ISA Transactions 47 (2008) 3–14

www.elsevier.com/locate/isatrans

Real time mass flow rate measurement using multiple fan beam optical
tomography
R. Abdul Rahim a,∗ , L.C. Leong a , K.S. Chan a , M.H. Rahiman b , J.F. Pang a
a Process Tomography Research Group (PROTOM), Department of Control and Instrumentation Engineering, Faculty of Electrical Engineering,
Universiti Teknologi Malaysia, 81310 UTM Skudai, Johor, Malaysia
b Department of Mechatronics, Universiti Malaysia Perlis, Malaysia

Received 12 October 2006; accepted 29 May 2007

Available online 20 August 2007

Abstract
This paper presents the implementing multiple fan beam projection technique using optical fibre sensors for a tomography system. From the
dynamic experiment of solid/gas flow using plastic beads in a gravity flow rig, the designed optical fibre sensors are reliable in measuring the mass
flow rate below 40% of flow. Another important matter that has been discussed is the image processing rate or IPR. Generally, the applied image
reconstruction algorithms, the construction of the sensor and also the designed software are considered to be reliable and suitable to perform
real-time image reconstruction and mass flow rate measurements.
c 2007, ISA. Published by Elsevier Ltd. All rights reserved.

Keywords: Optical tomography; Fan beam; Mass flowrate; Projection geometry; Sensors mapping

1. Introduction brought simultaneous improvements and cost reduction in

optoelectronic components. In the area of tomographic
Process tomography is a developing measurement technol-
imaging, an initial investigation into using optic fibres as
ogy. Over the last decade, the concept of tomographic imaging
measurement sensors in pneumatic conveying was started in
was not restricted to the medical field; it has been success-
Sheffield Hallam University [8].
fully developed into a reliable tool for imaging numerous
In using optical fibre for tomography imaging, the basic
industrial applications. Sensors in the measurement system
optical transmitter converts electrical input signals into
prove to be the most important part in acquiring the physical
modulated light for transmission over an optical fibre. Also, the
signal. The physical signal is then being converted, amplified
light beam from the transmitter is being received by the receiver
and fed into the computer to obtain the internal behaviours of
via the fibre optic. This configuration is being illustrated in
the investigated vessel. A conveyor with fast moving flow and
Fig. 1. With regard to its small physical size, it is believed that
rapid changing flow regime requires a sensor with very fast re-
using fibre optic will allow a higher number of optical sensors
sponse and also a fast data acquisition rate in order to prevent
to be installed, thus achieving high-resolution measurement in
data losses. The optical sensors have proved to possess the ad-
optical tomography. It is also said that the optical fibre sensors
vantages of fast responce, high performance, low cost and low
provide wide bandwidth which enables measurements to be
maintenance compared to electrical capacitance tomography.
performed on high speed flowing particles [9].
2. Optical fibre for tomography imaging
3. Switch-mode fan beam method
Mass production of optical fibre has reduced the cost of
the fibres significantly and more importantly the revolution Parallel projection and fan beam projection are two different
types of sensor arrangement methods. For sensor arrangement
∗ Corresponding address: Universiti Teknologi Malaysia, Research Manage-
method using parallel projection method, each transmitter and
ment Centre, 81310 UTM Skudai, Johor, Malaysia. Tel.: +60 7 5537804; fax: receiver only corresponds to each other. While for the fan
+60 7 5566177.
E-mail addresses: ruzairiabdulrahim@yahoo.co.uk, Ruzairi@fke.utm.my beam projection method, there may be more than one receiver
(R. Abdul Rahim). corresponding to a single or multiple light sources [1].

c 2007, ISA. Published by Elsevier Ltd. All rights reserved.

0019-0578/$ - see front matter
doi:10.1016/j.isatra.2007.05.005
4 R. Abdul Rahim et al. / ISA Transactions 47 (2008) 3–14

4. Hardware construction

A typical optical tomography system consists of the sensors’

array, signal control and conditioning circuit and also the
display unit, namely the computer. In this research, instead of
mounting the transmitters and receivers straight to the pipeline
as being done by Chan [2], Abdul Rahim [1], the optical
fibres are applied. A total of 64 fibre optics are used, with
32 connected from the sensor’s fixture (at the circumference
Fig. 1. Fibre optic configuration.
of the flow rig) to the emitters and the other 32 fibre optics
are connected to the receivers. The topology of the hardware
construction is shown in Fig. 3.
By referring to Fig. 3, the microcontroller is used to control
the duration of light projection, sample and hold digital input
and data acquisition system (DAS) synchronization signals.
Through the fibre optics, photodiodes detect the physical
signals (light beams) from the transmitters. In the signal
conditioning circuit, the physical signals are being converted
into voltage readings and then amplified. The analogue signals
go through the sample and hold circuit before being transferred
into DAS. DAS then converts the analogue signals into digital
signals. These digital signals are sent to the PC for image
reconstruction.

