electromagnetic sensing techniques for multi- phase flow
TRANSCRIPT
Electromagnetic Sensing Techniques For Multi-
Phase Flow Monitoring In Industrial Processes
A thesis submitted to the University of Manchester for the degree of
Doctor of Philosophy
in the Faculty of Science and Engineering
2018
By
Shupei Wang
School of Electrical and Electronic Engineering
2
TABLE OF CONTENTS
TABLE OF CONTENTS .............................................................................................. 2
LIST OF FIGURES ...................................................................................................... 7
LIST OF TABLES ...................................................................................................... 12
NOMENCLATURE .................................................................................................... 13
ABSTRACT ................................................................................................................ 18
DECLARATION ........................................................................................................ 19
COPYRIGHT STATEMENT .................................................................................... 20
ACKNOWLEDGEMENTS ........................................................................................ 21
LIST OF PUBLICATIONS ........................................................................................ 22
Chapter 1 Introduction ........................................................................................... 23
1.1 Motivation ...................................................................................................... 23
1.1.1 Clean-In-Place (CIP)................................................................................ 23
1.1.2 Industrial oil/water separation .................................................................. 24
1.2 Aims and objectives ........................................................................................ 25
1.3 Contributions .................................................................................................. 26
1.4 Organization of thesis ..................................................................................... 28
Chapter 2 Background of Sensing Techniques in the Monitoring of Industrial
Process with Multi-Phase Flow .................................................................................. 31
2.1 Existing sensing techniques in multi-phase flow measurement ........................ 31
2.1.1 Mechanical and hydraulic sensing ............................................................ 31
3
2.1.2 Electrical sensing ..................................................................................... 32
2.1.3 Acoustic, radiative and microwave sensing .............................................. 34
2.2 Monitoring CIP with Electrical Resistance Tomography (ERT) ...................... 38
2.2.1 State-of-the-art in monitoring CIP ............................................................ 38
2.2.2 Advantages of ERT in monitoring CIP ..................................................... 39
2.3 Measuring industrial oil-water batch separation process with Differential
Electromagnetic Inductive Sensor (DEMIS) .............................................................. 40
2.3.1 State-of-the-art in measuring industrial oil-water batch separation process40
2.3.2 Advantages of DEMIS in monitoring oil separation process ..................... 46
Chapter 3 Monitoring CIP Using Electrical Resistance Tomography (ERT) with
Dynamic Reference ..................................................................................................... 47
3.1 Fundamentals of ERT ..................................................................................... 47
3.1.1 Sensitivity and forward problem .............................................................. 47
3.1.2 Inverse problem and conventional algorithms .......................................... 52
3.2 Experimental setup and principles ................................................................... 55
3.2.1 Experimental setup .................................................................................. 55
3.2.2 Experimental procedures.......................................................................... 57
3.2.3 Measurement protocol ............................................................................. 58
3.2.4 Analytical principles ................................................................................ 59
3.3 Inverse calculation results with LBP and Tikhonov regularization .................. 59
3.3.1 Linear Back Projection (LBP) .................................................................. 59
3.3.2 Tikhonov regularization ........................................................................... 61
4
3.4 Algorithm optimization with dynamic reference ............................................. 65
3.4.1 Methodology ........................................................................................... 65
3.4.2 Optimization procedure ........................................................................... 66
3.5 Image reconstruction results calculated from Tikhonov regularization with
dynamic reference ..................................................................................................... 68
3.5.1 Average conductivity values .................................................................... 68
3.5.2 Maximum conductivity values ................................................................. 69
3.5.3 Image reconstruction comparison ............................................................. 70
3.6 Adopting Tikhonov regularization with dynamic reference in different cleaning
conditions ................................................................................................................. 72
3.6.1 Results under higher flow rate with T pipe ............................................... 72
3.6.2 Results with different pipe geometries ..................................................... 73
3.7 Summary ........................................................................................................ 76
Chapter 4 Electromagnetic Simulations: Modelling Three-Coil DEMIS and Oil-
Saline Batch Separation .............................................................................................. 78
4.1 Introduction of differential electromagnetic inductive sensor .......................... 78
4.1.1 Sensor structure ....................................................................................... 78
4.1.2 Sensitivity distribution ............................................................................. 79
4.1.3 Sensitivity distribution with different vessel radii..................................... 83
4.2 Electrical model of liquid-liquid separation ..................................................... 86
4.2.1 Liquid-liquid separation model ................................................................ 86
4.2.2 Effective conductivity model ................................................................... 89
5
4.3 Simulation of the electrical liquid-liquid separation model .............................. 91
4.4 Simulation results and discussions .................................................................. 93
4.5 Summary ........................................................................................................ 96
Chapter 5 Experimental System and Results In Monitoring Oil-Water
Separation With DEMIS ............................................................................................ 99
5.1 Experimental setup and testing strategy .......................................................... 99
5.1.1 Sensor system ........................................................................................ 100
5.1.2 Mixing and separation system ................................................................ 100
5.1.3 FPGA-based impedance analyser ........................................................... 101
5.1.4 Testing strategy ..................................................................................... 103
5.1.5 Choice of parameters ............................................................................. 104
5.2 Experiment and discussion ............................................................................ 105
5.2.1 Instrument performance inspections ....................................................... 106
5.2.2 Validation of sensitivity distribution ...................................................... 106
5.2.3 Experimental result example .................................................................. 107
5.2.4 Validation of sensor output .................................................................... 109
5.2.5 Experiment results under different mixing conditions ............................ 111
5.3 Summary ...................................................................................................... 115
Chapter 6 Conclusions and Future Works ........................................................... 118
6.1 Conclusions .................................................................................................. 118
6.1.1 Monitoring CIP using ERT with dynamic reference ............................... 118
6.1.2 Measuring oil-water separation process using DEMIS ........................... 120
6
6.2 Future work .................................................................................................. 122
REFERENCES ......................................................................................................... 124
7
LIST OF FIGURES
Figure 1-1 Schematic diagram of API pilot-scale separator ......................................... 25
Figure 2-1 Venturi meter ............................................................................................. 32
Figure 2-2 Ultrasound velocity profiler ....................................................................... 34
Figure 2-3 Common approach to implement radiative sensing on MPFs...................... 36
Figure 2-4 Schematic of radiography system with multiple radiative sources and 2D
detectors ..................................................................................................................... 36
Figure 2-5 Schematics of ultrasonic techniques in fouling detection. (a) Pulse echo
technique; (b) Transmission technique. ....................................................................... 38
Figure 2-6 Monitoring oil-water separation process with sight glass ............................ 40
Figure 2-7 Monitoring oil-water separation process with motive pressure sensor......... 41
Figure 2-8. Monitoring oil-water separation process with ultrasonic sensor ................. 42
Figure 2-9 Monitoring oil-water separation process with segmented sensor array ........ 44
Figure 3-1 Electromagnetic simulation model for sensitivity matrix calculation .......... 48
Figure 3-2 Sensitivity distributions in part of the projections ....................................... 49
Figure 3-3 Simulated CIP circuit ................................................................................. 55
Figure 3-4 'T-shape' pipe with the 4 ERT planes installed ........................................... 56
Figure 3-5 Schematic diagram of a straight pipe with a butterfly valve installed inside 56
Figure 3-6 Straight pipe with a butterfly valve installed inside .................................... 57
8
Figure 3-7 Contents to be measured in the test region under each step of the simulation
test. ............................................................................................................................. 58
Figure 3-8 Maximum conductivity values in all ERT planes during the cleaning process
calculated with LBP .................................................................................................... 60
Figure 3-9 Maximum conductivity values in all ERT planes during the cleaning process
calculated with LBP .................................................................................................... 61
Figure 3-10 Maximum conductivity values in all ERT planes during the cleaning
process calculated with Tikhonov regularization ......................................................... 64
Figure 3-11 Average conductivity values in all ERT planes during the cleaning process
calculated with Tikhonov regularization ...................................................................... 64
Figure 3-12 Relationship between measured boundary voltage and material conductivity
under same projection ................................................................................................. 65
Figure 3-13 Measured boundary voltages (U curves) comparison between pure water
and pure soil. .............................................................................................................. 65
Figure 3-14 Average conductivity values calculated from Tikhonov regularization with
dynamic reference ....................................................................................................... 69
Figure 3-15 Maximum conductivity values calculated from Tikhonov regularization
with dynamic reference ............................................................................................... 70
Figure 3-16 Maximum conductivity values in smaller scale ........................................ 70
Figure 3-17 Average and maximum conductivity values calculated from Tikhonov
Regularization with dynamic reference under higher flow rate. . a) Average conductivity
9
values; b) maximum conductivity values; c) maximum conductivity values in smaller
scale. ........................................................................................................................... 72
Figure 3-18 Average and maximum conductivity values calculated from Tikhonov
Regularization with dynamic reference under 4100 L/h and butterfly valve. a) Average
conductivity values; b) maximum conductivity values; c) maximum conductivity values
in smaller scale. .......................................................................................................... 74
Figure 3-19 Average and maximum conductivity values calculated from Tikhonov
Regularization with dynamic reference under 6200 L/h and butterfly valve. a) Average
conductivity values; b) maximum conductivity values; c) maximum conductivity values
in smaller scale. .......................................................................................................... 75
Figure 4-1 Schematic of differential electromagnetic inductive sensor ........................ 78
Figure 4-2 Sensitivity distribution in the axial cross-sectional testing area of a simulated
two-coil sensor system ................................................................................................ 82
Figure 4-3 Simplified and normalized planar sensitivity distribution ........................... 84
Figure 4-4 Planar sensitivity distribution under different testing vessel radii ............... 84
Figure 4-5 Average planar sensitivity values and standard deviation under different
testing area radii .......................................................................................................... 86
Figure 4-6 (a) 4 sections defined in oil/water separation system. (b) Height change of
the boundaries of 4 sections with time ......................................................................... 87
Figure 4-7 Interface heights in a theoretical oil/saline separation model under 350RPM
and oil fraction at 50% ............................................................................................... 92
Figure 4-8 Geometry of the simulation model. (a) 3D model; (b) 2D model. ............... 92
10
Figure 4-9 Induced voltage in both receiving coils and the differential output when oil
fraction is 50% and agitation speed is 350RPM ........................................................... 93
Figure 4-10 Simulated sensor outputs of two separation process from the same liquid
system agitated with same time length and different speeds ........................................ 95
Figure 4-11 Simulated sensor outputs of two separation process from the two liquid
system with different oil fractions agitated with same time length and speed ............... 95
Figure 5-1 Experimental system setup ......................................................................... 99
Figure 5-2 (a)Hardwood rod coated with black thermal plastic tube; (b) 3D-printed
plastic impeller ......................................................................................................... 100
Figure 5-3 System block diagram .............................................................................. 101
Figure 5-4 The instrument performance for measuring a voltage signal ..................... 106
Figure 5-5 Experimental test result for vertical sensitivity distribution with saline..... 107
Figure 5-6 Single test result under 1700RPM, 30 seconds mixing and 500 seconds
separation, oil fraction 50%....................................................................................... 108
Figure 5-7 Sensor output signal during separation process ........................................ 108
Figure 5-8 Part of the screenshots from the recorded video during the separation process
................................................................................................................................. 110
Figure 5-9 Comparison between sensor output and saline interface height change during
the separation process ............................................................................................... 110
11
Figure 5-10 Sensor outputs of repeated experiments under agitation speed of 1700 RPM,
oil fraction 50%. (a) mixing duration 30 seconds; (b) mixing duration 5 minutes; (c)
mixing duration 15 minutes. ...................................................................................... 112
Figure 5-11 Average sensor outputs of the repeated test under 1700 RPM during
separation process ..................................................................................................... 113
Figure 5-12 Average sensor outputs of the repeated test under 900RPM during
separation process with error bars ............................................................................. 113
Figure 5-13 Average sensor outputs of the repeated test at oil fraction of 33% under
900RPM during separation process with error bars ................................................... 114
12
LIST OF TABLES
Table 2-1 General comparison of Electrical Tomography methods .............................. 34
Table 3-1 Adjacent strategy on 16-electrode ERT system ............................................ 58
Table 3-2 Image reconstruction results under different regularization parameters during
cleaning process for Plane 4 ........................................................................................ 62
Table 3-3 The ratio of measured voltage for each projection between the soil and water
background ................................................................................................................. 66
Table 3-4 Reconstructed image in Plane 4 comparisons between conventional and
optimized Tihonov regularization................................................................................ 71
Table 4-1 The average and standard deviation of the sensitivity values under different
test region radii ........................................................................................................... 85
Table 4-2 Experimental profiles and model parameters adopted from [96] for H0=300
mm, D=154 mm, εp=0.65, and one hour agitation time ................................................ 91
13
NOMENCLATURE
Abbreviations and Acronyms
2D 2-dimentional
3D 3-dimentional
ADC analogue-to-digital converter
API American Petroleum Institute
C4D contactless capacitively coupled conductivity detector
CCD charge-coupled device
CIP Clean-In-Place
CMR Christian Michelsen Research
DAC digital-to-analogue converter
DEMIS differential electromagnetic inductive sensor
ECT Electrical Capacitance Tomography
EIS electrolyte-insulator-semiconductors
EIT Electrical Impedance Tomography
EMT Electromagnetic Tomography
ERT Electrical Resistance Tomography
FEM finite element method
FPGA field-programmable gate array
ILMS Inductive Level Monitoring System
LBP Linear Back Projection
MFMs multi-phase flow meters
MPFs multi-phase flows
NDT non-destructive testing
RPM Revolutions per minute
SNR signal-to-noise ratio
SoC system on a chip
14
Symbols
A1 upper sensing region of the DEMIS
A2 lower sensing region of the DEMIS
A magnetic vector potential
AR magnetic vector potential at the target point when the
receiving coil is excited by unit current
AT magnetic vector potential at the target point when the
transmitting coil is excited by unit current
B magnetic flux density
dl position vector of the corresponding segment of the
coil
D separation vessel diameter
E electrical field vector
ER local electric field vectors when the other coil is
excited by unit current and the testing area is empty
ET local electric field vectors when one of the two coils is
excited by unit current and the testing area is empty
E φ electric field vector in the corresponding pixel when
electrode pair one is injecting current and electrode
pair two is measuring the voltage
E ψ electric field vector in the corresponding pixel when
electrode pair two is injecting current and electrode
pair one is measuring the voltage
g gravitational acceleration
Δh1 thickness of the dense-packed zone
hc height of the interface between clear oil zone and
dense-packed zone
hp height of the interface between dense-packed zone and
sedimentation zone
hp1 height of the interface between the dense-packed zone
and sedimentation zone prior to the inflection point
hp2 height of the interface between the dense-packed zone
and sedimentation zone anterior to the inflection point
15
hs height of the interface between sedimentation zone
and clear water zone
hsi height of the interface between sedimentation zone
and clear water zone at inflection point
H magnetic field strength
H0 initial height of dispersion
H0il thickness of the clear oil layer
I identity matrix
I amplitude of the excitation current
j voltage projection number
J current vector in the coil
Jj,k normalized sensitivity of pixel k under voltage
projection j
Jp∗w
normalized sensitivity matrix
k pixel number
k1, k2, k3, k4 fitting constants without clear physical meanings
p total number of projections
r distance between the coil elements and the target point
in sensing region
R radius of the oil droplet
R1 upper receiving coil of the DEMIS
R2 lower receiving coil of the DEMIS
Sk planar sensitivity value on the kth cross-section plane
Sk,j the sensitivity value on the kth pixel under voltage
projective j
Sr relative sensitivity in a two-coil system
Sσ local conductivity sensitivity at a spatial point in the
testing area of a two-coil sensor system
Sck sensitivity value of the cth pixel on the kth cross-section
plane
t time
ti inflection point
T transmitting coil of the DEMIS
16
v buoyance velocity of an independent sphere oil droplet
in the continuous water phase
v0 initial sedimentation velocity of oil drops
vi sedimentation velocity of oil drops at the inflection
point
Vo output voltage of the DEMIS
V1 real part of the induced complex voltages of one of the
receiving coils
V2 real part of the induced complex voltages of the other
receiving coil
Ve detected signal of the empty space
Vmm median value of the 16 chosen voltages from measured
boundary voltages
VRm median value of the 16 chosen voltages from the
reference voltages
VR,j reference voltage in projection j
V∗ voltage change vector with dynamic reference
VR∗ dynamic reference voltages corresponding to each
frame
Vp∗1 voltage change vector
α Tikhonov regularization parameter
γ ratio between the average level of the measured
voltages and the reference
ΔV perturbation of the induced voltage
∆Vj voltage change in projection j
Δσ perturbation of conductivity
∆σk conductivity change of pixel k
ε0 initial oil hold-up fraction
εp oil holdup fraction in the dense-packed zone
εs oil fraction in the sedimentation zone
μ permeability of medium
μ0 vacuum permeability
μo dynamic viscosity of oil
17
ρ0 mass density of oil
ρw mass density of water
σ1 conductivity of the first liquid
σ2 conductivity of the second liquid
σmp effective conductivity of liquid as a mixture of liquids
with two different conductivities
σmc effective conductivity in the sedimentation zone
σR reference conductivity
σR,k reference conductivity of pixel k
σα conductivity change vector calculated by Tikhonov
regularization
σw∗1 conductivity change vector
σα∗ conductivity change vector calculated by Tikhonov
regularization with dynamic reference
ω excitation signal frequency
18
ABSTRACT
Multi-phase Flows (MPFs) exist in various chemical engineering and industrial
processes and the research on monitoring MPFs is of great importance. In this research,
electromagnetic sensing techniques in the monitoring of two industrial processes with
MPFs are studied, namely Clean-In-Place (CIP) and industrial oil-water separation.
In multifunctional food and detergent production lines, accurate identification of ending
point of the cleaning process for the previous product is crucial to ensure product
integrity. In this research, an optimization method with dynamic references based on
Tikhonov regularization is proposed and validated by monitoring a lab CIP circuit with
a commercial Electrical Resistance Tomography (ERT) system. The results prove that
the proposed method is capable of accurately identifying the ending point of CIP
process. Moreover, the comparisons made with several conventional image
reconstruction algorithms illustrate that significantly improved inverse calculation
results are obtained when the background conductivity largely differs from the reference
conductivity. Additionally, the feasibility of this novel approach is discussed.
Liquid-liquid separation is an important process in many chemical engineering
applications. The ability of monitoring this process, in particular with a non-contact
method is of high value. In this research, a novel sensing approach which adopts a
differential electromagnetic inductive sensor (DEMIS) and an FPGA-based (field-
programmable gate array) impedance analyser is proposed and implemented to monitor
the separation processes of an oil-in-water liquid system. The inductive sensor has a
concentric cylinder structure with its coils arranged differentially. It is optimised to
achieve a homogeneous sensitivity distribution in the sensing region. Electrical models
of the oil-saline separation processes are established. Experiments under different oil
and saline fractions, different agitation speeds and durations are conducted to validate
the capability of the system. Both simulation and experimental results have proved that
the proposed system is capable of monitoring oil/water separation process non-
intrusively and non-invasively with a relatively lower cost, higher reliability and less
complex structure, comparing to existing techniques.
