phase
flow regime. The water (Re = 54600 at 1.11 m/s) exhibits turbulent flow which will have a
flat radial velocity profile and therefore the velocity probability distribution will be
reasonably narrow close to the mean velocity. The much more viscous canola oil (Re = 820
at 1.11 m/s) exhibits laminar flow which is anticipated to have a Pouiselle radial velocity
profile. The corresponding velocity probability distribution for a Pouiselle radial profile is a
114
uniform probability density function with velocities ranging from 0 to 2vM. The theoretical
velocity distribution thus contains sharp (non-smooth) changes at the distribution boundaries
(i.e. 0 and 2vM). The Tikhonov penalty function (||Qp||2 in Equation 2.18) imposes
smoothness on the resulting distribution using the finite second derivative operator (Q).
Therefore the uniform velocity probability distribution for canola oil is very difficult to
capture experimentally as the non-smooth distribution will be penalised during inversion data
analysis. The laminar profile is reflected by the much broader velocity distribution (relative
to those of water) ranging from 0 to 2.5 m/s (~2vM). The measured T1 distribution does not
capture the bi-exponential behaviour of canola oil (as seen for the stationary oil measurement
in section 6.2.3). The two peaks are much more difficult to resolve for the flowing
experiments when considering the much lower SNR (~74 for the stationary measurement
compared to ~16 for flowing measurement at 6.0 m3/h) and the complex dual polarisation
model used to measure the T1 of flowing fluid.
The measurement procedure is conducted for ten different oil flowrates of 1.0 – 10.0
m3/h (corresponding to mean velocities of 0.18 – 1.83 m/s). Figure 6.13(a) shows the
effective polarisation as a function of separation distance for each flowrate, whilst Figure
6.13(b) shows the effective polarisation as a function of pre-polarisation time. For each
flowrate, the measured 2D velocity-T1 distribution has been separated into individual 1D
distributions with the velocity distributions (P(v)) shown in Figure 6.13(c) and the T1
distributions (P(T1)) displayed in Figure 6.13(d). Note that for flowrates of 1.0 and 2.0 m3/h,
none of the FID measurements at variable separation distance had an initial signal above the
SNR cut-off, therefore they are not used within the inversion and are not displayed in Figure
6.13(a).
115
Figure 6.13 Measured signal data and corresponding velocity and T1 distributions for single-phase
oil measurement. Flow measurements are obtained at flowrates of 1.0 – 10.0 m3/h (corresponding to
mean velocities of 0.18 – 1.83 m/s). (a) Effective polarisation as a function of polarisation-detection
separation distance and (b) Effective polarisation as a function of pre-polarisation time. For each of
the measured signals in (a) and (b) the model fit is also displayed. (c) The 1D velocity distributions
and (d) 1D T1 distributions for each oil flowrate obtained using the inversion procedure.
In Figure 6.13(a), the rate of exponential decay is much faster relative to the water
results in Figure 6.10(a) due to the much shorter T1 of canola oil relative to water. Only the
two highest flowrates (9.0 and 10.0 m3/h) have a measurable signal at the longer separation
distances (final measured point is 141 cm for measurements at 9.0 and 10.0 m3/h). In Figure
6.13(b); the influence of the FID measurements at varying pre-polarisation times is more
prominent for the oil measurements relative to water measurements. The FID measurements
observe a rapid polarisation according to the short T1 of oil (measured as ~81 ms under flow).
The velocity distributions in Figure 6.13(c) are relatively broad and typically range from ~ 0
0
2
4
6
8
10
12
14
16
40 80 120 160
NM
R S
ign
al
[µV
]
Separation distance [cm]
(a)
0
2
4
6
8
10
12
14
16
0 200 400 600
NM
R S
ign
al
[µV
]
Pre-polarising time [ms]
Flowrate[m3/h]
12345678910
Expt. data
Model fit
(b)
0
1
2
3
4
5
0 1 2 3 4 5
P(v
)
Velocity [m/s]
12345678910
Flowrate [m3/h]
(c)
0
1
2
3
4
5
0.01 0.1 1 10
P(T
1)
T1 [s]
12345678910
Flowrate [m3/h]
(d)
116
to twice the mean velocity indicating a laminar velocity profile (as discussed in relation to
the oil flow measurement example in Figure 6.12). Some of the velocity distributions display
two velocity peaks as a consequence of the data inversion; particularly at higher velocity (e.g.
at 7.0 and 9.0 m3/h). Furthermore, the T1 distributions each display a single peak with the
average log-mean T1 across the ten distributions being 81 ms. This is slightly shorter than
the log-mean T1 measured for stationary flow (115 ms – see section 6.2.3). The suggested
source of the discrepancy is inversion sensitivity and the model complexity. Using a complex
model in combination with the Tikhonov regularisation results in reduced inversion stability
and can lead to inaccurate measurement results (Mitchell et al., 2014). The incorporation of
complex calibration factors such as the T1/T2 calibration ratio (refer to section 6.2.5) and the
non-adiabatic signal loss (refer to Appendix A.3) can influence the position of the T1
distribution. The shorter oil T1 is much more sensitive to such parameters relative to the
longer water T1. The bi-exponential peak observed in the T1 distribution of stationary
measurements (see section 6.2.3) is not resolved for any of the flowing oil measurements;
this is attributed to the lower SNR and complex dual polarisation model as discussed in
relation to the oil flow measurement example at 6.0 m3/h in Figure 6.12.
The accuracy of the NMR flow measurement procedure for single-phase oil flows
can be validated against the rotameter flowrate measurements. The NMR measurement of
the oil flowrate is extracted from the 2D velocity-T1 distributions captured in Figure 6.13
using the component flowrate calculation procedure outlined in section 6.2.7. Figure 6.14(a)
shows a comparison of the NMR measured oil flowrate (qo,nmr) to the rotameter measured oil
flowrate (qo,rot). Figure 6.11(b) shows the deviations of the NMR measured flowrate (i.e.
qo,nmr – qo,rot).
117
Figure 6.14 Comparison of the NMR and rotameter measurements of oil flowrate. (a) Direct
comparison of the NMR measured oil flowrate to the rotameter measured oil flowrate. (b) Deviations
of the NMR measured flowrate relative to the rotameter measured oil flowrate.
Figure 6.14 shows excellent agreement between the NMR measured oil flowrate and
the rotameter measured oil flowrate. The mean error is -0.013 m3/h and the root mean square
error is calculated as 0.094 m3/h for the ten oil flowrate measurements. The accuracy
demonstrated for both single-phase oil and water flows gives confidence in applying the
measurement procedure to two-phase oil water flows as will be demonstrated in Chapter 7.
6.5 Conclusions
In this chapter the procedure for simultaneous measurement of the individual phase
flowrates in two-phase oil/water flows has been developed. The EFNMR flow meter is
extended to incorporate a dual polarisation scheme utilising the upstream Halbach array as
well as an electromagnetic pre-polarising coil which is integrated with the EFNMR detection
coil. The oil and water fractions are distinguished using the fluid spin-lattice relaxation time
(T1) which is captured using the polarised signal contrast. T1 relaxometry is demonstrated to
be an appropriate distinguishing mechanism; example measurements performed on
stationary oil/water mixtures accurately quantify the component phase fractions. The dual
polarisation system allows improved measurement sensitivity (relative to the initial EFNMR
system) with an effective velocity range of (0.10 to 3.0) m/s and effective T1 range of 50 ms
to 5.0 s. The corresponding NMR flow model has been developed simultaneously with the
apparatus in order to describe the NMR signal of fluid flowing through the upgraded flow
measurement system. An appropriate experimental measurement pulse sequence has been
0
2
4
6
8
10
12
0 2 4 6 8 10 12
NM
R m
ea
su
red
flo
wra
te
[m3/h
]
Rotameter measured flowrate [m3/h]
-0.3
-0.2
-0.1
0.0
0.1
0.2
0.3
0 2 4 6 8 10 12
NM
R f
low
rate
me
as
ure
me
nt
devia
tio
n [
m3/h
]
Rotameter measured flowrate [m3/h]
118
outlined which involves capturing FID signals at varying pre-polarising conditions. The
multi-FID signals are fit with a joint 2D velocity-T1 distribution using the developed NMR
flow model via a 2D Tikhonov regularisation algorithm. The performance of the NMR flow
measurement was examined for both single-phase water and oil flows and demonstrates
excellent agreement with in-line rotameter measurements. Furthermore; the velocity
distributions capture the appropriate single-phase flow classification; with laminar flows
(exhibited by oil) being distinguished from turbulent flow profiles (displayed by water) by
broader, sometimes bimodal, apparent velocity distributions.
119
Chapter 7. Quantitative characterisation of two-phase
oil/water flows
7.1 Introduction
In this chapter, the flow measurement methodology developed in Chapter 6 is applied
to simultaneously quantify phase flowrates for two-phase oil/water flows. Flow
measurements are performed across five different oil/water flow regimes. The 2D velocity-
T1 distributions are analysed in terms of two-phase flow characteristics such as the oil/water
velocity slip. The performance of the measurement system is analysed in each flow regime
via comparison against the individual phase flowrates.
7.2 Experimental methodology
7.2.1. Flow regimes
The measurement and analysis of two-phase oil/water flows requires consideration of
the range of possible liquid-liquid flow regimes. Details regarding the classification and
description of such flow regimes were discussed in section 3.1.3. For the experimental
oil/water flow loop used in this work (outlined in section 6.2.2), the visually observed flow
patterns have been classified into five different flow regime categories which are similar to
the regimes observed by Trallero et al. (1997). The flow regimes include; stratified flow
with mixing at the interface (St w/ mix), a dispersion of oil-in-water over a free water layer
(Do/w & w), a dual dispersion of water droplets in oil over oil droplets in water (Do/w &
Dw/o), a full oil-in-water emulsion (Eo/w) and a full water-in-oil emulsion (Ew/o). The oil
and water flowrates are not measurable (due to the rotameter metering range) at the low
flowrates required to achieve stratified flow (without mixing) thus this flow regime was not
incorporated in the experimental measurements performed. A summary of the five relevant
oil/water flow regimes (originally shown in Table 3.2) is presented here (in Table 7.1) for
reference with respect to the oil/water experiments conducted in this Chapter.
120
Table 7.1 Summary of the five relevant oil-water flow regimes.
Name Abbreviation Description Illustration
Stratified
flow with
mixing
St w/ mix. Fluids exist as segregated
layers with the presence of
small capillary waves at the
interface. Small quantities of
droplets of each fluid are
dispersed into the alternate
phase.
Dispersion
of oil-in-
water and
water
Do/w & w A dispersion layer exists at the
top of the pipe with oil
droplets entrained within a
continuous water phase. The
bottom layer consists of free
water alone.
Dispersion
of water-in-
oil and
dispersion
of oil-in-
water
Dw/o &
Do/w
A dual dispersion with; water
droplets dispersed into a
continuous oil layer (top)
above oil droplets dispersed
into a continuous water layer
(bottom).
Oil-in-water
emulsion
Eo/w A full emulsion of oil droplets
dispersed in a continuous
water layer.
Water-in-oil
emulsion
Ew/o A full emulsion of water
droplets dispersed in a
continuous oil layer.
An experimental matrix of oil and water superficial velocities to be used for
verification of the NMR flow measurement procedure is presented in Figure 7.1. The water
superficial velocities range from 0.18 – 2.03 m/s (corresponding to flowrates of 1.00 – 11.1
m3/h) and the oil superficial velocities range from 0.17 – 1.46 m/s (corresponding to flowrates
of 1.00 – 8.00 m3/h). The specified superficial velocities are chosen such the measurements
will cover a range of two-phase flow regimes. The visually observed flow regime for each
measurement point is indicated by the marker type and colour. A comparison to the flow
regime boundaries observed by Trallero et al. (1997) is provided (indicated with black lines);
121
this data is chosen for comparison due to the relatively similar experimental conditions;
experiments were conducted with tap water and mineral oil with a pipe diameter of 50.1 mm,
density ratio (ρo/ρw) of 0.85 and a viscosity ratio (μo/μw) of 29.6 (Trallero et al., 1997). In
this work, experiments are conducted with saline tap water and canola oil with a pipe
diameter of 44.0 mm, density ratio of 0.86 and a viscosity ratio of 44.0.
Figure 7.1 The experimental matrix of oil and water superficial velocities for two-phase flowing
measurements conducted in this work. The marker type and colour indicate the visually observed
flow regimes, whilst the flow regime boundaries from the work of Trallero et al. (1997) are indicated
by black lines.
The position of visually observed flow regimes (from this work) relative to the
observed flow regimes from Trallero et al. (1997) match relatively well. Exact agreement is
not anticipated as liquid-liquid flow regimes are dependent on a range of experimental
conditions (e.g. fluid densities, viscosities and wall wetting properties) (Angeli and Hewitt,
2000, Arirachakaran et al., 1989, Xu, 2007). The difference in flow regime transitions can
be attributed to the relative difference in viscosity ratios. In this work the high oil viscosity
(~55 cP) results in flow regime transitions at lower velocities (relative to the work of Trallero
et al. (1997)). For example, the transition from stratified flow with mixing (St w/mix) to a
dispersion of water droplets in oil over oil droplets in water (Do/w & Dw/o) occurs at vos ≈
0.10
1.00
0.10 1.00
Su
perf
icia
l w
ate
r ve
loc
ity,
vw
s[m
/s]
Superficial oil velocity, vos [m/s]
StratifiedStratified with mixing ()
Oil in water emulsion ()
Disp. o/w & w ()
Water in oil emulsion ()
Disp. w/o & Disp. o/w ()
122
0.50 m/s in this work compared to the transition at vos = 0.70 m/s observed by Trallero et al.
(1997).
The 34 experimental measurement points (shown in Figure 7.1) are categorised into
the five appropriate flow regimes according to visual observation. NMR measurements are
taken at each point according to the measurement procedure outlined in section 6.3. The
measured phase flowrates for oil and water are compared to in-line rotameter measurements
of the single-phase flowrates prior to mixing in the flow loop.
7.2.2. Fluid separation
The validation of experimental measurements requires reliable operation of the two-
phase flow loop; including adequate phase separation at the gravimetric separating tank.
Emulsion stability bottle tests were conducted in order to understand the separation behaviour
of the canola oil and water emulsions and are presented in Appendix C.1. The emulsions
(oil-in water as well as water-in-oil) were observed to be stable over a considerable timeframe
(~1 day). The formation of stable oil/water dispersions under flowing conditions prevents
adequate fluid separation at the gravimetric separator during measurements. This causes the
storage tanks to be contaminated with the alternate phase and thus the concentration of the
“pure fluids” discharged from the storage tanks changes throughout the duration of a
measurement. In general, the contamination is acceptable for the water storage tank as the
free water phase in the tank is denser than the emulsion and therefore free water will be
discharged from the bottom of the tank through the water side of the system (including the
water rotameter). However the presence of an emulsion is highly problematic for the oil
phase; the emulsion is denser than the free oil phase thus the emulsion sinks to the bottom of
the oil storage tank. The outlet at the bottom of the tank will discharge the emulsion (and
any water which has separated from the emulsion) when it should be discharging pure oil.
This means that water is contaminating the oil side of the system which is consequential for
both the rotameter and NMR flow measurement principles. The rotameter measurement is
dependent on the properties (e.g. density and viscosity) of the fluid being measured; thus it
is not suitable for measuring a fluid of changing composition. The NMR measurement
technique assumes constant fluid composition throughout the duration of the experiment.
123
Changing the fluid composition within the NMR measurement sequence will result in
unexpected changes (considering the assumed constant composition throughout a sequence)
in the NMR signals and the interpretation will yield incorrect flow measurement results.
The emulsion stability bottle tests (see results in Appendix C.1) observed significant
reduction in stability of oil-in-water emulsions in the presence of sodium chloride. Therefore
salt was introduced to the system (12 wt %) in order to enhance separation during trials with
oil-in-water emulsions. Measurements conducted in flow regimes with an oil-in-water
emulsion present (i.e. in the dispersion of oil-in-water & water flow regime as well as oil-in-
water emulsions) are anticipated to achieve sufficient phase separation. The NMR flowrate
measurements for such flow regimes may be confidently validated against the rotameter
measured flowrates.
Increasing salinity enhances the stability of water-in-oil emulsions (Ling et al., 2018).
Poor separation will be observed in flow regimes where water-in-oil emulsions are present
(i.e. the dispersion of water-in-oil and dispersion of oil-in-water flow regime, as well as the
water-in-oil emulsion flow regime). The fluid storage tanks will be contaminated leading to
changing fluid composition over the duration of an experiment (as discussed above). The
oil-side pump is designed to transport the canola oil; it provides a greater pressure head in
order to shift the more viscous canola oil. Therefore when water is inlet into the oil-pump,
the overall flowrate of the oil-side (and hence through the two-phase measurement section)
increases. As a result, the measurements cannot be reasonably validated when water-in-oil
emulsions are present. Furthermore, the emulsion contaminating the oil storage tank leads
to increasing water content as the experiment progresses.
For oil-in-water emulsions; reasonable separation can be achieved (on the order of
minutes), however oil-in-water emulsions are observed at high flowrates. The fluid
separation efficiency is dependent on the residence time of the mixture within the separator.
The separator residence time is inversely proportional to the total multiphase flowrate;
therefore the degree of fluid separation will be reduced as the flowrate increases. Partial
separation results in the composition in the multiphase pipeline measurement section
124
changing throughout the duration of a measurement. For example, for the measurement
presented in section 7.3.3, the overall fluid flowrate is 8.96 m3/h which means for a separation
volume of 128 L the separation residence time will be ~51s. Partial separation will result in
a slight increase in water concentration and thus small flowrate measurement errors are
anticipated.
7.3 Results and discussion
In the following sections, example NMR measurement results are presented for each
flow regime. The overall performance of the technique across all measurements is then
discussed in section 7.3.6.
7.3.1. Stratified flow with mixing
Ten measurements were visually observed to be in the stratified flow with mixing
flow regime (St w/ mix); consisting of two segregated layers of fluid in the pipe cross section
with water on the bottom layer and oil on the top layer. Some mixing (droplets from each
layer dispersing into the alternate layer) is observed at the interface of the two layers. The
degree of mixing (quantity and size of bubbles) increases as the overall fluid flowrate
increases. Figure 7.2 illustrates an example 2D velocity-T1 probability distribution for a
water flowrate of 1.00 m3/h and an oil flowrate of 0.95 m3/h. A table providing information
of key resultant properties is also included.
125
Results Oil Water
Measured fraction [%] 69.8 30.2
Log-mean T1 [s] 0.094 1.85
Mean velocity [m/s] 0.25 0.60
Rotameter flowrate [m3/h] 0.95 1.00
NMR flowrate [m3/h] 0.96 0.99
Figure 7.2 An example 2D velocity-T1 distribution for a stratified with mixing flow regime. The
water flowrate is 1.00 m3/h and the oil flowrate is 0.95 m3/h. The vertical dashed pink line indicates
the oil/water cut-off (T1,C = 600 ms) used to distinguish the two components. The 1D velocity
distribution is separated into the individual components. The table to the right indicates key resulting
parameters for both the oil and water phases.
The 2D velocity-T1 distribution illustrated in Figure 7.2 clearly shows two separate
regions with distinguishable T1 relaxation. The 2D distribution is reflective of the stratified
with mixing flow regime; the oil region (T1,LM = 0.094 s, vM = 0.25 m/s) displays a lower
velocity relative to the water region (T1,LM = 1.85 s, vM = 0.60 m/s). The measured velocity
difference is capturing the velocity slip that exists between the two liquid phases in the
stratified flow regime. The canola oil has a much higher viscosity (55 cP) relative to the
saline water (~1.25 cP). The higher viscosity results in a higher fluid-wall frictional force
for the canola oil leading to a lower velocity relative to the water phase. Note that for the
similar flowrates of 1.00 m3/h and 0.95 m3/h for water and oil respectively, the higher water
velocity will correspond to a higher oil fraction. The accuracy of the measurement for this
example is excellent; the NMR measured oil flowrate (0.96 m3/h) deviates from the rotameter
measured oil flowrate (0.95 m3/h) by only 0.01 m3/h, whilst the NMR measured water
flowrate (0.99 m3/h) deviates from the rotameter measured water flowrate (1.00 m3/h) by -
0.01 m3/h.
