This document presents a study on the use of an optical tomography system to measure the concentration of gas bubbles in a water column. The system employs a hydraulic flow rig and a hybrid back-projection algorithm to generate two-dimensional images of bubble distributions, which are important for applications in various fields, including medical and gas handling. Results demonstrate the capability of the system to accurately measure and visualize gas concentrations at different flow rates and bubble sizes.

1. ISA Transactions 51 (2012) 821–826
Contents lists available at SciVerse ScienceDirect
ISA Transactions
journal homepage: www.elsevier.com/locate/isatrans
Concentration measurements of bubbles in a water column using
an optical tomography system
S. Ibrahim a,n, Mohd Amri.Md Yunus a, R.G. Green b, K. Dutton b
a
b
Faculty of Electrical Engineering, Universiti Teknologi Malaysia, 81310 Skudai, Johor, Malaysia
Materials and Engineering Research Institute, Sheffield Hallam University, City Campus, Sheffield, S1 1WB, United Kingdom
a r t i c l e i n f o
abstract
Article history:
Received 21 June 2011
Received in revised form
27 April 2012
Accepted 27 April 2012
Available online 22 May 2012
Optical tomography provides a means for the determination of the spatial distribution of materials with
different optical density in a volume by non-intrusive means. This paper presents results of
concentration measurements of gas bubbles in a water column using an optical tomography system.
A hydraulic flow rig is used to generate vertical air–water two-phase flows with controllable bubble
flow rate. Two approaches are investigated. The first aims to obtain an average gas concentration at the
measurement section, the second aims to obtain a gas distribution profile by using tomographic
imaging. A hybrid back-projection algorithm is used to calculate concentration profiles from measured
sensor values to provide a tomographic image of the measurement cross-section. The algorithm
combines the characteristic of an optical sensor as a hard field sensor and the linear back projection
algorithm
& 2012 ISA. Published by Elsevier Ltd. All rights reserved.
Keywords:
Bubbles
Concentration
Optical tomography
Optical fibre sensors
Tomography
1. Introduction
In multi-phase flow measurement both the phase distribution
and the velocity profiles vary significantly with temporal and
spatial resolution. This is due to the different phases arranging
themselves in various ways. For a multi-phase flow, the flow
patterns are primarily functions of the volumetric fluxes of all
phases. The flow patterns are functions of superficial velocities or
pressure drops and are depicted in flow profiles. Hence it is
important to have a knowledge of the flow profiles in order to
design heat or heat and mass transfer equipment and to design
fluid-based conveying processes [1].
Information on gas concentration is vital in various applications. In the medical field, it is important to have information on
anaesthetic gas, oxygen or heliox concentration. In order to obtain
a precise bill, it is vital to record the exact heating value of the gas
for gas metering purpose. In gas burner appliances, flame optimisation is conducted for the purpose of efficiency and emission.
This is carried out by regulating the mixing ratio of air and gas.
Generally, accurate gas concentration measurement is vital in
many gas handling applications [2].
Many techniques have been employed for gas or bubble detection
in two-phase flows; both intrusive and non-intrusive measurement
n
Corresponding author. Tel.: þ60 19 7411 434; fax: þ60 7 55 66 272.
E-mail addresses: salleh@fke.utm.my (S. Ibrahim),
mus_utm@yahoo.com (M.Amri.M. Yunus), r.g.green@shu.ac.uk (R.G. Green),
k.dutton@shu.ac.uk (K. Dutton).
techniques were developed for such purpose. However, it is
important that sensors being used for measurement purposes
do not in any way perturb the quantity being measured. Point
sensors are not generally suitable as they disturb the flow field.
Non-intrusive techniques possess the advantage of not modifying
the flow field and they are suitable for laboratory tests [3].