4.1. Multi-projection technique

A new technique of implementing the switch-mode fan

beam method is being investigated in this paper. The
technique mentioned is the 2-projection technique. Preliminary
Fig. 2. Sensor configuration in switch-mode fan beam projection method. experiments show that the maximum achievable emission angle
for the fibre optic transmitter is about 30◦ , after the fibre optic
In the switch-mode fan beam method, a multiplexing is being lensed. 32 fibre optic transmitters are used in this
source is used and all the receivers produce the signals research; thus in order to make sure that the emission angle is
corresponding to each multiplexed source. Fig. 2 shows the approximately the same, each of the fibre optic emission angles
sensors’ configuration of fan beam projection method whereby is tested experimentally. The experimental method is illustrated
the transmitters and receivers are arranged alternately. in Fig. 4.
By assuming that the transmitter is the LED which has The fibre optic for the transmitter will be lensed until both
a wide transmission angle of 140◦ , the LED or Txn will be the Rx receive signals. This means that the emission angle has
turned on subsequently whereby the subscript n here represents reached 30◦ . All the 32 fibre optics for transmitters’ angles will
different LEDs [3]. be determined experimentally this way.

Fig. 3. Topology of the hardware construction.

 R. Abdul Rahim et al. / ISA Transactions 47 (2008) 3–14 5

Table 1
Configuration of transmitters and corresponding receivers

Transmitter Receiver Transmitter Receiver

Tx0 Rx13 Rx14 Rx15 Rx16 Rx17 Rx18 Tx16 Rx29 Rx30 Rx31 Rx0 Rx1 Rx2
Tx1 Rx14 Rx15 Rx16 Rx17 Rx18 Rx19 Tx17 Rx30 Rx31 Rx0 Rx1 Rx2 Rx3
Tx2 Rx15 Rx16 Rx17 Rx18 Rx19 Rx20 Tx18 Rx31 Rx0 Rx1 Rx2 Rx3 Rx4
Tx3 Rx16 Rx17 Rx18 Rx19 Rx20 Rx21 Tx19 Rx0 Rx1 Rx2 Rx3 Rx4 Rx5
Tx4 Rx17 Rx18 Rx19 Rx20 Rx21 Rx22 Tx20 Rx1 Rx2 Rx3 Rx4 Rx5 Rx6
Tx5 Rx18 Rx19 Rx20 Rx21 Rx22 Rx23 Tx21 Rx2 Rx3 Rx4 Rx5 Rx6 Rx7
Tx6 Rx19 Rx20 Rx21 Rx22 Rx23 Rx24 Tx22 Rx3 Rx4 Rx5 Rx6 Rx7 Rx8
Tx7 Rx20 Rx21 Rx22 Rx23 Rx24 Rx25 Tx23 Rx4 Rx5 Rx6 Rx7 Rx8 Rx9
Tx8 Rx21 Rx22 Rx23 Rx24 Rx25 Rx26 Tx24 Rx5 Rx6 Rx7 Rx8 Rx9 Rx10
Tx9 Rx22 Rx23 Rx24 Rx25 Rx26 Rx27 Tx25 Rx6 Rx7 Rx8 Rx9 Rx10 Rx11
Tx10 Rx23 Rx24 Rx25 Rx26 Rx27 Rx28 Tx26 Rx7 Rx8 Rx9 Rx10 Rx11 Rx12
Tx11 Rx24 Rx25 Rx26 Rx27 Rx28 Rx29 Tx27 Rx8 Rx9 Rx10 Rx11 Rx12 Rx13
Tx12 Rx25 Rx26 Rx27 Rx28 Rx29 Rx30 Tx28 Rx9 Rx10 Rx11 Rx12 Rx13 Rx14
Tx13 Rx26 Rx27 Rx28 Rx29 Rx30 Rx31 Tx29 Rx10 Rx11 Rx12 Rx13 Rx14 Rx15
Tx14 Rx27 Rx28 Rx29 Rx30 Rx31 Rx0 Tx30 Rx11 Rx12 Rx13 Rx14 Rx15 Rx16
Tx15 Rx28 Rx29 Rx30 Rx31 Rx0 Rx1 Tx31 Rx12 Rx13 Rx14 Rx15 Rx16 Rx17

Fig. 5. Comparison of DAR and resolution.