Oil
Water
126
7.3.2. Dispersion of oil-in-water & free water layer
Three measurements were visually observed as a dispersion of oil droplets in water
layer above a continuous water layer on the bottom of the pipe. Figure 7.3 illustrates an
example 2D velocity-T1 probability distribution in the dispersion of oil-in-water and water
(Do/w & w) flow regime for a water flowrate of 5.59 m3/h and an oil flowrate of 3.06 m3/h.
Results Oil Water
Measured fraction [%] 31.0 69.0
Log-mean T1 [s] 0.135 3.81
Mean velocity [m/s] 1.42 1.50
Rotameter flowrate [m3/h] 3.06 5.59
NMR flowrate [m3/h] 2.40 5.66
Figure 7.3 An example 2D velocity-T1 distribution for the dispersion of oil-in-water and water flow
regime. The water flowrate is 5.59 m3/h and the oil flowrate is 3.06 m3/h.
The 2D velocity-T1 distribution in Figure 7.3 for the dispersion of oil-in-water above
a water layer measurement displays three distinct regions. The region at short T1 corresponds
to the dispersed oil droplets in the top layer (fraction = 31.0 %, T1,LM = 0.135 s, vM = 1.42
m/s). The large dispersing water region at the lower velocity (fraction = 55.3 %, T1,LM = 3.74
s, vM = 1.31 m/s) corresponds to water with oil droplets dispersed within it. The small “free
water” region at high velocity (fraction = 13.7 %, T1,LM = 4.30 s, vM = 2.42 m/s) is water from
the water only layer (at the base of the pipeline). The free water layer is relatively small
(13.7 % of total pipeflow volume) as the oil flowrate is relatively high for a dispersion of oil-
in-water and water flow; meaning that the two-phase flow is beginning to approach the flow
regime boundary and transition towards a dual dispersion flow. The measurement is able to
capture the anticipated velocity slip quite well for this example; the fluids within the
dispersion layer (i.e. the dispersed oil drops and the water phase containing the oil) both have
very similar velocities which is anticipated for a continuous dispersion layer. The water only
phase (on the bottom layer of the pipe) is observed to be at a higher velocity relative to the
Free water
Dispersing water
Dispersed Oil
127
dispersion phase. This is expected as the dispersion will be more viscous relative to the free
water phase. The NMR measured water flowrate (qw,nmr = 5.66 m3/h) matches the rotameter
measured water flowrate (qw,rot = 5.59 m3/h) very well. However the NMR measured oil
flowrate (qo,nmr = 2.40 m3/h) is under-predicted relative to the rotameter measured oil
flowrate. (qo,rot = 3.06 m3/h). This discrepancy can possibly be attributed to the poor SNR
of oil flows at intermediate velocities. The introduction of the EPPC has partially alleviated
the issue of sensitivity to short T1 components, providing reasonable SNR for flows at low
velocities < 0.4 m/s. However there is still difficulty in providing an accurate inversion (due
to a poor SNR < 6) at intermediate velocities (~0.4 – 1.1 m/s) as this is between regions of
effective operation for each polarisation mechanism. The Halbach array does not provide
significant polarisation as the intermediate decay is considerable relative to higher velocities
(> 1.1 m/s), whilst the EPPC does not provide significant polarisation as the fluid residence
time within the EPPC is restricted relative to lower velocities (<0.4 m/s).
7.3.3. Oil-in-water emulsion
Eight of the two-phase oil/water flow measurements were observed to be an oil-in-
water emulsion (Eo/w); consisting of oil droplets fully emulsified into the continuous water
phase. Figure 7.4 illustrates an example 2D velocity-T1 probability distribution for the oil-
in-water flow regime for a water flowrate of 7.43 m3/h and an oil flowrate of 1.53 m3/h.
128
Results Oil Water
Measured fraction [%] 15.5 84.5
Log-mean T1 [s] 0.104 3.45
Mean velocity [m/s] 2.11 1.79
Rotameter flowrate [m3/h] 1.53 7.43
NMR flowrate [m3/h] 1.79 8.27
Figure 7.4 An example 2D velocity-T1 distribution for the oil-in-water flow regime. The water
flowrate is 7.43 m3/h and the oil flowrate is 1.53 m3/h.
The 2D distribution in Figure 7.4 displays a small region at short T1 corresponding to
the emulsified oil (T1,LM = 0.104 s, vM = 2.11 m/s) and a large region at high T1 corresponding
to water (T1,LM = 3.45 s, vM = 1.79 m/s). The velocity slip ratio is determined to be 1.18 from
the measured velocity-T1 distribution; this agrees well with the results of Rodriguez and
Oliemans (2006) who observed oil in water dispersions to have a slip ratio close to or slightly
greater than 1 (Xu et al., 2008). For this measurement, the NMR measured water flowrate
(qw,nmr = 8.27 m3/h) is over-predicted relative to the rotameter measured flowrate (qw,rot =
7.43 m3/h). The NMR measured oil flowrate (qo,nmr = 1.79 m3/h) is reasonably close to the
rotameter measured flowrate (qo,rot = 1.53 m3/h). The suggested source of discrepancy for
the water flowrate measurement is inadequate residence time (~51 s) for separation of the
oil-in-water emulsion (discussed in section 7.2.2).
7.3.4. Dispersion of oil-in-water & Dispersion of water-in-oil
Ten of the two-phase oil/water measurements were observed to be a dual dispersion
with a top layer consisting of water droplets dispersed into oil and a base pipeline layer of oil
droplets dispersed into water. Figure 7.5 illustrates an example 2D velocity-T1 probability
distribution for the dispersion of water-in-oil and dispersion of oil-in-water flow (Dw/o &
Do/w) regime for a water flowrate of 1.00 m3/h and an oil flowrate of 3.06 m3/h.
Emulsified Oil
Emulsifying water
129
Results Oil Water
Measured fraction [%] 66.5 33.5
Log-mean T1 [s] 0.144 2.59
Mean velocity [m/s] 0.93 0.78
Rotameter flowrate [m3/h] 3.06 1.00
NMR flowrate [m3/h] 3.38 1.43
Figure 7.5 An example 2D velocity-T1 distribution for the dispersion of water-in-oil and dispersion
of oil-in-water flow regime. The water flowrate is 3.06 m3/h and the oil flowrate is 1.00 m3/h.
The 2D velocity-T1 probability distribution in Figure 7.5 shows four distinct regions
(a fifth artefact peak is discussed below); two at low velocity (~ 0.5 m/s) and two at higher
velocity (~1.1 m/s). The high velocity region has a lower water-to-oil-ratio (WOR = 0.41)
relative to the low velocity region (WOR = 0.54). The viscosity ratio (μemulsion/μoil) is
approximated using the following equation for a medium emulsion (Sjoblom, 2001);
𝜇𝑒𝑚𝑢𝑙𝑠𝑖𝑜𝑛
𝜇𝑜𝑖𝑙= 𝑒5𝑥𝑤(1 − 3𝑥𝑤 − 4.5𝑥𝑤
2) (7.1)
where xw is the water-cut. For the high-velocity region, the estimated viscosity ratio is 2.2,
whilst for the low-velocity region, the estimated viscosity ratio is 2.9. Furthermore, visual
observation of the dual emulsion found the top layer to occupy a larger fraction of the cross-
sectional area. The lower WOR, lower estimated viscosity and visual observation of larger
observed fraction indicate that the high velocity layer is likely to be oil continuous. Thus it
is suggested that the low velocity regions are the oil-in-water dispersion (i.e. water
continuous layer) and the high velocity components are the water-in-oil dispersion (i.e. oil
continuous layer). The four regions can then be described as; (i) dispersed oil droplets with
low velocity and low T1 (fraction = 23.5 %, T1,LM = 0.436 s, vM = 0.45 m/s), (ii) dispersing
water with low velocity and high T1 (fraction = 12.7%, T1,LM = 5.18 s, vM = 0.61 m/s), (iii)
dispersing oil with high velocity and low T1 (fraction = 45.3%, T1,LM = 0.092 s, vM = 1.14
m/s) and (iv) dispersed water droplets with high velocity and high T1 (fraction = 18.6 %, T1,LM
Dispersing water
Dispersed water
Dispersed Oil
Dispersing Oil
130
= 1.92 s, vM = 1.01 m/s). The suggested distribution of components indicates that the slip
ratio is greater than one (S = voc/vwc = 2.08; where voc is the mean velocity of the oil
continuous phase and vwc is the mean velocity of the water continuous phase). The viscosity
of dual emulsions is considerably complex (Farah et al., 2005); therefore it is difficult to
predict the slip of the two dispersed phases.
The NMR measured water flowrate (qw,nmr = 1.43 m3/h) is over-predicted relative to
the rotameter measured water flowrate (qw,rot = 1.00 m3/h). The NMR measured oil flowrate
(qo,nmr = 3.38 m3/h) is also over-predicted relative to the rotameter measured oil flowrate.
(qo,rot = 3.06 m3/h). The flow measurement errors as well as the relaxation rate (T1,LM) are a
consequence of inadequate separation of the water-in-oil dispersion layer which will lead to
contamination of water in the oil tank (as discussed in section 7.2.2). The complex
multiphase changes in fluid fractions and velocities will cause the measured FID signals to
unexpectedly change. This can help explain the inversion artefact peak (at low and moderate
T1; such artefact peaks are resolved during inversion in order to fit unanticipated changes in
FID signals (e.g. from changing fluid composition). In the measurement presented in Figure
7.5; increasing water content (with a longer T1 relative to oil) causes the degree of
intermediate signal attenuation (between polarisation and detection) to be reduced. This
leads to an increasing FID signal amplitude as the measurement sequence progresses.
Consequently, the measured T1 values under flow vary significantly from the anticipated T1
values from stationary measurements (e.g. the dispersed oil droplets have T1,LM = 0.436 s
relative to the stationary T1,LM = 0.114 s). The inversion (and associated NMR signal model)
has shifted the dispersed oil droplet contribution to higher T1,LM as well as introducing an
artefact contribution (at T1,LM ~ 1.0 s, vM ~ 0.1 m/s) in order to fit to the observed FID signal
measurements. This will lead to inaccurate phase fraction measurements (e.g. the dispersed
oil droplets fit at higher the T1,LM of 0.436 s will have a larger phase fraction than the real
droplets at T1,LM of 0.114 s). The oil-side contamination also leads to increased overall
flowrate (when water is transported by the oil pump) which is observable in the flowrate
measurement errors. The overall flowrate is relatively low (4.06 m3/h) which means the
inadequate separation will have a small impact on the flowrate measurements (relative to the
example presented in 7.3.5).
131
7.3.5. Water-in-oil emulsion
The final three measurements were visually observed to be in the full water-in-oil
emulsion flow regime (Eo/w); consisting of water droplets fully emulsified into the
continuous oil phase. Figure 7.6 illustrates an example 2D velocity-T1 probability
distribution for the oil-in-water flow regime for a water flowrate of 1.00 m3/h and an oil
flowrate of 6.13 m3/h.
Results Oil Water
Measured fraction [%] 44.9 55.1
Log-mean T1 [s] 0.087 4.51
Mean velocity [m/s] 2.06 0.91
Rotameter flowrate [m3/h] 6.13 1.00
NMR flowrate [m3/h] 5.06 2.74
Figure 7.6 An example 2D velocity-T1 distribution for the emulsion of water-in-oil flow regime.
The water flowrate is 6.13 m3/h and the oil flowrate is 1.00 m3/h.
The 2D velocity-T1 probability distribution in Figure 7.6 shows two distinct regions
which are distinguishable by the fluid T1 relaxation distribution. The region at low T1
corresponds to emulsifying oil (T1,LM = 0.087 s, vM = 2.06 m/s) whilst the region at high T1
corresponds to emulsified water droplets (T1,LM = 4.51 s, vM = 0.91 m/s). For the water-in-
oil emulsion, inadequate separation of phases (discussed in section 7.2.2) will cause
significant (given the degree of emulsification) and rapid (considering the overall flowrate
of 7.13 m3/h) changes in phase fractions. In this example, phase contamination is evident
from two major effects (discussed in further detail below); (i) over-predicted oil velocity, and
(ii) over-predicted water fraction. Minor artefact peaks are also present which is to be
expected considering the FID signals will not behave according to the NMR signal model
when the fluid composition is changing.
Emulsifying Oil
Emulsified water
132
The over-predicted oil velocity is a consequence of signal misinterpretation. The
changes to phase fractions and velocities results in adverse changes in the measured FID
signal similar to the results presented in section 7.3.4. In this example, the 2D inversion (and
associated NMR signal model) has fit an artificially high oil velocity in order to account for
the observed changes in measured FID signals. Consequently, the velocity slip ratio is
measured as to be 2.26; which is very high relative to the anticipated slip ratio of 1 for a fully
emulsified stream where all components are travelling in the same continuous phase.
The second impact of over-predicted water fraction (NMR measured fraction is 55.1
%, whilst the anticipated fraction (for an assumed velocity slip of 1) is ~14 %) is a physical
effect of the oil-tank contamination. The emulsified water entrained into the oil continuous
phase will result in water being outlet through the oil side of the system. This causes the
overall water fraction to considerably increase in the duration of the NMR measurement
sequence (~21 minutes). The water pumped through the oil-side of the system will also have
an enhanced flowrate (as it is transported by the oil pump). As a result, the NMR measured
water flowrate (qw,nmr = 2.74 m3/h) is considerably higher than the rotameter measured water
flowrate (qw,rot = 1.00 m3/h) whilst the NMR measured oil flowrate (qo,nmr = 5.06 m3/h) is
underpredicted relative to the rotameter measured oil flowrate (qo,rot = 6.13 m3/h). The
overall flowrate in this example (7.13 m3/h) is higher relative to the example presented in
7.3.4 such that the impact of inadequate separation is greater.
7.3.6. Summary and discussion of results
The overall performance of the NMR measurement technique is now considered with
respect to all 34 two-phase oil/water measurements. Figure 7.7 presents a comparison
between NMR measured flowrates to in-line rotameter measurements for both oil and water
flowrates. Figure 7.7(a) shows results for the stratified with mixing, dispersion of oil-in-
water and water, as well as the oil-in-water emulsion flow regimes. These are flow regimes
where the experiments were successfully performed (i.e. with reasonable fluid separation)
such that the NMR flowrate measurements may be confidently validated by rotameter
flowrate measurements. Figure 7.7(b) presents results for the dispersion of water-in-oil and
dispersion of oil-in-water, as well as the water-in-oil emulsion flow regimes. These are the
133
flow regimes where inadequate oil and water separation was observed such that experiments
cannot be validated as very poor agreement is expected between NMR and rotameter flowrate
measurements (as discussed in section 7.2.2).
Figure 7.7 Direct comparison of the NMR measured flowrates and rotameter measured flowrates
(for both oil and water) for two-phase oil/water flow measurements. (a) Flowrate comparison for the
stratified with mixing (St w/ mix), dispersion of oil-in-water and water (Do/w & w), as well as the
oil-in-water emulsion (Eo/w) flow regimes. (b) Flowrate comparison for the dispersion of water-in-
oil and dispersion of oil-in-water (Dw/o & Do/w), as well as the water-in-oil emulsion (Ew/o) flow
regimes.
Figure 7.7(a) shows the NMR measured flowrates compare very well to the rotameter
measured flowrates for both oil and water for the stratified with mixing, dispersion of oil-in-
0
2
4
6
8
10
12
0 2 4 6 8 10 12
NM
R m
ea
su
red
flo
wra
te [
m3/h
]
Rotameter measured flowrate [m3/h]
Oil Water
St w/ mix
Do/w & w
Eo/w
(a)
0
2
4
6
8
10
12
0 2 4 6 8 10 12
NM
R m
ea
su
red
flo
wra
te [
m3/h
]
Rotameter measured flowrate [m3/h]
Oil WaterDw/o & o/w
Ew/o
(b)
134
water and water, as well as the oil-in-water emulsion flow regimes. Figure 7.7(b) shows very
poor agreement between NMR measured flowrates and rotameter measured flowrates for
both oil and water in the dispersion of water-in-oil and dispersion of oil-in-water flow regime,
as well as the water-in-oil emulsion flow regime. The NMR measured oil flowrates are
underpredicted relative to the rotameter oil flowrates, whilst the NMR measured water
flowrates are overpredicted relative to the rotameter water flowrates. The source of error for
the results presented in Figure 7.7(b) is the inadequate separation of water-in-oil emulsions
(as discussed in section 7.2.2).
A statistical comparison between the flow measurement techniques is performed in
order to quantify the relative performance in each flow regime. Figure 7.8 presents (a) the
mean error and (b) the root mean square error of the NMR flowrate measurements relative to
the rotameter measurement (for both oil and water) across the five flow regimes.
Figure 7.8 Statistical summary of NMR flow measurement performance in the five different flow
regimes. (a) Mean error (ME) of the NMR measured flowrate relative to rotameter measured
flowrates for each flow regime. (b) The root mean square error (RMSE) of the NMR measured
flowrate relative to rotameter measured flowrates in each flow regime.
The quantitative statistics shown in Figure 7.8 further reinforce the results presented
in Figure 7.7. Good agreement is observed for the first three flow regimes (St w/ mix, Do/w
& w, and Eo/w); the mean error is less than 0.10 m3/h across these flow regimes; except for
the oil flowrate measurement in the dispersion of oil-in-water and water flow regime. This
flow regime only had three experiments, and the example presented in Figure 7.3 (with a
-5
-4
-3
-2
-1
0
1
2
3
4
5
St. w/mix
Do/w &w
Eo/w Dw/o &Do/w
Ew/o
NM
R m
ea
su
red
flo
wra
te
ME
[m
3/h
]
Flow regime
WaterOil
(a)
0
1
2
3
4
5
St. w/mix
Do/w &w
Eo/w Dw/o &Do/w
Ew/o
NM
R m
ea
su
red
flo
wra
te
RM
SE
[m
3/h
]
Flow regime
WaterOil
(b)
135
deviation of -0.66 m3/h) demonstrates the source of this discrepancy. For mid-range oil
velocities (~1.4 m/s) there is loss of measurement sensitivity due to the poor SNR measured
in this velocity-T1 range. The velocity is between the effective regimes of the two polarising
mechanisms; the Halbach array achieves improved polarisation at higher velocities, whilst
the EPPC achieves improved polarisation at lower velocities.
Very good agreement is observed in the stratified with mixing flow regime (RMSE <
0.20 m3/h for both oil and water) as the two-phase flow stream is not emulsified, and the low
overall flowrate gives the fluid sufficient residence time for the two phases to separate in the
gravimetric separation tank. The flowrate agreement is marginally poorer for the oil-in-water
emulsion, primarily due to the reduced separation efficiency observed at higher flowrates.
The oil-in-water emulsion flow regime results in an emulsified mixture and requires a shorter
residence time (typically 1 minute, depending on the overall flowrate) for fluid separation.
This results in poorer comparison errors (RMSE of 0.55 m3/h for water and 0.32 m3/h for oil)
in the oil-in-water emulsion flow regime relative to the stratified with mixing flow regime.
The mean errors and RMSEs are significantly worse (absolute errors are greater than
2 m3/h) for both oil and water flowrate measurements in the dispersion of water-in-oil and
dispersion of oil-in-water flow regime, as well as the water-in-oil emulsion flow regime. The
mean errors highlight that the water flowrate is substantially over-predicted whilst the oil
flowrate is under-predicted in these flow regimes. The key source of error is the inability to
separate water-in-oil emulsions in a reasonable timeframe (for full detail, see section 7.2.2).
Water-in-oil emulsions are much more stable (separation occurs after ~ 1 day) relative to oil-
in-water emulsions (which separate within a few minutes – see appendix C.1). Poor fluid
separation of water-in-oil emulsions results in water significantly contaminating the oil tank
during the experimental timeframe and adversely impacting NMR and rotameter flowrate
measurements. Therefore, the system is currently unsuitable for validating measurements
when a water-in-oil emulsion is present. Future work will consider overcoming this issue
(using demulsifying agents as well as a larger flow loop with increased separation volume)
such that NMR measurements within the dispersion of water-in-oil and dispersion of oil-in-
water flow regime, or the full water-in-oil emulsion flow regime may be validated.
