You are on page 1of 13

IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 59, NO.

6, JUNE 2010 1507

A Multisensor Intelligent Device for Real-Time


Multiphase Flow Metering in Oil Fields
Mahmoud Meribout, Member, IEEE, Nabeel Z. Al-Rawahi, Member, IEEE, Ahmed M. Al-Naamany,
Ali Al-Bimani, Khamis Al-Busaidi, and Adel Meribout

Abstract—In this paper, a new multiphase flow metering device treatments, such as enhanced oil recovery, to be performed to
for real-time measurement of oil, gas, and water flow rates is increase the well productivity. In another application, a down-
presented. It is composed of several electrical and acoustic sensors hole monitoring of the multiphase flow allows well engineers
whose signals are digitalized and processed by a multilayer neural
network. This latest uses the physical models of multiphase fluids to control more effectively the propagation of the oil from the
to reduce the complexity of the parameter space while improving actual well by adequately controlling the array of valves in that
its accuracy. Furthermore, to overcome the uncertainties of the well (e.g., switching off the valve surrounded by high water-
electrical sensors in the range of 40%–60% and above 90% wa- cut fluid). A suitable instrument for such applications should be
ter-cut (i.e., ranges where most of the multiphase flow meter fail), capable of measuring the flow rates of each phase composing
two rings of high- and low-frequency ultrasonic sensors are used
for low and high gas fractions, respectively. The results of exper- the mixed fluid, which is usually oil, water, and gas. However,
iments that have been conducted in an in-house laboratory-scale such instrument, called a multiphase flow meter (MPFM), is
multiphase flow loop show that real-time classification for up to complex to design since it has to determine four unknown
90% gas fraction can be achieved with less than 10% relative error. parameters (i.e., oil fraction, water fraction, gas fraction, and
Index Terms—Artificial intelligence, capacitance and conduc- total flow rate), in addition to the fact that most of the available
tance probes, embedded systems design, gas flow rate measure- sensing techniques cannot simultaneously deal with all the three
ment, multiphase flow metering, neural network, ultrasonic waves, phases. Thus, use of several complementary sensors is required.
water-cut measurement. Dealing with three unknown parameters (i.e., oil fraction, water
fraction, and total flow rate) has been successfully achieved
I. I NTRODUCTION with less than 5% error within the whole water-cut range
[1]. These meters, however, are applicable only after complete
R EAL-TIME measurement of the volumetric flow rate of
process fluids (e.g., flow rate of the mixture, individ-
ual phase flow rates, and fractional phase volumes) prior to
liquid–gas separation using bulky and costly separation tanks.
Consequently, they are impractical for applications such as
separation is widely acknowledged to be a critical parameter monitoring flows of individual wells or remote onshore oil
in process control within the petrochemical industries since it fields. Thus, there is an increasing need for portable, cheap,
allows a more effective reservoir management. For example, in and safe MPFMs to be widely deployed in oil fields worldwide.
offshore production consisting of several nearby wells (Fig. 1), Following extensive research that has been done, most of
flow data gathered along a pipeline section of a given well can the existing multiphase flow metering instruments belong to
help in identifying how this well contributes to the aggregate the following two main technologies: 1) gamma-ray devices
flow and, hence, may help in locating a production anomaly, and 2) dielectric-based devices. However, these devices are
such as a water or gas breakthrough in the actual well. This either not safe (because of the use of radioactive sources) or
allows an easier localization of well stimulation or other well inaccurate in the case of the 40%–60% water-cut (i.e., the
percentage of water in the mixed fluid) or above the 90% water-
Manuscript received March 13, 2009; revised April 30, 2009. First published cut. This paper presents a new nonradioactive ultrasonic-based
September 25, 2009; current version published May 12, 2010. The Associate intelligent multisensor (IMS) multiphase flow metering device
Editor coordinating the review process for this paper was Dr. Dario Petri. for the determination of the volume flow rates of multiphase
M. Meribout was with the Department of Electrical and Computer Engi-
neering, Sultan Qaboos University, Muscat 123, Oman. He is now with the mixture components along a portion of a pipeline with different
Department of Electrical Engineering, Petroleum Institute, Abu Dhabi, United flow regimes and without prior separation of gas. In addition
Arab Emirates (e-mail: kassantina@hotmail.com). to homogenizing the flow, some properties of fluid mechanics
N. Z. Al-Rawahi is with the Department of Mechanical and Industrial
Engineering, Sultan Qaboos University, Muscat 123, Oman. are introduced into the pattern recognition algorithm and have
A. M. Al-Naamany is with the Department of Electrical and Computer led to the decomposition of the complicated nonlinear multi-
Engineering, Sultan Qaboos University, Muscat 123, Oman.
A. Al-Bimani is with the Department of Petroleum and Chemical Engineer-
dimensional spaces into simpler 2-D nonlinear spaces that are
ing, Sultan Qaboos University, Muscat 123, Oman. easily manageable with a dedicated multilayer neural network
K. Al-Busaidi is with Petroleum Development Oman, Muscat 113, Oman. algorithm. The system, which is compact and easily portable,
A. Meribout was with the Department of Electrical and Computer Engi-
neering, Sultan Qaboos University, Muscat 123, Oman. He is now with the has been implemented and extensively tested on a laboratory-
Department of Instrumentation and Control, Sonatrach Corporation, Algiers scale multiphase flow loop with various flow regimes, fluid
2533, Algeria. densities, and flow rates. Experimental results indicate that an
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org. error rate of ±10% can be achieved for real-time classification
Digital Object Identifier 10.1109/TIM.2009.2028210 of up to 90% gas fraction.

0018-9456/$26.00 © 2009 IEEE

Authorized licensed use limited to: Centro Federal de Educacao Tec do Rio de Janeiro. Downloaded on July 09,2021 at 23:25:01 UTC from IEEE Xplore. Restrictions apply.
1508 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 59, NO. 6, JUNE 2010

Fig. 1. Example of application of the MPFM in oil fields.