19
DECLARATION
No portion of the work referred to in this thesis has been submitted in support of an
application for another degree of qualification of this or any other university or other
institution of learning.
20
COPYRIGHT STATEMENT
i. The author of this thesis (including any appendices and/or schedules to this thesis)
owns certain copyright or related rights in it (the “Copyright”) and he has given The
University of Manchester certain rights to use such Copyright, including for
administrative purposes.
ii. Copies of this thesis, either in full or in extracts and whether in hard or electronic
copy, may be made only in accordance with the Copyright, Designs and Patents Act
1988 (as amended) and regulations issued under it or, where appropriate, in accordance
with licensing agreements which the University has from time to time. This page must
form part of any such copies made.
iii. The ownership of certain Copyright, patents, designs, trade marks and other
intellectual property (the “Intellectual Property”) and any reproductions of copyright
works in the thesis, for example graphs and tables (“Reproductions”), which may be
described in this thesis, may not be owned by the author and may be owned by third
parties. Such Intellectual Property and Reproductions cannot and must not be made
available for use without the prior written permission of the owner(s) of the relevant
Intellectual Property and/or Reproductions.
iv. Further information on the conditions under which disclosure, publication and
commercialisation of this thesis, the Copyright and any Intellectual Property and/or
Reproductions described in it may take place is available in the University IP Policy
(see http://documents.manchester.ac.uk/DocuInfo.aspx?DocID=487), in any relevant
Thesis restriction declarations deposited in the University Library, The University
Library’s regulations (see http://www.manchester.ac.uk/library/aboutus/regulations) and
in The University’s policy on Presentation of Theses.
21
ACKNOWLEDGEMENTS
I would like to express my sincere appreciation to everyone who supported and
encouraged me throughout the four years of my Ph.D. research. It has been an honour
and also wonderful experience to work in the group of Sensing, Imaging and Signal
Processing. In particular, I want to express my thanks to:
My supervisor, Dr. Wuliang Yin, for his constant guidance, encouragement, and
patience along the way. I am always grateful for becoming his student, as he not only
led me into the right path in research, but also taught me to be a good person in life. No
words could be strong enough to express my appreciation to him and I could have never
gone so far without him.
Dr. Ruozhou Hou in School of Chemical Engineering and Analytical Science, the
University of Manchester, for selflessly providing critical supports and discussions for
my research and becoming a good friend of mine in personal life. Your professionality
and enthusiasm would always inspire me.
My dear colleagues, Jorge Salas Avila, for helping me with the experimental setup and
developing the measurement system; Dr. Yuedong Xie, for the helpful suggestions and
guidance in simulations and thesis writing; Dr. Yang Tao, for the support in publication;
and many more in SISP group who offered me help and suggestions in many ways.
Moreover, I want to thank my parents, Mr. Liangmei Wang and Mrs. Jianhua Sun, for
their love and encouragement which supports me for the entire time. I would also like to
thank my beloved wife, Qingna Zhou, who helped me going through the hard time and
took care of me with everything she had.
At last, I want to thank our baby, who still lies in his/her mother’s belly, for fighting
together with dad and mom for their Ph.D. degrees in the last few months. You will
always be our most precious gift and I wish you all the best in your future life.
22
LIST OF PUBLICATIONS
Journal paper:
1. S. Wang, R. Hou, J. R. Salas Avila, Y. Tao and W. Yin, "A Novel Approach to
Measuring Separation Process of Oil–Saline Using Differential Electromagnetic
Inductive Sensor and FPGA-Based Impedance Analyzer," in IEEE Sensors Journal, vol.
18, no. 19, pp. 7980-7989, 1 Oct.1, 2018.
Conference papers:
1. S. Wang, W. Yin, “Monitoring Cleaning-In-Place by Electrical Resistance
Tomography with Dynamic References”, Imaging Systems and Techniques (IST), 2016
IEEE International Conference on. IEEE, 2016.
2. S. Wang, Y. Liu, K. Andrikopoulos, W. Yin, “Design of a Low-Cost Integrated
Electrical Resistance Tomography(ERT) System Based on Serial Bus”, Imaging
Systems and Techniques (IST), 2016 IEEE International Conference on. IEEE, 2016.
23
Chapter 1 Introduction
In this chapter, the motivation of the presented research is firstly illustrated. Then the
aims, objectives and contributions of the study are introduced, followed by the
organization of the whole thesis.
1.1 Motivation
The research on multi-phase flows (MPFs) is an important topic in fluid mechanics. The
term ‘multi-phase flows’ defines any fluid flow with more than one phase or component
and the behaviour of MPFs is proved to be extremely complicated[1]. Numerous
parameters of MPFs have been studied in the past, corresponding to different research
objectives, e.g. flow pattern, transient behaviour, component volume fraction, liquid
level detection etc.[2-5]. In this research, two important industrial processes which
involve multi-phase flows were studied, namely Clean-In-Place (CIP) and industrial oil-
water separation.
1.1.1 Clean-In-Place (CIP)
CIP is an important industrial process in food and detergent plants. It aims at the
removal/cleaning of soil inside the production line, which can endanger the process
sterility, without dismantling the plant[6]. It has always been a crucial topic, especially
for plants with multifunctional production lines, to monitor and analyse the process of
CIP. The main objective of studying CIP is to minimize water, chemical and waste-
treatment costs[7]. This requires accurate indication of the ending point (the moment
when the soil in the lose-loop pipe system is fully cleaned) and locating the most
difficult part to be cleaned.
24
Electrical Resistance Tomography (ERT) is a non-intrusive technology which has been
widely adopted in industrial process monitoring. It involves the acquisition of boundary
voltages from sensors (electrodes) located on the periphery of an object, such as a
process vessel, pipeline, etc.[8, 9]. The boundary voltages vary with the change of
external electric field (normally triggered by the injection of current) which is caused by
the change of conductivity distribution on the electrode plane[10]. With inverse
algorithms, the boundary voltages can be computed into reconstructed cross-sectional
images of the conductivity distributions. Because of its rapid measurement speed,
imaging ability and relatively low cost, ERT has become a popular technique in
industrial process monitoring and analysis.
In this research, ERT is adopted as the main technique to monitor and analyse the CIP
process. The main advantages of ERT in this scenario, comparing to other existing
techniques, can be concluded as follows:
1. Non-intrusive and non-hazardous;
2. Relatively low cost;
3. The ability to visualize the interior component distribution.
1.1.2 Industrial oil/water separation
Monitoring the separation process of crude oil extracted from an oil well is of great
importance in the oil industry. One of the most prevailing separation devices is the
American Petroleum Institute (API) oil/water separator which is illustrated in Figure 1-1.
At the rear of the API separator, oil flows over the edge of the variable height weir and
is recovered through the oil outlet pipe. Water, on the other hand, is directly disposed of
through the outlet pipe connected to the bottom of the primary separation tank. The
25
separation speed of oil and water is mainly dependent on the oil/water ratio, and the
interface level determines the control strategy of two outlet valve openings which
eventually controls the residence time of the oil and water output. The residence time
represents the probability distribution of time that oil/water stay inside the separator in
continuous operations[11]. It is therefore essential that the separation process and
oil/water ratio in the separation tank are monitored continuously in order to ensure a
high production efficiency and product consistency.
Figure 1-1 Schematic diagram of API pilot-scale separator
1.2 Aims and objectives
The aims of this research is to develop novel approaches or technologies for measuring
the CIP process and industrial oil/water separation process that have significant
advantages against conventional methodologies. Respectively, the research objectives
for the two industrial processes include:
For monitoring CIP process,
26
1. Developing a new approach of monitoring CIP process with ERT to accurately locate
the ending point of the whole process, i.e. locating the most difficult pipe section to be
cleaned and capture the moment when it is fully cleaned;
2. Visualizing the component distribution inside the test region to locate the most
difficult position to be cleaned;
3. Comparing the inverse results calculated from measured voltages of the lab-scale CIP
process with conventional algorithms and applying corresponding optimization method
to achieve more accurate results.
For measuring industrial oil-water separation process,
1. Designing a novel and reliable sensor system capable of measuring oil/water
separation process both non-intrusively and non-invasively. The sensor system should
be able to monitor the completion level of the separation process and locate the ending
point;
2. Conducting electrical simulations of monitoring the oil/water separation process with
proposed sensor system to achieve a better understanding of how the changes of the
physical parameters in the liquid system lead to the changes in the sensor output;
3. Testing the sensor system under different separation conditions and validating the
sensor output results.
1.3 Contributions
The author’s research has contributed to the sensing techniques in the monitoring of
industrial process with multi-phase flow in several ways. The significance and novelty
of the research can be concluded as follows:
27
In the research of monitoring CIP using ERT with dynamic reference:
1. For the first time, a novel optimization method with dynamic reference was proposed
based on conventional Tikhonov regularization. The method involves the construction
of dynamic references for the inverse calculation in ERT based on the average level of
measured voltages. The dynamic reference helps in reducing the distortion of the
inverse solution caused by the significantly different conductivities between the test
region and the reference;
2. By applying this optimization method to the inverse calculation of conductivity
distribution during CIP process, the reliability of ERT is improved. More accurate
results were achieved throughout the process, especially when the test regions were
dominated by high conductivity components. In addition, the advantage of conventional
Tikhonov regularization still remained, which is the accurate indication of the ending
point. Experiments were carried out under different CIP conditions, e.g. flow rate and
pipe geometry. The results clearly indicated the ending point of the CIP process and the
most difficult parts to be cleaned in the test region, which are the most critical
objectives of the research.
In the research of measuring oil/water separation process using differential
electromagnetic inductive sensors (DEMIS):
1. For the first time, a novel, cost-effective, non-intrusive and non-invasive technique
was developed for measuring oil/water separation process. The merits of DEMIS will be
explained in Chapter 2 by comparing the proposed technique with existing techniques in
measuring oil-water separation;
28
2. A novel optimization method was proposed to homogenize the vertical sensitivity
distribution in the test region of the DEMIS. The method involves decreasing the ratio
between the diameters of the separation vessel and the coils. An optimum ratio was
determined to achieve a relatively homogeneous vertical sensitivity distribution while
maintaining the sensitivity values and system signal-to-noise ratio at a reasonable level.
3. In the simulation part of this research, a dynamic electrical model was constructed
based on the physical liquid-liquid separation model. The physical model explains the
division of different sections in the liquid system during the separation process based on
the transient behaviour of oil droplets, as well as the height change of interfaces
between corresponding sections. Moreover, the equivalent conductivity value in each
section was calculated based on the conductivities of oil and water, oil/water fraction
calculated from the interface heights and Maxwell Garnett mixing formula. Thus a time-
varying electrical liquid-liquid separation model can be constructed in the
electromagnetic simulation toolkit. The above method successfully explained how the
changes in physical parameters lead to the change in the electrical output signal from
the sensor.
4. The output signal from the DEMIS during the oil-water separation process was
validated by comparing it with the physical interface (water interface) height change
synchronously recorded by a camera. It is proved that the changing trend of output
signal matches with the physical process and it could indicate the ending point of the
separation process.
1.4 Organization of thesis
Chapter 1 introduces the motivations, aims and objectives, contributions and
organization of the thesis.
29
Chapter 2 is a general review for the background of this research. Existing mainstream
sensing techniques of multi-phase flow was firstly introduced to achieve a general
understanding of the state-of-the-art in this area. Then the two industrial processes to be
studied in the author’s PhD work, namely CIP and industrial oil-water separation
process, together with the state-of-the-art monitoring techniques was introduced in
Section 2.2 and 2.3 respectively. In the end of each section, the novelty of the research
methodology was explained by comparing the techniques adopted in this research with
other existing techniques.
In Chapter 3, Electrical Resistance Tomography (ERT) is used as a new approach to
analyse the comprehensive spatial and time-varying conductivity changes during lab-
scale CIP processes, aiming at locating the most difficult point to be cleaned in the pipe
circuit and the ending point of the whole cleaning process. The chapter starts with
introducing the fundamentals of ERT, including the definition of forward problem and
inverse problem, and some conventional algorithms for imaging reconstruction. Then
the experimental setup and procedure of ERT measurement for monitoring CIP process
are explained, as well as the analytical principles for identifying the ending point of
cleaning process, and evaluating the overall performance of reconstructed images. With
the measured voltages, preliminary images were constructed with conventional
algorithms and the merits and drawbacks were pointed out. Then an optimization
method with dynamic reference was proposed to overcome the corresponding
drawbacks and the optimized results were compared with the original ones. In the end,
results achieved from optimized algorithm under different cleaning flow rate and pipe
geometries were also presented to justify the suitability of the optimization method
under different cleaning conditions.
30
In Chapter 4, the fundamentals of the proposed three-coil differential electromagnetic
inductive sensor was firstly introduced. In addition, a sensitivity analysis based on
analytical calculation and field value extraction is illustrated which gives rise to the
optimized design of the sensor. The optimization aims at homogenizing the vertical
sensitivity distribution within the sensing region. Moreover, an equivalent electrical
model of the liquid-liquid separation model is derived based upon which simulations are
carried out to explain the evolution of sensor output signal with respect to separation
process. Finally, the simulation results under different prior mixing conditions, e.g.
oil/water fraction, agitation speed and duration, are presented and discussed.
In Chapter 5, the design of the experimental system was firstly demonstrated, including
the sensor system, mixing and separation system and measurement system, followed by
the experimental strategy. Moreover, the validation of the experimental system and
results were carried out. The measurement system was validated by measuring a known-
magnitude voltage signal, while the validation of sensor system, specifically the
sensitivity distribution in the sensing region, was carried out by continuously adding
same amount of saline and inspect the corresponding sensor output change. A critical
validation of the sensor output during a complete oil-water separation process was also
demonstrated. It involves comparing the sensor output signal with the saline interface
height change recorded by a camera synchronously. Finally, experiments were
conducted under different prior mixing conditions, e.g. agitation speed, duration and
oil/saline fraction. The experimental results are compared with simulation results in
Chapter 4 to assess the performance of the proposed system.
Chapter 6 is the conclusion of this research and recommendations for future works.
31
Chapter 2 Background of Sensing Techniques in the
Monitoring of Industrial Process with Multi-
Phase Flow
In this chapter, existing mainstream sensing techniques in the research of multi-phase
flow were firstly introduced to achieve a general understanding of the state-of-the-art in
this area. Then the two industrial processes to be studied in the author’s PhD work,
namely CIP and industrial oil-water separation process, together with the state-of-the-art
monitoring techniques were introduced in Section 2.2 and 2.3 respectively. In the end of
each section, the novelty of the research methodology was explained by comparing the
techniques adopted by the author with other existing techniques.
2.1 Existing sensing techniques in multi-phase flow
measurement
There is a wide range of techniques that enables the evaluation and analysis of the
properties of MPFs. Based on the interaction between the emission signal and the
sample, the techniques can be categorized as mechanical sensing, hydraulic sensing,
electrical sensing, acoustic sensing, radiative sensing, and microwave sensing[12].
2.1.1 Mechanical and hydraulic sensing
Mechanical sensing involves the transmission of force or motion from the fluid to
instruments. One typical example is the conventional water meters which translate the
displacement or velocity profile of the water flow into the meter reading of flow rate.
Hydraulic sensing in general denotes the measurement of liquid pressure. Pressure
measurement is adopted in many industrial processes as liquid pressure is always
correlated with other fluid parameters such as liquid level, flow rate, etc. One example
is the venturi meter (Figure 2-1), which takes advantage of measuring the pressure
32
difference of the fluid when flowing through vessels with different diameters and thus
acquires the flow rate of the fluid[13].
Figure 2-1 Venturi meter
2.1.2 Electrical sensing
Electrical sensing in MPFs consists of a large group of techniques which involve the
measurement of electromagnetic parameters of the fluid with the aid of external
electromagnetic field and interpreting them into the information regarding component
motion, distribution or volume fraction.
Electrical tomography, for example, is a popular technique in the monitoring and
analysing of industrial process with MPFs. The term ‘electrical tomography’ normally
includes Electromagnetic Tomography (EMT), Electrical Capacitance Tomography
(ECT) and Electrical Impedance Tomography (EIT), while Electrical Resistance
Tomography (ERT) is considered to be a particular case of EIT[14]. The above
electrical tomography techniques share a common basic aim, which is to obtain a set of
measurements through sensors locating at the periphery of the test region so as to
achieve the component distribution inside (normally presented by a reconstructed image
of the cross-section) with certain inverse calculation algorithms[15-17]. The choice of
33
techniques in different applications mainly depends on the target material in the MPFs
to be monitored. EMT is capable of monitoring liquid metal, mineral, magnetic
materials and ionised water with the coil sensor array arranged outside the process
vessel[18]. It aims at acquiring the material conductivity and permeability distribution
with mutual inductance measurements[19, 20]. In terms of ECT, the target electrical
parameter to be measured is the permittivity of the material[21]. The measurement is
implemented by measuring the capacitance change between capacitive electrode plates
(usually eight or 16) mounted on the exterior surface of the process vessel[22, 23].
Similarly, EIT focuses on measuring the impedance (both conductivity and permittivity)
change between a set of electrodes[24]. As the conductivity measurement requires the
injection of current or voltage into the material, electrodes for EIT measurement are
usually installed on the interior surface of the process vessel to have direct contact with
the materials inside[25, 26]. In addition, ERT is an extreme case of EIT which only
focuses on the measurement of conductivity[27]. A general comparison of the electrical
tomography methods is illustrated in Table 2-1.
Other electrical sensing techniques have been proposed by various researchers. G.
Lucas and etc. developed a four-sensor probe system to measure non-conductive bubble
velocity profile in air-water and oil-water flows[28, 29]. This is achieved by calculating
the time intervals taken from the output signals from each of the four conductance
sensors located with the probe. Contactless capacitively coupled conductivity detector
(C4D) is a popular sensing technique in organic and biochemical applications. To detect
small inorganic ions as well as organic and biochemical species, two tubular electrodes,
namely actuator electrode and pick-up electrode, are arranged on the target tube, e.g. a
capillary, and connected capacitively[30]. Du and Zhe developed a novel inductive
pulse sensor to detect metallic wear debris in rotating machinery lubrication oil
34
circuit[31]. The sensor consists a two-layer planar coil capable of detecting the
inductance change caused by the passage of metallic debris. Other electrical sensing
techniques exist in different research areas and we will not numerate here.
Table 2-1 General comparison of Electrical Tomography methods
Methods Sensor arrangement Target
parameter Typical material
EMT
Coil Array
Self/mutual
inductance
Metals
Minerals Magnetic materials
Ionized water
ECT
Capacitive Plates
Capacitance
Oil
De-ionized water Non-metallic powders
Polymers
Burning gases
EIT
Electrode Array
Impedance
Water/saline Biological tissue
Geological materials
Semiconductors
2.1.3 Acoustic, radiative and microwave sensing
Figure 2-2 Ultrasound velocity profiler
35
Acoustic sensing techniques are also widely adopted in the monitoring of MPFs.