136
7.4 Conclusions
The performance of the NMR flow measurement procedure (which was developed in
Chapter 6) is examined for two-phase oil/water flows which were visually observed to be in
five different flow regimes. The NMR flow measurement technique performed very well in
the stratified with mixing, dispersion of oil-in-water and water, and oil-in-water emulsion
flow regimes, with very good agreement in each of these flow regimes. Furthermore, the
NMR flow measurement technique was successfully able to capture two-phase flow
characteristics; for example the oil/water velocity slip in the stratified with mixing flow
regime or the presence of three distinct regions in the dispersion of oil-in-water and water
flow regime. The flow measurement performance was much poorer in the dispersion of
water-in-oil and dispersion of oil-in-water flow regime, as well as the full water-in-oil
emulsion flow regime. The measurement error is a result of inadequate separation of water-
in-oil dispersions; this problem will be rectified in future work.
137
Chapter 8. Conclusions and future work
8.1 Conclusions
This thesis presented the development of an Earth’s field nuclear magnetic resonance
(EFNMR) multiphase flow meter (MPFM). The flow meter consists of an EFNMR r.f. coil
positioned downstream of a mobile pre-polarisation permanent magnet. The system excites
a 1H NMR signal from fluid moving through the flow meter. The free induction decay (FID)
signal is then acquired and appropriate signal interpretation techniques allow determination
of the composition and flowrate of the measured fluid stream.
Initial system development focussed on establishing an appropriate velocity
measurement technique. Velocity probability distributions were determined by fitting the
measured FID signal of a flowing water stream with a NMR fluid model using Tikhonov
regularisation. The measured velocity distributions were determined to be accurate with
validation against in-line rotameter measurements of water flowrate. The measured velocity
distributions capture the appropriate single-phase flow behaviour with good comparison to
theoretical turbulent distributions. The NMR measurement technique also demonstrated
robust behaviour, with further validation of artificially constructed multi-modal (two-pipe)
velocity distributions.
The system was then applied to analyse two-phase gas/liquid flows in the stratified
and slug flow regimes. The velocity measurement procedure allows accurate measurement
of the liquid velocity distribution in air/water flows. The signal interpretation technique was
applied in a ‘single-shot’ analysis to provide real-time tracking of the liquid holdup and
velocity (at 0.55 – 1.25 Hz). The NMR estimate of liquid holdup is shown to be comparable
138
against a simultaneous video track of holdup. Temporal variation in the liquid flowrate can
be used as a metric to distinguish stratified and slug flow.
The flow metering apparatus and associated signal analysis techniques were then
extended towards the analysis of two-phase oil/water flows. A dual polarisation scheme was
detailed which utilised an electromagnetic pre-polarising coil as well as the mobile
permanent magnet. The oil and water fractions are distinguished using the fluid spin-lattice
relaxation time (T1) which is captured using polarised signal contrast. The multi-FID signals
are fit with a joint 2D velocity-T1 distribution using the developed NMR flow model via a
2D Tikhonov regularisation algorithm. The subsequent analysis of the measured 2D
distributions enables measurement of key phase flow properties, including phase velocities,
fractions and flowrate. The NMR measurement procedure was used to perform accurate
flowrate measurements in the stratified flow with mixing, dispersion of oil-in-water and
water, and oil-in-water emulsion flow regimes. Furthermore; the measured 2D velocity-T1
distributions captured multiphase flow characteristics such as velocity slip. The procedure
could not be accurately validated for the dispersion of water-in-oil and dispersion of oil-in-
water flow regime, or the water-in-oil emulsion flow regime; these regimes produced a water-
in-oil emulsion which could not be sufficiently separated during the total measurement time
which led to flowrate quantification errors.
In summary, the EFNMR MPFM has overall demonstrated very good flow
measurement performance for single-phase flows, two-phase gas/liquid flows and two-phase
oil/water flows. The work presented in this thesis provides an excellent platform for further
developing the measurement system towards a fully capable NMR MPFM.
8.2 Future work
Three aspects of potential future work for the EFNMR MPFM project will be
discussed; (i) improvements to the flow loop equipment, (ii) improvements to the NMR
magnetic hardware, and (iii) advancing the flow measurement procedures.
139
8.2.1. Improving the flow loop equipment
The current flow loop has provided a good experimental set-up for validation of the
multiphase flow measurements presented in this thesis. However to further progress the
measurement system an improved experimental apparatus is required. Firstly a
demulsification technique should be considered to provide adequate separation of water-in-
oil emulsions. A chemical demulsifier will provide the simplest solution; however emulsion
stability testing will be required in order to ensure sufficient destabilisation of emulsions
(Zolfaghari et al., 2016). The appropriate demulsifier should provide separation of water-in-
oil emulsions in a reasonable timeframe (several minutes). This will allow NMR
measurements to be validated across all flow regimes, including the dispersion of water-in-
oil and dispersion of oil-in-water flow regime, as well as the water-in-oil emulsion flow
regime. A larger flow loop system will also be designed with greater separation capacity
(e.g. 1000 L compared to the current 128 L). This will allow the fluids a much longer
residence time to achieve reasonable separation.
The second important upgrade for the flow loop is incorporating realistic three-phase
flow. Three-phase measurements should first be demonstrated for oil-water-air mixtures
(including further development of the associated signal processing). However to truly
demonstrate the capability of the system, three-phase measurements with methane as the gas
phase should be performed. Pressurised methane (unlike air) will produce a measurable
NMR signal and can in principle be distinguished from water and oil via T1 relaxometry.
However the hydrogen index (influencing signal contribution) and T1 of methane is
dependent on pressure and temperature (Gerritsma and Trappeniers, 1971). Therefore
appropriate measurements of oil-water-methane mixtures will be required to develop and
validate the appropriate NMR measurement techniques in three-phase flows under a variety
of conditions. This will require a flow loop with high-pressure flow-lines as well as three-
phase separation in order to validate experimental results.
8.2.2. NMR magnetic hardware improvements
The measurements conducted in this thesis demonstrate that the dual polarisation
scheme is not the optimal method for improving the sensitivity to short T1 components (T1 <
140
100 ms). The impact of the Halbach array stray field on the Earth’s field detection coil has
previously been presented in regards to the measured magnetic field deviation as a function
of separation distance (Fridjonsson et al., 2014b). Further analysis of this effect is presented
in Appendix C.2. The stray field interference is significant due to the relative magnitudes of
the two magnetic fields (Halbach field is 0.3 T whilst Earth’s field is ~54 µT). If the detection
field strength is increased; then the relative impact of the Halbach array stray field on the
detecting field is reduced. A detection frequency of ~200 kHz is suggested to allow reduction
of the separation distance to ~5cm (subject to future magnetic field simulations and
experimental validation). Furthermore, utilising a system with a permanent magnet
producing the detection field eliminates the need for the measurement system to be
perpendicular to the Earth’s magnetic field. Figure 8.1 presents a new model SNR at the
assumed detection frequency of 200 kHz for the Halbach polarisation alone (calculated using
equation 6.13) applied across a velocity-T1 parameter space of (0 to 3) m/s and (0.01 to 10)
s with an assumed minimum separation distance of 5 cm. The higher SNR achievable would
allow the number of scans to be reduced (Nscans = 8) in order to achieve a shorter pulse
sequence measurement time of ~ 5 minutes.
Figure 8.1 Model SNR of fluid components in velocity-T1 space using the Halbach array alone at a
separation distance of 5 cm. The pink lines are contours for SNR = 5; highlighting the region where
reasonable interpretation of the measured signals is anticipated.
Figure 8.1 demonstrates that the proposed system is more sensitive to the short T1
components relative to the current dual polarisation system. There is still a region at low
velocity and low T1 with poor SNR. However if the Halbach array was brought to such a
141
short separation distance (of 5 cm), then the spin-lattice relaxation behaviour is considerably
more complicated during the transition from the Halbach array (polarising field) to Earth’s
field (intermediate decay) to the r.f. coil (detection field) such that some of the NMR flow
model assumptions may no longer apply. Magnetic field simulations should be conducted to
provide appropriate guidelines for the future magnetic system design of the NMR multiphase
flow meter.
The modification of the detection coil frequency will introduce a range of additional
benefits. The introduction of the dual polarisation scheme has resulted in an overly complex
NMR flow model. Using a complex model in combination with the Tikhonov regularisation
results in reduced stability of the inversion and can lead to inaccurate measurement results
(Mitchell et al., 2014). Removing the dual polarisation method will simplify the model and
improve inversion stability. A higher detection frequency would allow reduced ring-down
acquisition delay (Andrew and Jurga, 1987, Hoult, 1984) allowing a considerable increase in
the maximum measurable velocity (Zhen et al., 2018). For example; with the current system
(ring-down delay of 9 ms and detector length of 14 cm) the maximum measurable velocity
is ~3.5 m/s. This is achieved with an initial signal loss of 22.5 % (Sloss = vtdelay/LD, i.e. the
flush-out effect for te = tdelay as modelled in Equation 4.4). For a modified system with further
reduction in ring-down delay (e.g. 1 ms) albeit with a reduced detector length (7 cm), the
maximum measurable velocity (at Sloss = 22.5 %) would be ~15.8 m/s. The reduced ring-
down delay would also allow CPMG pulse sequences to be captured with considerably
reduced echo spacing. This provides a considerable advantage relative to the current FID
measurements as CPMG experiments are less sensitive to the magnetic field inhomogeneity
which changes as the Halbach array is shifted.
8.2.3. Advancing NMR flow measurement techniques
One of the key advantages of NMR over other multiphase measurement techniques
is the ability to non-invasively detect fluid behaviour and dynamics. The improvements to
the magnetic system detailed in section 8.2.3 could also incorporate additional magnetic
hardware improvements such that more advanced NMR measurements can be performed on
multiphase pipelines. For example; a major problem in oil and gas pipelines is the formation
142
and treatment of oil and water emulsions (Stewart and Arnold, 2008). The appropriate
treatment of oilfield emulsions requires consideration of emulsion characteristics, such as
droplet sizes. The application of NMR towards oil/water emulsion characterisation has been
well demonstrated (Fridjonsson et al., 2014a, Johns, 2009, Ling et al., 2016) including the
application of the EFNMR system towards droplet sizing (Fridjonsson et al., 2012). The
potential combination of droplet sizing techniques (e.g. Ling et al., 2017) with flow
measurement techniques presented in this thesis would provide a MPFM with excellent
oil/water characterisation capabilities. The in-situ characterisation would then provide
valuable information for emulsion destabilisation and treatment in oilfield production.
The multiphase flow metering industry has previously highlighted the potential of
tomographic techniques (Chaouki et al., 1997). Current MPFMs have considerable
measurement uncertainty due to the flow regime dependence of many commercial techniques
(Sætre et al., 2012). Capturing real-time images of multiphase flows would provide flow
regime identification; reducing measurement uncertainty and providing a valuable
information for flow assurance purposes (Ismail et al., 2005). NMR imaging has previously
demonstrated the ability to characterise and accurately quantify multiphase flow behaviour
across a range of chemical engineering applications (Gladden and Sederman, 2013, Gladden
and Sederman, 2017). The current EFNMR system has an inherently poor SNR, results in
an unreasonable timeframe and resolution for image acquisition (Halse et al., 2008).
However modern developments in statistical analysis (e.g. Bayesian analysis) have enabled
spatially quantitative measurements to be achieved with EFNMR (Fridjonsson et al., 2015,
Ross et al., 2012). The future low-field NMR system (discussed in section 8.2.2) could be
designed to capture magnetic resonance images of reasonable SNR. This could provide
tomographic analysis of the pipe cross-section; giving the potential for flow regime
identification which provides considerable benefits for flow assurance and production
monitoring purposes (Thorn et al., 1999).
143
References
Abouelwafa, M. S. A. & Kendall, E. J. M. 1980. The measurement of component ratios in
multiphase systems using alpha -ray attenuation. Journal of Physics E: Scientific
Instruments, 13, 341.
Abragam, A. 1955. Overhauser Effect in Nonmetals. Physical Review, 98, 1729-1735.
Abragam, A. 1961. The principles of nuclear magnetism, Clarendon Press.
Åbro, E. & Johansen, G. A. 1999. Improved void fraction determination by means of
multibeam gamma-ray attenuation measurements. Flow Measurement and
Instrumentation, 10, 99-108.
Al-Lababidi, S., Mba, D. & Addali, A. 2012. “Upstream Multiphase Flow Assurance
Monitoring Using Acoustic Emission”, in Sikorski, W. (ed) Acoustic Emission.
InTech Open, 217-250.
Al-Safran, E. 2009. Investigation and prediction of slug frequency in gas/liquid horizontal
pipe flow. Journal of Petroleum Science and Engineering, 69, 143-155.
Al‐Sheikh, J. N., Saunders, D. E. & Brodkey, R. S. 1970. Prediction of flow patterns in
horizontal two‐phase pipe flow. The Canadian Journal of Chemical Engineering,
48, 21-29.
Alboudwarej, H., Muhammad, M., Shahraki, A. K., Dubey, S., Vreenegoor, L. & Saleh, J.
M. 2007. Rheology of heavy-oil emulsions. SPE Production & Operations, 22, 285-
293.
Alessandri, S., Luchinat, C. & Parigi, G. 2005. Multi-exponential fit of NMRD data for
olive oil analysis. Conference on field cycling NMR relaxometry. Torino, Italy.
Allsopp, K., Wright, I., Lastockin, D., Mirotchnik, K. & Kantzas, A. 2001. Determination
of oil and water compositions of oil/water emulsions using low field NMR
relaxometry. Journal of Canadian Petroleum Technology, 40, 58-61.
Alvarado, V. & Manrique, E. 2010. Enhanced Oil Recovery: Field Planning and
Development Strategies, Elsevier Science.
Andersen, C. M., Frøst, M. B. & Viereck, N. 2010. Spectroscopic characterization of low-
and non-fat cream cheeses. International Dairy Journal, 20, 32-39.
Andreussi, P., Bendiksen, K. H. & Nydal, O. J. 1993. Void distribution in slug flow.
International Journal of Multiphase Flow, 19, 817-828.
Andreussi, P., Di Donfrancesco, A. & Messia, M. 1988. An impedance method for the
measurement of liquid hold-up in two-phase flow. International Journal of
Multiphase Flow, 14, 777-785.
144
Andrew, E. R. & Jurga, K. 1987. NMR probe with short recovery time. Journal of
Magnetic Resonance (1969), 73, 268-276.
Angeli, P. & Hewitt, G. F. 2000. Flow structure in horizontal oil–water flow. International
Journal of Multiphase Flow, 26, 1117-1140.
Appel, M., Freeman, J. & Pusiol, D. 2011. Robust Multi-Phase Flow Measurement Using
Magnetic Resonance Technology. SPE Middle East Oil and Gas Show and
Conference. Manama, Bahrain: Society of Petroleum Engineers.
Appelt, S., Häsing, F. W., Kühn, H., Perlo, J. & Blümich, B. 2005. Mobile high resolution
xenon nuclear magnetic resonance spectroscopy in the Earth’s magnetic field.
Physical Review Letters, 94, 197602.
Appelt, S., Kühn, H., Häsing, F. W. & Blümich, B. 2006. Chemical analysis by ultrahigh-
resolution nuclear magnetic resonance in the Earth’s magnetic field. Nature Physics,
2, 105.
Arirachakaran, S., Oglesby, K. D., Malinowsky, M. S., Shoham, O. & Brill, J. P. 1989. An
Analysis of Oil/Water Flow Phenomena in Horizontal Pipes. SPE Production
Operations Symposium. Oklahoma City, Oklahoma: Society of Petroleum
Engineers.
Armstrong, B. D., Lingwood, M. D., Mccarney, E. R., Brown, E. R., Blümler, P. & Han, S.
2008. Portable X-band system for solution state dynamic nuclear polarization.
Journal of Magnetic Resonance, 191, 273-281.
Armstrong, B. D., Soto, P., Shea, J.-E. & Han, S. 2009. Overhauser dynamic nuclear
polarization and molecular dynamics simulations using pyrroline and piperidine
ring nitroxide radicals. Journal of Magnetic Resonance, 200, 137-141.
Arnold, R., Burnett, D. B., Elphick, J., Galbrun, M., Hightower, M., Jiang, Z., Khan, M.,
Lavery, M. & Luffey, F. 2004. Managing water–from waste to resource. Oilfield
Review, 16, 26-41.
Arvoh, B. K., Hoffmann, R. & Halstensen, M. 2012. Estimation of volume fractions and
flow regime identification in multiphase flow based on gamma measurements and
multivariate calibration. Flow Measurement and Instrumentation, 23, 56-65.
Ashton, S. L., Cutmore, N. G., Roach, G. J., Watt, J. S., Zastawny, H. W. & Mcewan, A. J.
1994. Development and trial of microwave techniques for measurement of
multiphase flow of oil, water and gas. SPE Asia Pacific Oil and Gas Conference
and Exhibition. Melbourne: Society of Petroleum Engineers.
Atkinson, I., Theuveny, B., Berard, M., Conort, G., Groves, J., Lowe, T., McDiarmid, A.,
Mehdizadeh, P., Perciot, P. & Pinguet, B. 2004. A new horizon in multiphase flow
measurement. Oilfield Review, 16, 52-63.
Baker, O. 1953. Design of pipelines for the simultaneous flow of oil and gas. Fall Meeting
of the Petroleum Branch of AIME. Dallas: Society of Petroleum Engineers.
145
Baker, R. C. 2016. Flow Measurement Handbook, Cambridge University Press.
Barnea, D., Shoham, O. & Taitel, Y. 1980. Flow pattern characterization in two phase flow
by electrical conductance probe. International Journal of Multiphase Flow, 6, 387-
397.
Battocletti, J. H., Halbach, R. E., Salles‐Cunha, S. X. & Sances, A. 1981. The NMR blood
flowmeter—Theory and history. Medical physics, 8, 435-443.
Becerra, H. Q., Retamoso, C. & MacDonald, D. D. 2000. The corrosion of carbon steel in
oil-in-water emulsions under controlled hydrodynamic conditions. Corrosion
Science, 42, 561-575.
Becher, P. 1996. Encyclopedia of emulsion technology, New York, Marcel Dekker.
Beck, M. S. & Pląskowski, A. 1987. Cross Correlation Flowmeters, Their Design and
Application, Taylor & Francis.
Béné, G. J. 1980. Nuclear magnetism of liquid systems in the Earth field range. Physics
Reports, 58, 213-267.
Bensler, H., Delhaye, J. & Favreau, C. Measurement of interfacial area in bubbly flows by
means of an ultrasonic technique. National heat transfer conference, 1987.
Berge, F. E. Minimising platform space utilisation through the latest in multiphase
metering—a North Sea pilot. South East Asia Hydrocarbon Flow Measurement
Workshop, 2011 Kuala Lumpur.
Bernier, R. & Brennen, C. 1983. Use of the electromagnetic flowmeter in a two-phase flow.
International Journal of Multiphase Flow, 9, 251-257.
Blaney, S. & Yeung, H. 2008. Investigation of the exploitation of a fast-sampling single
gamma densitometer and pattern recognition to resolve the superficial phase
velocities and liquid phase water cut of vertically upward multiphase flows. Flow
Measurement and Instrumentation, 19, 57-66.
Bloch, F. 1946. Nuclear Induction. Physical Review, 70, 460-474.
Blümich, B., Casanova, F. & Appelt, S. 2009a. NMR at low magnetic fields. Chemical
Physics Letters, 477, 231-240.
Blümich, B., Mauler, J., Haber, A., Perlo, J., Danieli, E. & Casanova, F. 2009b. Mobile
NMR for geophysical analysis and materials testing. Petroleum Science, 6, 1-7.
Borges, G. R., Farias, G. B., Braz, T. M., Santos, L. M., Amaral, M. J., Fortuny, M.,
Franceschi, E., Dariva, C. & Santos, A. F. 2015. Use of near infrared for evaluation
of droplet size distribution and water content in water-in-crude oil emulsions in
pressurized pipeline. Fuel, 147, 43-52.
146
Borgia, G., Bortolotti, V., Brancolini, A., Brown, R. & Fantazzini, P. 1996. Developments
in core analysis by NMR measurements. Magnetic Resonance Imaging, 14, 751-
760.
Bovey, F. A., Mirau, P. A. & Gutowsky, H. S. 1988. Nuclear Magnetic Resonance
Spectroscopy, San Diego, Academic Press Inc.