II. P REVIOUS R ELATED W ORKS measurement. These MPFMs have the advantage of being safe.
However, their accuracy is weak in the water-cut range of
Multiphase flow is a complex phenomenon that is difficult
40%–60%. The reason is that within this range, the mixed fluid
to understand, predict, and model. Common single-phase char-
is neither totally conductive nor an isolator, thus leading to
acteristics, such as velocity profile, turbulence, and boundary
almost the same sensor outputs within this range. In addition,
layer, are thus inappropriate for describing the nature of such
these meters are inaccurate for the water-cut range greater than
flows. Thus, most of the existing MPFMs [2]–[8] rely on the
90%. Another solution proposed in [12] uses a water-cut meter
electrical (e.g., dielectric properties) and/or other nonelectrical
and a volumetric flow-meter for measuring the gas and liquid
properties (e.g., waves or energy propagation) of the individual
phases. This invention is complicated because it requires a
phases that compose the mixed fluid to proceed with a proper
positive displacement instrument so it can avoid the problem of
calibration using either pattern recognition or lookup table
slip between the gas and liquid phases. In addition, this system
techniques, but rarely with analytical equations.
does not appear to be effective for liquid fractions below about
MPFMs using gamma- or X-rays have been successfully
10%. Another field-programmable-gate-array-based device ad-
tested in several oil fields [11], [14]–[17]. Their principle is to
dressed in [5] has been presented to compute the total mass
emit one or several waves to determine the fractions of each
flow rate of the fluid passing through it. The device has no
individual flow composing the mixed fluid. Thus, by knowing
moving mechanical parts to wear out; therefore, its theoretical
the total flow rate of this fluid using the venturi meter, the
lifespan is almost infinite. However, the device cannot deliver
individual flow rates of oil, water, and gas can be determined.
the flow rates of each individual phase constituting the fluid and
However, these meters are radioactive and, thus, are not safe to
becomes inaccurate with the presence of the gas phase. Other
be deployed in hazardous oil fields. In addition, their accuracy
embedded and processor-based devices for the measurement
greatly decreases with the presence of gas (e.g., more than 20%
of the quantity of the fuel in the engine have been proposed
error for more than 90% gas). The reason is that the online
in [6] and [7], and their accuracy is claimed to be high and
gas–liquid separator embedded in these meters may not entirely
independent of the fluid properties (e.g., viscosity, temperature,
separate the gas phase from the liquid phase, inducing substan-
and density). However, these devices are only dedicated for one
tial errors to the water-cut meter, which is connected to the liq-
single phase and do not seem to be applicable for the gas phase.
uid outlet of the separator. To remedy the errors introduced by
Thus, it is clear that, still, a substantial amount of development
the online separator, other MPFMs do not use any separator but
efforts is required, particularly in improving the accuracy levels
instead mix the flow using mechanical mixers in such a way that
over the full range of flow multiphase flow conditions.
the slip velocity between all the three phases of the fluid become
negligible, making the individual velocities approximately
equal [9], [10]. However, this concept might be valid only if all
III. P ROBLEM D EFINITION AND E XPERIMENTAL S ETUP
components of the fluid are in the liquid phase since the liquid
flow rate is usually substantially different from the gas flow rate As previously mentioned, multiphase flow metering remains
in normal multiphase transportation because of density differ- a challenging task for several research groups and companies
ence [4]. MPFMs using this approach [11] have achieved more (e.g., in addition to oil companies, food processing, water
than 20% error in the experiments since the used equations treatment, and petrochemical companies need such device as
ignored any interaction between the gas and the liquid phases. well) since it requires to find out four unknown parameters. An
Other nonradioactive commercially available meters, which ideal meter would make independent direct measurements of
rely on the electrical properties of the mixed fluid, use the each of these parameters, which is not possible. One alternative
capacitance and conductance readings to determine oil, gas, and is to use the data fusion of several sensors to compute the
water fractions [18], [19]. In addition, they use pressure sensors, fractions α, β, and γ of gas, oil, and water, respectively. Then,
a temperature sensor, and cross correlation for liquid flow rate knowing the total volume flow rate T (in liters per minute) using

Authorized licensed use limited to: Centro Federal de Educacao Tec do Rio de Janeiro. Downloaded on July 09,2021 at 23:25:01 UTC from IEEE Xplore. Restrictions apply.
MERIBOUT et al.: MULTISENSOR DEVICE FOR REAL-TIME MULTIPHASE FLOW METERING IN OIL FIELDS 1509

additional sensors, the individual flow rates FG , FO , and FW


of gas, oil, and water, respectively, can be computed using the
following equations:

FG (in liters per minute) = T × α


FO (in liters per minute) = T × β
FW (in liters per minute) = T × γ. (1)

Another alternative, in case the total volume flow rate may not
be possible to measure (e.g., in case one of the three phases is
compressible and, thus, does not exhibit a constant volume flow
rate, which is the case of the gas phase), consists of measuring
the total mass flow rate M (since this parameter is always
constant) to provide flow rates according to the following
Fig. 2. Block diagram of the three-phase flow loop.
equations:

FG (in liters per minute) the pattern recognition algorithm to provide the following four
reference parameters:
M (in kilograms per minute) × α
=
ρG FGa
αa =
FO (in liters per minute) FGa + FWa + FOa
FOa
M (in kilograms per minute) × β βa =
= FGa + FWa + FOa
ρO
FWa
FW (in liters per minute) γa =
FGa + FWa + FOa
M (in kilograms per minute) × γ
= (2) Ma (in kilograms per minute)
ρW
= ρW × FWa + ρ0 × FOa + ρG × FGa (3)
where ρG , ρW , and ρO are the densities of gas, water, and oil,
respectively, to be provided to the MPFM during its calibration. where FWa , FOa , and FGa are the actual values of the water, oil,
Thus, as mentioned previously, and to simplify the above and gas volume flow rates (in liters per minute), respectively,
equations, some meters integrate an online gas–liquid separa- as indicated by the single-phase flow meters. The parameters
tor to separately calculate the parameters α and (β, γ). The αa , βa , and γa are the actual fractions of the gas, water, and
other approach, which consists of mixing the flow to provide oil, respectively, and Ma is the actual total mass flow rate.
similar individual flow rates, requires only the knowledge of The goal here is to design an appropriate pattern recognition
the total volume flow rate T to solve the individual flow rates: algorithm to generate in real time and with good accuracy
FG (in liters per minute) = FO (in liters per minute) = FW the flow rates of each phase FW , FO , and FG as close as
(in liters per minute) = T /3. However, both methods are possible to the actual fractions (i.e., FWa , FOa , and FGa ,
hardly applicable in practice for the reasons already mentioned respectively). Hence, during the calibration phase, the operator
in Section I. The IMS-MPFM proposed in this paper consists proceeds by adjusting the valves of the tanks and compressor
of solving the parameters α, β, and γ using four types of to provide various combinations of the flow rates in (3). For
sensors (i.e., capacitance, conductance, liquid ultrasonic, and each of these combinations, a host computer periodically grabs
air ultrasonic sensors), in addition to using the physical model the values of FWa , FOa , and FGa to compute the parameters
of the multiphase flow. The parameter M is solved using αa , βa , and γa while simultaneously collecting data from all
venturi and differential pressure sensors. More details on this the sensors of the IMS-MPFM. An offline calibration of this
meter are presented in Section V. The calibration and design meter is then performed to add new reference vectors into the
of the algorithmic part of the meter required the design and parameter spaces. These latest are then explored by an adequate
development of a multiphase phase flow loop (Fig. 2), within pattern recognition algorithm. Fig. 3 shows the scheduling of
which extensive experiments have been done. In this flow loop, data sampling and processing tasks. The pattern recognition
water and oil are carried out from two separate tanks of 2 m3 program is regularly triggered by a timer interrupt to perform
each under various regimes and flow rates, whereas the gas data acquisition of either one of the single-phase flow meters
phase is provided by a gas compressor. These flows are then or one of the sensors of the IMS-MPFM. This is followed by
mixed to cross the multiphase flow metering device under test. a noise removal and pattern recognition algorithm. The whole
Before the mixing point of the loop, three single-phase flow process corresponding to one sample takes 2 ms, which leads
meters, namely, water, oil, and gas flow meters, are inserted to 500 samples/s, providing to our IMS-MPFM the feature to
after the water tank, oil tank, and air compressor, respectively. handle total volume flow rates of up to 46 m3 /h since the
These single-phase flow meters constitute the references for diameter of the pipes in the flow loop is 2 in.