Different approaches have been developed by researchers. The conventional method is
investigating the propagation and attenuation of sound wave, which relate to component
fraction and flow regime of the MPFs[32, 33]. However, in the measurement of bubbly
MPFs, the accuracy of conventional acoustic sensing techniques highly depends on the
ratio between sound wave length and the size of the bubbles. Hence the research on
ultrasonic sensing techniques were developed. Takeda proposed the ultrasonic velocity
profiler by emitting an ultrasound pulse with a transducer and receiving the reflected
echo with the same transducer, as shown in Figure 2-2[34]. The velocity profile can be
obtained by analysing the delay of the reflected signal and Doppler shift frequency.
Similar measuring principle can be adopted to the measurement of the bubble velocity
in gas-liquid system and solid concentration in liquid-solid system[35, 36].
Radiative sensing techniques are also very popular in the measurement of MPFs
because of their high accuracy and efficiency. There are two mainstream radiative
sensing approaches, namely densitometry and radiography. The main objective to apply
radiative densitometry is to measure the component fraction in the MPFs. The common
approach to implement the measurement is to place a radiation source on one side of the
pipe with MPFs inside and measure the intensity change of radiation beam after it
passes through the flow, as shown in Figure 2-3[37].
36
Figure 2-3 Common approach to implement radiative sensing on MPFs
Bishop and James[38] applied neural networks analysis in the detection of air/oil/water
three-phase flow with dual energy gamma densitometry to avoid complex and difficult
flow modelling for phase configuration. Further work of flow regime identification in
combination with gamma-ray attenuation is carried out by Salgado and Pereira[39].
However, to achieve more accurate test results with radiation measurement methods, it
would require either a longer measurement period or a more intense radiation
source[40].
Figure 2-4 Schematic of radiography system with multiple radiative sources and 2D
detectors
On the other hand, radiography aims at visualizing and measuring the phase distribution
in MPFs. One typical example is X-ray radiography which is not only widely used in
multi-phase flow measurement but also in various non-destructive testing (NDT) and
medical applications[41]. The measurement approach is similar to that of the radiative
densitometry, however multiple 2-dimentional (2D) detectors are adopted to record the
37
X-ray attenuation map of the target object generated with multiple X-ray beam
sources[42]. An example of the schematic of radiography system is shown in Figure 2-4.
In addition, it is worth mentioning neutron radiography as another novel radiography
technique, which gives rise to the capability of imaging materials that are difficult to be
visualized by X-rays and gamma-rays as the attenuation and diffraction characteristics
of neutron rays are peculiar[43, 44].
Microwave sensing techniques are used in many commercial multi-phase flow meters
(MFMs)[45]. The basis of the microwave technique is that the transmission of
microwaves through MPFs, e.g. mixture of water, oil and gas, is affected by both the
fraction of water and the thickness of the liquids[46]. By measuring amplitude and
phase change of a microwave signal, the component fractions of the flow can be
investigated[47]. Moreover, Wu and etc. developed the microwave tomographic system
with microwave sensors arranged around the pipe, similar to that of electrical
tomography, to generate cross-sectional images based on the measurements of the
scattered microwave field[48, 49].
It is understandable that some other sensing techniques for multi-phase flow might be
omitted in this section. The objective of this section is to give a general understanding
of the current status in the monitoring and analysis of MPFs. Some existing techniques
developed exclusively for the targeted industrial processes in this research (CIP and oil-
water separation) will be introduced in later sections.
38
2.2 Monitoring CIP with Electrical Resistance Tomography
(ERT)
2.2.1 State-of-the-art in monitoring CIP
Figure 2-5 Schematics of ultrasonic techniques in fouling detection. (a) Pulse echo
technique; (b) Transmission technique.
The most crucial aim in monitoring CIP process is to detect the existence of fouling on
the interior pipe wall. Several ultrasonic and acoustic methods for fouling detection
have been tried by other researchers. Two approaches using ultrasonic sensing
technique have been proposed to determine the thickness of the fouling, namely pulse
echo technique and transmission technique[50]. The pulse echo technique detects the
reflected echo generated by the same ultrasonic transducer and the delay of the signal is
related to the boundary height in the target zone, whereas the transmission technique
applies a transmitter and a receiver on the opposite side of the pipe to allow the
ultrasound travel through the sensing region and investigate the existence of fouling by
analysing the attenuation of the signal. The schematics of the above two techniques are
illustrated in Figure 2-5[51, 52]. However, the accuracy and signal strength would be
significantly affected by the shape of the fouling layer, air bubbles and solid particles in
the flow. In addition, the cost of instrumentation limits the possibility of pervasive
applications in practical applications.
39
Another sensing technique based on acoustic vibration was developed for fouling
detection. The basis of this technique is that fouling layers on the interior pipe wall
would change the acoustic resonance frequency of the pipe during vibration[53]. The
measurement can be implemented by planting a vibration source on the interior pipe
wall and detect the acoustic wave with a microphone nearby. This method shows great
advantage with vessels of complex structure, such as plate heat exchangers. However,
this technique still suffers from its invasive nature[50].
Electrolyte–insulator–semiconductor (EIS) pH sensor has been adopted by researchers
to judge the completion of CIP process. However, this method can only monitor the
flow near the surface of the semiconductor which might lead to false reading of the
completion stage of the whole process[54].
Chen and Li etc. developed a method of detecting the milk fouling by measuring the
electrical resistance of the liquid and fouling layer between two parallel stainless steel
electrode installed on the pipe wall[55]. However, this technique is only capable of
identifying the existence of fouling in the sensing area, and no such information is given
regarding the location of the fouling on the pipe wall.
2.2.2 Advantages of ERT in monitoring CIP
Comparing to existing techniques for fouling detection, the significant advantages of
ERT can be concludes as follows:
1. Relatively low cost and immunity to the interference from bubbles and solid
particles, comparing to ultrasonic techniques;
2. Non-invasive nature, comparing to acoustic vibration techniques;
40
3. The ability to give comprehensive phase distribution information over the cross-
section of sensing area, comparing to EIS pH sensor and electrical resistance
measurement.
2.3 Measuring industrial oil-water batch separation process
with Differential Electromagnetic Inductive Sensor (DEMIS)
2.3.1 State-of-the-art in measuring industrial oil-water batch
separation process
Various techniques have been developed for the monitoring of oil-saline separation
process. The techniques can be concluded as three main types, the mechanical methods,
the ultrasonic sensors and the segmented sensor probes.
2.3.1.1 Mechanical methods
Figure 2-6 Monitoring oil-water separation process with sight glass
One conventional mechanical method involves the use of a sight glass, and the
inspection of oil-water level is conducted by human eyes[56]. Figure 2-6 illustrates a
typical sight glass installed on a separation vessel. The sight glass is vertically aligned
and connected to the separation vessel with two pipes. The oil-water interface height in
41
the sight glass is the same with that in the separation vessel and hence can be inspected
visually.
However, impurities such as wax and scale in the multiphase flow could coat the wall
and obstruct the interface observation through the sight glass. Possibility also exists that
the pipes connect sight glass with main separation vessel could be obstructed with
impurities mentioned above.
Another technique takes advantage of mechanical sensors which, for example, employs
a level gauge with a sliding displacer of chosen density between oil and saline[57, 58].
The performance of this technique would still tend to be affected by the cleanness of the
fluid, as the displacer could be immobilised[59].
Figure 2-7 Monitoring oil-water separation process with motive pressure sensor
Researches on single traversing pressure component driven by a motion system were
reported in[60]. The system mainly consists of a pressure sensor and a motion control
unit, which are connected by a flexible cable (Figure 2-7). The motion control unit
42
controls the vertical position of the pressure sensor in the separation vessel, in order to
detect the vertical pressure distribution and locate the oil-water interface.
One problem of the system was that the wear and tear of motive components would
reduce the stability for online monitoring and increase the maintenance cost.
2.3.1.2 Ultrasonic sensor
Ultrasonic sensor has the virtue of being contactless and non-invasive. The technique
involves one pair or multiple pairs of ultrasonic transmitter and receiver installed on the
separation vessel. The transmitter generates pulses that travel through the liquid system
and the receiver receives the reflected acoustic signal from the interface between
different phases[61]. The received signal is a function of the liquid densities and the
travelling speed of pulse echo in the liquids, hence the interface levels can be located.
An example of the function of ultrasonic sensor in monitoring oil-water separation
process is shown in Figure 2-8.
Figure 2-8. Monitoring oil-water separation process with ultrasonic sensor
43
It is worth mentioning that an ultrasonic interface level detector was developed in
Norway Christian Michelsen Research (CMR) in 2005[62]. The proposed ultrasonic
sensor was attached to the outside bottom of separator vessel and significantly reduced
the time and complexity of installation. The pulse echo travels through the vessel wall
as well as the liquid system. However, the accuracy of all ultrasonic sensors may drop
significantly due to signal attenuations caused by the presence of air bubbles, foams and
emulsions in the system[63].
2.3.1.3 Electrical Tomography
Electrical Capacitance Tomography (ECT) in particular was tested by Isaksenet et. al.
for interface measurements[64]. The basic principle of ECT is to construct the cross-
sectional images of permittivity distribution based on the inter-electrode mutual
capacitance measurements of electrode pairs surrounding the process vessel[65, 66].
The main limitation of ECT for the application of oil-saline separation monitoring is
that the presence of a conductive phase in the system has a tremendous impact on the
quality of the reconstructed images as it would strongly interfere electrical signals
acquired by the capacitive electrodes[67].
Bennett and William investigated the possibility of applying Electrical Resistance
Tomography to the monitoring of oil-water separation[68]. However, ERT
measurement requires the existence of conductive path between all electrode pairs.
Hence the attempt was only carried out on the deoiling hydrocyclone in which the
sensing region is fully occupied by liquids.
2.3.1.4 Segmented sensor array
Vertically segmented sensor arrays are also widely adopted in industrial applications.
The basic idea of this type of sensors is by combining relevant local information from
44
each individual sensor cell to achieve comprehensive spatial phase distribution
information. A common approach of adopting segmented sensor array in the monitoring
of oil-water separation process is shown in Figure 2-9.
Figure 2-9 Monitoring oil-water separation process with segmented sensor array
There are four commonly used sensor types, and they are nucleonic sensors, pressure
sensors, capacitance sensors and inductive sensors.
The nucleonic density measurement is considered to be very reliable and accurate in
monitoring the interface level of oil-saline separation process[67]. The high
performance is achieved by the technique of dual-energy gamma densitometry, which
translates the attenuation information corresponding to the absorption coefficient of
materials into density profiles[69, 70]. However, the hazardous nature of gamma
radiation leads to extra health and environmental concerns[71].
With respect to pressure sensors, local liquid density profiles are acquired by
eliminating liquid height, and gauging pressure components from the output signal[72].
45
In practical applications, however, the measurement accuracy is likely to degrade when
the sensor surface is covered by wax or scale impurities in the flow.
In terms of the segmented capacitance sensors, various attempts have been made. The
basis of segmented capacitance sensors is to measure the capacitance between the two
capacitive plates installed in each sensor to acquire local permittivity information[62,
73]. However, the capacitive plates still suffer from the fouling layer built up by the
impurities and the space between the plates may eventually be blocked.
As another electrical sensing methodology, electromagnetic inductive sensors are
widely applied to the level measurement of single phase conductive liquid, e.g.
monitoring sea level[74]. A typical inductive sensor system consists of a transmitting
coil and a receiving coil. Alternating current is injected into the transmitting coil which
generates a primary magnetic field. Eddy currents will be induced within the conductive
liquid giving rise to a secondary field which can be detected by the receiving coil. Based
upon the induced voltage of the receiving coil, the information of conductivity can be
deduced[75]. In order to measure the multiple interface levels that typically exist in a
primary oil separator, inductive sensors are distributed evenly along the vertical probe
from which sufficient spatial impedance distribution information can be acquired. One
example is the Inductive Level Monitoring System (ILMS) developed by the ABB
Group[76]. Although the proposed sensor probes were in direct contact with the fluid,
impurities hardly had significant effect on the measurement accuracy. However, the
saline, if leaking into the wires, would short-circuit the sensor and cause the
malfunction of the entire system.
46
2.3.2 Advantages of DEMIS in monitoring oil separation process
In this research, we present a non-intrusive and non-invasive differential
electromagnetic inductive sensor (DEMIS) for monitoring oil-saline separation process.
The drawbacks of existing techniques have been illustrated in the last section.
Comparing to existing inductive sensing system, there are two main advantages of
DEMIS. The first and most important advantage is that the DEMIS locates outside the
separation vessel so the direct contact with fluids and the potential risk of short-circuit
caused by leaking are eliminated. Moreover, the sensor is free from contamination
caused by the impurities in the liquids. In addition, the relatively larger number of
sensors for existing segmented inductive sensor system would indicate a more complex
data acquisition and processing system is needed, while the DEMIS system requires less.
47
Chapter 3 Monitoring CIP Using Electrical Resistance
Tomography (ERT) with Dynamic Reference
In this chapter, Electrical Resistance Tomography (ERT) is used as a new approach to
analyse the comprehensive spatial and time-varying conductivity changes during a CIP
process, aiming at locating the most difficult point to be cleaned in the pipe circuit and
also the ending point of the whole cleaning process. The chapter starts with introducing
the fundamentals of ERT, including the definition of forward problem and inverse
problem, and some conventional algorithms for imaging reconstruction. Then the
experimental setup and procedure for ERT measurement during CIP process are
explained, as well as the analytical principles for identifying the ending point of
cleaning process and evaluating the overall performance of reconstructed images in
comparison with practical conductivity distribution in the pipe. With the measured
voltages, preliminary images were reconstructed with conventional algorithms and the
merits and drawbacks were pointed out. Hence an optimization method was proposed to
overcome the corresponding drawbacks and the optimized results were compared with
the original ones. In the end, results achieved from the optimized algorithm under
different cleaning flow rate and pipe geometries are also presented to justify the
suitability of the optimization under different circumstances.
3.1 Fundamentals of ERT
3.1.1 Sensitivity and forward problem
3.1.1.1 Sensitivity
Geselowitz [77] and Lehr [78] derived the equation of how mutual impedance measured
by a four-electrode arrangement changes in response to the conductivity change in a
48
volume conductor. Based on their achievements, Murai and Kagawa presented the basic
theorem of sensitivity in Electrical Impedance Tomography as follows[79]:
Equation 3-1
𝑆𝑘,𝑗 = −∫ �� 𝜑 ∙ �� 𝜓𝑑𝑎 ≈ −�� 𝜑 ∙ �� 𝜓
k stands for pixel number in the reconstructed image. j stands for voltage projection
formed by two electrode pairs. �� 𝜑 and �� 𝜓 are the electric field vectors in the
corresponding pixel when one of the electrode pairs is acting as the transmitter injecting
current into the field, while the other electrode pair is acting as the receiver, from which
the boundary voltage measurement is taken.
To acquire the electric field strength under each projection, a two-dimensional
simulation model (Figure 3-1) is constructed according to the size of sensors and pipe
diameter in practical experiments with Ansys Maxwell (electromagnetic simulation
toolkit).
Figure 3-1 Electromagnetic simulation model for sensitivity matrix calculation
49
The simulation model consists of three parts.
1. Polyester pipe wall with an inner diameter of 38.1 mm and an outer diameter of
43.1 mm;
2. 16 copper electrodes attached to the inner pipe wall, where electrode 1 locates at
3 o'clock, electrodes 2 to 16 are sequenced in anticlockwise direction evenly;
3. Test region filled with tap water (conductivity 0.06 S/m).
The mesh number in the test region is 10000 and the voltage across the excitation
electrode pair is 5 V.
Figure 3-2 Sensitivity distributions in part of the projections
With the aid of field calculator in the software and MATLAB programming, the electric
field strength under each projection can be achieved and the sensitivity distribution
under each projection can be calculated through Equation 3-1. Figure 3-2 displays the
sensitivity distributions under some of the projections.
50
The sensitivity values decrease with the distance to the electrodes, as the electric field
strength decreases. In addition, the sensitivity distributions are determined by the
geometry of test region and the distribution and shapes of electrodes. In practical ERT
tests, the electrodes and test region are always fixed, which indicates all measurements
taken by ERT planes with same geometry and sensor structure share the same
sensitivity distributions.
3.1.1.2 Forward problem
The core equation of the forward problem based on the sensitivity theorem is in [80].For
the measurement on projection j,
Equation 3-2
∆𝑉𝑗
𝑉𝑅,𝑗≈ −
∑ ∆𝜎𝑘 ∙ 𝑠𝑗,𝑘𝑤𝑘=1
∑ 𝜎𝑅,𝑘 ∙ 𝑠𝑗,𝑘𝑤𝑘=1
where w is the total number of pixels in the reconstructed image; 𝑠𝑗,𝑘 is the sensitivity of
kth pixel in projection j; 𝑉𝑅,𝑗 and 𝜎𝑅,𝑘 are the reference voltage in projection j and the
reference conductivity of pixel k respectively; ∆𝑉𝑗 and ∆𝜎𝑘 are the voltage change in
projection j and the conductivity change of pixel k.
This explains how changes in the conductivity distribution lead to corresponding
changes in the measured boundary voltages. The reference background is normally a
homogeneous background with a fixed conductivity. In this research, the reference
background material is tap water with the conductivity of 0.06 S/m. Hence the equation
can be derived as:
Equation 3-3
∆Vj
VR,j≈ −
1
σR∙∑ ∆σk ∙ sj,k
wk=1
∑ sj,kwk=1
51
Where σR is the reference conductivity. For Jj,k =sj,k
∑ sj,kwk=1
, Equation 3-3 can be derived
as:
Equation 3-4
∆Vj
VR,j≈ −
1
σR∙ ∑ (Jj,k
w
k=1∙ ∆σk) = −
1
σR∙ [Jj,1, Jj,2 ⋯Jj,w] ∗ [
∆σ1
∆σ2
⋮∆σw
]
For a total number of p projections, the voltage change vector can be presented as:
Equation 3-5
Vp∗1 =
[ ∆V1
VR,1
∆V2
VR,2
⋮∆Vp
VR,p]
= −1
σR∙
[ J1,1 J1,2 ⋯ J1,w
J2,1 J2,2 ⋱ J2,w
J3,1 J3,2 ⋯ J3,w
⋮ ⋮ ⋯ ⋮Jp,1 Jp,2 ⋯ Jp,w]
∗ [
∆σ1
∆σ2
⋮∆σw
]
Which could be simplified as:
Equation 3-6
𝑉𝑝∗1 = −𝐽𝑝∗𝑤 ∗ 𝜎𝑤∗1
where 𝐽𝑝∗𝑤 is the normalized sensitivity matrix,
Equation 3-7
𝐽𝑝∗𝑤 =
[ J1,1 J1,2 ⋯ J1,w
J2,1 J2,2 ⋱ J2,w
J3,1 J3,2 ⋯ J3,w
⋮ ⋮ ⋯ ⋮Jp,1 Jp,2 ⋯ Jp,w]
𝑉𝑝∗1is the voltage change vector,
Equation 3-8
52
𝑉𝑝∗1 = [∆𝑉1
𝑉𝑅,1,∆𝑉2
𝑉𝑅,2⋯
∆𝑉𝑝𝑉𝑅,𝑝
]
−1
𝜎𝑤∗1is the conductivity change vector,
Equation 3-9
𝜎𝑤∗1 = [∆𝜎1
𝜎𝑅,∆𝜎2
𝜎𝑅⋯
∆𝜎𝑤
𝜎𝑅]−1
3.1.2 Inverse problem and conventional algorithms
3.1.2.1 Inverse problem
The basic idea of inverse problem in ERT is to reveal the relationship between changes
of conductivity distribution in the field and measured boundary voltages.