Brennen, C. E. 2005. Fundamentals of Multiphase Flow, New York, Cambridge University
Press.
British Petroleum 2017. BP Statistical Review of World Energy. Energy Economics.
London.
British Petroleum 2018. BP Energy Outlook. Energy Economics. London.
Brown, R. W., Haacke, E. M., Cheng, Y. C. N., Thompson, M. R. & Venkatesan, R. 2014.
Magnetic Resonance Imaging: Physical Principles and Sequence Design, Wiley.
Bruvik, E. & Hjertaker, B. 2016. Liquid characterization and measurement of fluid
properties for reduced uncertainty in multiphase flow area fraction measurements.
Flow Measurement and Instrumentation, 47, 10-18.
Callaghan, P., Eccles, C., Haskell, T., Langhorne, P. & Seymour, J. 1998. Earth's field
NMR in Antarctica: a pulsed gradient spin echo NMR study of restricted diffusion
in sea ice. Journal of Magnetic Resonance, 133, 148-154.
Callaghan, P., Eccles, C. & Seymour, J. 1997. An earth’s field nuclear magnetic resonance
apparatus suitable for pulsed gradient spin echo measurements of self-diffusion
under Antarctic conditions. Review of scientific instruments, 68, 4263-4270.
Callaghan, P. & Le Gros, M. 1982. Nuclear spins in the Earth’s magnetic field. American
Journal of Physics, 50, 709-713.
Callaghan, P. T. 1991. Principles of Nuclear Magnetic Resonance Microscopy, Oxford,
Clarendon Press.
Callaghan, P. T. 1999. Rheo-NMR: nuclear magnetic resonance and the rheology of
complex fluids. Reports on Progress in Physics, 62, 599.
Callaghan, P. T. 2006. Rheo-NMR and velocity imaging. Current Opinion in Colloid &
Interface Science, 11, 13-18.
Callaghan, P. T., Coy, A., Dykstra, R., Eccles, C. D., Halse, M. E., Hunter, M. W., Mercier,
O. R. & Robinson, J. N. 2007. New Zealand Developments in Earth's Field NMR.
Applied Magnetic Resonance, 32, 63-74.
Callaghan, P. T., Godefroy, S. & Ryland, B. N. 2003. Diffusion–relaxation correlation in
simple pore structures. Journal of Magnetic Resonance, 162.
Caprihan, A. & Fukushima, E. 1990. Flow measurements by NMR. Physical Reports, 198,
195-235.
147
Carr, H. Y. & Purcell, E. M. 1954. Effects of diffusion on free precession in nuclear
magnetic resonance experiments. Physical Review, 94, 630.
Ceccio, S. L. & George, D. L. 1996. A Review of Electrical Impedance Techniques for the
Measurement of Multiphase Flows. Journal of Fluids Engineering, 118, 391-399.
Chaouki, J., Larachi, F. & Dudukovic, M. P. 1997. Noninvasive tomographic and
velocimetric monitoring of multiphase flows. Industrial & Engineering Chemistry
Research, 36, 4476-4503.
Charles, M. E., Govier, G. W. & Hodgson, G. W. 1961. The horizontal pipeline flow of
equal density oil‐water mixtures. Canadian Journal of Chemical Engineering, 39,
27-36.
Charles, M. E. & Lilleleht, L. V. 1966. Correlation of pressure gradients for the stratified
laminar-turbulent pipeline flow of two immiscible liquids. The Canadian Journal of
Chemical Engineering, 44, 47-49.
Chatfield, C. 2016. The Analysis of Time Series: An Introduction, Sixth Edition, CRC Press.
Chun, M.-H. & Sung, C.-K. 1986. Parametric effects on the void fraction measurement by
capacitance transducers. International Journal of Multiphase Flow, 12, 627-640.
Coates, G. R., Xiao, L. & Prammer, M. G. 1999. NMR logging: principles and
applications, Houston, Haliburton Energy Services.
Conradi, M. S., Altobelli, S. A., Sowko, N. J., Conradi, S. H. & Fukushima, E. 2017. Pre-
polarization fields for earth’s field NMR: Fast discharge for use with short T1 and
large coils. Journal of Magnetic Resonance, 281, 241-245.
Constantinesco, A., Choquet, P., Cauffet, G., Fournier, J. M., Ravier, S., Drillon, J. M. &
Aubert, G. 1997. Low-field dedicated and desktop magnetic resonance imaging
systems for agricultural and food applications. Magnetic Resonance in Chemistry,
35, S69-S75.
Corneliussen, S., Couput, J., Dahl, E., Dykesteen, E., Frøysa, K.-E., Malde, E., Moestue,
H., Moksnes, P. O., Scheers, L. & Tunheim, H. 2005. Handbook of Multiphase
Flow Metering, Oslo, The Norwegian Society for Oil and Gas Measurement.
Costigan, G. & Whalley, P. B. 1997. Slug flow regime identification from dynamic void
fraction measurements in vertical air-water flows. International Journal of
Multiphase Flow, 23, 263-282.
Crawford, J. 1981. Marine and Offshore Pumping and Piping Systems, London,
Butterworths.
Da Silva, M. J. 2008. Impedance Sensors for Fast Multiphase Flow Measurement and
Imaging. Doktoringenieur, Technische Universität Dresden.
148
Dalitz, F., Cudaj, M., Maiwald, M. & Guthausen, G. 2012. Process and reaction monitoring
by low-field NMR spectroscopy. Progress in Nuclear Magnetic Resonance
Spectroscopy, 60, 52-70.
Danieli, E., Perlo, J., Blümich, B. & Casanova, F. 2010. Small magnets for portable NMR
spectrometers. Angewandte Chemie International Edition, 49, 4133-4135.
Danieli, E., Perlo, J., Duchateau, A. L. L., Verzijl, G. K. M., Litvinov, V. M., Blümich, B.
& Casanova, F. 2014. On-Line Monitoring of Chemical Reactions by using Bench-
Top Nuclear Magnetic Resonance Spectroscopy. ChemPhysChem, 15, 3060-3066.
De Leeuw, R. Liquid correction of Venturi meter readings in wet gas flow. North Sea Flow
Measurement Workshop, 1997. 335.
Delves, L. M. & Walsh, J. E. 1974. Numerical solution of integral equations, Michigan,
Clarendon Press.
Douglas-Westwood Limited 2004. The World Deepwater Report.
Dukler, A. E. & Hubbard, M. G. 1975. A model for gas-liquid slug flow in horizontal and
near horizontal tubes. Industrial & Engineering Chemistry Fundamentals, 14, 337-
347.
Dunn, K. J., Bergman, D. J. & Latorraca, G. A. 2002. Nuclear Magnetic Resonance:
Petrophysical and Logging Applications, Oxford, Elsevier Science.
Dykesteen, E., Hallanger, A., Hammer, E., Samnoy, E. & Thorn, R. 1985. Non-intrusive
three-component ratio measurement using an impedance sensor. Journal of Physics
E: Scientific Instruments, 18, 540.
Economides, M. J., Hill, A. D. & Ehlig-Economides, C. 1994. Petroleum Production
Systems, PTR Prentice Hall.
Falcimaigne, J. & Decarre, S. 2008. Multiphase Production: Pipeline Transport, Pumping
and Metering, Editions Technip.
Falcone, G., Hewitt, G. & Alimonti, C. 2009. Multiphase Flow Metering: Principles and
Applications, Elsevier Science.
Falcone, G., Hewitt, G. F., Alimonti, C. & Harrison, B. 2001. Multiphase Flow Metering:
Current Trends and Future Developments. SPE Annual Technical Conference and
Exhibition. New Orleans, Louisiana: Society of Petroleum Engineers.
Falcone, G., Hewitt, G. F., Alimonti, C. & Harrison, B. 2002. Multiphase flow metering:
current trends and future developments. Journal of Petroleum Technology, 54, 77-
84.
Farah, M. A., Oliveira, R. C., Caldas, J. N. & Rajagopal, K. 2005. Viscosity of water-in-oil
emulsions: Variation with temperature and water volume fraction. Journal of
Petroleum Science and Engineering, 48, 169-184.
149
Farrar, T. C. 1989. Introduction to Pulse NMR Spectroscopy, Farragut Press.
Fee, D. A. & O'dea, J. 2014. Technology for developing marginal offshore oilfields,
London, Taylor and Francis.
Fink, J. K. 2016. Guide to the Practical Use of Chemicals in Refineries and Pipelines,
Cambridge, Gulf Professional Publishing.
Folgerø, K., Lyng, S. E., Kocbach, J., Frøysa, K. E., Kleppe, K. & Åbro, E. Uncertainty
analysis of multiphase flow meters used for allocation measurements: field
experiences and future challenges. International North Sea Flow Meaurement
Workshop, 2013 Tønsberg.
Freedman, R. & Heaton, N. 2004. Fluid characterization using nuclear magnetic resonance
logging. Petrophysics, 45, 241-250.
Fridjonsson, E. O., Flux, L. S. & Johns, M. L. 2012. Determination of mean droplet sizes of
water-in-oil emulsions using an Earth’s field NMR instrument. Journal of Magnetic
Resonance, 221, 97-102.
Fridjonsson, E. O., Graham, B. F., Akhfash, M., May, E. F. & Johns, M. L. 2014a.
Optimized droplet sizing of water-in-crude oil emulsions using nuclear magnetic
resonance. Energy and Fuels, 28, 1756-1764.
Fridjonsson, E. O., Stanwix, P. L. & Johns, M. L. 2014b. Earth’s field NMR flow meter:
Preliminary quantitative measurements. Journal of Magnetic Resonance, 245, 110-
115.
Fridjonsson, E. O., Vogt, S. J., Vrouwenvelder, J. S. & Johns, M. L. 2015. Early non-
destructive biofouling detection in spiral wound RO membranes using a mobile
earth׳s field NMR. Journal of Membrane Science, 489, 227-236.
Frøystein, T., Kvandal, H. & Aakre, H. 2005. Dual energy gamma tomography system for
high pressure multiphase flow. Flow Measurement and Instrumentation, 16, 99-112.
Fukushima, E. 1999. Nuclear magnetic resonance as a tool to study flow. Ann. Rev. Fluid
Mech., 31, 95-123.
Fukushima, E. & Roeder, S. B. W. 1993. Experimental Pulse NMR: A Nuts and Bolts
Approach, Addison-Wesley.
García-Golding, F., Giallorenzo, M., Moreno, N. & Chang, V. 1995. Sensor for
determining the water content of oil-in-water emulsion by specific admittance
measurement. Sensors and Actuators A: Physical, 47, 337-341.
Gerritsma, C. J. & Trappeniers, N. J. 1971. Proton-spin-lattice relaxation and self-diffusion
in methanes: II. Physica, 51, 381-394.
Gidaspow, D. 1994. Multiphase Flow and Fluidization: Continuum and Kinetic Theory
Descriptions, San Diego, Academic Press Inc.
150
Gladden, L. F. 1994. Nuclear magnetic resonance in chemical engineering: principles and
applications. Chemical Engineering Science, 49, 3339-3408.
Gladden, L. F. & Alexander, P. 1996. Applications of nuclear magnetic resonance imaging
in process engineering. Measurement Science and Technology, 7, 423.
Gladden, L. F. & Sederman, A. J. 2013. Recent advances in flow MRI. Journal of Magnetic
Resonance, 229, 2-11.
Gladden, L. F. & Sederman, A. J. 2017. Magnetic resonance imaging and velocity mapping
in chemical engineering applications. Annual review of chemical and biomolecular
engineering, 8, 227-247.
Godefroy, S. & Callaghan, P. T. 2003. 2D relaxation/diffusion correlations in porous
media. Magnetic Resonance Imaging, 21.
Godefroy, S., Creamer, L. K., Watkinson, P. J. & Callaghan, P. T. 2003. The Use of 2D
Laplace Inversion in Food Material. In: Belton, P. S. (ed.) Magnetic Resonance in
Food Science: Latest Developments. Cambridge: Royal Society of Chemistry.
Golub, G. H., Heath, M. & Wahba, G. 1979. Generalized cross-validation as a method for
choosing a good ridge parameter. Technometrics, 21, 215-223.
Goncalves, J. L., Paiva, T. A., Carvalho, R. D. & Venturini, O. J. Application of the
ultrasonic technique for monitoring and measuring the phase fractions of liquid-gas-
solid mixtures. Brasil Offshore, 2011. Society of Petroleum Engineers.
Graf, V., Noack, F. & Béné, G. 1980. Proton spin T1 relaxation dispersion in liquid H2O by
slow proton‐exchange. The Journal of Chemical Physics, 72, 861-863.
Granwehr, J., Harel, E., Han, S., Garcia, S., Pines, A., Sen, P. N. & Song, Y.-Q. 2005.
Time-of-flight flow imaging using NMR remote detection. Physical review letters,
95, 075503.
Green, R. G. & Cunliffe, J. M. 1983. A frequency-modulated capacitance transducer for on-
line measurement of two-component fluid flow. Measurement, 1, 191-195.
Grombacher, D., Walbrecker, J. O. & Knight, R. 2013. Quantifying background magnetic-
field inhomogeneity for improved interpretation of free induction decay
measurements. Geophysics, 79, E11-E21.
Grunewald, E. & Knight, R. 2009. A laboratory study of NMR relaxation times and pore
coupling in heterogeneous media. Geophysics, 74, E215-E221.
Grunewald, E. & Knight, R. 2011. The effect of pore size and magnetic susceptibility on
the surface NMR relaxation parameter T2. Near Surface Geophysics, 9, 169-178.
Grunewald, E. & Knight, R. 2012. Nonexponential decay of the surface-NMR signal and
implications for water content estimation. Geophysics, 77, EN1-EN9.
Hahn, E. L. 1950. Spin echoes. Physical review, 80, 580.
151
Halbach, K. 1980. Design of permanent multipole magnets with oriented rare earth cobalt
material. Nuclear Instruments and Methods, 169, 1-10.
Halse, M. E. & Callaghan, P. T. 2008. A dynamic nuclear polarization strategy for multi-
dimensional Earth’s field NMR spectroscopy. Journal of Magnetic Resonance, 195,
162-168.
Halse, M. E., Coy, A., Dykstra, R., Eccles, C., Hunter, M., Ward, R. & Callaghan, P. T.
2006. A practical and flexible implementation of 3D MRI in the Earth’s magnetic
field. Journal of Magnetic Resonance, 182, 75-83.
Halse, M. E., Coy, A., Dykstra, R., Eccles, C. D., Hunter, M. W. & Callaghan, P. T. 2008.
Multidimensional Earth's-Field NMR. In: Codd, S. L. & Seymour, J. D. (eds.)
Magnetic Resonance Microscopy. Weinheim: John Wiley & Sons.
Hanratty, T. J. 2013. Physics of Gas-Liquid Flows, Cambridge University Press.
Harel, E., Granwehr, J., Seeley, J. A. & Pines, A. 2006. Multiphase imaging of gas flow in
a nanoporous material using remote-detection NMR. Nature materials, 5, 321.
Havre, K., Stornes, K. O. & Stray, H. 2000. Taming slug flow in pipelines. ABB review, 4,
55-63.
Hetsroni, G. 1982. Handbook of multiphase systems, Hemisphere Pub. Corp.
Hewitt, G. F., Delhaye, J. M. & Zuber, N. 2013. Multiphase Science and Technology,
Springer Berlin Heidelberg.
Hill, T., Fairhurst, C., Nelson, C., Becerra, H. & Bailey, R. 1996. Multiphase production
through hilly terrain pipelines in Cusiana oilfield, Colombia. SPE Annual Technical
Conference and Exhibition, 1996 Denver, USA.
Hills, B., Benamira, S., Marigheto, N. & Wright, K. 2004. T1-T2 correlation analysis of
complex foods. Applied Magnetic Resonance, 26.
Hogendoorn, J., Boer, A., Appel, M., De Jong, H. & De Leeuw, R. Magnetic resonance
technology - a new concept for multiphase flow measurement. International North
Sea Flow Measurement Workshop, 22-25 Oct 2013 Tønsberg.
Hogendoorn, J., Boer, A., Van Der Zande, M. & De Leeuw, R. 2016. Magnetic resonance
multiphase flowmeter: Pratical experiences. South East Asia Flow Measurement
Conference. Kuala Lumpur.
Hogendoorn, J., Boer, A., Van Der Zande, M., Moeleker, M., Appel, M., De Jong, H. & De
Leeuw, R. 2015. Magnetic resonance multiphase flowmeter: measuring principle
and multiple test results. Upstream Production Measurement Forum, 2015
Houston.
Hogendoorn, J., Boer, A., Zoeteweij, M., Bousché, O., Tromp, R., Leeuw, R. D., Moeleker,
P., Appel, M. & Jong, H. D. 2014. Magnetic Resonance Multiphase Flow Meter:
152
Gas Flow Measurement Principle and Wide Range Testing Results. International
North Sea Flow Measurement Workshop. St Andrews.
Hogendoorn, J., Zande, M. V. D., Zoeteweij, M., Tromp, R., Cerioni, L., Boer, A. &
Leeuw, R. D. 2017. Practical experiences obtained with the magnetic resonance
multiphase flowmeter. North Sea Flow Measurement Workshop. Sandefjord.
Holditch, S. & Chianelli, R. 2008. Factors That Will Influence Oil and Gas Supply and
Demand in the 21st Century. MRS Bulletin, 33, 317-323.
Holland, D. J., Blake, A., Tayler, A. B., Sederman, A. J. & Gladden, L. F. 2011. A
Bayesian approach to characterising multi-phase flows using magnetic resonance:
Application to bubble flows. Journal of Magnetic Resonance, 209, 83-87.
Holland, D. J. & Gladden, L. F. 2015. Software Developments: Improvements in Data
Analysis. In: Johns, M. L., Fridjonsson, E. O., Vogt, S. J. & Haber, A. (eds.) Mobile
NMR and MRI. Cambridge: Royal Society of Chemistry.
Hollingsworth, K. & Johns, M. 2003. Measurement of emulsion droplet sizes using PFG
NMR and regularization methods. Journal of Colloid and Interface Science, 258,
383-389.
Hollingsworth, K. G., Sederman, A. J., Buckley, C., Gladden, L. F. & Johns, M. L. 2004.
Fast emulsion droplet sizing using NMR self-diffusion measurements. Journal of
Colloid and Interface Science, 274, 244-250.
Hopper, T., Mandal, S., Cory, D., Hürlimann, M. & Song, Y.-Q. 2011. Low-frequency
NMR with a non-resonant circuit. Journal of Magnetic Resonance, 210, 69-74.
Hore, P. J. 2015. Nuclear Magnetic Resonance, Oxford, Oxford University Press.
Hornemann, J. A., Codd, S. L., Romanenko, K. R. & Seymour, J. D. 2009. T2–T2 exchange
in biofouled porous media. Diffusion Fundamentals, 10, 1.1-1.3.
Hoult, D. I. 1984. Fast recovery with a conventional probe. Journal of Magnetic
Resonance, 57, 394-403.
Hoyle, B. 1996. Process tomography using ultrasonic sensors. Measurement Science and
Technology, 7, 272.
Huang, Z., Wang, B. & Li, H. 2003. Application of electrical capacitance tomography to
the void fraction measurement of two-phase flow. IEEE Transactions on
Instrumentation and Measurement, 52, 7-12.
Hürlimann, M., Venkataramanan, L. & Flaum, C. 2002. The diffusion–spin relaxation time
distribution function as an experimental probe to characterize fluid mixtures in
porous media. The Journal of Chemical Physics, 117, 10223-10232.
Hurlimann, M. D., Burcaw, L. & Song, Y. Q. 2006. Quantitative characterization of food
products by two-dimensional D-T2 and T1-T2 distribution functions in a static
gradient. Journal of Colloid and Interface Science, 297, 303-11.
153
Hürlimann, M. D., Flaum, M., Venkataramanan, L., Flaum, C., Freedman, R. & Hirasaki,
G. J. 2003. Diffusion-relaxation distribution functions of sedimentary rocks in
different saturation states. Magnetic Resonance Imaging, 21, 305-310.
Hürlimann, M. D. & Heaton, N. J. 2015. NMR well logging. In: Johns, M. L., Fridjonsson,
E. O., Vogt, S. J. & Haber, A. (eds.) Mobile NMR and MRI. Cambridge: Royal
Society of Chemistry.