Authorized licensed use limited to: Centro Federal de Educacao Tec do Rio de Janeiro. Downloaded on July 09,2021 at 23:25:01 UTC from IEEE Xplore. Restrictions apply.
1510 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 59, NO. 6, JUNE 2010

Fig. 3. Scheduling of data sampling and processing tasks.

Fig. 4. Drawing of a generic venuri meter.

IV. P HYSICAL M ODEL OF THE MPFM


To properly value the gas and liquid production from the
wells under various flow regimes, a method that relies on
physical fluid equations to determine their individual flow rates
is required to achieve good accuracy. The purpose here is to Fig. 5. Total flow rate, function of the density, and the venturi probe.
perform adequate feature extractions to reduce the complexity
of the various parameter spaces. where K is a constant, which depends on the parameters D2
and D1 of the venturi (Fig. 4). Fig. 5 shows the plot of the M
function of the actual total mass flow rate [which is similar to
A. Total Flow Rate Determination Using the Venturi Probe the parameter M in (2)]. It can be seen that M is monotonic and
The two differential pressure sensors (p1 and p2 ) attached to cluttered around a straight line (Δ). Thus using the parameter
the venturi probe provide an indication of the total volume flow M as one feature to determine the total mass flow rate might
rate passing through it according to be appropriate. Note in Fig. 5 a few isolated samples for a
mass flow rate below 26 000 kg/h (e.g., samples A, B, and C).
T (in liters per minute) This is due to the limitation of the pressure sensors [p1 and p2
  in (5)] to accurately cope with relatively low pressure values.
 1 πD 2
2 × (p1 − p2 )
=  4 ×
2
× (4) Using an additional venturi probe for low pressure values would
4 ρ overcome this problem.
1− D 2
D1

where D1 and D2 are the higher and smaller diameters of the B. Estimation of the Fluid Density Using a Differential
throats of the venturi, p1 and p2 are the values of pressure at Pressure Sensor and a Venturi Sensor
these two points, and ρ is the fluid density. Fig. 4 shows a
generic drawing of a venturi meter. In multiphase metering, the differential pressure along a
Equation (4), however, is applicable only for uncompressible vertical pipe ΔP is a function of the density of the mixed fluid
fluids such as a liquid mixture and becomes less accurate in passing through it (ρ), augmented with its velocity v, i.e.,
the presence of gas. This is because the quantity of gas passing 1
through the meter is not the same as the total volume measured ΔP (in bars) = (ρ × g × h) + e × v 2 (6)
2
by the reference gas meter. Therefore, directly introducing (4)
to the pattern recognition algorithm would lead to high errors. where h is the distance separating the two pressure sensors, and
In our case, we considered the conservation of mass instead of e is a constant that depends on the internal diameter of the pipe
volume flow rate using the following equations: d, the length between the two pressure sensors h, the friction
 factor f , and the mixed density ρ (e = hf ρ/2d). This equation
M (in kilograms per minute) = K × ρ × (p1 − p2 ) (5) can provide useful information on the quantity of gas passing

Authorized licensed use limited to: Centro Federal de Educacao Tec do Rio de Janeiro. Downloaded on July 09,2021 at 23:25:01 UTC from IEEE Xplore. Restrictions apply.
MERIBOUT et al.: MULTISENSOR DEVICE FOR REAL-TIME MULTIPHASE FLOW METERING IN OIL FIELDS 1511

D. Driving Design Parameters for the Flow Loop and


Pattern Recognition Algorithm
Following the above discussions, an upstream pressure sen-
sor P1 was introduced in the flow loop near the gas flow meter
to provide the actual value of the gas passing through the

MPFM, i.e., FGa . Furthermore, the gas fraction is estimated
using the differential pressure and venturi sensors, as indicated
in (6) and (8). Finally, the mass flow rate M is selected as a
reference for to the pattern recognition algorithm.

V. O VERALL H ARDWARE A RCHITECTURE OF


Fig. 6. Differential pressure, as a function of the gas fraction, for different THE IMS-MPFM
ranges of mass flow rate.
As previously mentioned in Section III, our IMS-MPFM’s
measurement principle involves two groups of sensors (Fig. 7):
through the meter. The parameter space in Fig. 6, built during
one group that comprises two capacitances, a conductance,
the meter calibration, shows that ΔP tends to decrease with
and an ultrasonic array sensor that are used to determine the
an increase in the quantity of gas. In addition, some samples
phase fractions, and another group composed of the venturi
(samples A and B) have been found to provide the same differ-
and differential pressure sensors (DP in Fig. 7) that are used
ential pressure for different gas fractions because of the effect
for mass flow determination. These sensors are fixed into the
of the second term in (6) since their corresponding flow rates,
flow loop by means of flange connections and interfaced to the
measured by the venturi sensor, are different. It is worth noting
computer through both its peripheral component interconnect
that the plots shown in Fig. 6 are obtained by clustering the
and serial (i.e., RS232) buses. The capacitance and conductance
samples into different adjacent ranges of the mass flow rate M .
sensors are used to determine the conductivity and permittivity,
Within each of these clusters, the average value of differential
respectively, of the multiphase fluid, whereas the ultrasonic
pressures that correspond to the same gas fraction is computed.
array sensor provides an indication on the fluid composition in
Thus, the same value of the mass flow rate (e.g., mass flow rates
the regions where neither the capacitance nor the conductance
of 37 000 and 40 000 kg/h) provides the same samples in both
can respond accurately (for water-cut values ranging from 40%
adjacent ranges. This would lead to an easy validation of (6).
to 60%). This constitutes a significant advantage over other
Section VI-D would explain how this figure is explored during
commercially available multiphase meters that use either the
the first of the two phases composing the pattern recognition
conductance or capacitance sensors to determine the phase
algorithm to determine the fluid compositions.
fractions [9], [13], [15], [18], [19].