Mathematically, this process is the inverse version of Equation 3-6:
Equation 3-10
𝜎𝑤∗1 = −𝐽𝑤∗𝑝
−1𝑉𝑝∗1
Due to the soft field nature of electrical tomography, this is a severely ill-posed inverse
problem because of the large condition number of sensitivity matrix (normally 106) and
insufficient number of independent measurements (less than the number of pixels)[81].
Hence regularizations need to be adopted to minimize errors in the inverse calculations.
3.1.2.2 Conventional algorithms
3.1.2.2.1 Linear Back Projection (LBP)
LBP is one of the first algorithms adopted in the attempt to reconstruct images for the
electrical tomography. It is less accurate but with rapid response speed for computing
because of its simplicity. The basic idea is to consider the sensitivity matrix 𝐽 as a linear
53
mapping from conductivity vector space to boundary voltage vector space (which is not
true for the soft field domain), so the transposed sensitivity matrix 𝐽�� can be considered
as the related mapping from boundary voltage vector space to conductivity vector
space[21]. Hence,
Equation 3-11
𝜎𝑤∗1 = −𝐽𝑤∗𝑝
𝑇𝑉𝑝∗1
LBP is not chosen as the main algorithm in this application because the calculation is
based on an assumption which will reduce the accuracy of the inverse results and the
prediction of the ending point of the cleaning process requires accurate inverse results.
This will be proved in the Section 3.3.1.
3.1.2.2.2 Iterative algorithms
There are numerous existing iterative algorithms for ill-posed inverse problems, such as
Gauss-Newton, Landweber, conjugate gradient and etc.[80, 82, 83]. The common
approach for iterative algorithms is to reduce errors and converge the result to the true
solution through iterative calculations[84]. Iterative algorithms are very effective in
static experiments as they can trade off computing time to minimize the error in the
result. However, they are not adopted in this application because the potentially long
computing time lowers the efficiency for real time online tests.
3.1.2.2.3 Tikhonov regularization
The main objectives of employing regularization in ill-posed problems are to impose the
prior assumptions, which are the reference values in the case of ERT, on the solution
and at the same time filter out the high-frequency components of the solution, which
correspond to the smallest singular values of the sensitivity matrix[85]. However, when
54
there is a significant difference between the prior assumption and the solution, the latter
could be seriously distorted after regularization is adopted.
Tikhonov regularization is a widely used algorithm for soft field tomography. The
explicit formula pattern is[27]:
Equation 3-12
𝜎𝛼 = (𝐽𝑇𝐽 + 𝛼2𝐼)−1𝐽
𝑇𝑉
𝐽𝑇
stands for the transpose of the sensitivity matrix and 𝐼 is an identity matrix.𝛼2 is
known as the regularization parameter, which controls the convergence level of the
result. We use 𝛼 square to imply that it should always be positive.
Tikhonov regularization is chosen to be the basic algorithms in this research because the
main objective of monitoring CIP is to find out the ending point which in principle is
defined as the point when the last fragment of soil is cleaned. This implies that the
solution values would be very close to the reference values, i.e. the conductivity
distribution in the test region is close to the reference background, under which
circumstance regularization method is most suitable. However, in the earlier stage of the
cleaning process where the test region is dominated with soil of high conductivity,
significant difference lies between the average conductivity value of testing region and
that of the reference background. In this case, it is essential to carry out corresponding
optimizations on conventional Tikhonov regularization method in order to fulfil
comprehensive analytical purposes.
55
3.2 Experimental setup and principles
3.2.1 Experimental setup
The lab CIP circuit is shown in Figure 3-3. The left end of the testing area is connected
to the soil tank and water tap through a three-way valve from which the cleaning flow
rate can be adjusted. In the current study, a non-Newtonian, shear thinning shampoo
was used as the soil. The right end leads to the drain with a valve. There are 2
removable test areas into which transparent pipes with different geometry can be
installed.
Figure 3-3 Simulated CIP circuit
In the present research, two different pipe geometries are studied. A ‘T-shape’ pipe
shown in Figure 3-4 is installed in one of the test areas and the other test area is placed
with a normal straight pipe. A total of four ERT planes, each with 16 electrodes, are
installed in the test pipe Planes 1 to 3 are in the straight pipe section and plane 4 is in
the sealed concave bottom, which in principle should be the most difficult one to be
cleaned. Electrodes on the planes are connected to a commercial ERT system in order
to implement the measurements.
56
Figure 3-4 'T-shape' pipe with the 4 ERT planes installed
The second pipe sample is a 1.5 inches straight pipe with a butterfly valve installed in
the middle (Figure 3-5 and 3-6). This is mainly a 1.5 inches straight section, but with a
disc in the middle to mimic a 1.5 inches butterfly valve in a fully open position. The
geometry is equipped with four electrode planes, two at the upstream and two at the
downstream of the disc. Two inner planes are at the immediate upstream and
downstream of the disc respectively. Theoretically, the third plane at immediate
downstream of the valve should be the most difficult one to be cleaned.
Figure 3-5 Schematic diagram of a straight pipe with a butterfly valve installed inside
57
Figure 3-6 Straight pipe with a butterfly valve installed inside
3.2.2 Experimental procedures
The procedure of the simulated cleaning-in-place test can be concluded to steps as
follows:
Step 1: Switch the three-way valve to the branch connected to water tap and fully fill
the test regions with tap water. Then the ERT system is switched on to start current
injection and voltage measurement for 25 seconds while each group of complete
ERT measurement takes 1 second. The averaged measured voltage values within this
step are considered to be the original reference values to reduce the deviation caused
by measurement errors.
Step 2: Switch off the ERT system and switch the three-way valve to the soil tank.
Fill the test area with the soil. Then the ERT measurements will be reinitiated and
last for approximately 15 seconds.
Step 3: Switch the valve to water tap and start the CIP process.
Step 4: After the test regions are visually fully cleaned, keep measuring for 20
seconds.
58
In conclusion, the contents in the test regions to be measured under each step are tap
water, soil, the mixture of soil and tap water, tap water (Figure 3-7).
Figure 3-7 Contents to be measured in the test region under each step of the simulation
test.
3.2.3 Measurement protocol
The measurement protocol adopted in this research is Adjacent Electrode Pair strategy,
which was presented by Barber in 1984 and has been the most commonly used
measurement protocol for ERT[86]. It involves injecting current to a neighbouring
electrode pair and measuring the voltages across all other neighbouring electrode
pairs[87]. Table 3-1 shows an example of adopting adjacent strategy on an 16-electrode
ERT system.
Table 3-1 Adjacent strategy on 16-electrode ERT system
Current injection Measured electrode pairs Number of measurements
(1,2) (3,4), (4,5), …, (14,15), (15,16) 13
(2,3) (4,5), (5,6), …, (15,16), (16,1) 13
(3,4) (5,6), (6,7), …, (15,16), (16,1) 12
… … …
(13,14) (15,16), (16,1) 2
(14,15) (16,1) 1
Tap Water
Soil MixtureTap
Water
59
The number of independent measurements for a N-electrode ERT system is N(N-
3)/2[88]. Hence the total number of measurements for a 16-electrode system is 104.
3.2.4 Analytical principles
The average conductivity of the soil is approximately 4-6 S/m while that of tap water is
around 0.06 S/m. Therefore, the maximum conductivity value of each ERT frame, i.e.
the maximum pixel value in each reconstructed image is extracted independently as the
criterion of judging the remaining presence of the soil. The ending point of cleaning
process can be located when the maximum conductivity value equals to the reference
conductivity, which indicates the high conductivity component in test region (soil) is
fully cleaned. Also, the average conductivity value in each frame is calculated to
investigate the overall conductivity level in each experimental step described in Section
3.2.2.
3.3 Inverse calculation results with LBP and Tikhonov
regularization
366 frames of data were acquired for each ERT plane during the CIP process under the
water flow rate of 5400 L/h and with the injection current of 5.11 mA. The injection
current is set accordingly to the conductivity variation range in the test region to achieve
valid measured voltages. The imaging speed is one frame per second. The ending point
of cleaning process for the T junction plane, where the most difficult point in the pipe
was fully cleaned, was shown to be around frame 340 according to visual inspection.
3.3.1 Linear Back Projection (LBP)
The inverse solutions generated with LBP are shown in Figures 3-8 and Figure 3-9.
60
Figure 3-8 Maximum conductivity values in all ERT planes during the cleaning process
calculated with LBP
Figure 3-8 shows the maximum conductivity values of all four ERT planes during the
cleaning process. The corresponding stages are labelled across the curves according to
the experimental procedure in Section 3.2.2. The maximum conductivity values of all
four planes kept constant at the beginning of the test when reference measurements
were taken. The signal of Plane 4 (T junction plane) dropped and stayed steady around
frame 340, which complies with visually inspected result when the T junction was fully
cleaned. Meanwhile, the signal for other three planes at the straight pipe section
dropped rapidly after the cleaning process was initiated around frame 35. Hence, it
proved that the results are capable of indicating the ending point of the cleaning process
at the location of each ERT plane. However, the steady maximum conductivity values
of plane 2 and plane 3 slightly exceed the conductivity of tap water (0.06 S/m) when
they are fully cleaned. The maximum conductivity values of all four planes during step
two are lower than the actual soil conductivity (4-6 S/m).
Figure 3-9 demonstrates the average conductivity values of all four ERT planes
calculated with LBP during the cleaning process. The signals drift around the reference
conductivity value in a small range in Step 3, which is acceptable. However, at frame 25
61
when the component in the test regions was suddenly changed from tap water to the soil,
the average conductivity values of all 4 planes jumped below the reference value, which
did not reflect the actual processes of the experiment.
Figure 3-9 Maximum conductivity values in all ERT planes during the cleaning process
calculated with LBP
The results have proved that, despite the fact LBP is capable of locating the ending
point of the cleaning process, it is not suitable for the monitoring of CIP due to the lack
of accuracy and the poor ability to handle the significant change of component
conductivity in test region.
3.3.2 Tikhonov regularization
3.3.2.1 Selection of regularization parameter
The regularization parameter controls the weight given to the minimization of the side
constraint and an optimal regularization parameter should fairly balance the perturbation
error and the regularization error in the regularized solution[89]. Improper selection of
the parameter would either lead to over smoothing of the result or excessive error level.
Numerous methods of parameter selection exist in literature, but in practice, empirically
choosing the parameter is widely adopted. In this work, image reconstruction results
under different regularization parameters were compared with visual inspection results
62
during cleaning process to determine the choice of appropriate regularization parameter
(Table 3-2).
Table 3-2 Image reconstruction results under different regularization parameters during
cleaning process for Plane 4
Regularization
Parameters
Frame Numbers
10 100 200 300 333
0.0007
0.007
0.07
0.7
Colorbar
In Table 3-2, image reconstruction results of Plane 4 under four different regularization
parameters throughout the cleaning process are implemented. All images were
reconstructed under the same range of color bar. The results under same order of
magnitude of regularization parameter have neglectable differences, so only the results
under different orders of magnitude were compared. Images for 0.0007 and 0.007
indicate that the percentage of error generated is exorbitant and no valid information can
be identified from those images. Images of 0.7 have shown a typical over-smoothed
occasion which indicates that the regularization parameter is too large. Eventually, 0.07
is selected to be the proper regularization parameter adopted in latter study as the
images reconstructed can match with visually inspected results.
63
3.3.2.2 Results
The inverse solutions of all four ERT planes during the entire cleaning process
calculated with Tikhonov regularization are shown in Figure 3-10 and 3-11.
The maximum and average conductivity values in above figures are converged to a
much smaller range near reference conductivity value comparing to those of LBP while
the curves are spikier. But the accuracy of the solutions near and after ending point is
significantly improved. The maximum conductivity values after ending point of all 4
planes stay steady at the reference level (0.06 S/m) with minor drifting, which enables
the position of ending point to be identified clearly. As for the average conductivity
curves, they are also very spiky and vibrate within a small range before the ending
points. Similar to the case of LBP, they cannot represent the practical processes either.
The performance can be explained by the convergent nature of Tikhonov regularization.
When the conductivity of the target to be tested is significantly different from the
reference value, the result will be distorted. In this experiment, from frame 25 when the
background material was changed from water to soil, the inverse solution was severely
distorted. When the overall conductivity level trended near the reference level, the
inverse solutions became more and more accurate.
64
Figure 3-10 Maximum conductivity values in all ERT planes during the cleaning process
calculated with Tikhonov regularization
Figure 3-11 Average conductivity values in all ERT planes during the cleaning process
calculated with Tikhonov regularization
In conclusion, Tikhonov regularization is more suitable for monitoring the ending point
of the cleaning process. However, due to its nature of converging the results to reference
conductivity values, corresponding optimization methods need to be carried out to
achieve more accurate results during the entire cleaning process.
65
3.4 Algorithm optimization with dynamic reference
3.4.1 Methodology
Figure 3-12 Relationship between measured boundary voltage and material conductivity
under same projection
Figure 3-13 Measured boundary voltages (U curves) comparison between pure water and
pure soil.
As an example, two ERT planes with the same dimension and structure are filled
respectively by two uniform conductive liquids of different conductivities 𝜎1 and
𝜎2(Figure 3-12). The projection paths under the same electrode pairs in the two planes
have the same shape. Hence, with fixed injection current on the transmitter electrode
pairs, the ratio of the measured boundary voltage across a receiver electrode pair in
Plane 1 to that of the same positioned pair in Plane 2 will equal to the ratio of the liquid
conductivity in Plane 2 to that of Plane1.
Similarly, for other pairs of electrodes, the complete sets of boundary voltage
measurements taken from the two planes, which are also known to be the 'U curves',
0 20 40 60 80 1000
200
400
600
800
1000
1200
1400
Reference Voltage (Water)
Projection Number
Volta
ge
mV
0 20 40 60 80 1000
5
10
15
Measured Voltage (Soil)
Projection Number
Volta
ge
mV
66
should also be proportional. This can be proved by Figure 3-13. Despite the deviations
caused by system noise, the two curves basically stay in the same shape.
Table 3-3 The ratio of measured voltage for each projection between the soil and water
background
Current
Injection
Receiver Electrode Pairs
02-03 03-04 04-05 05-06 06-07 07-08 08-09 09-10 10-11 11-12 12-13 13-14 14-15 15-16
16-01 1 2 3 4 5 6 7 8 9 10 11 12 13
01-02 14 15 16 17 18 19 20 21 22 23 24 25 26
02-03 27 28 29 30 31 32 33 34 35 36 37 38
03-04 39 40 41 42 43 44 45 46 47 48 49
04-05 50 51 52 53 54 55 56 57 58 59
05-06 60 61 62 63 64 65 66 67 68
06-07 69 70 71 72 73 74 75 76
07-08 77 78 79 80 81 82 83
08-09 84 85 86 87 88 89
09-10 90 91 92 93 94
10-11 95 96 97 98
11-12 99 100 101
12-13 102 103
13-14 104
60<ratio<80
80<ratio<100 Average Ratio: 75
Ratio>100
For all the corresponding measured voltage values under all projections, we also
calculated the ratios between water and soil background and the results are shown in
Table 3-3. 85 out of 104 ratios stayed between 60-80, which is the actual ratio between
the conductivities of the soil and water in our investigation.
3.4.2 Optimization procedure
The cause of the distortions in previous results calculated with Tikhonov regularization
is the significant difference between the measured boundary voltages and reference
voltages. On the other hand, if the reference voltages can vary dynamically with the
change of background conductivity distribution to approach the level of measured
voltages, the distortions can hence be reduced or even eliminated.
67
The average voltage level can be determined in many ways. The method adopted in this
experiment can be concluded as follows:
a) Extracting the 16 voltage values corresponding to the measurements taken from
where the distances between transmitter electrode pair and receiver electrode pair are
the smallest and the accuracies are the highest. The choice of 16 voltage values are the
most accurate because the path of electric field is the shortest so as to reduce the error;
b) Reducing abnormal large values by limiting them to the average level of peak values
in the reference voltage;
c) The median value of the 16 measured voltages is adopted as the criterion to judge the
average voltage level. The median value is adopted instead of average value to reduce
the impact of extremely large values caused by measurement errors.
Based on above method, the ratio between the average level of the measured voltages
and the reference voltages can be determined:
Equation 3-13
𝛾 = Vmm/VRm
𝑉𝑚𝑚 is the median value of the 16 chosen voltages from measured boundary voltages
and 𝑉𝑅𝑚 is that of the reference voltages. Hence, the dynamic reference voltages
corresponding to each frame is:
Equation 3-14
𝑉𝑅∗ = 𝛾 ∙ 𝑉𝑅
Furthermore, the formula of optimized Tikhonov regularization with dynamic reference
can be derived as:
68
Equation 3-15
��𝛼∗ = (𝐽
𝑇𝐽 + 𝛼2𝐼)−1𝐽
𝑇𝑉∗
Where
Equation 3-16
��𝛼∗ = [
𝜎1 − 𝛾 ∙ 𝜎𝑅
𝛾 ∙ 𝜎𝑅,𝜎2 − 𝛾 ∙ 𝜎𝑅
𝛾 ∙ 𝜎𝑅⋯
𝜎𝑤 − 𝛾 ∙ 𝜎𝑅
𝛾 ∙ 𝜎𝑅]−1
Equation 3-17
𝑉∗ = [𝑉1 − 𝑉𝑅
∗
𝑉𝑅∗ ,
𝑉2 − 𝑉𝑅∗
𝑉𝑅∗ ⋯
𝑉𝑝 − 𝑉𝑅∗
𝑉𝑅∗ ]
−1
Hence the conductivity values calculated from Tikhonov regularization with dynamic
reference can be presented as:
Equation 3-18
��𝑤∗1 = [(��𝛼1∗ + 1) ∙ 𝛾 ∙ 𝜎𝑅, (��𝛼2
∗ + 1) ∙ 𝛾 ∙ 𝜎𝑅 ⋯(��𝛼𝑤∗ + 1) ∙ 𝛾 ∙ 𝜎𝑅]
3.5 Image reconstruction results calculated from Tikhonov
regularization with dynamic reference
3.5.1 Average conductivity values
Figure 3-14 illustrates the average conductivity values calculated from Tikhonov
regularization with dynamic reference in each frame. From frame 0 to around frame 25
when the testing regions were filled with tap water, the average conductivities of all
four ERT planes stayed at reference level (around 0.06 S/m). During frame 25 to 40, the
testing regions were filled by the soil with the average conductivity ranging from 4 to 6
S/m, as shown on the curves. The CIP process was then started from around frame 40.