Hürlimann, M. D. & Venkataramanan, L. 2002. Quantitative measurement of two-
dimensional distribution functions of diffusion and relaxation in grossly
inhomogeneous fields. Journal of Magnetic Resonance, 157, 31-42.
IBM Institute for Business Value 2010. Oil and Gas 2030 Meeting the growing demands
for energy in the coming decades. Somers, NY: IBM Global Business Services.
Iguchi, M. & Ilegbusi, O. J. 2013. Basic Transport Phenomena in Materials Engineering,
Springer Japan.
Igunnu, E. T. & Chen, G. Z. 2014. Produced water treatment technologies. International
Journal of Low-Carbon Technologies, 9, 157-177.
Ingle, J. D. & Crouch, S. R. 1988. Spectrochemical Analysis, Prentice Hall.
International Energy Agency 2017a. Energy Technology Perspectives. Paris.
International Energy Agency 2017b. World Energy Outlook. Paris.
Ismail, I., Gamio, J. C., Bukhari, S. F. A. & Yang, W. Q. 2005. Tomography for multi-
phase flow measurement in the oil industry. Flow Measurement and
Instrumentation, 16, 145-155.
Issa, R. & Kempf, M. 2003. Simulation of slug flow in horizontal and nearly horizontal
pipes with the two-fluid model. International journal of multiphase flow, 29, 69-95.
Istratov, A. A. & Vyvenko, O. F. 1999. Exponential analysis in physical phenomena.
Review of Scientific Instruments, 70, 1233-1257.
Jahn, F. 1998. Hydrocarbon Exploration and Production, Elsevier Science.
Jepson, W. P., Bhongale, S. & Gopal, M. 1996. Predictive model for sweet corrosion in
horizontal multiphase slug flow. Corrosion NACE International Conference and
Exposition, 1996 Houston, USA.
Jin, N., Xin, Z., Wang, J., Wang, Z., Jia, X. & Chen, W. 2008. Design and geometry
optimization of a conductivity probe with a vertical multiple electrode array for
measuring volume fraction and axial velocity of two-phase flow. Measurement
science and technology, 19, 045403.
Johns, M. L. 2009. NMR studies of emulsions. Current Opinion in Colloid & Interface
Science, 14, 178-183.
154
Jones, S. W., Amblard, A. & Favreau, C. 1986. Interaction of an ultrasonic wave with a
bubbly mixture. Experiments in Fluids, 4, 341-349.
Kantzas, A. 2004. On-line Measurement of Fluid Content and Fluid Properties for the Oil
and Gas Industry with Low Field NMR Relaxometry. Canadian School of
Hydrocarbon Measurement. Calgary, Alberta.
Kantzas, A., Kryuchkov, S. & Chandrasekaran, B. 2009. Advances in multiphase flow
measurements using magnetic resonance relaxometry. International Symposium on
Measurement Techniques for Multiphase Flows. Okinawa.
Keeler, J. 2005. Understanding NMR Spectroscopy, John Wiley & Sons.
Kentgens, A., Bart, J., Van Bentum, P., Brinkmann, A., Van Eck, E., Gardeniers, J. G.,
Janssen, J., Knijn, P., Vasa, S. & Verkuijlen, M. 2008. High-resolution liquid-and
solid-state nuclear magnetic resonance of nanoliter sample volumes using microcoil
detectors. The Journal of Chemical Physics, 128, 052202.
Kernevez, N. & Glenat, H. 1991. Description of a high sensitivity CW scalar DNP-NMR
magnetometer. IEEE Transactions on Magnetics, 27, 5402-5404.
Khan, A. S. 2014. Microwave Engineering: Concepts and Fundamentals, CRC Press.
Klaus, A., Tiddy, G. J. T., Solans, C., Harrar, A., Touraud, D. & Kunz, W. 2012. Effect of
salts on the phase behavior and the stability of nano-emulsions with rapeseed oil
and an extended surfactant. Langmuir, 28, 8318-8328.
Kleinberg, R. L. 2001a. NMR well logging at Schlumberger. Concepts in Magnetic
Resonance, 13, 396-403.
Kleinberg, R. L. 2001b. Well logging overview. Concepts in Magnetic Resonance, 13, 342-
343.
Kleinberg, R. L. & Jackson, J. A. 2001. An introduction to the history of NMR well
logging. Concepts in Magnetic Resonance, 13, 340-342.
Kleinberg, R. L. & Vinegar, H. J. 1996. NMR Properties of Reservoir Fluids. The Log
Analyst, 37, 20-32.
Klug, F. & Mayinger, F. 1994. Impedance based flow reconstruction-A novel flow
composition measuring technique for multi-phase-flows. Nuclear Engineering and
Design, 146, 35-42.
Knight, R., Grunewald, E., Irons, T., Dlubac, K., Song, Y., Bachman, H. N., Grau, B.,
Walsh, D., Abraham, J. D. & Cannia, J. 2012. Field experiment provides ground
truth for surface nuclear magnetic resonance measurement. Geophysical Research
Letters, 39.
Kokal, S. L. 2005. Crude Oil Emulsions: A State-Of-The-Art Review. SPE Production &
Facilities, 20, 5-13.
155
Kolev, N. I. 2011. Multiphase Flow Dynamics 1: Fundamentals, Springer Berlin
Heidelberg.
Korb, J. P. & Bryant, R. G. 2002. Magnetic field dependence of proton spin‐lattice
relaxation times. Magnetic Resonance in Medicine, 48, 21-26.
Krohne 2017. M-Phase 5000 Magnetic resonance multiphase flowmeter for the
simultaneous measurement of oil, gas and water. Dordrecht.
Krüger, G. J., Birke, A. & Weiss, R. 1996. Nuclear magnetic resonance (NMR) two-phase
mass flow measurements. Flow Measurement and Instrumentation, 7, 25-37.
Krüger, G. J., Haupt, J. & Weiss, R. A nuclear magnetic resonance method for the
investigation of two-phase flow. Measuring Techniques in Gas-Liquid Two-Phase
Flows, 1984 Nancy. Springer, 435-454.
Krynicki, K. 1966. Proton spin-lattice relaxation in pure water between 0°C and 100°C.
Physica, 32, 167-178.
Lakshmanan, S., Maru, W. A., Holland, D. J., Mantle, M. D. & Sederman, A. J. 2016.
Measurement of an oil–water flow using magnetic resonance imaging. Flow
Measurement and Instrumentation, 53, 161-171.
Leblond, J., Javelot, S., Lebrun, D. & Lebon, L. 1998. Two-phase flow characterization by
nuclear magnetic resonance. Nuclear Engineering and Design, 184, 229-237.
Lee, H., Sun, E., Ham, D. & Weissleder, R. 2008. Chip–NMR biosensor for detection and
molecular analysis of cells. Nature medicine, 14, 869.
Leeungculsatien, T. & Lucas, G. P. 2013. Measurement of velocity profiles in multiphase
flow using a multi-electrode electromagnetic flow meter. Flow Measurement and
Instrumentation, 31, 86-95.
Lemonnier, H. 2010. Nuclear magnetic resonance: A new tool for the validation of
multiphase multidimensional CFD codes. Nuclear Engineering and Design, 240,
2139-2147.
Leu, G., Fordham, E. J., Hürlimann, M. D. & Frulla, P. 2005. Fixed and pulsed gradient
diffusion methods in low-field core analysis. Magnetic Resonance Imaging, 23,
305-309.
Leupold, H. A., Potenziani, E. & Tilak, A. 1993. Adjustable multi-tesla permanent magnet
field sources. IEEE transactions on magnetics, 29, 2902-2904.
Levitt, M. H. 2013. Spin Dynamics: Basics of Nuclear Magnetic Resonance, Wiley.
Lin, P. Y. & Hanratty, T. J. 1987. Detection of slug flow from pressure measurements.
International Journal of Multiphase Flow, 13, 13-21.
Lin, Z. 1982. Two-phase flow measurements with sharp-edged orifices. International
Journal of Multiphase Flow, 8, 683-693.
156
Ling, N. N., Haber, A., Graham, B. F., Aman, Z. M., May, E. F., Fridjonsson, E. O. &
Johns, M. L. 2018. Quantifying the Effect of Salinity on Oilfield Water-in-Oil
Emulsion Stability. Energy & Fuels, 32, 10042-10049.
Ling, N. N. A., Haber, A., Hughes, T. J., Graham, B. F., May, E. F., Fridjonsson, E. O. &
Johns, M. L. 2016. NMR Studies of the Effect of CO2 on Oilfield Emulsion
Stability. Energy & Fuels, 30, 5555-5562.
Ling, N. N. A., Haber, A., May, E. F., Fridjonsson, E. O. & Johns, M. L. 2014. NMR
studies of emulsion microstructure approaching the phase inversion point. Colloids
and Surfaces A, 462, 244-251.
Ling, N. N. A., Haber, A., May, E. F., Fridjonsson, E. O. & Johns, M. L. 2017. By-line
NMR emulsion droplet sizing. Chemical Engineering Science, 160, 362-369.
Lovick, J. & Angeli, P. 2004. Experimental studies on the dual continuous flow pattern in
oil–water flows. International Journal of Multiphase Flow, 30, 139-157.
Lucas, T., Le Ray, D. & Davenel, A. 2005. Chilling and freezing of part-baked bread. Part
I: An MRI signal analysis. Journal of Food Engineering, 70, 139-149.
Lucas, T., Musse, M., Bornert, M., Davenel, A. & Quellec, S. 2012. Temperature mapping
in bread dough using SE and GE two-point MRI methods: experimental and
theoretical estimation of uncertainty. Magnetic Resonance Imaging, 30, 431-45.
Luiten, A., Knoet, F., Vlaardingerbroek, M. T. & Den Boer, J. A. 2013. Magnetic
Resonance Imaging: Theory and Practice, Berlin, Springer.
Lunde, E. H. & Hoseth, J. G. 1994. Subsea Multiphase Flow Meters. Subsea Control and
Data Acquisition: Proceedings of an international conference. London, UK:
Society of Underwater Technology.
Luthria, D. L. 2004. Oil Extraction and Analysis: Critical Issues and Competitive Studies,
Taylor & Francis.
Lysak, P. D., Jenkins, D. M., Capone, D. E. & Brown, W. L. 2008. Analytical model of an
ultrasonic cross-correlation flow meter, part 1: Stochastic modeling of turbulence.
Flow Measurement and Instrumentation, 19, 1-7.
Mandhane, J., Gregory, G. & Aziz, K. 1974. A flow pattern map for gas—liquid flow in
horizontal pipes. International Journal of Multiphase Flow, 1, 537-553.
Matsui, G. 1984. Identification of flow regimes in vertical gas-liquid two-phase flow using
differential pressure fluctuations. International Journal of Multiphase Flow, 10,
711-719.
May, E. F., Graham, B. F., Chauhan, A. S. & Trengove, R. D. 2008. Shear and Electrical
Property Measurements of Water-in-Oil Emulsions and Implications for Multiphase
Flow Meters. Energy & Fuels, 22, 3308-3316.
McAleese, S. 2000. Operational Aspects of Oil and Gas Well Testing, Elsevier Science.
157
McDowell, A. & Fukushima, E. 2008. Ultracompact NMR: 1H spectroscopy in a
subkilogram magnet. Applied Magnetic Resonance, 35, 185-195.
Mehdizadeh, P. 2001. Multiphase Measuring Advances Continue. Oil & Gas Journal, 99.
Mehdizadeh, P., Farchy, D. & Suarez, J. 2009. Multiphase Meter Production Well Testing
Applied to Low-GOR Mature Fields. SPE Production and Operations Symposium.
Oklahoma City, Oklahoma: Society of Petroleum Engineers.
Mehdizadeh, P., Marrelli, J. & Ting, V. C. Wet gas metering: trends in applications and
technical developments. SPE Annual Technical Conference and Exhibition, 2002.
Society of Petroleum Engineers.
Meiboom, S. & Gill, D. 1958. Modified spin-echo method for measuring nuclear relaxation
times. Review of Scientific Instruments, 29.
Merilo, M. 1977. Void fraction measurement with a rotating electric field conductance
gauge. Transactions of the American Society of Mechanical Engineers, Series C,
Journal of heat transfer, 99, 330-333.
Michaelides, E., Crowe, C. T. & Schwarzkopf, J. D. 2016. Multiphase Flow Handbook,
CRC Press.
Miller, G. 2008. Commingled multiphase flows - the metering challenge. Offshore, 68.
Millheim, K. K. The Next Major Challenge for Deep Water: Developing Marginal Fields.
Offshore Technology Conference, 2008. Offshore Technology Conference.
Mitchell, J., Chandrasekera, T. C. & Gladden, L. F. 2010a. Obtaining true transverse
relaxation time distributions in high-field NMR measurements of saturated porous
media: Removing the influence of internal gradients. Journal of Chemical Physics,
132, 244705.
Mitchell, J., Chandrasekera, T. C. & Gladden, L. F. 2012. Numerical estimation of
relaxation and diffusion distributions in two dimensions. Progress in Nuclear
Magnetic Resonance Spectroscopy, 62, 34-50.
Mitchell, J., Chandrasekera, T. C., Johns, M. L., Gladden, L. F. & Fordham, E. J. 2010b.
Nuclear magnetic resonance relaxation and diffusion in the presence of internal
gradients: The effect of magnetic field strength. Physical Review E, 81, 026101.
Mitchell, J. & Fordham, E. J. 2011. Emulation of petroleum well-logging D− T2
correlations on a standard benchtop spectrometer. Journal of Magnetic Resonance,
212, 394-401.
Mitchell, J., Gladden, L., Chandrasekera, T. & Fordham, E. 2014. Low-field permanent
magnets for industrial process and quality control. Progress in Nuclear Magnetic
Resonance Spectroscopy, 76, 1-60.
158
Mitchell, J., Griffith, J. D., Collins, J. H. P., Sederman, A. J., Gladden, L. F. & Johns, M. L.
2007. Validation of NMR relaxation exchange time measurements in porous media.
Journal of Chemical Physics, 127, 234701.
Mohoric, A. & Stepisnik, J. 2009. NMR in the Earth's magnetic field. Progress in Nuclear
Magnetic Resonance Spectroscopy, 54, 166-182.
Mohoric, A., Stepisnik, J., Kos, M. & Planinsi 1999. Self-diffusion imaging by spin echo in
Earth's magnetic field. Journal of Magnetic Resonance, 136, 22-6.
Moradi, M., Alvarado, V. & Huzurbazar, S. 2011. Effect of Salinity on Water-in-Crude Oil
Emulsion: Evaluation through Drop-Size Distribution Proxy. Energy & Fuels, 25,
260-268.
Morriss, C., Rossini, D., Straley, C., Tutunjian, P. & Vinegar, H. 1997. Core Analysis By
Low-field NMR. The Log Analyst, 38, 12.
Mortensen, M., Thybo, A. K., Bertram, H. C., Andersen, H. J. & Engelsen, S. B. 2005.
Cooking Effects on Water Distribution in Potatoes Using Nuclear Magnetic
Resonance Relaxation. Journal of Agricultural and Food Chemistry, 53, 5976-5981.
Moulé, A. J., Spence, M. M., Han, S.-I., Seeley, J. A., Pierce, K. L., Saxena, S. & Pines, A.
2003. Amplification of xenon NMR and MRI by remote detection. Proceedings of
the National Academy of Sciences, 100, 9122-9127.
Mueller, E., Deimling, M. & Reinhardt, E. R. 1986. Quantification of pulsatile flow in MRI
by an analysis of T2 changes in ECG‐gated multiecho experiments. Magnetic
Resonance in Medicine, 3, 331-335.
Mukherjee, H. & Brill, J. 1985a. Pressure drop correlations for inclined two-phase flow.
Journal of Energy Resources Technology, 107, 549-554.
Mukherjee, H. & Brill, J. P. 1985b. Empirical equations to predict flow patterns in two-
phase inclined flow. International Journal of Multiphase Flow, 11, 299-315.
Müller, C. R., Davidson, J. F., Dennis, J. S., Fennell, P., Gladden, L. F., Hayhurst, A. N.,
Mantle, M. D., Rees, A. & Sederman, A. J. 2006. Real-time measurement of
bubbling phenomena in a three-dimensional gas-fluidized bed using ultrafast
magnetic resonance imaging. Physical Review Letters, 96, 154504.
Müller, C. R., Holland, D. J., Davidson, J. F., Dennis, J. S., Gladden, L. F., Hayhurst, A.
N., Mantle, M. D. & Sederman, A. J. 2007. Rapid two-dimensional imaging of
bubbles and slugs in a three-dimensional, gas-solid, two-phase flow system using
ultrafast magnetic resonance. Physical Review E, 75, 020302.
Munn, K. & Smith, D. M. 1987. A NMR technique for the analysis of pore structure:
Numerical inversion of relaxation measurements. Journal of Colloid and Interface
Science, 119, 117-126.
159
Murdock, J. 1962. Two-phase flow measurement with orifices. Journal of Basic
Engineering, 84, 419-432.
Nassereddin, T., Melo, M., Al Hagin, K., Hariz, M., Raman, B. & Helal, R. Infra-Red
Water Cut Meter Field Trial Under Transient Conditions. Abu Dhabi International
Petroleum Exhibition and Conference, 2010. Society of Petroleum Engineers.
Nydal, O., Pintus, S. & Andreussi, P. 1992. Statistical characterization of slug flow in
horizontal pipes. International Journal of Multiphase Flow, 18, 439-453.
Nyfors, E. & Vainikainen, P. 1989. Industrial Microwave Sensors, Artech House.
Oliemans, R. V. A., Ooms, G., Wu, H. L. & Duijvestijn, A. 1987. Core-annular oil/water
flow: the turbulent-lubricating-film model and measurements in a 5 cm pipe loop.
International Journal of Multiphase Flow, 13, 23-31.
Olson, D. L., Peck, T. L., Webb, A. G., Magin, R. L. & Sweedler, J. V. 1995. High-
resolution microcoil 1H-NMR for mass-limited, nanoliter-volume samples. Science,
270, 1967-1970.
Ong, J. T., Oyeneyin, M. B., Coutts, E. J. & Maclean, I. M. 2004. In Well Nuclear
Magnetic Resonance (NMR) Multiphase Flowmeter in the Oil and Gas Industry.
SPE Annual Technical Conference and Exhibition. Houston, Texas: Society of
Petroleum Engineers.
Osán, T. M., Ollé, J. M., Carpinella, M., Cerioni, L. M. C., Pusiol, D. J., Appel, M.,
Freeman, J. & Espejo, I. 2011. Fast measurements of average flow velocity by low-
field 1H NMR. Journal of Magnetic Resonance, 209, 116-122.
Packard, M. & Varian, R. 1954. Free nuclear induction in the Earth's magnetic field.
Physical Review, 93, 1954.
Parent, A., Morin, M. & Lavigne, P. 1992. Propagation of super-Gaussian field
distributions. Optical and Quantum Electronics, 24, S1071-S1079.
Pearce, K. L., Rosenvold, K., Andersen, H. J. & Hopkins, D. L. 2011. Water distribution
and mobility in meat during the conversion of muscle to meat and ageing and the
impacts on fresh meat quality attributes—A review. Meat science, 89, 111-124.
Pedrotti, F. L., Pedrotti, L. M. & Pedrotti, L. S. 2017. Introduction to Optics, Cambridge
University Press.
Pendlebury, J. M., Smith, K., Unsworth, P., Greene, G. L. & Mampe, W. 1979. Precision
field averaging NMR magnetometer for low and high fields, using flowing water.
Review of Scientific Instruments, 50, 535-540.
Perles, C. E., Volpe, P. L. O. & Bombard, A. J. F. 2012. Study of the Cation and Salinity
Effect on Electrocoalescence of Water/Crude Oil Emulsions. Energy & Fuels, 26,
6914-6924.
160
Peshkovsky, A., Forguez, J., Cerioni, L. & Pusiol, D. 2005. RF probe recovery time
reduction with a novel active ringing suppression circuit. Journal of Magnetic
Resonance, 177, 67-73.
Piela, K., Delfos, R., Ooms, G., Westerweel, J. & Oliemans, R. V. A. 2008. On the phase
inversion process in an oil–water pipe flow. International Journal of Multiphase
Flow, 34, 665-677.