The word tomography comes from the Greek words ‘tomos’
which means a cut or slice and ‘graphein’ which means to
write [4]. Process tomography is a methodology in which the
internal characteristics of process vessel reaction or pipeline flows
are acquired from measurements on or outside the domain of
interest in a non-invasive fashion [5]. This paper describes an
optical tomography system which is used to reconstruct an image
from measurements obtained from several sensors placed around
the measurement section of a hydraulic flow rig. Light travelling
through a transparent medium suffers attenuation for various
reasons, including scattering and absorption. Different materials
cause varying levels of attenuation and it is this phenomenon that
forms the concept of optical tomography. The voltage generated
by the optical sensors is proportional to the level of received light.
It is related to the amount of attenuation in the path of the light
beam caused by the flow regime [6]. Information about the
optical characteristics of a flow can be obtained if a view
consisting of an optical emitter and detector pair are positioned
either side of the measurement section. A larger area can be
interrogated if several views are combined to form a projection.
The image of the flow can be reconstructed if several different
projections are utilised [7].
0019-0578/$ - see front matter & 2012 ISA. Published by Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.isatra.2012.04.010

2. 822
S. Ibrahim et al. / ISA Transactions 51 (2012) 821–826
2. The measurement system
In an optical tomography system, several groups of transmitter
and receiver pairs are used in a system to provide better solution
and to minimise the aliasing which occurs when two particles
intercept the same view [8]. In this project, four dichroic halogen
bulbs act as light projectors. The light receivers consist of an array
of optical fibre sensors arranged in a combination of two orthogonal and two rectilinear projections (Fig. 1). The orthogonal
projections each consist of an 8 Â 8 array of optical fibre sensors
whereas the rectilinear projections consist of an 11 Â 11 array of
optical fibre sensors (the numbers of sensors is chosen so that they
will give a balanced sensitivity). Thus the total number of optical
fibre sensors used is thirty eight. Ideally the two orthogonal and
two rectilinear projections system should be in the same plane.
However, if they were they would overlap each other and so two
of the projections have to be placed in a separate plane. These
planes are separated by only a few mm with the two orthogonal
projection system placed on top of the rectilinear projections with
respect to the direction of flow. Each optical fibre receiver has a
length of 200 cm from the flow pipe to the electronic circuit.
The receiver circuit is designed for signal conditioning using
amplifiers and filters. The final outputs of the circuit are electrical
signals consisting of a rectified voltage and an averaged voltage. The
rectified voltage enables unipolar data acquisition and consequent
signal processing using a PC. The output of the amplifier should be
proportional to the gas flow rate passing the associated sensor. If all
the averaged voltage amplifier outputs are summed, they should be
proportional to the gas flow rate indicated by the gas rotameter. All
the electronic circuits are placed in an earthed metal box to
minimise electrical noise pick-up. The rectified analogue signals
from an array of optical transducers, covering a cross-section of the
pipe, are converted into digital form by the Keithley Instruments
DAS-1800HC data acquisition system and passed into an image
reconstruction system. For concentration measurements, a sampling
frequency of 500 Hz per channel is chosen as the velocities and flow
rates associated with this project are relatively low (i.e. 0.2–0.3 m/s).
This enables two hundred and twenty-two points to be collected for
each of the thirty-eight channels, which allows 0.44 s of flow data to
be obtained.
Data acquired by the receiver circuit is processed using the hybrid
linear back projection algorithm, which has been described in a
recent paper by Ibrahim et al. [9] in order to generate two-dimensional images of the bubbles in water. The algorithm incorporates
both a priori knowledge and linear back projection (LBP) in order to
improve the accuracy of the image reconstruction. Since optical
sensors are hard field sensors, the material in the flow is assumed
only to vary the intensity of the received signal. This enables a priori
knowledge from the optical sensors to be used as a constraint in the
reconstruction. The optical sensor signal is conditioned so when no
Projector 2
Flow Pipe
Projector 1
Top Aluminium Plate
Optical Fibre Receivers
Optical Fibre
Receivers
Perspex
Optical Fibre Receivers
Projector 3
Perspex
Bottom Aluminium
Plate
Projector 4
Fig. 1. Arrangement of the projectors and optical fibres around the flow pipe—isometric view.

3. S. Ibrahim et al. / ISA Transactions 51 (2012) 821–826
objects block the path from light transmitter to the receiver the
sensor will produce a zero output value, neglecting the effect of noise
inherent in the system.