these techniques are able to achieve better timing performances

in data acquisition process compared with the conventional
single light projection.
The above multiple fan beam projection achievement can be
further verified by comparing the number of sensors and DAR
obtained by Chan [2] with the results achieved in this research,
whereby the multiple projection technique has a capability to
Fig. 4. Determining emission angle.
increase the resolution of the hardware system (a higher number
Due to the limitation of the fibre optic’s emitting angle, it is of sensors installed) and at the same time increasing the DAR
indeed an advantage to implement multiprojection to the light (shorter time needed for data conversion in one frame). The
projecting system without worrying about the overlapping of graph shown in Fig. 5 represents the improvement for the DAR
received photodiode signals. For the 32 transmitter model, Txn achieved by multiple projection technique in this research when
and Tx16+n will project light at the same time, for example, Tx0 compared to the single projection result achieved by Chan [2].
and Tx16 , Tx1 and Tx17 etc. will project light at the same time. The projection geometry of both the techniques is shown in
16 sets of projection will complete one frame of light emission. Fig. 6.
This will also minimize the total time for one frame of data to Considering that the 32 transmitters and 32 receivers are
be sent to the computer thus obtaining a higher data acquisition arranged alternately in clockwise starting from Tx0, Rx0, Tx1,
rate. Rx1 until Tx31 and Rx31, the transmitters emit light according
to this arrangement sequence. At a transmission angle of 30◦ ,
4.2. Multiple fan beam projection geometry each projection provides six light beams (six views) which
will be received by the corresponding receivers as tabulated in
Implementation of the multiple projection technique Table 1.
using switch-mode fan beam method yields two projection For the 2-projection technique, sixteen sets of projections
geometries which are the 2-projection geometry and the will complete one frame of light emission whereby Txn and
4-projection geometry. When two or four transmitters project Tx16+n (with n as the respective number projection ranging
light at the same time, the total time to complete one frame from 0 to 15) will transmit light at the same time. As for the
of data acquisition will be shorter, thus it is believed that both 4-projection technique, eight sets of projections will complete
6 R. Abdul Rahim et al. / ISA Transactions 47 (2008) 3–14

Fig. 6. Multiple projection geometry. (a) 2-projection (b) 4-projection.

Fig. 7. Optical fibre sensors arrangement on sensor’s fixture.

one frame of light emission with Txn , Tx8+n , Tx16+n and By distributing the fibre optics evenly around the
Tx24+n (taking n as the respective number projection ranging circumference of the fixture as shown in Fig. 6, the angle
from 0 to 7) transmitting light at the same time. between each emitter and its adjacent receiver viewed from the
The projection geometry of the sensors determines the centre of the circle is 5.625◦ (dimensions not to scale in figure).
relationship between the sensors’ arrangements and its The optical fibre sensors arrangement in Fig. 7 is seen
mathematical modeling. In this research, thirty-two pairs of from the top view. The sensor’s fixture is designed to be
optical sensors are employed. The optical sensors are divided bigger than the pipeline used. Tx and Rx represent the
into thirty-two infrared transmitters and thirty-two photodiode respective transmitters and receivers as they are arranged
receivers which are placed alternately. These optical sensors alternately in clockwise direction, starting from Tx0 and
are each coupled to separate fibre optics to the periphery of ending at Rx31. The 80 mm dimension is the diameter of the
the sensor’s fixture. The diameter of the sensor’s fixture which investigated pipe and the red gridlines are the 32 × 32 mapping
mounts the fibre optics is 100 mm and its circumference equal resolution which is applied in solving the forward and inverse
to 314.2 mm. problems.
 R. Abdul Rahim et al. / ISA Transactions 47 (2008) 3–14 7

Fig. 8. Optical fibre sensors mapping platform.

To simplify the mapping of optical fibre sensors to the two- Pn .x = the xth coordinate of the nth sensor.
dimension image plane, it is assumed that diameters of both Pn .y = the xth coordinate of the nth sensor.
optical fibre sensors (transmitters and receivers) are the same r = the radius of the circle which is 320 pixels.
which is 2.20 mm. This assumption is made because the size of θ = angle between emitter and its adjacent receiver viewed
the fibre optic’s inner core is 1 mm and after they are lensed, from the centre of the circle, which is 5.625◦ .
the surface of the inner core has an approximate diameter of
By using the mapping methods as above, the computer
2.20 mm. A two-dimension image plane which is made up of
generated coordinates of all the sensors are tabulated in Table 3.
640 × 640 pixels in Fig. 8 serves as the mapping platform for
From the 640 × 640 pixels, the resolution of the plane is
the optical fibre sensors.
divided into 40 × 40 rectangles. Each rectangle has a total of
The mapping of sensors on the image plane is important in 256 pixels. By definition, pixel represents the smallest unit of
solving the forward problem. Through Visual C++ program- an image and resolution refers to the clarity or sharpness of
ming, the image plane is set to 640 × 640 pixels using two- an image, initiated by the number of pixels per square inch
dimensional Cartesian coordinates. The top-left corner coordi- on a computer-generated display. When the transmitters and
nate is at (0, 0) and the right-bottom end coordinate is (640, receivers are distributed evenly on the periphery of the sensor’s
640) as shown in Fig. 8. To map the sensors, labeled as P0–P64, fixture, the equispaced interval [5] between each transmitter and
calculations are done by dividing the circle into four quadrants the subsequent receiver is 2.71 mm by using Eq. (1).
as presented in Fig. 8.
For all the quadrants in Fig. 9, sensors P0–P16 can be C − (2.20)n
E.S = (1)
mapped by using the drawings and equations in Table 2, n
whereby: whereby:
8 R. Abdul Rahim et al. / ISA Transactions 47 (2008) 3–14