Pinguet, B., Gaviria, F., Kemp, L., Graham, J., Coulder, C., Damas, C. & Ben Relem, K.
2010. First Ever Complete Evaluation of a Multiphase Flow Meter in SAGD and
Demonstration of the Performance Against Conventional Equipment. International
North Sea Flow Measurement Workshop. St Andrews.
Pinguet, B., Roux, G. & Hopman, N. S. 2006. Field Experience in Multiphase Gas-Well
Testing: The Benefit of the Combination of Venturi and Multienergy Gamma Ray.
SPE Annual Technical Conference and Exhibition. San Antonio: Society of
Petroleum Engineers.
Pinguet, B. G. 2010. Fundamentals of Multiphase Metering, Schlumberger.
Planinsic, G., Stepisink, J. & Kos, M. 1994. Relaxation-Time Measurement and Imaging in
the Earth′s Magnetic-Field. Journal of Magnetic Resonance, Series A, 110, 170-174.
Prud’homme, R. 2010. Multiphase Flow Concepts. Flows of Reactive Fluids. Boston:
Birkhäuser Boston.
Purcell, E. M., Torrey, H. C. & Pound, R. V. 1946. Resonance Absorption by Nuclear
Magnetic Moments in a Solid. Physical Review, 69, 37-38.
Rabi, I. I., Zacharias, J. R., Millman, S. & Kusch, P. 1938. A New Method of Measuring
Nuclear Magnetic Moment. Physical Review, 53, 318-318.
Raich, H. & Blümler, P. 2004. Design and construction of a dipolar Halbach array with a
homogeneous field from identical bar magnets: NMR Mandhalas. Concepts in
Magnetic Resonance Part B: Magnetic Resonance Engineering, 23B, 16-25.
Rajan, V., Ridley, R. & Rafa, K. 1993. Multiphase flow measurement techniques - A
review. Journal of Energy Resources Technology, 115.
Research and Markets 2016. Intelligent Flow Meter Market by Type, Communication
Protocol, Application, and Geography - Global Trend & Forecast to 2022. In:
Markets, R. A. (ed.) Automation and Process Control.
Robinson, J. N., Coy, A., Dykstra, R., Eccles, C. D., Hunter, M. W. & Callaghan, P. T.
2006. Two-dimensional NMR spectroscopy in Earth’s magnetic field. Journal of
Magnetic Resonance, 182, 343-347.
Rodriguez, I. H., Yamaguti, H. K., De Castro, M. S., Da Silva, M. J. & Rodriguez, O. M.
2011. Slip ratio in dispersed viscous oil–water pipe flow. Experimental Thermal
and Fluid Science, 35, 11-19.
161
Rodriguez, O. M. H. & Oliemans, R. V. A. 2006. Experimental study on oil–water flow in
horizontal and slightly inclined pipes. International Journal of Multiphase Flow, 32,
323-343.
Ross, A. & Stobie, G. Well testing—an evaluation of test separators and multiphase flow
meters International North Sea Flow Measurement Workshop, 2010 St Andrews.
Ross, J., Holland, D., Blake, A., Sederman, A. & Gladden, L. 2012. Extending the use of
Earth’s Field NMR using Bayesian methodology: application to particle sizing.
Journal of Magnetic Resonance, 222, 44-52.
Sætre, C., Johansen, G. A. & Tjugum, S. A. 2010. Salinity and flow regime independent
multiphase flow measurements. Flow Measurement and Instrumentation, 21, 454-
461.
Sætre, C., Johansen, G. A. & Tjugum, S. A. 2012. Tomographic multiphase flow
measurement. Applied Radiation and Isotopes, 70, 1080-1084.
Salgado, C. M., Pereira, C. M. N. A., Schirru, R. & Brandão, L. E. B. 2010. Flow regime
identification and volume fraction prediction in multiphase flows by means of
gamma-ray attenuation and artificial neural networks. Progress in Nuclear Energy,
52, 555-562.
Sanderson, M. L. & Yeung, H. 2002. Guidelines for the use of ultrasonic non-invasive
metering techniques. Flow Measurement and Instrumentation, 13, 125-142.
Sarkodie, K., Fergusson-Rees, A. & Diaz, P. 2018. A review of the application of non-
intrusive infrared sensing for gas–liquid flow characterization. The Journal of
Computational Multiphase Flows, 10, 43-56.
Scheers, A. M. & Noordhuis, B. R. 1999. Multi-phase and Wet Gas Flow Measurement.
5th Annual Multi-Phase Metering Conference. Aberdeen.
Scheers, A. M. & Slijkerman, W. F. J. 1996. Multiphase Flow Measurement Using
Multiple Energy Gamma Ray Absorption (MEGRA) Composition Measurement.
SPE Annual Technical Conference and Exhibition. Denver, Colorado: Society of
Petroleum Engineers.
Schramm, L. L. 2006. Emulsions, Foams, and Suspensions: Fundamentals and
Applications, Weinheim, Wiley.
Seymour, J. D. & Callaghan, P. T. 1997. Generalized approach to NMR analysis of flow
and dispersion in porous media. AIChE Journal, 43, 2096-2111.
Sheng, J. 2010. Modern Chemical Enhanced Oil Recovery: Theory and Practice, Elsevier
Science.
Shercliff, J. A. 1962. The theory of electromagnetic flow-measurement, Cambridge,
Cambridge University Press.
162
Shushakov, O. A. 1996. Groundwater NMR in conductive water. Geophysics, 61, 998-
1006.
Silva Elipe, M. V. & Milburn, R. R. 2016. Monitoring chemical reactions by low-field
benchtop NMR at 45 MHz: pros and cons. Magnetic Resonance in Chemistry, 54,
437-443.
Singer, J. 1959. Blood flow rates by nuclear magnetic resonance measurements. Science,
130, 1652-1653.
Sjoblom, J. 2001. Encyclopedic handbook of emulsion technology, CRC Press.
Slichter, C. P. 1996. Principles of Magnetic Resonance, Springer Berlin Heidelberg.
Song, Y.-Q. 2009. A 2D NMR method to characterize granular structure of dairy products.
Progress in Nuclear Magnetic Resonance Spectroscopy, 55, 324-334.
Song, Y.-Q. 2010. Recent progress of nuclear magnetic resonance applications in
sandstones and carbonate rocks. Vadose Zone Journal, 9, 828-834.
Song, Y.-Q. 2013. Magnetic Resonance of Porous Media (MRPM): A perspective. Journal
of Magnetic Resonance, 229, 12-24.
Song, Y. Q., Venkataramanan, L., Hürlimann, M. D., Flaum, M., Frulla, P. & Straley, C.
2002. T1–T2 Correlation Spectra Obtained Using a Fast Two-Dimensional Laplace
Inversion. Journal of Magnetic Resonance, 154, 261-268.
Speight, J. G. 2011. An Introduction to Petroleum Technology, Economics, and Politics,
Wiley.
Stapf, S. & Han, S. I. 2006. NMR Imaging in Chemical Engineering, Wiley.
Stepišnik, J., Eržen, V. & Kos, M. 1990. NMR imaging in the earth's magnetic field.
Magnetic Resonance in Medicine, 15, 386-391.
Steven, R. 2008. A dimensional analysis of two phase flow through a horizontally installed
Venturi flow meter. Flow Measurement and Instrumentation, 19, 342-349.
Steven, R. N. 2002. Wet gas metering with a horizontally mounted Venturi meter. Flow
Measurement and Instrumentation, 12, 361-372.
Stewart, M. & Arnold, K. 2008. Emulsions and Oil Treating Equipment: Selection, Sizing
and Troubleshooting, Elsevier Science.
Sukstanskii, A. L. & Yablonskiy, D. A. 2001. Theory of FID NMR Signal Dephasing
Induced by Mesoscopic Magnetic Field Inhomogeneities in Biological Systems.
Journal of Magnetic Resonance, 151, 107-117.
Sun, J.-Y. & Jepson, W. P. 1992. Slug flow characteristics and their effect on corrosion
rates in horizontal oil and gas pipelines. SPE Annual Technical Conference and
Exhibition, 1992 Washington DC, USA. Society of Petroleum Engineers.
163
Syre, B., Andersen, O. & Fosse, S. 2013. Using Multiphase Meters for Fiscal purposes - A
Field Experience. International North Sea Flow Measurement Workshop. Tonsberg.
Szilágyi, L. 1995. Chemical shifts in proteins come of age. Progress in Nuclear Magnetic
Resonance Spectroscopy, 27, 325-442.
Tadros, T. F. 2013. Emulsion Formation and Stability, Wiley.
Taitel, Y., Bornea, D. & Dukler, A. 1980. Modelling flow pattern transitions for steady
upward gas‐liquid flow in vertical tubes. AIChE Journal, 26, 345-354.
Taitel, Y. & Dukler, A. 1976. A model for predicting flow regime transitions in horizontal
and near horizontal gas‐liquid flow. AIChE journal, 22, 47-55.
Tayler, A. B., Holland, D. J., Sederman, A. J. & Gladden, L. F. 2012. Applications of ultra-
fast MRI to high voidage bubbly flow: measurement of bubble size distributions,
interfacial area and hydrodynamics. Chemical Engineering Science, 71, 468-483.
Thorn, R., Johansen, G. & Hammer, E. 1997. Recent developments in three-phase flow
measurement. Measurement Science and Technology, 8, 691.
Thorn, R., Johansen, G. A. & Hammer, E. A. Three-Phase Flow Measurement in the
Offshore Oil Industry: Is There a Place for Process Tomography? First World
Congress on Industrial Process Tomography, 1999 Buxton. 14-17.
Thorn, R., Johansen, G. A. & Hjertaker, B. T. 2013. Three-phase flow measurement in the
petroleum industry. Measurement Science and Technology, 24, 012003.
Thybo, A. K., Andersen, H. J., Karlsson, A. H., Dønstrup, S. & Stødkilde-Jørgensen, H.
2003. Low-field NMR relaxation and NMR-imaging as tools in differentiation
between potato sample and determination of dry matter content in potatoes. LWT -
Food Science and Technology, 36, 315-322.
Tikhonov, A. N., Goncharsky, A., Stepanov, V. V. & Yagola, A. G. 1995. Numerical
methods for the solution of ill-posed problems, Dordrecht, Kluwer Academic
Publishers.
Tjugum, S. A., Hjertaker, B. T. & Johansen, G. A. 2002. Multiphase flow regime
identification by multibeam gamma-ray densitometry. Measurement Science and
Technology, 13, 1319.
Trallero, J. L., Sarica, C. & Brill, J. P. 1997. A Study of Oil-Water Flow Patterns in
Horizontal Pipes. SPE Production & Facilities, 12, 165-172.
US Energy Information Administration 2017. International Energy Outlook. Washington
D.C.
Van Landeghem, M., Haber, A., D'espinose De Lacaillerie, J.-B. & Blümich, B. 2010.
Analysis of multisite 2D relaxation exchange NMR. Concepts in Magnetic
Resonance Part A, 36A, 153-169.
164
Vander, H. W., Genthe, W., Battocletti, J., Mccormick, W. & Snowball, H. 1968. NMR
applied to flow measurement. Instrumentation Technology.
Vashaee, S., Li, M., Newling, B., Macmillan, B., Marica, F., Kwak, H. T., Gao, J., Al-
Harbi, A. M. & Balcom, B. J. 2018. Local T1-T2 distribution measurements in
porous media. Journal of Magnetic Resonance, 287, 113-122.
Venkataramanan, L., Yi-Qiao, S. & Hurlimann, M. D. 2002. Solving Fredholm integrals of
the first kind with tensor product structure in 2 and 2.5 dimensions. IEEE
Transactions on Signal Processing, 50, 1017-1026.
Villarreal, J., Laverde, D. & Fuentes, C. 2006. Carbon-steel corrosion in multiphase slug
flow and CO2. Corrosion Science, 48, 2363-2379.
Vogt, S. J., Sanderlin, A. B., Seymour, J. D. & Codd, S. L. 2013. Permeability of a growing
biofilm in a porous media fluid flow analyzed by magnetic resonance displacement‐
relaxation correlations. Biotechnology and bioengineering, 110, 1366-1375.
Vold, R. L., Waugh, J. S., Klein, M. P. & Phelps, D. E. 1968. Measurement of spin
relaxation in complex systems. The Journal of Chemical Physics, 48, 3831-3832.
Wagner, L., Kalli, C., Fridjonsson, E. O., May, E., Zhen, J. & Johns, M. 2018.
Simultaneous quantification of aliphatic and aromatic hydrocarbons in produced
water analysis using mobile 1H NMR. Measurement Science and Technology, 29,
085501.
Wagner, L., Kalli, C., Fridjonsson, E. O., May, E. F., Stanwix, P. L., Graham, B. F.,
Carroll, M. R. J. & Johns, M. L. 2016. Quantitative produced water analysis using
mobile 1H NMR. Measurement Science and Technology, 27, 105501.
Wahba, G. 1977. Practical approximate solutions to linear operator equations when the data
are noisy. SIAM Journal on Numerical Analysis, 14, 651-667.
Wahba, G. 1982. Constrained regularization for ill posed linear operator equations, with
applications in meteorology and medicine. In: Gupta, S. S. & Berger, J. O. (eds.)
Statistical Decision Theory and Related Topics III. London: Academic Press.
Wang, J., Tian, G. & Lucas, G. 2007. Relationship between velocity profile and distribution
of induced potential for an electromagnetic flow meter. Flow Measurement and
Instrumentation, 18, 99-105.
Wang, X., Brandvik, A. & Alvarado, V. 2010. Probing Interfacial Water-in-Crude Oil
Emulsion Stability Controls Using Electrorheology. Energy & Fuels, 24, 6359-
6365.
Washburn, K. E. & Callaghan, P. T. 2006. Tracking Pore to Pore Exchange Using
Relaxation Exchange Spectroscopy. Physical Review Letters, 97, 175502.
Waters, G. S. 1955. A Measurement of the Earth's Magnetic Field by Nuclear Induction.
Nature, 176, 691.
165
Westbrook, C. & Talbot, J. 2018. MRI in Practice, Wiley.
Wu, H., Zhou, F. & Wu, Y. 2001. Intelligent identification system of flow regime of oil–
gas–water multiphase flow. International Journal of Multiphase Flow, 27, 459-475.
Wu, Z., Mccann, H., Davis, L. E., Hu, J., Fontes, A. & Xie, C. G. 2009. Microwave-
tomographic system for oil- and gas-multiphase-flow imaging. Measurement
Science and Technology, 20, 104026.
Xie, C. G. 2007. Measurement Of Multiphase Flow Water Fraction And Water‐cut. AIP
Conference Proceedings, 914, 232-239.
Xu, J.-Y., Wu, Y.-X., Feng, F.-F., Chang, Y. & Li, D.-H. 2008. Experimental investigation
on the slip between oil and water in horizontal pipes. Experimental Thermal and
Fluid Science, 33, 178-183.
Xu, L., Zhou, W., Li, X. & Wang, M. 2011. Wet-gas flow modeling for the straight section
of throat-extended venturi meter. IEEE transactions on Instrumentation and
Measurement, 60, 2080-2087.
Xu, X.-X. 2007. Study on oil–water two-phase flow in horizontal pipelines. Journal of
Petroleum Science and Engineering, 59, 43-58.
Yadigaroglu, G. & Hewitt, G. F. 2017. Introduction to Multiphase Flow: Basic Concepts,
Applications and Modelling, Springer International Publishing.
Yao, H.-Y., Li, Q.-P. & Wang, T. 2009. An Experimental Investigation On Pressure
Gradients In Horizontal Heavy Oil-Water Pipe Flows. The Nineteenth International
Offshore and Polar Engineering Conference. Osaka, Japan: International Society of
Offshore and Polar Engineers.
Yoder, J., Burum, B., Weeks, W., Lund, H., Buchanan, B. & Goldstein, D. 2016. The
World Market for Multiphase Flowmeters. In: Flow Research, I. (ed.). Wakefield.
Zabaras, G. J. 2000. Prediction of Slug Frequency for Gas/Liquid Flows. SPE Journal, 5,
252-258.
Zagarola, M. V., Perry, A. E. & Smits, A. J. 1997. Log laws or power laws: The scaling in
the overlap region. Physics of Fluids, 9, 2094-2100.
Zhen, J. Z., O'Neill, K. T., Fridjonsson, E. O., Stanwix, P. L. & Johns, M. L. 2018. A
resistive Q-switch for low-field NMR systems. Journal of Magnetic Resonance,
287, 33-40.
Zhong, X., Wu, Y., Li, D., Li, Z. & Li, Q. 2007. Effect of Slug Flow on Pipe Vibration in
Horizontal Pipeline. 5th International Symposium on Measurement Techniques for
Multiphase Flows, 2007 Macau, China. 537-541.
Zhu, Z. & Howe, D. 2001. Halbach permanent magnet machines and applications: a
review. IEE Proceedings-Electric Power Applications, 148, 299-308.
166
Zolfaghari, R., Fakhru’l-Razi, A., Abdullah, L. C., Elnashaie, S. S. E. H. & Pendashteh, A.
2016. Demulsification techniques of water-in-oil and oil-in-water emulsions in
petroleum industry. Separation and Purification Technology, 170, 377-407.
167
Appendices
Appendix A. Calibrations for NMR flow measurements
A.1 Magnetic field inhomogeneity
Experimental measurements acquire the free induction decay (FID) signal from fluids
flowing through the flow metering system. The FID signal obtained from a flowing medium
consists of two components; (i) stationary FID behaviour, and (ii) a ‘flush-out’ effect as
excited fluid leaves the detection volume according to fluid velocity. The flush-out effect
has been addressed in section 4.2.2; further detail regarding the stationary fluid behaviour is
provided here.
The stationary FID consists of T2 decay (due to spin-spin relaxation of the fluid) and
signal dephasing due to magnetic field inhomogeneities. For the single-phase water flow
measurements conducted in Chapter 4; capturing a FID of water in the pipe is sufficient to
calibrate the stationary FID behaviour. However for the two-phase oil/water measurements
conducted in Chapter 6; the impact of spin-spin relaxation on the stationary FID behaviour
cannot be calibrated as the fluid composition is unknown. Therefore the spin-spin relaxation
and magnetic field inhomogeneity decay effects must be segregated and accounted for
separately. The FID signal was previously given in equation 2.14; this can be considered in
terms of the two relevant components as follows;
𝑆(𝑡) = 𝑆0exp (−𝑡
𝑇2) exp(−(𝑅𝐼𝑡)2) (A.1)
where the first component (exp(-t/T2)) describes spin-spin relaxation behaviour and the
second component (exp(-(RI t)2)) describes magnetic field inhomogeneity decay behaviour in
the Earth’s magnetic field (with RI describing the Gaussian relaxation decay rate).
Measurement of the magnetic field inhomogeneity function is achieved by
comparison of a FID measurement to a standard CPMG measurement for a given sample.
The T2 decay component of the FID can be measured via a CPMG experiment; using spin
echoes to refocus spins which have been dephased by field inhomogeneity (Carr and Purcell,
1954, Meiboom and Gill, 1958). A measured CPMG signal decay (SCPMG) is described by;
168
𝑆𝐶𝑃𝑀𝐺(𝑇𝐸) = 𝑆0𝑒−
𝑇𝐸𝑇2 (A.2)
where S0 is the initial CPMG polarisation and TE is the echo time of the acquired signals. By
substituting equation A.2 into equation A.1 and performing a simple rearrangement, the field
inhomogeneity function may be described by;
𝐼(𝑡) = exp(−𝑅𝐼𝑡2) = 𝑆𝐹𝐼𝐷(𝑡)
𝑆𝐶𝑃𝑀𝐺 (A.3)
The field inhomogeneity function is determined from measurement of a stationary water
sample (in the pipeline); a CPMG measurement is acquired with 64 echoes at 100 ms spacing
whilst a FID measurement is obtained for an acquisition time of 819.2 ms. Both
measurements use Nscans = 32 and tpolz = 5 s. A T2 distribution is fit via the CPMG model
equation A.2 to the measured CPMG signal decay using 1D Tikhonov regularisation. The
FID (with the T2 decay calibrated according to the measured CPMG) is then fit with equation
A.1 using the Gaussian relaxation rate RI as a fitting constant, and thus the field
inhomogeneity function is determined. Figure A.1 shows the results of field inhomogeneity
function determination; including the measured CPMG signal and model fit, the measured
FID signal and model fit as well as the determined inhomogeneity distribution. Signals are
normalised to the back-extrapolated fit at zero time. Note that only the first 10 echoes are
shown for the CPMG decay for clarity of the other plots; however the full CPMG
measurement is repeated in Figure B.1(a).