The system was tested on the hydraulic flow rig shown in Fig. 2.
The measurement system is built around a vertical pipe 1.27 m long
with circular cross-section. Control of the water flow is effected by
the use of a pump and by various valves installed in the rig. Bubbles
are injected into the measurement section through two bubble
injectors placed at the base of the vertical section. The two small air
injectors are utilised to blow different sized gas bubbles from the
bottom of the pipe. Small bubbles are generated by a porous plug in
the base of the flow rig, producing bubbles which visually appear to
be in the range 10 mm. Large bubbles are produced by direct gas
injection into the flowing water. These bubbles are about 20 mm in
diameter. When the bubbles rise up and pass though the imaging
cross-section, the two-phase distribution over the imaging plane can
be measured. The time history of bubbles rising up through the
measurement section can be obtained in an off-line manner with
the data stored on the hard-disc. Control of bubble flow is achieved
through the use of two valves linked to the two bubble injectors. The
valves can control the size of the bubbles as well as generating
various flow regimes. The air pressure supply to the bubble injectors
can be varied from 0 to 420 kP. Throughout the experiment, gas is
injected at a constant pressure of 50 kP. The bubble collapsed as it
reached the surface of the water.
The flow pipe is made of perspex to enable visual observation
of the flow. The lower measurement section, consisting of thirtyeight sensors, is placed 62 cm above the gas injection points and
the second sensing array, which also consists of thirty-eight
sensors, is placed 15 cm downstream of the former. The measurement section is of modular construction and comprises a series of
perspex blocks 90 mm square with an 80 mm diameter central
bore so that when bolted together they provide a continuous
80 mm diameter internal flow passage. In order to reduce optical
distortion and to allow optical observation, a flat square shape
perspex window is used [10]. The flow rig is equipped with two
823
rotameters: a water rotameter (0–7 l/min) and a gas rotameter
(0–7 l/min). Each rotameter provides direct readings of the total
flow rate of water and bubbles respectively. For the experiments
described in this paper water always formed the continuous
phase and the gas flow was always in the bubbly regime. The
volumetric flow rate of bubbles ranged from 0 to 7 l/min.
3. Average concentration measurement
The measurements presented in this section consider the results
from each sensor as a continuous sample of the gas concentration
within its sensing field. The method was to obtain two hundred and
twenty-two samples for all thirty-eight sensors; each sensor was
Table 1
The flow rates of bubbles and the corresponding sum of pixel voltages for small
bubbles and large bubbles.
Flow rates
(l/min)
Sum of pixel
voltages for
small bubbles
(V)
Error
percentage of
small bubbles
(%)
Sum of pixel
voltages for
large bubbles
(V)
Error
percentage
of large
bubbles (%)
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
5.5
6
6.5
7
4.9
262.2
345.6
285.4
240.5
249.3
243.8
201.6
188
185.5
180.1
186.4
185.4
179.3
188.2
2
6.4
3.2
17.3
30.3
27.7
29.3
40.6
45.5
46.2
47.8
46.0
46.3
48.0
45.4
4.9
291.1
334.2
365.5
397.9
389.3
412
408.4
406.4
390.8
371.4
340.3
316.1
303.9
282.4
5.1
4.0
5.9
1.2
0.1
0.7
0.5
0.9
0.9
1.1
1.0
17.4
23.3
26.2
31.5
Fig. 2. The hydraulic flow rig.

4. 824
S. Ibrahim et al. / ISA Transactions 51 (2012) 821–826
Sum of pixels (V)
sampled at 500 Hz. The individual sensor gains were compensated for
in software. The mean value for each sensor was calculated at the
specified flow rate used with the standard linear back projection
algorithm to produce an image consisting of grey level pixels; white
represents maximum flow, black zero flow. These pixel grey levels are
represented by pixel voltages. The pixel voltages are summed over
the measurement cross-section. The values obtained for each flow
rate for small bubbles are compared with the values obtained for
large bubbles to observe the effect of flow rates and bubble size.