Table 2
Mapping sensors to image plane by using Cartesian coordinates

Quadrant Mathematical drawing Equations

Pn .x = r + r sin θ
Q1
Pn .y = r − r cos θ

Fig. 9. Dividing the circle into four quadrants.

image plane, each light beam will spread across each rectangle
in different weights as illustrated in Fig. 10.
The solution of forward problem generates a series of
Pn .x = r + r sin θ
sensitivity maps. And the inversion of these sensitivity
Q2 matrixes therefore provides a reconstruction of the optical
Pn .y = r + r cos θ
properties [10]. Basically, the number of generated sensitivity
maps is dependent to the number of projections for the
sensors. No matter whether the applied projection method is the
2-projection or 4-projection method, each light beam must be
sampled individually. Therefore, for the thirty-two transmitters
and the corresponding six receivers which receive light per
emission of each transmitter, the total projections are 192.
Pn .x = r − r sin θ This means that there will be 192 sensitivity maps generated
Q3
Pn .y = r + r cos θ as well. Before these sensitivity maps can be used for image
reconstruction, they must be normalized first.
Normalization of sensitivity maps is done when there is no
simple linear relationship between the weight distributions in
a rectangle for different light projections. For example, the
rectangle (25, 30) might have certain percentage of light passing
through for the light beam of transmitter 10 and receiver 25.
When transmitter 0 emits light to receiver 16, this light beam
Pn .x = r − r sin θ might pass through rectangle (25, 30) with a different weight
Q4
Pn .y = r − r cos θ distribution. If there are other light beams passing through the
same pixel as well, the weight distribution of the pixel will
become complicated. Thus, it is necessary to build a general
relationship between the pixels and all the projection beams
that pass through the pixels. A simple approach to normalize the
rectangle values in this research is engaged in a similar manner
as normalizing the pixel values in ECT so that they have the
values of zero and one when the rectangle contains the lower
E.S = Equispaced interval (mm). and higher weight distributions respectively.
C = Circumference (mm). As a reference to map the sensitivity maps onto the
n = A total of 64 optical fibre sensors. 2-dimension image plane, Fig. 8 is being referred. From the
These predetermined geometrical coordinates and rectangles 640 × 640 pixels plane, there are gridlines to divide the pixels
will be useful in sensor modelling and generating sensitivity into 40 × 40 rectangles where each rectangle contains 16 × 16
maps. pixels. When a light projection maps on the rectangles, a certain
number of pixels in once particular rectangle will be dubbed
4.3. Sensitivity maps into either ‘1’ for pixels covered with light beam or ‘0’ for
pixels not covered by light beam. The total number of ‘1’ pixels
From the fan beam projection properties [1], it is known of light beam will be counted for each rectangle which will be
that the fan beam projection can be seen as a point source of then represented as the total weight of the rectangle involved.
radiation that emanates as a fan shaped beam. The emanation In image reconstruction, the actual area of flow is from pixel
covers in many directions in different angles. For a given source 64 to pixel 575 in x-axis and y-axis. In this section the image
and detector combination, functions are calculated that describe plane is reconfigured so that all calculations and scanning of
the sensitivity of each source and measurement pair to changes pixels are done only for the part of the actual flow area. This
in optical properties within each pixel of the model [10]. Thus, is being done to avoid wasting unnecessary time and resources
when the projection beam is mapped onto the two-dimensional used to calculate and scan the whole 640 × 640 pixels. The
 R. Abdul Rahim et al. / ISA Transactions 47 (2008) 3–14 9