C. Compensation for the Gas Compressibility During the A. Ultrasonic Sensor


Meter Calibration
High-frequency ultrasound waves can easily be attenuated,
As mentioned in Section III, since the gas phase is com- particularly when their wavelength is smaller than the size of
pressible, the volume of gas passing through one section of bubbles of the gas in the liquid mixture [3]. However, the
the pipe might change when it passes through another one. measurement of the time of flight of the ultrasonic waves is
Consequently, the gas flow rate FGa , which is measured by more accurate when using high-frequency ultrasound waves
the reference single-phase flow meter, might be different from [11], [23]. Thus, the use of different types of ultrasound sensors

the one reaching the MPFM, i.e., FGa . The challenge consists is required to cover all flow regimes: low-frequency ultrasound

then of finding the value of FGa since it is this parameter that sensors (i.e., less than 30 kHz sensors) to determine the fluid
would be considered as the reference for the pattern recognition density in case of high gas fraction containing large gas bubbles
algorithm. Knowing that the gas volume flow rate T is related and high-frequency sensors (i.e., more than 2 MHz sensors)
to the pressure P , according to the following equation [12]: to more accurately determine the fluid density in case of low
u gas fraction with very small or inexistent gas bubbles. Hence,
T (in liters per minute) = (7) two ultrasonic sensors arrays are deployed in our IMS-MPFM.
P
Fig. 8 shows a perspective view of the ultrasound sensors.
where u is a constant that depends on the temperature, one can It comprises two rings of high- and low-frequency sensors,
∗ respectively, fixed to a flanged stainless-steel pipe via threaded
deduce the value of FGa using the following equation:
holders. In each of these two rings, the transit time of the
∗ P1 ultrasonic waves is used as another parameter by the pattern
FGA (in liters per minute) = FGA × (8) recognition algorithm for the determination of the mixed fluid
P0
density. The low-frequency sensors are fixed to the pipe by
where P1 and P0 are the pressure sensors near the gas flow leaving a thin gap of air between the mixed fluid and their
meter and the MPFM, respectively. active surface to satisfy the adaptation of impedance since

Authorized licensed use limited to: Centro Federal de Educacao Tec do Rio de Janeiro. Downloaded on July 09,2021 at 23:25:01 UTC from IEEE Xplore. Restrictions apply.
1512 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 59, NO. 6, JUNE 2010

Fig. 7. Overall architecture of the MPFM.

Fig. 9. Fixation of the ultrasound sensors on the pipe.

slot to transmit a short pulse ultrasonic wave of around 200 V


peak-to-peak amplitude in a time-multiplexed manner using a
bidirectional multiplexer–demultiplexer device.
The other 23-kHz sensors are time multiplexed with a burst
of ten pulses of 20 V peak-to-peak amplitude. The echoes
are then collected by the preamplifier and narrow-passband
filter to enhance the signal-to-noise ratio. The filtered signal
is then amplified further and converted into digital form and
transmitted to the reduced instruction set computer processor
of the transmitter to perform features extraction task such as
Fig. 8. Hardware schematic block of the ultrasound transducer. delay, amplitude, and fast Fourier transform measurement of
the echo signal. These vectors are then transferred to a host
computer (i.e., main processing unit) via its serial link for fur-
the acoustic impedance of these sensors is much higher than ther processing, together with other features that are collected
the liquid acoustic impedance (Fig. 9). These two arrays of from other sensors of the IMS-MPFM (e.g., capacitance and
sensors (ten sensors of 4 MHz and eight sensors of 23 kHz conductance sensors). Following extensive experiments, it was
resonance frequency each) are coupled to the multiflow flanged observed that all the above input vectors are a function not only
pipeline and placed along different cross-sectional portions of of the fluid density (ρ) but also of the total mass flow rate (M )
the multiflow pipeline. Each of the high-frequency sensors is (as addressed in Section III). The same phenomenon was also
allocated in a round-robin cycle with a periodic 100-ms time observed for both the conductance and capacitance sensors.

Authorized licensed use limited to: Centro Federal de Educacao Tec do Rio de Janeiro. Downloaded on July 09,2021 at 23:25:01 UTC from IEEE Xplore. Restrictions apply.
MERIBOUT et al.: MULTISENSOR DEVICE FOR REAL-TIME MULTIPHASE FLOW METERING IN OIL FIELDS 1513

Fig. 11. Plot showing the average capacitance response versus water-cut
Fig. 10. Plot showing average ultrasound delay versus water-cut (in percent) (in percent) for different ranges of the mass flow rate.
for different ranges of the mass flow rate.
40%, the average capacitance increases with the water-cut and
VI. IMS-MPFM C ALIBRATION AND THE reaches the saturation for water-cut greater than 60% from
P ATTERN R ECOGNITION A LGORITHM which a sudden jump occurs. This is because beyond this value
of the water-cut, the process fluid becomes mainly conductive,
A. IMS-MPFM Training Strategy
and therefore, the capacitance probe becomes short circuit. This
Prior to design the pattern recognition algorithm, pattern constitutes one of the main drawbacks of the capacitance probe.
extraction and meter calibration tasks were performed. Indi- However, similarly to the ultrasonic probe, the response of this
vidual flow rates corresponding to various flow regimes in the sensor is nonlinear and depends on the total mass flow rate. One
ranges [0%–100% water fraction], [0%–100% gas fraction], possible reason for these nonlinear responses is the fact that
and [200–600 liters/min total flow rate] have been carried out the multiphase flow exhibits a nonlinear function of the sensors
in the multiphase flow loop. For each of these experiments, the [e.g., the differential pressure in (6) depends on the temperature