Table 1 shows the sum of pixels for small bubbles and large bubbles
corresponding to various volumetric flow rates. The results in Table 1
are shown graphically in Fig. 3 and discussed in Section 5.
Gas Flow Rate Calibration
450
400
350
300
250
200
150
100
50
0
Measured
results
(Small
bubbles)
Measured
results (Big
bubbles)
Polynomial
regression
(Big bubbles)
Polynomial
regression
(Small
bubbles)
0
2
4
6
Flow Rate (l/min)
Fig. 3. Gas flow rate calibration graph.
8
4. Concentration profiles
The measurement system for concentration consists of thirtyeight sensors. Ideally, with zero flow, all sensors should have zero
output. In practice many of the sensors have an output voltage
due to factors such as drift, intrinsic noise and offset in operational amplifiers. To reduce these errors all the sensors were
sampled at 500 Hz for 0.44 s with no gas flow at the start of each
series of experiments. The root mean square voltage was then
calculated for each sensor to provide a zero flow reference
voltage. The zero flow reference voltages were used to correct
the gas flow measurements for offset errors in each sensor.
The experiments were conducted with the laboratory lights
switched off to ensure that the mains lighting did not affect the light
receivers. Measurements were made by energising all thirty-eight
optical sensors and monitoring their outputs at several gas flow rates
ranging from 0 l/min to 7 l/min. Throughout the experiment, water
flowed upwards at a volumetric flow rate of 3 l/min.
The following results show a sequence of images representing
the reconstructed fields of bubbles flowing in water, which were
generated at selected volumetric flow rates. A sequence of images
representing small bubbles flowing at a volumetric flow rate of
0.5 l/min, are shown on Fig. 4a–d. A sequence of images representing large bubbles generated at a volumetric gas flow rate of
0.5 l/min, are shown on Fig. 5a–d. The existence of various colours
in the matrix of Fig. 4a–d as well as Fig. 5a–d except the black colour
indicates the location of bubbles. Black colour indicates no flow. For
example in Fig. 4a, the concentration profile show the bubbles are
located at pixels (6,3), (4,5), (5,5), (2,6), (4,6), (5,6) and (6,6).
Fig. 4. (a) Matrix and concentration profile of the first sample representing small bubbles at 0.5 l/min. (b) Matrix and concentration profile of the second sample
representing small bubbles at 0.5 l/min. (c) Matrix and concentration profile of the third sample representing small bubbles at 0.5 l/min. (d) Matrix and concentration
profile of the fourth sample representing small bubbles at 0.5 l/min.

5. S. Ibrahim et al. / ISA Transactions 51 (2012) 821–826
825
Fig. 5. (a) Matrix and concentration profile of the first sample representing large bubbles at 0.5 l/min. (b) Matrix and concentration profile of the second sample
representing large bubbles at 0.5 l/min. (c) Matrix and concentration profile of the third sample representing large bubbles at 0.5 l/min. (d) Matrix and concentration
profile of the fourth sample representing large bubbles at 0.5 l/min.
5. Discussion of results
The gas flow rate calibration graph (Fig. 3) shows the sum of
voltages in all pixels within the flow pipe plotted against the
volumetric flow rate of the bubbles for both small bubbles and
large bubbles. The results shown in Table 1 give a noise level of
4.9 V, at zero flow, which corresponds to levels of 1.4% of
maximum flow reading for small bubbles and 1.2% for large
bubbles. The results obtained in section 4 indicate that the system
reacts to large and small bubbles in a similar manner. However,
the peak of the graph occurs at higher flow rates with large
bubbles than small bubbles. Fig. 3 shows the peak occurring at a
gas flow rate of 1 l/min for small bubbles and 3 l/min for large
bubbles. Empirical equations obtained using EXCEL software have
been fitted to the results. For small bubbles the equation used is
Sum of pixels ¼ À0:32x6 þ 7:76x5 þ 72:93x4 þ340:26x3
þ 810:16x2 þ 860:21x
ð1Þ
where x is the gas flow rate in l/min.