Table 3
The full coordinates of all the sensors from P0–P63
Sensor Coordinate Sensor Coordinate Sensor Coordinate Sensor Coordinate
P0 (351, 2) P16 (638, 351) P32 (289, 638) P48 (2, 289)
P1 (382, 7) P17 (633, 382) P33 (258, 633) P49 (7, 258)
P2 (412, 14) P18 (626, 412) P34 (228, 626) P50 (14, 228)
P3 (442, 25) P19 (615, 442) P35 (198, 615) P51 (25, 198)
P4 (470, 38) P20 (602, 470) P36 (170, 602) P52 (38, 170)
P5 (497, 54) P21 (586, 497) P37 (143, 586) P53 (54, 143)
P6 (523, 73) P22 (567, 523) P38 (117, 567) P54 (73, 117)
P7 (546, 94) P23 (546, 546) P39 (94, 546) P55 (94, 94)
P8 (567, 117) P24 (523, 567) P40 (73, 523) P56 (117, 73)
P9 (586, 143) P25 (497, 586) P41 (54, 497) P57 (143, 54)
P10 (602, 170) P26 (470, 602) P42 (38, 470) P58 (170, 38)
P11 (615, 198) P27 (442, 615) P43 (25, 442) P59 (198, 25)
P12 (626, 228) P28 (412, 626) P44 (14, 412) P60 (228, 14)
P13 (633, 258) P29 (382, 633) P45 (7, 382) P61 (258, 7)
P14 (638, 289) P30 (351, 638) P46 (2, 351) P62 (289, 2)
P15 (639, 320) P31 (320, 639) P47 (1, 320) P63 (320, 1)

Fig. 10. Fan beam projection map.

resolution of the actual image plane thus becomes 32 × 32 in Next is to sum and normalize all the 192 sensitivity maps
image reconstruction which contains 512 × 512 pixels. This using the following two approaches:
explanation can be further explained graphically in Fig. 11.
After the actual flow area is being identified, the sensitivity 1. Summing the maps according to the weights of the same
maps are generated using Visual C++ programming language rectangles from all projections and normalizing the maps
as shown in Eq. (2). (rectangle-based normalization).
511 X
511 2. Summing the maps based on weights of all the rectangles
X
Si, j (x, y) = Px,y (a, b) for individual light projection and normalizing the maps
a=0 b=0 (projection-based normalization).
Px,y (a, b) = 1;

if black = changed The rectangle-based normalization is to total up all the total
(2)
Px,y (a, b) = 0; if white = unchanged black pixels from all different projections in a same rectangle
whereby: using Eq. (3) and normalize the maps according to Eq. (4).

Si, j (x, y) = sensitivity map of light beams for all transmitters 32 X

X 16 
0 ≤ x < 32
and receivers. T 1(x, y) = Si, j (x, y) (3)
0 ≤ y < 32
Px,y (a, b) = an array to represent the total black pixels in i=0 j=0
rectangle-x y.  Si, j (x, y)

; for T 1(x, y) > 0
a and b = the ath column and bth row of the pixels in the N 1i, j (x, y) = T 1(x, y) (4)
actual flow plane. 0; for T 1(x, y) = 0
x and y = the xth column and yth row of the 32×32 rectangles
or resolution. whereby:
10 R. Abdul Rahim et al. / ISA Transactions 47 (2008) 3–14

Fig. 11. The actual flow image plane.

T 1(x, y) = the total or sum of the same element in rectangle- 5. Results and discussions
x y obtained from the 192 sensitivity maps.
Si, j (x, y) = the 192 sensitivity maps. 5.1. Mass flow rate results
N 1i, j (x, y) = the normalized sensitivity maps for light beams Mass flow rate or MFR can be defined as the number
of all Tx (0 ≤ i < 32) to Rx (0 ≤ j < 6) using the of grams or kilograms of flow that flows pass a given
rectangle-based normalization. cross-sectional area per second (unit of g/s or kg/s). For a
tomographic system with two layers of sensors (upstream and
Meanwhile, by using the projection-based normalization, the downstream), the mass flow rate of the flow can be obtained
total weights of pixels for all rectangles (32 rectangles) for each by multiplying the density of the flow material (g/m3 ) with
light projection is being summed by using Eq. (5) and then velocity of the flow (m/s) and also with the cross-sectional of
normalized as stated in Eq. (6). the pipe which the flowing objects flow through (m2 ). Here,
32 X
32 the velocity of the flow is being determined by performing
X
T 2(i, j) = Si, j (x, y) (5) cross-correlation of the concentration profiles for the upstream
x=0 y=0 and downstream sensors. However, for the single layer of
sensors implemented using the optical fibre fan beam method
T 2(i, j) > 0