169
Figure A.1 NMR water measurements used for field inhomogeneity function determination. A
CPMG measurement fit with a T2 distribution via equation A.2, a FID measurement fit with equation
A.1 and the resultant inhomogeneity function.
Figure A.1 illustrates the half-Gaussian line shape for the inhomogeneity distribution
and demonstrates that the model decay for the FID (according to equation A.1) fits reasonably
well to the measured data. The Gaussian relaxation rate is determined as 0.973 s-2 from the
model fit to the measured FID.
The application of the field inhomogeneity function to the measurement of flowing
data is complicated by local and temporal variation in Earth’s magnetic field. The proposed
flow measurement procedure applies variable polarisation-detection separation distances in
order to observe signal contrast according to T1. The movement of the Halbach array is
observed to interfere with the field inhomogeneity. In order to partially alleviate this
problem, the EFNMR detection coil is shimmed at each of the separation distances. To fully
account for the variations in the field inhomogeneity caused by shifting the Halbach array;
the field inhomogeneity function must be measured at each relevant separation distance. This
is demonstrated in Figure A.2; FID measurements are acquired with stationary water in the
pipe, with each FID signal fit with equation A.1 in order to determine the relevant field
inhomogeneity function for the logarithmically spaced separation distances of 45 – 150 cm.
Signals are normalised to the maximum value obtained from the back-extrapolated fit at zero
time.
0
0.2
0.4
0.6
0.8
1
1.2
0 0.2 0.4 0.6 0.8 1
No
rma
lis
ed
NM
R S
Ign
al
Time since excitation pulse [s]
Measured CPMG Model Fit to CPMGMeasured FID Model Fit to FIDInhomogeneity function
170
Figure A.2 Experimentally acquired FID signals at varying separation distances of 45 – 150 cm.
FID measurements are acquired with stationary water in the pipe running through the EFNMR
system. Signals are fit with the model equation A.1
The results shown in Figure A.2 demonstrate variation in the decay rate of the FID
measurements. The FID signals are relatively similar at separation distances above 62 cm
where the impact of the Halbach is minimal. FID measurements at shorter separation
distances cause more stray field interference with the EFNMR detection region causing an
increase in the observed decay rate (as separation distance is reduced). FID measurements
at separation distances closer than 45 cm begin to show considerably faster decay as the
Halbach magnet interferes with the EFNMR detection coil such that the FID is dominated by
inhomogeneity decay.
A.2 Relaxation in variable magnetic field strengths
The proposed dual polarisation system introduces complication due to the presence
of multiple magnetic field strengths through the NMR measurement system. Figure A.3
illustrates how fluid flowing through the system is subject to three separate magnetic fields
(Larmor frequency (ω0) is indicated for 1H precession); the field of the pre-polarising
Halbach magnet (~300 mT / 12,800 kHz), the Earth’s magnetic field alone (~0.054 mT/2.29
kHz) and the field created by the electromagnetic pre-polarising coil (EPPC, 18.7 mT, 920
0
0.2
0.4
0.6
0.8
1
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
NM
R S
ign
al [µ
V]
Time since excitation pulse [s]
LPD
[cm]45485154586266707580859096
103109116124132141150
Expt. data
Model fit
171
kHz). Following application of the EPPC, the coil is switched off and fluid will return to the
Earth’s magnetic field for excitation and detection.
Figure A.3 Illustration of the three separate magnetic field strengths present in the flow metering
system.
The spin-lattice relaxation rate (T1) of a fluid changes with field strength and therefore
fluid flowing through the system will experience variable T1 relaxation throughout different
sections of the system (Graf et al., 1980, Korb and Bryant, 2002). It is assumed that the T1
experienced due to the Halbach array’s magnetic field is the same as that measured for the
EPPC field. Despite the significant difference in field strength (300 mT for the Halbach
compared to 21.7 mT for the electromagnetic coil) this assumption appears reasonable; the
T1 of water is constant for higher field strengths (T1 of pure water at 250C is 2.89 s for B0 >
1.8 mT) (Graf et al., 1980) and whilst there is no data (to the author’s knowledge) of the T1
of canola oil at changing field strength, the T1 of olive oil is relatively constant between the
two field strengths (T1 of ~110 ms at 19 mT and T1 of ~125 ms at 300 mT) (Alessandri et al.,
2005).
In order to account for variable T1 relaxation, stationary fluid T1 measurements are
made at two different field strengths; in the Earth’s field and in the polarising field. T1
relaxometry measurements in the Earth’s magnetic field were previously presented in section
6.2.3 using an excitation delay sequence. The measurements demonstrated mono-
exponential relaxation behaviour for water with a T1,EF (the spin-lattice relaxation rate in the
earth’s magnetic field) of 3.1 s, whilst canola oil observes bi-exponential relaxation
behaviour with a log-mean T1 of 115 ms.
Pre-polarising Halbach Array (~300 mT / 12,800 kHz)
Intermediate section: Earth’s field (~0.054 mT / 2.29 kHz) Electromagnetic
pre-polarisation coil (18.7 mT / 920 kHz)
Detection: Earth’s field (~0.054 mT /
2.29 kHz)
172
A saturation recovery measurement is applied in order to measure the T1 for canola
oil and water in the polarising field. FID signals are acquired 64 polarisation times (i.e. tpolz)
which are logarithmically spaced from 10 ms – 10 s. The polarisation-excitation delay is
held constant (at the minimum of 60 ms). FID signals are back-extrapolated to determine the
pre-polarised signal (SP). The measured polarised signal is modelled using the following
equation;
𝑆𝑃(𝑡𝑝𝑜𝑙𝑧, 𝑇1) = 𝑆0 ൬1 − exp ൬−𝑡𝑝𝑜𝑙𝑧
𝑇1,𝑃𝐹൰൰ (A.4)
where S0 is the effective NMR signal after an infinite polarisation time in the electromagnetic
coil and T1,PF is the fluid spin-lattice relaxation constant in the polarising field. The
measurement is observing fluid magnetisation development in the polarising field, thus is
measuring T1 in the polarising field. A T1,PF distribution is fit with equation A.4 to the
measured polarised signal using the Tikhonov regularisation algorithm for both canola oil
and water. Figure A.4(a) shows the model signal fit to experimental data and Figure A.4(b)
shows the resultant probability distributions for T1,PF.
Figure A.4 Measuring T1 in the EPPC polarising field. (a) Regularised signal fit of equation A.4 to
measured signal data at variable polarisation times for canola oil and water. (b) Resultant T1,PF
probability distributions (P(T1,PF)) for both canola oil and water.
The polarising field T1 measurements show very similar results to the Earth’s field T1
measurement; with marginally higher relaxation rates. Water exhibits a mono-exponential
decay for T1,PF of 3.3 s, whilst the canola oil displays bi-exponential behaviour with the first
peak representing 62% of the sample with a log mean T1 of 71 ms and the second peak
representing 38% with a log mean T1 of 250 ms. The difference in the final polarised signal
0
20
40
60
80
100
0.01 0.1 1 10
NM
R S
ign
al
[µV
]
Polarisation time [s]
Water
Expt.data
Model fit
(a)
Oil
0
1
2
3
4
5
0.01 0.1 1 10
P(T
1,P
F)
T1,PF [s]
WaterOil
(b)
173
(95.0 µV for water and 51.9 µV for canola oil) can be attributed to the initial signal delay (60
ms) between polarisation and excitation.
In order to account for the complex relaxation behaviour in multiple fields during
flowing measurements, a ‘field calibration ratio’ is used. The model for NMR signal of a
flowing fluid (i.e. equation 6.12) is modified to account for the relevant T1 value in each field,
i.e. T1,PF is used for signal polarisation in the Halbach, whilst T1,EF is used for intermediate
signal decay in the Earth’s magnetic field. Hence equation 6.12 becomes;
𝑆(𝐿𝑃𝐷 , 𝑡𝑝𝑜𝑙𝑧, 𝑡𝑒 , 𝑣, 𝑇1,𝐸𝐹) = 𝑆0𝐻 (1 − 𝑒−
𝐿𝑃𝐻𝑣𝑇1,𝑃𝐹) (𝑒
−𝐿𝑃𝐷
𝑣𝑇1,𝐸𝐹) ൬1 −𝑡𝑒𝑣
𝐿𝐷൰ 𝑒
−൬𝑡𝑒∗𝑅𝑇𝑇1,𝐸𝐹
+(𝑅𝐼𝑡𝑒)2൰
for 𝑡𝑒 ≤𝐿𝐷
𝑣 (A.5)
however considering the measurement system uses T1 to differentiate oil and water; only one
T1 value should be used. The detection coil operates in the Earth’s magnetic field; thus it is
more relevant to measure T1,EF for fluid differentiation. The field calibration ratio; RF =
T1,PF/T1,EF is introduced and T1,PF is replaced;
𝑆(𝐿𝑃𝐷 , 𝑡𝑝𝑜𝑙𝑧, 𝑡𝑒 , 𝑣, 𝑇1,𝐸𝐹) = 𝑆0 (1 − 𝑒−
𝐿𝑃𝐻𝑣 𝑅𝐹𝑇1,𝐸𝐹) (𝑒
−𝐿𝑃𝐷
𝑣𝑇1,𝐸𝐹) (1 −𝑡𝑒𝑣
𝐿𝐷) 𝑒
−൬𝑡𝑒∗𝑅𝑇𝑇1,𝐸𝐹
+(𝑅𝐼𝑡𝑒)2൰
for 𝑡𝑒 ≤𝐿𝐷
𝑣 (A.6)
Incorporating the calibration ratio introduces additional measurement complexity, however
it can successfully quantify for the multi-relaxation behaviour at different field levels. The
field calibration ratio is not constant; the T1 measurements demonstrate that the different
fluids exhibit different changes of T1 between fields. Therefore separate field calibration
ratios are used for oil and water. The field calibration ratios are defined about the oil/water
T1 cut-off (T1,C = 600 ms, as discussed in section); for oil (T1,EF < 600 ms), RF = 1.08
(averaged for both canola oil components) and for water (T1,EF ≥ 600 ms), RF = 1.06.
A.3 Signal diminishment during repolarisation
During NMR measurements of flowing streams, an unexpected signal drop was
observed with the introduction of the electromagnetic pre-polarising coil. This effect is
174
demonstrated in Figure A.5; FID measurements of single-phase water flows are acquired at
three different velocities (0.37, 0.73 and 1.10 m/s) with two FIDs acquired for each velocity.
The first FID acquired is with only Halbach polarisation (at a separation distance LPD = 70
cm) – i.e. no re-polarisation at the EPPC is applied. The second FID acquired has the same
Halbach polarisation (LPD = 70 cm), however EPPC is switched on (polarising current of 6
A) for just 1 ms. Each FID measurement uses 64 scan averages.
Figure A.5 Demonstration of the signal diminishment with introduction of the EPPC. FID
measurements are shown for three different velocities (0.37, 0.73 and 1.10 m/s) with two FIDs
acquired for each velocity; one with only Halbach polarisation (no EPPC re-polarisation indicated by
diamonds) and the second with the EPPC switched on for 1 ms (with EPPC re-polarisation indicated
by crosses).
The FID signals were anticipated to have negligible difference (with and without
using the EPPC) considering the minimal polarisation time of 1 ms should result in negligible
polarisation of water molecules (water T1,PF was measured as 3.3 s). However Figure A.5
shows the initial FID signal amplitude for the measurements with the EPPC switched on for
1 ms is observed to drop relative to the FID signal without the EPPC.
In order to further understand the observed signal diminishment, similar experiments
are repeated for a water flows at a range of velocities (0.18 – 1.83 m/s) and with different
polarisation-excitation delays (tPE) of 60 – 400 ms. The variable polarisation-excitation delay
is applied in order to examine the variation in signal diminishment with time, without
influencing the fluid polarisation (which would occur if the polarisation time was increased).
0
10
20
30
40
50
0 0.1 0.2 0.3 0.4 0.5
NM
R S
ign
al [µ
V]
Time since excitation [s]
Velocity [m/s]
0.37
0.73
1.10
No EPPC re-polz.
With EPPC for 1 ms
175
For each velocity and delay time, two FIDs are measured (as in Figure A.5); the first without
the EPPC and the second with the EPPC switched on for 1 ms. The measured FID signals
are fit with velocity probability distributions (according to the methodology presented in
section 4.2.3) and the initial FID signal amplitude (SI) is determined by back-extrapolation
of the model fit such that time since excitation pulse is zero (i.e. te = 0). The signal
diminishment is quantified by the ratio of the FID signal amplitude with the EPPC switched
on (SI,EPPC) to the FID signal amplitude with only the Halbach (SI,Hal);
𝑆𝑑𝑖𝑚 =𝑆𝐼,E𝑃𝑃𝐶
𝑆𝐼,𝐻𝑎𝑙 (A.7)
Figure A.6 illustrates the signal diminishment at ten different water velocities (0.18 – 1.83
m/s) and four polarisation-excitation delays (60, 100, 200 and 400 ms).
Figure A.6 Experimental measurement of the signal diminishment as a function of velocity and the
polarisation-excitation delay (tPE).
The signal diminishment presented in Figure A.6 can be considered in terms of a
piecewise linear function of velocity and time segregated into three components. The first
piecewise component involves the signal ratio linearly decreasing as a function of velocity
(e.g. tPE = 60 ms, a linear decrease is observed for v < 1.5 m/s) until a transition point; this is
indicative of a flush-out effect from the detector region. Note that this component is not
observable for tPE = 400 ms as flush out has already occurred during the polarisation-
excitation delay. The transition point separates the first and second piecewise components
0.6
0.7
0.8
0.9
1
1.1
1.2
0 0.5 1 1.5 2
Sd
im
Velocity [m/s]
tPE
[ms]60100200400
176
and changes with the polarisation-excitation delay (e.g. for tPE = 60 ms, the transition velocity
is ~1.5 m/s, whilst for tPE = 100 ms, the transition velocity is ~0.9 m/s). The second piecewise
component observes the signal ratio linearly increasing with respect to velocity; this is
indicative of a flush-in effect into the detector region. The third and final component of the
piecewise function is a constant function of Sratio = 1; indicating that the diminishment is no
longer in effect and the FID signal amplitude with the EPPC switched on is the same as the
FID signal amplitude with only the Halbach. The suggested cause and a model to describe
observed signal variation is now considered.
The Magritek EFNMR system utilises a pre-polarising coil in order to achieve
reasonable signal sensitivity due to the increased polarising field strength (~18.7 mT
compared the Earth’s Field strength of ~ 54 μT) (Appelt et al., 2006, Blümich et al., 2009a,
Halse and Callaghan, 2008). One of the difficulties in using the pre-polarising field is
ensuring that the full pre-polarised magnetisation can be detected by allowing an adiabatic
field discharge (Conradi et al., 2017). The field discharge from the polarising field strength
to Earth’s field strength must be sufficiently slow to ensure an adiabatic field transition (e.g.
an adiabatic transition from a field of 0.05 T to Earth’s field requires a few milliseconds)
(Mohoric and Stepisnik, 2009). Whilst the transition is adiabatic for “switching off” (field
discharge) the pre-polarising coil, the transition is non-adiabatic for “switching on” the pre-
polarising coil. This is not a problem for the majority of measurements conducted using the
Earth’s field system as there is no significant pre-existing polarised signal which can be
disrupted when the polarising coil is switched on. However, the non-adiabatic transition
occurring for the polarisation coil “switch on” is important for the unusual set-up of the
EFNMR flow meter with a dual polarisation system. Fluid entering the EFNMR system will
have previously been magnetised or polarised by the upstream Halbach array and therefore
possess an initial magnetisation. The rate of polarising field accumulation is too fast (i.e. the
time for field “switch-on” is too short) and therefore the flowing fluid experiences a non-
adiabatic transition from Earth’s field to polarising field. This means that the pre-magnetised
signal (due to the Halbach) will not be able to fully re-orientate from the Earth’s field to the
pre-polarising field. Consequently, a signal loss is observable between experiments where
only the Halbach is used and experiments where both the Halbach and EPPC are used (as
177
demonstrated in Figures A.5 and A.6). The effect of the pre-polarising pulse on the
measurable signal in different experiments is illustrated in Figure A.7.
Figure A.7 Illustration of effective signal development during variable polarisation scenarios using
a pulse and collect experiment. (a) Signal variation for stationary fluid measurement (i.e. effectively
using the EPPC alone for pre-polarisation), (b) signal variation for a flowing measurement using the
Halbach alone (i.e. the EPPC is not applied), and (c) signal variation for flowing fluid measurements
Adiabatic “switch-off”
Non-adiabatic “switch-on” r.f. excitation
pulse
Measured FID
Signal polarisation
Polarising pulse
Intermediate decay
(a). Standard pulse and collect for stationary fluid.
Pulse sequence
Effective signal
r.f. excitation pulse
Incoming Halbach signal
(b). Pulse and collect for flowing measurements with Halbach only.
Pulse sequence
Effective signal
Measured FID
Measured FID
Adiabatic “switch-off”
Non-adiabatic “switch-on” r.f. excitation
pulse
Signal polarisation
Polarising pulse
Intermediate decay
(c). Pulse and collect for flowing measurements with Halbach and pre-polarising coil
Incoming Halbach signal
Pulse sequence
Effective signal
Signal loss
178
with the Halbach and the EPPC applied. The signal loss due to non-adiabatic field transition is only
relevant for the final scenario involving the dual polarisation mechanism.
Accounting for the non-adiabatic transition is complicated not only due to the fact
that the quantity of signal loss is unknown, but that the signal loss is a function of fluid
position within the coil. The EPPC and Earth’s field detection coil are both coaxial solenoid
coils and therefore have polarisation and detection field profiles (as a function of position
within the EFNMR apparatus) respectively. The transition from the polarisation field to
Earth’s field will therefore depend on the fluid’s position within the EFNMR apparatus.
Adiabatic transitions are more easily achievable near the centre of the coil where the signal
sensitivity is a maximum (Conradi et al., 2017). The direction of the polarising field (BP)
will nearly match the direction of the horizontal component of the Earth’s magnetic field at
the centre of the coil (Be). The direction of the polarising magnetic field will change as a
function of the position within the solenoid such that regions further away from the centre
will require larger reorientation from the polarising field to the Earth’s field. Therefore the
fluid further away from the centre of the coil will have a lower sensitivity and as a result the
non-adiabatic signal diminishment will be larger. The final consideration is the impact of
fluid outside the polarisation coil region which can flush through during detection. The
aforementioned signal effects are incorporated into the following simplistic model which
quantifies the signal variation due to the non-adiabatic transition. Fluid within the pipeline
is separated into regions of significance based on the fluid position relative to the polarisation
and detection zones. These regions are illustrated in Figure A.8 and a description of the
signal attenuation for each region is provided below the figure.
179
Figure A.8 Illustration of the impact of the non-adiabatic transition on different regions of fluid
within the pipeline. The length of the polarisation zone (LP), detection zone (LD) and intermediate
zones (LI) are indicated.
Figure A.8 highlights three regions of fluid within the pipeline; the initial fluid,
intermediate fluid and outside fluid. The “initial fluid” region (highlighted in blue) is fluid
in the detection region of the coil when the EPPC is switched on. The detection zone (of
length LD = 10 cm) reflects the sensitivity of the detection coil along the axis parallel to the
pipeline. The majority of the fluid in the detection region will experience an adiabatic
transition as the signal sensitivity is strongest towards the centre of the coil. However a full
adiabatic transition between fields is not possible due to the time of coil switch-on, thus a
small fraction of the polarised fluid (kP) will not transition from Earth’s field to the pre-
polarising field. This fraction of fluid which does achieve field transition will not be detected
following the excitation pulse (i.e. there is a partial impact causing signal loss). The
“intermediate fluid” region (highlighted in red) is fluid outside the detection zone of the
EFNMR apparatus but still within the polarisation zone (of length LP = 27 cm) of the EPPC.