The equation used the large bubbles is
in the centre of the pipe. This means that the majority of small
bubbles are confined to the central part of the measurement
cross-section and only affect a few sensors. As the flow rate
increases the bubbles get closer together.
In the case of small bubbles, the gas flow rate calibration graph
shows that initially from 0 to 1 l/min, the sum of the pixels increases,
but from 1.5 l/min to 4 l/min, the sum of the pixels begins to decrease
as the flow rate increases. The small gas bubbles are generally
confined to the centre of the pipe. As the volumetric flow rate
increases, more bubbles are released, resulting in only a few sensors
being affected and as such, the sensors as well as the electronics have
little time to recover from bubbles that flowed previously. The signal
conditioning causes the signal at high gas flow rate to gradually
reduce.
Large bubbles occupy a much larger cross section of the
conveyor than small bubbles and so the majority of the sensors
detect the presence of the bubbles. The sum of the pixels
increases over the flow rate of 0–3 l/min as shown in Fig. 3.
However, beyond the volumetric flow rate of 3 l/min, the sum of
the pixels begins to decrease as shown in Fig. 3.
Sum of pixels ¼ À0:31x6 þ 7:39x5 À67:63x4 þ 302:38x3
À694:29x2 þ 796:74x þ 10:76
ð2Þ
where x again is the gas flow rate in l/min.
The majority of measurements were made with circulating
water to ensure that the bubbles flowed upwards in the pipe.
However, the flowing water appeared to keep the small bubbles
6. Conclusions
In this paper, a non-intrusive concentration measurement of
bubbles flowing in a water column has been presented. The
results showed that the measurement system is able to fulfil
the original objectives of obtaining an average gas concentration

6. 826
S. Ibrahim et al. / ISA Transactions 51 (2012) 821–826
and gas distribution profiles. For future work, it is suggested that
experiments will be conducted over a larger range of flow
regimes. Ideally in an industrial environment, it is preferable to
use laser as the light source due to its monochromatic and
coherent characteristics. The resolution can be increased by
increasing the number of views per light sources for each pixel.
Further investigation using other types of reconstruction algorithms and different forms of filtering techniques should be
performed. The use of multi modality tomography should be
investigated in which the optical tomographic system can be
combined with other types of sensing with the aim of comparing
the accuracy of the measurements and increasing the understanding of the flow process.
Acknowledgement
The authors wish to acknowledge the assistance of Universiti
Teknologi Malaysia for providing the funds and resources in
carrying out this research.
References
[1] Dyakowski T. Process tomography applied to multi-phase flow measurement.
Measurement Science & Technology 1996;7:343–53.
[2] /http://www.sensirion.com/en/pdf/RD-notes/RD-note_February2012.pdfS.
[3] Rossi GL. Error analysis based development of a bubble velocity measurement chain. Flow Measurement and Instrumentation 1996;7(1):39–47.
[4] Northrop RB. Noninvasive instrumentation and measurement in medical
diagnosis. Boca Raton, FL: CRC Press; 2002.
[5] Wang M. Process tomography (editorial). Measurement Science and Technology 2006:17.
[6] Abdul Rahim R, Kok San C. Optical tomography imaging in pneumatic
conveyor. Sensors and Transducers Journal 2008;95(8):40–8.
[7] Daniels AR. Dual modality tomography for the monitoring of constituent
volumes in multi-component flow. PhD thesis. Sheffield Hallam University;
1996.
[8] Abdul Rahim R, Fea P J, Kok San C. Optical tomography sensor configuration
using two orthogonal and two rectilinear projection arrays. Flow Measurement and Instrumentation 2005;16:327–40.
[9] Ibrahim S, Green R G, Dutton K, Evans K, Goude A, Abdul Rahim R. Optical
sensor configurations for process tomography. Measurement science &
technology 1999;10:1079–86.
[10] King NW, Purfit GL, Pursley WC. A laboratory facility for the study of mixing
phenomena in water/oil flows. In: Proceedings of the symposium flow
metering and proving techniques in the offshore oil industry (Aberdeen);
1983: p. 233–41.