Si, j (x, y) in this research, the velocity of the flowing object cannot be
N 2i, j (x, y) = 0 ≤ i < 32 (6) determined by using the cross-correlation method.
T 2(i, j) 
0≤ j <6 Instead, an alternative approach introduced by Ruzairi [8]
whereby: is taken to obtain the mass flow rate for the solid/gas flow.
First, the solid material for the flow system must be determined
T 2(i, j) = the total of pixel weights for all rectangles in because different solid materials have different densities and
individual light projection. sizes and therefore yields different flow properties. Second,
by using the selected solid material, the specified flow rig
Si, j (x, y) = the 192 sensitivity maps.
must be calibrated. This will result in a relation between the
N 2i, j (x, y) = the normalized sensitivity maps for light concentration measurements and the manually obtained mass
beams of all Tx to Rx using the projection-based flow rates. The manually obtained mass flow rates are the actual
normalization. mass flow rates for the flowing system.
Instead of using the gravity conveying system with rotary
Since the normalized maps generated in Eqs. (4) and (6) has screw feeder, a range of filters are designed and used. This
a total of 192 maps each. is to solve the problems highlighted by Chan and Ruzairi [3]
 R. Abdul Rahim et al. / ISA Transactions 47 (2008) 3–14 11

Fig. 12. Flow rig calibration process.

whereby the resulting flow of the rotary screw feeder can cause
un-uniformed flow. The main aim to investigate the MFR in this
research is to provide an estimate of the range of mass flow that
will allow the optical fibre sensors to perform reliably. The flow
diagram of the calibration process is shown in Fig. 12 while the
experimental properties of the dynamic flow system are being
identified as follows:

5.2. Experimental conditions

Flow rig: Gravity flow rig in experimental scale.

Diameter of pipe used: 80 mm. Fig. 13. Flow rig calibration curve for plastic beads.
Length of pipe until the sensors: 1.2 m.
Flowing material: Plastic beads. ‘low’ until the light beams are not intercepted or when the user
Mean particle size of the plastic beads: Nominal diameter of stops the oscilloscope. Thus, the total time taken for the flow to
3 mm. commence is the duration of negative pulse width of the output
Filters: 10%, 20%, 30%, 40%, 50%, 60%, 70%, 80%, 90% waveform. The timer application can perform better than using
of the flow regime. the stopwatch because the start and stop time of the stopwatch is
Sensor array structure: 32 optical fibre transmitters and 32 user-operated and there is bound to be inaccuracy due to human
optical fibre receivers. error and delays.
Photo-sensors: Infrared emitters and photodiode receivers. When the flow has stopped, the collected plastic beads in
Peak wavelength of light source: 880 nm. the tank is weighed using the weighing machine. The weight
Type of projection: 4-projection switch-mode fan beam of the measured plastic beads and the total time needed for the
projection technique. flow are keyed into the computer for data analysis. After that,
Pixel grids/resolution: 32 × 32. the whole process is repeated until all the filters have beeen
With reference to the Fig. 12, the flow lines represents the used. Using a simple technique to obtain the best results in
flow of the calibration process while the dotted lines show the experiments, the process is repeated five times for each filter
flow data to the computer to be analysed. First, the plastic beads to get an average mass flow rate. The graph for average MFR
are manually filled into the hopper. At this time, the valve is still versus the filters is illustrated in Fig. 13.
closed to prevent flow from falling through the pipeline. Next, By referring to Fig. 13, although the points in the graph
the filter is inserted and then the timer is reset. When the valve could be fitted by a higher order polynomial, a first order
is opened, the plastic beads will flow through the pipeline into polynomial has been fitted to the data to simplify the equation
the tank and the timer will start to count when it senses the first for easier manipulation and calculation [8]. The equation of the
particle that intercepts its light beam. linear regression line [4] is formed using Eq. (1).
The timer used here comprised the infrared projection Average MFR (g/s) = 23.256 ∗ [Filter (%flow)] . (7)
circuits, photodiode receiver amplifying circuits, logic gates
and comparator circuits. The output of the timer is connected The sampling time is set at 5 kHz to obtain 100 sets of
to an oscilloscope to monitor the output waveform; the output concentration measurements. After the flow rig calibration has
will stay at ‘high’ until the light beams from the infrareds to been completed, the timer is uninstalled and the optical fibre
the photodiodes are being intercepted. The output will stay at sensor array is attached to the pipeline (the same location as the
12 R. Abdul Rahim et al. / ISA Transactions 47 (2008) 3–14

Fig. 14. Relationship between average MFR and iteration number.