The majority of the soil in straight pipe section (plane 1, 2 and 3) was flushed away and
the average conductivity dropped rapidly. The optimized ERT results indicate that plane
69
1 to 3 were cleaned around frame 40. As for plane 4, the majority of the soil still stayed
in the test region even after plane 1 to 3 were already fully cleaned. Then the amount of
soil starts to drop significantly from frame 50 until the plane is fully cleaned. In this
way, the average conductivity curves fully illustrate the whole process, which is an
important enhancement compared to conventional algorithms. This enables researchers
to analyse earlier stages of CIP.
Figure 3-14 Average conductivity values calculated from Tikhonov regularization with
dynamic reference
3.5.2 Maximum conductivity values
The maximum conductivity values of the four planes in each ERT frame calculated with
optimized Tikhonov regularization are shown in Figure 3-15 and Figure 3-16. Due to
system noise and the ill-conditioned nature of inverse problem, the maximum
conductivity values are slightly higher and spikier than the average conductivity values.
This is not critical as the main objective of analysing the maximum conductivity values
is to locate the ending point of the cleaning process. With the scale of Y axis narrowed
to 0~0.1 S/m, the ending point of all 4 planes can be clearly identified. Plane 1,2 and 3
in the straight pipe section were fully cleaned around frame 40, while plane 4 at the T
70
junction was fully cleaned around frame 340. Hence the ending point of the cleaning
process can be located at frame 340, which complies with the visually inspected result.
Figure 3-15 Maximum conductivity values calculated from Tikhonov regularization with
dynamic reference
Figure 3-16 Maximum conductivity values in smaller scale
3.5.3 Image reconstruction comparison
Table 3-4 illustrates the comparisons of reconstructed images between conventional
Tikhonov regularization and optimized Tikhonov regularization with dynamic reference.
Between frame 30 to 60, where the pipe is dominated by soil (highly conductive
71
component), the conventional Tikhonov regularization were not capable of providing
accurate reconstructed images as the average conductivity in the test region was
significantly different from the reference conductivity value. However, no significant
difference was discovered at later stage of the cleaning process (after frame 100) when
tiny amount of soil remained in the pipe and the average conductivity in the test region
was close to reference level. Above comparisons have shown the enhancement in image
reconstruction from the algorithm optimization at earlier stage of the cleaning process,
while the benefit of precisely locating the ending point still remained.
Table 3-4 Reconstructed image in Plane 4 comparisons between conventional and
optimized Tihonov regularization
Frame Numbers
10 30 60 100 200 300 333
Optimized
Algorithm
Conventional
Algorithm
72
3.6 Adopting Tikhonov regularization with dynamic reference
in different cleaning conditions
3.6.1 Results under higher flow rate with T pipe
Figure 3-17 Average and maximum conductivity values calculated from Tikhonov
Regularization with dynamic reference under higher flow rate. . a) Average conductivity
values; b) maximum conductivity values; c) maximum conductivity values in smaller scale.
The results of average and maximum conductivity values in Section 3.5 have shown the
capability of proposed optimization method with one set of measurement data from T
pipe against two conventional regularization algorithms. To validate its universal
feasibility, the optimized algorithm was applied to measurements taken under another
73
different flow rate, namely 8000 L/h. The visually inspected ending point of the
cleaning process (when the T junction was fully cleaned) was around Frame 170.
Results generated from Tikhonov regularization with dynamic reference are shown in
Figure 3-17.
Figures 3-17 a) is the average conductivity values in each frame, Figure 3-17 b) is the
maximum conductivity values in each frame and Figure 3-17 c) is Figure 3-17 b) in
smaller scale. Ending points and precise indication of conductivity level during the
whole process can still be observed from the two groups of optimized results,
respectively. Comparing with the 5400 L/h group, the ending point of the whole process
located at 170th frame, which is earlier than that in the 5400 L/h group. The above
results have proved the feasibility of adopting the optimization method with different
flow rates under the same pipe geometry.
3.6.2 Results with different pipe geometries
As the comparison group with T pipe under different flow rate has been implemented,
further investigation on a different pipe geometry was carried out as another
justification of universal feasibility study of this optimization method. The sample
group adopted a 1.5 inches straight pipe with a fully opened butterfly valve installed in
the middle (as introduced in Section 3.2.1). Four ERT planes were planted in the pipe,
two in the upstream of the valve and two in the downstream of the valve. The most
difficult ERT plane to be cleaned in principle is the third one (Plane 3) at immediate
downstream of the valve. Results calculated from the measurement data during the
cleaning process with the optimized Tikhonov regularization under two different flow
rates, namely 4100 L/h and 6200 L/h, are shown in Figures 3-18 and Figure 3-19.
74
Figure 3-18 Average and maximum conductivity values calculated from Tikhonov
Regularization with dynamic reference under 4100 L/h and butterfly valve. a) Average
conductivity values; b) maximum conductivity values; c) maximum conductivity values in
smaller scale.
The visually inspected ending point of the two cleaning process were around frame 75
and 70, respectively. The most difficult location to be cleaned indicated by the two
groups of results were both at Plane 3, which matched with theoretical assumption. The
ending point presented in Figure 3-18 c) and Figure 3-19 c) were around frame 75 and
70, which complied with the visually inspected result. For both groups of results, the
average conductivity values stayed between 5 to 7 S/m, despite very few spiky values.
75
The maximum conductivity values were slightly higher, but remained at reasonable
range. In conclusion, the results have proved that the optimization method is suitable for
ERT measurements in different pipe geometries.
Figure 3-19 Average and maximum conductivity values calculated from Tikhonov
Regularization with dynamic reference under 6200 L/h and butterfly valve. a) Average
conductivity values; b) maximum conductivity values; c) maximum conductivity values in
smaller scale.
76
3.7 Summary
In this chapter, a new approach to monitor and analyse the CIP process using ERT with
dynamic reference was proposed. With the fundamentals of ERT explained, three
conventional algorithms for the inverse problem calculation and image reconstruction
were compared, namely LBP, Tikhonov regularization and iterative algorithm.
Tikhonov regularization was proved to be more suitable in this research because of its
precision in locating the ending point of the cleaning process. However, the drawbacks
caused by the convergence nature of Tikhonov regularization should be eliminated to
achieve accurate inverse solution throughout the cleaning process, especially at the
earlier stage when the test regions were dominated by high conductivity components.
Thus, an optimization method with dynamic reference based on Tikhonov regularization
was proposed. The dynamic reference is generated by calculating the average level of
measured boundary voltages and correspondingly changing the reference voltages
proportionally. The result with optimized algorithm showed significant improvement
against conventional Tikhonov regularization. The merit of precisely locating the
ending point of the cleaning process remains, while the distortions caused by the
enormous conductivity difference between the testing region and reference. The
sharpness and reliability of the reconstructed images were also improved. In addition,
the results calculated with optimized algorithm and measured voltages under different
cleaning conditions, e.g. flow rate and pipe geometry, were compared to the visually
inspected results to justify the universal feasibility of the optimized algorithm in this
research.
However, drawbacks still exist in the optimization method. In the optimization
methodology, only uniform measured voltages, i.e. uniform U curves, were considered.
When the pipe is filled with roughly equal volume of soil and water, the shape of
77
measured voltage curve could be significantly different from the shape of original
reference voltage curve. Hence, the correspondingly generated reference will totally
distort the result. However, in this research, the majority of soil and mixture is flushed
away rapidly because of the high flow rate and the conductivity distribution within this
relatively short period is not the main objective, the drawback can thus be neglected
under this condition. If the dynamic reference optimization needs to be adopted in other
applications, this drawback should be considered in prior.
78
Chapter 4 Electromagnetic Simulations: Modelling
Three-Coil DEMIS and Oil-Saline Batch
Separation
In this chapter, the structure of the three-coil differential electromagnetic inductive
sensor was firstly introduced. In addition, a sensitivity analysis based on analytical
method is performed which gives rise to the optimized design of the sensor. Moreover,
an equivalent electrical model of the liquid-liquid separation model is derived based
upon which simulations are carried out to explain the evolution of sensor voltage with
respect to separation process. Finally, the simulation results are presented and discussed.
4.1 Introduction of differential electromagnetic inductive
sensor
4.1.1 Sensor structure
Figure 4-1 Schematic of differential electromagnetic inductive sensor
Comparing to conventional inductive sensors, the differential electromagnetic inductive
sensor consists of one cylindrical transmitting coil T and two cylindrical receiving coils
79
R1 and R2 (Figure 4-1). The three coils are aligned vertically, with the transmitting coil
located in the middle to form two symmetrical testing areas A1 and A2. The transmitting
coil is connected to the excitation signal of the impedance analyser. The two receiving
coils are connected to form a differential structure.
While functioning, the primary magnetic field generated by the alternating current in the
transmitting coil induces the second magnetic field in the materials. The phase and
magnitude of the secondary magnetic field is determined by the conductivity and
permittivity distribution in the sensing regions, leading to the change of the mutual
inductance between the transmitting coil and receiving coils. The output voltage
measured by impedance analyser can be presented as:
Equation 4-1
𝑉𝑜 = | 𝑉1 − 𝑉2|
Where V1 and V2 are the real parts of the induced complex voltages of the two receiving
coils, determined by the conductivity distributions in A1 and A2. When the distributions
in the two testing areas are the same, the output voltage after calibration ideally should
be zero. When the distributions in the two testing areas are different, the induced
voltages will become unbalanced, thus produce an output.
4.1.2 Sensitivity distribution
Rosell and etc. defined the relative sensitivity in a two-coil system as the relative
change at the receiver output produced by a conductivity perturbation[90]. The
mathematical form of the definition is:
Equation 4-2
𝑆𝑟 =(𝛥𝑉𝑉𝑒
)
𝛥𝜎
80
Where 𝛥𝑉 is the perturbation of the induced voltage, 𝑉𝑒 is the detected signal for the
empty space and 𝛥𝜎 is the perturbation of conductivity
Dyck and Lowther showed that the local conductivity sensitivity at a spatial point in the
testing area of a two-coil sensor system can be derived as[91]:
Equation 4-3
Sσ = ET ∙ ER
Where ET and ER
are the local electric field vectors when either of the two coils is
excited by unit current and the testing area is empty. Yin and Peyton considered the
motion of a conductive component in the system and its effect on the sensitivity
calculations[92]. In the current investigation, however, this effect is expected to be
negligible, since the velocity of the conductive medium in the separation process is
rather small. Xu et al. proposed a method of calculating the sensitivity distribution
based on field value extraction[93]. It is based on the calculation of electric field when
each of the coils is excited by unit current. Based on the definition of magnetic vector
potential 𝐴 (equation 4-4), the differential form of ampere circuital theorem (equation 4-
5) and the relationship between magnetic field strength �� and magnetic flux density ��
when the medium is evenly distributed in the field (equation 4-6), the magnetic vector
potential can be derived into the Poisson equation as shown in equation 4-7.
Equation 4-4
�� = 𝛻 × 𝐴
Equation 4-5
𝛻 × �� = 𝐽
Equation 4-6
81
�� = �� /𝜇
Equation 4-7
𝛻2𝐴 = −𝜇𝐽
In above equations, 𝜇 is the permeability of medium and 𝐽 is the current vector in the
coil. By splitting the excitation coil into certain numbers of elements and assuming that
the current element in each coil element can be presented in the form of 𝐼 · 𝑑𝑙 , where 𝐼
is the amplitude of the excitation current and 𝑑𝑙 is the position vector of the
corresponding segment of the coil, the Equation 4-7 can be derived as:
Equation 4-8
𝐴 =𝜇0
4𝜋∮
𝐼𝑑𝑙
𝑟
𝑙
𝜇0 is the vacuum permeability, and 𝑟 is the distance between the coil elements and the
target point in sensing region. When the electric and magnetic field is under sinusoidal
transformation, the differential form of Faraday Law of electromagnetic induction:
Equation 4-9
𝛻 × �� = −∂��
𝜕𝑡
can be derived as:
Equation 4-10
𝛻 × �� = −𝑗𝜔��
𝜔 is the excitation signal frequency. Hence according to equation 4-4, the electric field
at the target point can be calculated as:
Equation 4-11
82
�� = −𝑗𝜔𝐴
According to equation 4-3 and 4-11, the local conductivity sensitivity can be calculated
as:
Equation 4-12
Sσ = ET ∙ ER
= −𝜔2𝐴𝑇 · 𝐴𝑅
𝐴𝑇 and 𝐴𝑅
are the magnetic vector potential at the target point when the transmitting
coil or the receiving coil is excited by unit current, which can be calculated with
equation 4-8. Hence the sensitivity distribution in the sensing region can be achieved.
In this research, the sensitivity distribution of a simulated two-coil system with the coil
diameter of 150 mm and coil distance of 125 mm was studied. Both coils are divided
into 72 equal parts to calculate the current elements on them. The relative sensitivity
distribution of an axial cross-sectional testing area with the height and length of 150
mm was analysed. The cross-section was divided into 41×41 segments and the relative
sensitivity value of each segment is calculated. The result is shown in Figure 4-2.
Figure 4-2 Sensitivity distribution in the axial cross-sectional testing area of a simulated
two-coil sensor system
83
The above results have shown that the sensitivity value is relatively higher near the coils,
which indicates that the sensor output will be more sensitive to conductivity changes
near the coils. To implement the measurement of oil/water separation process in this
research, it is essential to achieve a homogeneous vertical sensitivity distribution in the
sensing region. Hence, corresponding optimization needs to be carried out.
4.1.3 Sensitivity distribution with different vessel radii
In order to analyse the spatial sensitivity distribution, the sensitivity values of all the
pixels on the same horizontal cross-section are added together to obtain the
corresponding planar sensitivity of the horizontal cross-section. The planar sensitivity
values represent the sensitivity of each horizontal plane with respect to the conductivity
change of homogeneously distributed material on the plane.
Equation 4-13
𝑆𝑘 = ∑𝑆𝑐𝑘
𝑛
𝑐=1
Where Sk is the planar sensitivity of the kth horizontal plane, Sck is the sensitivity of the
cth element in the kth plane and n is the number of elements adopted in analytical
solution. Figure 4-3 illustrates a simplified and normalized planar sensitivity
distribution profile between the transmitting coil and one of the receiving coils. The
sensing region being studied in this section is a cylinder with the height of 125mm and
the radius of 75mm. The cylindrical test region between the two coils in Figure 4-2 was
divided into 41 individual horizontal planes with corresponding planar sensitivity and
each plane is divided into 1681 segments.
84
Figure 4-3 Simplified and normalized planar sensitivity distribution
Figure 4-4 Planar sensitivity distribution under different testing vessel radii
The primary task is to reduce the variation of the planar sensitivity. Given a fixed size
of sensor coils, the central axial region has more uniform sensitivity than the peripheral
region of the coils. However, the absolute sensitivity is weaker in the central region. It
is therefore important to choose an appropriate vessel size. Planar sensitivity
85
distributions under different radii of the sensing region were calculated and are shown
in Figure 4-4, in which R stands for the radius of the coils.
It is clear that the variation of planar sensitivity distribution reduces with the testing
vessel radius, yet the absolute sensitivity value also drops. The average planar
sensitivity values were calculated and are listed in Table 4.1.
Table 4-1 The average and standard deviation of the sensitivity values under different test
region radii
Testing Vessel Radius 1/6R 2/6R 3/6R 4/6R 5/6R R
Average Sensitivity (V2/m
2) 0.1456 0.5731 1.2545 2.1428 3.1745 4.1515
Standard Deviation (V2/m
2) 0.0132 0.0443 0.0697 0.0787 0.2346 0.6611
To be more intuitive, the relationship of testing vessel radius, average sensitivity value
and sensitivity standard deviation are plotted in Figure 4-5. The standard deviations are
presented in the form of error bars.
Since we arbitrarily fix the size of the coil, the strategy in choosing the radius of the
vessel is to try to adopt a vessel with a smaller radius which, however, also needs to
satisfy the requirement of signal-to-noise ratio (SNR). In practical applications, the radii
of coils can be determined according to the radius of the separation vessel and the
optimum ratio achieved in this research. When the testing vessel radius reduced to 4/6
of the coil radius, the average sensitivity value is nearly halved while the deviation
range is nine times smaller than the original sensitivity distribution. In addition,
experimental tests showed that the average SNR with this testing vessel radius and the
FPGA-based impedance analyser (which will be introduced in Chapter 5) is around 60
dB which meet our measurement requirements. Hence, a mixing/separation vessel with
86
a diameter of 10 cm is selected for both the simulation model in later sections and
experimental setup in next chapter.
Figure 4-5 Average planar sensitivity values and standard deviation under different
testing area radii
4.2 Electrical model of liquid-liquid separation
4.2.1 Liquid-liquid separation model
A variety of liquid-liquid separation models have been proposed from either the
experimental or theoretical perspective[94]. Jeelani and Harland, in particular, have
detailed the basic principles governing the batch separation of oil and water[95]. In a
typical separation process, the liquid system can be divided into four zones, namely, a
clear oil zone, a dense-packed zone, a sedimentation zone and a clear water zone, as
shown in Figure 4-6 (a). The clear oil zone and clear water zone are occupied with a
single phase liquid, namely oil and water, respectively. The sedimentation zone is the
region where oil droplets are going through the buoyance process. After the buoyance
process, the oil droplets stack in the dense-packed zone and the bulk coalescence
87
process continues. The heights of the interfaces between clear oil zone and dense-
packed zone (hc), dense-packed zone and sedimentation zone (hp), sedimentation zone
and clear water zone (hs) are the main objectives to be investigated in this theory.
Figure 4-6 (a) 4 sections defined in oil/water separation system. (b) Height change of the
boundaries of 4 sections with time
In general, oil drops would ascend to form the dense-packed zone and the oil drops at
the top of the dense-packed zone would simultaneously coalesce to form the clear oil
zone. Therefore, the height of the interface between the clear oil zone and dense-packed
zone would decline, while the height of the interface between the sedimentation zone
and clear saline zone would increase with time. Since the speed of the sedimentation is
faster than that of the coalesce of oil drops, the height of the interface between the
dense-packed zone and sedimentation zone decreases until an inflection point ti when all
the oil drops are stacked in the dense-packed zone. After ti, the height of the dense-
packed zone gradually diminishes as oil drops continue to coalesce which gives rise to a
clear separation of oil and saline. The evolution of the interfaces between the liquid
zones can be depicted in Figure 4-6 (b).