This fluid is in two regions either side of the detection zone (both of length LI = 8.5 cm).
Note that the downstream intermediate region will not be detected (as fluid has already
passed through the detection region) however is incorporated into the model for consistency.
This region has significantly poorer sensitivity relative to the detection zone. The impact of
the non-adiabatic transition will result in a more significant fraction of fluid (kI) experiencing
a non-adiabatic transition. The “outside fluid” region (highlighted in green) is not impacted
by the non-adiabatic transition as the fluid is not within the polarising region of the coil. It
is assumed that this fluid which moves from outside the polarising region to within the
LD
LP
LI LI
Flow direction
Pipe
Signal regions Initial fluid – partial impact Intermediate fluid – significant impact Outside fluid – no impact
180
polarising region will effectively experience an adiabatic transition; the EPPC will have a
“polarisation profile” involving changing field strength with length. The velocity of the fluid
is low enough such that the moving fluid will experience a gradual change in field strength
as it moves into the polarising region. There is no signal change of the outside fluid initially,
however this fluid will become important as it moves into the detection zone replacing fluid
which has experienced a signal change.
The effect of the non-adiabatic transition will depend on the type of fluid (i.e. initial,
intermediate or outside) and the relative fraction of each fluid in the detector at a given time.
In order to aid the understanding of the signal loss model Table A.1 provides a description
of position, quantification of the relevant fluid fractions and a visual depiction of the regions
within the detection zone for five different cases at different times since the polarisation
switch on.
181
Table A.1 Movement of the separate signal variation regions at different times since the
polarisation switch on. The fraction of each region (initial, intermediate (interm.) and outside) in the
detection zone as well as a visual depiction of the effective region positions is also shown. The
detection region is indicated by dashed lines.
Table A.1 considers five different cases (A-E) at variable time since the polarisation
switch-on (tps). Case A is the initial position (i.e. tps = 0) where the “initial fluid” alone is
fully inside the detector zone. Note that this case is never applied as there is minimum
polarisation-excitation delay of 60 ms; however it is shown to assist understanding. Case B
shows the “intermediate fluid” flushing into the detection zone, whilst the “initial fluid” is
flushing out of the detector zone. Case C has the “intermediate fluid” fully within the
detection zone; note that this will only apply if LI ≤ LD. The “outside fluid” is beginning to
flush into the detection zone whilst the final fraction of the “initial fluid” is still flushing out
of the detector. In Case D the “initial fluid” has now completely left the detection zone and
the “intermediate fluid” is now flushing out of the detector. Finally in Case E, the
“intermediate fluid” has completely left the detection region and the “outside fluid” will
completely fill the detector. This means that the detector is now filled with fluid which was
Pipe
Detection region Pipe
Detection region Pipe
Detection region Pipe
Detection region
Pipe
Detection region
182
not impacted by the non-adiabatic transition and thus the signal will effectively be the same
as if the fluid was polarised by the Halbach alone.
In order to quantify the signal variation due to non-adiabatic field transition (SNA) at
a given time since polarisation pulse (tps), the following equation is applied;
𝑆𝑁𝐴(𝑡𝑝𝑠) = 𝑥𝑃(1 − 𝑘𝑃) + 𝑥𝐼(1 − 𝑘𝐼) + 𝑥𝑂 (A.8)
where xP is the fraction “initial fluid” within the detection zone, xI is the fraction of
“intermediate fluid” within the detection zone and xO is the fraction of “outside fluid” within
the detection zone. The values of xP, xI and xO are determined by selecting the relevant case
(according to the time since polarisation) from Table A.1, whilst kP and kI are fractions of the
“initial fluid” and “intermediate fluid” which observe non-adiabatic field transitions.
In order to validate the described model and estimate the values of the parameters kP
and kI, the experimentally measured signal diminishment ratios shown in Figure A.6 are fit
with equation A.8 using a least squares fitting method. The effective model fit is shown in
Figure A.9.
183
Figure A.9 Model fit of equation A.8 to experimentally measured signal diminishment ratios (SRatio)
as a function of fluid velocity and time since excitation pulse (tPE).
Figure A.9 shows good agreement between the experimental data and model fit; the
linear flush-in and flush-out effects are well captured by the model. During the application
of flow metering measurements and the signal loss model, the detector profile is described
by a super-Gaussian function (Parent et al., 1992). The super-Gaussian profile captures the
shape of the detection profile introduced by the detection solenoid coil much better than a
uniform profile of length LD. During the calculation of the signal loss model fractions (xP, xI
and xO); the detector profile is used instead of the single parameter LD in order to better
describe the signal loss as a function of fluid position. The super-Gaussian profile introduces
non-linearity into the model-fit which causes the slight curvature of the model-fit. The results
of the least squares model fit return parameter values of kP = 0.085 and kI = 0.253 indicating
8.5 % of the fluid in the “initial region” experiences a non-adiabatic transition, whilst 25.3
% of the fluid in the “intermediate region” experiences a non-adiabatic transition. Note that
in reality, the signal loss fractional parameters (i.e. kP and kI) will instead form a ‘signal loss
profile’ as a function of position with the detector (i.e. k(l) where l represents position within
the detector) and will depend on both the polarisation and detection profiles. However such
a ‘signal loss profile’ is too difficult to measure practically and therefore has been simplified
into the two signal loss parameters (kP and kI) in two discretised regions (the “inside” and
“intermediate” regions).
0.6
0.7
0.8
0.9
1
1.1
1.2
0 0.5 1 1.5 2
SR
ati
o
Velocity [m/s]
Expt data
Model fit
tPE
[ms]60
100200400
184
Appendix B. Fluid relaxometry
B.1 T2 measurements of canola oil and water
Measurements of the fluid T2 distributions (for stationary samples) are conducted in
order to assist understanding of the expected relaxation behaviour of oil and water. The T2
distribution of water is fit to measured CPMG data (64 echoes acquired at 100 ms spacing)
with equation A.2 using Tikhonov regularisation (as previously shown Figure A.1).
Determining the T2 distribution for canola oil is more complicated; due to considerable timing
delays, CPMG experiments in the EFNMR system are limited to echo times greater than 50
ms. Therefore for canola oil, which is anticipated to have a low T2 of ~100 ms (Luthria,
2004) the majority of observable signal decay will have occurred in the first few echoes (~6
echoes). Using such a small number of data points is unsuitable for regularisation analysis
in order to determine a distribution of relaxation parameters (e.g. the bi-exponential
distribution for canola oil) (Mitchell et al., 2014). Thus an alternative measurement and
analysis technique must be considered for determination of the T2 distribution for canola oil
in the EFNMR system.
A FID measurement has a much shorter acquisition delay (9 ms with the Q-switch)
and faster data spacing (dwell time of 50 µs) relative to a CPMG experiment in the EFNMR
system; allowing a reasonable number of points to be acquired for data analysis before the
observable signal decays. The magnetic field inhomogeneity function is first determined by
analysing the FID signal and CPMG measurement for a water sample (as shown in Figure
A.1). A canola oil sample is then placed in the system and a FID signal is acquired (Nscans =
32, tacq = 409.6 ms). The canola oil T2 distribution is fit to the measured FID signal with
equation A.1 using Tikhonov regularisation analysis. Figure B.1 displays the results of fluid
T2 analysis. Figure B.1(a) shows the measured CPMG signal data and corresponding fit for
the water sample as well as the measured FID signal and corresponding model fit for the
canola oil sample. Figure B.1(b) shows the resultant T2 probability distributions determined
via regularisation analysis.
185
Figure B.1 Measuring T2 of oil and water in the EFNMR system. (a) Regularised signal fit of
equation A.2 to measured CPMG signal data for water and the regularised signal fit of equation A.1
to measured FID signal data for canola oil. (b) The corresponding T2 distributions for water and
canola oil.
Figure B.1(b) shows similar results to the T1 measurements previously conducted
(both in the Earth’s field and the polarising field). The water exhibits a mono-exponential
decay with a T2 of 1.96 s. The canola oil displays bi-exponential behaviour with 50 % at 51
ms and 50 % at 139 ms. The difference in the initial signal in Figure B.1(a) can be attributed
to the initial signal delay (60 ms) between polarisation and excitation (the same effect as
observed in Figure A.4(a)). Note that for T2 measurements with lower SNR (e.g. during
flowing measurements), the two peaks can often be merged into a single peak. The overall
distribution (i.e. the bi-exponential distribution in Figure B.1(b)) log-mean T2 of 82 ms is a
more useful comparison to such results (with lower SNR).
The stationary T2 distributions are useful in understanding the expected decay
behaviour during flow; however such distributions will be difficult to measure under flow.
They will be particularly difficult to measure simultaneously with the T1,EF and velocity
distributions; therefore T1/T2 ratios are introduced in order to assist signal interpretation and
inversion. The T1/T2 ratio is defined around the oil/water T1 cut-off of T1,EF = 600 ms; with
oil (T1,EF < 600 ms) having RT = 1.35 (averaged for both canola oil components) and water
(T1,EF ≥ 600 ms) using RT = 1.63.
0
20
40
60
80
100
120
0.01 0.1 1 10
NM
R S
ign
al
[µV
]
Time since excitation [s]
Model fit
(a)Expt.data
Water (CPMG)
Oil (FID)
0
2
4
6
8
10
12
14
0.01 0.1 1 10
P(T
2)
T2 [s]
WaterOil
(b)
186
B.2 Comparative relaxometry measurements
Oil relaxation times (T1 and T2) were measured using an alternative NMR system (the
Magritek 2 MHz Rock Core Analyser (RCA)) in order to give confidence in the observed oil
relaxation behaviour in the EFNMR system. The RCA operates at a much higher frequency
(2 MHz relative to 2.29 kHz for the EFNMR system) and utilises a permanent magnet for the
polarising field (compared to the EPPC utilised in the Earth’s field system). This system
design means that the polarisation-excitation (tPE = 0.1 ms) and polarisation-acquisition delay
(tdelay = 0.1 ms) are much shorter relative to the EFNMR system (tPE = 60 ms and tdelay = 9
ms). The short system delays mean that it is possible to capture a much greater proportion
of the short relaxation behaviour during experimental measurements which gives greater
confidence in the measured relaxation distributions (relative to measurements in the EFNMR
systems). An inversion recovery measurement (Vold et al., 1968) was performed in order to
quantify the oil T1 distribution. The inversion recovery experiment (described in section
2.1.5) was performed with 32 logarithmically spaced inversion times from 0.1 to 5000 ms.
A T1 distribution is fit via the inversion recovery model (equation 2.9) to the measured signal
using Tikhonov regularisation. Figure B.2(a) shows the model signal fit to experimental data
and Figure B.2(b) shows the resultant probability distributions for T1.
Figure B.2 Measuring canola oil T1 in the Magritek 2 MHz Rock Core Analyser. (a) Normalised
signal fit of equation 2.9 to measured signal data for an inversion recovery measurement of canola
oil. (b) Resultant T1 probability distribution (P(T1)) for canola oil.
In Figure B.2(b) the bi-exponential relaxation behaviour is observed for the canola
oil with the first peak representing 60% of the sample with a log mean T1 of 70 ms and the
second peak representing 40% with a log mean T1 of 213 ms. This is consistent with the
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
0.1 1 10 100 1000 10000
No
rmalised
NM
R s
ign
al
[-]
Inversion time [ms]
ExperimentaldataModel fit
(a)
0
0.5
1
1.5
2
2.5
3
3.5
0.01 0.1 1
P(T
1)
T1 [s]
(b)
187
values observed with T1 analysis in the polarising field using the EFNMR system. Marginal
differences are anticipated; the measurement technique is slightly different (e.g. errors
introduced by the 60 ms delay between polarisation and excitation for the measurement using
the EFNMR system).
A CPMG measurement (Carr and Purcell, 1954, Meiboom and Gill, 1958) was
performed in order to quantify the oil T2 distribution. The CPMG measurement consists of
8192 echoes with an echo spacing (TE) of 0.1 ms. The CPMG is processed into 64
logarithmically spaced windows. A T2 distribution is fit via the model equation A.2 to the
measured signal using Tikhonov regularisation.
Figure B.3 Measuring canola oil T1 in the Magritek 2 MHz Rock Core Analyser. (a) Normalised
signal fit of equation A.2 to measured data for a CPMG measurement of canola oil. (b) Resultant T2
probability distribution (P(T2)) for canola oil.
Figure B.3(b) shows the bi-exponential T2 relaxation behaviour of canola oil with the
first peak representing 60% of the sample with a log mean T2 of 67 ms and the second peak
representing 40% with a log mean T2 of 197 ms. This is reasonably consistent with the values
observed with T2 analysis of canola oil using the EFNMR system. Some variation is
anticipated; the Earth’s field measurement utilises an FID measurement calibrated relative to
the inhomogeneity decay function (see appendix A.1 for further detail). Furthermore, the
SNR is higher in the RCA (SNR of 77 in the RCA relative to a SNR of 28 in the EFNMR
system) which allows improved resolution of the relaxation distribution components.
0
0.2
0.4
0.6
0.8
1
0 200 400 600 800 1000
No
rmalised
NM
R s
ign
al
[-]
Echo time [ms]
Experimental data
Model fit
(a)
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
0.01 0.1 1
P(T
2)
T2 [s]
(b)
188
Appendix C. Flow system design considerations
C.1 Separation of oil and water emulsions
The formation of emulsions (either oil-in-water, or water-in-oil) within production
pipelines is an ongoing problem in crude oil production and transportation (Alboudwarej et
al., 2007, Arnold et al.). In this work, canola oil and water flow simultaneously through
pipelines in the experimental apparatus which will result in the formation of emulsions at
appropriate flow conditions. The system is configured as a loop; therefore oil and water
mixtures require separation prior to returning to their individual storage tanks. A gravimetric
separator (with a separation volume of 128 L) is utilised to achieve mixture separation during
two-phase oil/water flowing experiments. The effectiveness of the separator is dependent on
the fluid residence time within the separator as well as the degree of emulsification of the
mixture. Emulsion stability bottle tests were conducted in order to understand the residence
time behaviour of the canola oil and water emulsions.
Emulsions were prepared in 80 mL batches using a high shear homogeniser (Miccra
D8, ART-moderne Labortechnik, Germany). Samples were mixed for 3 min at 16000 rpm
(determined as the mixing speed which is the pipe flow Reynolds number equivalent to the
maximum overall flowrate of 10.0 m3/h). Each sample is visually monitored, and
photographs are obtained over an hour to observe speed of phase separation. Figure C.1
shows photographs of an example emulsion stability test for a 50% canola oil/ 50% water
mixture at variable times since emulsion formation.
189
Figure C.1 Photographs of emulsion stability bottle tests for a 50% canola oil/ 50% water mixture.
Photographs are captured; (a) directly after, (b) 1 minute after, (c) 1 hour after, and (d) 1 day after
emulsion formation via mixing.
The emulsion in Figure C.1 was observed to be stable for the first 30 minutes. After
one hour the emulsion can be observed to have formed two distinct layers, albeit with each
phase appearing milky and opaque indicating the presence of the alternate phase within each
layer (i.e. the top layer is an emulsion of water droplets within oil and the bottom layer is an
emulsion of oil droplets within an water continuous phase). The emulsion is observed to
have fully separated into two clear layers (oil and water alone) after one day, indicating the
lack of droplets present within the layers.
The observed separation times are obviously inappropriate for application within the
flow loop system. For example, a flowrate of 10 m3/h would require an impractically large
10 m3 gravity separator to achieve the partial separation of fluids observed in Figure C.1(c).
A demulsification technique was required to enhance the separation and achieve separation
of emulsions in a reasonable timeframe (ideally on the order minutes for the separation
residence time). The simplest system modification would be the introduction of a chemical
demulsifier to the solution (Zolfaghari et al., 2016). The introduction of high concentrations
of sodium chloride (>10%) has been observed to reduce emulsion stability for crude oil and
water emulsions (Moradi et al., 2011, Wang et al., 2010) as well as canola oil and water
emulsions (Klaus et al., 2012). Therefore emulsion stability bottle tests were conducted with
12 wt % salt introduced to the mixture. Three emulsions were prepared; one with 25 % oil/
75 % water, a second with 50 % oil and 50 % water and a third with 75 % oil/ 25 % water.
The results of the bottle stability tests are summarised in Table C.1. Partial separation time
(a) Initial (b) 1 min (c) 1 hour (d) 1 day
190
indicates the time required to form two layers of approximately correct volume, whilst full
separation time indicates the time required for the individual layers to appear clear.
Table C.1 Summary of emulsion stability bottle test results.
Test Oil/water
[vol %]
NaCl
[wt %]
Partial
separation time
Full separation
time
A 50 / 50 0 ~ 1 hour ~ 1 day
B 25 / 75 12 < 1 min ~ 5 min
C 50 / 50 12 < 1 min ~ 5 min
D 75 / 25 12 ~ 1 hour ~ 1 day
The results of tests B and C indicate that sodium chloride can have a dramatic impact
on the stability of emulsions and effective phase separation. Sodium chloride is observed to
cause decreasing emulsion stability beyond a certain ionic strength value (which depends on
the emulsion system) (Perles et al., 2012). The sodium ions are suggested to interact with
the aqueous phase and alter the adsorption strength at the interface, thus impacting the droplet
stability. However sodium chloride is ineffective for test D where the mixture consists of 75
% oil and 25 % water; as water-in-oil emulsions show enhanced stability in the presence of
sodium chloride (Ling et al., 2018).
The results from the emulsion stability test lead to the introduction of 12 wt % sodium
chloride into the tap water storage tank of the flow loop system. The influence of salt on
emulsion stability is critical for the results of the two-phase oil water tests. For the flow
regimes where an oil-in-water emulsion forms (i.e. emulsion of oil-in-water and water as well
as full oil-in-water emulsions) the presence of the sodium chloride destabilises the emulsion
leading to phase separation in a reasonable time. For flow regimes where a water-in-oil
emulsion forms (i.e. dual emulsion of oil-in-water and water-in-oil as well as a full water-in-
oil emulsion) the sodium chloride is ineffective therefore the emulsion remains stable and is
unable to separate in a reasonable time.
C.2 Influence of the Halbach array stray field
The effect of the Halbach array stray field on the detection region is dependent on the
strength of the detecting field. In the case of the very low Earth’s magnetic field (~54 μT)
191
the impact of the Halbach array stray field is enough to significantly distort measurements at
separation distances lower than 45 cm. The impact of the Halbach array is quantified by
measuring the shift in Larmor frequency as a function of separation distance. The deviation
from the Earth’s magnetic field (ΔB0) can then be quantified using the following;
𝛥𝐵0(𝐿𝑃𝐷) = (𝜔(𝐿𝑃𝐷)− 𝜔0)
𝛾 (C.1)
where ω(LPD) is the measured Larmor frequency at a given separation distance, ω0 is the
Larmor frequency in the local Earth’s field (2294 kHz, when the Halbach array is not nearby)
and γ is the gyromagnetic ratio of 1H (42.58 MHz/T). Figure C.2 shows the change in local
magnetic field strength as a function of the polarisation to detection separation distance.
Figure C.2 Change in the local magnetic field strength as a function of the separation distance
between the polarisation and detection coil. The current measurement limit (at 45 cm) is indicated.
Figure C.2 shows the substantial impact of the Halbach array stray field on the Earth’s
field detection coil even at reasonably large separation distances (>45 cm). This can be
attributed to the considerable difference in magnitude between the two magnetic fields
(Halbach array field is 0.3 T whilst Earth’s field is ~54 µT). Figure C.2 shows that at the
current measurement limit of 45 cm, the magnetic field deviation is ~0.7 μT, corresponding
to a deviation of ~1.3 % relative to the local Earth’s magnetic field.
0
0.2
0.4
0.6
0.8
1
1.2
0 20 40 60 80
ΔB
0[μ
T]
Separation distance [cm]
Currentmeasurement limit
192
If the detection field is higher, the relative impact of the Halbach array stray field will
be reduced. Upon back-extrapolation of the measured signals in Figure C.2; a detection field
of ~70 kHz is estimated to produce a relative deviation of 1 % at a 5 cm separation distance.
However this is a considerable back-extrapolation and more detailed magnetic field
simulations and experimental validation should be conducted to validate a reasonable
detection frequency. A detection coil in the range of 200-500 kHz will be considered for
future development of the low field NMR flow meter.