Table 4
Percentage of flow obtained from concentration profiles

%Flow MFR (g/s)

LBP 10th iteration
0 0 0
10 265.79 190.729
20 554.39 475.952
30 824.98 784.808
40 1171.3 1179.454
50 1743.7 1777.036
60 2060.6 2163.671 Fig. 15. Relationships between average MFR and percentage of flow.
70 2176.7 2240.975
80 2161.6 2215.693
90 2008.8 2053.892 Table 5
100 2016.3 2089.983 Slopes and corrective coefficients for MFR graphs

Graph property Calibration LBP 10th iteration

timer). The whole process in Fig. 11 is repeated, only now the Slope, m 23.256 26 24.211
sensors are used to capture the voltage loss to perform image Slope ratio, r – 0.894 0.961
reconstruction. Through the data acquisition system, 100 sets Linear equation y = 23.256x y = 26x y = 24.211x
Corrective coefficient, K – 0.894 0.961
of data are taken per solid/gas filter. The sampling time is set at
5 kHz. These sensor losses are processed to obtain 100 sets of
concentration measurements. The average concentration profile It is observed that the average MFR acquired by the sensors
is obtained by adding all the concentration measurements and is only linear up to the range of 40% flow. This verifies the
the bulk value of concentration profile is divided by the number results obtained from the static flow model experiment using
of datasets, which are 100 in this case. There are 32 × 32 half flow whereby the half flow or 50% of flow cannot be
pixels used to form the reconstructed tomogram, thus in order measured by the sensors [6]. For further analysis, the linear
to obtain the percentage of flow for the concentration profile range of the average MFR from flow rig calibration, LBP
Eq. (8) is implemented. algorithm and the 10th iteration are integrated in Fig. 15.
 1024  It is observed that the linear regression of the calibration,
P
Vlbp[i] LBP and the 10th iteration lines have different slopes each. The
closer the graph is to the calibration line, the more accurate
 
 i=1
%Flow =   ∗ 100 (8)

 5120  the MFR is. A corrective coefficient is needed for the LBP
and 10th iteration graphs so that they behave the same way as
the calibrated data. To find the stated corrective coefficient, the
whereby: slopes of all the three graphs have to be calculated. The slopes
%Flow = percentage of flow obtained. and coefficients needed for the graphs are listed in Table 5
whereby slope ratio for calibration is not available because this
Vlbp[i] = the concentration profile of the image at ith pixel.
slope will be used as reference for comparison with LBP and
Eq. (6) to obtain the average MFR of the flow. Two image 10th iteration slopes.
reconstruction algorithms are used to process the measured The y and x term in the linear equations represents the
sensor values into concentration profiles [6]. Table 4 shows y-axis and x-axis of the MFR graphs. From the table, the slope
the results obtained from real-time data acquisition and Fig. 14 ratio, r of the 10th iteration is smaller than the r of LBP, thus
illustrates the relationship of the average MFR obtained for the it can be concluded the 10th iteration image reconstruction
LBP and iterative reconstruction algorithm. provides a more accurate image than the LBP with reference to
 R. Abdul Rahim et al. / ISA Transactions 47 (2008) 3–14 13

Table 6
Investigation of image processing rate using different computer speeds

Computer properties ISA slot availability Image processing rate (fps)

LBP 1st iteration 5th iteration 10th iteration
Intel Pentium III, 667 MHz Yes 12 8 3 2
Intel Pentium III, 933 MHz Yes 24 15 6 4
Intel Pentium IV, 1.5 GHz No 77 35 11 6
Intel Pentium IV, 2.4 GHz No 136 63 20 11