88
Yu and Mao combined above theory with the lifetime distribution of drops of equal size
at oil-water interface and developed a mathematical model to present the changing trend
of interfaces as functions of time[96]. The mathematical equations are presented as:
Before the inflection point:
Equation 4-14
hs = v0t − (v0 − vi)𝑡2
t𝑖
After the inflection point:
Equation 4-15
hp2 = (1 − 𝜀0)H0 + (1 −1
𝜀p)H0𝜀0(1 + k1𝑡
𝑘2)𝑒−𝑘3𝑡𝑘4
During the entire separation process:
Equation 4-16
hc = H0(1 − 𝜀0) + H0𝜀0(1 + k1𝑡𝑘2)𝑒−𝑘3𝑡𝑘4
Where hs denotes the interface height between the clear water zone and the
sedimentation zone prior to the inflection point ti. hc denotes the height of the interface
between the clear oil zone and dense-packed zone throughout the whole separation
process. The height of the interface between the dense-packed zone and sedimentation
zone prior and anterior to the inflection point ti are denoted as hp1 and hp2, respectively.
H0 is the initial height of dispersion, ε0 is the initial oil hold-up fraction, εp is the oil
holdup fraction in the dense-packed zone. The inflection point is represented as ti. The
initial sedimentation velocity of oil drops and the sedimentation velocity of oil drops at
the inflection point are denoted as v0 and vi respectively. The parameters k1, k2, k3 and k4
are fitting constants without clear physical meanings and they could be obtained by
89
fitting the equations with practical experiment results. In addition, Jeelani and Hartland
proposed the mathematical equation of the interface height between dense-packed zone
and sedimentation zone before inflection point as[95]:
Equation 4-17
hp1 = H0 −H0 − hsi
ti𝑡
Where The height hs at ti is represented as hsi. Hence, all the interface heights of the
theoretical oil/saline separation model can be obtained once the parameters in the
equations are obtained.
4.2.2 Effective conductivity model
In order to understand the induced voltage of the receiving coil, an electrical model
needs to be derived. Specifically, the effective conductivity of the liquid zones is
essential in the calculation of the voltage of receiving coil. The conductivity of clear
saline σ1 in our investigation is kept at 4 S/m while clear oil is considered to be
nonconductive. The oil hold-up fraction in dense-packed zone εp is considered to be
constant during the separation process. The effective conductivity of liquid as a mixture
of liquids with two different conductivities σ1 and σ2 can be calculated using the
Maxwell Garnett mixing formula[97]:
Equation 4-18
σmp = σ1 + 3𝜀pσ1
σ2 − σ1
σ2 + 2σ1 − 𝜀p(σ2 − σ1)
Where σmp is the effective conductivity. As the conductivity of oil σ2 is zero, the
equation can be simplified as:
Equation 4-19
90
σmp =2 − 2𝜀p
2 + 𝜀pσ1
The effective conductivity in the sedimentation zone σmc can also be calculated via the
similar approach. However, calculating the oil fraction in the sedimentation zone εs is
more complex, as it changes with time. Before the inflection point, the thickness of the
dense-packed zone should be:
Equation 4-20
Δh1 = hc − hp1
The thickness of the clear oil layer is correspondingly:
Equation 4-21
H0il = H0 − hc
Assuming oil droplets are evenly distributed in all cross-sections of the testing area,
thus the oil fraction in the sedimentation zone can be obtained as:
Equation 4-22
𝜀s =[H0𝜀0 − (H0il + Δh1𝜀p)]
hp1 − hs
Combining equations 4-19 and 4-22, the time-varying effective conductivity value in
sedimentation zone can be calculated.
Therefore, the whole electrical model of the separation process is obtained. In practical
scenario, the conductivity of crude oil is considered to be zero, and the real time
conductivity of saline can be measured through the saline released from saline outlet of
the separation vessel.
91
4.3 Simulation of the electrical liquid-liquid separation model
Table 4-2 Experimental profiles and model parameters adopted from [96] for H0=300 mm,
D=154 mm, εp=0.65, and one hour agitation time
Experiment No. 1 No. 2 No. 3
Agitation speed (RPM) 350 350 500
ε0 0.3 0.5 0.5
v0 (mm/s) 1.1 0.51 0.25
vi (mm/s) 0.08 0.15 0.06
ti (s) 153.7 165.0 348.9
hsi (mm) 91.4 54.9 52.6
k1 ( S-k2) 4.1801×10-9 1.4149×10-4 2.3575×10-4
k2 3.2264 1.8231 1.4828
k3 ( S-k4) 2.1941×10-5 1.0743×10-3 1.4160×10-3
k4 1.9770 1.4273 1.1780
Before conducting experiments, it is of benefit to study the sensor output by simulation
based upon the electrical model in Section 4.2 in order to gain a better insight of the
separation process. The model relies on additional sensors to acquire the parameters in
the interface height functions. For example, a digital charge-coupled device (CCD)
camera may be needed to capture the initial sedimentation velocity of oil drops v0 and
the sedimentation velocity of oil drops at the inflection point vi. In this section, we adopt
the parameters given in [96] and they are listed in Table 4-2. The parameters were
evaluated based on practical experiments carried out in a mixing vessel with the
diameter of 154 mm and height of 300 mm. With the interface heights and effective
conductivity values determined, simulations were carried out for the corresponding
separation models and coil sensors in Ansoft Maxwell (electromagnetic simulation
toolkit).
An example of calculated interface heights as functions of time with the model
parameters from experiment No. 2 is shown in Figure 4-7.
92
Figure 4-7 Interface heights in a theoretical oil/saline separation model under 350RPM
and oil fraction at 50%
The inflection point lies at 165 seconds after separation process begins. Comparing
Figure 4-7 with Figure 4-6(b), it can be told that the calculated interface height curves
comply with basic oil/water separation theories.
Figure 4-8 Geometry of the simulation model. (a) 3D model; (b) 2D model.
93
The 3D simulation geometry is shown in Figure 4-8(a). For the models applied in the
simulations, the transmitting coil is excited with a current of 1 A at 1 MHz and the
conductivity of saline is 4 S/m. The maximum length of mesh elements is set to be 1.5
mm. To improve simulation efficiency, a 2D simulation model of was constructed
(Figure 4-8(b)). It is the axial cross-section of the test region and sensor coils in the
right quadrant of the YZ plane which is symmetrical to the Z axis. Then we used the
default function in Maxwell to generate the solution of the cylindrical model.
4.4 Simulation results and discussions
Figure 4-9 Induced voltage in both receiving coils and the differential output when oil
fraction is 50% and agitation speed is 350RPM
With the No.2 separation profile in Table 4-2, the simulated real parts of the induced
voltages (which indicates the distribution of conductivity components) during the
separation process in both receiving coils (V1 and V2 respectively) and the differential
output are shown in Figure 4-9.
It can be seen that voltage in the upper receiving coil decreases with time. This is
because in the upper testing zone A1 (see Figure 4-1), when a separation process starts,
94
the volume of saline would gradually decrease while the volume of oil would increase.
The increasing voltage in the lower receiving coil can be explained similarly. The
overall differential output of the sensor system would increase with time.
For comparison, simulations under different separation speeds were considered. The
separation speed of the liquid system is mainly dependent on the buoyance and
coalescence of the oil droplets. The buoyance velocity of an independent sphere oil
droplet in the continuous water phase can be calculated by Stokes' law[98]:
Equation 4-23
𝑣 =2
9∙(𝜌𝑤 − 𝜌𝑜)
𝜇𝑜∙ 𝑔 ∙ 𝑅2
Where g is gravitational acceleration, ρ0 and ρw are the mass density of oil and water
respectively, μo is the dynamic viscosity of oil and R is the radius of the oil droplet. This
equation indicates that the buoyance speed decreases as the oil droplet size decreases.
Coalescence process can be divided into two categories, which are the binary
coalescence while the oil droplets ascend, and the bulk coalescence in the dense-packed
layer [99]. Smaller oil droplet size at the initial stage will lead to lower bulk coalescing
rate. Hence when the oil/saline liquid system is agitated with higher rotational speed,
smaller oil droplets will be generated, and the separation speed should be slower.
Figure 4-10 compares the simulated differential sensor outputs of two separation
process, using profiles No. 2 and No. 3 in Table 4-2. The liquid systems were agitated
with same duration under 350 RPM and 500 RPM, with the oil fraction being fixed at
0.5. The results indicate that at the oil fraction of 0.5, the separation time can be
significantly influenced by the agitation speed and the simulation result complies with
above hypothesis.
95
Figure 4-10 Simulated sensor outputs of two separation process from the same liquid
system agitated with same time length and different speeds
Figure 4-11 Simulated sensor outputs of two separation process from the two liquid system
with different oil fractions agitated with same time length and speed
Figure 4-11 illustrates the simulated sensor outputs of two separation process using
profiles No. 1 and No. 2 in Table 4-2. The two liquid systems have different oil holdup
fractions, namely 0.5 and 0.3 and they were both agitated with the rotational speed of
350 RPM for one hour.
The inflection point of these two separation processes are similar (153.7 seconds and
165 seconds). However, by reading the differential sensor output, the separation process
96
under the oil fraction of 0.3 seems to end earlier than that under the oil fraction of 0.5.
This is because when the oil fraction is 0.3, the separation process would go through the
stage that the lower sensing area A2 is completely occupied with clear saline, while the
separation process carries on in upper sensing area A1. The difference of saline volumes
in the two sensing areas will remain the same and the main factor that determines the
output of sensor system is the distribution of saline in A1. Ideally if the sensitivity
distribution in both sensing regions are homogeneous, the differential sensor output will
stay constant. However, small variations still exist in sensitivity distribution which led
to the minor deviation of the output. In addition, the final stable value of differential
output is lower when the oil fraction is lower. This is because the difference of saline
volume in two testing areas is smaller.
It is also noticed that the curves in above two figures are not perfectly smooth. This is
caused by the mesh noise.
4.5 Summary
The objective of this chapter was to understand and simulate the function of differential
electromagnetic inductive sensor during the monitoring of oil/water separation process.
The structure of the sensor was firstly illustrated and the sensitivity distribution inside
the sensing region was calculated. The goal was to investigate the variation of vertical
sensitivity distribution so that corresponding optimization method could be applied.
Afterwards, mathematical oil/water separation models were studied based on the theory
of liquid interface height variation and Maxwell Garnett mixing formula. Detailed
model parameters under different mixing speeds and oil fractions from previous work
by other researchers were applied to the proposed models. At last, electromagnetic
97
simulations were carried out with separation models and simulated sensor to validate
the theoretical sensor output during different oil/water separation processes.
In the beginning, based on the concentric cylindrical structure of the sensor, the axial
cross-sectional sensitivity distribution was calculated based on analytical solution and
field value extraction. Afterwards, the vertical sensitivity distribution was simplified
into planar sensitivity distribution with the assumption of homogeneous distribution of
oil droplets on horizontal planes. The calculated planar sensitivity values showed that
the vertical distribution needs to be homogenized. Hence the approach of reducing
sensing region radius was proposed. Since this approach was a trade-off between
absolute sensitivity values and sensitivity variation range, vertical planar sensitivity
distributions under different sensing region radii were calculated and an optimum
choice of the radius was made. The sensitivity variation was reduced by nine times
while the average sensitivity value was halved. The SNR of the sensor output based on
the chosen radius of the sensing region and measurement and sensor systems adopted in
Chapter 5 was also tested and proved to meeting our requirements of measurement.
Then the theoretical oil/water separation process was investigated as well as the
mathematical equations that present the variation of liquid interfaces during the
separation process. By applying separation parameters fitted with practical separation
processes to the equations and calculating effective conductivity values in
corresponding zones, separation models that can be used in electromagnetic simulations
were constructed. Differential electromagnetic inductive sensor and separation models
with different pre-mixing conditions and oil fractions are simulated in the
electromagnetic simulation software package to establish theoretical output of the
98
sensor during separation processes and gain a better understand of how the change of
physical parameters lead to the change of sensor output.
The simulation results showed that the sensor is able to illustrate the progress of
oil/water separation process. The difference in separation speed could be identified
through the output signal and the output response to different oil fractions were also
presented. Hence the feasibility of carrying out practical measurement of oil/water
separation process with proposed sensor system was validated.
99
Chapter 5 Experimental System and Results In
Monitoring Oil-Water Separation With DEMIS
In this chapter, the design of the experimental system was firstly demonstrated.
Moreover, the validation of the system and experimental results were carried out.
Finally, experiments were conducted under different process conditions, e.g. agitation
speed, duration and oil-saline fraction, to assess the performance of the proposed system.
5.1 Experimental setup and testing strategy
Figure 5-1 Experimental system setup
The experimental system consists of three main parts, a differential electromagnetic
inductive sensor, field-programmable gate array (FPGA)-based impedance analyser, and
mixing and separation system (Figure 5-1).
100
5.1.1 Sensor system
The three coils of the differential electromagnetic sensor were winded around a
transparent plastic vessel with an outer diameter of 150 mm and a height of 300 mm.
The distance between coils is 125 mm with the transmitting coil locating in the middle
and the two receiving coils lying on the top and bottom, respectively. Each coil has five
turns of plastic-isolated copper wire. The transmitting coil was connected to the
excitation end of the FPGA-based impedance analyser and the two receiver coils were
connected together to form the differential structure and then connected to the
impedance analyser for the measurement of output signals.
5.1.2 Mixing and separation system
Figure 5-2 (a)Hardwood rod coated with black thermal plastic tube; (b) 3D-printed plastic
impeller
The mixing and separation system consists of 2 parts, a stirrer and a plastic baffled
vessel (Fig. 2). The stirrer (RZR 1, Heidolph UK) could achieve a maximum rotational
speed of 1700 RPM. The original rod and impeller attached to the stirrer are metallic,
which will interfere with inductance measurements. Hence, they are replaced with a
hardwood rod coated with thermal plastic and a 3D-printed plastic impeller respectively
(Figure 5-2). The diameter of impeller is 4cm and the length of rod is 30 cm. These
101
dimensions are chosen to make sure that the distance between the stirrer and coils is big
enough to minimise any possible interference.
The vessel is cylindrical with a diameter of 10cm. The vessel diameter is 2/3 of the coils
as the sensitivity distribution investigation in Section 4.1.3 suggests this is the optimal
choice for measurements. Four full-length baffles are installed evenly on the inner
vessel wall to avoid air entrapment and surface fluctuation during the mixing process.
5.1.3 FPGA-based impedance analyser
Figure 5-3 System block diagram
A custom digital instrument is developed for measuring the impedance changes of the
sensor due to magnetic induction. The instrument generates a sinusoidal signal for
sensor excitation; digital signal demodulation is implemented to obtain the in-phase and
quadrature components of the sensor response. The Zynq-7020 system on a chip (SoC)
is the backbone of the system; this chip integrates a Xilinx 7-series FPGA and an ARM
102
dual Cortex-A9 based processor[100]. The instrument exploits the chip capabilities by
implementing the signal generation and I/Q demodulation modules using the hardware
benefits of the FPGA. The ARM processor is used for data transferring between the
FPGA and a host computer.
The block diagram of the system is shown in Figure 5-3. The main elements of the
system include an FPGA for signal generation and I/Q demodulation, analogue-to-
digital (ADC) and digital-to-analogue (DAC) converters, a front-end circuitry, and a
host PC for data log, display and control.
The main elements of the Analogue/Digital block of Fig. 8 are the DAC and ADC
circuits. For the DAC, an AD9767 from Analog Devices is used; it has two 14-bit
outputs and up to 125 MSPS update rate. DAC outputs are followed by a differential-
current to differential-voltage conversion and low-pass filtering stage. For the ADC an
AD6645 from Analog Devices is used; it has 14-bit resolution and a maximum
sampling rate of 105 MSPS. The reason to choose 14-bit resolution ADC is because it
can satisfy the requirement for SNR (above 80 dB) with a relatively lower cost. The
parallel digital output of the ADC is directly interfaced with the Zynq-7020 FPGA. The
ADC input stage includes differential voltage translation; input swing range is ±0.55 V
centred at 2.4 V.
The front-end circuitry conditions and amplifies the excitation and measured signals.
The output of instrument at the last amplification stage is composed of a differential
pair of power amplifiers. Output voltage amplitude is 16 Vrms and fed to the excitation
coil. The detection circuitry includes a RF transformer followed by several differential
receivers to amplify the measured signal and feed it to the ADC. Input voltage
amplitude is in millivolt range ~4 mV. Up to four measurement channels can be
103
multiplexed and the signal gain is programmable thought the host PC. A similar
architecture was presented in [75] for conductive flow measurements.
For all experiments presented, excitation frequency is set to 1 MHz. The sampling
frequency is 100 MHz. Data rate output is 25,000 samples per second (I/Q data).
Samples are sent to a PC through an Ethernet link.
The custom instrument architecture gives two main advantages in author's opinion.
1) The hardware front-end can be customized according to the sensor/sample needs and
experimental setup. A power amplifier is integrated for sensor excitation (16 Vrms as
stated in manuscript). A millivolt range (~4 mV) input is expected. Integrating active
amplification stages to commercial instruments commonly degrades instrument
performance.
2) High data-rate output at high SNR for observing processes with different dynamics.
An FPGA is used as the core of the instrument for digital signal synthesis and
demodulation. FPGA can implement digital demodulation at high speed rates (100 MHz
as stated in the manuscript). This gives the possibility to capture the dynamics of the
process with great detail during all the stages. Commonly, there is a compromise
between the sample rate and the SNR, a good balance between these two figures is
achieved with the custom instrument.
5.1.4 Testing strategy
The experimental investigation started with the validation of the instrument by
measuring voltage signals, followed by the investigation of the vertical sensitivity
distribution in the sensing region. Then a test result example was presented and
compared with camera recorded saline interface height profile for validation purpose.
104
Finally, experiments with different agitation speeds, mixing durations and oil fractions
were conducted, and the results were discussed. To simulate practical industrial
applications, the liquid level has been kept constant throughout, at the same height as
that of the top receiving coil. Two different oil fractions, 50% and 33%, were
investigated respectively. Tests of higher oil fractions were not conducted as the
separation process under such conditions were difficult to visualise. Results of two
different agitation speeds (900 RPM and 1700 RPM) were compared under the 50% oil
fraction and only 1700 RPM was tested under the 33% oil fraction as the mixing
efficiency was too low at lower agitation speeds. For each agitation speed, three levels
of mixing time, namely 30 seconds, 5 minutes and 15 minutes were investigated, in
order to compare the separation processes under different droplet sizes. Each test with
different combination of agitation speeds and mixing times were repeated for three
times to verify the repeatability of the experiments.
5.1.5 Choice of parameters
In this section, the choices of some of the critical parameters in the designed system are
discussed.
(1) Optimization with reduced vessel/coil diameter ratio
From the analytical result of the sensitivity distribution in the last chapter, the
optimum vessel/coil diameter ratio of 2:3 is applied to the designed system.
With this optimization method, the variation of sensor output during a oil/water
separation process becomes more smooth and the suddenly change of signal
caused by heterogeneous sensitivity distribution is avoided. Hence, a better
correlation between sensor output signal and oil/water interface height change
can be achieved.