actual flow rate of each phase FWa , FOa , and FGa , as described of the mixed fluid and is proportional to the square of the fluid
in Sections III and IV, have been measured by the reference wa- velocity]. The second reason is due to the uncertainties of the
ter, oil, and gas flow meters, respectively. These three variables sensors and their inability to cope with some situations. For
are the desired outputs for the pattern recognition algorithm, instance the information provided by the ultrasonic sensors is
whereas its inputs are the outputs of the IMS-MPFM sensors, not complete since these sensors do not cover the whole section
which are simultaneously grabbed with these three variables. of the pipe. The use of the 90◦ bent pipe elbows reduced to
The training process was done separately for the total flow some extent this effect, making the flow more homogeneous, al-
rate and phases composition to develop different parameter though this was not enough. Nevertheless, one of the best ways
spaces. Various pattern recognition algorithms relying on the to tackle the parameter spaces corresponding to Figs. 6, 10,
physical model of the fluid have then been applied on these and [11] is to provide a neural network for each range of the
parameter spaces to deduce the best one in terms of accuracy mass flow rate.
and computation time [20], [21]. In this paper, the use of the temperature variable as an addi-
tional vector for the proposed pattern recognition algorithm is
not considered since the multiphase flow loop was built within
B. Preliminary Compilation of the Learning Database for an indoor air-conditioned laboratory, where the temperature
Density Measurement was kept constant at room conditions (i.e., around 23 ◦ C).
In Fig. 10, four plots corresponding to four different flow However, since the temperature of the process fluid might vary
rates ranges are shown. Each of these plots corresponds to the function of other parameters than the room temperature (e.g.,
average delay for different water-cuts within a predefined range phase friction and internal pipe–fluid friction), the use of the
of the total mass flow rate M . As expected from Section V-A, temperature sensor would provide more accurate results and is,
the parameter space of this figure shows that the actual water- thus, considered as a future improvement factor for the next-
cut γa is inversely proportional to the average delay of the generation IMS-MPFM.
ultrasonic waves through the mixture. However, due to the
effect of the flow velocity (6), it was observed that in some
C. Pattern Recognition Algorithm
situations, the same water-cut values correspond to slightly dif-
ferent averaged ultrasound outputs (e.g., feature points A and B Fig. 12 shows the block diagram of the pattern recognition
or points C and D). This led us to use the venturi probe algorithm that has been embedded into the IMS-MPFM. The
as a second input to estimate the volume flow rate. Another algorithm performs in two sequential steps: First, it determines
observation in Fig. 10 is that γa follows a nonlinear and non- the fraction composition using the data fusion of the capaci-
parametric function of the average ultrasonic probe’s output. tance, conductance, ultrasound, pressure, and venturi sensors.
Fig. 11 shows another 2-D parameter space corresponding to Then, it computes the flow rate of the mixture using the
the capacitance probe. As expected, for water-cut ranges below estimated total density in addition to the differential pressure

Authorized licensed use limited to: Centro Federal de Educacao Tec do Rio de Janeiro. Downloaded on July 09,2021 at 23:25:01 UTC from IEEE Xplore. Restrictions apply.
1514 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 59, NO. 6, JUNE 2010

a) Controller design: In Fig. 13, the controller enables


only one output of the first three modules, depending on the
actual fraction ranges. For instance, from Fig. 14, which shows
the averaged error on the water-cut [i.e., the water fraction
γ ∗ 100 (in percent)] using either the ultrasonic, capacitance,
or conductance sensors following several experiments in our
laboratory-scale flow loop, it can be deduced that the lowest
amount of error is obtained when the capacitance and conduc-
tance sensors are used in the water-cut percent ranges [0–32]
and [47–81], respectively. The remaining other ranges, namely,
[33–46] and [82–100], are more suitable to be handled by the
ultrasonic sensors. This leads to consider the proposed device
as a good candidate to handle fluids with very high water-cut,
which are increasingly available in out-aged oil wells.
b) Neural network features for fluid composition: Fol-
lowing extensive experiments, where different number of hid-
den layers, number of nodes, and activation functions have
been evaluated, the logsig function, with one single hidden
layer and six neurons per hidden layer, was found to be the
most appropriate in terms of accuracy and speed (since larger
number of neurons in the hidden layer did not reduce the
error significantly). Hence, the first layer of each bank is fed
Fig. 12. Software architecture of the MPFM.
to two sensors: 1) the venturi sensor and 2) the conductance,
and venturi sensors. All the sensors are fed to spatial digital capacitance, ultrasonic, or differential pressure sensor. The last
filters to damp possible random noise, which are mainly due to layer generates the phase fractions α, β, or γ. The nodes in the
the vibration of the pipes of the flow loop and to the unsteady hidden layer are connected to all nodes in the adjacent layers.
state of the flow regime. To each sensor X, the following first- Each connection carries a weight wi,j . Hence, the output of a
order low-pass filter was applied: node j in the hidden layer can be expressed as follows:



(N/2)−1
5

Xaverage =
X(t + t0 )
(9) uj = gj wij × xi (10)
N i=1
t0 =−N/2

where gj is the logsig activation function, which, for a variable


where N is the window size of the spatial filter, which has been
x, can be expressed as follows [23]:
set to eight since this value could substantially reduce the level
of noise while leading to a low computation time by substituting 1
the division in (9) by simple right-shift instructions. logsig(x) = . (11)
1 + exp(−x)

D. Fluid Composition Determination Using a


Hierarchical Neural Network E. Flow Rate Determination Using a
The algorithm to determine the fluid composition is com- Hierarchical Neural Network
posed of four modules PR1, PR2, PR3, and PR4, which are con- Similarly to the fluid composition, another different multi-
nected to the conductance, capacitance, ultrasonic, and pressure layer neural network is used to determine the total mass flow
sensors, respectively (in addition to the venturi sensor) (Fig. 13) rate of the mixed fluid (Fig. 15). At first, the algorithm uses (6)
to determine the water fraction γ, oil fraction β, and gas fraction to provide an initial estimate of the total mass flow rate Ti using
α. As mentioned in Section IV-A [see (6)], the venturi and the variable θ as input, where (see Section IV)
differential pressure sensors are used to estimate the amount 
of gas within the mixed fluid (using PR4 module in Fig. 13). θ = ρ × (k − venturi) (12)
Following the 2-D parameter spaces shown in Fig. 6, the latter
performs in two sequential steps: First, it determines the class of where k is a constant function of the offset introduced by the
the actual sample using its corresponding venturi value. Then, it data acquisition card. The density of the fluid mixture is then
applies the neural network algorithm corresponding to that class computed as follows:
to determine the gas fraction. This two-step mechanism (i.e.,
classification and neural network step) is similarly performed ρ (in kilograms per cubic meter)
in PR3 and PR2 modules to explore the 2-D nonlinear param-
= γ + 0.81(β) + (1.2 × 10−3 × α)
eter spaces shown in Figs. 10 and 11, respectively. A similar
approach is adopted for thePR1 module as well. × 1000 kg/m3 × 100 (13)