the calibration graph. By applying the calculated K values into Although the functions above are being selected to execute
both image reconstruction algorithms, the new linear equation the image processing and display in a most optimized manner,
for the graphs is presented in Eq. (9). it is found out that the speed of the image processing rate is
dependent on the processing speed of the computer used. The
yLBP = (26.000 ∗ xLBP ) ∗ 0.897 (9a) image processing rate for the image reconstruction using LBP
y10th Ie = (24.211 ∗ x10th Ie ) ∗ 0.961. (9b) and iterative algorithms is investigated using different speed
specifications of the computer. The result of the investigation
Overall, the conclusion that can be drawn is that for the
is tabulated in Table 6.
plastic bead flow using the gravity flow rig on an experimental
During the investigation of the image processing rate, the
scale, the optical fibre sensors implemented in this research are
same GUI program is being run in four different computers
suitable to measure mass flow rates at the range of zero to not
with different processor speed. It is noticed that the IPR
more than 40% of the flow. The slope ratio, r , obtained proved
increases with the increase of processor’s speed as shown in
that the iterative method results in a better image for low density
Table 6. From the table also, the ISA slot availability is being
flows.
checked for each of the computer because the available Keithley
DAS1802HC data acquisition card in the laboratory requires
5.3. Image processing rate (IPR)
the ISA slot for communication. Recent Intel Pentium IV
computers have only the PCI slots and therefore are not suitable
The image processing rate is the comparative measurements to be used for a real-time data acquisition process although
of the time taken to process and display the image using they have the potential to achieve a higher image processing
different reconstruction algorithms. Based on the optimization rate. Instead, the Intel Pentium III 933 MHz computer is used
concept, a programmer should use fewer codes to accomplish because it has the fastest processor speed among the computers
the application and spend the minimum time to execute the with ISA slots. As a conclusion, it is thus proved that using
program written. In terms of optimizing the execution of the same programming, different computer speeds will result in
image reconstruction and displaying the image in bitmaps, a different image processing rates.
suitable timer and image display method are essential. In this
research, the Device Dependent Bitmap (DDB) method is used 6. Conclusions
to draw the tomogram because it is considered the fastest
way to construct and display an image compared to the other From the dynamic experiment of solid/gas flow using the
methods [7]. DDB is controlled by the device driver of display plastic beads in a gravity flow rig, the designed optical fibre
card and can only be displayed on the device with the correct sensors are reliable in measuring the mass flow rate below 40%
configuration. By comparing to the Device Independent Bitmap of flow.
(DIB), DDB can save processing time because DIB has to be Another important matter that has been discussed is the
converted to DDB before displaying the image onto screen. image processing rate or IPR. The IPR is dependent on
Next, is the concern of using the suitable timer to execute the optimization of certain functions used in processing the
the counting of time needed to process and display the image. image using different image reconstruction algorithms and also
From the research done by Pang [7], regarding the timing displaying the tomogram GUI. This research has opted for
functions found in Win32 library he has found out that the the most optimized applications of DDB display method and
high performance timer is the only timer that has the ability also the high performance timer which is able to measure
to measure a minimum time resolution of 1 µs while the rest of the minimum time resolution of 1 µs. The only factor that
the timers such as the GetTickCount, Clock, timeGetTIme and limits the IPR of this research is that fastest available processor
ftime timers can only count the minimum time resolution of speed of the computer with ISA slot is the Intel Pentium III
1 ms. Pang [7] has also done an experiment to find the standard 933 MHz, which has a slow processor speed compared to the
deviation of the timers by counting a routine’s execution Intel Pentium IV processors. The choice of computer is limited
time using 20 sets of data. The results showed that the high because the Keithley DAS 1802HC requires communication
performance timer has the lowest standard deviation value with the ISA slot and only the Intel Pentium III processors
which means that this timer is good in repeatability and has have this feature. Generally, the applied image reconstruction
the highest precision. Thus, based on his deduction, the high algorithms, the construction of the sensor and also the designed
performance timer is being applied in this research. software is considered to be reliable and is suitable to
14 R. Abdul Rahim et al. / ISA Transactions 47 (2008) 3–14

perform real-time image reconstruction and mass flow rate tomography sensor configuration using 4 parallel beam projections. In:
measurements with an overall accuracy of about 85%. 3rd international symposium on process tomography. 2004, p. 48–51.
[8] Ruzairi Abdul Rahim. A tomography imaging system for pneumatic
conveyors using optical fibres. Ph.D. thesis. Sheffield Hallam University;
References 1996.
[9] Ibrahim Sallehuddin. Measurement of gas bubbles in a vertical water
[1] Abdul Rahim R, Chan KS, Pang JF, Leong L. A hardware development column using optical tomography. Ph.D. thesis. Sheffield Hallam
for optical tomography system using switch mode fan beam projection. University; 2000.
Journal of Sensors & Actuators A: Physical 2005;120(1):277–90. [10] Xu H, Dehghani H, Pogue PW, Springett R, Paulsen KD, Dunn JF.
[2] Chan Kok San. Real time image reconstruction for fan beam optical Near-infrared imaging in the small animal brain: Optimization of fiber
tomography system. M.Sc. thesis. Universiti Teknologi Malaysia; 2002. positions. Journal of Biomedical Optics 2003;8(1):102–10.
[3] Chan KS, Ruzairi AR. Tomographic imaging of pneumatic conveyor
using optical sensor. In: 2nd world engineering congress. 2002.
[4] Green RG, Horbury NM, Abdul Rahim R, Dickin FJ, Naylor BD, R. Abdul Rahim received BEng. Degree with Honours
Pridmore TP. Optical fibre sensors for process tomography. Measurement in Electronic System And Control Engineering in
Science & Technology 1995;6:1699–704. 1992 from Sheffield City Polytechnic. He received his
[5] Kak AC, Slaney Malcolm. Principles of Computerized Tomographic Ph.D. in Instrumentation & Electronics Engineering
Imaging. Electronic ed. New York (United States): IEEE Press; 1999. from Sheffield Hallam University in 1996. Currently
[6] Leong Lai Chen. Implementation of multiple fan beam projection he is a Deputy Dean (Corporate Services), Research
Management Centre, Universiti Teknologi Malaysia.
technique in optical fibre process tomography. M.Sc. thesis. Universiti
His current research interests are process tomography
Teknologi Malaysia; 2005. and process measurement.
[7] Pang Jon Fea, Ruzairi Abdul Rahim, Chan Kok San. Infrared