105
(2) Number of turns of the coils
In the proposed sensor system, there are five turns of each coil. With the
increasing number of turns, the induced voltage in receiving coils will be
enhanced. However, it will also reduce the resonance frequency of the coils. To
avoid the measurement frequency meets the resonance frequency, we chose
relatively fewer number of turns;
(3) Excitation frequency and voltage
The excitation frequency and voltage are the highest we can achieve from the
measurement system to achieve maximum signal strength.
5.2 Experiment and discussion
In this section, experiments with the proposed sensor and mixing-separation system
were undertaken. Firstly, the validation of the FPGA-based impedance analyser was
carried out and the sensitivity distribution in the designed sensor system was
investigated. Then the sensor output signal was linked to the saline interface height
change recorded by camera to validate the sensor output. Finally, differences of the
sensor outputs were compared to the changes of mixing conditions such as agitation
speed, mixing duration and oil fraction. Based upon these validations and comparisons,
the sensor’s capability of identifying the completion level and separation speed under
different process conditions was investigated.
The separation process of oil and saline highly relies on the oil droplet size distribution.
The oil droplet size distribution is determined by the agitation speed and mixing
duration. With higher agitation speed and longer mixing duration, the separation speed
is expected to be slower.
106
5.2.1 Instrument performance inspections
Figure 5-4 shows the instrument performance for measuring a voltage signal. The
figure shows the nominal value of the input signal in the ‘x’ axis; voltage range is in
millivolt range. Instrument measured values for the magnitude is plotted in the ‘y’ axis.
A proportional increase in the measurement can be observed for the corresponding
increase in the input signal. For this experiment an SNR greater than 85 dB is achieved
for all measurement points.
Figure 5-4 The instrument performance for measuring a voltage signal
5.2.2 Validation of sensitivity distribution
Experimental tests were carried out to evaluate the actual sensitivity distribution by
continuously adding same volume of saline into the separation vessel. The
corresponding output signal change is shown in Figure 5-5. According to Equation 4-1
and the sensitivity theory, when the planar sensitivity distribution is uniform, the sensor
107
outputs tend to increase proportionally before the interface of saline reaches the
transmitting coil plane and drop proportionally after the interface exceeds the plane. The
-shaped curve in Figure 5-5 implies that the vertical sensitivity distribution is uniform
enough for the experiment.
Figure 5-5 Experimental test result for vertical sensitivity distribution with saline
5.2.3 Experimental result example
Figure 5-6 shows the sensor output during a complete test cycle. The oil fraction in this
test was 50%. Initially (i.e., t = 0) the oil and saline stayed in two separate phases, with
the top sensing region A1 being filled with the oil and the bottom sensing region A2
being filled with the saline. At this stage the differential sensor output would be at
maximum. Mixing then started under the agitation speed of 1700RPM. It can be seen
that the output signal started to drop rapidly as the system was getting homogeneous,
and the conductivity distribution differences in the two sensing regions were
disappearing. After 30 seconds, mixing was stopped, and the oil/saline phases were
allowed to separate naturally for 500 seconds. As the separation process proceeded, the
difference of saline volume between A1 and A2 gradually increased, which led to the
0 500 1000 1500 20000.12
0.125
0.13
0.135
0.14
0.145
0.15
Saline volume (cm3)
Se
nso
r o
utp
ut
(V)
108
increase of the sensor output. When the oil/saline distribution recovered to the initial
level (i.e. oil and saline are completely separated), the sensor output stayed steady and
indicated that the separation process is finished.
Figure 5-6 Single test result under 1700RPM, 30 seconds mixing and 500 seconds
separation, oil fraction 50%
For analytical purpose, the separation part of the curve is plotted in Figure 5-7.
Figure 5-7 Sensor output signal during separation process
109
The trend and shape of the test result are generally consistent with the simulated
differential output result in Figure 4-9. However, differences still exist:
(a) Offset and variation range
The variation range of the test result (from 0.0002 V to 0.0016 V) is different to that of
the simulation result (from 0 V to 0.0006 V). This was caused by the differences in the
excitation signal and coil structure. In the simulations, the transmitting coil was excited
with unit current, while in the experiments, the transmitting coil was excited with 16
Vrms voltage. In addition, the practical coil structure is imperfect comparing to that in
the simulations.
When the oil/saline liquid system is homogeneous, the output of an ideal sensor system
excluding the influence from background materials should be zero, e.g. the simulation
results. However, the imperfection of practical coil structure and the background
materials together generated an offset to the measurements.
(b) Initial separation speed
At the initial stage of the experimental separation process, the signal increases slower
than simulation result. This is because in the simulations, the liquid system started the
batch separation process from steady state. However, when the separation process was
initiated in practical experiments, the liquid system suffers from an initial turbulence
from the terminated mixing process which slows down the separation process.
5.2.4 Validation of sensor output
A camera was set up to record the separation of the oil saline mixture while the sensor
output was logged simultaneously. The oil fraction was 50% in this experiment and the
initial height of the interface between oil and saline was 125 mm. Firstly, the liquid
110
system was mixed for 30 seconds under the agitation speed of 1700 RPM. Then mixing
was stopped to allow the mixture to separate for 270 seconds. Figure 5-8 shows some of
the screenshots from the recorded video during the separation process . A clear interface
between saline and the and oil/saline mixture can be observed in the video. Changes of
the interface height were measured at against time and the results were compared with
the sensor output in Figure 5-9.
Figure 5-8 Part of the screenshots from the recorded video during the separation process
Figure 5-9 Comparison between sensor output and saline interface height change during
the separation process
The interface height indicates the completion level of the separation process. From the
comparison in Figure 5-9, it is clear that the overall trend of sensor signal change is
consistent with the observed interface height change. The signal started to increase
when separation began and became steady when the separation process was
111
approaching the end. However, differences exist between the two curves. This could be
explained with the structure of oil-saline separation system mentioned in Section 4.2.
The liquid above clear saline includes oil and oil/saline mixture. The sensor output
reflects the distribution of both clear saline and the saline component in oil/saline
mixture, while the saline interface height change captured by camera only indicates the
level of clear saline in the vessel. Hence the difference of the two curves is caused by
the saline component in oil/saline mixture.
In conclusion, the comparison between the sensor output signal and visually observed
saline interface height change shows that the proposed instrument is capable of
detecting the final separation stage of the process. This serves as an initial validation of
the application of DEMIS in monitoring oil-saline separation. Further validation which
involves on-site facilities would be useful and we leave it for further studies.
5.2.5 Experiment results under different mixing conditions
5.2.5.1 1700 RPM agitation speed and 50% oil fraction
Figure 5-10 shows the variations of the sensor output as a function of time when the
agitation speed and oil fraction were fixed at 1700 RPM and 50% respectively. Tests
with three different mixing durations, namely, 30 seconds, 5 minutes and 15 minutes
were investigated. Each experiment was repeated for three times and the corresponding
sensor outputs were recorded. It can be seen from Figure 5-10 that the results in each
experiment are highly repeatable and consistent.
112
Figure 5-10 Sensor outputs of repeated experiments under agitation speed of 1700 RPM,
oil fraction 50%. (a) mixing duration 30 seconds; (b) mixing duration 5 minutes; (c)
mixing duration 15 minutes.
In Figure 5-11, the averaged sensor outputs of the three repeat experiments during the
separation stage for the three different mixing durations are compared. The standard
deviation on each time point is also calculated and presented on the curves as error bars.
The starting points of the three curves are rather close to each other, which indicates that
the liquid systems were all fully mixed and the volume distributions of oil were similar
in all three groups when the separation started. However, the difference in separation
speeds under different mixing durations can be observed clearly. This can be attributed
to the different oil droplet sizes. The longer the mixing duration, the smaller the average
113
droplet size, and the slower the separation speed. After the separation completed, the
sensor output value of the three groups stayed approximately at the same level as the
one before mixing started. This shows that the oil and saline recovered to the state
where they were completely separated.
Figure 5-11 Average sensor outputs of the repeated test under 1700 RPM during
separation process
5.2.5.2 900 RPM agitation speed and 50% oil fraction
Figure 5-12 Average sensor outputs of the repeated test under 900RPM during separation
process with error bars
114
The experiment results when the agitation speed was 900 RPM are shown in Figure 5-
12. In this case, at the beginning of the separation process, the average initial sensor
output values of the three groups were different. The initial value was larger when the
mixing duration was shorter. This is mainly because the liquid system was not fully
mixed and there was an oil layer remaining at the top of the mixing vessel, resulting in a
higher differential output from the sensors when the separation started. In addition, the
standard deviations of the repeat tests were also getting larger comparing to those of the
1700 RPM agitation speed. When the liquid system was not fully mixed, the oil droplet
size distribution would become much more spread out, which led to large variations of
the differential sensor output. It should be noted that the overall separation speed was
still slower when the mixing duration became longer, as the resulted average oil droplet
size was correspondingly smaller.
5.2.5.3 Results under 1700RPM agitation speed and a lower oil
fraction of 33%
Figure 5-13 presents the results with the agitation speed of 1700RPM but a lower oil
fraction of 33%.
Figure 5-13 Average sensor outputs of the repeated test at oil fraction of 33% under
900RPM during separation process with error bars
115
It can be seen that the results are also highly consistent with the simulation results. The
difference in separation speed among the three groups is the result of different mixing
time, which leads to different average oil droplet sizes. Same as the simulation results,
the curves of separation process enter stable stage earlier than that of the groups with
the oil fraction of 50%. The reason has already been explained in the simulation result
as when the lower testing area is filled with clear water, the change of sensor output will
depend on the saline phase distribution in the top sensing region.
5.3 Summary
In this chapter, the experimental sensor and measurement systems were designed and
investigated. Firstly, a brief introduction of the sensor system and mixing/separation
system was presented. Then the architecture and advantages of the FPGA based
impedance analyser were illustrated.
In the experiments and discussion section, the validations of the proposed system and
sensor output results were firstly described. The validation of measurement system was
carried out by measuring a voltage with known magnitude. The validation of the sensor
system was implemented by continuously adding same amount of saline and investigate
the sensor output change so as to validate that the vertical planar sensitivity distribution
is homogenous. Then the sensor output measured during a complete mixing and
separation process was presented and compared with the simulation results in Chapter 4.
The comparison indicated that the sensor output complied with simulated result, while
the differences between them were explained. In addition, the sensor output during the
separation process was compared with the interface height change recorded by a camera
synchronously. The comparison made between the sensor output signal and saline
interface height during the same separation process indicates that the proposed
116
instrument is capable of detecting the final separation stage of the process. At last,
experiments under different oil-saline fractions (33% and 50%), different agitation
speeds (900 RPM and 1700 RPM) and durations (0.5 minutes, 5 minutes and 15
minutes) have been conducted. Each experiment was repeated three times and the error
of the sensor outputs for repeated experiments was within 16%.
The validations showed satisfactory and reliable performance from the proposed system
for measuring the mixing and separation process of oil and saline. The sensor output
during separation processes under different agitation conditions were in good agreement
with the simulation results in Chapter 4. The results indicate that the proposed sensor
system is able to measure the separation process of oil and saline under different process
conditions. Considering the practical oil-saline separation as a continuous process, the
measurement information could be interpreted into the interface location information in
the separation vessel and saline and oil outlet speed can be adjusted accordingly to
guarantee product quality. In conclusion, the non-intrusive and non-invasive nature of
the electromagnetic inductive sensing technique suggests it is a promising method for
in-situ monitoring of oil-saline separations in industrial applications. To apply this
technique in real world application, there are a few problems that remains to be solved:
(1) As the DEMIS is mounted outside the separation vessel and there is a gap
between coils and the vessel, an optimum sensor structure needs to be designed
to fit it in a pilot-scale separator and corresponding electromagnetic isolation
method needs to be developed to prevent environment interference;
(2) When the scale goes up from lab scale to pilot scale, corresponding
optimizations need to be carried out in order to enhance the signals;
117
(3) Real world oil/water separation is a continuous process, hence a correlation
method between sensor output signal and oil/water interface height needs to be
developed.
118
Chapter 6 Conclusions and Future Works
This chapter presents the conclusions drawn from the simulation and experiment results
in previous chapters. Based on the conclusions, recommendations for possible future
works are illustrated.
6.1 Conclusions
The research conducted in this thesis concerns two industrial processes, namely the CIP
and oil-water separation process. The techniques adopted to analyse the two processes
are ERT and DEMIS, respectively.
6.1.1 Monitoring CIP using ERT with dynamic reference
The aim of monitoring and analysing CIP are to locate the most difficult section to be
cleaned and indicate the ending point of the whole process, in order to eliminate
unnecessary waste of water, chemical and waste-treatment cost.
According to our first research objective, ERT is adopted as the main technique to
monitor CIP in this research for its non-intrusive and non-hazardous nature while the
cost is relatively low. To implement ERT measurements, certain number of electrodes
(normally a group of eight or 16) are mounted on the interior pipe wall at the target pipe
sections. The measurement involves injecting current to certain electrode pairs and
measuring the voltage across all other neighbouring electrode pairs, which is defined as
‘adjacent strategy’. The complete set of measured voltage is then calculated using the
sensitivity matrix acquired from electromagnetic simulation and certain inverse problem
algorithms to obtain the conductivity distribution at the cross-section of the test region.
119
The second research objective is to visualize the component distribution inside the test
region to locate the most difficult position to be cleaned. The conductivity distribution
presents the location of the last soil residue in the test region which indicates the most
difficult section to be cleaned, and the evolution of maximum conductivity value
indicates the ending point of the cleaning process. The inverse calculation is severely
ill-posed, thus the proper choice of algorithm is critical to achieve the most accurate
result.
According to the third research objective, three mainstream inverse algorithms were
compared, namely LBP, Tikhonov regularization and iterative algorithms. Tikhonov
regularization showed significant advantage with satisfying performance near the d x
ending point and relatively low computational cost. However, with the reference
conductivity chosen to be the conductivity of tap water, the inverse calculation results
during the stage when the test regions were occupied by highly conductive component
(soil and the mixture of soil and water) was distorted due to the convergence nature of
Tikhonov regularization. Hence a novel optimization method was proposed to eliminate
the distortion. The optimization was implemented by configurating dynamic reference
voltages at all ERT frames corresponding to the level of measured voltages. The
dynamic reference voltages are proportional to the reference voltages measured when
the test region was filled with tap water, and the ratio between them was calculated
under certain strategy. The optimized results showed significant improvement under
high conductivity background, while the advantage of indicating the ending point of the
CIP process still remained. In addition, the optimized Tikhonov regularization with
dynamic reference was applied to the measurement data taken with different pipe
geometries and water flow rates. The consistent performance has proved the universal
feasibility of this optimized algorithm in this research.
120
6.1.2 Measuring oil-water separation process using DEMIS
The aim of monitoring industrial oil/water separation process is to ensure high
production efficiency and product consistency.
According to the first research objective, we proposed a novel sensing technique,
namely Differential Electromagnetic Inductive Sensor (DEMIS), which is capable of
monitoring the oil-water separation process non-intrusively and non-invasively with a
relatively low cost and complexity. The sensor is based on two differentially connected
cylindrical receiving coils and one cylindrical transmitting coil lies in the middle of
them. The three coils form two neighbouring test regions of which the sensor output is
determined by the difference between the mutual inductances. In this research, only the
conductive component of the mutual inductance (saline) was taken into consideration.
To implement the second research objective, the sensitivity distribution in the test
region was mapped using field vector extraction. It is soon discovered that the
sensitivity distribution needs to be homogenized as the sensitivity value is significantly
higher near the coils. A corresponding optimization method was proposed by decreasing
the ratio between the diameters of the separation vessel and the coils. As this method is
a trade-off between the sensitivity value and homogeneity, an optimum ratio was chosen
to satisfy both SNR and measurement accuracy of the system. Moreover, in order to
understand how the change in physical parameters of the liquid system leads to the
change in electrical sensor output, electrical simulations of monitoring oil-water
separation process with the proposed sensor system were conducted. An electrical
simulation model was constructed based on physical oil-water separation model. In the
physical liquid-liquid separation model, the liquid system was divided into four zones
based on the behaviour of oil droplets. The evolution of interfaces between different
121
zones can be presented as functions of time with parameters of the liquid system. The
oil fraction of each zone was also calculated and the equivalent conductivities of them
were expressed as functions of time using Maxwell Garnett mixing formula. In this way
the electrical simulation model of the oil-water separation system was implemented. By
adopting the parameters of oil-water separation system provided in other literatures,
simulations of the sensor output were carried out during separation processes under
different pre-mixing conditions, e.g. agitation speed, mixing duration and oil fraction.
The results complied with theoretical assumptions which proves the possibility of
monitoring oil-water separation process with the proposed sensor.
Based on the design and simulation of the sensor system, an experimental system was
implemented to accomplish the third research objective. The experimental system
consists of the sensor system, measurement system and mixing/separation system. The
structure of the sensor system is the same with the simulation model. The main
instrument adopted in the measurement system is a FPGA-based impedance analyser
and the architecture and advantages of which was well illustrated. The ratio between the
diameters of the coils and the vessel adopted in the mixing/separation system follows
the optimum ratio discussed in previous research. In the experimental section,
validations of the measurement instrument and the sensitivity distribution of the sensor
were firstly carried out. Then the sensor output during an oil-saline separation process
was validated by comparing it to the saline interface height change recorded with a
camera synchronously. At last, experiments under different pre-mixing conditions were
conducted. The validations and results have proved that the proposed sensing technique
is a promising method for in-situ monitoring of oil-water separation in industrial
applications.
122
6.2 Future work
Based on the conclusions drawn from this research, future works are suggested in
following aspects:
1. For the research in monitoring CIP using ERT with dynamic reference,
(1) The proposed optimization method with dynamic reference was designed
exclusively for the monitoring and analysis of the CIP process. The existence
duration of soil-water mixture in the pipe is short enough to be neglected. Hence
the background material in the test region is always considered to be
homogenous or nearly homogenous, which leads to uniform measured voltages
comparing to the reference voltages. However, when the optimization method is
applied to the measurement of mixture which consists of similar amount of both
high and low conductivity components, additional research should be carried out
to acquire more accurate results.
(2) This research still stays in the laboratory stage. It would be interesting to design
a practical system which enables the proposed system and method to monitor in-
situ real-time CIP process in a product plant.
(3) It is worth considering to apply the proposed method to the monitoring of other
industrial processes which share similar criteria of measurement. As an example,
waste water treatment requires the complete removal of impurities such as
electrolyte or heavy metal ions which would increase the conductivity of water.
By using ERT with dynamic reference, the conductivity of the liquid flow can
be monitored to ensure above impurities are fully removed.