Authorized licensed use limited to: Centro Federal de Educacao Tec do Rio de Janeiro. Downloaded on July 09,2021 at 23:25:01 UTC from IEEE Xplore. Restrictions apply.
MERIBOUT et al.: MULTISENSOR DEVICE FOR REAL-TIME MULTIPHASE FLOW METERING IN OIL FIELDS 1515

Fig. 13. Hierarchical neural network for fluid fractions determination.

overcome during the second phase of the algorithm, where the


above density value ρ is fed together with the actual differential
pressure ΔPi into another neural network module of 14 neural
network banks to generate the mass flow rate residual ΔTi to be
aggregated to Ti . Note that the neural network blocks FFN-1-
to-n in Fig. 15, which are used for flow rate determination, are
different from those in Fig. 13, which are used to determine
the fluid composition. In addition, and similarly to Fig. 13,
one single hidden layer (six nodes in the hidden layer) with
logsig activation function was provided for each individual
neural network block since it provided the best satisfactory
results in terms of accuracy and execution time. Neural network
architectures with a larger number of hidden layers did not
substantially reduce the overall error.

VII. E XPERIMENTAL R ESULTS AND V ALIDATION


To cover the maximal ranges of fluid compositions and flow
Fig. 14. Average aggregated errors for conductance, capacitance, and ultra- rates, extensive experiments have been performed to produce
sonic sensors. the following two databases:
1) a flow rate database with 240 sample elements (of ρ,
venturi, and ΔP probes) to cover mass flow rates M from
0 to 47 000 kg/h;
2) a phase fractions database with 470 sample elements (of
conductance, capacitance, ultrasonic, ρ, venturi, and ΔP
probes) to cover flow composition from 0% to 100% for
αa , βa , and γa , respectively.

A. Comparison of Accuracy for Offline Experiments


Prior to starting the online experiments, an offline evaluation
Fig. 15. Hierarchical neural network for flow rate determination. of the two above databases and the associated hierarchical
neural network algorithms has been performed for various
where the three parameters α, β, and γ are determined during levels of noise that can easily be aggregated to sensors outputs
the fluid composition determination phase. since these latest are intended to operate in harsh environ-
This first estimation phase is performed by exploring the ments. Thus, slight variations of the sensors output character-
parameter space of Fig. 5. However, as shown in this figure, istics might be expected. Hence, various levels of noise are
different combinations of the density and venturi probe might added to the simulated input signals: signal-to-noise ratio = ∞
lead to more than one candidate total mass flow rate. This is (no noise), 40 dB, 30 dB, and 20 dB. The evaluation criteria

Authorized licensed use limited to: Centro Federal de Educacao Tec do Rio de Janeiro. Downloaded on July 09,2021 at 23:25:01 UTC from IEEE Xplore. Restrictions apply.
1516 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 59, NO. 6, JUNE 2010

TABLE I
NEURAL NETWORK RESULTS FOR VARIOUS LEVELS OF NOISE ADDED TO THE TRAINING AND VALIDATION SETS

Fig. 16. Plots showing (a) the water-cut response of the (b) capacitance, (c) conductance, and (d) ultrasound sensors.

use both the maximum error and the standard deviation σ (i.e., Following the construction of the two above databases, ex-
standard deviation of the error between expected and computed tensive online experiments have been conducted on the three
distances). phase flow loop under various flow regimes, water-cut, and
In addition to the samples of the two databases, 140 new flow-rate ranges using the platform shown in Fig. 7 and using
validation tests have been used during this evaluation phase. the same above two databases.
Table I summarizes the results.
It can be seen that a greater precision is obtained when the
B. Ultrasound Sensors for Accurate Flow Composition
noise level added to the input is the smallest. This sensitivity
can be strongly reduced by carrying out the training phase with As mentioned in Sections I and II, most existing multi-
noisy input data. The other observation is that overall, both the phase flow metering devices are usually not accurate when
error and standard deviation are higher for the flow rate than dealing with high water-cut values. In addition, neither the
for the flow composition determination. This might be due to conductance nor capacitance sensors can easily detect the
the inaccuracy of the venturi probe to cope accurately with the oil–water composition in the range of 40%–60%. Fig. 16
whole range of the mass flow rate since it is used as a class shows the merit of using our ultrasonic probe within this
discriminator only for the flow composition (see Fig. 13) and as range [Fig. 16(d)], where neither the capacitance [Fig. 16(b)]
a neural network input for flow rate determination. This might nor the conductance [Fig. 16(c)] sensors could respond [11].
be overcome by using more than one venturi probe to cover The error was 2.3% and −0.26% for water and oil fractions,
smaller mass flow rate ranges. respectively.

Authorized licensed use limited to: Centro Federal de Educacao Tec do Rio de Janeiro. Downloaded on July 09,2021 at 23:25:01 UTC from IEEE Xplore. Restrictions apply.
MERIBOUT et al.: MULTISENSOR DEVICE FOR REAL-TIME MULTIPHASE FLOW METERING IN OIL FIELDS 1517

than 90%. Fig. 20 shows the experimental results for the total
flow rate determination in the [0–40 m3 /h] range. The MPFM
could also easily track the reference total flow rate, leading to
a small relative error of 2.5%. One of the difficulties that we
have faced while conducting the tests is the relatively short
time of the experiments due to the small storage capacity of
the water and oil tanks (i.e., 1 m3 capacity for each tank). Our
future plan is to build a larger flow loop with longer pipes
(to help create flow regimes more similar to the ones obtained
in the oil field) and larger tanks (more than 12 m3 capacity).
In addition, an emulsion layer detection device to detect the
pure oil and water interfaces would be installed in the storage
tank to continuously run the experiments while avoiding the
introduction of the emulsion layer into the flow loop. The
overall relative errors are provided in Section VII-D.

D. Accuracy of the Proposed Multiphase Flow Meter for


Online Experiments
For online experiments, the error (relative error) was mea-
sured in terms of percentage uncertainty relative to the flow
Fig. 17. Experimental results for density measurement for (a) water-cut and
(b) gas fraction.
rates of each phase. i.e., oil, gas, and water flow rates. The errors
for water, gas, and oil, respectively, are measured as follows:
FWa − FW
Error(W) [in percent] = × 100% (14)
FW
FGa − FG
Error(G) [in percent] = × 100% (15)
FG
FOa − FO
Error(O) [in percent] = × 100% (16)
FO
where, as mentioned in Section III, FW , FG , and FO are the
flow rates of water, gas, and oil, which are measured by the
IMS-MPFM, respectively, and FWa , FGa , and FOa are
the actual flow rates of water, gas, and oil, which are measured
by the single-phase flow meter. Table II illustrates the overall
Fig. 18. Experimental results for gas fraction determination in the [0%–60%] relative error for few experiments (i.e., 13 experiments). Hence,
range.
the aggregate relative error for each of the three phases (which
is the average of the absolute values of the relative errors
C. Experimental Results for Simultaneous Flow and through all the experiments) was less than 10%. Using the
Density Determination same database, the same result was obtained for hundreds of
other experiments (351 experiments) that have been conducted
Fig. 17(a) and (b) shows the results of several experiments for
for different combinations (i.e., different total flow rates up to
the determination of the water and gas fractions, respectively. It
45 m3 /h and different phases [0%–100%]). This is an improve-
can be observed that both the water-cut and gas fraction pro-
ment over other traditional MPFMs where the relative error is
vided by the proposed hierarchical neural network could track
more than 10% in most of the cases.
in real time (i.e., up to 2 ms sampling rate) the reference in the
ranges [20%–100% water-cut] and [0%–40% gas], respectively.
VIII. C ONCLUSION
In these experiments, the flow rate of each phase was randomly
increased and decreased. For both directions, the IMS-MPFM This paper has shown a new cost-effective approach for
could track even the short-period fluctuations of the reference. multiphase flow measurement. It consists of a nonradioac-
These fluctuations are due to the gas phase, where the substan- tive IMS-MPFM, which is composed of several electrical and
tial pipe vibrations caused by the gas compressor did not help acoustic sensors and does not include any moving part. Both its
on providing a steady gas flow. This phenomenon disappears hardware and software parts were presented. This latest consists
when the gas is completely omitted—a situation in which the of two hierarchical neural network algorithms to determine the
IMS-MPFM can provide even more accurate results, regardless phase fractions and total mass flow rate, respectively. Results of
of the flow regime [1]. Figs. 18 and 19 show other experimental extensive online experiments, which have been conducted in an
results for gas fractions in the [0%–60%] and [0%–100%] in-house multiphase flow loop, show that the meter can track
ranges, respectively. Hence, the MPFM could accurately track multiphase flow rates of more than 90% gas for an average
the variations of the gas phase, even for gas fractions greater relative error of less than 10%. The other equally important

Authorized licensed use limited to: Centro Federal de Educacao Tec do Rio de Janeiro. Downloaded on July 09,2021 at 23:25:01 UTC from IEEE Xplore. Restrictions apply.
1518 IEEE TRANSACTIONS ON INSTRUMENTATION AND MEASUREMENT, VOL. 59, NO. 6, JUNE 2010

Fig. 19. Experimental results for gas fraction determination in the [0%–100%] range.

R EFERENCES
[1] M. M. A. Al-Naamani and K. Al-Busaidi, “An industrial-prototype
acoustic array for real-time emulsion layer detection in oil storage tanks,”
IEEE Trans. Instrum. Meas., 2009. to be published.
[2] C. H. Lo, Y. K. Wong, and A. B. Rad, “Intelligent system for process
supervision and fault diagnostic in dynamic physical systems,” IEEE
Trans. Ind. Electron., vol. 53, no. 2, pp. 581–592, Apr. 2006.
[3] J. Chaoki, L. Larachi, and M. P. Dudoković, Non-Invasive Monitoring of
Multiphase Flows. Amsterdam, The Netherlands: Elsevier, 2006.
[4] T. S. Whitaker, “A review of multiphase flowmeters and future devel-
opment potential,” in Proc. 6th Int. Conf. FLOMEKO, Seoul, Korea,
Oct. 2005, pp. 628–634.
[5] M. Zamora, “An FPGA implementation of a digital Coriolis mass flow
Fig. 20. Experimental results for flow rate determination in the [0%–100%] metering drive system,” IEEE Trans. Ind. Electron., vol. 55, no. 7,
range. pp. 2820–2831, Jul. 2008.
[6] D. D’Aessandro, “Application of microprocessor to fuel measurement
TABLE II system,” IEEE Trans. Ind. Electron., vol. IE-30, no. 2, pp. 164–170,
RELATIVE ERROR OF THE PROPOSED MPFM May 1983.
[7] M. Kawai, H. Miyagi, J. Nakano, and Y. Kondo, “Toyota’s new
microprocessor-based diesel engine control for passenger cars,” IEEE
Trans. Ind. Electron., vol. 32, no. 4, pp. 289–293, Nov. 1995.
[8] P. K. Chande, “Ultrasonic flow velocity sensor based on picosecond tim-
ing system,” IEEE Trans. Ind. Electron., vol. IE-33, no. 2, pp. 162–165,
May 1986.
[9] M. J. W. Povey, Ultrasonic Techniques for Fluid Characterization.
New York: Academic, 2005.
[10] G. E. P. Box, W. G. Hunter, and J. S. Hunter, Statistics for Experimenters.
New York: Wiley, 2004.
[11] J. W. Murdock, “Two phase flow measurement with orifices,” Trans.
ASME, J. Basic Eng., vol. 84, pp. 419–433, Dec. 2006.
[12] D. Farch and J. Agar, “Apparatus and method for measuring two- or
three-phase fluid flow utilizing one or more momentum flow meters and a
volumetric flow meter,” U.S. Patent 5 461 930, Sep. 3, 1996.
[13] R. H. Boll and W. L. Ghering, “System for measuring entrained solid
flow,” U.S. Patent 4 231 262, Nov. 4, 1980.
[14] G. Yanfeng, Z. Jinwu, and S. Gang, “Measurement of two-phase flow rate
based on slotted orifice couple and neural network ensemble,” in Proc.
IEEE Int. Conf. Inf. Acquisition, 2006, pp. 1037–1041.
[15] V. R. Bom, M. C. Clarijs, C. W. E. van Eijk, Z. I. Kolar, J. Frieling,
L. A. Scheers, and G. J. Miller,, “Accuracy aspects in multiphase flow
metering,” IEEE Trans. Nucl. Sci., vol. 48, no. 6, pp. 2335–2339,
Dec. 2001.
[16] E. J. Morton, R. D. Luggar, M. J. Key, A. Kundu, L. M. N. Tavora, and
W. B. Gilboy, “Development of a high speed X-ray tomography system
for multiphase flow imaging,” IEEE Trans. Nucl. Sci., vol. 46, no. 3,
pp. 380–384, Jun. 1999.
[17] S. Gehrke and K.-E. Wirth, “Application of conventional- and dual-energy
X-ray tomography in process engineering,” IEEE Sensors J., vol. 5, no. 2,
advantage of this meter is that by integrating an in-house made pp. 183–187, Apr. 2005.
ultrasonic sonar array, it could overcome the limitations of tra- [18] L. Sun, D. Chen, and G. Zheng, “Analysis of performance and capacitance
ditional electrical sensors to operate in the full water-cut range sensitivity distributions of sensor for electrical capacitance tomography
system,” in Proc. 6th WCICA, 2006, vol. 1, pp. 4977–4981.
(i.e., 40%–60% water-cut) and for high water-cut. These are [19] S. I. Al-Mously and A. Y. Ahmed, “The use of a coaxial capacitor as
some reasons why the proposed flow measurement technique a capacitance sensor for phase percentage determination in multiphase
may constitute a new class of flow metering technology not pipelines,” in Proc. 10th MELECON, 2000, vol. 2, pp. 742–745.
[20] S. Jung and S. su Kim, “Hardware implementation of a real-time neural
only in petrochemical applications, but in other industries as network controller with a DSP and an FPGA for nonlinear systems,” IEEE
well (e.g., food and mine industries). Trans. Ind. Electron., vol. 54, no. 1, pp. 265–271, Feb. 2007.