2. For the research in measuring industrial oil-water separation process with
DEMIS,
123
(1) The differential structure of the sensor is currently implemented by connecting
the two receiving coils differentially. However, theoretically the sensor output
would be the same when two liquid systems share opposite fractions of oil and
saline, e.g. one liquid system has 30% of oil and the other liquid system has 30%
of saline. It would be worth considering to connect the two receiving coils
separately to the measurement system to acquire the induced voltages
independently and implement the differential calculation through data
processing. When the induced voltage in the upper receiving coil is zero, it
indicates that the upper test region is fully occupied with oil and the oil faction is
over 50%. In this way, the confusion on the oil fraction could be eliminated;
(2) In practical scenario, it is essential to know the conductivity of water in the
separation vessel to determine the offset value of sensor output. Hence a real-
time conductivity measurement system connected to the water outlet of the
vessel should be designed and embedded to the current system;
(3) Also in practical scenario, as the diameter of the coils is larger than that of the
separation vessel, further discussion and research need to be carried out with
field engineers to find a solution of installing the sensor to the separation vessel
without obstructing the separation process;
(4) The in-situ environment is complicated and the sensor signal could be interfered
in many ways. Hence proper electromagnetic isolation would be essential to
guarantee the reliability of the measurements.
124
REFERENCES
1. Brennen, C.E., Introduction to Multiphase Flow, in Fundamentals of Multiphase
Flow. 2005, Cambridge University Press New York. p. 1-28.
2. McQuillan, K. and P. Whalley, Flow patterns in vertical two-phase flow.
International Journal of Multiphase Flow, 1985. 11(2): p. 161-175.
3. Taitel, Y., O. Shoham, and J. Brill, Simplified transient solution and simulation
of two-phase flow in pipelines. Chemical engineering science, 1989. 44(6): p.
1353-1359.
4. Gregory, G., M. Nicholson, and K. Aziz, Correlation of the liquid volume
fraction in the slug for horizontal gas-liquid slug flow. International Journal of
Multiphase Flow, 1978. 4(1): p. 33-39.
5. Yang, W., M. Brant, and M. Beck, A multi-interface level measurement system
using a segmented capacitance sensor for oil separators. Measurement Science
and Technology, 1994. 5(9): p. 1177.
6. J. Mendes, A.P., L.F. Melo, Monitoring cleaning-in-place of shampoo films
using nanovibration technology. Sensors and Actuators, 2009. B 136: p. 376-
382.
7. J.C. Stewart, D.A.S., Clean in Place. Chemical Engineering, 1996. 103(1): p. 72.
8. M.S.Beck, Process Tomography: Principles, Techniques and Applications. 1995:
Butterworth-Heinemann Ltd.
9. Scott, D.M., Process Imaging For Automatic Control. 2005: Taylor&Francis
Group.
10. Borcea, L., Electrical impedance tomography. Inverse Problems, 2002. 18: p.
R99-R136.
11. Behin, J. and M. Aghajari, Influence of water level on oil–water separation by
residence time distribution curves investigations. Separation and purification
technology, 2008. 64(1): p. 48-55.
12. Falcone, G., Multiphase Flow Metering Principles. Developments in Petroleum
Science, 2009. 54: p. 33-45.
13. Nasr, G. and N. Connor, Natural Gas Engineering and Safety Challenges. 2016:
Springer.
14. H.S. Tapp, A.J.P., E.K. Kemsley, R.H. Wilson, Chemical engineering
applications of electrical process tomography. Sensors and Actuators, 2003. B
92: p. 17-24.
15. York, T.A., Status of electrical tomography in industrial applications. Journal of
Electronic Imaging, 2001. 10(3): p. 608-620.
16. W.Q. Yang, D.M.S., T.A. York and H. McCann, An image-reconstruction
algorithm based on Landweber's iteration method for electrical-capacitance
tomography. Meas. Sci. Technol., 1999. 10: p. 1065-1069.
17. Fraser Dickin, M.W., Electrical resistance tomography for process applications.
Meas. Sci. Technol., 1995. 7: p. 247-260.
18. Peyton, A., et al. Development of electromagnetic tomography (EMT) for
industrial applications. Part 1: Sensor design and instrumentation. in 1st World
Congress on Industrial Process Tomography. 1999.
19. Yu, Z., et al., Imaging system based on electromagnetic tomography (EMT).
Electronics Letters, 1993. 29(7): p. 625-626.
125
20. Al-Zeibak, S. and N. Saunders, A feasibility study of in vivo electromagnetic
imaging. Physics in Medicine & Biology, 1993. 38(1): p. 151.
21. Yang, W. and L. Peng, Image reconstruction algorithms for electrical
capacitance tomography. Measurement Science and Technology, 2002. 14(1): p.
R1.
22. Yang, W., Hardware design of electrical capacitance tomography systems.
Measurement Science and Technology, 1996. 7(3): p. 225.
23. Yang, W., Design of electrical capacitance tomography sensors. Measurement
Science and Technology, 2010. 21(4): p. 042001.
24. M. Cheney, D.I.a.J.C.N., Electrical impedance tomography. SIAM review, 1999.
41(1): p. 85-101.
25. George, D., et al., Validation of electrical-impedance tomography for
measurements of material distribution in two-phase flows. International Journal
of Multiphase Flow, 2000. 26(4): p. 549-581.
26. Heikkinen, L., et al., Real time three-dimensional electrical impedance
tomography applied in multiphase flow imaging. Measurement Science and
Technology, 2006. 17(8): p. 2083.
27. Holder, D.S., Electrical impedance tomography: methods, history and
applications. 2004: CRC Press.
28. Lucas, G. and R. Mishra, Measurement of bubble velocity components in a
swirling gas–liquid pipe flow using a local four-sensor conductance probe.
Measurement Science and Technology, 2005. 16(3): p. 749.
29. Gary Lucas, X.Z., Suman Pradhan, Optimisation of four-sensor probes for
measuring bubble velocity components in bubbly air-water and oil-water flows.
Flow Measurement and Instrumentation, 2011. 22: p. 50-63.
30. Kubáň, P. and P.C. Hauser, A review of the recent achievements in capacitively
coupled contactless conductivity detection. Analytica chimica acta, 2008. 607(1):
p. 15-29.
31. Du, L. and J. Zhe, A high throughput inductive pulse sensor for online oil debris
monitoring. Tribology International, 2011. 44(2): p. 175-179.
32. Ardron, K. and R. Duffey, Acoustic wave propagation in a flowing liquid-
vapour mixture. International Journal of Multiphase Flow, 1978. 4(3): p. 303-
322.
33. Atkinson, C. and H. Kytömaa, Acoustic wave speed and attenuation in
suspensions. International Journal of multiphase flow, 1992. 18(4): p. 577-592.
34. Takeda, Y., Velocity profile measurement by ultrasonic Doppler method.
Experimental thermal and fluid science, 1995. 10(4): p. 444-453.
35. Murai, Y., et al., Turbulent bubbly channel flow investigated by ultrasound
velocity profiler. Journal of Fluid Science and Technology, 2006. 1(1): p. 12-23.
36. Wang, T., et al., Application of Doppler ultrasound velocimetry in multiphase
flow. Chemical Engineering Journal, 2003. 92(1-3): p. 111-122.
37. Thorn, R., G. Johansen, and E. Hammer, Recent developments in three-phase
flow measurement. Measurement Science and Technology, 1997. 8(7): p. 691.
38. C. M. Bishop, G.D.J., Analysis of multiphase flows using dual-energy gamma
densitometry and neural networks. Nuclear Instruments and Methods in Physics
Research, 1993. A327: p. 580-593.
39. César Marques Salgado, C.M.N.A.P., Roberto Schirru, Luis E.B. Brandão, Flow
regime identification and volume fraction prediction in multiphase flows by
means of gamma-ray attenuation and artificial neural networks. Progress in
Nuclear Energy, 2010. 52: p. 555-562.
126
40. Van Santen, H., Z.I. Kolar, and A.M. Scheers, Photon Energy Selection for Dual
Energy γ- and/or X-ray Absorption Composition Measurements in Oil-Water-
Gas Mixtures. Nuclear Geophysics, 1995. 9(3): p. 193-202.
41. Narabayashi, T., et al., MEASUREMENT OF PHASE DISTRIBUTION
PHENOMENA IN HIGH PRESSURE STEAM-WATER TWO PHASE FLOW.
Dynamics of Two-Phase Flows, 1992: p. 423-438.
42. Heindel, T.J., A review of X-ray flow visualization with applications to
multiphase flows. Journal of Fluids Engineering, 2011. 133(7): p. 074001.
43. Mishima, K. and T. Hibiki, Development of high-frame-rate neutron
radiography and quantitative measurement method for multiphase flow research.
Nuclear Engineering and Design, 1998. 184(2-3): p. 183-201.
44. Takenaka, N., et al., Application of neutron radiography to visualization of
multiphase flows. Flow Measurement and Instrumentation, 1990. 1(3): p. 149-
156.
45. Falcone, G., et al. Multiphase flow metering: current trends and future
developments. in SPE annual technical conference and exhibition. 2001. Society
of Petroleum Engineers.
46. Ashton, S., et al. Development and trial of microwave techniques for
measurement of multiphase flow of oil, water and gas. in SPE Asia Pacific Oil
and Gas Conference. 1994. Society of Petroleum Engineers.
47. Wylie, S., A. Shaw, and A. Al-Shamma'a, RF sensor for multiphase flow
measurement through an oil pipeline. Measurement Science and Technology,
2006. 17(8): p. 2141.
48. Wu, Z., et al., Microwave-tomographic system for oil-and gas-multiphase-flow
imaging. Measurement Science and Technology, 2009. 20(10): p. 104026.
49. Wu, Z., Developing a microwave tomographic system for multiphase flow
imaging: advances and challenges. Transactions of the Institute of Measurement
and Control, 2015. 37(6): p. 760-768.
50. Withers, P.M., Ultrasonic, acoustic and optical techniques for the non-invasive
detection of fouling in food processing equipment. Trends in food science &
technology, 1996. 7(9): p. 293-298.
51. T.R. Hay, J.L.R., Fouling detection in the food industry using ultrasonic guided
waves. Food Control, 2003. 14: p. 481–488.
52. K.R. Lohr, J.L.R., Ultrasonic guided wave and acoustic impact methods for pipe
fouling detection. Journal of Food Engineering, 2003. 56: p. 315–324.
53. da Silva, J.J., et al., Non-invasive fast detection of internal fouling layers in
tubes and ducts by acoustic vibration analysis. IEEE Transactions on
Instrumentation and Measurement, 2009. 58(1): p. 108-114.
54. Schöning M J, B.D., Rolka D, et al., CIP (cleaning-in-place) suitable “non-
glass” pH sensor based on a Ta 2 O 5-gate EIS structure. Sensors and Actuators,
2005. 111(B): p. 423-429.
55. Chen, X.D., et al., On-line fouling/cleaning detection by measuring electric
resistance––equipment development and application to milk fouling detection
and chemical cleaning monitoring. Journal of Food Engineering, 2004. 61(2): p.
181-189.
56. Fowles, G., Flow, level and pressure measurement in the water industry. 1993:
Butterworth-Heinemann.
57. Nyce, D., Magnetostrictive linear position sensors. Sensors-the Journal of
Applied Sensing Technology, 1999. 16(11): p. 38-47.
127
58. Pasquale, M., Mechanical sensors and actuators. Sensors and Actuators A:
Physical, 2003. 106(1): p. 142-148.
59. Jaworski, A.J. and G. Meng, On-line measurement of separation dynamics in
primary gas/oil/water separators: challenges and technical solutions—a review.
Journal of Petroleum Science and Engineering, 2009. 68(1): p. 47-59.
60. Arvoh, B.K., N.-O. Skeie, and M. Halstensen, Estimation of gas/liquid and
oil/water interface levels in an oil/water/gas separator based on pressure
measurements and regression modelling. Separation and Purification
Technology, 2013. 107: p. 204-210.
61. Hauptmann, P., N. Hoppe, and A. Püttmer, Application of ultrasonic sensors in
the process industry. Measurement Science and Technology, 2002. 13(8): p.
R73.
62. Bukhari, S.F.A. and W. Yang, Multi-interface level sensors and new
development in monitoring and control of oil separators. Sensors, 2006. 6(4): p.
380-389.
63. Hjertaker, B.T., G.A. Johansen, and P. Jackson. Recent developments in
hydrocarbon separator interface imaging. in Proc. of SPIE. 2001.
64. Isaksen, O., A. Dico, and E.A. Hammer, A capacitance-based tomography
system for interface measurement in separation vessels. Measurement Science
and Technology, 1994. 5(10): p. 1262.
65. Yang, W., Further developments in an ac-based capacitance tomography system.
Review of Scientific Instruments, 2001. 72(10): p. 3902-3907.
66. Shafquet, A., et al., Estimation of gas void formation in statically cooled waxy
crude oil using online capacitance measurement. International Journal of
Multiphase Flow, 2015. 75: p. 257-266.
67. Yang, W., Sensors and instrumentation for monitoring and control of multi-
phase separation. Measurement and Control, 2006. 39(6): p. 178-184.
68. Bennett, M. and R. Williams, Monitoring the operation of an oil/water
separator using impedance tomography. Minerals Engineering, 2004. 17(5): p.
605-614.
69. Abouelwafa, M. and E. Kendall, The measurement of component ratios in
multiphase systems using alpha-ray attenuation. Journal of Physics E: Scientific
Instruments, 1980. 13(3): p. 341.
70. Kumara, W., B. Halvorsen, and M. Melaaen, Single-beam gamma densitometry
measurements of oil–water flow in horizontal and slightly inclined pipes.
International Journal of Multiphase Flow, 2010. 36(6): p. 467-480.
71. Meribout, M., A. Al Naamany, and K. Al Busaidi, Interface layers detection in
oil field tanks: a critical review, in Expert Systems for Human, Materials and
Automation. 2011, InTech.
72. Skeie, N.O. and M. Halstensen, Level estimation in oil/water separators based
on multiple pressure sensors and multivariate calibration. Journal of
Chemometrics, 2010. 24(7‐8): p. 387-398.
73. Jaworski, A.J. and T. Dyakowski, Measurements of oil–water separation
dynamics in primary separation systems using distributed capacitance sensors.
Flow Measurement and Instrumentation, 2005. 16(2): p. 113-127.
74. Yin, W., et al., Simultaneous noncontact measurement of water level and
conductivity. IEEE Transactions on Instrumentation and Measurement, 2008.
57(11): p. 2665-2669.
75. Yang, T., et al., A high frequency digital induction system for condutive flow
level measurements. Flow Measurement and Instrumentation, 2014. 37: p. 83-91.
128
76. Asskildt, K. and P. Hansson, New measuring sensor for level detection in subsea
separators. ABB Review, 1999(4): p. 11-17.
77. Geselowitz, D.B., An Application of Electrocardiographic Lead Theory to
Impedance Plethysmography. IEEE Transactions on Bio-medical Engineering,
1971. BME-18(No. 1): p. 38-41.
78. Lehr, J., A Vector Derivation Useful in Impedance Plethysmographic Field
Calculations. IEEE Transactions on Biomedical Engineering, 1972. BME-19(2):
p. 156-157.
79. Murai, T. and Y. Kagawa, Electrical impedance computed tomography based on
a finite element model. Biomedical Engineering, IEEE Transactions on, 1985(3):
p. 177-184.
80. Wang, M., Inverse solutions for electrical impedance tomography based on
conjugate gradients methods. Measurement Science and Technology, 2001.
13(1): p. 101.
81. Yang, W., et al., An image-reconstruction algorithm based on Landweber's
iteration method for electrical-capacitance tomography. Measurement Science
and Technology, 1999. 10(11): p. 1065.
82. Li, Y. and W. Yang, Image reconstruction by nonlinear Landweber iteration for
complicated distributions. Measurement Science and Technology, 2008. 19(9): p.
094014.
83. Schweiger, M., S.R. Arridge, and I. Nissila, Gauss-Newton method for image
reconstruction in diffuse optical tomography. Phys Med Biol, 2005. 50(10): p.
2365-86.
84. Yorkey, T.J., J.G. Webster, and W.J. Tompkins, Comparing reconstruction
algorithms for electrical impedance tomography. Biomedical Engineering, IEEE
Transactions on, 1987(11): p. 843-852.
85. Polydorides, N., Image reconstruction algorithms for soft-field tomography.
2002, University of Manchester: UMIST.
86. D.C. Barber, B.H.B., I.L. Freeston, Imaging Spatial Distributions of Resistivity
Using Applied Potential Tomography -APT. Information Processing in Medical
Imaging, 1984: p. 446-462.
87. P.J. Holden, M.W., and R. Mann, Imaging stirred-vessel macromixing using
electrical resistance tomography. American Institute of Chemical Engineers,
1998. 44(4): p. 780-790.
88. X. Deng, F.D., L. Xu, X. Liu and L. Xu, The design of a dual-plane ERT system
for cross correlation measurement of bubbly gas/liquid pipe flow. Meas. Sci.
Technol., 2001. 12: p. 1024-1031.
89. Krawczyk-StańDo, D. and M. Rudnicki, Regularization parameter selection in
discrete ill-posed problems—the use of the U-curve. International Journal of
Applied Mathematics and Computer Science, 2007. 17(2): p. 157-164.
90. Rosell, J., R. Casanas, and H. Scharfetter, Sensitivity maps and system
requirements for magnetic induction tomography using a planar gradiometer.
Physiological measurement, 2001. 22(1): p. 121.
91. Dyck, D., D. Lowther, and E. Freeman, A method of computing the sensitivity of
electromagnetic quantities to changes in materials and sources. IEEE
Transactions on Magnetics, 1994. 30(5): p. 3415-3418.
92. Yin, W. and A.J. Peyton, Sensitivity formulation including velocity effects for
electromagnetic induction systems. Magnetics, IEEE Transactions on, 2010.
46(5): p. 1172-1176.
129
93. Xu, K., et al., Sensitivity derivation and calculation of electromagnetic
tomography (EMT) sensor based on field value extraction. Chinese Journal of
Sensors and Actuators, 2011. 24(4): p. 543-547.
94. Al-Zuhair, S., Using liquid–liquid deep settling model in determining the design
parameters of crude palm oil settler. Separation and purification technology,
2004. 35(2): p. 133-140.
95. Jeelani, S. and S. Hartland, Effect of dispersion properties on the separation of
batch liquid− liquid dispersions. Industrial & engineering chemistry research,
1998. 37(2): p. 547-554.
96. Yu, G.Z. and Z.S. Mao, Sedimentation and coalescence profiles in liquid‐liquid
batch settling experiments. Chemical engineering & technology, 2004. 27(4): p.
407-413.
97. Faraj, Y., et al., Measurement of vertical oil-in-water two-phase flow using dual-
modality ERT–EMF system. Flow Measurement and Instrumentation, 2015. 46:
p. 255-261.
98. Batchelor, G., An Introduction to Fluid Mechanics, 615 pp. Cambridge
University Press, New York, 1967.
99. Jeelani, S.A.K. and S. Hartland, Prediction of steady state dispersion height
from batch settling data. AIChE journal, 1985. 31(5): p. 711-720.
100. Xilinx, Zynq-7000 All Programmable SoC.