Authorized licensed use limited to: Centro Federal de Educacao Tec do Rio de Janeiro. Downloaded on July 09,2021 at 23:25:01 UTC from IEEE Xplore. Restrictions apply.
MERIBOUT et al.: MULTISENSOR DEVICE FOR REAL-TIME MULTIPHASE FLOW METERING IN OIL FIELDS 1519

[21] J. Chen, Z. Zhou, X. Zhang, and T. Yu, “Terahertz process tomog- Ahmed M. Al-Naamany received the B.Sc. degree
raphy using in multiphase flow measurement,” in Proc. WCICA, 2004, (with honors) in multidisciplinary engineering from
pp. 3727–3729. Widener University, Chester, PA, in 1986, the M.Sc.
[22] L. C. Lynnworth, Ultrasonic Measurements for Process Control. degree in electrical engineering and computer con-
New York: Academic, 2006. trol from Drexel University, Philadelphia, PA, in
[23] H. Zhuang, K.-S. Low, and W.-Y. Yau, “A pulsed neural network with on- 1990, and the Ph.D. degree from the University of
chip learning and its practical applications,” IEEE Trans. Ind. Electron., Manchester, Manchester, U.K., in 1995.
vol. 54, no. 1, pp. 34–42, Feb. 2007. He lectured on computer architecture, computer
control, and artificial intelligence applications. Since
1996, he has been with the Department of Electrical
and Computer Engineering, Sultan Qaboos Univer-
sity, Muscat, Oman, where he is currently an Assistant Professor.
Dr. Al-Naamany is a member of the IEEE Computer Society and the IEEE
Control Systems Society. He was the recipient of a Fulbright Scholarship.
Mahmoud Meribout (S’85–M’91) received the
Ph.D. degree in electronics engineering from the Uni-
versity of Technology of Compiegne, Compiegne, Ali Al-Bimani received the M.Sc. and Ph.D. degrees
France, in January 3, 1995. in petroleum engineering from the University of
From November 1995 to October 2000, he was Southern California, Los Angeles.
with the NTT Corporation, Japan, and then with the He is currently an Associate Professor with he
NEC Corporation, Japan, where he has been involved Department of Petroleum and Chemical Engineer-
in several projects related to embedded systems de- ing, Sultan Qaboos University, Muscat, Oman. His
sign. In November 2000, he joined the Department of research interests include well testing, reservoir en-
Electrical and Computer Engineering, Sultan Qaboos gineering, rock and fluid properties, and computer
University, Muscat, Oman. Since August 2008, he applications in petroleum engineering.
has been an Associate Professor with the Department of Electrical Engineering,
Petroleum Institute, Abu Dhabi, United Arab Emirates. He is the holder of
several Japanese and American patents related to embedded system design.
Dr. Meribout received the NTT Best Award in 1998 for his research and Khamis Al-Busaidi received the M.Sc. and Ph.D. degrees in petroleum engi-
development records.
neering from the University of Manchester, Manchester, U.K.
He is currently with Petroleum Development Oman (PDO), Muscat, Oman,
where he has been involved in several industrial projects such as smart fields
and multiphase flow meters. He has been active on introducing these technolo-
gies to PDO and is an active consultant for Sultan Qaboos University, Muscat.

Nabeel Z. Al-Rawahi (S’94–M’97) received the Adel Meribout received the B.Eng. degree in elec-
Ph.D. degree in mechanical engineering from the trical and computer engineering from the University
University of Michigan, Ann Arbor, in 2002. of Constantine, Constantine, Algeria, in 1995.
He was an Assistant Professor with the University From January 2007 to February 2008, he was a
of Michigan, where he was involved in several in- Research Assistant with the Department of Electrical
dustrial and academic projects involving multiphase and Computer Engineering, Sultan Qaboos Univer-
flow and desalination. He is currently an Assistant sity, Muscat, Oman. He is currently with the De-
Professor with the Department of Mechanical and partment of Instrumentation and Control, Sonatrach
Industrial Engineering, Sultan Qaboos University, Corporation, Algiers, Algeria (an Algerian oil com-
Muscat, Oman. His research interests include multi- pany), as an Engineer. His research interests include
phase flow, computational fluid dynamics, desalina- embedded Instrumentation design and implementa-
tion processes, and renewable energy. tion in the areas of oil and gas.

Authorized licensed use limited to: Centro Federal de Educacao Tec do Rio de Janeiro. Downloaded on July 09,2021 at 23:25:01 UTC from IEEE Xplore. Restrictions apply.

You might also like